1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2001 Sistina Software (UK) Limited.
4 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 *
6 * This file is released under the GPL.
7 */
8
9 #include "dm-core.h"
10 #include "dm-rq.h"
11
12 #include <linux/module.h>
13 #include <linux/vmalloc.h>
14 #include <linux/blkdev.h>
15 #include <linux/blk-integrity.h>
16 #include <linux/namei.h>
17 #include <linux/ctype.h>
18 #include <linux/string.h>
19 #include <linux/slab.h>
20 #include <linux/interrupt.h>
21 #include <linux/mutex.h>
22 #include <linux/delay.h>
23 #include <linux/atomic.h>
24 #include <linux/blk-mq.h>
25 #include <linux/mount.h>
26 #include <linux/dax.h>
27
28 #define DM_MSG_PREFIX "table"
29
30 #define NODE_SIZE L1_CACHE_BYTES
31 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
32 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
33
34 /*
35 * Similar to ceiling(log_size(n))
36 */
int_log(unsigned int n,unsigned int base)37 static unsigned int int_log(unsigned int n, unsigned int base)
38 {
39 int result = 0;
40
41 while (n > 1) {
42 n = dm_div_up(n, base);
43 result++;
44 }
45
46 return result;
47 }
48
49 /*
50 * Calculate the index of the child node of the n'th node k'th key.
51 */
get_child(unsigned int n,unsigned int k)52 static inline unsigned int get_child(unsigned int n, unsigned int k)
53 {
54 return (n * CHILDREN_PER_NODE) + k;
55 }
56
57 /*
58 * Return the n'th node of level l from table t.
59 */
get_node(struct dm_table * t,unsigned int l,unsigned int n)60 static inline sector_t *get_node(struct dm_table *t,
61 unsigned int l, unsigned int n)
62 {
63 return t->index[l] + (n * KEYS_PER_NODE);
64 }
65
66 /*
67 * Return the highest key that you could lookup from the n'th
68 * node on level l of the btree.
69 */
high(struct dm_table * t,unsigned int l,unsigned int n)70 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
71 {
72 for (; l < t->depth - 1; l++)
73 n = get_child(n, CHILDREN_PER_NODE - 1);
74
75 if (n >= t->counts[l])
76 return (sector_t) -1;
77
78 return get_node(t, l, n)[KEYS_PER_NODE - 1];
79 }
80
81 /*
82 * Fills in a level of the btree based on the highs of the level
83 * below it.
84 */
setup_btree_index(unsigned int l,struct dm_table * t)85 static int setup_btree_index(unsigned int l, struct dm_table *t)
86 {
87 unsigned int n, k;
88 sector_t *node;
89
90 for (n = 0U; n < t->counts[l]; n++) {
91 node = get_node(t, l, n);
92
93 for (k = 0U; k < KEYS_PER_NODE; k++)
94 node[k] = high(t, l + 1, get_child(n, k));
95 }
96
97 return 0;
98 }
99
100 /*
101 * highs, and targets are managed as dynamic arrays during a
102 * table load.
103 */
alloc_targets(struct dm_table * t,unsigned int num)104 static int alloc_targets(struct dm_table *t, unsigned int num)
105 {
106 sector_t *n_highs;
107 struct dm_target *n_targets;
108
109 /*
110 * Allocate both the target array and offset array at once.
111 */
112 n_highs = kvcalloc(num, sizeof(struct dm_target) + sizeof(sector_t),
113 GFP_KERNEL);
114 if (!n_highs)
115 return -ENOMEM;
116
117 n_targets = (struct dm_target *) (n_highs + num);
118
119 memset(n_highs, -1, sizeof(*n_highs) * num);
120 kvfree(t->highs);
121
122 t->num_allocated = num;
123 t->highs = n_highs;
124 t->targets = n_targets;
125
126 return 0;
127 }
128
dm_table_create(struct dm_table ** result,blk_mode_t mode,unsigned int num_targets,struct mapped_device * md)129 int dm_table_create(struct dm_table **result, blk_mode_t mode,
130 unsigned int num_targets, struct mapped_device *md)
131 {
132 struct dm_table *t;
133
134 if (num_targets > DM_MAX_TARGETS)
135 return -EOVERFLOW;
136
137 t = kzalloc(sizeof(*t), GFP_KERNEL);
138
139 if (!t)
140 return -ENOMEM;
141
142 INIT_LIST_HEAD(&t->devices);
143 init_rwsem(&t->devices_lock);
144
145 if (!num_targets)
146 num_targets = KEYS_PER_NODE;
147
148 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
149
150 if (!num_targets) {
151 kfree(t);
152 return -EOVERFLOW;
153 }
154
155 if (alloc_targets(t, num_targets)) {
156 kfree(t);
157 return -ENOMEM;
158 }
159
160 t->type = DM_TYPE_NONE;
161 t->mode = mode;
162 t->md = md;
163 t->flush_bypasses_map = true;
164 *result = t;
165 return 0;
166 }
167
free_devices(struct list_head * devices,struct mapped_device * md)168 static void free_devices(struct list_head *devices, struct mapped_device *md)
169 {
170 struct list_head *tmp, *next;
171
172 list_for_each_safe(tmp, next, devices) {
173 struct dm_dev_internal *dd =
174 list_entry(tmp, struct dm_dev_internal, list);
175 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
176 dm_device_name(md), dd->dm_dev->name);
177 dm_put_table_device(md, dd->dm_dev);
178 kfree(dd);
179 }
180 }
181
182 static void dm_table_destroy_crypto_profile(struct dm_table *t);
183
dm_table_destroy(struct dm_table * t)184 void dm_table_destroy(struct dm_table *t)
185 {
186 if (!t)
187 return;
188
189 /* free the indexes */
190 if (t->depth >= 2)
191 kvfree(t->index[t->depth - 2]);
192
193 /* free the targets */
194 for (unsigned int i = 0; i < t->num_targets; i++) {
195 struct dm_target *ti = dm_table_get_target(t, i);
196
197 if (ti->type->dtr)
198 ti->type->dtr(ti);
199
200 dm_put_target_type(ti->type);
201 }
202
203 kvfree(t->highs);
204
205 /* free the device list */
206 free_devices(&t->devices, t->md);
207
208 dm_free_md_mempools(t->mempools);
209
210 dm_table_destroy_crypto_profile(t);
211
212 kfree(t);
213 }
214
215 /*
216 * See if we've already got a device in the list.
217 */
find_device(struct list_head * l,dev_t dev)218 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
219 {
220 struct dm_dev_internal *dd;
221
222 list_for_each_entry(dd, l, list)
223 if (dd->dm_dev->bdev->bd_dev == dev)
224 return dd;
225
226 return NULL;
227 }
228
229 /*
230 * If possible, this checks an area of a destination device is invalid.
231 */
device_area_is_invalid(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)232 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
233 sector_t start, sector_t len, void *data)
234 {
235 struct queue_limits *limits = data;
236 struct block_device *bdev = dev->bdev;
237 sector_t dev_size = bdev_nr_sectors(bdev);
238 unsigned short logical_block_size_sectors =
239 limits->logical_block_size >> SECTOR_SHIFT;
240
241 if (!dev_size)
242 return 0;
243
244 if ((start >= dev_size) || (start + len > dev_size)) {
245 DMERR("%s: %pg too small for target: start=%llu, len=%llu, dev_size=%llu",
246 dm_device_name(ti->table->md), bdev,
247 (unsigned long long)start,
248 (unsigned long long)len,
249 (unsigned long long)dev_size);
250 return 1;
251 }
252
253 /*
254 * If the target is mapped to zoned block device(s), check
255 * that the zones are not partially mapped.
256 */
257 if (bdev_is_zoned(bdev)) {
258 unsigned int zone_sectors = bdev_zone_sectors(bdev);
259
260 if (start & (zone_sectors - 1)) {
261 DMERR("%s: start=%llu not aligned to h/w zone size %u of %pg",
262 dm_device_name(ti->table->md),
263 (unsigned long long)start,
264 zone_sectors, bdev);
265 return 1;
266 }
267
268 /*
269 * Note: The last zone of a zoned block device may be smaller
270 * than other zones. So for a target mapping the end of a
271 * zoned block device with such a zone, len would not be zone
272 * aligned. We do not allow such last smaller zone to be part
273 * of the mapping here to ensure that mappings with multiple
274 * devices do not end up with a smaller zone in the middle of
275 * the sector range.
