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#include "dm-ima.h"
12
13#include <linux/init.h>
14#include <linux/module.h>
15#include <linux/mutex.h>
16#include <linux/sched/mm.h>
17#include <linux/sched/signal.h>
18#include <linux/blkpg.h>
19#include <linux/bio.h>
20#include <linux/mempool.h>
21#include <linux/dax.h>
22#include <linux/slab.h>
23#include <linux/idr.h>
24#include <linux/uio.h>
25#include <linux/hdreg.h>
26#include <linux/delay.h>
27#include <linux/wait.h>
28#include <linux/pr.h>
29#include <linux/refcount.h>
30#include <linux/part_stat.h>
31#include <linux/blk-crypto.h>
32#include <linux/blk-crypto-profile.h>
33
34#define DM_MSG_PREFIX "core"
35
36/*
37 * Cookies are numeric values sent with CHANGE and REMOVE
38 * uevents while resuming, removing or renaming the device.
39 */
40#define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE"
41#define DM_COOKIE_LENGTH 24
42
43/*
44 * For REQ_POLLED fs bio, this flag is set if we link mapped underlying
45 * dm_io into one list, and reuse bio->bi_private as the list head. Before
46 * ending this fs bio, we will recover its ->bi_private.
47 */
48#define REQ_DM_POLL_LIST REQ_DRV
49
50static const char *_name = DM_NAME;
51
52static unsigned int major = 0;
53static unsigned int _major = 0;
54
55static DEFINE_IDR(_minor_idr);
56
57static DEFINE_SPINLOCK(_minor_lock);
58
59static void do_deferred_remove(struct work_struct *w);
60
61static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
62
63static struct workqueue_struct *deferred_remove_workqueue;
64
65atomic_t dm_global_event_nr = ATOMIC_INIT(0);
66DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
67
68void dm_issue_global_event(void)
69{
70 atomic_inc(&dm_global_event_nr);
71 wake_up(&dm_global_eventq);
72}
73
74DEFINE_STATIC_KEY_FALSE(stats_enabled);
75DEFINE_STATIC_KEY_FALSE(swap_bios_enabled);
76DEFINE_STATIC_KEY_FALSE(zoned_enabled);
77
78/*
79 * One of these is allocated (on-stack) per original bio.
80 */
81struct clone_info {
82 struct dm_table *map;
83 struct bio *bio;
84 struct dm_io *io;
85 sector_t sector;
86 unsigned sector_count;
87 bool is_abnormal_io:1;
88 bool submit_as_polled:1;
89};
90
91static inline struct dm_target_io *clone_to_tio(struct bio *clone)
92{
93 return container_of(clone, struct dm_target_io, clone);
94}
95
96void *dm_per_bio_data(struct bio *bio, size_t data_size)
97{
98 if (!dm_tio_flagged(clone_to_tio(bio), DM_TIO_INSIDE_DM_IO))
99 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
100 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
101}
102EXPORT_SYMBOL_GPL(dm_per_bio_data);
103
104struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
105{
106 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
107 if (io->magic == DM_IO_MAGIC)
108 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
109 BUG_ON(io->magic != DM_TIO_MAGIC);
110 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
111}
112EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
113
114unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
115{
116 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
117}
118EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
119
120#define MINOR_ALLOCED ((void *)-1)
121
122#define DM_NUMA_NODE NUMA_NO_NODE
123static int dm_numa_node = DM_NUMA_NODE;
124
125#define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
126static int swap_bios = DEFAULT_SWAP_BIOS;
127static int get_swap_bios(void)
128{
129 int latch = READ_ONCE(swap_bios);
130 if (unlikely(latch <= 0))
131 latch = DEFAULT_SWAP_BIOS;
132 return latch;
133}
134
135struct table_device {
136 struct list_head list;
137 refcount_t count;
138 struct dm_dev dm_dev;
139};
140
141/*
142 * Bio-based DM's mempools' reserved IOs set by the user.
143 */
144#define RESERVED_BIO_BASED_IOS 16
145static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
146
147static int __dm_get_module_param_int(int *module_param, int min, int max)
148{
149 int param = READ_ONCE(*module_param);
150 int modified_param = 0;
151 bool modified = true;
152
153 if (param < min)
154 modified_param = min;
155 else if (param > max)
156 modified_param = max;
157 else
158 modified = false;
159
160 if (modified) {
161 (void)cmpxchg(module_param, param, modified_param);
162 param = modified_param;
163 }
164
165 return param;
166}
167
168unsigned __dm_get_module_param(unsigned *module_param,
169 unsigned def, unsigned max)
170{
171 unsigned param = READ_ONCE(*module_param);
172 unsigned modified_param = 0;
173
174 if (!param)
175 modified_param = def;
176 else if (param > max)
177 modified_param = max;
178
179 if (modified_param) {
180 (void)cmpxchg(module_param, param, modified_param);
181 param = modified_param;
182 }
183
184 return param;
185}
186
187unsigned dm_get_reserved_bio_based_ios(void)
188{
189 return __dm_get_module_param(&reserved_bio_based_ios,
190 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
191}
192EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
193
194static unsigned dm_get_numa_node(void)
195{
196 return __dm_get_module_param_int(&dm_numa_node,
197 DM_NUMA_NODE, num_online_nodes() - 1);
198}
199
200static int __init local_init(void)
201{
202 int r;
203
204 r = dm_uevent_init();
205 if (r)
206 return r;
207
208 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
209 if (!deferred_remove_workqueue) {
210 r = -ENOMEM;
211 goto out_uevent_exit;
212 }
213
214 _major = major;
215 r = register_blkdev(_major, _name);
216 if (r < 0)
217 goto out_free_workqueue;
218
219 if (!_major)
220 _major = r;
221
222 return 0;
223
224out_free_workqueue:
225 destroy_workqueue(deferred_remove_workqueue);
226out_uevent_exit:
227 dm_uevent_exit();
228
229 return r;
230}
231
232static void local_exit(void)
233{
234 flush_scheduled_work();
235 destroy_workqueue(deferred_remove_workqueue);
236
237 unregister_blkdev(_major, _name);
238 dm_uevent_exit();
239
240 _major = 0;
241
242 DMINFO("cleaned up");
243}
244
245static int (*_inits[])(void) __initdata = {
246 local_init,
247 dm_target_init,
248 dm_linear_init,
249 dm_stripe_init,
250 dm_io_init,
251 dm_kcopyd_init,
252 dm_interface_init,
253 dm_statistics_init,
254};
255
256static void (*_exits[])(void) = {
257 local_exit,
258 dm_target_exit,
259 dm_linear_exit,
260 dm_stripe_exit,
261 dm_io_exit,
262 dm_kcopyd_exit,
263 dm_interface_exit,
264 dm_statistics_exit,
265};
266
267static int __init dm_init(void)
268{
269 const int count = ARRAY_SIZE(_inits);
270 int r, i;
271
272#if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
273 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
274 " Duplicate IMA measurements will not be recorded in the IMA log.");
275#endif
276
277 for (i = 0; i < count; i++) {
278 r = _inits[i]();
279 if (r)
280 goto bad;
281 }
282
283 return 0;
284bad:
285 while (i--)
286 _exits[i]();
287
288 return r;
289}
290
291static void __exit dm_exit(void)
292{
293 int i = ARRAY_SIZE(_exits);
294
295 while (i--)
296 _exits[i]();
297
298 /*
299 * Should be empty by this point.
300 */
301 idr_destroy(&_minor_idr);
302}
303
304/*
305 * Block device functions
306 */
307int dm_deleting_md(struct mapped_device *md)
308{
309 return test_bit(DMF_DELETING, &md->flags);
310}
311
312static int dm_blk_open(struct block_device *bdev, fmode_t mode)
313{
314 struct mapped_device *md;
315
316 spin_lock(&_minor_lock);
317
318 md = bdev->bd_disk->private_data;
319 if (!md)
320 goto out;
321
322 if (test_bit(DMF_FREEING, &md->flags) ||
323 dm_deleting_md(md)) {
324 md = NULL;
325 goto out;
326 }
327
328 dm_get(md);
329 atomic_inc(&md->open_count);
330out:
331 spin_unlock(&_minor_lock);
332
333 return md ? 0 : -ENXIO;
334}
335
336static void dm_blk_close(struct gendisk *disk, fmode_t mode)
337{
338 struct mapped_device *md;
339
340 spin_lock(&_minor_lock);
341
342 md = disk->private_data;
343 if (WARN_ON(!md))
344 goto out;
345
346 if (atomic_dec_and_test(&md->open_count) &&
347 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
348 queue_work(deferred_remove_workqueue, &deferred_remove_work);
349
350 dm_put(md);
351out:
352 spin_unlock(&_minor_lock);
353}
354
355int dm_open_count(struct mapped_device *md)
356{
357 return atomic_read(&md->open_count);
358}
359
360/*
361 * Guarantees nothing is using the device before it's deleted.
362 */
363int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
364{
365 int r = 0;
366
367 spin_lock(&_minor_lock);
368
369 if (dm_open_count(md)) {
370 r = -EBUSY;
371 if (mark_deferred)
372 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
373 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
374 r = -EEXIST;
375 else
376 set_bit(DMF_DELETING, &md->flags);
377
378 spin_unlock(&_minor_lock);
379
380 return r;
381}
382
383int dm_cancel_deferred_remove(struct mapped_device *md)
384{
385 int r = 0;
386
387 spin_lock(&_minor_lock);
388
389 if (test_bit(DMF_DELETING, &md->flags))
390 r = -EBUSY;
391 else
392 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
393
394 spin_unlock(&_minor_lock);
395
396 return r;
397}
398
399static void do_deferred_remove(struct work_struct *w)
400{
401 dm_deferred_remove();
402}
403
404static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
405{
406 struct mapped_device *md = bdev->bd_disk->private_data;
407
408 return dm_get_geometry(md, geo);
409}
410
411static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
412 struct block_device **bdev)
413{
414 struct dm_target *ti;
415 struct dm_table *map;
416 int r;
417
418retry:
419 r = -ENOTTY;
420 map = dm_get_live_table(md, srcu_idx);
421 if (!map || !dm_table_get_size(map))
422 return r;
423
424 /* We only support devices that have a single target */
425 if (map->num_targets != 1)
426 return r;
427
428 ti = dm_table_get_target(map, 0);
429 if (!ti->type->prepare_ioctl)
430 return r;
431
432 if (dm_suspended_md(md))
433 return -EAGAIN;
434
435 r = ti->type->prepare_ioctl(ti, bdev);
436 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
437 dm_put_live_table(md, *srcu_idx);
438 msleep(10);
439 goto retry;
440 }
441
442 return r;
443}
444
445static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
446{
447 dm_put_live_table(md, srcu_idx);
448}
449
450static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
451 unsigned int cmd, unsigned long arg)
452{
453 struct mapped_device *md = bdev->bd_disk->private_data;
454 int r, srcu_idx;
455
456 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
457 if (r < 0)
458 goto out;
459
460 if (r > 0) {
461 /*
462 * Target determined this ioctl is being issued against a
463 * subset of the parent bdev; require extra privileges.
464 */
465 if (!capable(CAP_SYS_RAWIO)) {
466 DMDEBUG_LIMIT(
467 "%s: sending ioctl %x to DM device without required privilege.",
468 current->comm, cmd);
469 r = -ENOIOCTLCMD;
470 goto out;
471 }
472 }
473
474 if (!bdev->bd_disk->fops->ioctl)
475 r = -ENOTTY;
476 else
477 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
478out:
479 dm_unprepare_ioctl(md, srcu_idx);
480 return r;
481}
482
483u64 dm_start_time_ns_from_clone(struct bio *bio)
484{
485 return jiffies_to_nsecs(clone_to_tio(bio)->io->start_time);
486}
487EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
488
489static bool bio_is_flush_with_data(struct bio *bio)
490{
491 return ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size);
492}
493
494static void dm_io_acct(struct dm_io *io, bool end)
495{
496 struct dm_stats_aux *stats_aux = &io->stats_aux;
497 unsigned long start_time = io->start_time;
498 struct mapped_device *md = io->md;
499 struct bio *bio = io->orig_bio;
500 unsigned int sectors;
501
502 /*
503 * If REQ_PREFLUSH set, don't account payload, it will be
504 * submitted (and accounted) after this flush completes.
505 */
506 if (bio_is_flush_with_data(bio))
507 sectors = 0;
508 else if (likely(!(dm_io_flagged(io, DM_IO_WAS_SPLIT))))
509 sectors = bio_sectors(bio);
510 else
511 sectors = io->sectors;
512
513 if (!end)
514 bdev_start_io_acct(bio->bi_bdev, sectors, bio_op(bio),
515 start_time);
516 else
517 bdev_end_io_acct(bio->bi_bdev, bio_op(bio), start_time);
518
519 if (static_branch_unlikely(&stats_enabled) &&
520 unlikely(dm_stats_used(&md->stats))) {
521 sector_t sector;
522
523 if (likely(!dm_io_flagged(io, DM_IO_WAS_SPLIT)))
524 sector = bio->bi_iter.bi_sector;
525 else
526 sector = bio_end_sector(bio) - io->sector_offset;
527
528 dm_stats_account_io(&md->stats, bio_data_dir(bio),
529 sector, sectors,
530 end, start_time, stats_aux);
531 }
532}
533
534static void __dm_start_io_acct(struct dm_io *io)
535{
536 dm_io_acct(io, false);
537}
538
539static void dm_start_io_acct(struct dm_io *io, struct bio *clone)
540{
541 /*
542 * Ensure IO accounting is only ever started once.
543 */
544 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
545 return;
546
547 /* Expect no possibility for race unless DM_TIO_IS_DUPLICATE_BIO. */
548 if (!clone || likely(dm_tio_is_normal(clone_to_tio(clone)))) {
549 dm_io_set_flag(io, DM_IO_ACCOUNTED);
550 } else {
551 unsigned long flags;
552 /* Can afford locking given DM_TIO_IS_DUPLICATE_BIO */
553 spin_lock_irqsave(&io->lock, flags);
554 if (dm_io_flagged(io, DM_IO_ACCOUNTED)) {
555 spin_unlock_irqrestore(&io->lock, flags);
556 return;
557 }
558 dm_io_set_flag(io, DM_IO_ACCOUNTED);
559 spin_unlock_irqrestore(&io->lock, flags);
560 }
561
562 __dm_start_io_acct(io);
563}
564
565static void dm_end_io_acct(struct dm_io *io)
566{
567 dm_io_acct(io, true);
568}
569
570static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
571{
572 struct dm_io *io;
573 struct dm_target_io *tio;
574 struct bio *clone;
575
576 clone = bio_alloc_clone(NULL, bio, GFP_NOIO, &md->mempools->io_bs);
577 tio = clone_to_tio(clone);
578 tio->flags = 0;
579 dm_tio_set_flag(tio, DM_TIO_INSIDE_DM_IO);
580 tio->io = NULL;
581
582 io = container_of(tio, struct dm_io, tio);
583 io->magic = DM_IO_MAGIC;
584 io->status = BLK_STS_OK;
585
586 /* one ref is for submission, the other is for completion */
587 atomic_set(&io->io_count, 2);
588 this_cpu_inc(*md->pending_io);
589 io->orig_bio = bio;
590 io->md = md;
591 spin_lock_init(&io->lock);
592 io->start_time = jiffies;
593 io->flags = 0;
594
595 if (static_branch_unlikely(&stats_enabled))
596 dm_stats_record_start(&md->stats, &io->stats_aux);
597
598 return io;
599}
600
601static void free_io(struct dm_io *io)
602{
603 bio_put(&io->tio.clone);
604}
605
606static struct bio *alloc_tio(struct clone_info *ci, struct dm_target *ti,
607 unsigned target_bio_nr, unsigned *len, gfp_t gfp_mask)
608{
609 struct mapped_device *md = ci->io->md;
610 struct dm_target_io *tio;
611 struct bio *clone;
612
613 if (!ci->io->tio.io) {
614 /* the dm_target_io embedded in ci->io is available */
615 tio = &ci->io->tio;
616 /* alloc_io() already initialized embedded clone */
617 clone = &tio->clone;
618 } else {
619 clone = bio_alloc_clone(NULL, ci->bio, gfp_mask,
620 &md->mempools->bs);
621 if (!clone)
622 return NULL;
623
624 /* REQ_DM_POLL_LIST shouldn't be inherited */
625 clone->bi_opf &= ~REQ_DM_POLL_LIST;
626
627 tio = clone_to_tio(clone);
628 tio->flags = 0; /* also clears DM_TIO_INSIDE_DM_IO */
629 }
630
631 tio->magic = DM_TIO_MAGIC;
632 tio->io = ci->io;
633 tio->ti = ti;
634 tio->target_bio_nr = target_bio_nr;
635 tio->len_ptr = len;
636 tio->old_sector = 0;
637
638 /* Set default bdev, but target must bio_set_dev() before issuing IO */
639 clone->bi_bdev = md->disk->part0;
640 if (unlikely(ti->needs_bio_set_dev))
641 bio_set_dev(clone, md->disk->part0);
642
643 if (len) {
644 clone->bi_iter.bi_size = to_bytes(*len);
645 if (bio_integrity(clone))
646 bio_integrity_trim(clone);
647 }
648
649 return clone;
650}
651
652static void free_tio(struct bio *clone)
653{
654 if (dm_tio_flagged(clone_to_tio(clone), DM_TIO_INSIDE_DM_IO))
655 return;
656 bio_put(clone);
657}
658
659/*
660 * Add the bio to the list of deferred io.
661 */
662static void queue_io(struct mapped_device *md, struct bio *bio)
663{
664 unsigned long flags;
665
666 spin_lock_irqsave(&md->deferred_lock, flags);
667 bio_list_add(&md->deferred, bio);
668 spin_unlock_irqrestore(&md->deferred_lock, flags);
669 queue_work(md->wq, &md->work);
670}
671
672/*
673 * Everyone (including functions in this file), should use this
674 * function to access the md->map field, and make sure they call
675 * dm_put_live_table() when finished.
676 */
677struct dm_table *dm_get_live_table(struct mapped_device *md,
678 int *srcu_idx) __acquires(md->io_barrier)
679{
680 *srcu_idx = srcu_read_lock(&md->io_barrier);
681
682 return srcu_dereference(md->map, &md->io_barrier);
683}
684
685void dm_put_live_table(struct mapped_device *md,
686 int srcu_idx) __releases(md->io_barrier)
687{
688 srcu_read_unlock(&md->io_barrier, srcu_idx);
689}
690
691void dm_sync_table(struct mapped_device *md)
692{
693 synchronize_srcu(&md->io_barrier);
694 synchronize_rcu_expedited();
695}
696
697/*
698 * A fast alternative to dm_get_live_table/dm_put_live_table.
699 * The caller must not block between these two functions.
700 */
701static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
702{
703 rcu_read_lock();
704 return rcu_dereference(md->map);
705}
706
707static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
708{
709 rcu_read_unlock();
710}
711
712static inline struct dm_table *dm_get_live_table_bio(struct mapped_device *md,
713 int *srcu_idx, blk_opf_t bio_opf)
714{
715 if (bio_opf & REQ_NOWAIT)
716 return dm_get_live_table_fast(md);
717 else
718 return dm_get_live_table(md, srcu_idx);
719}
720
721static inline void dm_put_live_table_bio(struct mapped_device *md, int srcu_idx,
722 blk_opf_t bio_opf)
723{
724 if (bio_opf & REQ_NOWAIT)
725 dm_put_live_table_fast(md);
726 else
727 dm_put_live_table(md, srcu_idx);
728}
729
730static char *_dm_claim_ptr = "I belong to device-mapper";
731
732/*
733 * Open a table device so we can use it as a map destination.
734 */
735static struct table_device *open_table_device(struct mapped_device *md,
736 dev_t dev, fmode_t mode)
737{
738 struct table_device *td;
739 struct block_device *bdev;
740 u64 part_off;
741 int r;
742
743 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
744 if (!td)
745 return ERR_PTR(-ENOMEM);
746 refcount_set(&td->count, 1);
747
748 bdev = blkdev_get_by_dev(dev, mode | FMODE_EXCL, _dm_claim_ptr);
749 if (IS_ERR(bdev)) {
750 r = PTR_ERR(bdev);
751 goto out_free_td;
752 }
753
754 /*
755 * We can be called before the dm disk is added. In that case we can't
756 * register the holder relation here. It will be done once add_disk was
757 * called.
758 */
759 if (md->disk->slave_dir) {
760 r = bd_link_disk_holder(bdev, md->disk);
761 if (r)
762 goto out_blkdev_put;
763 }
764
765 td->dm_dev.mode = mode;
766 td->dm_dev.bdev = bdev;
767 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev, &part_off, NULL, NULL);
768 format_dev_t(td->dm_dev.name, dev);
769 list_add(&td->list, &md->table_devices);
770 return td;
771
772out_blkdev_put:
773 blkdev_put(bdev, mode | FMODE_EXCL);
774out_free_td:
775 kfree(td);
776 return ERR_PTR(r);
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 (md->disk->slave_dir)
785 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
786 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
787 put_dax(td->dm_dev.dax_dev);
788 list_del(&td->list);
789 kfree(td);
790}
791
792static struct table_device *find_table_device(struct list_head *l, dev_t dev,
793 fmode_t mode)
794{
795 struct table_device *td;
796
797 list_for_each_entry(td, l, list)
798 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
799 return td;
800
801 return NULL;
802}
803
804int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
805 struct dm_dev **result)
806{
807 struct table_device *td;
808
809 mutex_lock(&md->table_devices_lock);
810 td = find_table_device(&md->table_devices, dev, mode);
811 if (!td) {
812 td = open_table_device(md, dev, mode);
813 if (IS_ERR(td)) {
814 mutex_unlock(&md->table_devices_lock);
815 return PTR_ERR(td);
816 }
817 } else {
818 refcount_inc(&td->count);
819 }
820 mutex_unlock(&md->table_devices_lock);
821
822 *result = &td->dm_dev;
823 return 0;
824}
825
826void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
827{
828 struct table_device *td = container_of(d, struct table_device, dm_dev);
829
830 mutex_lock(&md->table_devices_lock);
831 if (refcount_dec_and_test(&td->count))
832 close_table_device(td, md);
833 mutex_unlock(&md->table_devices_lock);
834}
835
836/*
837 * Get the geometry associated with a dm device
838 */
839int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
840{
841 *geo = md->geometry;
842
843 return 0;
844}
845
846/*
847 * Set the geometry of a device.
