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 */
37static 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 */
52static 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 */
60static 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 */
70static 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 */
85static 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: l + 1, n: 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 */
104static 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(n: num, size: 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(addr: t->highs);
121
122 t->num_allocated = num;
123 t->highs = n_highs;
124 t->targets = n_targets;
125
126 return 0;
127}
128
129int 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 = kzalloc(size: sizeof(*t), GFP_KERNEL);
133
134 if (!t)
135 return -ENOMEM;
136
137 INIT_LIST_HEAD(list: &t->devices);
138 init_rwsem(&t->devices_lock);
139
140 if (!num_targets)
141 num_targets = KEYS_PER_NODE;
142
143 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
144
145 if (!num_targets) {
146 kfree(objp: t);
147 return -ENOMEM;
148 }
149
150 if (alloc_targets(t, num: num_targets)) {
151 kfree(objp: t);
152 return -ENOMEM;
153 }
154
155 t->type = DM_TYPE_NONE;
156 t->mode = mode;
157 t->md = md;
158 *result = t;
159 return 0;
160}
161
162static void free_devices(struct list_head *devices, struct mapped_device *md)
163{
164 struct list_head *tmp, *next;
165
166 list_for_each_safe(tmp, next, devices) {
167 struct dm_dev_internal *dd =
168 list_entry(tmp, struct dm_dev_internal, list);
169 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
170 dm_device_name(md), dd->dm_dev->name);
171 dm_put_table_device(md, d: dd->dm_dev);
172 kfree(objp: dd);
173 }
174}
175
176static void dm_table_destroy_crypto_profile(struct dm_table *t);
177
178void dm_table_destroy(struct dm_table *t)
179{
180 if (!t)
181 return;
182
183 /* free the indexes */
184 if (t->depth >= 2)
185 kvfree(addr: t->index[t->depth - 2]);
186
187 /* free the targets */
188 for (unsigned int i = 0; i < t->num_targets; i++) {
189 struct dm_target *ti = dm_table_get_target(t, index: i);
190
191 if (ti->type->dtr)
192 ti->type->dtr(ti);
193
194 dm_put_target_type(tt: ti->type);
195 }
196
197 kvfree(addr: t->highs);
198
199 /* free the device list */
200 free_devices(devices: &t->devices, md: t->md);
201
202 dm_free_md_mempools(pools: t->mempools);
203
204 dm_table_destroy_crypto_profile(t);
205
206 kfree(objp: t);
207}
208
209/*
210 * See if we've already got a device in the list.
211 */
212static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
213{
214 struct dm_dev_internal *dd;
215
216 list_for_each_entry(dd, l, list)
217 if (dd->dm_dev->bdev->bd_dev == dev)
218 return dd;
219
220 return NULL;
221}
222
223/*
224 * If possible, this checks an area of a destination device is invalid.
225 */
226static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
227 sector_t start, sector_t len, void *data)
228{
229 struct queue_limits *limits = data;
230 struct block_device *bdev = dev->bdev;
231 sector_t dev_size = bdev_nr_sectors(bdev);
232 unsigned short logical_block_size_sectors =
233 limits->logical_block_size >> SECTOR_SHIFT;
234
235 if (!dev_size)
236 return 0;
237
238 if ((start >= dev_size) || (start + len > dev_size)) {
239 DMERR("%s: %pg too small for target: start=%llu, len=%llu, dev_size=%llu",
240 dm_device_name(ti->table->md), bdev,
241 (unsigned long long)start,
242 (unsigned long long)len,
243 (unsigned long long)dev_size);
244 return 1;
245 }
246
247 /*
248 * If the target is mapped to zoned block device(s), check
249 * that the zones are not partially mapped.
250 */
251 if (bdev_is_zoned(bdev)) {
252 unsigned int zone_sectors = bdev_zone_sectors(bdev);
253
254 if (start & (zone_sectors - 1)) {
255 DMERR("%s: start=%llu not aligned to h/w zone size %u of %pg",
256 dm_device_name(ti->table->md),
257 (unsigned long long)start,
258 zone_sectors, bdev);
259 return 1;
260 }
261
262 /*
263 * Note: The last zone of a zoned block device may be smaller
264 * than other zones. So for a target mapping the end of a
265 * zoned block device with such a zone, len would not be zone
266 * aligned. We do not allow such last smaller zone to be part
267 * of the mapping here to ensure that mappings with multiple
268 * devices do not end up with a smaller zone in the middle of
269 * the sector range.
270 */
271 if (len & (zone_sectors - 1)) {
272 DMERR("%s: len=%llu not aligned to h/w zone size %u of %pg",
273 dm_device_name(ti->table->md),
274 (unsigned long long)len,
275 zone_sectors, bdev);
276 return 1;
277 }
278 }
279
280 if (logical_block_size_sectors <= 1)
281 return 0;
282
283 if (start & (logical_block_size_sectors - 1)) {
284 DMERR("%s: start=%llu not aligned to h/w logical block size %u of %pg",
285 dm_device_name(ti->table->md),
286 (unsigned long long)start,
287 limits->logical_block_size, bdev);
288 return 1;
289 }
290
291 if (len & (logical_block_size_sectors - 1)) {
292 DMERR("%s: len=%llu not aligned to h/w logical block size %u of %pg",
293 dm_device_name(ti->table->md),
294 (unsigned long long)len,
295 limits->logical_block_size, bdev);
296 return 1;
297 }
298
299 return 0;
300}
301
302/*
303 * This upgrades the mode on an already open dm_dev, being
304 * careful to leave things as they were if we fail to reopen the
305 * device and not to touch the existing bdev field in case
306 * it is accessed concurrently.
307 */
308static int upgrade_mode(struct dm_dev_internal *dd, blk_mode_t new_mode,
309 struct mapped_device *md)
310{
311 int r;
312 struct dm_dev *old_dev, *new_dev;
313
314 old_dev = dd->dm_dev;
315
316 r = dm_get_table_device(md, dev: dd->dm_dev->bdev->bd_dev,
317 mode: dd->dm_dev->mode | new_mode, result: &new_dev);
318 if (r)
319 return r;
320
321 dd->dm_dev = new_dev;
322 dm_put_table_device(md, d: old_dev);
323
324 return 0;
325}
326
327/*
328 * Add a device to the list, or just increment the usage count if
329 * it's already present.
330 *
331 * Note: the __ref annotation is because this function can call the __init
332 * marked early_lookup_bdev when called during early boot code from dm-init.c.