276 */
277 if (len & (zone_sectors - 1)) {
278 DMERR("%s: len=%llu not aligned to h/w zone size %u of %pg",
279 dm_device_name(ti->table->md),
280 (unsigned long long)len,
281 zone_sectors, bdev);
282 return 1;
283 }
284 }
285
286 if (logical_block_size_sectors <= 1)
287 return 0;
288
289 if (start & (logical_block_size_sectors - 1)) {
290 DMERR("%s: start=%llu not aligned to h/w logical block size %u of %pg",
291 dm_device_name(ti->table->md),
292 (unsigned long long)start,
293 limits->logical_block_size, bdev);
294 return 1;
295 }
296
297 if (len & (logical_block_size_sectors - 1)) {
298 DMERR("%s: len=%llu not aligned to h/w logical block size %u of %pg",
299 dm_device_name(ti->table->md),
300 (unsigned long long)len,
301 limits->logical_block_size, bdev);
302 return 1;
303 }
304
305 return 0;
306 }
307
308 /*
309 * This upgrades the mode on an already open dm_dev, being
310 * careful to leave things as they were if we fail to reopen the
311 * device and not to touch the existing bdev field in case
312 * it is accessed concurrently.
313 */
upgrade_mode(struct dm_dev_internal * dd,blk_mode_t new_mode,struct mapped_device * md)314 static int upgrade_mode(struct dm_dev_internal *dd, blk_mode_t new_mode,
315 struct mapped_device *md)
316 {
317 int r;
318 struct dm_dev *old_dev, *new_dev;
319
320 old_dev = dd->dm_dev;
321
322 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
323 dd->dm_dev->mode | new_mode, &new_dev);
324 if (r)
325 return r;
326
327 dd->dm_dev = new_dev;
328 dm_put_table_device(md, old_dev);
329
330 return 0;
331 }
332
333 /*
334 * Note: the __ref annotation is because this function can call the __init
335 * marked early_lookup_bdev when called during early boot code from dm-init.c.
336 */
dm_devt_from_path(const char * path,dev_t * dev_p)337 int __ref dm_devt_from_path(const char *path, dev_t *dev_p)
338 {
339 int r;
340 dev_t dev;
341 unsigned int major, minor;
342 char dummy;
343
344 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
345 /* Extract the major/minor numbers */
346 dev = MKDEV(major, minor);
347 if (MAJOR(dev) != major || MINOR(dev) != minor)
348 return -EOVERFLOW;
349 } else {
350 r = lookup_bdev(path, &dev);
351 #ifndef MODULE
352 if (r && system_state < SYSTEM_RUNNING)
353 r = early_lookup_bdev(path, &dev);
354 #endif
355 if (r)
356 return r;
357 }
358 *dev_p = dev;
359 return 0;
360 }
361 EXPORT_SYMBOL(dm_devt_from_path);
362
363 /*
364 * Add a device to the list, or just increment the usage count if
365 * it's already present.
366 */
dm_get_device(struct dm_target * ti,const char * path,blk_mode_t mode,struct dm_dev ** result)367 int dm_get_device(struct dm_target *ti, const char *path, blk_mode_t mode,
368 struct dm_dev **result)
369 {
370 int r;
371 dev_t dev;
372 struct dm_dev_internal *dd;
373 struct dm_table *t = ti->table;
374
375 BUG_ON(!t);
376
377 r = dm_devt_from_path(path, &dev);
378 if (r)
379 return r;
380
381 if (dev == disk_devt(t->md->disk))
382 return -EINVAL;
383
384 down_write(&t->devices_lock);
385
386 dd = find_device(&t->devices, dev);
387 if (!dd) {
388 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
389 if (!dd) {
390 r = -ENOMEM;
391 goto unlock_ret_r;
392 }
393
394 r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev);
395 if (r) {
396 kfree(dd);
397 goto unlock_ret_r;
398 }
399
400 refcount_set(&dd->count, 1);
401 list_add(&dd->list, &t->devices);
402 goto out;
403
404 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
405 r = upgrade_mode(dd, mode, t->md);
406 if (r)
407 goto unlock_ret_r;
408 }
409 refcount_inc(&dd->count);
410 out:
411 up_write(&t->devices_lock);
412 *result = dd->dm_dev;
413 return 0;
414
415 unlock_ret_r:
416 up_write(&t->devices_lock);
417 return r;
418 }
419 EXPORT_SYMBOL(dm_get_device);
420
dm_set_device_limits(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)421 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
422 sector_t start, sector_t len, void *data)
423 {
424 struct queue_limits *limits = data;
425 struct block_device *bdev = dev->bdev;
426 struct request_queue *q = bdev_get_queue(bdev);
427
428 if (unlikely(!q)) {
429 DMWARN("%s: Cannot set limits for nonexistent device %pg",
430 dm_device_name(ti->table->md), bdev);
431 return 0;
432 }
433
434 if (blk_stack_limits(limits, &q->limits,
435 get_start_sect(bdev) + start) < 0)
436 DMWARN("%s: adding target device %pg caused an alignment inconsistency: "
437 "physical_block_size=%u, logical_block_size=%u, "
438 "alignment_offset=%u, start=%llu",
439 dm_device_name(ti->table->md), bdev,
440 q->limits.physical_block_size,
441 q->limits.logical_block_size,
442 q->limits.alignment_offset,
443 (unsigned long long) start << SECTOR_SHIFT);
444
445 /*
446 * Only stack the integrity profile if the target doesn't have native
447 * integrity support.
448 */
449 if (!dm_target_has_integrity(ti->type))
450 queue_limits_stack_integrity_bdev(limits, bdev);
451 return 0;
452 }
453
454 /*
455 * Decrement a device's use count and remove it if necessary.
456 */
dm_put_device(struct dm_target * ti,struct dm_dev * d)457 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
458 {
459 int found = 0;
460 struct dm_table *t = ti->table;
461 struct list_head *devices = &t->devices;
462 struct dm_dev_internal *dd;
463
464 down_write(&t->devices_lock);
465
466 list_for_each_entry(dd, devices, list) {
467 if (dd->dm_dev == d) {
468 found = 1;
469 break;
470 }
471 }
472 if (!found) {
473 DMERR("%s: device %s not in table devices list",
474 dm_device_name(t->md), d->name);
475 goto unlock_ret;
476 }
477 if (refcount_dec_and_test(&dd->count)) {
478 dm_put_table_device(t->md, d);
479 list_del(&dd->list);
480 kfree(dd);
481 }
482
483 unlock_ret:
484 up_write(&t->devices_lock);
485 }
486 EXPORT_SYMBOL(dm_put_device);
487
488 /*
489 * Checks to see if the target joins onto the end of the table.
490 */
adjoin(struct dm_table * t,struct dm_target * ti)491 static int adjoin(struct dm_table *t, struct dm_target *ti)
492 {
493 struct dm_target *prev;
494
495 if (!t->num_targets)
496 return !ti->begin;
497
498 prev = &t->targets[t->num_targets - 1];
499 return (ti->begin == (prev->begin + prev->len));
500 }
501
502 /*
503 * Used to dynamically allocate the arg array.
504 *
505 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
506 * process messages even if some device is suspended. These messages have a
507 * small fixed number of arguments.
508 *
509 * On the other hand, dm-switch needs to process bulk data using messages and
510 * excessive use of GFP_NOIO could cause trouble.
511 */
realloc_argv(unsigned int * size,char ** old_argv)512 static char **realloc_argv(unsigned int *size, char **old_argv)
513 {
514 char **argv;
515 unsigned int new_size;
516 gfp_t gfp;
517
518 if (*size) {
519 new_size = *size * 2;
520 gfp = GFP_KERNEL;
521 } else {
522 new_size = 8;
523 gfp = GFP_NOIO;
524 }
525 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
526 if (argv) {
527 if (old_argv)
528 memcpy(argv, old_argv, *size * sizeof(*argv));
529 *size = new_size;
530 }
531
532 kfree(old_argv);
533 return argv;
534 }
535
536 /*
537 * Destructively splits up the argument list to pass to ctr.
