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/keyslot-manager.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 static const char *_name = DM_NAME;
44
45 static unsigned int major = 0;
46 static unsigned int _major = 0;
47
48 static DEFINE_IDR(_minor_idr);
49
50 static DEFINE_SPINLOCK(_minor_lock);
51
52 static void do_deferred_remove(struct work_struct *w);
53
54 static DECLARE_WORK(deferred_remove_work, do_deferred_remove);
55
56 static struct workqueue_struct *deferred_remove_workqueue;
57
58 atomic_t dm_global_event_nr = ATOMIC_INIT(0);
59 DECLARE_WAIT_QUEUE_HEAD(dm_global_eventq);
60
dm_issue_global_event(void)61 void dm_issue_global_event(void)
62 {
63 atomic_inc(&dm_global_event_nr);
64 wake_up(&dm_global_eventq);
65 }
66
67 /*
68 * One of these is allocated (on-stack) per original bio.
69 */
70 struct clone_info {
71 struct dm_table *map;
72 struct bio *bio;
73 struct dm_io *io;
74 sector_t sector;
75 unsigned sector_count;
76 };
77
78 #define DM_TARGET_IO_BIO_OFFSET (offsetof(struct dm_target_io, clone))
79 #define DM_IO_BIO_OFFSET \
80 (offsetof(struct dm_target_io, clone) + offsetof(struct dm_io, tio))
81
dm_per_bio_data(struct bio * bio,size_t data_size)82 void *dm_per_bio_data(struct bio *bio, size_t data_size)
83 {
84 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
85 if (!tio->inside_dm_io)
86 return (char *)bio - DM_TARGET_IO_BIO_OFFSET - data_size;
87 return (char *)bio - DM_IO_BIO_OFFSET - data_size;
88 }
89 EXPORT_SYMBOL_GPL(dm_per_bio_data);
90
dm_bio_from_per_bio_data(void * data,size_t data_size)91 struct bio *dm_bio_from_per_bio_data(void *data, size_t data_size)
92 {
93 struct dm_io *io = (struct dm_io *)((char *)data + data_size);
94 if (io->magic == DM_IO_MAGIC)
95 return (struct bio *)((char *)io + DM_IO_BIO_OFFSET);
96 BUG_ON(io->magic != DM_TIO_MAGIC);
97 return (struct bio *)((char *)io + DM_TARGET_IO_BIO_OFFSET);
98 }
99 EXPORT_SYMBOL_GPL(dm_bio_from_per_bio_data);
100
dm_bio_get_target_bio_nr(const struct bio * bio)101 unsigned dm_bio_get_target_bio_nr(const struct bio *bio)
102 {
103 return container_of(bio, struct dm_target_io, clone)->target_bio_nr;
104 }
105 EXPORT_SYMBOL_GPL(dm_bio_get_target_bio_nr);
106
107 #define MINOR_ALLOCED ((void *)-1)
108
109 #define DM_NUMA_NODE NUMA_NO_NODE
110 static int dm_numa_node = DM_NUMA_NODE;
111
112 #define DEFAULT_SWAP_BIOS (8 * 1048576 / PAGE_SIZE)
113 static int swap_bios = DEFAULT_SWAP_BIOS;
get_swap_bios(void)114 static int get_swap_bios(void)
115 {
116 int latch = READ_ONCE(swap_bios);
117 if (unlikely(latch <= 0))
118 latch = DEFAULT_SWAP_BIOS;
119 return latch;
120 }
121
122 /*
123 * For mempools pre-allocation at the table loading time.
124 */
125 struct dm_md_mempools {
126 struct bio_set bs;
127 struct bio_set io_bs;
128 };
129
130 struct table_device {
131 struct list_head list;
132 refcount_t count;
133 struct dm_dev dm_dev;
134 };
135
136 /*
137 * Bio-based DM's mempools' reserved IOs set by the user.
138 */
139 #define RESERVED_BIO_BASED_IOS 16
140 static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS;
141
__dm_get_module_param_int(int * module_param,int min,int max)142 static int __dm_get_module_param_int(int *module_param, int min, int max)
143 {
144 int param = READ_ONCE(*module_param);
145 int modified_param = 0;
146 bool modified = true;
147
148 if (param < min)
149 modified_param = min;
150 else if (param > max)
151 modified_param = max;
152 else
153 modified = false;
154
155 if (modified) {
156 (void)cmpxchg(module_param, param, modified_param);
157 param = modified_param;
158 }
159
160 return param;
161 }
162
__dm_get_module_param(unsigned * module_param,unsigned def,unsigned max)163 unsigned __dm_get_module_param(unsigned *module_param,
164 unsigned def, unsigned max)
165 {
166 unsigned param = READ_ONCE(*module_param);
167 unsigned modified_param = 0;
168
169 if (!param)
170 modified_param = def;
171 else if (param > max)
172 modified_param = max;
173
174 if (modified_param) {
175 (void)cmpxchg(module_param, param, modified_param);
176 param = modified_param;
177 }
178
179 return param;
180 }
181
dm_get_reserved_bio_based_ios(void)182 unsigned dm_get_reserved_bio_based_ios(void)
183 {
184 return __dm_get_module_param(&reserved_bio_based_ios,
185 RESERVED_BIO_BASED_IOS, DM_RESERVED_MAX_IOS);
186 }
187 EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios);
188
dm_get_numa_node(void)189 static unsigned dm_get_numa_node(void)
190 {
191 return __dm_get_module_param_int(&dm_numa_node,
192 DM_NUMA_NODE, num_online_nodes() - 1);
193 }
194
local_init(void)195 static int __init local_init(void)
196 {
197 int r;
198
199 r = dm_uevent_init();
200 if (r)
201 return r;
202
203 deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1);
204 if (!deferred_remove_workqueue) {
205 r = -ENOMEM;
206 goto out_uevent_exit;
207 }
208
209 _major = major;
210 r = register_blkdev(_major, _name);
211 if (r < 0)
212 goto out_free_workqueue;
213
214 if (!_major)
215 _major = r;
216
217 return 0;
218
219 out_free_workqueue:
220 destroy_workqueue(deferred_remove_workqueue);
221 out_uevent_exit:
222 dm_uevent_exit();
223
224 return r;
225 }
226
local_exit(void)227 static void local_exit(void)
228 {
229 flush_scheduled_work();
230 destroy_workqueue(deferred_remove_workqueue);
231
232 unregister_blkdev(_major, _name);
233 dm_uevent_exit();
234
235 _major = 0;
236
237 DMINFO("cleaned up");
238 }
239
240 static int (*_inits[])(void) __initdata = {
241 local_init,
242 dm_target_init,
243 dm_linear_init,
244 dm_stripe_init,
245 dm_io_init,
246 dm_kcopyd_init,
247 dm_interface_init,
248 dm_statistics_init,
249 };
250
251 static void (*_exits[])(void) = {
252 local_exit,
253 dm_target_exit,
254 dm_linear_exit,
255 dm_stripe_exit,
256 dm_io_exit,
257 dm_kcopyd_exit,
258 dm_interface_exit,
259 dm_statistics_exit,
260 };
261
dm_init(void)262 static int __init dm_init(void)
263 {
264 const int count = ARRAY_SIZE(_inits);
265 int r, i;
266
267 #if (IS_ENABLED(CONFIG_IMA) && !IS_ENABLED(CONFIG_IMA_DISABLE_HTABLE))
268 DMWARN("CONFIG_IMA_DISABLE_HTABLE is disabled."
269 " Duplicate IMA measurements will not be recorded in the IMA log.");
270 #endif
271
272 for (i = 0; i < count; i++) {
273 r = _inits[i]();
274 if (r)
275 goto bad;
276 }
277
278 return 0;
279 bad:
280 while (i--)
281 _exits[i]();
282
283 return r;
284 }
285
dm_exit(void)286 static void __exit dm_exit(void)
287 {
288 int i = ARRAY_SIZE(_exits);
289
290 while (i--)
291 _exits[i]();
292
293 /*
294 * Should be empty by this point.
295 */
296 idr_destroy(&_minor_idr);
297 }
298
299 /*
300 * Block device functions
301 */
dm_deleting_md(struct mapped_device * md)302 int dm_deleting_md(struct mapped_device *md)
303 {
304 return test_bit(DMF_DELETING, &md->flags);
305 }
306
dm_blk_open(struct block_device * bdev,fmode_t mode)307 static int dm_blk_open(struct block_device *bdev, fmode_t mode)
308 {
309 struct mapped_device *md;
310
311 spin_lock(&_minor_lock);
312
313 md = bdev->bd_disk->private_data;
314 if (!md)
315 goto out;
316
317 if (test_bit(DMF_FREEING, &md->flags) ||
318 dm_deleting_md(md)) {
319 md = NULL;
320 goto out;
321 }
322
323 dm_get(md);
324 atomic_inc(&md->open_count);
325 out:
326 spin_unlock(&_minor_lock);
327
328 return md ? 0 : -ENXIO;
329 }
330
dm_blk_close(struct gendisk * disk,fmode_t mode)331 static void dm_blk_close(struct gendisk *disk, fmode_t mode)
332 {
333 struct mapped_device *md;
334
335 spin_lock(&_minor_lock);
336
337 md = disk->private_data;
338 if (WARN_ON(!md))
339 goto out;
340
341 if (atomic_dec_and_test(&md->open_count) &&
342 (test_bit(DMF_DEFERRED_REMOVE, &md->flags)))
343 queue_work(deferred_remove_workqueue, &deferred_remove_work);
344
345 dm_put(md);
346 out:
347 spin_unlock(&_minor_lock);
348 }
349
dm_open_count(struct mapped_device * md)350 int dm_open_count(struct mapped_device *md)
351 {
352 return atomic_read(&md->open_count);
353 }
354
355 /*
356 * Guarantees nothing is using the device before it's deleted.
357 */
dm_lock_for_deletion(struct mapped_device * md,bool mark_deferred,bool only_deferred)358 int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred)
359 {
360 int r = 0;
361
362 spin_lock(&_minor_lock);
363
364 if (dm_open_count(md)) {
365 r = -EBUSY;
366 if (mark_deferred)
367 set_bit(DMF_DEFERRED_REMOVE, &md->flags);
368 } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags))
369 r = -EEXIST;
370 else
371 set_bit(DMF_DELETING, &md->flags);
372
373 spin_unlock(&_minor_lock);
374
375 return r;
376 }
377
dm_cancel_deferred_remove(struct mapped_device * md)378 int dm_cancel_deferred_remove(struct mapped_device *md)
379 {
380 int r = 0;
381
382 spin_lock(&_minor_lock);
383
384 if (test_bit(DMF_DELETING, &md->flags))
385 r = -EBUSY;
386 else
387 clear_bit(DMF_DEFERRED_REMOVE, &md->flags);
388
389 spin_unlock(&_minor_lock);
390
391 return r;
392 }
393
do_deferred_remove(struct work_struct * w)394 static void do_deferred_remove(struct work_struct *w)
395 {
396 dm_deferred_remove();
397 }
398
dm_blk_getgeo(struct block_device * bdev,struct hd_geometry * geo)399 static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
400 {
401 struct mapped_device *md = bdev->bd_disk->private_data;
402
403 return dm_get_geometry(md, geo);
404 }
405
dm_prepare_ioctl(struct mapped_device * md,int * srcu_idx,struct block_device ** bdev)406 static int dm_prepare_ioctl(struct mapped_device *md, int *srcu_idx,
407 struct block_device **bdev)
408 {
409 struct dm_target *tgt;
410 struct dm_table *map;
411 int r;
412
413 retry:
414 r = -ENOTTY;
415 map = dm_get_live_table(md, srcu_idx);
416 if (!map || !dm_table_get_size(map))
417 return r;
418
419 /* We only support devices that have a single target */
420 if (dm_table_get_num_targets(map) != 1)
421 return r;
422
423 tgt = dm_table_get_target(map, 0);
424 if (!tgt->type->prepare_ioctl)
425 return r;
426
427 if (dm_suspended_md(md))
428 return -EAGAIN;
429
430 r = tgt->type->prepare_ioctl(tgt, bdev);
431 if (r == -ENOTCONN && !fatal_signal_pending(current)) {
432 dm_put_live_table(md, *srcu_idx);
433 msleep(10);
434 goto retry;
435 }
436
437 return r;
438 }
439
dm_unprepare_ioctl(struct mapped_device * md,int srcu_idx)440 static void dm_unprepare_ioctl(struct mapped_device *md, int srcu_idx)
441 {
442 dm_put_live_table(md, srcu_idx);
443 }
444
dm_blk_ioctl(struct block_device * bdev,fmode_t mode,unsigned int cmd,unsigned long arg)445 static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode,
446 unsigned int cmd, unsigned long arg)
447 {
448 struct mapped_device *md = bdev->bd_disk->private_data;
449 int r, srcu_idx;
450
451 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
452 if (r < 0)
453 goto out;
454
455 if (r > 0) {
456 /*
457 * Target determined this ioctl is being issued against a
458 * subset of the parent bdev; require extra privileges.