848 */
849int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
850{
851 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
852
853 if (geo->start > sz) {
854 DMERR("Start sector is beyond the geometry limits.");
855 return -EINVAL;
856 }
857
858 md->geometry = *geo;
859
860 return 0;
861}
862
863static int __noflush_suspending(struct mapped_device *md)
864{
865 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
866}
867
868static void dm_requeue_add_io(struct dm_io *io, bool first_stage)
869{
870 struct mapped_device *md = io->md;
871
872 if (first_stage) {
873 struct dm_io *next = md->requeue_list;
874
875 md->requeue_list = io;
876 io->next = next;
877 } else {
878 bio_list_add_head(&md->deferred, io->orig_bio);
879 }
880}
881
882static void dm_kick_requeue(struct mapped_device *md, bool first_stage)
883{
884 if (first_stage)
885 queue_work(md->wq, &md->requeue_work);
886 else
887 queue_work(md->wq, &md->work);
888}
889
890/*
891 * Return true if the dm_io's original bio is requeued.
892 * io->status is updated with error if requeue disallowed.
893 */
894static bool dm_handle_requeue(struct dm_io *io, bool first_stage)
895{
896 struct bio *bio = io->orig_bio;
897 bool handle_requeue = (io->status == BLK_STS_DM_REQUEUE);
898 bool handle_polled_eagain = ((io->status == BLK_STS_AGAIN) &&
899 (bio->bi_opf & REQ_POLLED));
900 struct mapped_device *md = io->md;
901 bool requeued = false;
902
903 if (handle_requeue || handle_polled_eagain) {
904 unsigned long flags;
905
906 if (bio->bi_opf & REQ_POLLED) {
907 /*
908 * Upper layer won't help us poll split bio
909 * (io->orig_bio may only reflect a subset of the
910 * pre-split original) so clear REQ_POLLED.
911 */
912 bio_clear_polled(bio);
913 }
914
915 /*
916 * Target requested pushing back the I/O or
917 * polled IO hit BLK_STS_AGAIN.
918 */
919 spin_lock_irqsave(&md->deferred_lock, flags);
920 if ((__noflush_suspending(md) &&
921 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) ||
922 handle_polled_eagain || first_stage) {
923 dm_requeue_add_io(io, first_stage);
924 requeued = true;
925 } else {
926 /*
927 * noflush suspend was interrupted or this is
928 * a write to a zoned target.
929 */
930 io->status = BLK_STS_IOERR;
931 }
932 spin_unlock_irqrestore(&md->deferred_lock, flags);
933 }
934
935 if (requeued)
936 dm_kick_requeue(md, first_stage);
937
938 return requeued;
939}
940
941static void __dm_io_complete(struct dm_io *io, bool first_stage)
942{
943 struct bio *bio = io->orig_bio;
944 struct mapped_device *md = io->md;
945 blk_status_t io_error;
946 bool requeued;
947
948 requeued = dm_handle_requeue(io, first_stage);
949 if (requeued && first_stage)
950 return;
951
952 io_error = io->status;
953 if (dm_io_flagged(io, DM_IO_ACCOUNTED))
954 dm_end_io_acct(io);
955 else if (!io_error) {
956 /*
957 * Must handle target that DM_MAPIO_SUBMITTED only to
958 * then bio_endio() rather than dm_submit_bio_remap()
959 */
960 __dm_start_io_acct(io);
961 dm_end_io_acct(io);
962 }
963 free_io(io);
964 smp_wmb();
965 this_cpu_dec(*md->pending_io);
966
967 /* nudge anyone waiting on suspend queue */
968 if (unlikely(wq_has_sleeper(&md->wait)))
969 wake_up(&md->wait);
970
971 /* Return early if the original bio was requeued */
972 if (requeued)
973 return;
974
975 if (bio_is_flush_with_data(bio)) {
976 /*
977 * Preflush done for flush with data, reissue
978 * without REQ_PREFLUSH.
979 */
980 bio->bi_opf &= ~REQ_PREFLUSH;
981 queue_io(md, bio);
982 } else {
983 /* done with normal IO or empty flush */
984 if (io_error)
985 bio->bi_status = io_error;
986 bio_endio(bio);
987 }
988}
989
990static void dm_wq_requeue_work(struct work_struct *work)
991{
992 struct mapped_device *md = container_of(work, struct mapped_device,
993 requeue_work);
994 unsigned long flags;
995 struct dm_io *io;
996
997 /* reuse deferred lock to simplify dm_handle_requeue */
998 spin_lock_irqsave(&md->deferred_lock, flags);
999 io = md->requeue_list;
1000 md->requeue_list = NULL;
1001 spin_unlock_irqrestore(&md->deferred_lock, flags);
1002
1003 while (io) {
1004 struct dm_io *next = io->next;
1005
1006 dm_io_rewind(io, &md->disk->bio_split);
1007
1008 io->next = NULL;
1009 __dm_io_complete(io, false);
1010 io = next;
1011 }
1012}
1013
1014/*
1015 * Two staged requeue:
1016 *
1017 * 1) io->orig_bio points to the real original bio, and the part mapped to
1018 * this io must be requeued, instead of other parts of the original bio.
1019 *
1020 * 2) io->orig_bio points to new cloned bio which matches the requeued dm_io.
1021 */
1022static void dm_io_complete(struct dm_io *io)
1023{
1024 bool first_requeue;
1025
1026 /*
1027 * Only dm_io that has been split needs two stage requeue, otherwise
1028 * we may run into long bio clone chain during suspend and OOM could
1029 * be triggered.
1030 *
1031 * Also flush data dm_io won't be marked as DM_IO_WAS_SPLIT, so they
1032 * also aren't handled via the first stage requeue.
1033 */
1034 if (dm_io_flagged(io, DM_IO_WAS_SPLIT))
1035 first_requeue = true;
1036 else
1037 first_requeue = false;
1038
1039 __dm_io_complete(io, first_requeue);
1040}
1041
1042/*
1043 * Decrements the number of outstanding ios that a bio has been
1044 * cloned into, completing the original io if necc.
1045 */
1046static inline void __dm_io_dec_pending(struct dm_io *io)
1047{
1048 if (atomic_dec_and_test(&io->io_count))
1049 dm_io_complete(io);
1050}
1051
1052static void dm_io_set_error(struct dm_io *io, blk_status_t error)
1053{
1054 unsigned long flags;
1055
1056 /* Push-back supersedes any I/O errors */
1057 spin_lock_irqsave(&io->lock, flags);
1058 if (!(io->status == BLK_STS_DM_REQUEUE &&
1059 __noflush_suspending(io->md))) {
1060 io->status = error;
1061 }
1062 spin_unlock_irqrestore(&io->lock, flags);
1063}
1064
1065static void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
1066{
1067 if (unlikely(error))
1068 dm_io_set_error(io, error);
1069
1070 __dm_io_dec_pending(io);
1071}
1072
1073void disable_discard(struct mapped_device *md)
1074{
1075 struct queue_limits *limits = dm_get_queue_limits(md);
1076
1077 /* device doesn't really support DISCARD, disable it */
1078 limits->max_discard_sectors = 0;
1079}
1080
1081void disable_write_zeroes(struct mapped_device *md)
1082{
1083 struct queue_limits *limits = dm_get_queue_limits(md);
1084
1085 /* device doesn't really support WRITE ZEROES, disable it */
1086 limits->max_write_zeroes_sectors = 0;
1087}
1088
1089static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
1090{
1091 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
1092}
1093
1094static void clone_endio(struct bio *bio)
1095{
1096 blk_status_t error = bio->bi_status;
1097 struct dm_target_io *tio = clone_to_tio(bio);
1098 struct dm_target *ti = tio->ti;
1099 dm_endio_fn endio = ti->type->end_io;
1100 struct dm_io *io = tio->io;
1101 struct mapped_device *md = io->md;
1102
1103 if (unlikely(error == BLK_STS_TARGET)) {
1104 if (bio_op(bio) == REQ_OP_DISCARD &&
1105 !bdev_max_discard_sectors(bio->bi_bdev))
1106 disable_discard(md);
1107 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
1108 !bdev_write_zeroes_sectors(bio->bi_bdev))
1109 disable_write_zeroes(md);
1110 }
1111
1112 if (static_branch_unlikely(&zoned_enabled) &&
1113 unlikely(bdev_is_zoned(bio->bi_bdev)))
1114 dm_zone_endio(io, bio);
1115
1116 if (endio) {
1117 int r = endio(ti, bio, &error);
1118 switch (r) {
1119 case DM_ENDIO_REQUEUE:
1120 if (static_branch_unlikely(&zoned_enabled)) {
1121 /*
1122 * Requeuing writes to a sequential zone of a zoned
1123 * target will break the sequential write pattern:
1124 * fail such IO.
1125 */
1126 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
1127 error = BLK_STS_IOERR;
1128 else
1129 error = BLK_STS_DM_REQUEUE;
1130 } else
1131 error = BLK_STS_DM_REQUEUE;
1132 fallthrough;
1133 case DM_ENDIO_DONE:
1134 break;
1135 case DM_ENDIO_INCOMPLETE:
1136 /* The target will handle the io */
1137 return;
1138 default:
1139 DMCRIT("unimplemented target endio return value: %d", r);
1140 BUG();
1141 }
1142 }
1143
1144 if (static_branch_unlikely(&swap_bios_enabled) &&
1145 unlikely(swap_bios_limit(ti, bio)))
1146 up(&md->swap_bios_semaphore);
1147
1148 free_tio(bio);
1149 dm_io_dec_pending(io, error);
1150}
1151
1152/*
1153 * Return maximum size of I/O possible at the supplied sector up to the current
1154 * target boundary.
1155 */
1156static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
1157 sector_t target_offset)
1158{
1159 return ti->len - target_offset;
1160}
1161
1162static sector_t max_io_len(struct dm_target *ti, sector_t sector)
1163{
1164 sector_t target_offset = dm_target_offset(ti, sector);
1165 sector_t len = max_io_len_target_boundary(ti, target_offset);
1166
1167 /*
1168 * Does the target need to split IO even further?
1169 * - varied (per target) IO splitting is a tenet of DM; this
1170 * explains why stacked chunk_sectors based splitting via
1171 * bio_split_to_limits() isn't possible here.
1172 */
1173 if (!ti->max_io_len)
1174 return len;
1175 return min_t(sector_t, len,
1176 min(queue_max_sectors(ti->table->md->queue),
1177 blk_chunk_sectors_left(target_offset, ti->max_io_len)));
1178}
1179
1180int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1181{
1182 if (len > UINT_MAX) {
1183 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1184 (unsigned long long)len, UINT_MAX);
1185 ti->error = "Maximum size of target IO is too large";
1186 return -EINVAL;
1187 }
1188
1189 ti->max_io_len = (uint32_t) len;
1190
1191 return 0;
1192}
1193EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1194
1195static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
1196 sector_t sector, int *srcu_idx)
1197 __acquires(md->io_barrier)
1198{
1199 struct dm_table *map;
1200 struct dm_target *ti;
1201
1202 map = dm_get_live_table(md, srcu_idx);
1203 if (!map)
1204 return NULL;
1205
1206 ti = dm_table_find_target(map, sector);
1207 if (!ti)
1208 return NULL;
1209
1210 return ti;
1211}
1212
1213static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1214 long nr_pages, enum dax_access_mode mode, void **kaddr,
1215 pfn_t *pfn)
1216{
1217 struct mapped_device *md = dax_get_private(dax_dev);
1218 sector_t sector = pgoff * PAGE_SECTORS;
1219 struct dm_target *ti;
1220 long len, ret = -EIO;
1221 int srcu_idx;
1222
1223 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1224
1225 if (!ti)
1226 goto out;
1227 if (!ti->type->direct_access)
1228 goto out;
1229 len = max_io_len(ti, sector) / PAGE_SECTORS;
1230 if (len < 1)
1231 goto out;
1232 nr_pages = min(len, nr_pages);
1233 ret = ti->type->direct_access(ti, pgoff, nr_pages, mode, kaddr, pfn);
1234
1235 out:
1236 dm_put_live_table(md, srcu_idx);
1237
1238 return ret;
1239}
1240
1241static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1242 size_t nr_pages)
1243{
1244 struct mapped_device *md = dax_get_private(dax_dev);
1245 sector_t sector = pgoff * PAGE_SECTORS;
1246 struct dm_target *ti;
1247 int ret = -EIO;
1248 int srcu_idx;
1249
1250 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1251
1252 if (!ti)
1253 goto out;
1254 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1255 /*
1256 * ->zero_page_range() is mandatory dax operation. If we are
1257 * here, something is wrong.
1258 */
1259 goto out;
1260 }
1261 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1262 out:
1263 dm_put_live_table(md, srcu_idx);
1264
1265 return ret;
1266}
1267
1268static size_t dm_dax_recovery_write(struct dax_device *dax_dev, pgoff_t pgoff,
1269 void *addr, size_t bytes, struct iov_iter *i)
1270{
1271 struct mapped_device *md = dax_get_private(dax_dev);
1272 sector_t sector = pgoff * PAGE_SECTORS;
1273 struct dm_target *ti;
1274 int srcu_idx;
1275 long ret = 0;
1276
1277 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1278 if (!ti || !ti->type->dax_recovery_write)
1279 goto out;
1280
1281 ret = ti->type->dax_recovery_write(ti, pgoff, addr, bytes, i);
1282out:
1283 dm_put_live_table(md, srcu_idx);
1284 return ret;
1285}
1286
1287/*
1288 * A target may call dm_accept_partial_bio only from the map routine. It is
1289 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1290 * operations, REQ_OP_ZONE_APPEND (zone append writes) and any bio serviced by
1291 * __send_duplicate_bios().
1292 *
1293 * dm_accept_partial_bio informs the dm that the target only wants to process
1294 * additional n_sectors sectors of the bio and the rest of the data should be
1295 * sent in a next bio.
1296 *
1297 * A diagram that explains the arithmetics:
1298 * +--------------------+---------------+-------+
1299 * | 1 | 2 | 3 |
1300 * +--------------------+---------------+-------+
1301 *
1302 * <-------------- *tio->len_ptr --------------->
1303 * <----- bio_sectors ----->
1304 * <-- n_sectors -->
1305 *
1306 * Region 1 was already iterated over with bio_advance or similar function.
1307 * (it may be empty if the target doesn't use bio_advance)
1308 * Region 2 is the remaining bio size that the target wants to process.
1309 * (it may be empty if region 1 is non-empty, although there is no reason
1310 * to make it empty)
1311 * The target requires that region 3 is to be sent in the next bio.
1312 *
1313 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1314 * the partially processed part (the sum of regions 1+2) must be the same for all
1315 * copies of the bio.
1316 */
1317void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1318{
1319 struct dm_target_io *tio = clone_to_tio(bio);
1320 struct dm_io *io = tio->io;
1321 unsigned bio_sectors = bio_sectors(bio);
1322
1323 BUG_ON(dm_tio_flagged(tio, DM_TIO_IS_DUPLICATE_BIO));
1324 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1325 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1326 BUG_ON(bio_sectors > *tio->len_ptr);
1327 BUG_ON(n_sectors > bio_sectors);
1328
1329 *tio->len_ptr -= bio_sectors - n_sectors;
1330 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1331
1332 /*
1333 * __split_and_process_bio() may have already saved mapped part
1334 * for accounting but it is being reduced so update accordingly.
1335 */
1336 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1337 io->sectors = n_sectors;
1338 io->sector_offset = bio_sectors(io->orig_bio);
1339}
1340EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1341
1342/*
1343 * @clone: clone bio that DM core passed to target's .map function
1344 * @tgt_clone: clone of @clone bio that target needs submitted
1345 *
1346 * Targets should use this interface to submit bios they take
1347 * ownership of when returning DM_MAPIO_SUBMITTED.
1348 *
1349 * Target should also enable ti->accounts_remapped_io
1350 */
1351void dm_submit_bio_remap(struct bio *clone, struct bio *tgt_clone)
1352{
1353 struct dm_target_io *tio = clone_to_tio(clone);
1354 struct dm_io *io = tio->io;
1355
1356 /* establish bio that will get submitted */
1357 if (!tgt_clone)
1358 tgt_clone = clone;
1359
1360 /*
1361 * Account io->origin_bio to DM dev on behalf of target
1362 * that took ownership of IO with DM_MAPIO_SUBMITTED.
1363 */
1364 dm_start_io_acct(io, clone);
1365
1366 trace_block_bio_remap(tgt_clone, disk_devt(io->md->disk),
1367 tio->old_sector);
1368 submit_bio_noacct(tgt_clone);
1369}
1370EXPORT_SYMBOL_GPL(dm_submit_bio_remap);
1371
1372static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1373{
1374 mutex_lock(&md->swap_bios_lock);
1375 while (latch < md->swap_bios) {
1376 cond_resched();
1377 down(&md->swap_bios_semaphore);
1378 md->swap_bios--;
1379 }
1380 while (latch > md->swap_bios) {
1381 cond_resched();
1382 up(&md->swap_bios_semaphore);
1383 md->swap_bios++;
1384 }
1385 mutex_unlock(&md->swap_bios_lock);
1386}
1387
1388static void __map_bio(struct bio *clone)
1389{
1390 struct dm_target_io *tio = clone_to_tio(clone);
1391 struct dm_target *ti = tio->ti;
1392 struct dm_io *io = tio->io;
1393 struct mapped_device *md = io->md;
1394 int r;
1395
1396 clone->bi_end_io = clone_endio;
1397
1398 /*
1399 * Map the clone.
1400 */
1401 tio->old_sector = clone->bi_iter.bi_sector;
1402
1403 if (static_branch_unlikely(&swap_bios_enabled) &&
1404 unlikely(swap_bios_limit(ti, clone))) {
1405 int latch = get_swap_bios();
1406 if (unlikely(latch != md->swap_bios))
1407 __set_swap_bios_limit(md, latch);
1408 down(&md->swap_bios_semaphore);
1409 }
1410
1411 if (static_branch_unlikely(&zoned_enabled)) {
1412 /*
1413 * Check if the IO needs a special mapping due to zone append
1414 * emulation on zoned target. In this case, dm_zone_map_bio()
1415 * calls the target map operation.
1416 */
1417 if (unlikely(dm_emulate_zone_append(md)))
1418 r = dm_zone_map_bio(tio);
1419 else
1420 r = ti->type->map(ti, clone);
1421 } else
1422 r = ti->type->map(ti, clone);
1423
1424 switch (r) {
1425 case DM_MAPIO_SUBMITTED:
1426 /* target has assumed ownership of this io */
1427 if (!ti->accounts_remapped_io)
1428 dm_start_io_acct(io, clone);
1429 break;
1430 case DM_MAPIO_REMAPPED:
1431 dm_submit_bio_remap(clone, NULL);
1432 break;
1433 case DM_MAPIO_KILL:
1434 case DM_MAPIO_REQUEUE:
1435 if (static_branch_unlikely(&swap_bios_enabled) &&
1436 unlikely(swap_bios_limit(ti, clone)))
1437 up(&md->swap_bios_semaphore);
1438 free_tio(clone);
1439 if (r == DM_MAPIO_KILL)
1440 dm_io_dec_pending(io, BLK_STS_IOERR);
1441 else
1442 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1443 break;
1444 default:
1445 DMCRIT("unimplemented target map return value: %d", r);
1446 BUG();
1447 }
1448}
1449
1450static void setup_split_accounting(struct clone_info *ci, unsigned len)
1451{
1452 struct dm_io *io = ci->io;
1453
1454 if (ci->sector_count > len) {
1455 /*
1456 * Split needed, save the mapped part for accounting.
1457 * NOTE: dm_accept_partial_bio() will update accordingly.
1458 */
1459 dm_io_set_flag(io, DM_IO_WAS_SPLIT);
1460 io->sectors = len;
1461 io->sector_offset = bio_sectors(ci->bio);
1462 }
1463}
1464
1465static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1466 struct dm_target *ti, unsigned num_bios)
1467{
1468 struct bio *bio;
1469 int try;
1470
1471 for (try = 0; try < 2; try++) {
1472 int bio_nr;
1473
1474 if (try)
1475 mutex_lock(&ci->io->md->table_devices_lock);
1476 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1477 bio = alloc_tio(ci, ti, bio_nr, NULL,
1478 try ? GFP_NOIO : GFP_NOWAIT);
1479 if (!bio)
1480 break;
1481
1482 bio_list_add(blist, bio);
1483 }
1484 if (try)
1485 mutex_unlock(&ci->io->md->table_devices_lock);
1486 if (bio_nr == num_bios)
1487 return;
1488
1489 while ((bio = bio_list_pop(blist)))
1490 free_tio(bio);
1491 }
1492}
1493
1494static int __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1495 unsigned int num_bios, unsigned *len)
1496{
1497 struct bio_list blist = BIO_EMPTY_LIST;
1498 struct bio *clone;
1499 unsigned int ret = 0;
1500
1501 switch (num_bios) {
1502 case 0:
1503 break;
1504 case 1:
1505 if (len)
1506 setup_split_accounting(ci, *len);
1507 clone = alloc_tio(ci, ti, 0, len, GFP_NOIO);
1508 __map_bio(clone);
1509 ret = 1;
1510 break;
1511 default:
1512 /* dm_accept_partial_bio() is not supported with shared tio->len_ptr */
1513 alloc_multiple_bios(&blist, ci, ti, num_bios);
1514 while ((clone = bio_list_pop(&blist))) {
1515 dm_tio_set_flag(clone_to_tio(clone), DM_TIO_IS_DUPLICATE_BIO);
1516 __map_bio(clone);
1517 ret += 1;
1518 }
1519 break;
1520 }
1521
1522 return ret;
1523}
1524
1525static void __send_empty_flush(struct clone_info *ci)
1526{
1527 struct dm_table *t = ci->map;
1528 struct bio flush_bio;
1529
1530 /*
1531 * Use an on-stack bio for this, it's safe since we don't
1532 * need to reference it after submit. It's just used as
1533 * the basis for the clone(s).