538 */
dm_split_args(int * argc,char *** argvp,char * input)539 int dm_split_args(int *argc, char ***argvp, char *input)
540 {
541 char *start, *end = input, *out, **argv = NULL;
542 unsigned int array_size = 0;
543
544 *argc = 0;
545
546 if (!input) {
547 *argvp = NULL;
548 return 0;
549 }
550
551 argv = realloc_argv(&array_size, argv);
552 if (!argv)
553 return -ENOMEM;
554
555 while (1) {
556 /* Skip whitespace */
557 start = skip_spaces(end);
558
559 if (!*start)
560 break; /* success, we hit the end */
561
562 /* 'out' is used to remove any back-quotes */
563 end = out = start;
564 while (*end) {
565 /* Everything apart from '\0' can be quoted */
566 if (*end == '\\' && *(end + 1)) {
567 *out++ = *(end + 1);
568 end += 2;
569 continue;
570 }
571
572 if (isspace(*end))
573 break; /* end of token */
574
575 *out++ = *end++;
576 }
577
578 /* have we already filled the array ? */
579 if ((*argc + 1) > array_size) {
580 argv = realloc_argv(&array_size, argv);
581 if (!argv)
582 return -ENOMEM;
583 }
584
585 /* we know this is whitespace */
586 if (*end)
587 end++;
588
589 /* terminate the string and put it in the array */
590 *out = '\0';
591 argv[*argc] = start;
592 (*argc)++;
593 }
594
595 *argvp = argv;
596 return 0;
597 }
598
dm_set_stacking_limits(struct queue_limits * limits)599 static void dm_set_stacking_limits(struct queue_limits *limits)
600 {
601 blk_set_stacking_limits(limits);
602 limits->features |= BLK_FEAT_IO_STAT | BLK_FEAT_NOWAIT | BLK_FEAT_POLL;
603 }
604
605 /*
606 * Impose necessary and sufficient conditions on a devices's table such
607 * that any incoming bio which respects its logical_block_size can be
608 * processed successfully. If it falls across the boundary between
609 * two or more targets, the size of each piece it gets split into must
610 * be compatible with the logical_block_size of the target processing it.
611 */
validate_hardware_logical_block_alignment(struct dm_table * t,struct queue_limits * limits)612 static int validate_hardware_logical_block_alignment(struct dm_table *t,
613 struct queue_limits *limits)
614 {
615 /*
616 * This function uses arithmetic modulo the logical_block_size
617 * (in units of 512-byte sectors).
618 */
619 unsigned short device_logical_block_size_sects =
620 limits->logical_block_size >> SECTOR_SHIFT;
621
622 /*
623 * Offset of the start of the next table entry, mod logical_block_size.
624 */
625 unsigned short next_target_start = 0;
626
627 /*
628 * Given an aligned bio that extends beyond the end of a
629 * target, how many sectors must the next target handle?
630 */
631 unsigned short remaining = 0;
632
633 struct dm_target *ti;
634 struct queue_limits ti_limits;
635 unsigned int i;
636
637 /*
638 * Check each entry in the table in turn.
639 */
640 for (i = 0; i < t->num_targets; i++) {
641 ti = dm_table_get_target(t, i);
642
643 dm_set_stacking_limits(&ti_limits);
644
645 /* combine all target devices' limits */
646 if (ti->type->iterate_devices)
647 ti->type->iterate_devices(ti, dm_set_device_limits,
648 &ti_limits);
649
650 /*
651 * If the remaining sectors fall entirely within this
652 * table entry are they compatible with its logical_block_size?
653 */
654 if (remaining < ti->len &&
655 remaining & ((ti_limits.logical_block_size >>
656 SECTOR_SHIFT) - 1))
657 break; /* Error */
658
659 next_target_start =
660 (unsigned short) ((next_target_start + ti->len) &
661 (device_logical_block_size_sects - 1));
662 remaining = next_target_start ?
663 device_logical_block_size_sects - next_target_start : 0;
664 }
665
666 if (remaining) {
667 DMERR("%s: table line %u (start sect %llu len %llu) "
668 "not aligned to h/w logical block size %u",
669 dm_device_name(t->md), i,
670 (unsigned long long) ti->begin,
671 (unsigned long long) ti->len,
672 limits->logical_block_size);
673 return -EINVAL;
674 }
675
676 return 0;
677 }
678
dm_table_add_target(struct dm_table * t,const char * type,sector_t start,sector_t len,char * params)679 int dm_table_add_target(struct dm_table *t, const char *type,
680 sector_t start, sector_t len, char *params)
681 {
682 int r = -EINVAL, argc;
683 char **argv;
684 struct dm_target *ti;
685
686 if (t->singleton) {
687 DMERR("%s: target type %s must appear alone in table",
688 dm_device_name(t->md), t->targets->type->name);
689 return -EINVAL;
690 }
691
692 BUG_ON(t->num_targets >= t->num_allocated);
693
694 ti = t->targets + t->num_targets;
695 memset(ti, 0, sizeof(*ti));
696
697 if (!len) {
698 DMERR("%s: zero-length target", dm_device_name(t->md));
699 return -EINVAL;
700 }
701 if (start + len < start || start + len > LLONG_MAX >> SECTOR_SHIFT) {
702 DMERR("%s: too large device", dm_device_name(t->md));
703 return -EINVAL;
704 }
705
706 ti->type = dm_get_target_type(type);
707 if (!ti->type) {
708 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
709 return -EINVAL;
710 }
711
712 if (dm_target_needs_singleton(ti->type)) {
713 if (t->num_targets) {
714 ti->error = "singleton target type must appear alone in table";
715 goto bad;
716 }
717 t->singleton = true;
718 }
719
720 if (dm_target_always_writeable(ti->type) &&
721 !(t->mode & BLK_OPEN_WRITE)) {
722 ti->error = "target type may not be included in a read-only table";
723 goto bad;
724 }
725
726 if (t->immutable_target_type) {
727 if (t->immutable_target_type != ti->type) {
728 ti->error = "immutable target type cannot be mixed with other target types";
729 goto bad;
730 }
731 } else if (dm_target_is_immutable(ti->type)) {
732 if (t->num_targets) {
733 ti->error = "immutable target type cannot be mixed with other target types";
734 goto bad;
735 }
736 t->immutable_target_type = ti->type;
737 }
738
739 ti->table = t;
740 ti->begin = start;
741 ti->len = len;
742 ti->error = "Unknown error";
743
744 /*
745 * Does this target adjoin the previous one ?
746 */
747 if (!adjoin(t, ti)) {
748 ti->error = "Gap in table";
749 goto bad;
750 }
751
752 r = dm_split_args(&argc, &argv, params);
753 if (r) {
754 ti->error = "couldn't split parameters";
755 goto bad;
756 }
757
758 r = ti->type->ctr(ti, argc, argv);
759 kfree(argv);
760 if (r)
761 goto bad;
762
763 t->highs[t->num_targets++] = ti->begin + ti->len - 1;
764
765 if (!ti->num_discard_bios && ti->discards_supported)
766 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
767 dm_device_name(t->md), type);
768
769 if (ti->limit_swap_bios && !static_key_enabled(&swap_bios_enabled.key))
770 static_branch_enable(&swap_bios_enabled);
771
772 if (!ti->flush_bypasses_map)
773 t->flush_bypasses_map = false;
774
775 return 0;
776
777 bad:
778 DMERR("%s: %s: %s (%pe)", dm_device_name(t->md), type, ti->error, ERR_PTR(r));
779 dm_put_target_type(ti->type);
780 return r;
781 }
782
783 /*
784 * Target argument parsing helpers.