459 */
460 if (!capable(CAP_SYS_RAWIO)) {
461 DMDEBUG_LIMIT(
462 "%s: sending ioctl %x to DM device without required privilege.",
463 current->comm, cmd);
464 r = -ENOIOCTLCMD;
465 goto out;
466 }
467 }
468
469 if (!bdev->bd_disk->fops->ioctl)
470 r = -ENOTTY;
471 else
472 r = bdev->bd_disk->fops->ioctl(bdev, mode, cmd, arg);
473 out:
474 dm_unprepare_ioctl(md, srcu_idx);
475 return r;
476 }
477
dm_start_time_ns_from_clone(struct bio * bio)478 u64 dm_start_time_ns_from_clone(struct bio *bio)
479 {
480 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
481 struct dm_io *io = tio->io;
482
483 return jiffies_to_nsecs(io->start_time);
484 }
485 EXPORT_SYMBOL_GPL(dm_start_time_ns_from_clone);
486
start_io_acct(struct dm_io * io)487 static void start_io_acct(struct dm_io *io)
488 {
489 struct mapped_device *md = io->md;
490 struct bio *bio = io->orig_bio;
491
492 io->start_time = bio_start_io_acct(bio);
493 if (unlikely(dm_stats_used(&md->stats)))
494 dm_stats_account_io(&md->stats, bio_data_dir(bio),
495 bio->bi_iter.bi_sector, bio_sectors(bio),
496 false, 0, &io->stats_aux);
497 }
498
end_io_acct(struct mapped_device * md,struct bio * bio,unsigned long start_time,struct dm_stats_aux * stats_aux)499 static void end_io_acct(struct mapped_device *md, struct bio *bio,
500 unsigned long start_time, struct dm_stats_aux *stats_aux)
501 {
502 unsigned long duration = jiffies - start_time;
503
504 bio_end_io_acct(bio, start_time);
505
506 if (unlikely(dm_stats_used(&md->stats)))
507 dm_stats_account_io(&md->stats, bio_data_dir(bio),
508 bio->bi_iter.bi_sector, bio_sectors(bio),
509 true, duration, stats_aux);
510
511 /* nudge anyone waiting on suspend queue */
512 if (unlikely(wq_has_sleeper(&md->wait)))
513 wake_up(&md->wait);
514 }
515
alloc_io(struct mapped_device * md,struct bio * bio)516 static struct dm_io *alloc_io(struct mapped_device *md, struct bio *bio)
517 {
518 struct dm_io *io;
519 struct dm_target_io *tio;
520 struct bio *clone;
521
522 clone = bio_alloc_bioset(GFP_NOIO, 0, &md->io_bs);
523 if (!clone)
524 return NULL;
525
526 tio = container_of(clone, struct dm_target_io, clone);
527 tio->inside_dm_io = true;
528 tio->io = NULL;
529
530 io = container_of(tio, struct dm_io, tio);
531 io->magic = DM_IO_MAGIC;
532 io->status = 0;
533 atomic_set(&io->io_count, 1);
534 io->orig_bio = bio;
535 io->md = md;
536 spin_lock_init(&io->endio_lock);
537
538 start_io_acct(io);
539
540 return io;
541 }
542
free_io(struct mapped_device * md,struct dm_io * io)543 static void free_io(struct mapped_device *md, struct dm_io *io)
544 {
545 bio_put(&io->tio.clone);
546 }
547
alloc_tio(struct clone_info * ci,struct dm_target * ti,unsigned target_bio_nr,gfp_t gfp_mask)548 static struct dm_target_io *alloc_tio(struct clone_info *ci, struct dm_target *ti,
549 unsigned target_bio_nr, gfp_t gfp_mask)
550 {
551 struct dm_target_io *tio;
552
553 if (!ci->io->tio.io) {
554 /* the dm_target_io embedded in ci->io is available */
555 tio = &ci->io->tio;
556 } else {
557 struct bio *clone = bio_alloc_bioset(gfp_mask, 0, &ci->io->md->bs);
558 if (!clone)
559 return NULL;
560
561 tio = container_of(clone, struct dm_target_io, clone);
562 tio->inside_dm_io = false;
563 }
564
565 tio->magic = DM_TIO_MAGIC;
566 tio->io = ci->io;
567 tio->ti = ti;
568 tio->target_bio_nr = target_bio_nr;
569
570 return tio;
571 }
572
free_tio(struct dm_target_io * tio)573 static void free_tio(struct dm_target_io *tio)
574 {
575 if (tio->inside_dm_io)
576 return;
577 bio_put(&tio->clone);
578 }
579
580 /*
581 * Add the bio to the list of deferred io.
582 */
queue_io(struct mapped_device * md,struct bio * bio)583 static void queue_io(struct mapped_device *md, struct bio *bio)
584 {
585 unsigned long flags;
586
587 spin_lock_irqsave(&md->deferred_lock, flags);
588 bio_list_add(&md->deferred, bio);
589 spin_unlock_irqrestore(&md->deferred_lock, flags);
590 queue_work(md->wq, &md->work);
591 }
592
593 /*
594 * Everyone (including functions in this file), should use this
595 * function to access the md->map field, and make sure they call
596 * dm_put_live_table() when finished.
597 */
dm_get_live_table(struct mapped_device * md,int * srcu_idx)598 struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier)
599 {
600 *srcu_idx = srcu_read_lock(&md->io_barrier);
601
602 return srcu_dereference(md->map, &md->io_barrier);
603 }
604
dm_put_live_table(struct mapped_device * md,int srcu_idx)605 void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier)
606 {
607 srcu_read_unlock(&md->io_barrier, srcu_idx);
608 }
609
dm_sync_table(struct mapped_device * md)610 void dm_sync_table(struct mapped_device *md)
611 {
612 synchronize_srcu(&md->io_barrier);
613 synchronize_rcu_expedited();
614 }
615
616 /*
617 * A fast alternative to dm_get_live_table/dm_put_live_table.
618 * The caller must not block between these two functions.
619 */
dm_get_live_table_fast(struct mapped_device * md)620 static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU)
621 {
622 rcu_read_lock();
623 return rcu_dereference(md->map);
624 }
625
dm_put_live_table_fast(struct mapped_device * md)626 static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU)
627 {
628 rcu_read_unlock();
629 }
630
631 static char *_dm_claim_ptr = "I belong to device-mapper";
632
633 /*
634 * Open a table device so we can use it as a map destination.
635 */
open_table_device(struct table_device * td,dev_t dev,struct mapped_device * md)636 static int open_table_device(struct table_device *td, dev_t dev,
637 struct mapped_device *md)
638 {
639 struct block_device *bdev;
640
641 int r;
642
643 BUG_ON(td->dm_dev.bdev);
644
645 bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _dm_claim_ptr);
646 if (IS_ERR(bdev))
647 return PTR_ERR(bdev);
648
649 r = bd_link_disk_holder(bdev, dm_disk(md));
650 if (r) {
651 blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL);
652 return r;
653 }
654
655 td->dm_dev.bdev = bdev;
656 td->dm_dev.dax_dev = fs_dax_get_by_bdev(bdev);
657 return 0;
658 }
659
660 /*
661 * Close a table device that we've been using.
662 */
close_table_device(struct table_device * td,struct mapped_device * md)663 static void close_table_device(struct table_device *td, struct mapped_device *md)
664 {
665 if (!td->dm_dev.bdev)
666 return;
667
668 bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md));
669 blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL);
670 put_dax(td->dm_dev.dax_dev);
671 td->dm_dev.bdev = NULL;
672 td->dm_dev.dax_dev = NULL;
673 }
674
find_table_device(struct list_head * l,dev_t dev,fmode_t mode)675 static struct table_device *find_table_device(struct list_head *l, dev_t dev,
676 fmode_t mode)
677 {
678 struct table_device *td;
679
680 list_for_each_entry(td, l, list)
681 if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode)
682 return td;
683
684 return NULL;
685 }
686
dm_get_table_device(struct mapped_device * md,dev_t dev,fmode_t mode,struct dm_dev ** result)687 int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode,
688 struct dm_dev **result)
689 {
690 int r;
691 struct table_device *td;
692
693 mutex_lock(&md->table_devices_lock);
694 td = find_table_device(&md->table_devices, dev, mode);
695 if (!td) {
696 td = kmalloc_node(sizeof(*td), GFP_KERNEL, md->numa_node_id);
697 if (!td) {
698 mutex_unlock(&md->table_devices_lock);
699 return -ENOMEM;
700 }
701
702 td->dm_dev.mode = mode;
703 td->dm_dev.bdev = NULL;
704
705 if ((r = open_table_device(td, dev, md))) {
706 mutex_unlock(&md->table_devices_lock);
707 kfree(td);
708 return r;
709 }
710
711 format_dev_t(td->dm_dev.name, dev);
712
713 refcount_set(&td->count, 1);
714 list_add(&td->list, &md->table_devices);
715 } else {
716 refcount_inc(&td->count);
717 }
718 mutex_unlock(&md->table_devices_lock);
719
720 *result = &td->dm_dev;
721 return 0;
722 }
723
dm_put_table_device(struct mapped_device * md,struct dm_dev * d)724 void dm_put_table_device(struct mapped_device *md, struct dm_dev *d)
725 {
726 struct table_device *td = container_of(d, struct table_device, dm_dev);
727
728 mutex_lock(&md->table_devices_lock);
729 if (refcount_dec_and_test(&td->count)) {
730 close_table_device(td, md);
731 list_del(&td->list);
732 kfree(td);
733 }
734 mutex_unlock(&md->table_devices_lock);
735 }
736
free_table_devices(struct list_head * devices)737 static void free_table_devices(struct list_head *devices)
738 {
739 struct list_head *tmp, *next;
740
741 list_for_each_safe(tmp, next, devices) {
742 struct table_device *td = list_entry(tmp, struct table_device, list);
743
744 DMWARN("dm_destroy: %s still exists with %d references",
745 td->dm_dev.name, refcount_read(&td->count));
746 kfree(td);
747 }
748 }
749
750 /*
751 * Get the geometry associated with a dm device
752 */
dm_get_geometry(struct mapped_device * md,struct hd_geometry * geo)753 int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo)
754 {
755 *geo = md->geometry;
756
757 return 0;
758 }
759
760 /*
761 * Set the geometry of a device.
762 */
dm_set_geometry(struct mapped_device * md,struct hd_geometry * geo)763 int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo)
764 {
765 sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors;
766
767 if (geo->start > sz) {
768 DMWARN("Start sector is beyond the geometry limits.");
769 return -EINVAL;
770 }
771
772 md->geometry = *geo;
773
774 return 0;
775 }
776
__noflush_suspending(struct mapped_device * md)777 static int __noflush_suspending(struct mapped_device *md)
778 {
779 return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
780 }
781
782 /*
783 * Decrements the number of outstanding ios that a bio has been
784 * cloned into, completing the original io if necc.
785 */
dm_io_dec_pending(struct dm_io * io,blk_status_t error)786 void dm_io_dec_pending(struct dm_io *io, blk_status_t error)
787 {
788 unsigned long flags;
789 blk_status_t io_error;
790 struct bio *bio;
791 struct mapped_device *md = io->md;
792 unsigned long start_time = 0;
793 struct dm_stats_aux stats_aux;
794
795 /* Push-back supersedes any I/O errors */
796 if (unlikely(error)) {
797 spin_lock_irqsave(&io->endio_lock, flags);
798 if (!(io->status == BLK_STS_DM_REQUEUE && __noflush_suspending(md)))
799 io->status = error;
800 spin_unlock_irqrestore(&io->endio_lock, flags);
801 }
802
803 if (atomic_dec_and_test(&io->io_count)) {
804 bio = io->orig_bio;
805 if (io->status == BLK_STS_DM_REQUEUE) {
806 /*
807 * Target requested pushing back the I/O.
808 */
809 spin_lock_irqsave(&md->deferred_lock, flags);
810 if (__noflush_suspending(md) &&
811 !WARN_ON_ONCE(dm_is_zone_write(md, bio))) {
812 /* NOTE early return due to BLK_STS_DM_REQUEUE below */
813 bio_list_add_head(&md->deferred, bio);
814 } else {
815 /*
816 * noflush suspend was interrupted or this is
817 * a write to a zoned target.
818 */
819 io->status = BLK_STS_IOERR;
820 }
821 spin_unlock_irqrestore(&md->deferred_lock, flags);
822 }
823
824 io_error = io->status;
825 start_time = io->start_time;
826 stats_aux = io->stats_aux;
827 free_io(md, io);
828 end_io_acct(md, bio, start_time, &stats_aux);
829
830 if (io_error == BLK_STS_DM_REQUEUE)
831 return;
832
833 if ((bio->bi_opf & REQ_PREFLUSH) && bio->bi_iter.bi_size) {
834 /*
835 * Preflush done for flush with data, reissue
836 * without REQ_PREFLUSH.