333 */
334int __ref dm_get_device(struct dm_target *ti, const char *path, blk_mode_t mode,
335 struct dm_dev **result)
336{
337 int r;
338 dev_t dev;
339 unsigned int major, minor;
340 char dummy;
341 struct dm_dev_internal *dd;
342 struct dm_table *t = ti->table;
343
344 BUG_ON(!t);
345
346 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
347 /* Extract the major/minor numbers */
348 dev = MKDEV(major, minor);
349 if (MAJOR(dev) != major || MINOR(dev) != minor)
350 return -EOVERFLOW;
351 } else {
352 r = lookup_bdev(pathname: path, dev: &dev);
353#ifndef MODULE
354 if (r && system_state < SYSTEM_RUNNING)
355 r = early_lookup_bdev(pathname: path, dev: &dev);
356#endif
357 if (r)
358 return r;
359 }
360 if (dev == disk_devt(disk: t->md->disk))
361 return -EINVAL;
362
363 down_write(sem: &t->devices_lock);
364
365 dd = find_device(l: &t->devices, dev);
366 if (!dd) {
367 dd = kmalloc(size: sizeof(*dd), GFP_KERNEL);
368 if (!dd) {
369 r = -ENOMEM;
370 goto unlock_ret_r;
371 }
372
373 r = dm_get_table_device(md: t->md, dev, mode, result: &dd->dm_dev);
374 if (r) {
375 kfree(objp: dd);
376 goto unlock_ret_r;
377 }
378
379 refcount_set(r: &dd->count, n: 1);
380 list_add(new: &dd->list, head: &t->devices);
381 goto out;
382
383 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
384 r = upgrade_mode(dd, new_mode: mode, md: t->md);
385 if (r)
386 goto unlock_ret_r;
387 }
388 refcount_inc(r: &dd->count);
389out:
390 up_write(sem: &t->devices_lock);
391 *result = dd->dm_dev;
392 return 0;
393
394unlock_ret_r:
395 up_write(sem: &t->devices_lock);
396 return r;
397}
398EXPORT_SYMBOL(dm_get_device);
399
400static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
401 sector_t start, sector_t len, void *data)
402{
403 struct queue_limits *limits = data;
404 struct block_device *bdev = dev->bdev;
405 struct request_queue *q = bdev_get_queue(bdev);
406
407 if (unlikely(!q)) {
408 DMWARN("%s: Cannot set limits for nonexistent device %pg",
409 dm_device_name(ti->table->md), bdev);
410 return 0;
411 }
412
413 if (blk_stack_limits(t: limits, b: &q->limits,
414 offset: get_start_sect(bdev) + start) < 0)
415 DMWARN("%s: adding target device %pg caused an alignment inconsistency: "
416 "physical_block_size=%u, logical_block_size=%u, "
417 "alignment_offset=%u, start=%llu",
418 dm_device_name(ti->table->md), bdev,
419 q->limits.physical_block_size,
420 q->limits.logical_block_size,
421 q->limits.alignment_offset,
422 (unsigned long long) start << SECTOR_SHIFT);
423 return 0;
424}
425
426/*
427 * Decrement a device's use count and remove it if necessary.
428 */
429void dm_put_device(struct dm_target *ti, struct dm_dev *d)
430{
431 int found = 0;
432 struct dm_table *t = ti->table;
433 struct list_head *devices = &t->devices;
434 struct dm_dev_internal *dd;
435
436 down_write(sem: &t->devices_lock);
437
438 list_for_each_entry(dd, devices, list) {
439 if (dd->dm_dev == d) {
440 found = 1;
441 break;
442 }
443 }
444 if (!found) {
445 DMERR("%s: device %s not in table devices list",
446 dm_device_name(t->md), d->name);
447 goto unlock_ret;
448 }
449 if (refcount_dec_and_test(r: &dd->count)) {
450 dm_put_table_device(md: t->md, d);
451 list_del(entry: &dd->list);
452 kfree(objp: dd);
453 }
454
455unlock_ret:
456 up_write(sem: &t->devices_lock);
457}
458EXPORT_SYMBOL(dm_put_device);
459
460/*
461 * Checks to see if the target joins onto the end of the table.
462 */
463static int adjoin(struct dm_table *t, struct dm_target *ti)
464{
465 struct dm_target *prev;
466
467 if (!t->num_targets)
468 return !ti->begin;
469
470 prev = &t->targets[t->num_targets - 1];
471 return (ti->begin == (prev->begin + prev->len));
472}
473
474/*
475 * Used to dynamically allocate the arg array.
476 *
477 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
478 * process messages even if some device is suspended. These messages have a
479 * small fixed number of arguments.
480 *
481 * On the other hand, dm-switch needs to process bulk data using messages and
482 * excessive use of GFP_NOIO could cause trouble.
483 */
484static char **realloc_argv(unsigned int *size, char **old_argv)
485{
486 char **argv;
487 unsigned int new_size;
488 gfp_t gfp;
489
490 if (*size) {
491 new_size = *size * 2;
492 gfp = GFP_KERNEL;
493 } else {
494 new_size = 8;
495 gfp = GFP_NOIO;
496 }
497 argv = kmalloc_array(n: new_size, size: sizeof(*argv), flags: gfp);
498 if (argv && old_argv) {
499 memcpy(argv, old_argv, *size * sizeof(*argv));
500 *size = new_size;
501 }
502
503 kfree(objp: old_argv);
504 return argv;
505}
506
507/*
508 * Destructively splits up the argument list to pass to ctr.
509 */
510int dm_split_args(int *argc, char ***argvp, char *input)
511{
512 char *start, *end = input, *out, **argv = NULL;
513 unsigned int array_size = 0;
514
515 *argc = 0;
516
517 if (!input) {
518 *argvp = NULL;
519 return 0;
520 }
521
522 argv = realloc_argv(size: &array_size, old_argv: argv);
523 if (!argv)
524 return -ENOMEM;
525
526 while (1) {
527 /* Skip whitespace */
528 start = skip_spaces(end);
529
530 if (!*start)
531 break; /* success, we hit the end */
532
533 /* 'out' is used to remove any back-quotes */
534 end = out = start;
535 while (*end) {
536 /* Everything apart from '\0' can be quoted */
537 if (*end == '\\' && *(end + 1)) {
538 *out++ = *(end + 1);
539 end += 2;
540 continue;
541 }
542
543 if (isspace(*end))
544 break; /* end of token */
545
546 *out++ = *end++;
547 }
548
549 /* have we already filled the array ? */
550 if ((*argc + 1) > array_size) {
551 argv = realloc_argv(size: &array_size, old_argv: argv);
552 if (!argv)
553 return -ENOMEM;
554 }
555
556 /* we know this is whitespace */
557 if (*end)
558 end++;
559
560 /* terminate the string and put it in the array */
561 *out = '\0';
562 argv[*argc] = start;
563 (*argc)++;
564 }
565
566 *argvp = argv;
567 return 0;
568}
569
570/*
571 * Impose necessary and sufficient conditions on a devices's table such
572 * that any incoming bio which respects its logical_block_size can be
573 * processed successfully. If it falls across the boundary between
574 * two or more targets, the size of each piece it gets split into must
575 * be compatible with the logical_block_size of the target processing it.
576 */
577static int validate_hardware_logical_block_alignment(struct dm_table *t,
578 struct queue_limits *limits)
579{
580 /*
581 * This function uses arithmetic modulo the logical_block_size
582 * (in units of 512-byte sectors).
583 */
584 unsigned short device_logical_block_size_sects =
585 limits->logical_block_size >> SECTOR_SHIFT;
586
587 /*
588 * Offset of the start of the next table entry, mod logical_block_size.
589 */
590 unsigned short next_target_start = 0;
591
592 /*
593 * Given an aligned bio that extends beyond the end of a
594 * target, how many sectors must the next target handle?
595 */
596 unsigned short remaining = 0;
597
598 struct dm_target *ti;
599 struct queue_limits ti_limits;
600 unsigned int i;
601
602 /*
603 * Check each entry in the table in turn.
604 */
605 for (i = 0; i < t->num_targets; i++) {
606 ti = dm_table_get_target(t, index: i);
607
608 blk_set_stacking_limits(lim: &ti_limits);
609
610 /* combine all target devices' limits */
611 if (ti->type->iterate_devices)
612 ti->type->iterate_devices(ti, dm_set_device_limits,
613 &ti_limits);
614
615 /*
616 * If the remaining sectors fall entirely within this
617 * table entry are they compatible with its logical_block_size?
618 */
619 if (remaining < ti->len &&
620 remaining & ((ti_limits.logical_block_size >>
621 SECTOR_SHIFT) - 1))
622 break; /* Error */
623
624 next_target_start =
625 (unsigned short) ((next_target_start + ti->len) &
626 (device_logical_block_size_sects - 1));
627 remaining = next_target_start ?