785 */
validate_next_arg(const struct dm_arg * arg,struct dm_arg_set * arg_set,unsigned int * value,char ** error,unsigned int grouped)786 static int validate_next_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
787 unsigned int *value, char **error, unsigned int grouped)
788 {
789 const char *arg_str = dm_shift_arg(arg_set);
790 char dummy;
791
792 if (!arg_str ||
793 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
794 (*value < arg->min) ||
795 (*value > arg->max) ||
796 (grouped && arg_set->argc < *value)) {
797 *error = arg->error;
798 return -EINVAL;
799 }
800
801 return 0;
802 }
803
dm_read_arg(const struct dm_arg * arg,struct dm_arg_set * arg_set,unsigned int * value,char ** error)804 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
805 unsigned int *value, char **error)
806 {
807 return validate_next_arg(arg, arg_set, value, error, 0);
808 }
809 EXPORT_SYMBOL(dm_read_arg);
810
dm_read_arg_group(const struct dm_arg * arg,struct dm_arg_set * arg_set,unsigned int * value,char ** error)811 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
812 unsigned int *value, char **error)
813 {
814 return validate_next_arg(arg, arg_set, value, error, 1);
815 }
816 EXPORT_SYMBOL(dm_read_arg_group);
817
dm_shift_arg(struct dm_arg_set * as)818 const char *dm_shift_arg(struct dm_arg_set *as)
819 {
820 char *r;
821
822 if (as->argc) {
823 as->argc--;
824 r = *as->argv;
825 as->argv++;
826 return r;
827 }
828
829 return NULL;
830 }
831 EXPORT_SYMBOL(dm_shift_arg);
832
dm_consume_args(struct dm_arg_set * as,unsigned int num_args)833 void dm_consume_args(struct dm_arg_set *as, unsigned int num_args)
834 {
835 BUG_ON(as->argc < num_args);
836 as->argc -= num_args;
837 as->argv += num_args;
838 }
839 EXPORT_SYMBOL(dm_consume_args);
840
__table_type_bio_based(enum dm_queue_mode table_type)841 static bool __table_type_bio_based(enum dm_queue_mode table_type)
842 {
843 return (table_type == DM_TYPE_BIO_BASED ||
844 table_type == DM_TYPE_DAX_BIO_BASED);
845 }
846
__table_type_request_based(enum dm_queue_mode table_type)847 static bool __table_type_request_based(enum dm_queue_mode table_type)
848 {
849 return table_type == DM_TYPE_REQUEST_BASED;
850 }
851
dm_table_set_type(struct dm_table * t,enum dm_queue_mode type)852 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
853 {
854 t->type = type;
855 }
856 EXPORT_SYMBOL_GPL(dm_table_set_type);
857
858 /* validate the dax capability of the target device span */
device_not_dax_capable(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)859 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
860 sector_t start, sector_t len, void *data)
861 {
862 if (dev->dax_dev)
863 return false;
864
865 DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev);
866 return true;
867 }
868
869 /* Check devices support synchronous DAX */
device_not_dax_synchronous_capable(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)870 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
871 sector_t start, sector_t len, void *data)
872 {
873 return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
874 }
875
dm_table_supports_dax(struct dm_table * t,iterate_devices_callout_fn iterate_fn)876 static bool dm_table_supports_dax(struct dm_table *t,
877 iterate_devices_callout_fn iterate_fn)
878 {
879 /* Ensure that all targets support DAX. */
880 for (unsigned int i = 0; i < t->num_targets; i++) {
881 struct dm_target *ti = dm_table_get_target(t, i);
882
883 if (!ti->type->direct_access)
884 return false;
885
886 if (dm_target_is_wildcard(ti->type) ||
887 !ti->type->iterate_devices ||
888 ti->type->iterate_devices(ti, iterate_fn, NULL))
889 return false;
890 }
891
892 return true;
893 }
894
device_is_rq_stackable(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)895 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
896 sector_t start, sector_t len, void *data)
897 {
898 struct block_device *bdev = dev->bdev;
899 struct request_queue *q = bdev_get_queue(bdev);
900
901 /* request-based cannot stack on partitions! */
902 if (bdev_is_partition(bdev))
903 return false;
904
905 return queue_is_mq(q);
906 }
907
dm_table_determine_type(struct dm_table * t)908 static int dm_table_determine_type(struct dm_table *t)
909 {
910 unsigned int bio_based = 0, request_based = 0, hybrid = 0;
911 struct dm_target *ti;
912 struct list_head *devices = dm_table_get_devices(t);
913 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
914
915 if (t->type != DM_TYPE_NONE) {
916 /* target already set the table's type */
917 if (t->type == DM_TYPE_BIO_BASED) {
918 /* possibly upgrade to a variant of bio-based */
919 goto verify_bio_based;
920 }
921 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
922 goto verify_rq_based;
923 }
924
925 for (unsigned int i = 0; i < t->num_targets; i++) {
926 ti = dm_table_get_target(t, i);
927 if (dm_target_hybrid(ti))
928 hybrid = 1;
929 else if (dm_target_request_based(ti))
930 request_based = 1;
931 else
932 bio_based = 1;
933
934 if (bio_based && request_based) {
935 DMERR("Inconsistent table: different target types can't be mixed up");
936 return -EINVAL;
937 }
938 }
939
940 if (hybrid && !bio_based && !request_based) {
941 /*
942 * The targets can work either way.
943 * Determine the type from the live device.
944 * Default to bio-based if device is new.
945 */
946 if (__table_type_request_based(live_md_type))
947 request_based = 1;
948 else
949 bio_based = 1;
950 }
951
952 if (bio_based) {
953 verify_bio_based:
954 /* We must use this table as bio-based */
955 t->type = DM_TYPE_BIO_BASED;
956 if (dm_table_supports_dax(t, device_not_dax_capable) ||
957 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
958 t->type = DM_TYPE_DAX_BIO_BASED;
959 }
960 return 0;
961 }
962
963 BUG_ON(!request_based); /* No targets in this table */
964
965 t->type = DM_TYPE_REQUEST_BASED;
966
967 verify_rq_based:
968 /*
969 * Request-based dm supports only tables that have a single target now.
970 * To support multiple targets, request splitting support is needed,
971 * and that needs lots of changes in the block-layer.
972 * (e.g. request completion process for partial completion.)
973 */
974 if (t->num_targets > 1) {
975 DMERR("request-based DM doesn't support multiple targets");
976 return -EINVAL;
977 }
978
979 if (list_empty(devices)) {
980 int srcu_idx;
981 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
982
983 /* inherit live table's type */
984 if (live_table)
985 t->type = live_table->type;
986 dm_put_live_table(t->md, srcu_idx);
987 return 0;
988 }
989
990 ti = dm_table_get_immutable_target(t);
991 if (!ti) {
992 DMERR("table load rejected: immutable target is required");
993 return -EINVAL;
994 } else if (ti->max_io_len) {
995 DMERR("table load rejected: immutable target that splits IO is not supported");
996 return -EINVAL;
997 }
998
999 /* Non-request-stackable devices can't be used for request-based dm */
1000 if (!ti->type->iterate_devices ||
1001 !ti->type->iterate_devices(ti, device_is_rq_stackable, NULL)) {
1002 DMERR("table load rejected: including non-request-stackable devices");
1003 return -EINVAL;
1004 }
1005
1006 return 0;
1007 }
1008
dm_table_get_type(struct dm_table * t)1009 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1010 {
1011 return t->type;
1012 }
1013
dm_table_get_immutable_target_type(struct dm_table * t)1014 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1015 {
1016 return t->immutable_target_type;
1017 }
1018
dm_table_get_immutable_target(struct dm_table * t)1019 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1020 {
1021 /* Immutable target is implicitly a singleton */
1022 if (t->num_targets > 1 ||
1023 !dm_target_is_immutable(t->targets[0].type))
1024 return NULL;
1025
1026 return t->targets;
1027 }
1028
dm_table_get_wildcard_target(struct dm_table * t)1029 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1030 {
1031 for (unsigned int i = 0; i < t->num_targets; i++) {
1032 struct dm_target *ti = dm_table_get_target(t, i);
1033
1034 if (dm_target_is_wildcard(ti->type))
1035 return ti;
1036 }
1037
1038 return NULL;
1039 }
1040
dm_table_request_based(struct dm_table * t)1041 bool dm_table_request_based(struct dm_table *t)
1042 {
1043 return __table_type_request_based(dm_table_get_type(t));
1044 }
1045
dm_table_alloc_md_mempools(struct dm_table * t,struct mapped_device * md)1046 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1047 {
1048 enum dm_queue_mode type = dm_table_get_type(t);
1049 unsigned int per_io_data_size = 0, front_pad, io_front_pad;
1050 unsigned int min_pool_size = 0, pool_size;
1051 struct dm_md_mempools *pools;
1052 unsigned int bioset_flags = 0;
1053
1054 if (unlikely(type == DM_TYPE_NONE)) {
1055 DMERR("no table type is set, can't allocate mempools");
1056 return -EINVAL;
1057 }
1058
1059 pools = kzalloc_node(sizeof(*pools), GFP_KERNEL, md->numa_node_id);
1060 if (!pools)
1061 return -ENOMEM;
1062
1063 if (type == DM_TYPE_REQUEST_BASED) {
1064 pool_size = dm_get_reserved_rq_based_ios();
1065 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
1066 goto init_bs;
1067 }
1068
1069 if (md->queue->limits.features & BLK_FEAT_POLL)
1070 bioset_flags |= BIOSET_PERCPU_CACHE;
1071
1072 for (unsigned int i = 0; i < t->num_targets; i++) {
1073 struct dm_target *ti = dm_table_get_target(t, i);
1074
1075 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1076 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1077 }
1078 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
1079 front_pad = roundup(per_io_data_size,
1080 __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
1081
1082 io_front_pad = roundup(per_io_data_size,
1083 __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
1084 if (bioset_init(&pools->io_bs, pool_size, io_front_pad, bioset_flags))
1085 goto out_free_pools;
1086 init_bs:
1087 if (bioset_init(&pools->bs, pool_size, front_pad, 0))
1088 goto out_free_pools;
1089
1090 t->mempools = pools;
1091 return 0;
1092
1093 out_free_pools:
1094 dm_free_md_mempools(pools);
1095 return -ENOMEM;
1096 }
1097
setup_indexes(struct dm_table * t)1098 static int setup_indexes(struct dm_table *t)
1099 {
1100 int i;
1101 unsigned int total = 0;
1102 sector_t *indexes;
1103
1104 /* allocate the space for *all* the indexes */
1105 for (i = t->depth - 2; i >= 0; i--) {
1106 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1107 total += t->counts[i];
1108 }
1109
1110 indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1111 if (!indexes)
1112 return -ENOMEM;
1113
1114 /* set up internal nodes, bottom-up */
1115 for (i = t->depth - 2; i >= 0; i--) {
1116 t->index[i] = indexes;
1117 indexes += (KEYS_PER_NODE * t->counts[i]);
1118 setup_btree_index(i, t);
1119 }
1120
1121 return 0;
1122 }
1123
1124 /*
1125 * Builds the btree to index the map.