837 */
838 bio->bi_opf &= ~REQ_PREFLUSH;
839 queue_io(md, bio);
840 } else {
841 /* done with normal IO or empty flush */
842 if (io_error)
843 bio->bi_status = io_error;
844 bio_endio(bio);
845 }
846 }
847 }
848
disable_discard(struct mapped_device * md)849 void disable_discard(struct mapped_device *md)
850 {
851 struct queue_limits *limits = dm_get_queue_limits(md);
852
853 /* device doesn't really support DISCARD, disable it */
854 limits->max_discard_sectors = 0;
855 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, md->queue);
856 }
857
disable_write_same(struct mapped_device * md)858 void disable_write_same(struct mapped_device *md)
859 {
860 struct queue_limits *limits = dm_get_queue_limits(md);
861
862 /* device doesn't really support WRITE SAME, disable it */
863 limits->max_write_same_sectors = 0;
864 }
865
disable_write_zeroes(struct mapped_device * md)866 void disable_write_zeroes(struct mapped_device *md)
867 {
868 struct queue_limits *limits = dm_get_queue_limits(md);
869
870 /* device doesn't really support WRITE ZEROES, disable it */
871 limits->max_write_zeroes_sectors = 0;
872 }
873
swap_bios_limit(struct dm_target * ti,struct bio * bio)874 static bool swap_bios_limit(struct dm_target *ti, struct bio *bio)
875 {
876 return unlikely((bio->bi_opf & REQ_SWAP) != 0) && unlikely(ti->limit_swap_bios);
877 }
878
clone_endio(struct bio * bio)879 static void clone_endio(struct bio *bio)
880 {
881 blk_status_t error = bio->bi_status;
882 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
883 struct dm_io *io = tio->io;
884 struct mapped_device *md = tio->io->md;
885 dm_endio_fn endio = tio->ti->type->end_io;
886 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
887
888 if (unlikely(error == BLK_STS_TARGET)) {
889 if (bio_op(bio) == REQ_OP_DISCARD &&
890 !q->limits.max_discard_sectors)
891 disable_discard(md);
892 else if (bio_op(bio) == REQ_OP_WRITE_SAME &&
893 !q->limits.max_write_same_sectors)
894 disable_write_same(md);
895 else if (bio_op(bio) == REQ_OP_WRITE_ZEROES &&
896 !q->limits.max_write_zeroes_sectors)
897 disable_write_zeroes(md);
898 }
899
900 if (blk_queue_is_zoned(q))
901 dm_zone_endio(io, bio);
902
903 if (endio) {
904 int r = endio(tio->ti, bio, &error);
905 switch (r) {
906 case DM_ENDIO_REQUEUE:
907 /*
908 * Requeuing writes to a sequential zone of a zoned
909 * target will break the sequential write pattern:
910 * fail such IO.
911 */
912 if (WARN_ON_ONCE(dm_is_zone_write(md, bio)))
913 error = BLK_STS_IOERR;
914 else
915 error = BLK_STS_DM_REQUEUE;
916 fallthrough;
917 case DM_ENDIO_DONE:
918 break;
919 case DM_ENDIO_INCOMPLETE:
920 /* The target will handle the io */
921 return;
922 default:
923 DMWARN("unimplemented target endio return value: %d", r);
924 BUG();
925 }
926 }
927
928 if (unlikely(swap_bios_limit(tio->ti, bio))) {
929 struct mapped_device *md = io->md;
930 up(&md->swap_bios_semaphore);
931 }
932
933 free_tio(tio);
934 dm_io_dec_pending(io, error);
935 }
936
937 /*
938 * Return maximum size of I/O possible at the supplied sector up to the current
939 * target boundary.
940 */
max_io_len_target_boundary(struct dm_target * ti,sector_t target_offset)941 static inline sector_t max_io_len_target_boundary(struct dm_target *ti,
942 sector_t target_offset)
943 {
944 return ti->len - target_offset;
945 }
946
max_io_len(struct dm_target * ti,sector_t sector)947 static sector_t max_io_len(struct dm_target *ti, sector_t sector)
948 {
949 sector_t target_offset = dm_target_offset(ti, sector);
950 sector_t len = max_io_len_target_boundary(ti, target_offset);
951 sector_t max_len;
952
953 /*
954 * Does the target need to split IO even further?
955 * - varied (per target) IO splitting is a tenet of DM; this
956 * explains why stacked chunk_sectors based splitting via
957 * blk_max_size_offset() isn't possible here. So pass in
958 * ti->max_io_len to override stacked chunk_sectors.
959 */
960 if (ti->max_io_len) {
961 max_len = blk_max_size_offset(ti->table->md->queue,
962 target_offset, ti->max_io_len);
963 if (len > max_len)
964 len = max_len;
965 }
966
967 return len;
968 }
969
dm_set_target_max_io_len(struct dm_target * ti,sector_t len)970 int dm_set_target_max_io_len(struct dm_target *ti, sector_t len)
971 {
972 if (len > UINT_MAX) {
973 DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)",
974 (unsigned long long)len, UINT_MAX);
975 ti->error = "Maximum size of target IO is too large";
976 return -EINVAL;
977 }
978
979 ti->max_io_len = (uint32_t) len;
980
981 return 0;
982 }
983 EXPORT_SYMBOL_GPL(dm_set_target_max_io_len);
984
dm_dax_get_live_target(struct mapped_device * md,sector_t sector,int * srcu_idx)985 static struct dm_target *dm_dax_get_live_target(struct mapped_device *md,
986 sector_t sector, int *srcu_idx)
987 __acquires(md->io_barrier)
988 {
989 struct dm_table *map;
990 struct dm_target *ti;
991
992 map = dm_get_live_table(md, srcu_idx);
993 if (!map)
994 return NULL;
995
996 ti = dm_table_find_target(map, sector);
997 if (!ti)
998 return NULL;
999
1000 return ti;
1001 }
1002
dm_dax_direct_access(struct dax_device * dax_dev,pgoff_t pgoff,long nr_pages,void ** kaddr,pfn_t * pfn)1003 static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff,
1004 long nr_pages, void **kaddr, pfn_t *pfn)
1005 {
1006 struct mapped_device *md = dax_get_private(dax_dev);
1007 sector_t sector = pgoff * PAGE_SECTORS;
1008 struct dm_target *ti;
1009 long len, ret = -EIO;
1010 int srcu_idx;
1011
1012 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1013
1014 if (!ti)
1015 goto out;
1016 if (!ti->type->direct_access)
1017 goto out;
1018 len = max_io_len(ti, sector) / PAGE_SECTORS;
1019 if (len < 1)
1020 goto out;
1021 nr_pages = min(len, nr_pages);
1022 ret = ti->type->direct_access(ti, pgoff, nr_pages, kaddr, pfn);
1023
1024 out:
1025 dm_put_live_table(md, srcu_idx);
1026
1027 return ret;
1028 }
1029
dm_dax_supported(struct dax_device * dax_dev,struct block_device * bdev,int blocksize,sector_t start,sector_t len)1030 static bool dm_dax_supported(struct dax_device *dax_dev, struct block_device *bdev,
1031 int blocksize, sector_t start, sector_t len)
1032 {
1033 struct mapped_device *md = dax_get_private(dax_dev);
1034 struct dm_table *map;
1035 bool ret = false;
1036 int srcu_idx;
1037
1038 map = dm_get_live_table(md, &srcu_idx);
1039 if (!map)
1040 goto out;
1041
1042 ret = dm_table_supports_dax(map, device_not_dax_capable, &blocksize);
1043
1044 out:
1045 dm_put_live_table(md, srcu_idx);
1046
1047 return ret;
1048 }
1049
dm_dax_copy_from_iter(struct dax_device * dax_dev,pgoff_t pgoff,void * addr,size_t bytes,struct iov_iter * i)1050 static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1051 void *addr, size_t bytes, struct iov_iter *i)
1052 {
1053 struct mapped_device *md = dax_get_private(dax_dev);
1054 sector_t sector = pgoff * PAGE_SECTORS;
1055 struct dm_target *ti;
1056 long ret = 0;
1057 int srcu_idx;
1058
1059 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1060
1061 if (!ti)
1062 goto out;
1063 if (!ti->type->dax_copy_from_iter) {
1064 ret = copy_from_iter(addr, bytes, i);
1065 goto out;
1066 }
1067 ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i);
1068 out:
1069 dm_put_live_table(md, srcu_idx);
1070
1071 return ret;
1072 }
1073
dm_dax_copy_to_iter(struct dax_device * dax_dev,pgoff_t pgoff,void * addr,size_t bytes,struct iov_iter * i)1074 static size_t dm_dax_copy_to_iter(struct dax_device *dax_dev, pgoff_t pgoff,
1075 void *addr, size_t bytes, struct iov_iter *i)
1076 {
1077 struct mapped_device *md = dax_get_private(dax_dev);
1078 sector_t sector = pgoff * PAGE_SECTORS;
1079 struct dm_target *ti;
1080 long ret = 0;
1081 int srcu_idx;
1082
1083 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1084
1085 if (!ti)
1086 goto out;
1087 if (!ti->type->dax_copy_to_iter) {
1088 ret = copy_to_iter(addr, bytes, i);
1089 goto out;
1090 }
1091 ret = ti->type->dax_copy_to_iter(ti, pgoff, addr, bytes, i);
1092 out:
1093 dm_put_live_table(md, srcu_idx);
1094
1095 return ret;
1096 }
1097
dm_dax_zero_page_range(struct dax_device * dax_dev,pgoff_t pgoff,size_t nr_pages)1098 static int dm_dax_zero_page_range(struct dax_device *dax_dev, pgoff_t pgoff,
1099 size_t nr_pages)
1100 {
1101 struct mapped_device *md = dax_get_private(dax_dev);
1102 sector_t sector = pgoff * PAGE_SECTORS;
1103 struct dm_target *ti;
1104 int ret = -EIO;
1105 int srcu_idx;
1106
1107 ti = dm_dax_get_live_target(md, sector, &srcu_idx);
1108
1109 if (!ti)
1110 goto out;
1111 if (WARN_ON(!ti->type->dax_zero_page_range)) {
1112 /*
1113 * ->zero_page_range() is mandatory dax operation. If we are
1114 * here, something is wrong.
1115 */
1116 goto out;
1117 }
1118 ret = ti->type->dax_zero_page_range(ti, pgoff, nr_pages);
1119 out:
1120 dm_put_live_table(md, srcu_idx);
1121
1122 return ret;
1123 }
1124
1125 /*
1126 * A target may call dm_accept_partial_bio only from the map routine. It is
1127 * allowed for all bio types except REQ_PREFLUSH, REQ_OP_ZONE_* zone management
1128 * operations and REQ_OP_ZONE_APPEND (zone append writes).
1129 *
1130 * dm_accept_partial_bio informs the dm that the target only wants to process
1131 * additional n_sectors sectors of the bio and the rest of the data should be
1132 * sent in a next bio.
1133 *
1134 * A diagram that explains the arithmetics:
1135 * +--------------------+---------------+-------+
1136 * | 1 | 2 | 3 |
1137 * +--------------------+---------------+-------+
1138 *
1139 * <-------------- *tio->len_ptr --------------->
1140 * <------- bi_size ------->
1141 * <-- n_sectors -->
1142 *
1143 * Region 1 was already iterated over with bio_advance or similar function.
1144 * (it may be empty if the target doesn't use bio_advance)
1145 * Region 2 is the remaining bio size that the target wants to process.
1146 * (it may be empty if region 1 is non-empty, although there is no reason
1147 * to make it empty)
1148 * The target requires that region 3 is to be sent in the next bio.
1149 *
1150 * If the target wants to receive multiple copies of the bio (via num_*bios, etc),
1151 * the partially processed part (the sum of regions 1+2) must be the same for all
1152 * copies of the bio.
1153 */
dm_accept_partial_bio(struct bio * bio,unsigned n_sectors)1154 void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors)
1155 {
1156 struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone);
1157 unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT;
1158
1159 BUG_ON(bio->bi_opf & REQ_PREFLUSH);
1160 BUG_ON(op_is_zone_mgmt(bio_op(bio)));
1161 BUG_ON(bio_op(bio) == REQ_OP_ZONE_APPEND);
1162 BUG_ON(bi_size > *tio->len_ptr);
1163 BUG_ON(n_sectors > bi_size);
1164
1165 *tio->len_ptr -= bi_size - n_sectors;
1166 bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT;
1167 }
1168 EXPORT_SYMBOL_GPL(dm_accept_partial_bio);
1169
__set_swap_bios_limit(struct mapped_device * md,int latch)1170 static noinline void __set_swap_bios_limit(struct mapped_device *md, int latch)
1171 {
1172 mutex_lock(&md->swap_bios_lock);
1173 while (latch < md->swap_bios) {
1174 cond_resched();
1175 down(&md->swap_bios_semaphore);
1176 md->swap_bios--;
1177 }
1178 while (latch > md->swap_bios) {
1179 cond_resched();
1180 up(&md->swap_bios_semaphore);
1181 md->swap_bios++;
1182 }
1183 mutex_unlock(&md->swap_bios_lock);
1184 }
1185
__map_bio(struct dm_target_io * tio)1186 static blk_qc_t __map_bio(struct dm_target_io *tio)
1187 {
1188 int r;
1189 sector_t sector;
1190 struct bio *clone = &tio->clone;
1191 struct dm_io *io = tio->io;
1192 struct dm_target *ti = tio->ti;
1193 blk_qc_t ret = BLK_QC_T_NONE;
1194
1195 clone->bi_end_io = clone_endio;
1196
1197 /*
1198 * Map the clone. If r == 0 we don't need to do
1199 * anything, the target has assumed ownership of
1200 * this io.