628 device_logical_block_size_sects - next_target_start : 0;
629 }
630
631 if (remaining) {
632 DMERR("%s: table line %u (start sect %llu len %llu) "
633 "not aligned to h/w logical block size %u",
634 dm_device_name(t->md), i,
635 (unsigned long long) ti->begin,
636 (unsigned long long) ti->len,
637 limits->logical_block_size);
638 return -EINVAL;
639 }
640
641 return 0;
642}
643
644int dm_table_add_target(struct dm_table *t, const char *type,
645 sector_t start, sector_t len, char *params)
646{
647 int r = -EINVAL, argc;
648 char **argv;
649 struct dm_target *ti;
650
651 if (t->singleton) {
652 DMERR("%s: target type %s must appear alone in table",
653 dm_device_name(t->md), t->targets->type->name);
654 return -EINVAL;
655 }
656
657 BUG_ON(t->num_targets >= t->num_allocated);
658
659 ti = t->targets + t->num_targets;
660 memset(ti, 0, sizeof(*ti));
661
662 if (!len) {
663 DMERR("%s: zero-length target", dm_device_name(t->md));
664 return -EINVAL;
665 }
666
667 ti->type = dm_get_target_type(name: type);
668 if (!ti->type) {
669 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
670 return -EINVAL;
671 }
672
673 if (dm_target_needs_singleton(ti->type)) {
674 if (t->num_targets) {
675 ti->error = "singleton target type must appear alone in table";
676 goto bad;
677 }
678 t->singleton = true;
679 }
680
681 if (dm_target_always_writeable(ti->type) &&
682 !(t->mode & BLK_OPEN_WRITE)) {
683 ti->error = "target type may not be included in a read-only table";
684 goto bad;
685 }
686
687 if (t->immutable_target_type) {
688 if (t->immutable_target_type != ti->type) {
689 ti->error = "immutable target type cannot be mixed with other target types";
690 goto bad;
691 }
692 } else if (dm_target_is_immutable(ti->type)) {
693 if (t->num_targets) {
694 ti->error = "immutable target type cannot be mixed with other target types";
695 goto bad;
696 }
697 t->immutable_target_type = ti->type;
698 }
699
700 if (dm_target_has_integrity(ti->type))
701 t->integrity_added = 1;
702
703 ti->table = t;
704 ti->begin = start;
705 ti->len = len;
706 ti->error = "Unknown error";
707
708 /*
709 * Does this target adjoin the previous one ?
710 */
711 if (!adjoin(t, ti)) {
712 ti->error = "Gap in table";
713 goto bad;
714 }
715
716 r = dm_split_args(argc: &argc, argvp: &argv, input: params);
717 if (r) {
718 ti->error = "couldn't split parameters";
719 goto bad;
720 }
721
722 r = ti->type->ctr(ti, argc, argv);
723 kfree(objp: argv);
724 if (r)
725 goto bad;
726
727 t->highs[t->num_targets++] = ti->begin + ti->len - 1;
728
729 if (!ti->num_discard_bios && ti->discards_supported)
730 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
731 dm_device_name(t->md), type);
732
733 if (ti->limit_swap_bios && !static_key_enabled(&swap_bios_enabled.key))
734 static_branch_enable(&swap_bios_enabled);
735
736 return 0;
737
738 bad:
739 DMERR("%s: %s: %s (%pe)", dm_device_name(t->md), type, ti->error, ERR_PTR(r));
740 dm_put_target_type(tt: ti->type);
741 return r;
742}
743
744/*
745 * Target argument parsing helpers.
746 */
747static int validate_next_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
748 unsigned int *value, char **error, unsigned int grouped)
749{
750 const char *arg_str = dm_shift_arg(as: arg_set);
751 char dummy;
752
753 if (!arg_str ||
754 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
755 (*value < arg->min) ||
756 (*value > arg->max) ||
757 (grouped && arg_set->argc < *value)) {
758 *error = arg->error;
759 return -EINVAL;
760 }
761
762 return 0;
763}
764
765int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
766 unsigned int *value, char **error)
767{
768 return validate_next_arg(arg, arg_set, value, error, grouped: 0);
769}
770EXPORT_SYMBOL(dm_read_arg);
771
772int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
773 unsigned int *value, char **error)
774{
775 return validate_next_arg(arg, arg_set, value, error, grouped: 1);
776}
777EXPORT_SYMBOL(dm_read_arg_group);
778
779const char *dm_shift_arg(struct dm_arg_set *as)
780{
781 char *r;
782
783 if (as->argc) {
784 as->argc--;
785 r = *as->argv;
786 as->argv++;
787 return r;
788 }
789
790 return NULL;
791}
792EXPORT_SYMBOL(dm_shift_arg);
793
794void dm_consume_args(struct dm_arg_set *as, unsigned int num_args)
795{
796 BUG_ON(as->argc < num_args);
797 as->argc -= num_args;
798 as->argv += num_args;
799}
800EXPORT_SYMBOL(dm_consume_args);
801
802static bool __table_type_bio_based(enum dm_queue_mode table_type)
803{
804 return (table_type == DM_TYPE_BIO_BASED ||
805 table_type == DM_TYPE_DAX_BIO_BASED);
806}
807
808static bool __table_type_request_based(enum dm_queue_mode table_type)
809{
810 return table_type == DM_TYPE_REQUEST_BASED;
811}
812
813void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
814{
815 t->type = type;
816}
817EXPORT_SYMBOL_GPL(dm_table_set_type);
818
819/* validate the dax capability of the target device span */
820static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
821 sector_t start, sector_t len, void *data)
822{
823 if (dev->dax_dev)
824 return false;
825
826 DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev);
827 return true;
828}
829
830/* Check devices support synchronous DAX */
831static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
832 sector_t start, sector_t len, void *data)
833{
834 return !dev->dax_dev || !dax_synchronous(dax_dev: dev->dax_dev);
835}
836
837static bool dm_table_supports_dax(struct dm_table *t,
838 iterate_devices_callout_fn iterate_fn)
839{
840 /* Ensure that all targets support DAX. */
841 for (unsigned int i = 0; i < t->num_targets; i++) {
842 struct dm_target *ti = dm_table_get_target(t, index: i);
843
844 if (!ti->type->direct_access)
845 return false;
846
847 if (dm_target_is_wildcard(ti->type) ||
848 !ti->type->iterate_devices ||
849 ti->type->iterate_devices(ti, iterate_fn, NULL))
850 return false;
851 }
852
853 return true;
854}
855
856static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
857 sector_t start, sector_t len, void *data)
858{
859 struct block_device *bdev = dev->bdev;
860 struct request_queue *q = bdev_get_queue(bdev);
861
862 /* request-based cannot stack on partitions! */
863 if (bdev_is_partition(bdev))
864 return false;
865
866 return queue_is_mq(q);
867}
868
869static int dm_table_determine_type(struct dm_table *t)
870{
871 unsigned int bio_based = 0, request_based = 0, hybrid = 0;
872 struct dm_target *ti;
873 struct list_head *devices = dm_table_get_devices(t);
874 enum dm_queue_mode live_md_type = dm_get_md_type(md: t->md);
875
876 if (t->type != DM_TYPE_NONE) {
877 /* target already set the table's type */
878 if (t->type == DM_TYPE_BIO_BASED) {
879 /* possibly upgrade to a variant of bio-based */
880 goto verify_bio_based;
881 }
882 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
883 goto verify_rq_based;
884 }
885
886 for (unsigned int i = 0; i < t->num_targets; i++) {
887 ti = dm_table_get_target(t, index: i);
888 if (dm_target_hybrid(ti))
889 hybrid = 1;
890 else if (dm_target_request_based(ti))
891 request_based = 1;
892 else
893 bio_based = 1;
894
895 if (bio_based && request_based) {
896 DMERR("Inconsistent table: different target types can't be mixed up");
897 return -EINVAL;
898 }
899 }
900
901 if (hybrid && !bio_based && !request_based) {
902 /*
903 * The targets can work either way.