1534 */
1535 bio_init(&flush_bio, ci->io->md->disk->part0, NULL, 0,
1536 REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC);
1537
1538 ci->bio = &flush_bio;
1539 ci->sector_count = 0;
1540 ci->io->tio.clone.bi_iter.bi_size = 0;
1541
1542 for (unsigned int i = 0; i < t->num_targets; i++) {
1543 unsigned int bios;
1544 struct dm_target *ti = dm_table_get_target(t, i);
1545
1546 atomic_add(ti->num_flush_bios, &ci->io->io_count);
1547 bios = __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1548 atomic_sub(ti->num_flush_bios - bios, &ci->io->io_count);
1549 }
1550
1551 /*
1552 * alloc_io() takes one extra reference for submission, so the
1553 * reference won't reach 0 without the following subtraction
1554 */
1555 atomic_sub(1, &ci->io->io_count);
1556
1557 bio_uninit(ci->bio);
1558}
1559
1560static void __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1561 unsigned num_bios)
1562{
1563 unsigned len;
1564 unsigned int bios;
1565
1566 len = min_t(sector_t, ci->sector_count,
1567 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1568
1569 atomic_add(num_bios, &ci->io->io_count);
1570 bios = __send_duplicate_bios(ci, ti, num_bios, &len);
1571 /*
1572 * alloc_io() takes one extra reference for submission, so the
1573 * reference won't reach 0 without the following (+1) subtraction
1574 */
1575 atomic_sub(num_bios - bios + 1, &ci->io->io_count);
1576
1577 ci->sector += len;
1578 ci->sector_count -= len;
1579}
1580
1581static bool is_abnormal_io(struct bio *bio)
1582{
1583 enum req_op op = bio_op(bio);
1584
1585 if (op != REQ_OP_READ && op != REQ_OP_WRITE && op != REQ_OP_FLUSH) {
1586 switch (op) {
1587 case REQ_OP_DISCARD:
1588 case REQ_OP_SECURE_ERASE:
1589 case REQ_OP_WRITE_ZEROES:
1590 return true;
1591 default:
1592 break;
1593 }
1594 }
1595
1596 return false;
1597}
1598
1599static blk_status_t __process_abnormal_io(struct clone_info *ci,
1600 struct dm_target *ti)
1601{
1602 unsigned num_bios = 0;
1603
1604 switch (bio_op(ci->bio)) {
1605 case REQ_OP_DISCARD:
1606 num_bios = ti->num_discard_bios;
1607 break;
1608 case REQ_OP_SECURE_ERASE:
1609 num_bios = ti->num_secure_erase_bios;
1610 break;
1611 case REQ_OP_WRITE_ZEROES:
1612 num_bios = ti->num_write_zeroes_bios;
1613 break;
1614 default:
1615 break;
1616 }
1617
1618 /*
1619 * Even though the device advertised support for this type of
1620 * request, that does not mean every target supports it, and
1621 * reconfiguration might also have changed that since the
1622 * check was performed.
1623 */
1624 if (unlikely(!num_bios))
1625 return BLK_STS_NOTSUPP;
1626
1627 __send_changing_extent_only(ci, ti, num_bios);
1628 return BLK_STS_OK;
1629}
1630
1631/*
1632 * Reuse ->bi_private as dm_io list head for storing all dm_io instances
1633 * associated with this bio, and this bio's bi_private needs to be
1634 * stored in dm_io->data before the reuse.
1635 *
1636 * bio->bi_private is owned by fs or upper layer, so block layer won't
1637 * touch it after splitting. Meantime it won't be changed by anyone after
1638 * bio is submitted. So this reuse is safe.
1639 */
1640static inline struct dm_io **dm_poll_list_head(struct bio *bio)
1641{
1642 return (struct dm_io **)&bio->bi_private;
1643}
1644
1645static void dm_queue_poll_io(struct bio *bio, struct dm_io *io)
1646{
1647 struct dm_io **head = dm_poll_list_head(bio);
1648
1649 if (!(bio->bi_opf & REQ_DM_POLL_LIST)) {
1650 bio->bi_opf |= REQ_DM_POLL_LIST;
1651 /*
1652 * Save .bi_private into dm_io, so that we can reuse
1653 * .bi_private as dm_io list head for storing dm_io list
1654 */
1655 io->data = bio->bi_private;
1656
1657 /* tell block layer to poll for completion */
1658 bio->bi_cookie = ~BLK_QC_T_NONE;
1659
1660 io->next = NULL;
1661 } else {
1662 /*
1663 * bio recursed due to split, reuse original poll list,
1664 * and save bio->bi_private too.
1665 */
1666 io->data = (*head)->data;
1667 io->next = *head;
1668 }
1669
1670 *head = io;
1671}
1672
1673/*
1674 * Select the correct strategy for processing a non-flush bio.
1675 */
1676static blk_status_t __split_and_process_bio(struct clone_info *ci)
1677{
1678 struct bio *clone;
1679 struct dm_target *ti;
1680 unsigned len;
1681
1682 ti = dm_table_find_target(ci->map, ci->sector);
1683 if (unlikely(!ti))
1684 return BLK_STS_IOERR;
1685
1686 if (unlikely((ci->bio->bi_opf & REQ_NOWAIT) != 0) &&
1687 unlikely(!dm_target_supports_nowait(ti->type)))
1688 return BLK_STS_NOTSUPP;
1689
1690 if (unlikely(ci->is_abnormal_io))
1691 return __process_abnormal_io(ci, ti);
1692
1693 /*
1694 * Only support bio polling for normal IO, and the target io is
1695 * exactly inside the dm_io instance (verified in dm_poll_dm_io)
1696 */
1697 ci->submit_as_polled = !!(ci->bio->bi_opf & REQ_POLLED);
1698
1699 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1700 setup_split_accounting(ci, len);
1701 clone = alloc_tio(ci, ti, 0, &len, GFP_NOIO);
1702 __map_bio(clone);
1703
1704 ci->sector += len;
1705 ci->sector_count -= len;
1706
1707 return BLK_STS_OK;
1708}
1709
1710static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1711 struct dm_table *map, struct bio *bio, bool is_abnormal)
1712{
1713 ci->map = map;
1714 ci->io = alloc_io(md, bio);
1715 ci->bio = bio;
1716 ci->is_abnormal_io = is_abnormal;
1717 ci->submit_as_polled = false;
1718 ci->sector = bio->bi_iter.bi_sector;
1719 ci->sector_count = bio_sectors(bio);
1720
1721 /* Shouldn't happen but sector_count was being set to 0 so... */
1722 if (static_branch_unlikely(&zoned_enabled) &&
1723 WARN_ON_ONCE(op_is_zone_mgmt(bio_op(bio)) && ci->sector_count))
1724 ci->sector_count = 0;
1725}
1726
1727/*
1728 * Entry point to split a bio into clones and submit them to the targets.
1729 */
1730static void dm_split_and_process_bio(struct mapped_device *md,
1731 struct dm_table *map, struct bio *bio)
1732{
1733 struct clone_info ci;
1734 struct dm_io *io;
1735 blk_status_t error = BLK_STS_OK;
1736 bool is_abnormal;
1737
1738 is_abnormal = is_abnormal_io(bio);
1739 if (unlikely(is_abnormal)) {
1740 /*
1741 * Use bio_split_to_limits() for abnormal IO (e.g. discard, etc)
1742 * otherwise associated queue_limits won't be imposed.
1743 */
1744 bio = bio_split_to_limits(bio);
1745 if (!bio)
1746 return;
1747 }
1748
1749 init_clone_info(&ci, md, map, bio, is_abnormal);
1750 io = ci.io;
1751
1752 if (bio->bi_opf & REQ_PREFLUSH) {
1753 __send_empty_flush(&ci);
1754 /* dm_io_complete submits any data associated with flush */
1755 goto out;
1756 }
1757
1758 error = __split_and_process_bio(&ci);
1759 if (error || !ci.sector_count)
1760 goto out;
1761 /*
1762 * Remainder must be passed to submit_bio_noacct() so it gets handled
1763 * *after* bios already submitted have been completely processed.
1764 */
1765 bio_trim(bio, io->sectors, ci.sector_count);
1766 trace_block_split(bio, bio->bi_iter.bi_sector);
1767 bio_inc_remaining(bio);
1768 submit_bio_noacct(bio);
1769out:
1770 /*
1771 * Drop the extra reference count for non-POLLED bio, and hold one
1772 * reference for POLLED bio, which will be released in dm_poll_bio
1773 *
1774 * Add every dm_io instance into the dm_io list head which is stored
1775 * in bio->bi_private, so that dm_poll_bio can poll them all.
1776 */
1777 if (error || !ci.submit_as_polled) {
1778 /*
1779 * In case of submission failure, the extra reference for
1780 * submitting io isn't consumed yet
1781 */
1782 if (error)
1783 atomic_dec(&io->io_count);
1784 dm_io_dec_pending(io, error);
1785 } else
1786 dm_queue_poll_io(bio, io);
1787}
1788
1789static void dm_submit_bio(struct bio *bio)
1790{
1791 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1792 int srcu_idx;
1793 struct dm_table *map;
1794 blk_opf_t bio_opf = bio->bi_opf;
1795
1796 map = dm_get_live_table_bio(md, &srcu_idx, bio_opf);
1797
1798 /* If suspended, or map not yet available, queue this IO for later */
1799 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) ||
1800 unlikely(!map)) {
1801 if (bio->bi_opf & REQ_NOWAIT)
1802 bio_wouldblock_error(bio);
1803 else if (bio->bi_opf & REQ_RAHEAD)
1804 bio_io_error(bio);
1805 else
1806 queue_io(md, bio);
1807 goto out;
1808 }
1809
1810 dm_split_and_process_bio(md, map, bio);
1811out:
1812 dm_put_live_table_bio(md, srcu_idx, bio_opf);
1813}
1814
1815static bool dm_poll_dm_io(struct dm_io *io, struct io_comp_batch *iob,
1816 unsigned int flags)
1817{
1818 WARN_ON_ONCE(!dm_tio_is_normal(&io->tio));
1819
1820 /* don't poll if the mapped io is done */
1821 if (atomic_read(&io->io_count) > 1)
1822 bio_poll(&io->tio.clone, iob, flags);
1823
1824 /* bio_poll holds the last reference */
1825 return atomic_read(&io->io_count) == 1;
1826}
1827
1828static int dm_poll_bio(struct bio *bio, struct io_comp_batch *iob,
1829 unsigned int flags)
1830{
1831 struct dm_io **head = dm_poll_list_head(bio);
1832 struct dm_io *list = *head;
1833 struct dm_io *tmp = NULL;
1834 struct dm_io *curr, *next;
1835
1836 /* Only poll normal bio which was marked as REQ_DM_POLL_LIST */
1837 if (!(bio->bi_opf & REQ_DM_POLL_LIST))
1838 return 0;
1839
1840 WARN_ON_ONCE(!list);
1841
1842 /*
1843 * Restore .bi_private before possibly completing dm_io.
1844 *
1845 * bio_poll() is only possible once @bio has been completely
1846 * submitted via submit_bio_noacct()'s depth-first submission.
1847 * So there is no dm_queue_poll_io() race associated with
1848 * clearing REQ_DM_POLL_LIST here.
1849 */
1850 bio->bi_opf &= ~REQ_DM_POLL_LIST;
1851 bio->bi_private = list->data;
1852
1853 for (curr = list, next = curr->next; curr; curr = next, next =
1854 curr ? curr->next : NULL) {
1855 if (dm_poll_dm_io(curr, iob, flags)) {
1856 /*
1857 * clone_endio() has already occurred, so no
1858 * error handling is needed here.
1859 */
1860 __dm_io_dec_pending(curr);
1861 } else {
1862 curr->next = tmp;
1863 tmp = curr;
1864 }
1865 }
1866
1867 /* Not done? */
1868 if (tmp) {
1869 bio->bi_opf |= REQ_DM_POLL_LIST;
1870 /* Reset bio->bi_private to dm_io list head */
1871 *head = tmp;
1872 return 0;
1873 }
1874 return 1;
1875}
1876
1877/*-----------------------------------------------------------------
1878 * An IDR is used to keep track of allocated minor numbers.
1879 *---------------------------------------------------------------*/
1880static void free_minor(int minor)
1881{
1882 spin_lock(&_minor_lock);
1883 idr_remove(&_minor_idr, minor);
1884 spin_unlock(&_minor_lock);
1885}
1886
1887/*
1888 * See if the device with a specific minor # is free.
1889 */
1890static int specific_minor(int minor)
1891{
1892 int r;
1893
1894 if (minor >= (1 << MINORBITS))
1895 return -EINVAL;
1896
1897 idr_preload(GFP_KERNEL);
1898 spin_lock(&_minor_lock);
1899
1900 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1901
1902 spin_unlock(&_minor_lock);
1903 idr_preload_end();
1904 if (r < 0)
1905 return r == -ENOSPC ? -EBUSY : r;
1906 return 0;
1907}
1908
1909static int next_free_minor(int *minor)
1910{
1911 int r;
1912
1913 idr_preload(GFP_KERNEL);
1914 spin_lock(&_minor_lock);
1915
1916 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1917
1918 spin_unlock(&_minor_lock);
1919 idr_preload_end();
1920 if (r < 0)
1921 return r;
1922 *minor = r;
1923 return 0;
1924}
1925
1926static const struct block_device_operations dm_blk_dops;
1927static const struct block_device_operations dm_rq_blk_dops;
1928static const struct dax_operations dm_dax_ops;
1929
1930static void dm_wq_work(struct work_struct *work);
1931
1932#ifdef CONFIG_BLK_INLINE_ENCRYPTION
1933static void dm_queue_destroy_crypto_profile(struct request_queue *q)
1934{
1935 dm_destroy_crypto_profile(q->crypto_profile);
1936}
1937
1938#else /* CONFIG_BLK_INLINE_ENCRYPTION */
1939
1940static inline void dm_queue_destroy_crypto_profile(struct request_queue *q)
1941{
1942}
1943#endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1944
1945static void cleanup_mapped_device(struct mapped_device *md)
1946{
1947 if (md->wq)
1948 destroy_workqueue(md->wq);
1949 dm_free_md_mempools(md->mempools);
1950
1951 if (md->dax_dev) {
1952 dax_remove_host(md->disk);
1953 kill_dax(md->dax_dev);
1954 put_dax(md->dax_dev);
1955 md->dax_dev = NULL;
1956 }
1957
1958 dm_cleanup_zoned_dev(md);
1959 if (md->disk) {
1960 spin_lock(&_minor_lock);
1961 md->disk->private_data = NULL;
1962 spin_unlock(&_minor_lock);
1963 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1964 struct table_device *td;
1965
1966 dm_sysfs_exit(md);
1967 list_for_each_entry(td, &md->table_devices, list) {
1968 bd_unlink_disk_holder(td->dm_dev.bdev,
1969 md->disk);
1970 }
1971
1972 /*
1973 * Hold lock to make sure del_gendisk() won't concurrent
1974 * with open/close_table_device().
1975 */
1976 mutex_lock(&md->table_devices_lock);
1977 del_gendisk(md->disk);
1978 mutex_unlock(&md->table_devices_lock);
1979 }
1980 dm_queue_destroy_crypto_profile(md->queue);
1981 put_disk(md->disk);
1982 }
1983
1984 if (md->pending_io) {
1985 free_percpu(md->pending_io);
1986 md->pending_io = NULL;
1987 }
1988
1989 cleanup_srcu_struct(&md->io_barrier);
1990
1991 mutex_destroy(&md->suspend_lock);
1992 mutex_destroy(&md->type_lock);
1993 mutex_destroy(&md->table_devices_lock);
1994 mutex_destroy(&md->swap_bios_lock);
1995
1996 dm_mq_cleanup_mapped_device(md);
1997}
1998
1999/*
2000 * Allocate and initialise a blank device with a given minor.
2001 */
2002static struct mapped_device *alloc_dev(int minor)
2003{
2004 int r, numa_node_id = dm_get_numa_node();
2005 struct mapped_device *md;
2006 void *old_md;
2007
2008 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
2009 if (!md) {
2010 DMERR("unable to allocate device, out of memory.");
2011 return NULL;
2012 }
2013
2014 if (!try_module_get(THIS_MODULE))
2015 goto bad_module_get;
2016
2017 /* get a minor number for the dev */
2018 if (minor == DM_ANY_MINOR)
2019 r = next_free_minor(&minor);
2020 else
2021 r = specific_minor(minor);
2022 if (r < 0)
2023 goto bad_minor;
2024
2025 r = init_srcu_struct(&md->io_barrier);
2026 if (r < 0)
2027 goto bad_io_barrier;
2028
2029 md->numa_node_id = numa_node_id;
2030 md->init_tio_pdu = false;
2031 md->type = DM_TYPE_NONE;
2032 mutex_init(&md->suspend_lock);
2033 mutex_init(&md->type_lock);
2034 mutex_init(&md->table_devices_lock);
2035 spin_lock_init(&md->deferred_lock);
2036 atomic_set(&md->holders, 1);
2037 atomic_set(&md->open_count, 0);
2038 atomic_set(&md->event_nr, 0);
2039 atomic_set(&md->uevent_seq, 0);
2040 INIT_LIST_HEAD(&md->uevent_list);
2041 INIT_LIST_HEAD(&md->table_devices);
2042 spin_lock_init(&md->uevent_lock);
2043
2044 /*
2045 * default to bio-based until DM table is loaded and md->type
2046 * established. If request-based table is loaded: blk-mq will
2047 * override accordingly.
2048 */
2049 md->disk = blk_alloc_disk(md->numa_node_id);
2050 if (!md->disk)
2051 goto bad;
2052 md->queue = md->disk->queue;
2053
2054 init_waitqueue_head(&md->wait);
2055 INIT_WORK(&md->work, dm_wq_work);
2056 INIT_WORK(&md->requeue_work, dm_wq_requeue_work);
2057 init_waitqueue_head(&md->eventq);
2058 init_completion(&md->kobj_holder.completion);
2059
2060 md->requeue_list = NULL;
2061 md->swap_bios = get_swap_bios();
2062 sema_init(&md->swap_bios_semaphore, md->swap_bios);
2063 mutex_init(&md->swap_bios_lock);
2064
2065 md->disk->major = _major;
2066 md->disk->first_minor = minor;
2067 md->disk->minors = 1;
2068 md->disk->flags |= GENHD_FL_NO_PART;
2069 md->disk->fops = &dm_blk_dops;
2070 md->disk->private_data = md;
2071 sprintf(md->disk->disk_name, "dm-%d", minor);
2072
2073 if (IS_ENABLED(CONFIG_FS_DAX)) {
2074 md->dax_dev = alloc_dax(md, &dm_dax_ops);
2075 if (IS_ERR(md->dax_dev)) {
2076 md->dax_dev = NULL;
2077 goto bad;
2078 }
2079 set_dax_nocache(md->dax_dev);
2080 set_dax_nomc(md->dax_dev);
2081 if (dax_add_host(md->dax_dev, md->disk))
2082 goto bad;
2083 }
2084
2085 format_dev_t(md->name, MKDEV(_major, minor));
2086
2087 md->wq = alloc_workqueue("kdmflush/%s", WQ_MEM_RECLAIM, 0, md->name);
2088 if (!md->wq)
2089 goto bad;
2090
2091 md->pending_io = alloc_percpu(unsigned long);
2092 if (!md->pending_io)
2093 goto bad;
2094
2095 dm_stats_init(&md->stats);
2096
2097 /* Populate the mapping, nobody knows we exist yet */
2098 spin_lock(&_minor_lock);
2099 old_md = idr_replace(&_minor_idr, md, minor);
2100 spin_unlock(&_minor_lock);
2101
2102 BUG_ON(old_md != MINOR_ALLOCED);
2103
2104 return md;
2105
2106bad:
2107 cleanup_mapped_device(md);
2108bad_io_barrier:
2109 free_minor(minor);
2110bad_minor:
2111 module_put(THIS_MODULE);
2112bad_module_get:
2113 kvfree(md);
2114 return NULL;
2115}
2116
2117static void unlock_fs(struct mapped_device *md);
2118
2119static void free_dev(struct mapped_device *md)
2120{
2121 int minor = MINOR(disk_devt(md->disk));
2122
2123 unlock_fs(md);
2124
2125 cleanup_mapped_device(md);
2126
2127 WARN_ON_ONCE(!list_empty(&md->table_devices));
2128 dm_stats_cleanup(&md->stats);
2129 free_minor(minor);
2130
2131 module_put(THIS_MODULE);
2132 kvfree(md);
2133}
2134
2135/*
2136 * Bind a table to the device.
2137 */
2138static void event_callback(void *context)
2139{
2140 unsigned long flags;
2141 LIST_HEAD(uevents);
2142 struct mapped_device *md = (struct mapped_device *) context;
2143
2144 spin_lock_irqsave(&md->uevent_lock, flags);
2145 list_splice_init(&md->uevent_list, &uevents);
2146 spin_unlock_irqrestore(&md->uevent_lock, flags);
2147
2148 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2149
2150 atomic_inc(&md->event_nr);
2151 wake_up(&md->eventq);
2152 dm_issue_global_event();
2153}
2154
2155/*
2156 * Returns old map, which caller must destroy.
2157 */
2158static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2159 struct queue_limits *limits)
2160{
2161 struct dm_table *old_map;
2162 sector_t size;
2163 int ret;
2164
2165 lockdep_assert_held(&md->suspend_lock);
2166
2167 size = dm_table_get_size(t);
2168
2169 /*
2170 * Wipe any geometry if the size of the table changed.
2171 */
2172 if (size != dm_get_size(md))
2173 memset(&md->geometry, 0, sizeof(md->geometry));
2174
2175 if (!get_capacity(md->disk))
2176 set_capacity(md->disk, size);
2177 else
2178 set_capacity_and_notify(md->disk, size);
2179
2180 dm_table_event_callback(t, event_callback, md);
2181
2182 if (dm_table_request_based(t)) {
2183 /*
2184 * Leverage the fact that request-based DM targets are
2185 * immutable singletons - used to optimize dm_mq_queue_rq.
2186 */
2187 md->immutable_target = dm_table_get_immutable_target(t);
2188
2189 /*
2190 * There is no need to reload with request-based dm because the
2191 * size of front_pad doesn't change.
2192 *
2193 * Note for future: If you are to reload bioset, prep-ed
2194 * requests in the queue may refer to bio from the old bioset,
2195 * so you must walk through the queue to unprep.
2196 */
2197 if (!md->mempools) {
2198 md->mempools = t->mempools;
2199 t->mempools = NULL;
2200 }
2201 } else {
2202 /*
2203 * The md may already have mempools that need changing.
2204 * If so, reload bioset because front_pad may have changed
2205 * because a different table was loaded.