1126 */
dm_table_build_index(struct dm_table * t)1127 static int dm_table_build_index(struct dm_table *t)
1128 {
1129 int r = 0;
1130 unsigned int leaf_nodes;
1131
1132 /* how many indexes will the btree have ? */
1133 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1134 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1135
1136 /* leaf layer has already been set up */
1137 t->counts[t->depth - 1] = leaf_nodes;
1138 t->index[t->depth - 1] = t->highs;
1139
1140 if (t->depth >= 2)
1141 r = setup_indexes(t);
1142
1143 return r;
1144 }
1145
1146 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1147
1148 struct dm_crypto_profile {
1149 struct blk_crypto_profile profile;
1150 struct mapped_device *md;
1151 };
1152
dm_keyslot_evict_callback(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1153 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1154 sector_t start, sector_t len, void *data)
1155 {
1156 const struct blk_crypto_key *key = data;
1157
1158 blk_crypto_evict_key(dev->bdev, key);
1159 return 0;
1160 }
1161
1162 /*
1163 * When an inline encryption key is evicted from a device-mapper device, evict
1164 * it from all the underlying devices.
1165 */
dm_keyslot_evict(struct blk_crypto_profile * profile,const struct blk_crypto_key * key,unsigned int slot)1166 static int dm_keyslot_evict(struct blk_crypto_profile *profile,
1167 const struct blk_crypto_key *key, unsigned int slot)
1168 {
1169 struct mapped_device *md =
1170 container_of(profile, struct dm_crypto_profile, profile)->md;
1171 struct dm_table *t;
1172 int srcu_idx;
1173
1174 t = dm_get_live_table(md, &srcu_idx);
1175 if (!t)
1176 goto put_live_table;
1177
1178 for (unsigned int i = 0; i < t->num_targets; i++) {
1179 struct dm_target *ti = dm_table_get_target(t, i);
1180
1181 if (!ti->type->iterate_devices)
1182 continue;
1183 ti->type->iterate_devices(ti, dm_keyslot_evict_callback,
1184 (void *)key);
1185 }
1186
1187 put_live_table:
1188 dm_put_live_table(md, srcu_idx);
1189 return 0;
1190 }
1191
1192 static int
device_intersect_crypto_capabilities(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1193 device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev,
1194 sector_t start, sector_t len, void *data)
1195 {
1196 struct blk_crypto_profile *parent = data;
1197 struct blk_crypto_profile *child =
1198 bdev_get_queue(dev->bdev)->crypto_profile;
1199
1200 blk_crypto_intersect_capabilities(parent, child);
1201 return 0;
1202 }
1203
dm_destroy_crypto_profile(struct blk_crypto_profile * profile)1204 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1205 {
1206 struct dm_crypto_profile *dmcp = container_of(profile,
1207 struct dm_crypto_profile,
1208 profile);
1209
1210 if (!profile)
1211 return;
1212
1213 blk_crypto_profile_destroy(profile);
1214 kfree(dmcp);
1215 }
1216
dm_table_destroy_crypto_profile(struct dm_table * t)1217 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1218 {
1219 dm_destroy_crypto_profile(t->crypto_profile);
1220 t->crypto_profile = NULL;
1221 }
1222
1223 /*
1224 * Constructs and initializes t->crypto_profile with a crypto profile that
1225 * represents the common set of crypto capabilities of the devices described by
1226 * the dm_table. However, if the constructed crypto profile doesn't support all
1227 * crypto capabilities that are supported by the current mapped_device, it
1228 * returns an error instead, since we don't support removing crypto capabilities
1229 * on table changes. Finally, if the constructed crypto profile is "empty" (has
1230 * no crypto capabilities at all), it just sets t->crypto_profile to NULL.
1231 */
dm_table_construct_crypto_profile(struct dm_table * t)1232 static int dm_table_construct_crypto_profile(struct dm_table *t)
1233 {
1234 struct dm_crypto_profile *dmcp;
1235 struct blk_crypto_profile *profile;
1236 unsigned int i;
1237 bool empty_profile = true;
1238
1239 dmcp = kmalloc(sizeof(*dmcp), GFP_KERNEL);
1240 if (!dmcp)
1241 return -ENOMEM;
1242 dmcp->md = t->md;
1243
1244 profile = &dmcp->profile;
1245 blk_crypto_profile_init(profile, 0);
1246 profile->ll_ops.keyslot_evict = dm_keyslot_evict;
1247 profile->max_dun_bytes_supported = UINT_MAX;
1248 memset(profile->modes_supported, 0xFF,
1249 sizeof(profile->modes_supported));
1250 profile->key_types_supported = ~0;
1251
1252 for (i = 0; i < t->num_targets; i++) {
1253 struct dm_target *ti = dm_table_get_target(t, i);
1254
1255 if (!dm_target_passes_crypto(ti->type)) {
1256 blk_crypto_intersect_capabilities(profile, NULL);
1257 break;
1258 }
1259 if (!ti->type->iterate_devices)
1260 continue;
1261 ti->type->iterate_devices(ti,
1262 device_intersect_crypto_capabilities,
1263 profile);
1264 }
1265
1266 if (t->md->queue &&
1267 !blk_crypto_has_capabilities(profile,
1268 t->md->queue->crypto_profile)) {
1269 DMERR("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1270 dm_destroy_crypto_profile(profile);
1271 return -EINVAL;
1272 }
1273
1274 /*
1275 * If the new profile doesn't actually support any crypto capabilities,
1276 * we may as well represent it with a NULL profile.
1277 */
1278 for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) {
1279 if (profile->modes_supported[i]) {
1280 empty_profile = false;
1281 break;
1282 }
1283 }
1284
1285 if (empty_profile) {
1286 dm_destroy_crypto_profile(profile);
1287 profile = NULL;
1288 }
1289
1290 /*
1291 * t->crypto_profile is only set temporarily while the table is being
1292 * set up, and it gets set to NULL after the profile has been
1293 * transferred to the request_queue.