904 * Determine the type from the live device.
905 * Default to bio-based if device is new.
906 */
907 if (__table_type_request_based(table_type: live_md_type))
908 request_based = 1;
909 else
910 bio_based = 1;
911 }
912
913 if (bio_based) {
914verify_bio_based:
915 /* We must use this table as bio-based */
916 t->type = DM_TYPE_BIO_BASED;
917 if (dm_table_supports_dax(t, iterate_fn: device_not_dax_capable) ||
918 (list_empty(head: devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
919 t->type = DM_TYPE_DAX_BIO_BASED;
920 }
921 return 0;
922 }
923
924 BUG_ON(!request_based); /* No targets in this table */
925
926 t->type = DM_TYPE_REQUEST_BASED;
927
928verify_rq_based:
929 /*
930 * Request-based dm supports only tables that have a single target now.
931 * To support multiple targets, request splitting support is needed,
932 * and that needs lots of changes in the block-layer.
933 * (e.g. request completion process for partial completion.)
934 */
935 if (t->num_targets > 1) {
936 DMERR("request-based DM doesn't support multiple targets");
937 return -EINVAL;
938 }
939
940 if (list_empty(head: devices)) {
941 int srcu_idx;
942 struct dm_table *live_table = dm_get_live_table(md: t->md, srcu_idx: &srcu_idx);
943
944 /* inherit live table's type */
945 if (live_table)
946 t->type = live_table->type;
947 dm_put_live_table(md: t->md, srcu_idx);
948 return 0;
949 }
950
951 ti = dm_table_get_immutable_target(t);
952 if (!ti) {
953 DMERR("table load rejected: immutable target is required");
954 return -EINVAL;
955 } else if (ti->max_io_len) {
956 DMERR("table load rejected: immutable target that splits IO is not supported");
957 return -EINVAL;
958 }
959
960 /* Non-request-stackable devices can't be used for request-based dm */
961 if (!ti->type->iterate_devices ||
962 !ti->type->iterate_devices(ti, device_is_rq_stackable, NULL)) {
963 DMERR("table load rejected: including non-request-stackable devices");
964 return -EINVAL;
965 }
966
967 return 0;
968}
969
970enum dm_queue_mode dm_table_get_type(struct dm_table *t)
971{
972 return t->type;
973}
974
975struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
976{
977 return t->immutable_target_type;
978}
979
980struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
981{
982 /* Immutable target is implicitly a singleton */
983 if (t->num_targets > 1 ||
984 !dm_target_is_immutable(t->targets[0].type))
985 return NULL;
986
987 return t->targets;
988}
989
990struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
991{
992 for (unsigned int i = 0; i < t->num_targets; i++) {
993 struct dm_target *ti = dm_table_get_target(t, index: i);
994
995 if (dm_target_is_wildcard(ti->type))
996 return ti;
997 }
998
999 return NULL;
1000}
1001
1002bool dm_table_bio_based(struct dm_table *t)
1003{
1004 return __table_type_bio_based(table_type: dm_table_get_type(t));
1005}
1006
1007bool dm_table_request_based(struct dm_table *t)
1008{
1009 return __table_type_request_based(table_type: dm_table_get_type(t));
1010}
1011
1012static bool dm_table_supports_poll(struct dm_table *t);
1013
1014static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1015{
1016 enum dm_queue_mode type = dm_table_get_type(t);
1017 unsigned int per_io_data_size = 0, front_pad, io_front_pad;
1018 unsigned int min_pool_size = 0, pool_size;
1019 struct dm_md_mempools *pools;
1020
1021 if (unlikely(type == DM_TYPE_NONE)) {
1022 DMERR("no table type is set, can't allocate mempools");
1023 return -EINVAL;
1024 }
1025
1026 pools = kzalloc_node(size: sizeof(*pools), GFP_KERNEL, node: md->numa_node_id);
1027 if (!pools)
1028 return -ENOMEM;
1029
1030 if (type == DM_TYPE_REQUEST_BASED) {
1031 pool_size = dm_get_reserved_rq_based_ios();
1032 front_pad = offsetof(struct dm_rq_clone_bio_info, clone);
1033 goto init_bs;
1034 }
1035
1036 for (unsigned int i = 0; i < t->num_targets; i++) {
1037 struct dm_target *ti = dm_table_get_target(t, index: i);
1038
1039 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1040 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1041 }
1042 pool_size = max(dm_get_reserved_bio_based_ios(), min_pool_size);
1043 front_pad = roundup(per_io_data_size,
1044 __alignof__(struct dm_target_io)) + DM_TARGET_IO_BIO_OFFSET;
1045
1046 io_front_pad = roundup(per_io_data_size,
1047 __alignof__(struct dm_io)) + DM_IO_BIO_OFFSET;
1048 if (bioset_init(&pools->io_bs, pool_size, io_front_pad,
1049 flags: dm_table_supports_poll(t) ? BIOSET_PERCPU_CACHE : 0))
1050 goto out_free_pools;
1051 if (t->integrity_supported &&
1052 bioset_integrity_create(&pools->io_bs, pool_size))
1053 goto out_free_pools;
1054init_bs:
1055 if (bioset_init(&pools->bs, pool_size, front_pad, flags: 0))
1056 goto out_free_pools;
1057 if (t->integrity_supported &&
1058 bioset_integrity_create(&pools->bs, pool_size))
1059 goto out_free_pools;
1060
1061 t->mempools = pools;
1062 return 0;
1063
1064out_free_pools:
1065 dm_free_md_mempools(pools);
1066 return -ENOMEM;
1067}
1068
1069static int setup_indexes(struct dm_table *t)
1070{
1071 int i;
1072 unsigned int total = 0;
1073 sector_t *indexes;
1074
1075 /* allocate the space for *all* the indexes */
1076 for (i = t->depth - 2; i >= 0; i--) {
1077 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1078 total += t->counts[i];
1079 }
1080
1081 indexes = kvcalloc(n: total, NODE_SIZE, GFP_KERNEL);
1082 if (!indexes)
1083 return -ENOMEM;
1084
1085 /* set up internal nodes, bottom-up */
1086 for (i = t->depth - 2; i >= 0; i--) {
1087 t->index[i] = indexes;
1088 indexes += (KEYS_PER_NODE * t->counts[i]);
1089 setup_btree_index(l: i, t);
1090 }
1091
1092 return 0;
1093}
1094
1095/*
1096 * Builds the btree to index the map.
1097 */
1098static int dm_table_build_index(struct dm_table *t)
1099{
1100 int r = 0;
1101 unsigned int leaf_nodes;
1102
1103 /* how many indexes will the btree have ? */
1104 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1105 t->depth = 1 + int_log(n: leaf_nodes, CHILDREN_PER_NODE);
1106
1107 /* leaf layer has already been set up */
1108 t->counts[t->depth - 1] = leaf_nodes;
1109 t->index[t->depth - 1] = t->highs;
1110
1111 if (t->depth >= 2)
1112 r = setup_indexes(t);
1113
1114 return r;
1115}
1116
1117static bool integrity_profile_exists(struct gendisk *disk)
1118{
1119 return !!blk_get_integrity(disk);
1120}
1121
1122/*
1123 * Get a disk whose integrity profile reflects the table's profile.
1124 * Returns NULL if integrity support was inconsistent or unavailable.