2206 */
2207 dm_free_md_mempools(md->mempools);
2208 md->mempools = t->mempools;
2209 t->mempools = NULL;
2210 }
2211
2212 ret = dm_table_set_restrictions(t, md->queue, limits);
2213 if (ret) {
2214 old_map = ERR_PTR(ret);
2215 goto out;
2216 }
2217
2218 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2219 rcu_assign_pointer(md->map, (void *)t);
2220 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2221
2222 if (old_map)
2223 dm_sync_table(md);
2224out:
2225 return old_map;
2226}
2227
2228/*
2229 * Returns unbound table for the caller to free.
2230 */
2231static struct dm_table *__unbind(struct mapped_device *md)
2232{
2233 struct dm_table *map = rcu_dereference_protected(md->map, 1);
2234
2235 if (!map)
2236 return NULL;
2237
2238 dm_table_event_callback(map, NULL, NULL);
2239 RCU_INIT_POINTER(md->map, NULL);
2240 dm_sync_table(md);
2241
2242 return map;
2243}
2244
2245/*
2246 * Constructor for a new device.
2247 */
2248int dm_create(int minor, struct mapped_device **result)
2249{
2250 struct mapped_device *md;
2251
2252 md = alloc_dev(minor);
2253 if (!md)
2254 return -ENXIO;
2255
2256 dm_ima_reset_data(md);
2257
2258 *result = md;
2259 return 0;
2260}
2261
2262/*
2263 * Functions to manage md->type.
2264 * All are required to hold md->type_lock.
2265 */
2266void dm_lock_md_type(struct mapped_device *md)
2267{
2268 mutex_lock(&md->type_lock);
2269}
2270
2271void dm_unlock_md_type(struct mapped_device *md)
2272{
2273 mutex_unlock(&md->type_lock);
2274}
2275
2276void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2277{
2278 BUG_ON(!mutex_is_locked(&md->type_lock));
2279 md->type = type;
2280}
2281
2282enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2283{
2284 return md->type;
2285}
2286
2287struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2288{
2289 return md->immutable_target_type;
2290}
2291
2292/*
2293 * The queue_limits are only valid as long as you have a reference
2294 * count on 'md'.
2295 */
2296struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2297{
2298 BUG_ON(!atomic_read(&md->holders));
2299 return &md->queue->limits;
2300}
2301EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2302
2303/*
2304 * Setup the DM device's queue based on md's type
2305 */
2306int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2307{
2308 enum dm_queue_mode type = dm_table_get_type(t);
2309 struct queue_limits limits;
2310 struct table_device *td;
2311 int r;
2312
2313 switch (type) {
2314 case DM_TYPE_REQUEST_BASED:
2315 md->disk->fops = &dm_rq_blk_dops;
2316 r = dm_mq_init_request_queue(md, t);
2317 if (r) {
2318 DMERR("Cannot initialize queue for request-based dm mapped device");
2319 return r;
2320 }
2321 break;
2322 case DM_TYPE_BIO_BASED:
2323 case DM_TYPE_DAX_BIO_BASED:
2324 break;
2325 case DM_TYPE_NONE:
2326 WARN_ON_ONCE(true);
2327 break;
2328 }
2329
2330 r = dm_calculate_queue_limits(t, &limits);
2331 if (r) {
2332 DMERR("Cannot calculate initial queue limits");
2333 return r;
2334 }
2335 r = dm_table_set_restrictions(t, md->queue, &limits);
2336 if (r)
2337 return r;
2338
2339 /*
2340 * Hold lock to make sure add_disk() and del_gendisk() won't concurrent
2341 * with open_table_device() and close_table_device().
2342 */
2343 mutex_lock(&md->table_devices_lock);
2344 r = add_disk(md->disk);
2345 mutex_unlock(&md->table_devices_lock);
2346 if (r)
2347 return r;
2348
2349 /*
2350 * Register the holder relationship for devices added before the disk
2351 * was live.
2352 */
2353 list_for_each_entry(td, &md->table_devices, list) {
2354 r = bd_link_disk_holder(td->dm_dev.bdev, md->disk);
2355 if (r)
2356 goto out_undo_holders;
2357 }
2358
2359 r = dm_sysfs_init(md);
2360 if (r)
2361 goto out_undo_holders;
2362
2363 md->type = type;
2364 return 0;
2365
2366out_undo_holders:
2367 list_for_each_entry_continue_reverse(td, &md->table_devices, list)
2368 bd_unlink_disk_holder(td->dm_dev.bdev, md->disk);
2369 mutex_lock(&md->table_devices_lock);
2370 del_gendisk(md->disk);
2371 mutex_unlock(&md->table_devices_lock);
2372 return r;
2373}
2374
2375struct mapped_device *dm_get_md(dev_t dev)
2376{
2377 struct mapped_device *md;
2378 unsigned minor = MINOR(dev);
2379
2380 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2381 return NULL;
2382
2383 spin_lock(&_minor_lock);
2384
2385 md = idr_find(&_minor_idr, minor);
2386 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2387 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2388 md = NULL;
2389 goto out;
2390 }
2391 dm_get(md);
2392out:
2393 spin_unlock(&_minor_lock);
2394
2395 return md;
2396}
2397EXPORT_SYMBOL_GPL(dm_get_md);
2398
2399void *dm_get_mdptr(struct mapped_device *md)
2400{
2401 return md->interface_ptr;
2402}
2403
2404void dm_set_mdptr(struct mapped_device *md, void *ptr)
2405{
2406 md->interface_ptr = ptr;
2407}
2408
2409void dm_get(struct mapped_device *md)
2410{
2411 atomic_inc(&md->holders);
2412 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2413}
2414
2415int dm_hold(struct mapped_device *md)
2416{
2417 spin_lock(&_minor_lock);
2418 if (test_bit(DMF_FREEING, &md->flags)) {
2419 spin_unlock(&_minor_lock);
2420 return -EBUSY;
2421 }
2422 dm_get(md);
2423 spin_unlock(&_minor_lock);
2424 return 0;
2425}
2426EXPORT_SYMBOL_GPL(dm_hold);
2427
2428const char *dm_device_name(struct mapped_device *md)
2429{
2430 return md->name;
2431}
2432EXPORT_SYMBOL_GPL(dm_device_name);
2433
2434static void __dm_destroy(struct mapped_device *md, bool wait)
2435{
2436 struct dm_table *map;
2437 int srcu_idx;
2438
2439 might_sleep();
2440
2441 spin_lock(&_minor_lock);
2442 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2443 set_bit(DMF_FREEING, &md->flags);
2444 spin_unlock(&_minor_lock);
2445
2446 blk_mark_disk_dead(md->disk);
2447
2448 /*
2449 * Take suspend_lock so that presuspend and postsuspend methods
2450 * do not race with internal suspend.
2451 */
2452 mutex_lock(&md->suspend_lock);
2453 map = dm_get_live_table(md, &srcu_idx);
2454 if (!dm_suspended_md(md)) {
2455 dm_table_presuspend_targets(map);
2456 set_bit(DMF_SUSPENDED, &md->flags);
2457 set_bit(DMF_POST_SUSPENDING, &md->flags);
2458 dm_table_postsuspend_targets(map);
2459 }
2460 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2461 dm_put_live_table(md, srcu_idx);
2462 mutex_unlock(&md->suspend_lock);
2463
2464 /*
2465 * Rare, but there may be I/O requests still going to complete,
2466 * for example. Wait for all references to disappear.
2467 * No one should increment the reference count of the mapped_device,
2468 * after the mapped_device state becomes DMF_FREEING.
2469 */
2470 if (wait)
2471 while (atomic_read(&md->holders))
2472 msleep(1);
2473 else if (atomic_read(&md->holders))
2474 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2475 dm_device_name(md), atomic_read(&md->holders));
2476
2477 dm_table_destroy(__unbind(md));
2478 free_dev(md);
2479}
2480
2481void dm_destroy(struct mapped_device *md)
2482{
2483 __dm_destroy(md, true);
2484}
2485
2486void dm_destroy_immediate(struct mapped_device *md)
2487{
2488 __dm_destroy(md, false);
2489}
2490
2491void dm_put(struct mapped_device *md)
2492{
2493 atomic_dec(&md->holders);
2494}
2495EXPORT_SYMBOL_GPL(dm_put);
2496
2497static bool dm_in_flight_bios(struct mapped_device *md)
2498{
2499 int cpu;
2500 unsigned long sum = 0;
2501
2502 for_each_possible_cpu(cpu)
2503 sum += *per_cpu_ptr(md->pending_io, cpu);
2504
2505 return sum != 0;
2506}
2507
2508static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2509{
2510 int r = 0;
2511 DEFINE_WAIT(wait);
2512
2513 while (true) {
2514 prepare_to_wait(&md->wait, &wait, task_state);
2515
2516 if (!dm_in_flight_bios(md))
2517 break;
2518
2519 if (signal_pending_state(task_state, current)) {
2520 r = -EINTR;
2521 break;
2522 }
2523
2524 io_schedule();
2525 }
2526 finish_wait(&md->wait, &wait);
2527
2528 smp_rmb();
2529
2530 return r;
2531}
2532
2533static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2534{
2535 int r = 0;
2536
2537 if (!queue_is_mq(md->queue))
2538 return dm_wait_for_bios_completion(md, task_state);
2539
2540 while (true) {
2541 if (!blk_mq_queue_inflight(md->queue))
2542 break;
2543
2544 if (signal_pending_state(task_state, current)) {
2545 r = -EINTR;
2546 break;
2547 }
2548
2549 msleep(5);
2550 }
2551
2552 return r;
2553}
2554
2555/*
2556 * Process the deferred bios
2557 */
2558static void dm_wq_work(struct work_struct *work)
2559{
2560 struct mapped_device *md = container_of(work, struct mapped_device, work);
2561 struct bio *bio;
2562
2563 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2564 spin_lock_irq(&md->deferred_lock);
2565 bio = bio_list_pop(&md->deferred);
2566 spin_unlock_irq(&md->deferred_lock);
2567
2568 if (!bio)
2569 break;
2570
2571 submit_bio_noacct(bio);
2572 }
2573}
2574
2575static void dm_queue_flush(struct mapped_device *md)
2576{
2577 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2578 smp_mb__after_atomic();
2579 queue_work(md->wq, &md->work);
2580}
2581
2582/*
2583 * Swap in a new table, returning the old one for the caller to destroy.
2584 */
2585struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2586{
2587 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2588 struct queue_limits limits;
2589 int r;
2590
2591 mutex_lock(&md->suspend_lock);
2592
2593 /* device must be suspended */
2594 if (!dm_suspended_md(md))
2595 goto out;
2596
2597 /*
2598 * If the new table has no data devices, retain the existing limits.
2599 * This helps multipath with queue_if_no_path if all paths disappear,
2600 * then new I/O is queued based on these limits, and then some paths
2601 * reappear.
2602 */
2603 if (dm_table_has_no_data_devices(table)) {
2604 live_map = dm_get_live_table_fast(md);
2605 if (live_map)
2606 limits = md->queue->limits;
2607 dm_put_live_table_fast(md);
2608 }
2609
2610 if (!live_map) {
2611 r = dm_calculate_queue_limits(table, &limits);
2612 if (r) {
2613 map = ERR_PTR(r);
2614 goto out;
2615 }
2616 }
2617
2618 map = __bind(md, table, &limits);
2619 dm_issue_global_event();
2620
2621out:
2622 mutex_unlock(&md->suspend_lock);
2623 return map;
2624}
2625
2626/*
2627 * Functions to lock and unlock any filesystem running on the
2628 * device.
2629 */
2630static int lock_fs(struct mapped_device *md)
2631{
2632 int r;
2633
2634 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2635
2636 r = freeze_bdev(md->disk->part0);
2637 if (!r)
2638 set_bit(DMF_FROZEN, &md->flags);
2639 return r;
2640}
2641
2642static void unlock_fs(struct mapped_device *md)
2643{
2644 if (!test_bit(DMF_FROZEN, &md->flags))
2645 return;
2646 thaw_bdev(md->disk->part0);
2647 clear_bit(DMF_FROZEN, &md->flags);
2648}
2649
2650/*
2651 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2652 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2653 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2654 *
2655 * If __dm_suspend returns 0, the device is completely quiescent
2656 * now. There is no request-processing activity. All new requests
2657 * are being added to md->deferred list.
2658 */
2659static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2660 unsigned suspend_flags, unsigned int task_state,
2661 int dmf_suspended_flag)
2662{
2663 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2664 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2665 int r;
2666
2667 lockdep_assert_held(&md->suspend_lock);
2668
2669 /*
2670 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2671 * This flag is cleared before dm_suspend returns.
2672 */
2673 if (noflush)
2674 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2675 else
2676 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2677
2678 /*
2679 * This gets reverted if there's an error later and the targets
2680 * provide the .presuspend_undo hook.
2681 */
2682 dm_table_presuspend_targets(map);
2683
2684 /*
2685 * Flush I/O to the device.
2686 * Any I/O submitted after lock_fs() may not be flushed.
2687 * noflush takes precedence over do_lockfs.
2688 * (lock_fs() flushes I/Os and waits for them to complete.)
2689 */
2690 if (!noflush && do_lockfs) {
2691 r = lock_fs(md);
2692 if (r) {
2693 dm_table_presuspend_undo_targets(map);
2694 return r;
2695 }
2696 }
2697
2698 /*
2699 * Here we must make sure that no processes are submitting requests
2700 * to target drivers i.e. no one may be executing
2701 * dm_split_and_process_bio from dm_submit_bio.
2702 *
2703 * To get all processes out of dm_split_and_process_bio in dm_submit_bio,
2704 * we take the write lock. To prevent any process from reentering
2705 * dm_split_and_process_bio from dm_submit_bio and quiesce the thread
2706 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2707 * flush_workqueue(md->wq).
2708 */
2709 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2710 if (map)
2711 synchronize_srcu(&md->io_barrier);
2712
2713 /*
2714 * Stop md->queue before flushing md->wq in case request-based
2715 * dm defers requests to md->wq from md->queue.
2716 */
2717 if (dm_request_based(md))
2718 dm_stop_queue(md->queue);
2719
2720 flush_workqueue(md->wq);
2721
2722 /*
2723 * At this point no more requests are entering target request routines.
2724 * We call dm_wait_for_completion to wait for all existing requests
2725 * to finish.
2726 */
2727 r = dm_wait_for_completion(md, task_state);
2728 if (!r)
2729 set_bit(dmf_suspended_flag, &md->flags);
2730
2731 if (noflush)
2732 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2733 if (map)
2734 synchronize_srcu(&md->io_barrier);
2735
2736 /* were we interrupted ? */
2737 if (r < 0) {
2738 dm_queue_flush(md);
2739
2740 if (dm_request_based(md))
2741 dm_start_queue(md->queue);
2742
2743 unlock_fs(md);
2744 dm_table_presuspend_undo_targets(map);
2745 /* pushback list is already flushed, so skip flush */
2746 }
2747
2748 return r;
2749}
2750
2751/*
2752 * We need to be able to change a mapping table under a mounted
2753 * filesystem. For example we might want to move some data in
2754 * the background. Before the table can be swapped with
2755 * dm_bind_table, dm_suspend must be called to flush any in
2756 * flight bios and ensure that any further io gets deferred.
2757 */
2758/*
2759 * Suspend mechanism in request-based dm.
2760 *
2761 * 1. Flush all I/Os by lock_fs() if needed.
2762 * 2. Stop dispatching any I/O by stopping the request_queue.
2763 * 3. Wait for all in-flight I/Os to be completed or requeued.
2764 *
2765 * To abort suspend, start the request_queue.
2766 */
2767int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2768{
2769 struct dm_table *map = NULL;
2770 int r = 0;
2771
2772retry:
2773 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2774
2775 if (dm_suspended_md(md)) {
2776 r = -EINVAL;
2777 goto out_unlock;
2778 }
2779
2780 if (dm_suspended_internally_md(md)) {
2781 /* already internally suspended, wait for internal resume */
2782 mutex_unlock(&md->suspend_lock);
2783 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2784 if (r)
2785 return r;
2786 goto retry;
2787 }
2788
2789 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2790
2791 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2792 if (r)
2793 goto out_unlock;
2794
2795 set_bit(DMF_POST_SUSPENDING, &md->flags);
2796 dm_table_postsuspend_targets(map);
2797 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2798
2799out_unlock:
2800 mutex_unlock(&md->suspend_lock);
2801 return r;
2802}
2803
2804static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2805{
2806 if (map) {
2807 int r = dm_table_resume_targets(map);
2808 if (r)
2809 return r;
2810 }
2811
2812 dm_queue_flush(md);
2813
2814 /*
2815 * Flushing deferred I/Os must be done after targets are resumed
2816 * so that mapping of targets can work correctly.
2817 * Request-based dm is queueing the deferred I/Os in its request_queue.
2818 */
2819 if (dm_request_based(md))
2820 dm_start_queue(md->queue);
2821
2822 unlock_fs(md);
2823
2824 return 0;
2825}
2826
2827int dm_resume(struct mapped_device *md)
2828{
2829 int r;
2830 struct dm_table *map = NULL;
2831
2832retry:
2833 r = -EINVAL;
2834 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2835
2836 if (!dm_suspended_md(md))
2837 goto out;
2838
2839 if (dm_suspended_internally_md(md)) {
2840 /* already internally suspended, wait for internal resume */
2841 mutex_unlock(&md->suspend_lock);
2842 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2843 if (r)
2844 return r;
2845 goto retry;
2846 }
2847
2848 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2849 if (!map || !dm_table_get_size(map))
2850 goto out;
2851
2852 r = __dm_resume(md, map);
2853 if (r)
2854 goto out;
2855
2856 clear_bit(DMF_SUSPENDED, &md->flags);
2857out:
2858 mutex_unlock(&md->suspend_lock);
2859
2860 return r;
2861}
2862
2863/*
2864 * Internal suspend/resume works like userspace-driven suspend. It waits
2865 * until all bios finish and prevents issuing new bios to the target drivers.
2866 * It may be used only from the kernel.
2867 */
2868
2869static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2870{
2871 struct dm_table *map = NULL;
2872
2873 lockdep_assert_held(&md->suspend_lock);
2874
2875 if (md->internal_suspend_count++)
2876 return; /* nested internal suspend */
2877
2878 if (dm_suspended_md(md)) {
2879 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2880 return; /* nest suspend */
2881 }
2882
2883 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2884
2885 /*
2886 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2887 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2888 * would require changing .presuspend to return an error -- avoid this
2889 * until there is a need for more elaborate variants of internal suspend.
2890 */
2891 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2892 DMF_SUSPENDED_INTERNALLY);
2893
2894 set_bit(DMF_POST_SUSPENDING, &md->flags);
2895 dm_table_postsuspend_targets(map);
2896 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2897}
2898
2899static void __dm_internal_resume(struct mapped_device *md)
2900{
2901 BUG_ON(!md->internal_suspend_count);
2902
2903 if (--md->internal_suspend_count)
2904 return; /* resume from nested internal suspend */
2905
2906 if (dm_suspended_md(md))
2907 goto done; /* resume from nested suspend */
2908
2909 /*
2910 * NOTE: existing callers don't need to call dm_table_resume_targets
2911 * (which may fail -- so best to avoid it for now by passing NULL map)
2912 */
2913 (void) __dm_resume(md, NULL);
2914
2915done:
2916 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2917 smp_mb__after_atomic();
2918 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2919}
2920
2921void dm_internal_suspend_noflush(struct mapped_device *md)
2922{
2923 mutex_lock(&md->suspend_lock);
2924 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2925 mutex_unlock(&md->suspend_lock);
2926}
2927EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2928
2929void dm_internal_resume(struct mapped_device *md)
2930{
2931 mutex_lock(&md->suspend_lock);
2932 __dm_internal_resume(md);
2933 mutex_unlock(&md->suspend_lock);
2934}
2935EXPORT_SYMBOL_GPL(dm_internal_resume);
2936
2937/*
2938 * Fast variants of internal suspend/resume hold md->suspend_lock,
2939 * which prevents interaction with userspace-driven suspend.
2940 */
2941
2942void dm_internal_suspend_fast(struct mapped_device *md)
2943{
2944 mutex_lock(&md->suspend_lock);
2945 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2946 return;
2947
2948 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2949 synchronize_srcu(&md->io_barrier);
2950 flush_workqueue(md->wq);
2951 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2952}
2953EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2954
2955void dm_internal_resume_fast(struct mapped_device *md)
2956{
2957 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2958 goto done;
2959
2960 dm_queue_flush(md);
2961
2962done:
2963 mutex_unlock(&md->suspend_lock);
2964}
2965EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2966
2967/*-----------------------------------------------------------------
2968 * Event notification.
2969 *---------------------------------------------------------------*/
2970int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2971 unsigned cookie)
2972{
2973 int r;
2974 unsigned noio_flag;
2975 char udev_cookie[DM_COOKIE_LENGTH];
2976 char *envp[] = { udev_cookie, NULL };
2977
2978 noio_flag = memalloc_noio_save();
2979
2980 if (!cookie)
2981 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2982 else {
2983 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2984 DM_COOKIE_ENV_VAR_NAME, cookie);
2985 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2986 action, envp);
2987 }
2988
2989 memalloc_noio_restore(noio_flag);
2990
2991 return r;
2992}
2993
2994uint32_t dm_next_uevent_seq(struct mapped_device *md)
2995{
2996 return atomic_add_return(1, &md->uevent_seq);
2997}
2998
2999uint32_t dm_get_event_nr(struct mapped_device *md)
3000{
3001 return atomic_read(&md->event_nr);
3002}
3003
3004int dm_wait_event(struct mapped_device *md, int event_nr)
3005{
3006 return wait_event_interruptible(md->eventq,
3007 (event_nr != atomic_read(&md->event_nr)));
3008}
3009
3010void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
3011{
3012 unsigned long flags;
3013
3014 spin_lock_irqsave(&md->uevent_lock, flags);
3015 list_add(elist, &md->uevent_list);
3016 spin_unlock_irqrestore(&md->uevent_lock, flags);
3017}
3018
3019/*
3020 * The gendisk is only valid as long as you have a reference
3021 * count on 'md'.