1201 */
1202 dm_io_inc_pending(io);
1203 sector = clone->bi_iter.bi_sector;
1204
1205 if (unlikely(swap_bios_limit(ti, clone))) {
1206 struct mapped_device *md = io->md;
1207 int latch = get_swap_bios();
1208 if (unlikely(latch != md->swap_bios))
1209 __set_swap_bios_limit(md, latch);
1210 down(&md->swap_bios_semaphore);
1211 }
1212
1213 /*
1214 * Check if the IO needs a special mapping due to zone append emulation
1215 * on zoned target. In this case, dm_zone_map_bio() calls the target
1216 * map operation.
1217 */
1218 if (dm_emulate_zone_append(io->md))
1219 r = dm_zone_map_bio(tio);
1220 else
1221 r = ti->type->map(ti, clone);
1222
1223 switch (r) {
1224 case DM_MAPIO_SUBMITTED:
1225 break;
1226 case DM_MAPIO_REMAPPED:
1227 /* the bio has been remapped so dispatch it */
1228 trace_block_bio_remap(clone, bio_dev(io->orig_bio), sector);
1229 ret = submit_bio_noacct(clone);
1230 break;
1231 case DM_MAPIO_KILL:
1232 if (unlikely(swap_bios_limit(ti, clone))) {
1233 struct mapped_device *md = io->md;
1234 up(&md->swap_bios_semaphore);
1235 }
1236 free_tio(tio);
1237 dm_io_dec_pending(io, BLK_STS_IOERR);
1238 break;
1239 case DM_MAPIO_REQUEUE:
1240 if (unlikely(swap_bios_limit(ti, clone))) {
1241 struct mapped_device *md = io->md;
1242 up(&md->swap_bios_semaphore);
1243 }
1244 free_tio(tio);
1245 dm_io_dec_pending(io, BLK_STS_DM_REQUEUE);
1246 break;
1247 default:
1248 DMWARN("unimplemented target map return value: %d", r);
1249 BUG();
1250 }
1251
1252 return ret;
1253 }
1254
bio_setup_sector(struct bio * bio,sector_t sector,unsigned len)1255 static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len)
1256 {
1257 bio->bi_iter.bi_sector = sector;
1258 bio->bi_iter.bi_size = to_bytes(len);
1259 }
1260
1261 /*
1262 * Creates a bio that consists of range of complete bvecs.
1263 */
clone_bio(struct dm_target_io * tio,struct bio * bio,sector_t sector,unsigned len)1264 static int clone_bio(struct dm_target_io *tio, struct bio *bio,
1265 sector_t sector, unsigned len)
1266 {
1267 struct bio *clone = &tio->clone;
1268 int r;
1269
1270 __bio_clone_fast(clone, bio);
1271
1272 r = bio_crypt_clone(clone, bio, GFP_NOIO);
1273 if (r < 0)
1274 return r;
1275
1276 if (bio_integrity(bio)) {
1277 if (unlikely(!dm_target_has_integrity(tio->ti->type) &&
1278 !dm_target_passes_integrity(tio->ti->type))) {
1279 DMWARN("%s: the target %s doesn't support integrity data.",
1280 dm_device_name(tio->io->md),
1281 tio->ti->type->name);
1282 return -EIO;
1283 }
1284
1285 r = bio_integrity_clone(clone, bio, GFP_NOIO);
1286 if (r < 0)
1287 return r;
1288 }
1289
1290 bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector));
1291 clone->bi_iter.bi_size = to_bytes(len);
1292
1293 if (bio_integrity(bio))
1294 bio_integrity_trim(clone);
1295
1296 return 0;
1297 }
1298
alloc_multiple_bios(struct bio_list * blist,struct clone_info * ci,struct dm_target * ti,unsigned num_bios)1299 static void alloc_multiple_bios(struct bio_list *blist, struct clone_info *ci,
1300 struct dm_target *ti, unsigned num_bios)
1301 {
1302 struct dm_target_io *tio;
1303 int try;
1304
1305 if (!num_bios)
1306 return;
1307
1308 if (num_bios == 1) {
1309 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1310 bio_list_add(blist, &tio->clone);
1311 return;
1312 }
1313
1314 for (try = 0; try < 2; try++) {
1315 int bio_nr;
1316 struct bio *bio;
1317
1318 if (try)
1319 mutex_lock(&ci->io->md->table_devices_lock);
1320 for (bio_nr = 0; bio_nr < num_bios; bio_nr++) {
1321 tio = alloc_tio(ci, ti, bio_nr, try ? GFP_NOIO : GFP_NOWAIT);
1322 if (!tio)
1323 break;
1324
1325 bio_list_add(blist, &tio->clone);
1326 }
1327 if (try)
1328 mutex_unlock(&ci->io->md->table_devices_lock);
1329 if (bio_nr == num_bios)
1330 return;
1331
1332 while ((bio = bio_list_pop(blist))) {
1333 tio = container_of(bio, struct dm_target_io, clone);
1334 free_tio(tio);
1335 }
1336 }
1337 }
1338
__clone_and_map_simple_bio(struct clone_info * ci,struct dm_target_io * tio,unsigned * len)1339 static blk_qc_t __clone_and_map_simple_bio(struct clone_info *ci,
1340 struct dm_target_io *tio, unsigned *len)
1341 {
1342 struct bio *clone = &tio->clone;
1343
1344 tio->len_ptr = len;
1345
1346 __bio_clone_fast(clone, ci->bio);
1347 if (len)
1348 bio_setup_sector(clone, ci->sector, *len);
1349
1350 return __map_bio(tio);
1351 }
1352
__send_duplicate_bios(struct clone_info * ci,struct dm_target * ti,unsigned num_bios,unsigned * len)1353 static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti,
1354 unsigned num_bios, unsigned *len)
1355 {
1356 struct bio_list blist = BIO_EMPTY_LIST;
1357 struct bio *bio;
1358 struct dm_target_io *tio;
1359
1360 alloc_multiple_bios(&blist, ci, ti, num_bios);
1361
1362 while ((bio = bio_list_pop(&blist))) {
1363 tio = container_of(bio, struct dm_target_io, clone);
1364 (void) __clone_and_map_simple_bio(ci, tio, len);
1365 }
1366 }
1367
__send_empty_flush(struct clone_info * ci)1368 static int __send_empty_flush(struct clone_info *ci)
1369 {
1370 unsigned target_nr = 0;
1371 struct dm_target *ti;
1372 struct bio flush_bio;
1373
1374 /*
1375 * Use an on-stack bio for this, it's safe since we don't
1376 * need to reference it after submit. It's just used as
1377 * the basis for the clone(s).
1378 */
1379 bio_init(&flush_bio, NULL, 0);
1380 flush_bio.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC;
1381 bio_set_dev(&flush_bio, ci->io->md->disk->part0);
1382
1383 ci->bio = &flush_bio;
1384 ci->sector_count = 0;
1385
1386 BUG_ON(bio_has_data(ci->bio));
1387 while ((ti = dm_table_get_target(ci->map, target_nr++)))
1388 __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL);
1389
1390 bio_uninit(ci->bio);
1391 return 0;
1392 }
1393
__clone_and_map_data_bio(struct clone_info * ci,struct dm_target * ti,sector_t sector,unsigned * len)1394 static int __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti,
1395 sector_t sector, unsigned *len)
1396 {
1397 struct bio *bio = ci->bio;
1398 struct dm_target_io *tio;
1399 int r;
1400
1401 tio = alloc_tio(ci, ti, 0, GFP_NOIO);
1402 tio->len_ptr = len;
1403 r = clone_bio(tio, bio, sector, *len);
1404 if (r < 0) {
1405 free_tio(tio);
1406 return r;
1407 }
1408 (void) __map_bio(tio);
1409
1410 return 0;
1411 }
1412
__send_changing_extent_only(struct clone_info * ci,struct dm_target * ti,unsigned num_bios)1413 static int __send_changing_extent_only(struct clone_info *ci, struct dm_target *ti,
1414 unsigned num_bios)
1415 {
1416 unsigned len;
1417
1418 /*
1419 * Even though the device advertised support for this type of
1420 * request, that does not mean every target supports it, and
1421 * reconfiguration might also have changed that since the
1422 * check was performed.
1423 */
1424 if (!num_bios)
1425 return -EOPNOTSUPP;
1426
1427 len = min_t(sector_t, ci->sector_count,
1428 max_io_len_target_boundary(ti, dm_target_offset(ti, ci->sector)));
1429
1430 __send_duplicate_bios(ci, ti, num_bios, &len);
1431
1432 ci->sector += len;
1433 ci->sector_count -= len;
1434
1435 return 0;
1436 }
1437
is_abnormal_io(struct bio * bio)1438 static bool is_abnormal_io(struct bio *bio)
1439 {
1440 bool r = false;
1441
1442 switch (bio_op(bio)) {
1443 case REQ_OP_DISCARD:
1444 case REQ_OP_SECURE_ERASE:
1445 case REQ_OP_WRITE_SAME:
1446 case REQ_OP_WRITE_ZEROES:
1447 r = true;
1448 break;
1449 }
1450
1451 return r;
1452 }
1453
__process_abnormal_io(struct clone_info * ci,struct dm_target * ti,int * result)1454 static bool __process_abnormal_io(struct clone_info *ci, struct dm_target *ti,
1455 int *result)
1456 {
1457 struct bio *bio = ci->bio;
1458 unsigned num_bios = 0;
1459
1460 switch (bio_op(bio)) {
1461 case REQ_OP_DISCARD:
1462 num_bios = ti->num_discard_bios;
1463 break;
1464 case REQ_OP_SECURE_ERASE:
1465 num_bios = ti->num_secure_erase_bios;
1466 break;
1467 case REQ_OP_WRITE_SAME:
1468 num_bios = ti->num_write_same_bios;
1469 break;
1470 case REQ_OP_WRITE_ZEROES:
1471 num_bios = ti->num_write_zeroes_bios;
1472 break;
1473 default:
1474 return false;
1475 }
1476
1477 *result = __send_changing_extent_only(ci, ti, num_bios);
1478 return true;
1479 }
1480
1481 /*
1482 * Select the correct strategy for processing a non-flush bio.
1483 */
__split_and_process_non_flush(struct clone_info * ci)1484 static int __split_and_process_non_flush(struct clone_info *ci)
1485 {
1486 struct dm_target *ti;
1487 unsigned len;
1488 int r;
1489
1490 ti = dm_table_find_target(ci->map, ci->sector);
1491 if (!ti)
1492 return -EIO;
1493
1494 if (__process_abnormal_io(ci, ti, &r))
1495 return r;
1496
1497 len = min_t(sector_t, max_io_len(ti, ci->sector), ci->sector_count);
1498
1499 r = __clone_and_map_data_bio(ci, ti, ci->sector, &len);
1500 if (r < 0)
1501 return r;
1502
1503 ci->sector += len;
1504 ci->sector_count -= len;
1505
1506 return 0;
1507 }
1508
init_clone_info(struct clone_info * ci,struct mapped_device * md,struct dm_table * map,struct bio * bio)1509 static void init_clone_info(struct clone_info *ci, struct mapped_device *md,
1510 struct dm_table *map, struct bio *bio)
1511 {
1512 ci->map = map;
1513 ci->io = alloc_io(md, bio);
1514 ci->sector = bio->bi_iter.bi_sector;
1515 }
1516
1517 #define __dm_part_stat_sub(part, field, subnd) \
1518 (part_stat_get(part, field) -= (subnd))
1519
1520 /*
1521 * Entry point to split a bio into clones and submit them to the targets.
1522 */
__split_and_process_bio(struct mapped_device * md,struct dm_table * map,struct bio * bio)1523 static blk_qc_t __split_and_process_bio(struct mapped_device *md,
1524 struct dm_table *map, struct bio *bio)
1525 {
1526 struct clone_info ci;
1527 blk_qc_t ret = BLK_QC_T_NONE;
1528 int error = 0;
1529
1530 init_clone_info(&ci, md, map, bio);
1531
1532 if (bio->bi_opf & REQ_PREFLUSH) {
1533 error = __send_empty_flush(&ci);
1534 /* dm_io_dec_pending submits any data associated with flush */
1535 } else if (op_is_zone_mgmt(bio_op(bio))) {
1536 ci.bio = bio;
1537 ci.sector_count = 0;
1538 error = __split_and_process_non_flush(&ci);
1539 } else {
1540 ci.bio = bio;
1541 ci.sector_count = bio_sectors(bio);
1542 error = __split_and_process_non_flush(&ci);
1543 if (ci.sector_count && !error) {
1544 /*
1545 * Remainder must be passed to submit_bio_noacct()
1546 * so that it gets handled *after* bios already submitted
1547 * have been completely processed.
1548 * We take a clone of the original to store in
1549 * ci.io->orig_bio to be used by end_io_acct() and
1550 * for dec_pending to use for completion handling.
1551 */
1552 struct bio *b = bio_split(bio, bio_sectors(bio) - ci.sector_count,
1553 GFP_NOIO, &md->queue->bio_split);
1554 ci.io->orig_bio = b;
1555
1556 /*
1557 * Adjust IO stats for each split, otherwise upon queue
1558 * reentry there will be redundant IO accounting.