1125 */
1126static struct gendisk *dm_table_get_integrity_disk(struct dm_table *t)
1127{
1128 struct list_head *devices = dm_table_get_devices(t);
1129 struct dm_dev_internal *dd = NULL;
1130 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1131
1132 for (unsigned int i = 0; i < t->num_targets; i++) {
1133 struct dm_target *ti = dm_table_get_target(t, index: i);
1134
1135 if (!dm_target_passes_integrity(ti->type))
1136 goto no_integrity;
1137 }
1138
1139 list_for_each_entry(dd, devices, list) {
1140 template_disk = dd->dm_dev->bdev->bd_disk;
1141 if (!integrity_profile_exists(disk: template_disk))
1142 goto no_integrity;
1143 else if (prev_disk &&
1144 blk_integrity_compare(prev_disk, template_disk) < 0)
1145 goto no_integrity;
1146 prev_disk = template_disk;
1147 }
1148
1149 return template_disk;
1150
1151no_integrity:
1152 if (prev_disk)
1153 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1154 dm_device_name(t->md),
1155 prev_disk->disk_name,
1156 template_disk->disk_name);
1157 return NULL;
1158}
1159
1160/*
1161 * Register the mapped device for blk_integrity support if the
1162 * underlying devices have an integrity profile. But all devices may
1163 * not have matching profiles (checking all devices isn't reliable
1164 * during table load because this table may use other DM device(s) which
1165 * must be resumed before they will have an initialized integity
1166 * profile). Consequently, stacked DM devices force a 2 stage integrity
1167 * profile validation: First pass during table load, final pass during
1168 * resume.
1169 */
1170static int dm_table_register_integrity(struct dm_table *t)
1171{
1172 struct mapped_device *md = t->md;
1173 struct gendisk *template_disk = NULL;
1174
1175 /* If target handles integrity itself do not register it here. */
1176 if (t->integrity_added)
1177 return 0;
1178
1179 template_disk = dm_table_get_integrity_disk(t);
1180 if (!template_disk)
1181 return 0;
1182
1183 if (!integrity_profile_exists(disk: dm_disk(md))) {
1184 t->integrity_supported = true;
1185 /*
1186 * Register integrity profile during table load; we can do
1187 * this because the final profile must match during resume.
1188 */
1189 blk_integrity_register(dm_disk(md),
1190 blk_get_integrity(disk: template_disk));
1191 return 0;
1192 }
1193
1194 /*
1195 * If DM device already has an initialized integrity
1196 * profile the new profile should not conflict.
1197 */
1198 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1199 DMERR("%s: conflict with existing integrity profile: %s profile mismatch",
1200 dm_device_name(t->md),
1201 template_disk->disk_name);
1202 return 1;
1203 }
1204
1205 /* Preserve existing integrity profile */
1206 t->integrity_supported = true;
1207 return 0;
1208}
1209
1210#ifdef CONFIG_BLK_INLINE_ENCRYPTION
1211
1212struct dm_crypto_profile {
1213 struct blk_crypto_profile profile;
1214 struct mapped_device *md;
1215};
1216
1217static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1218 sector_t start, sector_t len, void *data)
1219{
1220 const struct blk_crypto_key *key = data;
1221
1222 blk_crypto_evict_key(bdev: dev->bdev, key);
1223 return 0;
1224}
1225
1226/*
1227 * When an inline encryption key is evicted from a device-mapper device, evict
1228 * it from all the underlying devices.
1229 */
1230static int dm_keyslot_evict(struct blk_crypto_profile *profile,
1231 const struct blk_crypto_key *key, unsigned int slot)
1232{
1233 struct mapped_device *md =
1234 container_of(profile, struct dm_crypto_profile, profile)->md;
1235 struct dm_table *t;
1236 int srcu_idx;
1237
1238 t = dm_get_live_table(md, srcu_idx: &srcu_idx);
1239 if (!t)
1240 return 0;
1241
1242 for (unsigned int i = 0; i < t->num_targets; i++) {
1243 struct dm_target *ti = dm_table_get_target(t, index: i);
1244
1245 if (!ti->type->iterate_devices)
1246 continue;
1247 ti->type->iterate_devices(ti, dm_keyslot_evict_callback,
1248 (void *)key);
1249 }
1250
1251 dm_put_live_table(md, srcu_idx);
1252 return 0;
1253}
1254
1255static int
1256device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev,
1257 sector_t start, sector_t len, void *data)
1258{
1259 struct blk_crypto_profile *parent = data;
1260 struct blk_crypto_profile *child =
1261 bdev_get_queue(bdev: dev->bdev)->crypto_profile;
1262
1263 blk_crypto_intersect_capabilities(parent, child);
1264 return 0;
1265}
1266
1267void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1268{
1269 struct dm_crypto_profile *dmcp = container_of(profile,
1270 struct dm_crypto_profile,
1271 profile);
1272
1273 if (!profile)
1274 return;
1275
1276 blk_crypto_profile_destroy(profile);
1277 kfree(objp: dmcp);
1278}
1279
1280static void dm_table_destroy_crypto_profile(struct dm_table *t)
1281{
1282 dm_destroy_crypto_profile(profile: t->crypto_profile);
1283 t->crypto_profile = NULL;
1284}
1285
1286/*
1287 * Constructs and initializes t->crypto_profile with a crypto profile that
1288 * represents the common set of crypto capabilities of the devices described by
1289 * the dm_table. However, if the constructed crypto profile doesn't support all
1290 * crypto capabilities that are supported by the current mapped_device, it
1291 * returns an error instead, since we don't support removing crypto capabilities
1292 * on table changes. Finally, if the constructed crypto profile is "empty" (has
1293 * no crypto capabilities at all), it just sets t->crypto_profile to NULL.
1294 */
1295static int dm_table_construct_crypto_profile(struct dm_table *t)
1296{
1297 struct dm_crypto_profile *dmcp;
1298 struct blk_crypto_profile *profile;
1299 unsigned int i;
1300 bool empty_profile = true;
1301
1302 dmcp = kmalloc(size: sizeof(*dmcp), GFP_KERNEL);
1303 if (!dmcp)
1304 return -ENOMEM;
1305 dmcp->md = t->md;
1306
1307 profile = &dmcp->profile;
1308 blk_crypto_profile_init(profile, num_slots: 0);
1309 profile->ll_ops.keyslot_evict = dm_keyslot_evict;
1310 profile->max_dun_bytes_supported = UINT_MAX;
1311 memset(profile->modes_supported, 0xFF,
1312 sizeof(profile->modes_supported));
1313
1314 for (i = 0; i < t->num_targets; i++) {
1315 struct dm_target *ti = dm_table_get_target(t, index: i);
1316
1317 if (!dm_target_passes_crypto(ti->type)) {
1318 blk_crypto_intersect_capabilities(parent: profile, NULL);
1319 break;
1320 }
1321 if (!ti->type->iterate_devices)
1322 continue;
1323 ti->type->iterate_devices(ti,
1324 device_intersect_crypto_capabilities,
1325 profile);
1326 }
1327
1328 if (t->md->queue &&
1329 !blk_crypto_has_capabilities(target: profile,
1330 reference: t->md->queue->crypto_profile)) {
1331 DMERR("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1332 dm_destroy_crypto_profile(profile);
1333 return -EINVAL;
1334 }
1335
1336 /*
1337 * If the new profile doesn't actually support any crypto capabilities,
1338 * we may as well represent it with a NULL profile.
1339 */
1340 for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) {
1341 if (profile->modes_supported[i]) {
1342 empty_profile = false;
1343 break;
1344 }
1345 }
1346
1347 if (empty_profile) {
1348 dm_destroy_crypto_profile(profile);
1349 profile = NULL;
1350 }
1351
1352 /*
1353 * t->crypto_profile is only set temporarily while the table is being
1354 * set up, and it gets set to NULL after the profile has been
1355 * transferred to the request_queue.