1294 */
1295 t->crypto_profile = profile;
1296
1297 return 0;
1298 }
1299
dm_update_crypto_profile(struct request_queue * q,struct dm_table * t)1300 static void dm_update_crypto_profile(struct request_queue *q,
1301 struct dm_table *t)
1302 {
1303 if (!t->crypto_profile)
1304 return;
1305
1306 /* Make the crypto profile less restrictive. */
1307 if (!q->crypto_profile) {
1308 blk_crypto_register(t->crypto_profile, q);
1309 } else {
1310 blk_crypto_update_capabilities(q->crypto_profile,
1311 t->crypto_profile);
1312 dm_destroy_crypto_profile(t->crypto_profile);
1313 }
1314 t->crypto_profile = NULL;
1315 }
1316
1317 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1318
dm_table_construct_crypto_profile(struct dm_table * t)1319 static int dm_table_construct_crypto_profile(struct dm_table *t)
1320 {
1321 return 0;
1322 }
1323
dm_destroy_crypto_profile(struct blk_crypto_profile * profile)1324 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1325 {
1326 }
1327
dm_table_destroy_crypto_profile(struct dm_table * t)1328 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1329 {
1330 }
1331
dm_update_crypto_profile(struct request_queue * q,struct dm_table * t)1332 static void dm_update_crypto_profile(struct request_queue *q,
1333 struct dm_table *t)
1334 {
1335 }
1336
1337 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1338
1339 /*
1340 * Prepares the table for use by building the indices,
1341 * setting the type, and allocating mempools.
1342 */
dm_table_complete(struct dm_table * t)1343 int dm_table_complete(struct dm_table *t)
1344 {
1345 int r;
1346
1347 r = dm_table_determine_type(t);
1348 if (r) {
1349 DMERR("unable to determine table type");
1350 return r;
1351 }
1352
1353 r = dm_table_build_index(t);
1354 if (r) {
1355 DMERR("unable to build btrees");
1356 return r;
1357 }
1358
1359 r = dm_table_construct_crypto_profile(t);
1360 if (r) {
1361 DMERR("could not construct crypto profile.");
1362 return r;
1363 }
1364
1365 r = dm_table_alloc_md_mempools(t, t->md);
1366 if (r)
1367 DMERR("unable to allocate mempools");
1368
1369 return r;
1370 }
1371
1372 static DEFINE_MUTEX(_event_lock);
dm_table_event_callback(struct dm_table * t,void (* fn)(void *),void * context)1373 void dm_table_event_callback(struct dm_table *t,
1374 void (*fn)(void *), void *context)
1375 {
1376 mutex_lock(&_event_lock);
1377 t->event_fn = fn;
1378 t->event_context = context;
1379 mutex_unlock(&_event_lock);
1380 }
1381
dm_table_event(struct dm_table * t)1382 void dm_table_event(struct dm_table *t)
1383 {
1384 mutex_lock(&_event_lock);
1385 if (t->event_fn)
1386 t->event_fn(t->event_context);
1387 mutex_unlock(&_event_lock);
1388 }
1389 EXPORT_SYMBOL(dm_table_event);
1390
dm_table_get_size(struct dm_table * t)1391 inline sector_t dm_table_get_size(struct dm_table *t)
1392 {
1393 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1394 }
1395 EXPORT_SYMBOL(dm_table_get_size);
1396
1397 /*
1398 * Search the btree for the correct target.
1399 *
1400 * Caller should check returned pointer for NULL
1401 * to trap I/O beyond end of device.
1402 */
dm_table_find_target(struct dm_table * t,sector_t sector)1403 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1404 {
1405 unsigned int l, n = 0, k = 0;
1406 sector_t *node;
1407
1408 if (unlikely(sector >= dm_table_get_size(t)))
1409 return NULL;
1410
1411 for (l = 0; l < t->depth; l++) {
1412 n = get_child(n, k);
1413 node = get_node(t, l, n);
1414
1415 for (k = 0; k < KEYS_PER_NODE; k++)
1416 if (node[k] >= sector)
1417 break;
1418 }
1419
1420 return &t->targets[(KEYS_PER_NODE * n) + k];
1421 }
1422
1423 /*
1424 * type->iterate_devices() should be called when the sanity check needs to
1425 * iterate and check all underlying data devices. iterate_devices() will
1426 * iterate all underlying data devices until it encounters a non-zero return
1427 * code, returned by whether the input iterate_devices_callout_fn, or
1428 * iterate_devices() itself internally.
1429 *
1430 * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1431 * iterate multiple underlying devices internally, in which case a non-zero
1432 * return code returned by iterate_devices_callout_fn will stop the iteration
1433 * in advance.
1434 *
1435 * Cases requiring _any_ underlying device supporting some kind of attribute,
1436 * should use the iteration structure like dm_table_any_dev_attr(), or call
1437 * it directly. @func should handle semantics of positive examples, e.g.
1438 * capable of something.
1439 *
1440 * Cases requiring _all_ underlying devices supporting some kind of attribute,
1441 * should use the iteration structure like dm_table_supports_nowait() or
1442 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1443 * uses an @anti_func that handle semantics of counter examples, e.g. not
1444 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1445 */
dm_table_any_dev_attr(struct dm_table * t,iterate_devices_callout_fn func,void * data)1446 static bool dm_table_any_dev_attr(struct dm_table *t,
1447 iterate_devices_callout_fn func, void *data)
1448 {
1449 for (unsigned int i = 0; i < t->num_targets; i++) {
1450 struct dm_target *ti = dm_table_get_target(t, i);
1451
1452 if (ti->type->iterate_devices &&
1453 ti->type->iterate_devices(ti, func, data))
1454 return true;
1455 }
1456
1457 return false;
1458 }
1459
count_device(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1460 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1461 sector_t start, sector_t len, void *data)
1462 {
1463 unsigned int *num_devices = data;
1464
1465 (*num_devices)++;
1466
1467 return 0;
1468 }
1469
1470 /*
1471 * Check whether a table has no data devices attached using each
1472 * target's iterate_devices method.
1473 * Returns false if the result is unknown because a target doesn't
1474 * support iterate_devices.
1475 */
dm_table_has_no_data_devices(struct dm_table * t)1476 bool dm_table_has_no_data_devices(struct dm_table *t)
1477 {
1478 for (unsigned int i = 0; i < t->num_targets; i++) {
1479 struct dm_target *ti = dm_table_get_target(t, i);
1480 unsigned int num_devices = 0;
1481
1482 if (!ti->type->iterate_devices)
1483 return false;
1484
1485 ti->type->iterate_devices(ti, count_device, &num_devices);
1486 if (num_devices)
1487 return false;
1488 }
1489
1490 return true;
1491 }
1492
device_not_zoned(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1493 static int device_not_zoned(struct dm_target *ti, struct dm_dev *dev,
1494 sector_t start, sector_t len, void *data)
1495 {
1496 bool *zoned = data;
1497
1498 return bdev_is_zoned(dev->bdev) != *zoned;
1499 }
1500
device_is_zoned_model(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1501 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1502 sector_t start, sector_t len, void *data)
1503 {
1504 return bdev_is_zoned(dev->bdev);
1505 }
1506
1507 /*
1508 * Check the device zoned model based on the target feature flag. If the target
1509 * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1510 * also accepted but all devices must have the same zoned model. If the target
1511 * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1512 * zoned model with all zoned devices having the same zone size.
1513 */
dm_table_supports_zoned(struct dm_table * t,bool zoned)1514 static bool dm_table_supports_zoned(struct dm_table *t, bool zoned)
1515 {
1516 for (unsigned int i = 0; i < t->num_targets; i++) {
1517 struct dm_target *ti = dm_table_get_target(t, i);
1518
1519 /*
1520 * For the wildcard target (dm-error), if we do not have a
1521 * backing device, we must always return false. If we have a
1522 * backing device, the result must depend on checking zoned
1523 * model, like for any other target. So for this, check directly
1524 * if the target backing device is zoned as we get "false" when
1525 * dm-error was set without a backing device.
1526 */
1527 if (dm_target_is_wildcard(ti->type) &&
1528 !ti->type->iterate_devices(ti, device_is_zoned_model, NULL))
1529 return false;
1530
1531 if (dm_target_supports_zoned_hm(ti->type)) {
1532 if (!ti->type->iterate_devices ||
1533 ti->type->iterate_devices(ti, device_not_zoned,
1534 &zoned))
1535 return false;
1536 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1537 if (zoned)
1538 return false;
1539 }
1540 }
1541
1542 return true;
1543 }
1544
device_not_matches_zone_sectors(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1545 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1546 sector_t start, sector_t len, void *data)
1547 {
1548 unsigned int *zone_sectors = data;
1549
1550 if (!bdev_is_zoned(dev->bdev))
1551 return 0;
1552 return bdev_zone_sectors(dev->bdev) != *zone_sectors;
1553 }
1554
1555 /*
1556 * Check consistency of zoned model and zone sectors across all targets. For
1557 * zone sectors, if the destination device is a zoned block device, it shall
1558 * have the specified zone_sectors.