1559 * NOTE: this is a stop-gap fix, a proper fix involves
1560 * significant refactoring of DM core's bio splitting
1561 * (by eliminating DM's splitting and just using bio_split)
1562 */
1563 part_stat_lock();
1564 __dm_part_stat_sub(dm_disk(md)->part0,
1565 sectors[op_stat_group(bio_op(bio))], ci.sector_count);
1566 part_stat_unlock();
1567
1568 bio_chain(b, bio);
1569 trace_block_split(b, bio->bi_iter.bi_sector);
1570 ret = submit_bio_noacct(bio);
1571 }
1572 }
1573
1574 /* drop the extra reference count */
1575 dm_io_dec_pending(ci.io, errno_to_blk_status(error));
1576 return ret;
1577 }
1578
dm_submit_bio(struct bio * bio)1579 static blk_qc_t dm_submit_bio(struct bio *bio)
1580 {
1581 struct mapped_device *md = bio->bi_bdev->bd_disk->private_data;
1582 blk_qc_t ret = BLK_QC_T_NONE;
1583 int srcu_idx;
1584 struct dm_table *map;
1585
1586 map = dm_get_live_table(md, &srcu_idx);
1587 if (unlikely(!map)) {
1588 DMERR_LIMIT("%s: mapping table unavailable, erroring io",
1589 dm_device_name(md));
1590 bio_io_error(bio);
1591 goto out;
1592 }
1593
1594 /* If suspended, queue this IO for later */
1595 if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) {
1596 if (bio->bi_opf & REQ_NOWAIT)
1597 bio_wouldblock_error(bio);
1598 else if (bio->bi_opf & REQ_RAHEAD)
1599 bio_io_error(bio);
1600 else
1601 queue_io(md, bio);
1602 goto out;
1603 }
1604
1605 /*
1606 * Use blk_queue_split() for abnormal IO (e.g. discard, writesame, etc)
1607 * otherwise associated queue_limits won't be imposed.
1608 */
1609 if (is_abnormal_io(bio))
1610 blk_queue_split(&bio);
1611
1612 ret = __split_and_process_bio(md, map, bio);
1613 out:
1614 dm_put_live_table(md, srcu_idx);
1615 return ret;
1616 }
1617
1618 /*-----------------------------------------------------------------
1619 * An IDR is used to keep track of allocated minor numbers.
1620 *---------------------------------------------------------------*/
free_minor(int minor)1621 static void free_minor(int minor)
1622 {
1623 spin_lock(&_minor_lock);
1624 idr_remove(&_minor_idr, minor);
1625 spin_unlock(&_minor_lock);
1626 }
1627
1628 /*
1629 * See if the device with a specific minor # is free.
1630 */
specific_minor(int minor)1631 static int specific_minor(int minor)
1632 {
1633 int r;
1634
1635 if (minor >= (1 << MINORBITS))
1636 return -EINVAL;
1637
1638 idr_preload(GFP_KERNEL);
1639 spin_lock(&_minor_lock);
1640
1641 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT);
1642
1643 spin_unlock(&_minor_lock);
1644 idr_preload_end();
1645 if (r < 0)
1646 return r == -ENOSPC ? -EBUSY : r;
1647 return 0;
1648 }
1649
next_free_minor(int * minor)1650 static int next_free_minor(int *minor)
1651 {
1652 int r;
1653
1654 idr_preload(GFP_KERNEL);
1655 spin_lock(&_minor_lock);
1656
1657 r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT);
1658
1659 spin_unlock(&_minor_lock);
1660 idr_preload_end();
1661 if (r < 0)
1662 return r;
1663 *minor = r;
1664 return 0;
1665 }
1666
1667 static const struct block_device_operations dm_blk_dops;
1668 static const struct block_device_operations dm_rq_blk_dops;
1669 static const struct dax_operations dm_dax_ops;
1670
1671 static void dm_wq_work(struct work_struct *work);
1672
1673 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
dm_queue_destroy_keyslot_manager(struct request_queue * q)1674 static void dm_queue_destroy_keyslot_manager(struct request_queue *q)
1675 {
1676 dm_destroy_keyslot_manager(q->ksm);
1677 }
1678
1679 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1680
dm_queue_destroy_keyslot_manager(struct request_queue * q)1681 static inline void dm_queue_destroy_keyslot_manager(struct request_queue *q)
1682 {
1683 }
1684 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1685
cleanup_mapped_device(struct mapped_device * md)1686 static void cleanup_mapped_device(struct mapped_device *md)
1687 {
1688 if (md->wq)
1689 destroy_workqueue(md->wq);
1690 bioset_exit(&md->bs);
1691 bioset_exit(&md->io_bs);
1692
1693 if (md->dax_dev) {
1694 kill_dax(md->dax_dev);
1695 put_dax(md->dax_dev);
1696 md->dax_dev = NULL;
1697 }
1698
1699 if (md->disk) {
1700 spin_lock(&_minor_lock);
1701 md->disk->private_data = NULL;
1702 spin_unlock(&_minor_lock);
1703 if (dm_get_md_type(md) != DM_TYPE_NONE) {
1704 dm_sysfs_exit(md);
1705 del_gendisk(md->disk);
1706 }
1707 dm_queue_destroy_keyslot_manager(md->queue);
1708 blk_cleanup_disk(md->disk);
1709 }
1710
1711 cleanup_srcu_struct(&md->io_barrier);
1712
1713 mutex_destroy(&md->suspend_lock);
1714 mutex_destroy(&md->type_lock);
1715 mutex_destroy(&md->table_devices_lock);
1716 mutex_destroy(&md->swap_bios_lock);
1717
1718 dm_mq_cleanup_mapped_device(md);
1719 dm_cleanup_zoned_dev(md);
1720 }
1721
1722 /*
1723 * Allocate and initialise a blank device with a given minor.
1724 */
alloc_dev(int minor)1725 static struct mapped_device *alloc_dev(int minor)
1726 {
1727 int r, numa_node_id = dm_get_numa_node();
1728 struct mapped_device *md;
1729 void *old_md;
1730
1731 md = kvzalloc_node(sizeof(*md), GFP_KERNEL, numa_node_id);
1732 if (!md) {
1733 DMWARN("unable to allocate device, out of memory.");
1734 return NULL;
1735 }
1736
1737 if (!try_module_get(THIS_MODULE))
1738 goto bad_module_get;
1739
1740 /* get a minor number for the dev */
1741 if (minor == DM_ANY_MINOR)
1742 r = next_free_minor(&minor);
1743 else
1744 r = specific_minor(minor);
1745 if (r < 0)
1746 goto bad_minor;
1747
1748 r = init_srcu_struct(&md->io_barrier);
1749 if (r < 0)
1750 goto bad_io_barrier;
1751
1752 md->numa_node_id = numa_node_id;
1753 md->init_tio_pdu = false;
1754 md->type = DM_TYPE_NONE;
1755 mutex_init(&md->suspend_lock);
1756 mutex_init(&md->type_lock);
1757 mutex_init(&md->table_devices_lock);
1758 spin_lock_init(&md->deferred_lock);
1759 atomic_set(&md->holders, 1);
1760 atomic_set(&md->open_count, 0);
1761 atomic_set(&md->event_nr, 0);
1762 atomic_set(&md->uevent_seq, 0);
1763 INIT_LIST_HEAD(&md->uevent_list);
1764 INIT_LIST_HEAD(&md->table_devices);
1765 spin_lock_init(&md->uevent_lock);
1766
1767 /*
1768 * default to bio-based until DM table is loaded and md->type
1769 * established. If request-based table is loaded: blk-mq will
1770 * override accordingly.
1771 */
1772 md->disk = blk_alloc_disk(md->numa_node_id);
1773 if (!md->disk)
1774 goto bad;
1775 md->queue = md->disk->queue;
1776
1777 init_waitqueue_head(&md->wait);
1778 INIT_WORK(&md->work, dm_wq_work);
1779 init_waitqueue_head(&md->eventq);
1780 init_completion(&md->kobj_holder.completion);
1781
1782 md->swap_bios = get_swap_bios();
1783 sema_init(&md->swap_bios_semaphore, md->swap_bios);
1784 mutex_init(&md->swap_bios_lock);
1785
1786 md->disk->major = _major;
1787 md->disk->first_minor = minor;
1788 md->disk->minors = 1;
1789 md->disk->fops = &dm_blk_dops;
1790 md->disk->queue = md->queue;
1791 md->disk->private_data = md;
1792 sprintf(md->disk->disk_name, "dm-%d", minor);
1793
1794 if (IS_ENABLED(CONFIG_DAX_DRIVER)) {
1795 md->dax_dev = alloc_dax(md, md->disk->disk_name,
1796 &dm_dax_ops, 0);
1797 if (IS_ERR(md->dax_dev))
1798 goto bad;
1799 }
1800
1801 format_dev_t(md->name, MKDEV(_major, minor));
1802
1803 md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0);
1804 if (!md->wq)
1805 goto bad;
1806
1807 dm_stats_init(&md->stats);
1808
1809 /* Populate the mapping, nobody knows we exist yet */
1810 spin_lock(&_minor_lock);
1811 old_md = idr_replace(&_minor_idr, md, minor);
1812 spin_unlock(&_minor_lock);
1813
1814 BUG_ON(old_md != MINOR_ALLOCED);
1815
1816 return md;
1817
1818 bad:
1819 cleanup_mapped_device(md);
1820 bad_io_barrier:
1821 free_minor(minor);
1822 bad_minor:
1823 module_put(THIS_MODULE);
1824 bad_module_get:
1825 kvfree(md);
1826 return NULL;
1827 }
1828
1829 static void unlock_fs(struct mapped_device *md);
1830
free_dev(struct mapped_device * md)1831 static void free_dev(struct mapped_device *md)
1832 {
1833 int minor = MINOR(disk_devt(md->disk));
1834
1835 unlock_fs(md);
1836
1837 cleanup_mapped_device(md);
1838
1839 free_table_devices(&md->table_devices);
1840 dm_stats_cleanup(&md->stats);
1841 free_minor(minor);
1842
1843 module_put(THIS_MODULE);
1844 kvfree(md);
1845 }
1846
__bind_mempools(struct mapped_device * md,struct dm_table * t)1847 static int __bind_mempools(struct mapped_device *md, struct dm_table *t)
1848 {
1849 struct dm_md_mempools *p = dm_table_get_md_mempools(t);
1850 int ret = 0;
1851
1852 if (dm_table_bio_based(t)) {
1853 /*
1854 * The md may already have mempools that need changing.
1855 * If so, reload bioset because front_pad may have changed
1856 * because a different table was loaded.
1857 */
1858 bioset_exit(&md->bs);
1859 bioset_exit(&md->io_bs);
1860
1861 } else if (bioset_initialized(&md->bs)) {
1862 /*
1863 * There's no need to reload with request-based dm
1864 * because the size of front_pad doesn't change.
1865 * Note for future: If you are to reload bioset,
1866 * prep-ed requests in the queue may refer
1867 * to bio from the old bioset, so you must walk
1868 * through the queue to unprep.
1869 */
1870 goto out;
1871 }
1872
1873 BUG_ON(!p ||
1874 bioset_initialized(&md->bs) ||
1875 bioset_initialized(&md->io_bs));
1876
1877 ret = bioset_init_from_src(&md->bs, &p->bs);
1878 if (ret)
1879 goto out;
1880 ret = bioset_init_from_src(&md->io_bs, &p->io_bs);
1881 if (ret)
1882 bioset_exit(&md->bs);
1883 out:
1884 /* mempool bind completed, no longer need any mempools in the table */
1885 dm_table_free_md_mempools(t);
1886 return ret;
1887 }
1888
1889 /*
1890 * Bind a table to the device.
1891 */
event_callback(void * context)1892 static void event_callback(void *context)
1893 {
1894 unsigned long flags;
1895 LIST_HEAD(uevents);
1896 struct mapped_device *md = (struct mapped_device *) context;
1897
1898 spin_lock_irqsave(&md->uevent_lock, flags);
1899 list_splice_init(&md->uevent_list, &uevents);
1900 spin_unlock_irqrestore(&md->uevent_lock, flags);
1901
1902 dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj);
1903
1904 atomic_inc(&md->event_nr);
1905 wake_up(&md->eventq);
1906 dm_issue_global_event();
1907 }
1908
1909 /*
1910 * Returns old map, which caller must destroy.
1911 */
__bind(struct mapped_device * md,struct dm_table * t,struct queue_limits * limits)1912 static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
1913 struct queue_limits *limits)
1914 {
1915 struct dm_table *old_map;
1916 struct request_queue *q = md->queue;
1917 bool request_based = dm_table_request_based(t);
1918 sector_t size;
1919 int ret;
1920
1921 lockdep_assert_held(&md->suspend_lock);
1922
1923 size = dm_table_get_size(t);
1924
1925 /*
1926 * Wipe any geometry if the size of the table changed.
1927 */
1928 if (size != dm_get_size(md))
1929 memset(&md->geometry, 0, sizeof(md->geometry));
1930
1931 if (!get_capacity(md->disk))
1932 set_capacity(md->disk, size);
1933 else
1934 set_capacity_and_notify(md->disk, size);
1935
1936 dm_table_event_callback(t, event_callback, md);
1937
1938 /*
1939 * The queue hasn't been stopped yet, if the old table type wasn't
1940 * for request-based during suspension. So stop it to prevent
1941 * I/O mapping before resume.