1356 */
1357 t->crypto_profile = profile;
1358
1359 return 0;
1360}
1361
1362static void dm_update_crypto_profile(struct request_queue *q,
1363 struct dm_table *t)
1364{
1365 if (!t->crypto_profile)
1366 return;
1367
1368 /* Make the crypto profile less restrictive. */
1369 if (!q->crypto_profile) {
1370 blk_crypto_register(profile: t->crypto_profile, q);
1371 } else {
1372 blk_crypto_update_capabilities(dst: q->crypto_profile,
1373 src: t->crypto_profile);
1374 dm_destroy_crypto_profile(profile: t->crypto_profile);
1375 }
1376 t->crypto_profile = NULL;
1377}
1378
1379#else /* CONFIG_BLK_INLINE_ENCRYPTION */
1380
1381static int dm_table_construct_crypto_profile(struct dm_table *t)
1382{
1383 return 0;
1384}
1385
1386void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1387{
1388}
1389
1390static void dm_table_destroy_crypto_profile(struct dm_table *t)
1391{
1392}
1393
1394static void dm_update_crypto_profile(struct request_queue *q,
1395 struct dm_table *t)
1396{
1397}
1398
1399#endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1400
1401/*
1402 * Prepares the table for use by building the indices,
1403 * setting the type, and allocating mempools.
1404 */
1405int dm_table_complete(struct dm_table *t)
1406{
1407 int r;
1408
1409 r = dm_table_determine_type(t);
1410 if (r) {
1411 DMERR("unable to determine table type");
1412 return r;
1413 }
1414
1415 r = dm_table_build_index(t);
1416 if (r) {
1417 DMERR("unable to build btrees");
1418 return r;
1419 }
1420
1421 r = dm_table_register_integrity(t);
1422 if (r) {
1423 DMERR("could not register integrity profile.");
1424 return r;
1425 }
1426
1427 r = dm_table_construct_crypto_profile(t);
1428 if (r) {
1429 DMERR("could not construct crypto profile.");
1430 return r;
1431 }
1432
1433 r = dm_table_alloc_md_mempools(t, md: t->md);
1434 if (r)
1435 DMERR("unable to allocate mempools");
1436
1437 return r;
1438}
1439
1440static DEFINE_MUTEX(_event_lock);
1441void dm_table_event_callback(struct dm_table *t,
1442 void (*fn)(void *), void *context)
1443{
1444 mutex_lock(&_event_lock);
1445 t->event_fn = fn;
1446 t->event_context = context;
1447 mutex_unlock(lock: &_event_lock);
1448}
1449
1450void dm_table_event(struct dm_table *t)
1451{
1452 mutex_lock(&_event_lock);
1453 if (t->event_fn)
1454 t->event_fn(t->event_context);
1455 mutex_unlock(lock: &_event_lock);
1456}
1457EXPORT_SYMBOL(dm_table_event);
1458
1459inline sector_t dm_table_get_size(struct dm_table *t)
1460{
1461 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1462}
1463EXPORT_SYMBOL(dm_table_get_size);
1464
1465/*
1466 * Search the btree for the correct target.
1467 *
1468 * Caller should check returned pointer for NULL
1469 * to trap I/O beyond end of device.
1470 */
1471struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1472{
1473 unsigned int l, n = 0, k = 0;
1474 sector_t *node;
1475
1476 if (unlikely(sector >= dm_table_get_size(t)))
1477 return NULL;
1478
1479 for (l = 0; l < t->depth; l++) {
1480 n = get_child(n, k);
1481 node = get_node(t, l, n);
1482
1483 for (k = 0; k < KEYS_PER_NODE; k++)
1484 if (node[k] >= sector)
1485 break;
1486 }
1487
1488 return &t->targets[(KEYS_PER_NODE * n) + k];
1489}
1490
1491static int device_not_poll_capable(struct dm_target *ti, struct dm_dev *dev,
1492 sector_t start, sector_t len, void *data)
1493{
1494 struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1495
1496 return !test_bit(QUEUE_FLAG_POLL, &q->queue_flags);
1497}
1498
1499/*
1500 * type->iterate_devices() should be called when the sanity check needs to
1501 * iterate and check all underlying data devices. iterate_devices() will
1502 * iterate all underlying data devices until it encounters a non-zero return
1503 * code, returned by whether the input iterate_devices_callout_fn, or
1504 * iterate_devices() itself internally.
1505 *
1506 * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1507 * iterate multiple underlying devices internally, in which case a non-zero
1508 * return code returned by iterate_devices_callout_fn will stop the iteration
1509 * in advance.
1510 *
1511 * Cases requiring _any_ underlying device supporting some kind of attribute,
1512 * should use the iteration structure like dm_table_any_dev_attr(), or call
1513 * it directly. @func should handle semantics of positive examples, e.g.
1514 * capable of something.
1515 *
1516 * Cases requiring _all_ underlying devices supporting some kind of attribute,
1517 * should use the iteration structure like dm_table_supports_nowait() or
1518 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1519 * uses an @anti_func that handle semantics of counter examples, e.g. not
1520 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1521 */
1522static bool dm_table_any_dev_attr(struct dm_table *t,
1523 iterate_devices_callout_fn func, void *data)
1524{
1525 for (unsigned int i = 0; i < t->num_targets; i++) {
1526 struct dm_target *ti = dm_table_get_target(t, index: i);
1527
1528 if (ti->type->iterate_devices &&
1529 ti->type->iterate_devices(ti, func, data))
1530 return true;
1531 }
1532
1533 return false;
1534}
1535
1536static int count_device(struct dm_target *ti, struct dm_dev *dev,
1537 sector_t start, sector_t len, void *data)
1538{
1539 unsigned int *num_devices = data;
1540
1541 (*num_devices)++;
1542
1543 return 0;
1544}
1545
1546static bool dm_table_supports_poll(struct dm_table *t)
1547{
1548 for (unsigned int i = 0; i < t->num_targets; i++) {
1549 struct dm_target *ti = dm_table_get_target(t, index: i);
1550
1551 if (!ti->type->iterate_devices ||
1552 ti->type->iterate_devices(ti, device_not_poll_capable, NULL))
1553 return false;
1554 }
1555
1556 return true;
1557}
1558
1559/*
1560 * Check whether a table has no data devices attached using each
1561 * target's iterate_devices method.
1562 * Returns false if the result is unknown because a target doesn't
1563 * support iterate_devices.
1564 */
1565bool dm_table_has_no_data_devices(struct dm_table *t)
1566{
1567 for (unsigned int i = 0; i < t->num_targets; i++) {
1568 struct dm_target *ti = dm_table_get_target(t, index: i);
1569 unsigned int num_devices = 0;
1570
1571 if (!ti->type->iterate_devices)
1572 return false;
1573
1574 ti->type->iterate_devices(ti, count_device, &num_devices);
1575 if (num_devices)
1576 return false;
1577 }
1578
1579 return true;
1580}
1581
1582static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1583 sector_t start, sector_t len, void *data)
1584{
1585 struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1586 enum blk_zoned_model *zoned_model = data;
1587
1588 return blk_queue_zoned_model(q) != *zoned_model;
1589}
1590
1591static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1592 sector_t start, sector_t len, void *data)
1593{
1594 struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1595
1596 return blk_queue_zoned_model(q) != BLK_ZONED_NONE;
1597}
1598
1599/*
1600 * Check the device zoned model based on the target feature flag. If the target
1601 * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1602 * also accepted but all devices must have the same zoned model. If the target
1603 * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1604 * zoned model with all zoned devices having the same zone size.
1605 */
1606static bool dm_table_supports_zoned_model(struct dm_table *t,
1607 enum blk_zoned_model zoned_model)
1608{
1609 for (unsigned int i = 0; i < t->num_targets; i++) {
1610 struct dm_target *ti = dm_table_get_target(t, index: i);
1611
1612 /*
1613 * For the wildcard target (dm-error), if we do not have a
1614 * backing device, we must always return false. If we have a
1615 * backing device, the result must depend on checking zoned
1616 * model, like for any other target. So for this, check directly
1617 * if the target backing device is zoned as we get "false" when
1618 * dm-error was set without a backing device.