3022 */
3023struct gendisk *dm_disk(struct mapped_device *md)
3024{
3025 return md->disk;
3026}
3027EXPORT_SYMBOL_GPL(dm_disk);
3028
3029struct kobject *dm_kobject(struct mapped_device *md)
3030{
3031 return &md->kobj_holder.kobj;
3032}
3033
3034struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
3035{
3036 struct mapped_device *md;
3037
3038 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
3039
3040 spin_lock(&_minor_lock);
3041 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
3042 md = NULL;
3043 goto out;
3044 }
3045 dm_get(md);
3046out:
3047 spin_unlock(&_minor_lock);
3048
3049 return md;
3050}
3051
3052int dm_suspended_md(struct mapped_device *md)
3053{
3054 return test_bit(DMF_SUSPENDED, &md->flags);
3055}
3056
3057static int dm_post_suspending_md(struct mapped_device *md)
3058{
3059 return test_bit(DMF_POST_SUSPENDING, &md->flags);
3060}
3061
3062int dm_suspended_internally_md(struct mapped_device *md)
3063{
3064 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
3065}
3066
3067int dm_test_deferred_remove_flag(struct mapped_device *md)
3068{
3069 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
3070}
3071
3072int dm_suspended(struct dm_target *ti)
3073{
3074 return dm_suspended_md(ti->table->md);
3075}
3076EXPORT_SYMBOL_GPL(dm_suspended);
3077
3078int dm_post_suspending(struct dm_target *ti)
3079{
3080 return dm_post_suspending_md(ti->table->md);
3081}
3082EXPORT_SYMBOL_GPL(dm_post_suspending);
3083
3084int dm_noflush_suspending(struct dm_target *ti)
3085{
3086 return __noflush_suspending(ti->table->md);
3087}
3088EXPORT_SYMBOL_GPL(dm_noflush_suspending);
3089
3090void dm_free_md_mempools(struct dm_md_mempools *pools)
3091{
3092 if (!pools)
3093 return;
3094
3095 bioset_exit(&pools->bs);
3096 bioset_exit(&pools->io_bs);
3097
3098 kfree(pools);
3099}
3100
3101struct dm_pr {
3102 u64 old_key;
3103 u64 new_key;
3104 u32 flags;
3105 bool abort;
3106 bool fail_early;
3107 int ret;
3108 enum pr_type type;
3109};
3110
3111static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
3112 struct dm_pr *pr)
3113{
3114 struct mapped_device *md = bdev->bd_disk->private_data;
3115 struct dm_table *table;
3116 struct dm_target *ti;
3117 int ret = -ENOTTY, srcu_idx;
3118
3119 table = dm_get_live_table(md, &srcu_idx);
3120 if (!table || !dm_table_get_size(table))
3121 goto out;
3122
3123 /* We only support devices that have a single target */
3124 if (table->num_targets != 1)
3125 goto out;
3126 ti = dm_table_get_target(table, 0);
3127
3128 if (dm_suspended_md(md)) {
3129 ret = -EAGAIN;
3130 goto out;
3131 }
3132
3133 ret = -EINVAL;
3134 if (!ti->type->iterate_devices)
3135 goto out;
3136
3137 ti->type->iterate_devices(ti, fn, pr);
3138 ret = 0;
3139out:
3140 dm_put_live_table(md, srcu_idx);
3141 return ret;
3142}
3143
3144/*
3145 * For register / unregister we need to manually call out to every path.
3146 */
3147static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
3148 sector_t start, sector_t len, void *data)
3149{
3150 struct dm_pr *pr = data;
3151 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3152 int ret;
3153
3154 if (!ops || !ops->pr_register) {
3155 pr->ret = -EOPNOTSUPP;
3156 return -1;
3157 }
3158
3159 ret = ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
3160 if (!ret)
3161 return 0;
3162
3163 if (!pr->ret)
3164 pr->ret = ret;
3165
3166 if (pr->fail_early)
3167 return -1;
3168
3169 return 0;
3170}
3171
3172static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
3173 u32 flags)
3174{
3175 struct dm_pr pr = {
3176 .old_key = old_key,
3177 .new_key = new_key,
3178 .flags = flags,
3179 .fail_early = true,
3180 .ret = 0,
3181 };
3182 int ret;
3183
3184 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
3185 if (ret) {
3186 /* Didn't even get to register a path */
3187 return ret;
3188 }
3189
3190 if (!pr.ret)
3191 return 0;
3192 ret = pr.ret;
3193
3194 if (!new_key)
3195 return ret;
3196
3197 /* unregister all paths if we failed to register any path */
3198 pr.old_key = new_key;
3199 pr.new_key = 0;
3200 pr.flags = 0;
3201 pr.fail_early = false;
3202 (void) dm_call_pr(bdev, __dm_pr_register, &pr);
3203 return ret;
3204}
3205
3206
3207static int __dm_pr_reserve(struct dm_target *ti, struct dm_dev *dev,
3208 sector_t start, sector_t len, void *data)
3209{
3210 struct dm_pr *pr = data;
3211 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3212
3213 if (!ops || !ops->pr_reserve) {
3214 pr->ret = -EOPNOTSUPP;
3215 return -1;
3216 }
3217
3218 pr->ret = ops->pr_reserve(dev->bdev, pr->old_key, pr->type, pr->flags);
3219 if (!pr->ret)
3220 return -1;
3221
3222 return 0;
3223}
3224
3225static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
3226 u32 flags)
3227{
3228 struct dm_pr pr = {
3229 .old_key = key,
3230 .flags = flags,
3231 .type = type,
3232 .fail_early = false,
3233 .ret = 0,
3234 };
3235 int ret;
3236
3237 ret = dm_call_pr(bdev, __dm_pr_reserve, &pr);
3238 if (ret)
3239 return ret;
3240
3241 return pr.ret;
3242}
3243
3244/*
3245 * If there is a non-All Registrants type of reservation, the release must be
3246 * sent down the holding path. For the cases where there is no reservation or
3247 * the path is not the holder the device will also return success, so we must
3248 * try each path to make sure we got the correct path.
3249 */
3250static int __dm_pr_release(struct dm_target *ti, struct dm_dev *dev,
3251 sector_t start, sector_t len, void *data)
3252{
3253 struct dm_pr *pr = data;
3254 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3255
3256 if (!ops || !ops->pr_release) {
3257 pr->ret = -EOPNOTSUPP;
3258 return -1;
3259 }
3260
3261 pr->ret = ops->pr_release(dev->bdev, pr->old_key, pr->type);
3262 if (pr->ret)
3263 return -1;
3264
3265 return 0;
3266}
3267
3268static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
3269{
3270 struct dm_pr pr = {
3271 .old_key = key,
3272 .type = type,
3273 .fail_early = false,
3274 };
3275 int ret;
3276
3277 ret = dm_call_pr(bdev, __dm_pr_release, &pr);
3278 if (ret)
3279 return ret;
3280
3281 return pr.ret;
3282}
3283
3284static int __dm_pr_preempt(struct dm_target *ti, struct dm_dev *dev,
3285 sector_t start, sector_t len, void *data)
3286{
3287 struct dm_pr *pr = data;
3288 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
3289
3290 if (!ops || !ops->pr_preempt) {
3291 pr->ret = -EOPNOTSUPP;
3292 return -1;
3293 }
3294
3295 pr->ret = ops->pr_preempt(dev->bdev, pr->old_key, pr->new_key, pr->type,
3296 pr->abort);
3297 if (!pr->ret)
3298 return -1;
3299
3300 return 0;
3301}
3302
3303static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
3304 enum pr_type type, bool abort)
3305{
3306 struct dm_pr pr = {
3307 .new_key = new_key,
3308 .old_key = old_key,
3309 .type = type,
3310 .fail_early = false,
3311 };
3312 int ret;
3313
3314 ret = dm_call_pr(bdev, __dm_pr_preempt, &pr);
3315 if (ret)
3316 return ret;
3317
3318 return pr.ret;
3319}
3320
3321static int dm_pr_clear(struct block_device *bdev, u64 key)
3322{
3323 struct mapped_device *md = bdev->bd_disk->private_data;
3324 const struct pr_ops *ops;
3325 int r, srcu_idx;
3326
3327 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3328 if (r < 0)
3329 goto out;
3330
3331 ops = bdev->bd_disk->fops->pr_ops;
3332 if (ops && ops->pr_clear)
3333 r = ops->pr_clear(bdev, key);
3334 else
3335 r = -EOPNOTSUPP;
3336out:
3337 dm_unprepare_ioctl(md, srcu_idx);
3338 return r;
3339}
3340
3341static const struct pr_ops dm_pr_ops = {
3342 .pr_register = dm_pr_register,
3343 .pr_reserve = dm_pr_reserve,
3344 .pr_release = dm_pr_release,
3345 .pr_preempt = dm_pr_preempt,
3346 .pr_clear = dm_pr_clear,
3347};
3348
3349static const struct block_device_operations dm_blk_dops = {
3350 .submit_bio = dm_submit_bio,
3351 .poll_bio = dm_poll_bio,
3352 .open = dm_blk_open,
3353 .release = dm_blk_close,
3354 .ioctl = dm_blk_ioctl,
3355 .getgeo = dm_blk_getgeo,
3356 .report_zones = dm_blk_report_zones,
3357 .pr_ops = &dm_pr_ops,
3358 .owner = THIS_MODULE
3359};
3360
3361static const struct block_device_operations dm_rq_blk_dops = {
3362 .open = dm_blk_open,
3363 .release = dm_blk_close,
3364 .ioctl = dm_blk_ioctl,
3365 .getgeo = dm_blk_getgeo,
3366 .pr_ops = &dm_pr_ops,
3367 .owner = THIS_MODULE
3368};
3369
3370static const struct dax_operations dm_dax_ops = {
3371 .direct_access = dm_dax_direct_access,
3372 .zero_page_range = dm_dax_zero_page_range,
3373 .recovery_write = dm_dax_recovery_write,
3374};
3375
3376/*
3377 * module hooks
3378 */
3379module_init(dm_init);
3380module_exit(dm_exit);
3381
3382module_param(major, uint, 0);
3383MODULE_PARM_DESC(major, "The major number of the device mapper");
3384
3385module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3386MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3387
3388module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3389MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3390
3391module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3392MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3393
3394MODULE_DESCRIPTION(DM_NAME " driver");
3395MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3396MODULE_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
53static void do_deferred_remove(struct work_struct *w);
54
55static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
56
57/*
58 * For bio-based dm.
59 * One of these is allocated per bio.
60 */
61struct dm_io {
62 struct mapped_device *md;
63 int error;
64 atomic_t io_count;
65 struct bio *bio;
66 unsigned long start_time;
67 spinlock_t endio_lock;
68 struct dm_stats_aux stats_aux;
69};
70
71/*
72 * For request-based dm.
73 * One of these is allocated per request.
74 */
75struct dm_rq_target_io {
76 struct mapped_device *md;
77 struct dm_target *ti;
78 struct request *orig, clone;
79 int error;
80 union map_info info;
81};
82
83/*
84 * For request-based dm - the bio clones we allocate are embedded in these
85 * structs.
86 *
87 * We allocate these with bio_alloc_bioset, using the front_pad parameter when
88 * the bioset is created - this means the bio has to come at the end of the
89 * struct.
90 */
91struct dm_rq_clone_bio_info {
92 struct bio *orig;
93 struct dm_rq_target_io *tio;
94 struct bio clone;
95};
96
97union map_info *dm_get_rq_mapinfo(struct request *rq)
98{
99 if (rq && rq->end_io_data)
100 return &((struct dm_rq_target_io *)rq->end_io_data)->info;
101 return NULL;
102}
103EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo);
104
105#define MINOR_ALLOCED ((void *)-1)
106
107/*
108 * Bits for the md->flags field.
109 */
110#define DMF_BLOCK_IO_FOR_SUSPEND 0
111#define DMF_SUSPENDED 1
112#define DMF_FROZEN 2
113#define DMF_FREEING 3
114#define DMF_DELETING 4
115#define DMF_NOFLUSH_SUSPENDING 5
116#define DMF_MERGE_IS_OPTIONAL 6
117#define DMF_DEFERRED_REMOVE 7
118
119/*
120 * A dummy definition to make RCU happy.
121 * struct dm_table should never be dereferenced in this file.
122 */
123struct dm_table {
124 int undefined__;
125};
126
127/*
128 * Work processed by per-device workqueue.
129 */
130struct mapped_device {
131 struct srcu_struct io_barrier;
132 struct mutex suspend_lock;
133 atomic_t holders;
134 atomic_t open_count;
135
136 /*
137 * The current mapping.
138 * Use dm_get_live_table{_fast} or take suspend_lock for
139 * dereference.
140 */
141 struct dm_table *map;
142
143 unsigned long flags;
144
145 struct request_queue *queue;
146 unsigned type;
147 /* Protect queue and type against concurrent access. */
148 struct mutex type_lock;
149
150 struct target_type *immutable_target_type;
151
152 struct gendisk *disk;
153 char name[16];
154
155 void *interface_ptr;
156
157 /*
158 * A list of ios that arrived while we were suspended.
159 */
160 atomic_t pending[2];
161 wait_queue_head_t wait;
162 struct work_struct work;
163 struct bio_list deferred;
164 spinlock_t deferred_lock;
165
166 /*
167 * Processing queue (flush)
168 */
169 struct workqueue_struct *wq;
170
171 /*
172 * io objects are allocated from here.
173 */
174 mempool_t *io_pool;
175
176 struct bio_set *bs;
177
178 /*
179 * Event handling.
180 */
181 atomic_t event_nr;
182 wait_queue_head_t eventq;
183 atomic_t uevent_seq;
184 struct list_head uevent_list;
185 spinlock_t uevent_lock; /* Protect access to uevent_list */
186
187 /*
188 * freeze/thaw support require holding onto a super block
189 */
190 struct super_block *frozen_sb;
191 struct block_device *bdev;
192
193 /* forced geometry settings */
194 struct hd_geometry geometry;
195
196 /* kobject and completion */
197 struct dm_kobject_holder kobj_holder;
198
199 /* zero-length flush that will be cloned and submitted to targets */
200 struct bio flush_bio;
201
202 struct dm_stats stats;
203};
204
205/*
206 * For mempools pre-allocation at the table loading time.
207 */
208struct dm_md_mempools {
209 mempool_t *io_pool;
210 struct bio_set *bs;
211};
212
213#define RESERVED_BIO_BASED_IOS 16
214#define RESERVED_REQUEST_BASED_IOS 256
215#define RESERVED_MAX_IOS 1024
216static struct kmem_cache *_io_cache;
217static struct kmem_cache *_rq_tio_cache;
218
219/*
220 * Bio-based DM's mempools' reserved IOs set by the user.
221 */
222static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
223
224/*
225 * Request-based DM's mempools' reserved IOs set by the user.
226 */
227static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS;
228
229static unsigned __dm_get_reserved_ios(unsigned *reserved_ios,
230 unsigned def, unsigned max)
231{
232 unsigned ios = ACCESS_ONCE(*reserved_ios);
233 unsigned modified_ios = 0;
234
235 if (!ios)
236 modified_ios = def;
237 else if (ios > max)
238 modified_ios = max;
239
240 if (modified_ios) {
241 (void)cmpxchg(reserved_ios, ios, modified_ios);
242 ios = modified_ios;
243 }
244
245 return ios;
246}
247
248unsigned dm_get_reserved_bio_based_ios(void)
249{
250 return __dm_get_reserved_ios(&reserved_bio_based_ios,
251 RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS);
252}
253EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
254
255unsigned dm_get_reserved_rq_based_ios(void)
256{
257 return __dm_get_reserved_ios(&reserved_rq_based_ios,
258 RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS);
259}
260EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios);
261
262static int __init local_init(void)
263{
264 int r = -ENOMEM;
265
266 /* allocate a slab for the dm_ios */
267 _io_cache = KMEM_CACHE(dm_io, 0);
268 if (!_io_cache)
269 return r;
270
271 _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0);
272 if (!_rq_tio_cache)
273 goto out_free_io_cache;
274
275 r = dm_uevent_init();
276 if (r)
277 goto out_free_rq_tio_cache;
278
279 _major = major;
280 r = register_blkdev(_major, _name);
281 if (r < 0)
282 goto out_uevent_exit;
283
284 if (!_major)
285 _major = r;
286
287 return 0;
288
289out_uevent_exit:
290 dm_uevent_exit();
291out_free_rq_tio_cache:
292 kmem_cache_destroy(_rq_tio_cache);
293out_free_io_cache:
294 kmem_cache_destroy(_io_cache);
295
296 return r;
297}
298
299static void local_exit(void)
300{
301 flush_scheduled_work();
302
303 kmem_cache_destroy(_rq_tio_cache);
304 kmem_cache_destroy(_io_cache);
305 unregister_blkdev(_major, _name);
306 dm_uevent_exit();
307
308 _major = 0;
309
310 DMINFO("cleaned up");
311}
312
313static int (*_inits[])(void) __initdata = {
314 local_init,
315 dm_target_init,
316 dm_linear_init,
317 dm_stripe_init,
318 dm_io_init,
319 dm_kcopyd_init,
320 dm_interface_init,
321 dm_statistics_init,
322};
323
324static void (*_exits[])(void) = {
325 local_exit,
326 dm_target_exit,
327 dm_linear_exit,
328 dm_stripe_exit,
329 dm_io_exit,
330 dm_kcopyd_exit,
331 dm_interface_exit,
332 dm_statistics_exit,
333};
334
335static int __init dm_init(void)
336{
337 const int count = ARRAY_SIZE(_inits);
338
339 int r, i;
340
341 for (i = 0; i < count; i++) {
342 r = _inits[i]();
343 if (r)
344 goto bad;
345 }
346
347 return 0;
348
349 bad:
350 while (i--)
351 _exits[i]();
352
353 return r;
354}
355
356static void __exit dm_exit(void)
357{
358 int i = ARRAY_SIZE(_exits);
359
360 while (i--)
361 _exits[i]();
362
363 /*
364 * Should be empty by this point.
365 */
366 idr_destroy(&_minor_idr);
367}
368
369/*
370 * Block device functions
371 */
372int dm_deleting_md(struct mapped_device *md)
373{
374 return test_bit(DMF_DELETING, &md->flags);
375}
376
377static int dm_blk_open(struct block_device *bdev, fmode_t mode)
378{
379 struct mapped_device *md;
380
381 spin_lock(&_minor_lock);
382
383 md = bdev->bd_disk->private_data;
384 if (!md)
385 goto out;
386
387 if (test_bit(DMF_FREEING, &md->flags) ||
388 dm_deleting_md(md)) {
389 md = NULL;
390 goto out;
391 }
392
393 dm_get(md);
394 atomic_inc(&md->open_count);
395
396out:
397 spin_unlock(&_minor_lock);
398
399 return md ? 0 : -ENXIO;
400}
401
402static void dm_blk_close(struct gendisk *disk, fmode_t mode)
403{
404 struct mapped_device *md = disk->private_data;
405
406 spin_lock(&_minor_lock);
407
408 if (atomic_dec_and_test(&md->open_count) &&
409 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
410 schedule_work(&deferred_remove_work);
411
412 dm_put(md);
413
414 spin_unlock(&_minor_lock);
415}
416
417int dm_open_count(struct mapped_device *md)
418{
419 return atomic_read(&md->open_count);
420}
421
422/*
423 * Guarantees nothing is using the device before it's deleted.
424 */
425int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
426{
427 int r = 0;
428
429 spin_lock(&_minor_lock);
430
431 if (dm_open_count(md)) {
432 r = -EBUSY;
433 if (mark_deferred)
434 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
435 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
436 r = -EEXIST;
437 else
438 set_bit(DMF_DELETING, &md->flags);
439
440 spin_unlock(&_minor_lock);
441
442 return r;
443}
444
445int dm_cancel_deferred_remove(struct mapped_device *md)
446{
447 int r = 0;
448
449 spin_lock(&_minor_lock);
450
451 if (test_bit(DMF_DELETING, &md->flags))
452 r = -EBUSY;
453 else
454 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
455
456 spin_unlock(&_minor_lock);
457
458 return r;
459}
460
461static void do_deferred_remove(struct work_struct *w)
462{
463 dm_deferred_remove();
464}
465
466sector_t dm_get_size(struct mapped_device *md)
467{
468 return get_capacity(md->disk);
469}
470
471struct request_queue *dm_get_md_queue(struct mapped_device *md)
472{
473 return md->queue;
474}
475
476struct dm_stats *dm_get_stats(struct mapped_device *md)
477{
478 return &md->stats;
479}
480
481static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
482{
483 struct mapped_device *md = bdev->bd_disk->private_data;
484
485 return dm_get_geometry(md, geo);
486}
487
488static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
489 unsigned int cmd, unsigned long arg)
490{
491 struct mapped_device *md = bdev->bd_disk->private_data;
492 int srcu_idx;
493 struct dm_table *map;
494 struct dm_target *tgt;
495 int r = -ENOTTY;
496
497retry:
498 map = dm_get_live_table(md, &srcu_idx);
499
500 if (!map || !dm_table_get_size(map))
501 goto out;
502
503 /* We only support devices that have a single target */
504 if (dm_table_get_num_targets(map) != 1)
505 goto out;
506
507 tgt = dm_table_get_target(map, 0);
508
509 if (dm_suspended_md(md)) {
510 r = -EAGAIN;
511 goto out;
512 }
513
514 if (tgt->type->ioctl)
515 r = tgt->type->ioctl(tgt, cmd, arg);
516
517out:
518 dm_put_live_table(md, srcu_idx);
519
520 if (r == -ENOTCONN) {
521 msleep(10);
522 goto retry;
523 }
524
525 return r;
526}
527
528static struct dm_io *alloc_io(struct mapped_device *md)
529{
530 return mempool_alloc(md->io_pool, GFP_NOIO);
531}
532
533static void free_io(struct mapped_device *md, struct dm_io *io)
534{
535 mempool_free(io, md->io_pool);
536}
537
538static void free_tio(struct mapped_device *md, struct dm_target_io *tio)
539{
540 bio_put(&tio->clone);
541}
542
543static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md,
544 gfp_t gfp_mask)
545{
546 return mempool_alloc(md->io_pool, gfp_mask);
547}
548
549static void free_rq_tio(struct dm_rq_target_io *tio)
550{
551 mempool_free(tio, tio->md->io_pool);
552}
553
554static int md_in_flight(struct mapped_device *md)
555{
556 return atomic_read(&md->pending[READ]) +
557 atomic_read(&md->pending[WRITE]);
558}
559
560static void start_io_acct(struct dm_io *io)
561{
562 struct mapped_device *md = io->md;
563 struct bio *bio = io->bio;
564 int cpu;
565 int rw = bio_data_dir(bio);
566
567 io->start_time = jiffies;
568
569 cpu = part_stat_lock();
570 part_round_stats(cpu, &dm_disk(md)->part0);
571 part_stat_unlock();
572 atomic_set(&dm_disk(md)->part0.in_flight[rw],
573 atomic_inc_return(&md->pending[rw]));
574
575 if (unlikely(dm_stats_used(&md->stats)))
576 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
577 bio_sectors(bio), false, 0, &io->stats_aux);
578}
579
580static void end_io_acct(struct dm_io *io)
581{
582 struct mapped_device *md = io->md;
583 struct bio *bio = io->bio;
584 unsigned long duration = jiffies - io->start_time;
585 int pending, cpu;
586 int rw = bio_data_dir(bio);
587
588 cpu = part_stat_lock();
589 part_round_stats(cpu, &dm_disk(md)->part0);
590 part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration);
591 part_stat_unlock();
592
593 if (unlikely(dm_stats_used(&md->stats)))
594 dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector,
595 bio_sectors(bio), true, duration, &io->stats_aux);
596
597 /*
598 * After this is decremented the bio must not be touched if it is
599 * a flush.