1942 * This must be done before setting the queue restrictions,
1943 * because request-based dm may be run just after the setting.
1944 */
1945 if (request_based)
1946 dm_stop_queue(q);
1947
1948 if (request_based) {
1949 /*
1950 * Leverage the fact that request-based DM targets are
1951 * immutable singletons - used to optimize dm_mq_queue_rq.
1952 */
1953 md->immutable_target = dm_table_get_immutable_target(t);
1954 }
1955
1956 ret = __bind_mempools(md, t);
1957 if (ret) {
1958 old_map = ERR_PTR(ret);
1959 goto out;
1960 }
1961
1962 ret = dm_table_set_restrictions(t, q, limits);
1963 if (ret) {
1964 old_map = ERR_PTR(ret);
1965 goto out;
1966 }
1967
1968 old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
1969 rcu_assign_pointer(md->map, (void *)t);
1970 md->immutable_target_type = dm_table_get_immutable_target_type(t);
1971
1972 if (old_map)
1973 dm_sync_table(md);
1974
1975 out:
1976 return old_map;
1977 }
1978
1979 /*
1980 * Returns unbound table for the caller to free.
1981 */
__unbind(struct mapped_device * md)1982 static struct dm_table *__unbind(struct mapped_device *md)
1983 {
1984 struct dm_table *map = rcu_dereference_protected(md->map, 1);
1985
1986 if (!map)
1987 return NULL;
1988
1989 dm_table_event_callback(map, NULL, NULL);
1990 RCU_INIT_POINTER(md->map, NULL);
1991 dm_sync_table(md);
1992
1993 return map;
1994 }
1995
1996 /*
1997 * Constructor for a new device.
1998 */
dm_create(int minor,struct mapped_device ** result)1999 int dm_create(int minor, struct mapped_device **result)
2000 {
2001 struct mapped_device *md;
2002
2003 md = alloc_dev(minor);
2004 if (!md)
2005 return -ENXIO;
2006
2007 dm_ima_reset_data(md);
2008
2009 *result = md;
2010 return 0;
2011 }
2012
2013 /*
2014 * Functions to manage md->type.
2015 * All are required to hold md->type_lock.
2016 */
dm_lock_md_type(struct mapped_device * md)2017 void dm_lock_md_type(struct mapped_device *md)
2018 {
2019 mutex_lock(&md->type_lock);
2020 }
2021
dm_unlock_md_type(struct mapped_device * md)2022 void dm_unlock_md_type(struct mapped_device *md)
2023 {
2024 mutex_unlock(&md->type_lock);
2025 }
2026
dm_set_md_type(struct mapped_device * md,enum dm_queue_mode type)2027 void dm_set_md_type(struct mapped_device *md, enum dm_queue_mode type)
2028 {
2029 BUG_ON(!mutex_is_locked(&md->type_lock));
2030 md->type = type;
2031 }
2032
dm_get_md_type(struct mapped_device * md)2033 enum dm_queue_mode dm_get_md_type(struct mapped_device *md)
2034 {
2035 return md->type;
2036 }
2037
dm_get_immutable_target_type(struct mapped_device * md)2038 struct target_type *dm_get_immutable_target_type(struct mapped_device *md)
2039 {
2040 return md->immutable_target_type;
2041 }
2042
2043 /*
2044 * The queue_limits are only valid as long as you have a reference
2045 * count on 'md'.
2046 */
dm_get_queue_limits(struct mapped_device * md)2047 struct queue_limits *dm_get_queue_limits(struct mapped_device *md)
2048 {
2049 BUG_ON(!atomic_read(&md->holders));
2050 return &md->queue->limits;
2051 }
2052 EXPORT_SYMBOL_GPL(dm_get_queue_limits);
2053
2054 /*
2055 * Setup the DM device's queue based on md's type
2056 */
dm_setup_md_queue(struct mapped_device * md,struct dm_table * t)2057 int dm_setup_md_queue(struct mapped_device *md, struct dm_table *t)
2058 {
2059 enum dm_queue_mode type = dm_table_get_type(t);
2060 struct queue_limits limits;
2061 int r;
2062
2063 switch (type) {
2064 case DM_TYPE_REQUEST_BASED:
2065 md->disk->fops = &dm_rq_blk_dops;
2066 r = dm_mq_init_request_queue(md, t);
2067 if (r) {
2068 DMERR("Cannot initialize queue for request-based dm mapped device");
2069 return r;
2070 }
2071 break;
2072 case DM_TYPE_BIO_BASED:
2073 case DM_TYPE_DAX_BIO_BASED:
2074 break;
2075 case DM_TYPE_NONE:
2076 WARN_ON_ONCE(true);
2077 break;
2078 }
2079
2080 r = dm_calculate_queue_limits(t, &limits);
2081 if (r) {
2082 DMERR("Cannot calculate initial queue limits");
2083 return r;
2084 }
2085 r = dm_table_set_restrictions(t, md->queue, &limits);
2086 if (r)
2087 return r;
2088
2089 add_disk(md->disk);
2090
2091 r = dm_sysfs_init(md);
2092 if (r) {
2093 del_gendisk(md->disk);
2094 return r;
2095 }
2096 md->type = type;
2097 return 0;
2098 }
2099
dm_get_md(dev_t dev)2100 struct mapped_device *dm_get_md(dev_t dev)
2101 {
2102 struct mapped_device *md;
2103 unsigned minor = MINOR(dev);
2104
2105 if (MAJOR(dev) != _major || minor >= (1 << MINORBITS))
2106 return NULL;
2107
2108 spin_lock(&_minor_lock);
2109
2110 md = idr_find(&_minor_idr, minor);
2111 if (!md || md == MINOR_ALLOCED || (MINOR(disk_devt(dm_disk(md))) != minor) ||
2112 test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2113 md = NULL;
2114 goto out;
2115 }
2116 dm_get(md);
2117 out:
2118 spin_unlock(&_minor_lock);
2119
2120 return md;
2121 }
2122 EXPORT_SYMBOL_GPL(dm_get_md);
2123
dm_get_mdptr(struct mapped_device * md)2124 void *dm_get_mdptr(struct mapped_device *md)
2125 {
2126 return md->interface_ptr;
2127 }
2128
dm_set_mdptr(struct mapped_device * md,void * ptr)2129 void dm_set_mdptr(struct mapped_device *md, void *ptr)
2130 {
2131 md->interface_ptr = ptr;
2132 }
2133
dm_get(struct mapped_device * md)2134 void dm_get(struct mapped_device *md)
2135 {
2136 atomic_inc(&md->holders);
2137 BUG_ON(test_bit(DMF_FREEING, &md->flags));
2138 }
2139
dm_hold(struct mapped_device * md)2140 int dm_hold(struct mapped_device *md)
2141 {
2142 spin_lock(&_minor_lock);
2143 if (test_bit(DMF_FREEING, &md->flags)) {
2144 spin_unlock(&_minor_lock);
2145 return -EBUSY;
2146 }
2147 dm_get(md);
2148 spin_unlock(&_minor_lock);
2149 return 0;
2150 }
2151 EXPORT_SYMBOL_GPL(dm_hold);
2152
dm_device_name(struct mapped_device * md)2153 const char *dm_device_name(struct mapped_device *md)
2154 {
2155 return md->name;
2156 }
2157 EXPORT_SYMBOL_GPL(dm_device_name);
2158
__dm_destroy(struct mapped_device * md,bool wait)2159 static void __dm_destroy(struct mapped_device *md, bool wait)
2160 {
2161 struct dm_table *map;
2162 int srcu_idx;
2163
2164 might_sleep();
2165
2166 spin_lock(&_minor_lock);
2167 idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md))));
2168 set_bit(DMF_FREEING, &md->flags);
2169 spin_unlock(&_minor_lock);
2170
2171 blk_set_queue_dying(md->queue);
2172
2173 /*
2174 * Take suspend_lock so that presuspend and postsuspend methods
2175 * do not race with internal suspend.
2176 */
2177 mutex_lock(&md->suspend_lock);
2178 map = dm_get_live_table(md, &srcu_idx);
2179 if (!dm_suspended_md(md)) {
2180 dm_table_presuspend_targets(map);
2181 set_bit(DMF_SUSPENDED, &md->flags);
2182 set_bit(DMF_POST_SUSPENDING, &md->flags);
2183 dm_table_postsuspend_targets(map);
2184 }
2185 /* dm_put_live_table must be before msleep, otherwise deadlock is possible */
2186 dm_put_live_table(md, srcu_idx);
2187 mutex_unlock(&md->suspend_lock);
2188
2189 /*
2190 * Rare, but there may be I/O requests still going to complete,
2191 * for example. Wait for all references to disappear.
2192 * No one should increment the reference count of the mapped_device,
2193 * after the mapped_device state becomes DMF_FREEING.
2194 */
2195 if (wait)
2196 while (atomic_read(&md->holders))
2197 msleep(1);
2198 else if (atomic_read(&md->holders))
2199 DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)",
2200 dm_device_name(md), atomic_read(&md->holders));
2201
2202 dm_table_destroy(__unbind(md));
2203 free_dev(md);
2204 }
2205
dm_destroy(struct mapped_device * md)2206 void dm_destroy(struct mapped_device *md)
2207 {
2208 __dm_destroy(md, true);
2209 }
2210
dm_destroy_immediate(struct mapped_device * md)2211 void dm_destroy_immediate(struct mapped_device *md)
2212 {
2213 __dm_destroy(md, false);
2214 }
2215
dm_put(struct mapped_device * md)2216 void dm_put(struct mapped_device *md)
2217 {
2218 atomic_dec(&md->holders);
2219 }
2220 EXPORT_SYMBOL_GPL(dm_put);
2221
md_in_flight_bios(struct mapped_device * md)2222 static bool md_in_flight_bios(struct mapped_device *md)
2223 {
2224 int cpu;
2225 struct block_device *part = dm_disk(md)->part0;
2226 long sum = 0;
2227
2228 for_each_possible_cpu(cpu) {
2229 sum += part_stat_local_read_cpu(part, in_flight[0], cpu);
2230 sum += part_stat_local_read_cpu(part, in_flight[1], cpu);
2231 }
2232
2233 return sum != 0;
2234 }
2235
dm_wait_for_bios_completion(struct mapped_device * md,unsigned int task_state)2236 static int dm_wait_for_bios_completion(struct mapped_device *md, unsigned int task_state)
2237 {
2238 int r = 0;
2239 DEFINE_WAIT(wait);
2240
2241 while (true) {
2242 prepare_to_wait(&md->wait, &wait, task_state);
2243
2244 if (!md_in_flight_bios(md))
2245 break;
2246
2247 if (signal_pending_state(task_state, current)) {
2248 r = -EINTR;
2249 break;
2250 }
2251
2252 io_schedule();
2253 }
2254 finish_wait(&md->wait, &wait);
2255
2256 return r;
2257 }
2258
dm_wait_for_completion(struct mapped_device * md,unsigned int task_state)2259 static int dm_wait_for_completion(struct mapped_device *md, unsigned int task_state)
2260 {
2261 int r = 0;
2262
2263 if (!queue_is_mq(md->queue))
2264 return dm_wait_for_bios_completion(md, task_state);
2265
2266 while (true) {
2267 if (!blk_mq_queue_inflight(md->queue))
2268 break;
2269
2270 if (signal_pending_state(task_state, current)) {
2271 r = -EINTR;
2272 break;
2273 }
2274
2275 msleep(5);
2276 }
2277
2278 return r;
2279 }
2280
2281 /*
2282 * Process the deferred bios
2283 */
dm_wq_work(struct work_struct * work)2284 static void dm_wq_work(struct work_struct *work)
2285 {
2286 struct mapped_device *md = container_of(work, struct mapped_device, work);
2287 struct bio *bio;
2288
2289 while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) {
2290 spin_lock_irq(&md->deferred_lock);
2291 bio = bio_list_pop(&md->deferred);
2292 spin_unlock_irq(&md->deferred_lock);
2293
2294 if (!bio)
2295 break;
2296
2297 submit_bio_noacct(bio);
2298 }
2299 }
2300
dm_queue_flush(struct mapped_device * md)2301 static void dm_queue_flush(struct mapped_device *md)
2302 {
2303 clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2304 smp_mb__after_atomic();
2305 queue_work(md->wq, &md->work);
2306 }
2307
2308 /*
2309 * Swap in a new table, returning the old one for the caller to destroy.
2310 */
dm_swap_table(struct mapped_device * md,struct dm_table * table)2311 struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table)
2312 {
2313 struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL);
2314 struct queue_limits limits;
2315 int r;
2316
2317 mutex_lock(&md->suspend_lock);
2318
2319 /* device must be suspended */
2320 if (!dm_suspended_md(md))
2321 goto out;
2322
2323 /*
2324 * If the new table has no data devices, retain the existing limits.
2325 * This helps multipath with queue_if_no_path if all paths disappear,
2326 * then new I/O is queued based on these limits, and then some paths
2327 * reappear.