1619 */
1620 if (dm_target_is_wildcard(ti->type) &&
1621 !ti->type->iterate_devices(ti, device_is_zoned_model, NULL))
1622 return false;
1623
1624 if (dm_target_supports_zoned_hm(ti->type)) {
1625 if (!ti->type->iterate_devices ||
1626 ti->type->iterate_devices(ti, device_not_zoned_model,
1627 &zoned_model))
1628 return false;
1629 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1630 if (zoned_model == BLK_ZONED_HM)
1631 return false;
1632 }
1633 }
1634
1635 return true;
1636}
1637
1638static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1639 sector_t start, sector_t len, void *data)
1640{
1641 unsigned int *zone_sectors = data;
1642
1643 if (!bdev_is_zoned(bdev: dev->bdev))
1644 return 0;
1645 return bdev_zone_sectors(bdev: dev->bdev) != *zone_sectors;
1646}
1647
1648/*
1649 * Check consistency of zoned model and zone sectors across all targets. For
1650 * zone sectors, if the destination device is a zoned block device, it shall
1651 * have the specified zone_sectors.
1652 */
1653static int validate_hardware_zoned_model(struct dm_table *t,
1654 enum blk_zoned_model zoned_model,
1655 unsigned int zone_sectors)
1656{
1657 if (zoned_model == BLK_ZONED_NONE)
1658 return 0;
1659
1660 if (!dm_table_supports_zoned_model(t, zoned_model)) {
1661 DMERR("%s: zoned model is not consistent across all devices",
1662 dm_device_name(t->md));
1663 return -EINVAL;
1664 }
1665
1666 /* Check zone size validity and compatibility */
1667 if (!zone_sectors || !is_power_of_2(n: zone_sectors))
1668 return -EINVAL;
1669
1670 if (dm_table_any_dev_attr(t, func: device_not_matches_zone_sectors, data: &zone_sectors)) {
1671 DMERR("%s: zone sectors is not consistent across all zoned devices",
1672 dm_device_name(t->md));
1673 return -EINVAL;
1674 }
1675
1676 return 0;
1677}
1678
1679/*
1680 * Establish the new table's queue_limits and validate them.
1681 */
1682int dm_calculate_queue_limits(struct dm_table *t,
1683 struct queue_limits *limits)
1684{
1685 struct queue_limits ti_limits;
1686 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1687 unsigned int zone_sectors = 0;
1688
1689 blk_set_stacking_limits(lim: limits);
1690
1691 for (unsigned int i = 0; i < t->num_targets; i++) {
1692 struct dm_target *ti = dm_table_get_target(t, index: i);
1693
1694 blk_set_stacking_limits(lim: &ti_limits);
1695
1696 if (!ti->type->iterate_devices) {
1697 /* Set I/O hints portion of queue limits */
1698 if (ti->type->io_hints)
1699 ti->type->io_hints(ti, &ti_limits);
1700 goto combine_limits;
1701 }
1702
1703 /*
1704 * Combine queue limits of all the devices this target uses.
1705 */
1706 ti->type->iterate_devices(ti, dm_set_device_limits,
1707 &ti_limits);
1708
1709 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1710 /*
1711 * After stacking all limits, validate all devices
1712 * in table support this zoned model and zone sectors.
1713 */
1714 zoned_model = ti_limits.zoned;
1715 zone_sectors = ti_limits.chunk_sectors;
1716 }
1717
1718 /* Set I/O hints portion of queue limits */
1719 if (ti->type->io_hints)
1720 ti->type->io_hints(ti, &ti_limits);
1721
1722 /*
1723 * Check each device area is consistent with the target's
1724 * overall queue limits.
1725 */
1726 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1727 &ti_limits))
1728 return -EINVAL;
1729
1730combine_limits:
1731 /*
1732 * Merge this target's queue limits into the overall limits
1733 * for the table.
1734 */
1735 if (blk_stack_limits(t: limits, b: &ti_limits, offset: 0) < 0)
1736 DMWARN("%s: adding target device (start sect %llu len %llu) "
1737 "caused an alignment inconsistency",
1738 dm_device_name(t->md),
1739 (unsigned long long) ti->begin,
1740 (unsigned long long) ti->len);
1741 }
1742
1743 /*
1744 * Verify that the zoned model and zone sectors, as determined before
1745 * any .io_hints override, are the same across all devices in the table.
1746 * - this is especially relevant if .io_hints is emulating a disk-managed
1747 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1748 * BUT...
1749 */
1750 if (limits->zoned != BLK_ZONED_NONE) {
1751 /*
1752 * ...IF the above limits stacking determined a zoned model
1753 * validate that all of the table's devices conform to it.
1754 */
1755 zoned_model = limits->zoned;
1756 zone_sectors = limits->chunk_sectors;
1757 }
1758 if (validate_hardware_zoned_model(t, zoned_model, zone_sectors))
1759 return -EINVAL;
1760
1761 return validate_hardware_logical_block_alignment(t, limits);
1762}
1763
1764/*
1765 * Verify that all devices have an integrity profile that matches the
1766 * DM device's registered integrity profile. If the profiles don't
1767 * match then unregister the DM device's integrity profile.
1768 */
1769static void dm_table_verify_integrity(struct dm_table *t)
1770{
1771 struct gendisk *template_disk = NULL;
1772
1773 if (t->integrity_added)
1774 return;
1775
1776 if (t->integrity_supported) {
1777 /*
1778 * Verify that the original integrity profile
1779 * matches all the devices in this table.
1780 */
1781 template_disk = dm_table_get_integrity_disk(t);
1782 if (template_disk &&
1783 blk_integrity_compare(dm_disk(md: t->md), template_disk) >= 0)
1784 return;
1785 }
1786
1787 if (integrity_profile_exists(disk: dm_disk(md: t->md))) {
1788 DMWARN("%s: unable to establish an integrity profile",
1789 dm_device_name(t->md));
1790 blk_integrity_unregister(dm_disk(md: t->md));
1791 }
1792}
1793
1794static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1795 sector_t start, sector_t len, void *data)
1796{
1797 unsigned long flush = (unsigned long) data;
1798 struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1799
1800 return (q->queue_flags & flush);
1801}
1802
1803static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1804{
1805 /*
1806 * Require at least one underlying device to support flushes.
1807 * t->devices includes internal dm devices such as mirror logs
1808 * so we need to use iterate_devices here, which targets
1809 * supporting flushes must provide.