600 */
601 pending = atomic_dec_return(&md->pending[rw]);
602 atomic_set(&dm_disk(md)->part0.in_flight[rw], pending);
603 pending += atomic_read(&md->pending[rw^0x1]);
604
605 /* nudge anyone waiting on suspend queue */
606 if (!pending)
607 wake_up(&md->wait);
608}
609
610/*
611 * Add the bio to the list of deferred io.
612 */
613static void queue_io(struct mapped_device *md, struct bio *bio)
614{
615 unsigned long flags;
616
617 spin_lock_irqsave(&md->deferred_lock, flags);
618 bio_list_add(&md->deferred, bio);
619 spin_unlock_irqrestore(&md->deferred_lock, flags);
620 queue_work(md->wq, &md->work);
621}
622
623/*
624 * Everyone (including functions in this file), should use this
625 * function to access the md->map field, and make sure they call
626 * dm_put_live_table() when finished.
627 */
628struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
629{
630 *srcu_idx = srcu_read_lock(&md->io_barrier);
631
632 return srcu_dereference(md->map, &md->io_barrier);
633}
634
635void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
636{
637 srcu_read_unlock(&md->io_barrier, srcu_idx);
638}
639
640void dm_sync_table(struct mapped_device *md)
641{
642 synchronize_srcu(&md->io_barrier);
643 synchronize_rcu_expedited();
644}
645
646/*
647 * A fast alternative to dm_get_live_table/dm_put_live_table.
648 * The caller must not block between these two functions.
649 */
650static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
651{
652 rcu_read_lock();
653 return rcu_dereference(md->map);
654}
655
656static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
657{
658 rcu_read_unlock();
659}
660
661/*
662 * Get the geometry associated with a dm device
663 */
664int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
665{
666 *geo = md->geometry;
667
668 return 0;
669}
670
671/*
672 * Set the geometry of a device.
673 */
674int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
675{
676 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
677
678 if (geo->start > sz) {
679 DMWARN("Start sector is beyond the geometry limits.");
680 return -EINVAL;
681 }
682
683 md->geometry = *geo;
684
685 return 0;
686}
687
688/*-----------------------------------------------------------------
689 * CRUD START:
690 * A more elegant soln is in the works that uses the queue
691 * merge fn, unfortunately there are a couple of changes to
692 * the block layer that I want to make for this. So in the
693 * interests of getting something for people to use I give
694 * you this clearly demarcated crap.
695 *---------------------------------------------------------------*/
696
697static int __noflush_suspending(struct mapped_device *md)
698{
699 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
700}
701
702/*
703 * Decrements the number of outstanding ios that a bio has been
704 * cloned into, completing the original io if necc.
705 */
706static void dec_pending(struct dm_io *io, int error)
707{
708 unsigned long flags;
709 int io_error;
710 struct bio *bio;
711 struct mapped_device *md = io->md;
712
713 /* Push-back supersedes any I/O errors */
714 if (unlikely(error)) {
715 spin_lock_irqsave(&io->endio_lock, flags);
716 if (!(io->error > 0 && __noflush_suspending(md)))
717 io->error = error;
718 spin_unlock_irqrestore(&io->endio_lock, flags);
719 }
720
721 if (atomic_dec_and_test(&io->io_count)) {
722 if (io->error == DM_ENDIO_REQUEUE) {
723 /*
724 * Target requested pushing back the I/O.
725 */
726 spin_lock_irqsave(&md->deferred_lock, flags);
727 if (__noflush_suspending(md))
728 bio_list_add_head(&md->deferred, io->bio);
729 else
730 /* noflush suspend was interrupted. */
731 io->error = -EIO;
732 spin_unlock_irqrestore(&md->deferred_lock, flags);
733 }
734
735 io_error = io->error;
736 bio = io->bio;
737 end_io_acct(io);
738 free_io(md, io);
739
740 if (io_error == DM_ENDIO_REQUEUE)
741 return;
742
743 if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) {
744 /*
745 * Preflush done for flush with data, reissue
746 * without REQ_FLUSH.
747 */
748 bio->bi_rw &= ~REQ_FLUSH;
749 queue_io(md, bio);
750 } else {
751 /* done with normal IO or empty flush */
752 trace_block_bio_complete(md->queue, bio, io_error);
753 bio_endio(bio, io_error);
754 }
755 }
756}
757
758static void clone_endio(struct bio *bio, int error)
759{
760 int r = 0;
761 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
762 struct dm_io *io = tio->io;
763 struct mapped_device *md = tio->io->md;
764 dm_endio_fn endio = tio->ti->type->end_io;
765
766 if (!bio_flagged(bio, BIO_UPTODATE) && !error)
767 error = -EIO;
768
769 if (endio) {
770 r = endio(tio->ti, bio, error);
771 if (r < 0 || r == DM_ENDIO_REQUEUE)
772 /*
773 * error and requeue request are handled
774 * in dec_pending().
775 */
776 error = r;
777 else if (r == DM_ENDIO_INCOMPLETE)
778 /* The target will handle the io */
779 return;
780 else if (r) {
781 DMWARN("unimplemented target endio return value: %d", r);
782 BUG();
783 }
784 }
785
786 free_tio(md, tio);
787 dec_pending(io, error);
788}
789
790/*
791 * Partial completion handling for request-based dm
792 */
793static void end_clone_bio(struct bio *clone, int error)
794{
795 struct dm_rq_clone_bio_info *info =
796 container_of(clone, struct dm_rq_clone_bio_info, clone);
797 struct dm_rq_target_io *tio = info->tio;
798 struct bio *bio = info->orig;
799 unsigned int nr_bytes = info->orig->bi_iter.bi_size;
800
801 bio_put(clone);
802
803 if (tio->error)
804 /*
805 * An error has already been detected on the request.
806 * Once error occurred, just let clone->end_io() handle
807 * the remainder.
808 */
809 return;
810 else if (error) {
811 /*
812 * Don't notice the error to the upper layer yet.
813 * The error handling decision is made by the target driver,
814 * when the request is completed.
815 */
816 tio->error = error;
817 return;
818 }
819
820 /*
821 * I/O for the bio successfully completed.
822 * Notice the data completion to the upper layer.
823 */
824
825 /*
826 * bios are processed from the head of the list.
827 * So the completing bio should always be rq->bio.
828 * If it's not, something wrong is happening.
829 */
830 if (tio->orig->bio != bio)
831 DMERR("bio completion is going in the middle of the request");
832
833 /*
834 * Update the original request.
835 * Do not use blk_end_request() here, because it may complete
836 * the original request before the clone, and break the ordering.
837 */
838 blk_update_request(tio->orig, 0, nr_bytes);
839}
840
841/*
842 * Don't touch any member of the md after calling this function because
843 * the md may be freed in dm_put() at the end of this function.
844 * Or do dm_get() before calling this function and dm_put() later.
845 */
846static void rq_completed(struct mapped_device *md, int rw, int run_queue)
847{
848 atomic_dec(&md->pending[rw]);
849
850 /* nudge anyone waiting on suspend queue */
851 if (!md_in_flight(md))
852 wake_up(&md->wait);
853
854 /*
855 * Run this off this callpath, as drivers could invoke end_io while
856 * inside their request_fn (and holding the queue lock). Calling
857 * back into ->request_fn() could deadlock attempting to grab the
858 * queue lock again.
859 */
860 if (run_queue)
861 blk_run_queue_async(md->queue);
862
863 /*
864 * dm_put() must be at the end of this function. See the comment above
865 */
866 dm_put(md);
867}
868
869static void free_rq_clone(struct request *clone)
870{
871 struct dm_rq_target_io *tio = clone->end_io_data;
872
873 blk_rq_unprep_clone(clone);
874 free_rq_tio(tio);
875}
876
877/*
878 * Complete the clone and the original request.
879 * Must be called without queue lock.
880 */
881static void dm_end_request(struct request *clone, int error)
882{
883 int rw = rq_data_dir(clone);
884 struct dm_rq_target_io *tio = clone->end_io_data;
885 struct mapped_device *md = tio->md;
886 struct request *rq = tio->orig;
887
888 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
889 rq->errors = clone->errors;
890 rq->resid_len = clone->resid_len;
891
892 if (rq->sense)
893 /*
894 * We are using the sense buffer of the original
895 * request.
896 * So setting the length of the sense data is enough.
897 */
898 rq->sense_len = clone->sense_len;
899 }
900
901 free_rq_clone(clone);
902 blk_end_request_all(rq, error);
903 rq_completed(md, rw, true);
904}
905
906static void dm_unprep_request(struct request *rq)
907{
908 struct request *clone = rq->special;
909
910 rq->special = NULL;
911 rq->cmd_flags &= ~REQ_DONTPREP;
912
913 free_rq_clone(clone);
914}
915
916/*
917 * Requeue the original request of a clone.
918 */
919void dm_requeue_unmapped_request(struct request *clone)
920{
921 int rw = rq_data_dir(clone);
922 struct dm_rq_target_io *tio = clone->end_io_data;
923 struct mapped_device *md = tio->md;
924 struct request *rq = tio->orig;
925 struct request_queue *q = rq->q;
926 unsigned long flags;
927
928 dm_unprep_request(rq);
929
930 spin_lock_irqsave(q->queue_lock, flags);
931 blk_requeue_request(q, rq);
932 spin_unlock_irqrestore(q->queue_lock, flags);
933
934 rq_completed(md, rw, 0);
935}
936EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request);
937
938static void __stop_queue(struct request_queue *q)
939{
940 blk_stop_queue(q);
941}
942
943static void stop_queue(struct request_queue *q)
944{
945 unsigned long flags;
946
947 spin_lock_irqsave(q->queue_lock, flags);
948 __stop_queue(q);
949 spin_unlock_irqrestore(q->queue_lock, flags);
950}
951
952static void __start_queue(struct request_queue *q)
953{
954 if (blk_queue_stopped(q))
955 blk_start_queue(q);
956}
957
958static void start_queue(struct request_queue *q)
959{
960 unsigned long flags;
961
962 spin_lock_irqsave(q->queue_lock, flags);
963 __start_queue(q);
964 spin_unlock_irqrestore(q->queue_lock, flags);
965}
966
967static void dm_done(struct request *clone, int error, bool mapped)
968{
969 int r = error;
970 struct dm_rq_target_io *tio = clone->end_io_data;
971 dm_request_endio_fn rq_end_io = NULL;
972
973 if (tio->ti) {
974 rq_end_io = tio->ti->type->rq_end_io;
975
976 if (mapped && rq_end_io)
977 r = rq_end_io(tio->ti, clone, error, &tio->info);
978 }
979
980 if (r <= 0)
981 /* The target wants to complete the I/O */
982 dm_end_request(clone, r);
983 else if (r == DM_ENDIO_INCOMPLETE)
984 /* The target will handle the I/O */
985 return;
986 else if (r == DM_ENDIO_REQUEUE)
987 /* The target wants to requeue the I/O */
988 dm_requeue_unmapped_request(clone);
989 else {
990 DMWARN("unimplemented target endio return value: %d", r);
991 BUG();
992 }
993}
994
995/*
996 * Request completion handler for request-based dm
997 */
998static void dm_softirq_done(struct request *rq)
999{
1000 bool mapped = true;
1001 struct request *clone = rq->completion_data;
1002 struct dm_rq_target_io *tio = clone->end_io_data;
1003
1004 if (rq->cmd_flags & REQ_FAILED)
1005 mapped = false;
1006
1007 dm_done(clone, tio->error, mapped);
1008}
1009
1010/*
1011 * Complete the clone and the original request with the error status
1012 * through softirq context.
1013 */
1014static void dm_complete_request(struct request *clone, int error)
1015{
1016 struct dm_rq_target_io *tio = clone->end_io_data;
1017 struct request *rq = tio->orig;
1018
1019 tio->error = error;
1020 rq->completion_data = clone;
1021 blk_complete_request(rq);
1022}
1023
1024/*
1025 * Complete the not-mapped clone and the original request with the error status
1026 * through softirq context.
1027 * Target's rq_end_io() function isn't called.
1028 * This may be used when the target's map_rq() function fails.
1029 */
1030void dm_kill_unmapped_request(struct request *clone, int error)
1031{
1032 struct dm_rq_target_io *tio = clone->end_io_data;
1033 struct request *rq = tio->orig;
1034
1035 rq->cmd_flags |= REQ_FAILED;
1036 dm_complete_request(clone, error);
1037}
1038EXPORT_SYMBOL_GPL(dm_kill_unmapped_request);
1039
1040/*
1041 * Called with the queue lock held
1042 */
1043static void end_clone_request(struct request *clone, int error)
1044{
1045 /*
1046 * For just cleaning up the information of the queue in which
1047 * the clone was dispatched.
1048 * The clone is *NOT* freed actually here because it is alloced from
1049 * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags.
1050 */
1051 __blk_put_request(clone->q, clone);
1052
1053 /*
1054 * Actual request completion is done in a softirq context which doesn't
1055 * hold the queue lock. Otherwise, deadlock could occur because:
1056 * - another request may be submitted by the upper level driver
1057 * of the stacking during the completion
1058 * - the submission which requires queue lock may be done
1059 * against this queue
1060 */
1061 dm_complete_request(clone, error);
1062}
1063
1064/*
1065 * Return maximum size of I/O possible at the supplied sector up to the current
1066 * target boundary.
1067 */
1068static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti)
1069{
1070 sector_t target_offset = dm_target_offset(ti, sector);
1071
1072 return ti->len - target_offset;
1073}
1074
1075static sector_t max_io_len(sector_t sector, struct dm_target *ti)
1076{
1077 sector_t len = max_io_len_target_boundary(sector, ti);
1078 sector_t offset, max_len;
1079
1080 /*
1081 * Does the target need to split even further?
1082 */
1083 if (ti->max_io_len) {
1084 offset = dm_target_offset(ti, sector);
1085 if (unlikely(ti->max_io_len & (ti->max_io_len - 1)))
1086 max_len = sector_div(offset, ti->max_io_len);
1087 else
1088 max_len = offset & (ti->max_io_len - 1);
1089 max_len = ti->max_io_len - max_len;
1090
1091 if (len > max_len)
1092 len = max_len;
1093 }
1094
1095 return len;
1096}
1097
1098int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
1099{
1100 if (len > UINT_MAX) {
1101 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
1102 (unsigned long long)len, UINT_MAX);
1103 ti->error = "Maximum size of target IO is too large";
1104 return -EINVAL;
1105 }
1106
1107 ti->max_io_len = (uint32_t) len;
1108
1109 return 0;
1110}
1111EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
1112
1113static void __map_bio(struct dm_target_io *tio)
1114{
1115 int r;
1116 sector_t sector;
1117 struct mapped_device *md;
1118 struct bio *clone = &tio->clone;
1119 struct dm_target *ti = tio->ti;
1120
1121 clone->bi_end_io = clone_endio;
1122
1123 /*
1124 * Map the clone. If r == 0 we don't need to do
1125 * anything, the target has assumed ownership of
1126 * this io.
1127 */
1128 atomic_inc(&tio->io->io_count);
1129 sector = clone->bi_iter.bi_sector;
1130 r = ti->type->map(ti, clone);
1131 if (r == DM_MAPIO_REMAPPED) {
1132 /* the bio has been remapped so dispatch it */
1133
1134 trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone,
1135 tio->io->bio->bi_bdev->bd_dev, sector);
1136
1137 generic_make_request(clone);
1138 } else if (r < 0 || r == DM_MAPIO_REQUEUE) {
1139 /* error the io and bail out, or requeue it if needed */
1140 md = tio->io->md;
1141 dec_pending(tio->io, r);
1142 free_tio(md, tio);
1143 } else if (r) {
1144 DMWARN("unimplemented target map return value: %d", r);
1145 BUG();
1146 }
1147}
1148
1149struct clone_info {
1150 struct mapped_device *md;
1151 struct dm_table *map;
1152 struct bio *bio;
1153 struct dm_io *io;
1154 sector_t sector;
1155 sector_t sector_count;
1156};
1157
1158static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len)
1159{
1160 bio->bi_iter.bi_sector = sector;
1161 bio->bi_iter.bi_size = to_bytes(len);
1162}
1163
1164/*
1165 * Creates a bio that consists of range of complete bvecs.
1166 */
1167static void clone_bio(struct dm_target_io *tio, struct bio *bio,
1168 sector_t sector, unsigned len)
1169{
1170 struct bio *clone = &tio->clone;
1171
1172 __bio_clone_fast(clone, bio);
1173
1174 if (bio_integrity(bio))
1175 bio_integrity_clone(clone, bio, GFP_NOIO);
1176
1177 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1178 clone->bi_iter.bi_size = to_bytes(len);
1179
1180 if (bio_integrity(bio))
1181 bio_integrity_trim(clone, 0, len);
1182}
1183
1184static struct dm_target_io *alloc_tio(struct clone_info *ci,
1185 struct dm_target *ti, int nr_iovecs,
1186 unsigned target_bio_nr)
1187{
1188 struct dm_target_io *tio;
1189 struct bio *clone;
1190
1191 clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs);
1192 tio = container_of(clone, struct dm_target_io, clone);
1193
1194 tio->io = ci->io;
1195 tio->ti = ti;
1196 tio->target_bio_nr = target_bio_nr;
1197
1198 return tio;
1199}
1200
1201static void __clone_and_map_simple_bio(struct clone_info *ci,
1202 struct dm_target *ti,
1203 unsigned target_bio_nr, sector_t len)
1204{
1205 struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr);
1206 struct bio *clone = &tio->clone;
1207
1208 /*
1209 * Discard requests require the bio's inline iovecs be initialized.
1210 * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush
1211 * and discard, so no need for concern about wasted bvec allocations.
1212 */
1213 __bio_clone_fast(clone, ci->bio);
1214 if (len)
1215 bio_setup_sector(clone, ci->sector, len);
1216
1217 __map_bio(tio);
1218}
1219
1220static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1221 unsigned num_bios, sector_t len)
1222{
1223 unsigned target_bio_nr;
1224
1225 for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++)
1226 __clone_and_map_simple_bio(ci, ti, target_bio_nr, len);
1227}
1228
1229static int __send_empty_flush(struct clone_info *ci)
1230{
1231 unsigned target_nr = 0;
1232 struct dm_target *ti;
1233
1234 BUG_ON(bio_has_data(ci->bio));
1235 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1236 __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0);
1237
1238 return 0;
1239}
1240
1241static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1242 sector_t sector, unsigned len)
1243{
1244 struct bio *bio = ci->bio;
1245 struct dm_target_io *tio;
1246 unsigned target_bio_nr;
1247 unsigned num_target_bios = 1;
1248
1249 /*
1250 * Does the target want to receive duplicate copies of the bio?
1251 */
1252 if (bio_data_dir(bio) == WRITE && ti->num_write_bios)
1253 num_target_bios = ti->num_write_bios(ti, bio);
1254
1255 for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) {
1256 tio = alloc_tio(ci, ti, 0, target_bio_nr);
1257 clone_bio(tio, bio, sector, len);
1258 __map_bio(tio);
1259 }
1260}
1261
1262typedef unsigned (*get_num_bios_fn)(struct dm_target *ti);
1263
1264static unsigned get_num_discard_bios(struct dm_target *ti)
1265{
1266 return ti->num_discard_bios;
1267}
1268
1269static unsigned get_num_write_same_bios(struct dm_target *ti)
1270{
1271 return ti->num_write_same_bios;
1272}
1273
1274typedef bool (*is_split_required_fn)(struct dm_target *ti);
1275
1276static bool is_split_required_for_discard(struct dm_target *ti)
1277{
1278 return ti->split_discard_bios;
1279}
1280
1281static int __send_changing_extent_only(struct clone_info *ci,
1282 get_num_bios_fn get_num_bios,
1283 is_split_required_fn is_split_required)
1284{
1285 struct dm_target *ti;
1286 sector_t len;
1287 unsigned num_bios;
1288
1289 do {
1290 ti = dm_table_find_target(ci->map, ci->sector);
1291 if (!dm_target_is_valid(ti))
1292 return -EIO;
1293
1294 /*
1295 * Even though the device advertised support for this type of
1296 * request, that does not mean every target supports it, and
1297 * reconfiguration might also have changed that since the
1298 * check was performed.
1299 */
1300 num_bios = get_num_bios ? get_num_bios(ti) : 0;
1301 if (!num_bios)
1302 return -EOPNOTSUPP;
1303
1304 if (is_split_required && !is_split_required(ti))
1305 len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti));
1306 else
1307 len = min(ci->sector_count, max_io_len(ci->sector, ti));
1308
1309 __send_duplicate_bios(ci, ti, num_bios, len);
1310
1311 ci->sector += len;
1312 } while (ci->sector_count -= len);
1313
1314 return 0;
1315}
1316
1317static int __send_discard(struct clone_info *ci)
1318{
1319 return __send_changing_extent_only(ci, get_num_discard_bios,
1320 is_split_required_for_discard);
1321}
1322
1323static int __send_write_same(struct clone_info *ci)
1324{
1325 return __send_changing_extent_only(ci, get_num_write_same_bios, NULL);
1326}
1327
1328/*
1329 * Select the correct strategy for processing a non-flush bio.