2328 */
2329 if (dm_table_has_no_data_devices(table)) {
2330 live_map = dm_get_live_table_fast(md);
2331 if (live_map)
2332 limits = md->queue->limits;
2333 dm_put_live_table_fast(md);
2334 }
2335
2336 if (!live_map) {
2337 r = dm_calculate_queue_limits(table, &limits);
2338 if (r) {
2339 map = ERR_PTR(r);
2340 goto out;
2341 }
2342 }
2343
2344 map = __bind(md, table, &limits);
2345 dm_issue_global_event();
2346
2347 out:
2348 mutex_unlock(&md->suspend_lock);
2349 return map;
2350 }
2351
2352 /*
2353 * Functions to lock and unlock any filesystem running on the
2354 * device.
2355 */
lock_fs(struct mapped_device * md)2356 static int lock_fs(struct mapped_device *md)
2357 {
2358 int r;
2359
2360 WARN_ON(test_bit(DMF_FROZEN, &md->flags));
2361
2362 r = freeze_bdev(md->disk->part0);
2363 if (!r)
2364 set_bit(DMF_FROZEN, &md->flags);
2365 return r;
2366 }
2367
unlock_fs(struct mapped_device * md)2368 static void unlock_fs(struct mapped_device *md)
2369 {
2370 if (!test_bit(DMF_FROZEN, &md->flags))
2371 return;
2372 thaw_bdev(md->disk->part0);
2373 clear_bit(DMF_FROZEN, &md->flags);
2374 }
2375
2376 /*
2377 * @suspend_flags: DM_SUSPEND_LOCKFS_FLAG and/or DM_SUSPEND_NOFLUSH_FLAG
2378 * @task_state: e.g. TASK_INTERRUPTIBLE or TASK_UNINTERRUPTIBLE
2379 * @dmf_suspended_flag: DMF_SUSPENDED or DMF_SUSPENDED_INTERNALLY
2380 *
2381 * If __dm_suspend returns 0, the device is completely quiescent
2382 * now. There is no request-processing activity. All new requests
2383 * are being added to md->deferred list.
2384 */
__dm_suspend(struct mapped_device * md,struct dm_table * map,unsigned suspend_flags,unsigned int task_state,int dmf_suspended_flag)2385 static int __dm_suspend(struct mapped_device *md, struct dm_table *map,
2386 unsigned suspend_flags, unsigned int task_state,
2387 int dmf_suspended_flag)
2388 {
2389 bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG;
2390 bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG;
2391 int r;
2392
2393 lockdep_assert_held(&md->suspend_lock);
2394
2395 /*
2396 * DMF_NOFLUSH_SUSPENDING must be set before presuspend.
2397 * This flag is cleared before dm_suspend returns.
2398 */
2399 if (noflush)
2400 set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2401 else
2402 DMDEBUG("%s: suspending with flush", dm_device_name(md));
2403
2404 /*
2405 * This gets reverted if there's an error later and the targets
2406 * provide the .presuspend_undo hook.
2407 */
2408 dm_table_presuspend_targets(map);
2409
2410 /*
2411 * Flush I/O to the device.
2412 * Any I/O submitted after lock_fs() may not be flushed.
2413 * noflush takes precedence over do_lockfs.
2414 * (lock_fs() flushes I/Os and waits for them to complete.)
2415 */
2416 if (!noflush && do_lockfs) {
2417 r = lock_fs(md);
2418 if (r) {
2419 dm_table_presuspend_undo_targets(map);
2420 return r;
2421 }
2422 }
2423
2424 /*
2425 * Here we must make sure that no processes are submitting requests
2426 * to target drivers i.e. no one may be executing
2427 * __split_and_process_bio from dm_submit_bio.
2428 *
2429 * To get all processes out of __split_and_process_bio in dm_submit_bio,
2430 * we take the write lock. To prevent any process from reentering
2431 * __split_and_process_bio from dm_submit_bio and quiesce the thread
2432 * (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND and call
2433 * flush_workqueue(md->wq).
2434 */
2435 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2436 if (map)
2437 synchronize_srcu(&md->io_barrier);
2438
2439 /*
2440 * Stop md->queue before flushing md->wq in case request-based
2441 * dm defers requests to md->wq from md->queue.
2442 */
2443 if (dm_request_based(md))
2444 dm_stop_queue(md->queue);
2445
2446 flush_workqueue(md->wq);
2447
2448 /*
2449 * At this point no more requests are entering target request routines.
2450 * We call dm_wait_for_completion to wait for all existing requests
2451 * to finish.
2452 */
2453 r = dm_wait_for_completion(md, task_state);
2454 if (!r)
2455 set_bit(dmf_suspended_flag, &md->flags);
2456
2457 if (noflush)
2458 clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags);
2459 if (map)
2460 synchronize_srcu(&md->io_barrier);
2461
2462 /* were we interrupted ? */
2463 if (r < 0) {
2464 dm_queue_flush(md);
2465
2466 if (dm_request_based(md))
2467 dm_start_queue(md->queue);
2468
2469 unlock_fs(md);
2470 dm_table_presuspend_undo_targets(map);
2471 /* pushback list is already flushed, so skip flush */
2472 }
2473
2474 return r;
2475 }
2476
2477 /*
2478 * We need to be able to change a mapping table under a mounted
2479 * filesystem. For example we might want to move some data in
2480 * the background. Before the table can be swapped with
2481 * dm_bind_table, dm_suspend must be called to flush any in
2482 * flight bios and ensure that any further io gets deferred.
2483 */
2484 /*
2485 * Suspend mechanism in request-based dm.
2486 *
2487 * 1. Flush all I/Os by lock_fs() if needed.
2488 * 2. Stop dispatching any I/O by stopping the request_queue.
2489 * 3. Wait for all in-flight I/Os to be completed or requeued.
2490 *
2491 * To abort suspend, start the request_queue.
2492 */
dm_suspend(struct mapped_device * md,unsigned suspend_flags)2493 int dm_suspend(struct mapped_device *md, unsigned suspend_flags)
2494 {
2495 struct dm_table *map = NULL;
2496 int r = 0;
2497
2498 retry:
2499 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2500
2501 if (dm_suspended_md(md)) {
2502 r = -EINVAL;
2503 goto out_unlock;
2504 }
2505
2506 if (dm_suspended_internally_md(md)) {
2507 /* already internally suspended, wait for internal resume */
2508 mutex_unlock(&md->suspend_lock);
2509 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2510 if (r)
2511 return r;
2512 goto retry;
2513 }
2514
2515 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2516
2517 r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE, DMF_SUSPENDED);
2518 if (r)
2519 goto out_unlock;
2520
2521 set_bit(DMF_POST_SUSPENDING, &md->flags);
2522 dm_table_postsuspend_targets(map);
2523 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2524
2525 out_unlock:
2526 mutex_unlock(&md->suspend_lock);
2527 return r;
2528 }
2529
__dm_resume(struct mapped_device * md,struct dm_table * map)2530 static int __dm_resume(struct mapped_device *md, struct dm_table *map)
2531 {
2532 if (map) {
2533 int r = dm_table_resume_targets(map);
2534 if (r)
2535 return r;
2536 }
2537
2538 dm_queue_flush(md);
2539
2540 /*
2541 * Flushing deferred I/Os must be done after targets are resumed
2542 * so that mapping of targets can work correctly.
2543 * Request-based dm is queueing the deferred I/Os in its request_queue.
2544 */
2545 if (dm_request_based(md))
2546 dm_start_queue(md->queue);
2547
2548 unlock_fs(md);
2549
2550 return 0;
2551 }
2552
dm_resume(struct mapped_device * md)2553 int dm_resume(struct mapped_device *md)
2554 {
2555 int r;
2556 struct dm_table *map = NULL;
2557
2558 retry:
2559 r = -EINVAL;
2560 mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING);
2561
2562 if (!dm_suspended_md(md))
2563 goto out;
2564
2565 if (dm_suspended_internally_md(md)) {
2566 /* already internally suspended, wait for internal resume */
2567 mutex_unlock(&md->suspend_lock);
2568 r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE);
2569 if (r)
2570 return r;
2571 goto retry;
2572 }
2573
2574 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2575 if (!map || !dm_table_get_size(map))
2576 goto out;
2577
2578 r = __dm_resume(md, map);
2579 if (r)
2580 goto out;
2581
2582 clear_bit(DMF_SUSPENDED, &md->flags);
2583 out:
2584 mutex_unlock(&md->suspend_lock);
2585
2586 return r;
2587 }
2588
2589 /*
2590 * Internal suspend/resume works like userspace-driven suspend. It waits
2591 * until all bios finish and prevents issuing new bios to the target drivers.
2592 * It may be used only from the kernel.
2593 */
2594
__dm_internal_suspend(struct mapped_device * md,unsigned suspend_flags)2595 static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags)
2596 {
2597 struct dm_table *map = NULL;
2598
2599 lockdep_assert_held(&md->suspend_lock);
2600
2601 if (md->internal_suspend_count++)
2602 return; /* nested internal suspend */
2603
2604 if (dm_suspended_md(md)) {
2605 set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2606 return; /* nest suspend */
2607 }
2608
2609 map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock));
2610
2611 /*
2612 * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is
2613 * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend
2614 * would require changing .presuspend to return an error -- avoid this
2615 * until there is a need for more elaborate variants of internal suspend.
2616 */
2617 (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE,
2618 DMF_SUSPENDED_INTERNALLY);
2619
2620 set_bit(DMF_POST_SUSPENDING, &md->flags);
2621 dm_table_postsuspend_targets(map);
2622 clear_bit(DMF_POST_SUSPENDING, &md->flags);
2623 }
2624
__dm_internal_resume(struct mapped_device * md)2625 static void __dm_internal_resume(struct mapped_device *md)
2626 {
2627 BUG_ON(!md->internal_suspend_count);
2628
2629 if (--md->internal_suspend_count)
2630 return; /* resume from nested internal suspend */
2631
2632 if (dm_suspended_md(md))
2633 goto done; /* resume from nested suspend */
2634
2635 /*
2636 * NOTE: existing callers don't need to call dm_table_resume_targets
2637 * (which may fail -- so best to avoid it for now by passing NULL map)
2638 */
2639 (void) __dm_resume(md, NULL);
2640
2641 done:
2642 clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2643 smp_mb__after_atomic();
2644 wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY);
2645 }
2646
dm_internal_suspend_noflush(struct mapped_device * md)2647 void dm_internal_suspend_noflush(struct mapped_device *md)
2648 {
2649 mutex_lock(&md->suspend_lock);
2650 __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG);
2651 mutex_unlock(&md->suspend_lock);
2652 }
2653 EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush);
2654
dm_internal_resume(struct mapped_device * md)2655 void dm_internal_resume(struct mapped_device *md)
2656 {
2657 mutex_lock(&md->suspend_lock);
2658 __dm_internal_resume(md);
2659 mutex_unlock(&md->suspend_lock);
2660 }
2661 EXPORT_SYMBOL_GPL(dm_internal_resume);
2662
2663 /*
2664 * Fast variants of internal suspend/resume hold md->suspend_lock,
2665 * which prevents interaction with userspace-driven suspend.
2666 */
2667
dm_internal_suspend_fast(struct mapped_device * md)2668 void dm_internal_suspend_fast(struct mapped_device *md)
2669 {
2670 mutex_lock(&md->suspend_lock);
2671 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2672 return;
2673
2674 set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags);
2675 synchronize_srcu(&md->io_barrier);
2676 flush_workqueue(md->wq);
2677 dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE);
2678 }
2679 EXPORT_SYMBOL_GPL(dm_internal_suspend_fast);
2680
dm_internal_resume_fast(struct mapped_device * md)2681 void dm_internal_resume_fast(struct mapped_device *md)
2682 {
2683 if (dm_suspended_md(md) || dm_suspended_internally_md(md))
2684 goto done;
2685
2686 dm_queue_flush(md);
2687
2688 done:
2689 mutex_unlock(&md->suspend_lock);
2690 }
2691 EXPORT_SYMBOL_GPL(dm_internal_resume_fast);
2692
2693 /*-----------------------------------------------------------------
2694 * Event notification.
2695 *---------------------------------------------------------------*/
dm_kobject_uevent(struct mapped_device * md,enum kobject_action action,unsigned cookie)2696 int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action,
2697 unsigned cookie)
2698 {
2699 int r;
2700 unsigned noio_flag;
2701 char udev_cookie[DM_COOKIE_LENGTH];
2702 char *envp[] = { udev_cookie, NULL };
2703
2704 noio_flag = memalloc_noio_save();
2705
2706 if (!cookie)
2707 r = kobject_uevent(&disk_to_dev(md->disk)->kobj, action);
2708 else {
2709 snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u",
2710 DM_COOKIE_ENV_VAR_NAME, cookie);
2711 r = kobject_uevent_env(&disk_to_dev(md->disk)->kobj,
2712 action, envp);
2713 }
2714
2715 memalloc_noio_restore(noio_flag);
2716
2717 return r;
2718 }
2719
dm_next_uevent_seq(struct mapped_device * md)2720 uint32_t dm_next_uevent_seq(struct mapped_device *md)
2721 {
2722 return atomic_add_return(1, &md->uevent_seq);
2723 }
2724
dm_get_event_nr(struct mapped_device * md)2725 uint32_t dm_get_event_nr(struct mapped_device *md)
2726 {
2727 return atomic_read(&md->event_nr);
2728 }
2729
dm_wait_event(struct mapped_device * md,int event_nr)2730 int dm_wait_event(struct mapped_device *md, int event_nr)
2731 {
2732 return wait_event_interruptible(md->eventq,
2733 (event_nr != atomic_read(&md->event_nr)));
2734 }
2735
dm_uevent_add(struct mapped_device * md,struct list_head * elist)2736 void dm_uevent_add(struct mapped_device *md, struct list_head *elist)
2737 {
2738 unsigned long flags;
2739
2740 spin_lock_irqsave(&md->uevent_lock, flags);
2741 list_add(elist, &md->uevent_list);
2742 spin_unlock_irqrestore(&md->uevent_lock, flags);
2743 }
2744
2745 /*
2746 * The gendisk is only valid as long as you have a reference
2747 * count on 'md'.