1559 */
validate_hardware_zoned(struct dm_table * t,bool zoned,unsigned int zone_sectors)1560 static int validate_hardware_zoned(struct dm_table *t, bool zoned,
1561 unsigned int zone_sectors)
1562 {
1563 if (!zoned)
1564 return 0;
1565
1566 if (!dm_table_supports_zoned(t, zoned)) {
1567 DMERR("%s: zoned model is not consistent across all devices",
1568 dm_device_name(t->md));
1569 return -EINVAL;
1570 }
1571
1572 /* Check zone size validity and compatibility */
1573 if (!zone_sectors || !is_power_of_2(zone_sectors))
1574 return -EINVAL;
1575
1576 if (dm_table_any_dev_attr(t, device_not_matches_zone_sectors, &zone_sectors)) {
1577 DMERR("%s: zone sectors is not consistent across all zoned devices",
1578 dm_device_name(t->md));
1579 return -EINVAL;
1580 }
1581
1582 return 0;
1583 }
1584
1585 /*
1586 * Establish the new table's queue_limits and validate them.
1587 */
dm_calculate_queue_limits(struct dm_table * t,struct queue_limits * limits)1588 int dm_calculate_queue_limits(struct dm_table *t,
1589 struct queue_limits *limits)
1590 {
1591 struct queue_limits ti_limits;
1592 unsigned int zone_sectors = 0;
1593 bool zoned = false;
1594
1595 dm_set_stacking_limits(limits);
1596
1597 t->integrity_supported = true;
1598 for (unsigned int i = 0; i < t->num_targets; i++) {
1599 struct dm_target *ti = dm_table_get_target(t, i);
1600
1601 if (!dm_target_passes_integrity(ti->type))
1602 t->integrity_supported = false;
1603 }
1604
1605 for (unsigned int i = 0; i < t->num_targets; i++) {
1606 struct dm_target *ti = dm_table_get_target(t, i);
1607
1608 dm_set_stacking_limits(&ti_limits);
1609
1610 if (!ti->type->iterate_devices) {
1611 /* Set I/O hints portion of queue limits */
1612 if (ti->type->io_hints)
1613 ti->type->io_hints(ti, &ti_limits);
1614 goto combine_limits;
1615 }
1616
1617 /*
1618 * Combine queue limits of all the devices this target uses.
1619 */
1620 ti->type->iterate_devices(ti, dm_set_device_limits,
1621 &ti_limits);
1622
1623 if (!zoned && (ti_limits.features & BLK_FEAT_ZONED)) {
1624 /*
1625 * After stacking all limits, validate all devices
1626 * in table support this zoned model and zone sectors.
1627 */
1628 zoned = (ti_limits.features & BLK_FEAT_ZONED);
1629 zone_sectors = ti_limits.chunk_sectors;
1630 }
1631
1632 /* Set I/O hints portion of queue limits */
1633 if (ti->type->io_hints)
1634 ti->type->io_hints(ti, &ti_limits);
1635
1636 /*
1637 * Check each device area is consistent with the target's
1638 * overall queue limits.
1639 */
1640 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1641 &ti_limits))
1642 return -EINVAL;
1643
1644 combine_limits:
1645 /*
1646 * Merge this target's queue limits into the overall limits
1647 * for the table.
1648 */
1649 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1650 DMWARN("%s: adding target device (start sect %llu len %llu) "
1651 "caused an alignment inconsistency",
1652 dm_device_name(t->md),
1653 (unsigned long long) ti->begin,
1654 (unsigned long long) ti->len);
1655
1656 if (t->integrity_supported ||
1657 dm_target_has_integrity(ti->type)) {
1658 if (!queue_limits_stack_integrity(limits, &ti_limits)) {
1659 DMWARN("%s: adding target device (start sect %llu len %llu) "
1660 "disabled integrity support due to incompatibility",
1661 dm_device_name(t->md),
1662 (unsigned long long) ti->begin,
1663 (unsigned long long) ti->len);
1664 t->integrity_supported = false;
1665 }
1666 }
1667 }
1668
1669 /*
1670 * Verify that the zoned model and zone sectors, as determined before
1671 * any .io_hints override, are the same across all devices in the table.
1672 * - this is especially relevant if .io_hints is emulating a disk-managed
1673 * zoned model on host-managed zoned block devices.
1674 * BUT...
1675 */
1676 if (limits->features & BLK_FEAT_ZONED) {
1677 /*
1678 * ...IF the above limits stacking determined a zoned model
1679 * validate that all of the table's devices conform to it.
1680 */
1681 zoned = limits->features & BLK_FEAT_ZONED;
1682 zone_sectors = limits->chunk_sectors;
1683 }
1684 if (validate_hardware_zoned(t, zoned, zone_sectors))
1685 return -EINVAL;
1686
1687 return validate_hardware_logical_block_alignment(t, limits);
1688 }
1689
1690 /*
1691 * Check if a target requires flush support even if none of the underlying
1692 * devices need it (e.g. to persist target-specific metadata).
1693 */
dm_table_supports_flush(struct dm_table * t)1694 static bool dm_table_supports_flush(struct dm_table *t)
1695 {
1696 for (unsigned int i = 0; i < t->num_targets; i++) {
1697 struct dm_target *ti = dm_table_get_target(t, i);
1698
1699 if (ti->num_flush_bios && ti->flush_supported)
1700 return true;
1701 }
1702
1703 return false;
1704 }
1705
device_dax_write_cache_enabled(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1706 static int device_dax_write_cache_enabled(struct dm_target *ti,
1707 struct dm_dev *dev, sector_t start,
1708 sector_t len, void *data)
1709 {
1710 struct dax_device *dax_dev = dev->dax_dev;
1711
1712 if (!dax_dev)
1713 return false;
1714
1715 if (dax_write_cache_enabled(dax_dev))
1716 return true;
1717 return false;
1718 }
1719
device_not_write_zeroes_capable(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1720 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1721 sector_t start, sector_t len, void *data)
1722 {
1723 struct request_queue *q = bdev_get_queue(dev->bdev);
1724
1725 return !q->limits.max_write_zeroes_sectors;
1726 }
1727
dm_table_supports_write_zeroes(struct dm_table * t)1728 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1729 {
1730 for (unsigned int i = 0; i < t->num_targets; i++) {
1731 struct dm_target *ti = dm_table_get_target(t, i);
1732
1733 if (!ti->num_write_zeroes_bios)
1734 return false;
1735
1736 if (!ti->type->iterate_devices ||
1737 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1738 return false;
1739 }
1740
1741 return true;
1742 }
1743
dm_table_supports_nowait(struct dm_table * t)1744 static bool dm_table_supports_nowait(struct dm_table *t)
1745 {
1746 for (unsigned int i = 0; i < t->num_targets; i++) {
1747 struct dm_target *ti = dm_table_get_target(t, i);
1748
1749 if (!dm_target_supports_nowait(ti->type))
1750 return false;
1751 }
1752
1753 return true;
1754 }
1755
device_not_discard_capable(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1756 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1757 sector_t start, sector_t len, void *data)
1758 {
1759 return !bdev_max_discard_sectors(dev->bdev);
1760 }
1761
dm_table_supports_discards(struct dm_table * t)1762 static bool dm_table_supports_discards(struct dm_table *t)
1763 {
1764 for (unsigned int i = 0; i < t->num_targets; i++) {
1765 struct dm_target *ti = dm_table_get_target(t, i);
1766
1767 if (!ti->num_discard_bios)
1768 return false;
1769
1770 /*
1771 * Either the target provides discard support (as implied by setting
1772 * 'discards_supported') or it relies on _all_ data devices having
1773 * discard support.