1330 */
1331static int __split_and_process_non_flush(struct clone_info *ci)
1332{
1333 struct bio *bio = ci->bio;
1334 struct dm_target *ti;
1335 unsigned len;
1336
1337 if (unlikely(bio->bi_rw & REQ_DISCARD))
1338 return __send_discard(ci);
1339 else if (unlikely(bio->bi_rw & REQ_WRITE_SAME))
1340 return __send_write_same(ci);
1341
1342 ti = dm_table_find_target(ci->map, ci->sector);
1343 if (!dm_target_is_valid(ti))
1344 return -EIO;
1345
1346 len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count);
1347
1348 __clone_and_map_data_bio(ci, ti, ci->sector, len);
1349
1350 ci->sector += len;
1351 ci->sector_count -= len;
1352
1353 return 0;
1354}
1355
1356/*
1357 * Entry point to split a bio into clones and submit them to the targets.
1358 */
1359static void __split_and_process_bio(struct mapped_device *md,
1360 struct dm_table *map, struct bio *bio)
1361{
1362 struct clone_info ci;
1363 int error = 0;
1364
1365 if (unlikely(!map)) {
1366 bio_io_error(bio);
1367 return;
1368 }
1369
1370 ci.map = map;
1371 ci.md = md;
1372 ci.io = alloc_io(md);
1373 ci.io->error = 0;
1374 atomic_set(&ci.io->io_count, 1);
1375 ci.io->bio = bio;
1376 ci.io->md = md;
1377 spin_lock_init(&ci.io->endio_lock);
1378 ci.sector = bio->bi_iter.bi_sector;
1379
1380 start_io_acct(ci.io);
1381
1382 if (bio->bi_rw & REQ_FLUSH) {
1383 ci.bio = &ci.md->flush_bio;
1384 ci.sector_count = 0;
1385 error = __send_empty_flush(&ci);
1386 /* dec_pending submits any data associated with flush */
1387 } else {
1388 ci.bio = bio;
1389 ci.sector_count = bio_sectors(bio);
1390 while (ci.sector_count && !error)
1391 error = __split_and_process_non_flush(&ci);
1392 }
1393
1394 /* drop the extra reference count */
1395 dec_pending(ci.io, error);
1396}
1397/*-----------------------------------------------------------------
1398 * CRUD END
1399 *---------------------------------------------------------------*/
1400
1401static int dm_merge_bvec(struct request_queue *q,
1402 struct bvec_merge_data *bvm,
1403 struct bio_vec *biovec)
1404{
1405 struct mapped_device *md = q->queuedata;
1406 struct dm_table *map = dm_get_live_table_fast(md);
1407 struct dm_target *ti;
1408 sector_t max_sectors;
1409 int max_size = 0;
1410
1411 if (unlikely(!map))
1412 goto out;
1413
1414 ti = dm_table_find_target(map, bvm->bi_sector);
1415 if (!dm_target_is_valid(ti))
1416 goto out;
1417
1418 /*
1419 * Find maximum amount of I/O that won't need splitting
1420 */
1421 max_sectors = min(max_io_len(bvm->bi_sector, ti),
1422 (sector_t) BIO_MAX_SECTORS);
1423 max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size;
1424 if (max_size < 0)
1425 max_size = 0;
1426
1427 /*
1428 * merge_bvec_fn() returns number of bytes
1429 * it can accept at this offset
1430 * max is precomputed maximal io size
1431 */
1432 if (max_size && ti->type->merge)
1433 max_size = ti->type->merge(ti, bvm, biovec, max_size);
1434 /*
1435 * If the target doesn't support merge method and some of the devices
1436 * provided their merge_bvec method (we know this by looking at
1437 * queue_max_hw_sectors), then we can't allow bios with multiple vector
1438 * entries. So always set max_size to 0, and the code below allows
1439 * just one page.
1440 */
1441 else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9)
1442
1443 max_size = 0;
1444
1445out:
1446 dm_put_live_table_fast(md);
1447 /*
1448 * Always allow an entire first page
1449 */
1450 if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT))
1451 max_size = biovec->bv_len;
1452
1453 return max_size;
1454}
1455
1456/*
1457 * The request function that just remaps the bio built up by
1458 * dm_merge_bvec.
1459 */
1460static void _dm_request(struct request_queue *q, struct bio *bio)
1461{
1462 int rw = bio_data_dir(bio);
1463 struct mapped_device *md = q->queuedata;
1464 int cpu;
1465 int srcu_idx;
1466 struct dm_table *map;
1467
1468 map = dm_get_live_table(md, &srcu_idx);
1469
1470 cpu = part_stat_lock();
1471 part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]);
1472 part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio));
1473 part_stat_unlock();
1474
1475 /* if we're suspended, we have to queue this io for later */
1476 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1477 dm_put_live_table(md, srcu_idx);
1478
1479 if (bio_rw(bio) != READA)
1480 queue_io(md, bio);
1481 else
1482 bio_io_error(bio);
1483 return;
1484 }
1485
1486 __split_and_process_bio(md, map, bio);
1487 dm_put_live_table(md, srcu_idx);
1488 return;
1489}
1490
1491int dm_request_based(struct mapped_device *md)
1492{
1493 return blk_queue_stackable(md->queue);
1494}
1495
1496static void dm_request(struct request_queue *q, struct bio *bio)
1497{
1498 struct mapped_device *md = q->queuedata;
1499
1500 if (dm_request_based(md))
1501 blk_queue_bio(q, bio);
1502 else
1503 _dm_request(q, bio);
1504}
1505
1506void dm_dispatch_request(struct request *rq)
1507{
1508 int r;
1509
1510 if (blk_queue_io_stat(rq->q))
1511 rq->cmd_flags |= REQ_IO_STAT;
1512
1513 rq->start_time = jiffies;
1514 r = blk_insert_cloned_request(rq->q, rq);
1515 if (r)
1516 dm_complete_request(rq, r);
1517}
1518EXPORT_SYMBOL_GPL(dm_dispatch_request);
1519
1520static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig,
1521 void *data)
1522{
1523 struct dm_rq_target_io *tio = data;
1524 struct dm_rq_clone_bio_info *info =
1525 container_of(bio, struct dm_rq_clone_bio_info, clone);
1526
1527 info->orig = bio_orig;
1528 info->tio = tio;
1529 bio->bi_end_io = end_clone_bio;
1530
1531 return 0;
1532}
1533
1534static int setup_clone(struct request *clone, struct request *rq,
1535 struct dm_rq_target_io *tio)
1536{
1537 int r;
1538
1539 r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC,
1540 dm_rq_bio_constructor, tio);
1541 if (r)
1542 return r;
1543
1544 clone->cmd = rq->cmd;
1545 clone->cmd_len = rq->cmd_len;
1546 clone->sense = rq->sense;
1547 clone->buffer = rq->buffer;
1548 clone->end_io = end_clone_request;
1549 clone->end_io_data = tio;
1550
1551 return 0;
1552}
1553
1554static struct request *clone_rq(struct request *rq, struct mapped_device *md,
1555 gfp_t gfp_mask)
1556{
1557 struct request *clone;
1558 struct dm_rq_target_io *tio;
1559
1560 tio = alloc_rq_tio(md, gfp_mask);
1561 if (!tio)
1562 return NULL;
1563
1564 tio->md = md;
1565 tio->ti = NULL;
1566 tio->orig = rq;
1567 tio->error = 0;
1568 memset(&tio->info, 0, sizeof(tio->info));
1569
1570 clone = &tio->clone;
1571 if (setup_clone(clone, rq, tio)) {
1572 /* -ENOMEM */
1573 free_rq_tio(tio);
1574 return NULL;
1575 }
1576
1577 return clone;
1578}
1579
1580/*
1581 * Called with the queue lock held.
1582 */
1583static int dm_prep_fn(struct request_queue *q, struct request *rq)
1584{
1585 struct mapped_device *md = q->queuedata;
1586 struct request *clone;
1587
1588 if (unlikely(rq->special)) {
1589 DMWARN("Already has something in rq->special.");
1590 return BLKPREP_KILL;
1591 }
1592
1593 clone = clone_rq(rq, md, GFP_ATOMIC);
1594 if (!clone)
1595 return BLKPREP_DEFER;
1596
1597 rq->special = clone;
1598 rq->cmd_flags |= REQ_DONTPREP;
1599
1600 return BLKPREP_OK;
1601}
1602
1603/*
1604 * Returns:
1605 * 0 : the request has been processed (not requeued)
1606 * !0 : the request has been requeued
1607 */
1608static int map_request(struct dm_target *ti, struct request *clone,
1609 struct mapped_device *md)
1610{
1611 int r, requeued = 0;
1612 struct dm_rq_target_io *tio = clone->end_io_data;
1613
1614 tio->ti = ti;
1615 r = ti->type->map_rq(ti, clone, &tio->info);
1616 switch (r) {
1617 case DM_MAPIO_SUBMITTED:
1618 /* The target has taken the I/O to submit by itself later */
1619 break;
1620 case DM_MAPIO_REMAPPED:
1621 /* The target has remapped the I/O so dispatch it */
1622 trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)),
1623 blk_rq_pos(tio->orig));
1624 dm_dispatch_request(clone);
1625 break;
1626 case DM_MAPIO_REQUEUE:
1627 /* The target wants to requeue the I/O */
1628 dm_requeue_unmapped_request(clone);
1629 requeued = 1;
1630 break;
1631 default:
1632 if (r > 0) {
1633 DMWARN("unimplemented target map return value: %d", r);
1634 BUG();
1635 }
1636
1637 /* The target wants to complete the I/O */
1638 dm_kill_unmapped_request(clone, r);
1639 break;
1640 }
1641
1642 return requeued;
1643}
1644
1645static struct request *dm_start_request(struct mapped_device *md, struct request *orig)
1646{
1647 struct request *clone;
1648
1649 blk_start_request(orig);
1650 clone = orig->special;
1651 atomic_inc(&md->pending[rq_data_dir(clone)]);
1652
1653 /*
1654 * Hold the md reference here for the in-flight I/O.
1655 * We can't rely on the reference count by device opener,
1656 * because the device may be closed during the request completion
1657 * when all bios are completed.
1658 * See the comment in rq_completed() too.
1659 */
1660 dm_get(md);
1661
1662 return clone;
1663}
1664
1665/*
1666 * q->request_fn for request-based dm.
1667 * Called with the queue lock held.
1668 */
1669static void dm_request_fn(struct request_queue *q)
1670{
1671 struct mapped_device *md = q->queuedata;
1672 int srcu_idx;
1673 struct dm_table *map = dm_get_live_table(md, &srcu_idx);
1674 struct dm_target *ti;
1675 struct request *rq, *clone;
1676 sector_t pos;
1677
1678 /*
1679 * For suspend, check blk_queue_stopped() and increment
1680 * ->pending within a single queue_lock not to increment the
1681 * number of in-flight I/Os after the queue is stopped in
1682 * dm_suspend().
1683 */
1684 while (!blk_queue_stopped(q)) {
1685 rq = blk_peek_request(q);
1686 if (!rq)
1687 goto delay_and_out;
1688
1689 /* always use block 0 to find the target for flushes for now */
1690 pos = 0;
1691 if (!(rq->cmd_flags & REQ_FLUSH))
1692 pos = blk_rq_pos(rq);
1693
1694 ti = dm_table_find_target(map, pos);
1695 if (!dm_target_is_valid(ti)) {
1696 /*
1697 * Must perform setup, that dm_done() requires,
1698 * before calling dm_kill_unmapped_request
1699 */
1700 DMERR_LIMIT("request attempted access beyond the end of device");
1701 clone = dm_start_request(md, rq);
1702 dm_kill_unmapped_request(clone, -EIO);
1703 continue;
1704 }
1705
1706 if (ti->type->busy && ti->type->busy(ti))
1707 goto delay_and_out;
1708
1709 clone = dm_start_request(md, rq);
1710
1711 spin_unlock(q->queue_lock);
1712 if (map_request(ti, clone, md))
1713 goto requeued;
1714
1715 BUG_ON(!irqs_disabled());
1716 spin_lock(q->queue_lock);
1717 }
1718
1719 goto out;
1720
1721requeued:
1722 BUG_ON(!irqs_disabled());
1723 spin_lock(q->queue_lock);
1724
1725delay_and_out:
1726 blk_delay_queue(q, HZ / 10);
1727out:
1728 dm_put_live_table(md, srcu_idx);
1729}
1730
1731int dm_underlying_device_busy(struct request_queue *q)
1732{
1733 return blk_lld_busy(q);
1734}
1735EXPORT_SYMBOL_GPL(dm_underlying_device_busy);
1736
1737static int dm_lld_busy(struct request_queue *q)
1738{
1739 int r;
1740 struct mapped_device *md = q->queuedata;
1741 struct dm_table *map = dm_get_live_table_fast(md);
1742
1743 if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))
1744 r = 1;
1745 else
1746 r = dm_table_any_busy_target(map);
1747
1748 dm_put_live_table_fast(md);
1749
1750 return r;
1751}
1752
1753static int dm_any_congested(void *congested_data, int bdi_bits)
1754{
1755 int r = bdi_bits;
1756 struct mapped_device *md = congested_data;
1757 struct dm_table *map;
1758
1759 if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
1760 map = dm_get_live_table_fast(md);
1761 if (map) {
1762 /*
1763 * Request-based dm cares about only own queue for
1764 * the query about congestion status of request_queue
1765 */
1766 if (dm_request_based(md))
1767 r = md->queue->backing_dev_info.state &
1768 bdi_bits;
1769 else
1770 r = dm_table_any_congested(map, bdi_bits);
1771 }
1772 dm_put_live_table_fast(md);
1773 }
1774
1775 return r;
1776}
1777
1778/*-----------------------------------------------------------------
1779 * An IDR is used to keep track of allocated minor numbers.
1780 *---------------------------------------------------------------*/
1781static void free_minor(int minor)
1782{
1783 spin_lock(&_minor_lock);
1784 idr_remove(&_minor_idr, minor);
1785 spin_unlock(&_minor_lock);
1786}
1787
1788/*
1789 * See if the device with a specific minor # is free.
1790 */
1791static int specific_minor(int minor)
1792{
1793 int r;
1794
1795 if (minor >= (1 << MINORBITS))
1796 return -EINVAL;
1797
1798 idr_preload(GFP_KERNEL);
1799 spin_lock(&_minor_lock);
1800
1801 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1802
1803 spin_unlock(&_minor_lock);
1804 idr_preload_end();
1805 if (r < 0)
1806 return r == -ENOSPC ? -EBUSY : r;
1807 return 0;
1808}
1809
1810static int next_free_minor(int *minor)
1811{
1812 int r;
1813
1814 idr_preload(GFP_KERNEL);
1815 spin_lock(&_minor_lock);
1816
1817 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1818
1819 spin_unlock(&_minor_lock);
1820 idr_preload_end();
1821 if (r < 0)
1822 return r;
1823 *minor = r;
1824 return 0;
1825}
1826
1827static const struct block_device_operations dm_blk_dops;
1828
1829static void dm_wq_work(struct work_struct *work);
1830
1831static void dm_init_md_queue(struct mapped_device *md)
1832{
1833 /*
1834 * Request-based dm devices cannot be stacked on top of bio-based dm
1835 * devices. The type of this dm device has not been decided yet.
1836 * The type is decided at the first table loading time.
1837 * To prevent problematic device stacking, clear the queue flag
1838 * for request stacking support until then.
1839 *
1840 * This queue is new, so no concurrency on the queue_flags.
1841 */
1842 queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue);
1843
1844 md->queue->queuedata = md;
1845 md->queue->backing_dev_info.congested_fn = dm_any_congested;
1846 md->queue->backing_dev_info.congested_data = md;
1847 blk_queue_make_request(md->queue, dm_request);
1848 blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY);
1849 blk_queue_merge_bvec(md->queue, dm_merge_bvec);
1850}
1851
1852/*
1853 * Allocate and initialise a blank device with a given minor.
1854 */
1855static struct mapped_device *alloc_dev(int minor)
1856{
1857 int r;
1858 struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL);
1859 void *old_md;
1860
1861 if (!md) {
1862 DMWARN("unable to allocate device, out of memory.");
1863 return NULL;
1864 }
1865
1866 if (!try_module_get(THIS_MODULE))
1867 goto bad_module_get;
1868
1869 /* get a minor number for the dev */
1870 if (minor == DM_ANY_MINOR)
1871 r = next_free_minor(&minor);
1872 else
1873 r = specific_minor(minor);
1874 if (r < 0)
1875 goto bad_minor;
1876
1877 r = init_srcu_struct(&md->io_barrier);
1878 if (r < 0)
1879 goto bad_io_barrier;
1880
1881 md->type = DM_TYPE_NONE;
1882 mutex_init(&md->suspend_lock);
1883 mutex_init(&md->type_lock);
1884 spin_lock_init(&md->deferred_lock);
1885 atomic_set(&md->holders, 1);
1886 atomic_set(&md->open_count, 0);
1887 atomic_set(&md->event_nr, 0);
1888 atomic_set(&md->uevent_seq, 0);
1889 INIT_LIST_HEAD(&md->uevent_list);
1890 spin_lock_init(&md->uevent_lock);
1891
1892 md->queue = blk_alloc_queue(GFP_KERNEL);
1893 if (!md->queue)
1894 goto bad_queue;
1895
1896 dm_init_md_queue(md);
1897
1898 md->disk = alloc_disk(1);
1899 if (!md->disk)
1900 goto bad_disk;
1901
1902 atomic_set(&md->pending[0], 0);
1903 atomic_set(&md->pending[1], 0);
1904 init_waitqueue_head(&md->wait);
1905 INIT_WORK(&md->work, dm_wq_work);
1906 init_waitqueue_head(&md->eventq);
1907 init_completion(&md->kobj_holder.completion);
1908
1909 md->disk->major = _major;
1910 md->disk->first_minor = minor;
1911 md->disk->fops = &dm_blk_dops;
1912 md->disk->queue = md->queue;
1913 md->disk->private_data = md;
1914 sprintf(md->disk->disk_name, "dm-%d", minor);
1915 add_disk(md->disk);
1916 format_dev_t(md->name, MKDEV(_major, minor));
1917
1918 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1919 if (!md->wq)
1920 goto bad_thread;
1921
1922 md->bdev = bdget_disk(md->disk, 0);
1923 if (!md->bdev)
1924 goto bad_bdev;
1925
1926 bio_init(&md->flush_bio);
1927 md->flush_bio.bi_bdev = md->bdev;
1928 md->flush_bio.bi_rw = WRITE_FLUSH;
1929
1930 dm_stats_init(&md->stats);
1931
1932 /* Populate the mapping, nobody knows we exist yet */
1933 spin_lock(&_minor_lock);
1934 old_md = idr_replace(&_minor_idr, md, minor);
1935 spin_unlock(&_minor_lock);
1936
1937 BUG_ON(old_md != MINOR_ALLOCED);
1938
1939 return md;
1940
1941bad_bdev:
1942 destroy_workqueue(md->wq);
1943bad_thread:
1944 del_gendisk(md->disk);
1945 put_disk(md->disk);
1946bad_disk:
1947 blk_cleanup_queue(md->queue);
1948bad_queue:
1949 cleanup_srcu_struct(&md->io_barrier);
1950bad_io_barrier:
1951 free_minor(minor);
1952bad_minor:
1953 module_put(THIS_MODULE);
1954bad_module_get:
1955 kfree(md);
1956 return NULL;
1957}
1958
1959static void unlock_fs(struct mapped_device *md);
1960
1961static void free_dev(struct mapped_device *md)
1962{
1963 int minor = MINOR(disk_devt(md->disk));
1964
1965 unlock_fs(md);
1966 bdput(md->bdev);
1967 destroy_workqueue(md->wq);
1968 if (md->io_pool)
1969 mempool_destroy(md->io_pool);
1970 if (md->bs)
1971 bioset_free(md->bs);
1972 blk_integrity_unregister(md->disk);
1973 del_gendisk(md->disk);
1974 cleanup_srcu_struct(&md->io_barrier);
1975 free_minor(minor);
1976
1977 spin_lock(&_minor_lock);
1978 md->disk->private_data = NULL;
1979 spin_unlock(&_minor_lock);
1980
1981 put_disk(md->disk);
1982 blk_cleanup_queue(md->queue);
1983 dm_stats_cleanup(&md->stats);
1984 module_put(THIS_MODULE);
1985 kfree(md);
1986}
1987
1988static void __bind_mempools(struct mapped_device *md, struct dm_table *t)
1989{
1990 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1991
1992 if (md->io_pool && md->bs) {
1993 /* The md already has necessary mempools. */
1994 if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) {
1995 /*
1996 * Reload bioset because front_pad may have changed
1997 * because a different table was loaded.
1998 */
1999 bioset_free(md->bs);
2000 md->bs = p->bs;
2001 p->bs = NULL;
2002 } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) {
2003 /*
2004 * There's no need to reload with request-based dm
2005 * because the size of front_pad doesn't change.
2006 * Note for future: If you are to reload bioset,
2007 * prep-ed requests in the queue may refer
2008 * to bio from the old bioset, so you must walk
2009 * through the queue to unprep.
2010 */
2011 }
2012 goto out;
2013 }
2014
2015 BUG_ON(!p || md->io_pool || md->bs);
2016
2017 md->io_pool = p->io_pool;
2018 p->io_pool = NULL;
2019 md->bs = p->bs;
2020 p->bs = NULL;
2021
2022out:
2023 /* mempool bind completed, now no need any mempools in the table */
2024 dm_table_free_md_mempools(t);
2025}
2026
2027/*
2028 * Bind a table to the device.
2029 */
2030static void event_callback(void *context)
2031{
2032 unsigned long flags;
2033 LIST_HEAD(uevents);
2034 struct mapped_device *md = (struct mapped_device *) context;
2035
2036 spin_lock_irqsave(&md->uevent_lock, flags);
2037 list_splice_init(&md->uevent_list, &uevents);
2038 spin_unlock_irqrestore(&md->uevent_lock, flags);
2039
2040 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
2041
2042 atomic_inc(&md->event_nr);
2043 wake_up(&md->eventq);
2044}
2045
2046/*
2047 * Protected by md->suspend_lock obtained by dm_swap_table().
2048 */
2049static void __set_size(struct mapped_device *md, sector_t size)
2050{
2051 set_capacity(md->disk, size);
2052
2053 i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
2054}
2055
2056/*
2057 * Return 1 if the queue has a compulsory merge_bvec_fn function.
2058 *
2059 * If this function returns 0, then the device is either a non-dm
2060 * device without a merge_bvec_fn, or it is a dm device that is
2061 * able to split any bios it receives that are too big.