2748 */
dm_disk(struct mapped_device * md)2749 struct gendisk *dm_disk(struct mapped_device *md)
2750 {
2751 return md->disk;
2752 }
2753 EXPORT_SYMBOL_GPL(dm_disk);
2754
dm_kobject(struct mapped_device * md)2755 struct kobject *dm_kobject(struct mapped_device *md)
2756 {
2757 return &md->kobj_holder.kobj;
2758 }
2759
dm_get_from_kobject(struct kobject * kobj)2760 struct mapped_device *dm_get_from_kobject(struct kobject *kobj)
2761 {
2762 struct mapped_device *md;
2763
2764 md = container_of(kobj, struct mapped_device, kobj_holder.kobj);
2765
2766 spin_lock(&_minor_lock);
2767 if (test_bit(DMF_FREEING, &md->flags) || dm_deleting_md(md)) {
2768 md = NULL;
2769 goto out;
2770 }
2771 dm_get(md);
2772 out:
2773 spin_unlock(&_minor_lock);
2774
2775 return md;
2776 }
2777
dm_suspended_md(struct mapped_device * md)2778 int dm_suspended_md(struct mapped_device *md)
2779 {
2780 return test_bit(DMF_SUSPENDED, &md->flags);
2781 }
2782
dm_post_suspending_md(struct mapped_device * md)2783 static int dm_post_suspending_md(struct mapped_device *md)
2784 {
2785 return test_bit(DMF_POST_SUSPENDING, &md->flags);
2786 }
2787
dm_suspended_internally_md(struct mapped_device * md)2788 int dm_suspended_internally_md(struct mapped_device *md)
2789 {
2790 return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags);
2791 }
2792
dm_test_deferred_remove_flag(struct mapped_device * md)2793 int dm_test_deferred_remove_flag(struct mapped_device *md)
2794 {
2795 return test_bit(DMF_DEFERRED_REMOVE, &md->flags);
2796 }
2797
dm_suspended(struct dm_target * ti)2798 int dm_suspended(struct dm_target *ti)
2799 {
2800 return dm_suspended_md(ti->table->md);
2801 }
2802 EXPORT_SYMBOL_GPL(dm_suspended);
2803
dm_post_suspending(struct dm_target * ti)2804 int dm_post_suspending(struct dm_target *ti)
2805 {
2806 return dm_post_suspending_md(ti->table->md);
2807 }
2808 EXPORT_SYMBOL_GPL(dm_post_suspending);
2809
dm_noflush_suspending(struct dm_target * ti)2810 int dm_noflush_suspending(struct dm_target *ti)
2811 {
2812 return __noflush_suspending(ti->table->md);
2813 }
2814 EXPORT_SYMBOL_GPL(dm_noflush_suspending);
2815
dm_alloc_md_mempools(struct mapped_device * md,enum dm_queue_mode type,unsigned integrity,unsigned per_io_data_size,unsigned min_pool_size)2816 struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, enum dm_queue_mode type,
2817 unsigned integrity, unsigned per_io_data_size,
2818 unsigned min_pool_size)
2819 {
2820 struct dm_md_mempools *pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
2821 unsigned int pool_size = 0;
2822 unsigned int front_pad, io_front_pad;
2823 int ret;
2824
2825 if (!pools)
2826 return NULL;
2827
2828 switch (type) {
2829 case DM_TYPE_BIO_BASED:
2830 case DM_TYPE_DAX_BIO_BASED:
2831 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
2832 front_pad = roundup(per_io_data_size, __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
2833 io_front_pad = roundup(per_io_data_size, __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
2834 ret = bioset_init(&pools->io_bs, pool_size, io_front_pad, 0);
2835 if (ret)
2836 goto out;
2837 if (integrity && bioset_integrity_create(&pools->io_bs, pool_size))
2838 goto out;
2839 break;
2840 case DM_TYPE_REQUEST_BASED:
2841 pool_size = max(dm_get_reserved_rq_based_ios(), min_pool_size);
2842 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
2843 /* per_io_data_size is used for blk-mq pdu at queue allocation */
2844 break;
2845 default:
2846 BUG();
2847 }
2848
2849 ret = bioset_init(&pools->bs, pool_size, front_pad, 0);
2850 if (ret)
2851 goto out;
2852
2853 if (integrity && bioset_integrity_create(&pools->bs, pool_size))
2854 goto out;
2855
2856 return pools;
2857
2858 out:
2859 dm_free_md_mempools(pools);
2860
2861 return NULL;
2862 }
2863
dm_free_md_mempools(struct dm_md_mempools * pools)2864 void dm_free_md_mempools(struct dm_md_mempools *pools)
2865 {
2866 if (!pools)
2867 return;
2868
2869 bioset_exit(&pools->bs);
2870 bioset_exit(&pools->io_bs);
2871
2872 kfree(pools);
2873 }
2874
2875 struct dm_pr {
2876 u64 old_key;
2877 u64 new_key;
2878 u32 flags;
2879 bool fail_early;
2880 };
2881
dm_call_pr(struct block_device * bdev,iterate_devices_callout_fn fn,void * data)2882 static int dm_call_pr(struct block_device *bdev, iterate_devices_callout_fn fn,
2883 void *data)
2884 {
2885 struct mapped_device *md = bdev->bd_disk->private_data;
2886 struct dm_table *table;
2887 struct dm_target *ti;
2888 int ret = -ENOTTY, srcu_idx;
2889
2890 table = dm_get_live_table(md, &srcu_idx);
2891 if (!table || !dm_table_get_size(table))
2892 goto out;
2893
2894 /* We only support devices that have a single target */
2895 if (dm_table_get_num_targets(table) != 1)
2896 goto out;
2897 ti = dm_table_get_target(table, 0);
2898
2899 ret = -EINVAL;
2900 if (!ti->type->iterate_devices)
2901 goto out;
2902
2903 ret = ti->type->iterate_devices(ti, fn, data);
2904 out:
2905 dm_put_live_table(md, srcu_idx);
2906 return ret;
2907 }
2908
2909 /*
2910 * For register / unregister we need to manually call out to every path.
2911 */
__dm_pr_register(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)2912 static int __dm_pr_register(struct dm_target *ti, struct dm_dev *dev,
2913 sector_t start, sector_t len, void *data)
2914 {
2915 struct dm_pr *pr = data;
2916 const struct pr_ops *ops = dev->bdev->bd_disk->fops->pr_ops;
2917
2918 if (!ops || !ops->pr_register)
2919 return -EOPNOTSUPP;
2920 return ops->pr_register(dev->bdev, pr->old_key, pr->new_key, pr->flags);
2921 }
2922
dm_pr_register(struct block_device * bdev,u64 old_key,u64 new_key,u32 flags)2923 static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
2924 u32 flags)
2925 {
2926 struct dm_pr pr = {
2927 .old_key = old_key,
2928 .new_key = new_key,
2929 .flags = flags,
2930 .fail_early = true,
2931 };
2932 int ret;
2933
2934 ret = dm_call_pr(bdev, __dm_pr_register, &pr);
2935 if (ret && new_key) {
2936 /* unregister all paths if we failed to register any path */
2937 pr.old_key = new_key;
2938 pr.new_key = 0;
2939 pr.flags = 0;
2940 pr.fail_early = false;
2941 dm_call_pr(bdev, __dm_pr_register, &pr);
2942 }
2943
2944 return ret;
2945 }
2946
dm_pr_reserve(struct block_device * bdev,u64 key,enum pr_type type,u32 flags)2947 static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
2948 u32 flags)
2949 {
2950 struct mapped_device *md = bdev->bd_disk->private_data;
2951 const struct pr_ops *ops;
2952 int r, srcu_idx;
2953
2954 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2955 if (r < 0)
2956 goto out;
2957
2958 ops = bdev->bd_disk->fops->pr_ops;
2959 if (ops && ops->pr_reserve)
2960 r = ops->pr_reserve(bdev, key, type, flags);
2961 else
2962 r = -EOPNOTSUPP;
2963 out:
2964 dm_unprepare_ioctl(md, srcu_idx);
2965 return r;
2966 }
2967
dm_pr_release(struct block_device * bdev,u64 key,enum pr_type type)2968 static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
2969 {
2970 struct mapped_device *md = bdev->bd_disk->private_data;
2971 const struct pr_ops *ops;
2972 int r, srcu_idx;
2973
2974 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2975 if (r < 0)
2976 goto out;
2977
2978 ops = bdev->bd_disk->fops->pr_ops;
2979 if (ops && ops->pr_release)
2980 r = ops->pr_release(bdev, key, type);
2981 else
2982 r = -EOPNOTSUPP;
2983 out:
2984 dm_unprepare_ioctl(md, srcu_idx);
2985 return r;
2986 }
2987
dm_pr_preempt(struct block_device * bdev,u64 old_key,u64 new_key,enum pr_type type,bool abort)2988 static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
2989 enum pr_type type, bool abort)
2990 {
2991 struct mapped_device *md = bdev->bd_disk->private_data;
2992 const struct pr_ops *ops;
2993 int r, srcu_idx;
2994
2995 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
2996 if (r < 0)
2997 goto out;
2998
2999 ops = bdev->bd_disk->fops->pr_ops;
3000 if (ops && ops->pr_preempt)
3001 r = ops->pr_preempt(bdev, old_key, new_key, type, abort);
3002 else
3003 r = -EOPNOTSUPP;
3004 out:
3005 dm_unprepare_ioctl(md, srcu_idx);
3006 return r;
3007 }
3008
dm_pr_clear(struct block_device * bdev,u64 key)3009 static int dm_pr_clear(struct block_device *bdev, u64 key)
3010 {
3011 struct mapped_device *md = bdev->bd_disk->private_data;
3012 const struct pr_ops *ops;
3013 int r, srcu_idx;
3014
3015 r = dm_prepare_ioctl(md, &srcu_idx, &bdev);
3016 if (r < 0)
3017 goto out;
3018
3019 ops = bdev->bd_disk->fops->pr_ops;
3020 if (ops && ops->pr_clear)
3021 r = ops->pr_clear(bdev, key);
3022 else
3023 r = -EOPNOTSUPP;
3024 out:
3025 dm_unprepare_ioctl(md, srcu_idx);
3026 return r;
3027 }
3028
3029 static const struct pr_ops dm_pr_ops = {
3030 .pr_register = dm_pr_register,
3031 .pr_reserve = dm_pr_reserve,
3032 .pr_release = dm_pr_release,
3033 .pr_preempt = dm_pr_preempt,
3034 .pr_clear = dm_pr_clear,
3035 };
3036
3037 static const struct block_device_operations dm_blk_dops = {
3038 .submit_bio = dm_submit_bio,
3039 .open = dm_blk_open,
3040 .release = dm_blk_close,
3041 .ioctl = dm_blk_ioctl,
3042 .getgeo = dm_blk_getgeo,
3043 .report_zones = dm_blk_report_zones,
3044 .pr_ops = &dm_pr_ops,
3045 .owner = THIS_MODULE
3046 };
3047
3048 static const struct block_device_operations dm_rq_blk_dops = {
3049 .open = dm_blk_open,
3050 .release = dm_blk_close,
3051 .ioctl = dm_blk_ioctl,
3052 .getgeo = dm_blk_getgeo,
3053 .pr_ops = &dm_pr_ops,
3054 .owner = THIS_MODULE
3055 };
3056
3057 static const struct dax_operations dm_dax_ops = {
3058 .direct_access = dm_dax_direct_access,
3059 .dax_supported = dm_dax_supported,
3060 .copy_from_iter = dm_dax_copy_from_iter,
3061 .copy_to_iter = dm_dax_copy_to_iter,
3062 .zero_page_range = dm_dax_zero_page_range,
3063 };
3064
3065 /*
3066 * module hooks
3067 */
3068 module_init(dm_init);
3069 module_exit(dm_exit);
3070
3071 module_param(major, uint, 0);
3072 MODULE_PARM_DESC(major, "The major number of the device mapper");
3073
3074 module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR);
3075 MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools");
3076
3077 module_param(dm_numa_node, int, S_IRUGO | S_IWUSR);
3078 MODULE_PARM_DESC(dm_numa_node, "NUMA node for DM device memory allocations");
3079
3080 module_param(swap_bios, int, S_IRUGO | S_IWUSR);
3081 MODULE_PARM_DESC(swap_bios, "Maximum allowed inflight swap IOs");
3082
3083 MODULE_DESCRIPTION(DM_NAME " driver");
3084 MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>");
3085 MODULE_LICENSE("GPL");
3086