1774 */
1775 if (!ti->discards_supported &&
1776 (!ti->type->iterate_devices ||
1777 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1778 return false;
1779 }
1780
1781 return true;
1782 }
1783
device_not_secure_erase_capable(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1784 static int device_not_secure_erase_capable(struct dm_target *ti,
1785 struct dm_dev *dev, sector_t start,
1786 sector_t len, void *data)
1787 {
1788 return !bdev_max_secure_erase_sectors(dev->bdev);
1789 }
1790
dm_table_supports_secure_erase(struct dm_table * t)1791 static bool dm_table_supports_secure_erase(struct dm_table *t)
1792 {
1793 for (unsigned int i = 0; i < t->num_targets; i++) {
1794 struct dm_target *ti = dm_table_get_target(t, i);
1795
1796 if (!ti->num_secure_erase_bios)
1797 return false;
1798
1799 if (!ti->type->iterate_devices ||
1800 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1801 return false;
1802 }
1803
1804 return true;
1805 }
1806
device_not_atomic_write_capable(struct dm_target * ti,struct dm_dev * dev,sector_t start,sector_t len,void * data)1807 static int device_not_atomic_write_capable(struct dm_target *ti,
1808 struct dm_dev *dev, sector_t start,
1809 sector_t len, void *data)
1810 {
1811 return !bdev_can_atomic_write(dev->bdev);
1812 }
1813
dm_table_supports_atomic_writes(struct dm_table * t)1814 static bool dm_table_supports_atomic_writes(struct dm_table *t)
1815 {
1816 for (unsigned int i = 0; i < t->num_targets; i++) {
1817 struct dm_target *ti = dm_table_get_target(t, i);
1818
1819 if (!dm_target_supports_atomic_writes(ti->type))
1820 return false;
1821
1822 if (!ti->type->iterate_devices)
1823 return false;
1824
1825 if (ti->type->iterate_devices(ti,
1826 device_not_atomic_write_capable, NULL)) {
1827 return false;
1828 }
1829 }
1830 return true;
1831 }
1832
dm_table_set_restrictions(struct dm_table * t,struct request_queue * q,struct queue_limits * limits)1833 int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1834 struct queue_limits *limits)
1835 {
1836 int r;
1837
1838 if (!dm_table_supports_nowait(t))
1839 limits->features &= ~BLK_FEAT_NOWAIT;
1840
1841 /*
1842 * The current polling impementation does not support request based
1843 * stacking.
1844 */
1845 if (!__table_type_bio_based(t->type))
1846 limits->features &= ~BLK_FEAT_POLL;
1847
1848 if (!dm_table_supports_discards(t)) {
1849 limits->max_hw_discard_sectors = 0;
1850 limits->discard_granularity = 0;
1851 limits->discard_alignment = 0;
1852 }
1853
1854 if (!dm_table_supports_write_zeroes(t))
1855 limits->max_write_zeroes_sectors = 0;
1856
1857 if (!dm_table_supports_secure_erase(t))
1858 limits->max_secure_erase_sectors = 0;
1859
1860 if (dm_table_supports_flush(t))
1861 limits->features |= BLK_FEAT_WRITE_CACHE | BLK_FEAT_FUA;
1862
1863 if (dm_table_supports_dax(t, device_not_dax_capable)) {
1864 limits->features |= BLK_FEAT_DAX;
1865 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable))
1866 set_dax_synchronous(t->md->dax_dev);
1867 } else
1868 limits->features &= ~BLK_FEAT_DAX;
1869
1870 if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
1871 dax_write_cache(t->md->dax_dev, true);
1872
1873 /* For a zoned table, setup the zone related queue attributes. */
1874 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
1875 (limits->features & BLK_FEAT_ZONED)) {
1876 r = dm_set_zones_restrictions(t, q, limits);
1877 if (r)
1878 return r;
1879 }
1880
1881 if (dm_table_supports_atomic_writes(t))
1882 limits->features |= BLK_FEAT_ATOMIC_WRITES;
1883
1884 r = queue_limits_set(q, limits);
1885 if (r)
1886 return r;
1887
1888 /*
1889 * Now that the limits are set, check the zones mapped by the table
1890 * and setup the resources for zone append emulation if necessary.
1891 */
1892 if (IS_ENABLED(CONFIG_BLK_DEV_ZONED) &&
1893 (limits->features & BLK_FEAT_ZONED)) {
1894 r = dm_revalidate_zones(t, q);
1895 if (r)
1896 return r;
1897 }
1898
1899 dm_update_crypto_profile(q, t);
1900 return 0;
1901 }
1902
dm_table_get_devices(struct dm_table * t)1903 struct list_head *dm_table_get_devices(struct dm_table *t)
1904 {
1905 return &t->devices;
1906 }
1907
dm_table_get_mode(struct dm_table * t)1908 blk_mode_t dm_table_get_mode(struct dm_table *t)
1909 {
1910 return t->mode;
1911 }
1912 EXPORT_SYMBOL(dm_table_get_mode);
1913
1914 enum suspend_mode {
1915 PRESUSPEND,
1916 PRESUSPEND_UNDO,
1917 POSTSUSPEND,
1918 };
1919
suspend_targets(struct dm_table * t,enum suspend_mode mode)1920 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1921 {
1922 lockdep_assert_held(&t->md->suspend_lock);
1923
1924 for (unsigned int i = 0; i < t->num_targets; i++) {
1925 struct dm_target *ti = dm_table_get_target(t, i);
1926
1927 switch (mode) {
1928 case PRESUSPEND:
1929 if (ti->type->presuspend)
1930 ti->type->presuspend(ti);
1931 break;
1932 case PRESUSPEND_UNDO:
1933 if (ti->type->presuspend_undo)
1934 ti->type->presuspend_undo(ti);
1935 break;
1936 case POSTSUSPEND:
1937 if (ti->type->postsuspend)
1938 ti->type->postsuspend(ti);
1939 break;
1940 }
1941 }
1942 }
1943
dm_table_presuspend_targets(struct dm_table * t)1944 void dm_table_presuspend_targets(struct dm_table *t)
1945 {
1946 if (!t)
1947 return;
1948
1949 suspend_targets(t, PRESUSPEND);
1950 }
1951
dm_table_presuspend_undo_targets(struct dm_table * t)1952 void dm_table_presuspend_undo_targets(struct dm_table *t)
1953 {
1954 if (!t)
1955 return;
1956
1957 suspend_targets(t, PRESUSPEND_UNDO);
1958 }
1959
dm_table_postsuspend_targets(struct dm_table * t)1960 void dm_table_postsuspend_targets(struct dm_table *t)
1961 {
1962 if (!t)
1963 return;
1964
1965 suspend_targets(t, POSTSUSPEND);
1966 }
1967
dm_table_resume_targets(struct dm_table * t)1968 int dm_table_resume_targets(struct dm_table *t)
1969 {
1970 unsigned int i;
1971 int r = 0;
1972
1973 lockdep_assert_held(&t->md->suspend_lock);
1974
1975 for (i = 0; i < t->num_targets; i++) {
1976 struct dm_target *ti = dm_table_get_target(t, i);
1977
1978 if (!ti->type->preresume)
1979 continue;
1980
1981 r = ti->type->preresume(ti);
1982 if (r) {
1983 DMERR("%s: %s: preresume failed, error = %d",
1984 dm_device_name(t->md), ti->type->name, r);
1985 return r;
1986 }
1987 }
1988
1989 for (i = 0; i < t->num_targets; i++) {
1990 struct dm_target *ti = dm_table_get_target(t, i);
1991
1992 if (ti->type->resume)
1993 ti->type->resume(ti);
1994 }
1995
1996 return 0;
1997 }
1998
dm_table_get_md(struct dm_table * t)1999 struct mapped_device *dm_table_get_md(struct dm_table *t)
2000 {
2001 return t->md;
2002 }
2003 EXPORT_SYMBOL(dm_table_get_md);
2004
dm_table_device_name(struct dm_table * t)2005 const char *dm_table_device_name(struct dm_table *t)
2006 {
2007 return dm_device_name(t->md);
2008 }
2009 EXPORT_SYMBOL_GPL(dm_table_device_name);
2010
dm_table_run_md_queue_async(struct dm_table * t)2011 void dm_table_run_md_queue_async(struct dm_table *t)
2012 {
2013 if (!dm_table_request_based(t))
2014 return;
2015
2016 if (t->md->queue)
2017 blk_mq_run_hw_queues(t->md->queue, true);
2018 }
2019 EXPORT_SYMBOL(dm_table_run_md_queue_async);
2020
2021