2062 */
2063int dm_queue_merge_is_compulsory(struct request_queue *q)
2064{
2065 struct mapped_device *dev_md;
2066
2067 if (!q->merge_bvec_fn)
2068 return 0;
2069
2070 if (q->make_request_fn == dm_request) {
2071 dev_md = q->queuedata;
2072 if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags))
2073 return 0;
2074 }
2075
2076 return 1;
2077}
2078
2079static int dm_device_merge_is_compulsory(struct dm_target *ti,
2080 struct dm_dev *dev, sector_t start,
2081 sector_t len, void *data)
2082{
2083 struct block_device *bdev = dev->bdev;
2084 struct request_queue *q = bdev_get_queue(bdev);
2085
2086 return dm_queue_merge_is_compulsory(q);
2087}
2088
2089/*
2090 * Return 1 if it is acceptable to ignore merge_bvec_fn based
2091 * on the properties of the underlying devices.
2092 */
2093static int dm_table_merge_is_optional(struct dm_table *table)
2094{
2095 unsigned i = 0;
2096 struct dm_target *ti;
2097
2098 while (i < dm_table_get_num_targets(table)) {
2099 ti = dm_table_get_target(table, i++);
2100
2101 if (ti->type->iterate_devices &&
2102 ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL))
2103 return 0;
2104 }
2105
2106 return 1;
2107}
2108
2109/*
2110 * Returns old map, which caller must destroy.
2111 */
2112static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
2113 struct queue_limits *limits)
2114{
2115 struct dm_table *old_map;
2116 struct request_queue *q = md->queue;
2117 sector_t size;
2118 int merge_is_optional;
2119
2120 size = dm_table_get_size(t);
2121
2122 /*
2123 * Wipe any geometry if the size of the table changed.
2124 */
2125 if (size != dm_get_size(md))
2126 memset(&md->geometry, 0, sizeof(md->geometry));
2127
2128 __set_size(md, size);
2129
2130 dm_table_event_callback(t, event_callback, md);
2131
2132 /*
2133 * The queue hasn't been stopped yet, if the old table type wasn't
2134 * for request-based during suspension. So stop it to prevent
2135 * I/O mapping before resume.
2136 * This must be done before setting the queue restrictions,
2137 * because request-based dm may be run just after the setting.
2138 */
2139 if (dm_table_request_based(t) && !blk_queue_stopped(q))
2140 stop_queue(q);
2141
2142 __bind_mempools(md, t);
2143
2144 merge_is_optional = dm_table_merge_is_optional(t);
2145
2146 old_map = md->map;
2147 rcu_assign_pointer(md->map, t);
2148 md->immutable_target_type = dm_table_get_immutable_target_type(t);
2149
2150 dm_table_set_restrictions(t, q, limits);
2151 if (merge_is_optional)
2152 set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2153 else
2154 clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags);
2155 dm_sync_table(md);
2156
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 = md->map;
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 struct mapped_device *md;
2183
2184 md = alloc_dev(minor);
2185 if (!md)
2186 return -ENXIO;
2187
2188 dm_sysfs_init(md);
2189
2190 *result = md;
2191 return 0;
2192}
2193
2194/*
2195 * Functions to manage md->type.
2196 * All are required to hold md->type_lock.
2197 */
2198void dm_lock_md_type(struct mapped_device *md)
2199{
2200 mutex_lock(&md->type_lock);
2201}
2202
2203void dm_unlock_md_type(struct mapped_device *md)
2204{
2205 mutex_unlock(&md->type_lock);
2206}
2207
2208void dm_set_md_type(struct mapped_device *md, unsigned type)
2209{
2210 BUG_ON(!mutex_is_locked(&md->type_lock));
2211 md->type = type;
2212}
2213
2214unsigned dm_get_md_type(struct mapped_device *md)
2215{
2216 BUG_ON(!mutex_is_locked(&md->type_lock));
2217 return md->type;
2218}
2219
2220struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2221{
2222 return md->immutable_target_type;
2223}
2224
2225/*
2226 * The queue_limits are only valid as long as you have a reference
2227 * count on 'md'.
2228 */
2229struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2230{
2231 BUG_ON(!atomic_read(&md->holders));
2232 return &md->queue->limits;
2233}
2234EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2235
2236/*
2237 * Fully initialize a request-based queue (->elevator, ->request_fn, etc).
2238 */
2239static int dm_init_request_based_queue(struct mapped_device *md)
2240{
2241 struct request_queue *q = NULL;
2242
2243 if (md->queue->elevator)
2244 return 1;
2245
2246 /* Fully initialize the queue */
2247 q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL);
2248 if (!q)
2249 return 0;
2250
2251 md->queue = q;
2252 dm_init_md_queue(md);
2253 blk_queue_softirq_done(md->queue, dm_softirq_done);
2254 blk_queue_prep_rq(md->queue, dm_prep_fn);
2255 blk_queue_lld_busy(md->queue, dm_lld_busy);
2256
2257 elv_register_queue(md->queue);
2258
2259 return 1;
2260}
2261
2262/*
2263 * Setup the DM device's queue based on md's type
2264 */
2265int dm_setup_md_queue(struct mapped_device *md)
2266{
2267 if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) &&
2268 !dm_init_request_based_queue(md)) {
2269 DMWARN("Cannot initialize queue for request-based mapped device");
2270 return -EINVAL;
2271 }
2272
2273 return 0;
2274}
2275
2276static struct mapped_device *dm_find_md(dev_t dev)
2277{
2278 struct mapped_device *md;
2279 unsigned minor = MINOR(dev);
2280
2281 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2282 return NULL;
2283
2284 spin_lock(&_minor_lock);
2285
2286 md = idr_find(&_minor_idr, minor);
2287 if (md && (md == MINOR_ALLOCED ||
2288 (MINOR(disk_devt(dm_disk(md))) != minor) ||
2289 dm_deleting_md(md) ||
2290 test_bit(DMF_FREEING, &md->flags))) {
2291 md = NULL;
2292 goto out;
2293 }
2294
2295out:
2296 spin_unlock(&_minor_lock);
2297
2298 return md;
2299}
2300
2301struct mapped_device *dm_get_md(dev_t dev)
2302{
2303 struct mapped_device *md = dm_find_md(dev);
2304
2305 if (md)
2306 dm_get(md);
2307
2308 return md;
2309}
2310EXPORT_SYMBOL_GPL(dm_get_md);
2311
2312void *dm_get_mdptr(struct mapped_device *md)
2313{
2314 return md->interface_ptr;
2315}
2316
2317void dm_set_mdptr(struct mapped_device *md, void *ptr)
2318{
2319 md->interface_ptr = ptr;
2320}
2321
2322void dm_get(struct mapped_device *md)
2323{
2324 atomic_inc(&md->holders);
2325 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2326}
2327
2328const char *dm_device_name(struct mapped_device *md)
2329{
2330 return md->name;
2331}
2332EXPORT_SYMBOL_GPL(dm_device_name);
2333
2334static void __dm_destroy(struct mapped_device *md, bool wait)
2335{
2336 struct dm_table *map;
2337 int srcu_idx;
2338
2339 might_sleep();
2340
2341 spin_lock(&_minor_lock);
2342 map = dm_get_live_table(md, &srcu_idx);
2343 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2344 set_bit(DMF_FREEING, &md->flags);
2345 spin_unlock(&_minor_lock);
2346
2347 if (!dm_suspended_md(md)) {
2348 dm_table_presuspend_targets(map);
2349 dm_table_postsuspend_targets(map);
2350 }
2351
2352 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2353 dm_put_live_table(md, srcu_idx);
2354
2355 /*
2356 * Rare, but there may be I/O requests still going to complete,
2357 * for example. Wait for all references to disappear.
2358 * No one should increment the reference count of the mapped_device,
2359 * after the mapped_device state becomes DMF_FREEING.
2360 */
2361 if (wait)
2362 while (atomic_read(&md->holders))
2363 msleep(1);
2364 else if (atomic_read(&md->holders))
2365 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2366 dm_device_name(md), atomic_read(&md->holders));
2367
2368 dm_sysfs_exit(md);
2369 dm_table_destroy(__unbind(md));
2370 free_dev(md);
2371}
2372
2373void dm_destroy(struct mapped_device *md)
2374{
2375 __dm_destroy(md, true);
2376}
2377
2378void dm_destroy_immediate(struct mapped_device *md)
2379{
2380 __dm_destroy(md, false);
2381}
2382
2383void dm_put(struct mapped_device *md)
2384{
2385 atomic_dec(&md->holders);
2386}
2387EXPORT_SYMBOL_GPL(dm_put);
2388
2389static int dm_wait_for_completion(struct mapped_device *md, int interruptible)
2390{
2391 int r = 0;
2392 DECLARE_WAITQUEUE(wait, current);
2393
2394 add_wait_queue(&md->wait, &wait);
2395
2396 while (1) {
2397 set_current_state(interruptible);
2398
2399 if (!md_in_flight(md))
2400 break;
2401
2402 if (interruptible == TASK_INTERRUPTIBLE &&
2403 signal_pending(current)) {
2404 r = -EINTR;
2405 break;
2406 }
2407
2408 io_schedule();
2409 }
2410 set_current_state(TASK_RUNNING);
2411
2412 remove_wait_queue(&md->wait, &wait);
2413
2414 return r;
2415}
2416
2417/*
2418 * Process the deferred bios
2419 */
2420static void dm_wq_work(struct work_struct *work)
2421{
2422 struct mapped_device *md = container_of(work, struct mapped_device,
2423 work);
2424 struct bio *c;
2425 int srcu_idx;
2426 struct dm_table *map;
2427
2428 map = dm_get_live_table(md, &srcu_idx);
2429
2430 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2431 spin_lock_irq(&md->deferred_lock);
2432 c = bio_list_pop(&md->deferred);
2433 spin_unlock_irq(&md->deferred_lock);
2434
2435 if (!c)
2436 break;
2437
2438 if (dm_request_based(md))
2439 generic_make_request(c);
2440 else
2441 __split_and_process_bio(md, map, c);
2442 }
2443
2444 dm_put_live_table(md, srcu_idx);
2445}
2446
2447static void dm_queue_flush(struct mapped_device *md)
2448{
2449 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2450 smp_mb__after_clear_bit();
2451 queue_work(md->wq, &md->work);
2452}
2453
2454/*
2455 * Swap in a new table, returning the old one for the caller to destroy.
2456 */
2457struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2458{
2459 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2460 struct queue_limits limits;
2461 int r;
2462
2463 mutex_lock(&md->suspend_lock);
2464
2465 /* device must be suspended */
2466 if (!dm_suspended_md(md))
2467 goto out;
2468
2469 /*
2470 * If the new table has no data devices, retain the existing limits.
2471 * This helps multipath with queue_if_no_path if all paths disappear,
2472 * then new I/O is queued based on these limits, and then some paths
2473 * reappear.
2474 */
2475 if (dm_table_has_no_data_devices(table)) {
2476 live_map = dm_get_live_table_fast(md);
2477 if (live_map)
2478 limits = md->queue->limits;
2479 dm_put_live_table_fast(md);
2480 }
2481
2482 if (!live_map) {
2483 r = dm_calculate_queue_limits(table, &limits);
2484 if (r) {
2485 map = ERR_PTR(r);
2486 goto out;
2487 }
2488 }
2489
2490 map = __bind(md, table, &limits);
2491
2492out:
2493 mutex_unlock(&md->suspend_lock);
2494 return map;
2495}
2496
2497/*
2498 * Functions to lock and unlock any filesystem running on the
2499 * device.
2500 */
2501static int lock_fs(struct mapped_device *md)
2502{
2503 int r;
2504
2505 WARN_ON(md->frozen_sb);
2506
2507 md->frozen_sb = freeze_bdev(md->bdev);
2508 if (IS_ERR(md->frozen_sb)) {
2509 r = PTR_ERR(md->frozen_sb);
2510 md->frozen_sb = NULL;
2511 return r;
2512 }
2513
2514 set_bit(DMF_FROZEN, &md->flags);
2515
2516 return 0;
2517}
2518
2519static void unlock_fs(struct mapped_device *md)
2520{
2521 if (!test_bit(DMF_FROZEN, &md->flags))
2522 return;
2523
2524 thaw_bdev(md->bdev, md->frozen_sb);
2525 md->frozen_sb = NULL;
2526 clear_bit(DMF_FROZEN, &md->flags);
2527}
2528
2529/*
2530 * We need to be able to change a mapping table under a mounted
2531 * filesystem. For example we might want to move some data in
2532 * the background. Before the table can be swapped with
2533 * dm_bind_table, dm_suspend must be called to flush any in
2534 * flight bios and ensure that any further io gets deferred.
2535 */
2536/*
2537 * Suspend mechanism in request-based dm.
2538 *
2539 * 1. Flush all I/Os by lock_fs() if needed.
2540 * 2. Stop dispatching any I/O by stopping the request_queue.
2541 * 3. Wait for all in-flight I/Os to be completed or requeued.
2542 *
2543 * To abort suspend, start the request_queue.
2544 */
2545int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2546{
2547 struct dm_table *map = NULL;
2548 int r = 0;
2549 int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0;
2550 int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0;
2551
2552 mutex_lock(&md->suspend_lock);
2553
2554 if (dm_suspended_md(md)) {
2555 r = -EINVAL;
2556 goto out_unlock;
2557 }
2558
2559 map = md->map;
2560
2561 /*
2562 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2563 * This flag is cleared before dm_suspend returns.
2564 */
2565 if (noflush)
2566 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2567
2568 /* This does not get reverted if there's an error later. */
2569 dm_table_presuspend_targets(map);
2570
2571 /*
2572 * Flush I/O to the device.
2573 * Any I/O submitted after lock_fs() may not be flushed.
2574 * noflush takes precedence over do_lockfs.
2575 * (lock_fs() flushes I/Os and waits for them to complete.)
2576 */
2577 if (!noflush && do_lockfs) {
2578 r = lock_fs(md);
2579 if (r)
2580 goto out_unlock;
2581 }
2582
2583 /*
2584 * Here we must make sure that no processes are submitting requests
2585 * to target drivers i.e. no one may be executing
2586 * __split_and_process_bio. This is called from dm_request and
2587 * dm_wq_work.
2588 *
2589 * To get all processes out of __split_and_process_bio in dm_request,
2590 * we take the write lock. To prevent any process from reentering
2591 * __split_and_process_bio from dm_request and quiesce the thread
2592 * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call
2593 * flush_workqueue(md->wq).
2594 */
2595 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2596 synchronize_srcu(&md->io_barrier);
2597
2598 /*
2599 * Stop md->queue before flushing md->wq in case request-based
2600 * dm defers requests to md->wq from md->queue.
2601 */
2602 if (dm_request_based(md))
2603 stop_queue(md->queue);
2604
2605 flush_workqueue(md->wq);
2606
2607 /*
2608 * At this point no more requests are entering target request routines.
2609 * We call dm_wait_for_completion to wait for all existing requests
2610 * to finish.
2611 */
2612 r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE);
2613
2614 if (noflush)
2615 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2616 synchronize_srcu(&md->io_barrier);
2617
2618 /* were we interrupted ? */
2619 if (r < 0) {
2620 dm_queue_flush(md);
2621
2622 if (dm_request_based(md))
2623 start_queue(md->queue);
2624
2625 unlock_fs(md);
2626 goto out_unlock; /* pushback list is already flushed, so skip flush */
2627 }
2628
2629 /*
2630 * If dm_wait_for_completion returned 0, the device is completely
2631 * quiescent now. There is no request-processing activity. All new
2632 * requests are being added to md->deferred list.
2633 */
2634
2635 set_bit(DMF_SUSPENDED, &md->flags);
2636
2637 dm_table_postsuspend_targets(map);
2638
2639out_unlock:
2640 mutex_unlock(&md->suspend_lock);
2641 return r;
2642}
2643
2644int dm_resume(struct mapped_device *md)
2645{
2646 int r = -EINVAL;
2647 struct dm_table *map = NULL;
2648
2649 mutex_lock(&md->suspend_lock);
2650 if (!dm_suspended_md(md))
2651 goto out;
2652
2653 map = md->map;
2654 if (!map || !dm_table_get_size(map))
2655 goto out;
2656
2657 r = dm_table_resume_targets(map);
2658 if (r)
2659 goto out;
2660
2661 dm_queue_flush(md);
2662
2663 /*
2664 * Flushing deferred I/Os must be done after targets are resumed
2665 * so that mapping of targets can work correctly.
2666 * Request-based dm is queueing the deferred I/Os in its request_queue.
2667 */
2668 if (dm_request_based(md))
2669 start_queue(md->queue);
2670
2671 unlock_fs(md);
2672
2673 clear_bit(DMF_SUSPENDED, &md->flags);
2674
2675 r = 0;
2676out:
2677 mutex_unlock(&md->suspend_lock);
2678
2679 return r;
2680}
2681
2682/*
2683 * Internal suspend/resume works like userspace-driven suspend. It waits
2684 * until all bios finish and prevents issuing new bios to the target drivers.
2685 * It may be used only from the kernel.
2686 *
2687 * Internal suspend holds md->suspend_lock, which prevents interaction with
2688 * userspace-driven suspend.
2689 */
2690
2691void dm_internal_suspend(struct mapped_device *md)
2692{
2693 mutex_lock(&md->suspend_lock);
2694 if (dm_suspended_md(md))
2695 return;
2696
2697 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2698 synchronize_srcu(&md->io_barrier);
2699 flush_workqueue(md->wq);
2700 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2701}
2702
2703void dm_internal_resume(struct mapped_device *md)
2704{
2705 if (dm_suspended_md(md))
2706 goto done;
2707
2708 dm_queue_flush(md);
2709
2710done:
2711 mutex_unlock(&md->suspend_lock);
2712}
2713
2714/*-----------------------------------------------------------------
2715 * Event notification.
2716 *---------------------------------------------------------------*/
2717int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2718 unsigned cookie)
2719{
2720 char udev_cookie[DM_COOKIE_LENGTH];
2721 char *envp[] = { udev_cookie, NULL };
2722
2723 if (!cookie)
2724 return kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2725 else {
2726 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2727 DM_COOKIE_ENV_VAR_NAME, cookie);
2728 return kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2729 action, envp);
2730 }
2731}
2732
2733uint32_t dm_next_uevent_seq(struct mapped_device *md)
2734{
2735 return atomic_add_return(1, &md->uevent_seq);
2736}
2737
2738uint32_t dm_get_event_nr(struct mapped_device *md)
2739{
2740 return atomic_read(&md->event_nr);
2741}
2742
2743int dm_wait_event(struct mapped_device *md, int event_nr)
2744{
2745 return wait_event_interruptible(md->eventq,
2746 (event_nr != atomic_read(&md->event_nr)));
2747}
2748
2749void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2750{
2751 unsigned long flags;
2752
2753 spin_lock_irqsave(&md->uevent_lock, flags);
2754 list_add(elist, &md->uevent_list);
2755 spin_unlock_irqrestore(&md->uevent_lock, flags);
2756}
2757
2758/*
2759 * The gendisk is only valid as long as you have a reference
2760 * count on 'md'.
2761 */
2762struct gendisk *dm_disk(struct mapped_device *md)
2763{
2764 return md->disk;
2765}
2766
2767struct kobject *dm_kobject(struct mapped_device *md)
2768{
2769 return &md->kobj_holder.kobj;
2770}
2771
2772struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2773{
2774 struct mapped_device *md;
2775
2776 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2777
2778 if (test_bit(DMF_FREEING, &md->flags) ||
2779 dm_deleting_md(md))
2780 return NULL;
2781
2782 dm_get(md);
2783 return md;
2784}
2785
2786int dm_suspended_md(struct mapped_device *md)
2787{
2788 return test_bit(DMF_SUSPENDED, &md->flags);
2789}
2790
2791int dm_test_deferred_remove_flag(struct mapped_device *md)
2792{
2793 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2794}
2795
2796int dm_suspended(struct dm_target *ti)
2797{
2798 return dm_suspended_md(dm_table_get_md(ti->table));
2799}
2800EXPORT_SYMBOL_GPL(dm_suspended);
2801
2802int dm_noflush_suspending(struct dm_target *ti)
2803{
2804 return __noflush_suspending(dm_table_get_md(ti->table));
2805}
2806EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2807
2808struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size)
2809{
2810 struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL);
2811 struct kmem_cache *cachep;
2812 unsigned int pool_size;
2813 unsigned int front_pad;
2814
2815 if (!pools)
2816 return NULL;
2817
2818 if (type == DM_TYPE_BIO_BASED) {
2819 cachep = _io_cache;
2820 pool_size = dm_get_reserved_bio_based_ios();
2821 front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone);
2822 } else if (type == DM_TYPE_REQUEST_BASED) {
2823 cachep = _rq_tio_cache;
2824 pool_size = dm_get_reserved_rq_based_ios();
2825 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2826 /* per_bio_data_size is not used. See __bind_mempools(). */
2827 WARN_ON(per_bio_data_size != 0);
2828 } else
2829 goto out;
2830
2831 pools->io_pool = mempool_create_slab_pool(pool_size, cachep);
2832 if (!pools->io_pool)
2833 goto out;
2834
2835 pools->bs = bioset_create(pool_size, front_pad);
2836 if (!pools->bs)
2837 goto out;
2838
2839 if (integrity && bioset_integrity_create(pools->bs, pool_size))
2840 goto out;
2841
2842 return pools;
2843
2844out:
2845 dm_free_md_mempools(pools);
2846
2847 return NULL;
2848}
2849
2850void dm_free_md_mempools(struct dm_md_mempools *pools)
2851{
2852 if (!pools)
2853 return;
2854
2855 if (pools->io_pool)
2856 mempool_destroy(pools->io_pool);
2857
2858 if (pools->bs)
2859 bioset_free(pools->bs);
2860
2861 kfree(pools);
2862}
2863
2864static const struct block_device_operations dm_blk_dops = {
2865 .open = dm_blk_open,
2866 .release = dm_blk_close,
2867 .ioctl = dm_blk_ioctl,
2868 .getgeo = dm_blk_getgeo,
2869 .owner = THIS_MODULE
2870};
2871
2872/*
2873 * module hooks
2874 */
2875module_init(dm_init);
2876module_exit(dm_exit);
2877
2878module_param(major, uint, 0);
2879MODULE_PARM_DESC(major, "The major number of the device mapper");
2880
2881module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
2882MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
2883
2884module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR);
2885MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools");
2886
2887MODULE_DESCRIPTION(DM_NAME " driver");
2888MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
2889MODULE_LICENSE("GPL");