1810 */
1811 for (unsigned int i = 0; i < t->num_targets; i++) {
1812 struct dm_target *ti = dm_table_get_target(t, index: i);
1813
1814 if (!ti->num_flush_bios)
1815 continue;
1816
1817 if (ti->flush_supported)
1818 return true;
1819
1820 if (ti->type->iterate_devices &&
1821 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1822 return true;
1823 }
1824
1825 return false;
1826}
1827
1828static int device_dax_write_cache_enabled(struct dm_target *ti,
1829 struct dm_dev *dev, sector_t start,
1830 sector_t len, void *data)
1831{
1832 struct dax_device *dax_dev = dev->dax_dev;
1833
1834 if (!dax_dev)
1835 return false;
1836
1837 if (dax_write_cache_enabled(dax_dev))
1838 return true;
1839 return false;
1840}
1841
1842static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1843 sector_t start, sector_t len, void *data)
1844{
1845 return !bdev_nonrot(bdev: dev->bdev);
1846}
1847
1848static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1849 sector_t start, sector_t len, void *data)
1850{
1851 struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1852
1853 return !blk_queue_add_random(q);
1854}
1855
1856static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1857 sector_t start, sector_t len, void *data)
1858{
1859 struct request_queue *q = bdev_get_queue(bdev: dev->bdev);
1860
1861 return !q->limits.max_write_zeroes_sectors;
1862}
1863
1864static bool dm_table_supports_write_zeroes(struct dm_table *t)
1865{
1866 for (unsigned int i = 0; i < t->num_targets; i++) {
1867 struct dm_target *ti = dm_table_get_target(t, index: i);
1868
1869 if (!ti->num_write_zeroes_bios)
1870 return false;
1871
1872 if (!ti->type->iterate_devices ||
1873 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1874 return false;
1875 }
1876
1877 return true;
1878}
1879
1880static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1881 sector_t start, sector_t len, void *data)
1882{
1883 return !bdev_nowait(bdev: dev->bdev);
1884}
1885
1886static bool dm_table_supports_nowait(struct dm_table *t)
1887{
1888 for (unsigned int i = 0; i < t->num_targets; i++) {
1889 struct dm_target *ti = dm_table_get_target(t, index: i);
1890
1891 if (!dm_target_supports_nowait(ti->type))
1892 return false;
1893
1894 if (!ti->type->iterate_devices ||
1895 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1896 return false;
1897 }
1898
1899 return true;
1900}
1901
1902static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1903 sector_t start, sector_t len, void *data)
1904{
1905 return !bdev_max_discard_sectors(bdev: dev->bdev);
1906}
1907
1908static bool dm_table_supports_discards(struct dm_table *t)
1909{
1910 for (unsigned int i = 0; i < t->num_targets; i++) {
1911 struct dm_target *ti = dm_table_get_target(t, index: i);
1912
1913 if (!ti->num_discard_bios)
1914 return false;
1915
1916 /*
1917 * Either the target provides discard support (as implied by setting
1918 * 'discards_supported') or it relies on _all_ data devices having
1919 * discard support.
1920 */
1921 if (!ti->discards_supported &&
1922 (!ti->type->iterate_devices ||
1923 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1924 return false;
1925 }
1926
1927 return true;
1928}
1929
1930static int device_not_secure_erase_capable(struct dm_target *ti,
1931 struct dm_dev *dev, sector_t start,
1932 sector_t len, void *data)
1933{
1934 return !bdev_max_secure_erase_sectors(bdev: dev->bdev);
1935}
1936
1937static bool dm_table_supports_secure_erase(struct dm_table *t)
1938{
1939 for (unsigned int i = 0; i < t->num_targets; i++) {
1940 struct dm_target *ti = dm_table_get_target(t, index: i);
1941
1942 if (!ti->num_secure_erase_bios)
1943 return false;
1944
1945 if (!ti->type->iterate_devices ||
1946 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1947 return false;
1948 }
1949
1950 return true;
1951}
1952
1953static int device_requires_stable_pages(struct dm_target *ti,
1954 struct dm_dev *dev, sector_t start,
1955 sector_t len, void *data)
1956{
1957 return bdev_stable_writes(bdev: dev->bdev);
1958}
1959
1960int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1961 struct queue_limits *limits)
1962{
1963 bool wc = false, fua = false;
1964 int r;
1965
1966 /*
1967 * Copy table's limits to the DM device's request_queue
1968 */
1969 q->limits = *limits;
1970
1971 if (dm_table_supports_nowait(t))
1972 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1973 else
1974 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1975
1976 if (!dm_table_supports_discards(t)) {
1977 q->limits.max_discard_sectors = 0;
1978 q->limits.max_hw_discard_sectors = 0;
1979 q->limits.discard_granularity = 0;
1980 q->limits.discard_alignment = 0;
1981 q->limits.discard_misaligned = 0;
1982 }
1983
1984 if (!dm_table_supports_secure_erase(t))
1985 q->limits.max_secure_erase_sectors = 0;
1986
1987 if (dm_table_supports_flush(t, flush: (1UL << QUEUE_FLAG_WC))) {
1988 wc = true;
1989 if (dm_table_supports_flush(t, flush: (1UL << QUEUE_FLAG_FUA)))
1990 fua = true;
1991 }
1992 blk_queue_write_cache(q, enabled: wc, fua);
1993
1994 if (dm_table_supports_dax(t, iterate_fn: device_not_dax_capable)) {
1995 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1996 if (dm_table_supports_dax(t, iterate_fn: device_not_dax_synchronous_capable))
1997 set_dax_synchronous(t->md->dax_dev);
1998 } else
1999 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
2000
2001 if (dm_table_any_dev_attr(t, func: device_dax_write_cache_enabled, NULL))
2002 dax_write_cache(dax_dev: t->md->dax_dev, wc: true);
2003
2004 /* Ensure that all underlying devices are non-rotational. */
2005 if (dm_table_any_dev_attr(t, func: device_is_rotational, NULL))
2006 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2007 else
2008 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2009
2010 if (!dm_table_supports_write_zeroes(t))
2011 q->limits.max_write_zeroes_sectors = 0;
2012
2013 dm_table_verify_integrity(t);
2014
2015 /*
2016 * Some devices don't use blk_integrity but still want stable pages
2017 * because they do their own checksumming.
2018 * If any underlying device requires stable pages, a table must require
2019 * them as well. Only targets that support iterate_devices are considered:
2020 * don't want error, zero, etc to require stable pages.
2021 */
2022 if (dm_table_any_dev_attr(t, func: device_requires_stable_pages, NULL))
2023 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2024 else
2025 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2026
2027 /*
2028 * Determine whether or not this queue's I/O timings contribute
2029 * to the entropy pool, Only request-based targets use this.
2030 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2031 * have it set.
2032 */
2033 if (blk_queue_add_random(q) &&
2034 dm_table_any_dev_attr(t, func: device_is_not_random, NULL))
2035 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2036
2037 /*
2038 * For a zoned target, setup the zones related queue attributes
2039 * and resources necessary for zone append emulation if necessary.
2040 */
2041 if (blk_queue_is_zoned(q)) {
2042 r = dm_set_zones_restrictions(t, q);
2043 if (r)
2044 return r;
2045 if (!static_key_enabled(&zoned_enabled.key))
2046 static_branch_enable(&zoned_enabled);
2047 }
2048
2049 dm_update_crypto_profile(q, t);
2050 disk_update_readahead(disk: t->md->disk);
2051
2052 /*
2053 * Check for request-based device is left to
2054 * dm_mq_init_request_queue()->blk_mq_init_allocated_queue().
2055 *
2056 * For bio-based device, only set QUEUE_FLAG_POLL when all
2057 * underlying devices supporting polling.
2058 */
2059 if (__table_type_bio_based(table_type: t->type)) {
2060 if (dm_table_supports_poll(t))
2061 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
2062 else
2063 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
2064 }
2065
2066 return 0;
2067}
2068
2069struct list_head *dm_table_get_devices(struct dm_table *t)
2070{
2071 return &t->devices;
2072}
2073
2074blk_mode_t dm_table_get_mode(struct dm_table *t)
2075{
2076 return t->mode;
2077}
2078EXPORT_SYMBOL(dm_table_get_mode);
2079
2080enum suspend_mode {
2081 PRESUSPEND,
2082 PRESUSPEND_UNDO,
2083 POSTSUSPEND,
2084};
2085
2086static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2087{
2088 lockdep_assert_held(&t->md->suspend_lock);
2089
2090 for (unsigned int i = 0; i < t->num_targets; i++) {
2091 struct dm_target *ti = dm_table_get_target(t, index: i);
2092
2093 switch (mode) {
2094 case PRESUSPEND:
2095 if (ti->type->presuspend)
2096 ti->type->presuspend(ti);
2097 break;
2098 case PRESUSPEND_UNDO:
2099 if (ti->type->presuspend_undo)
2100 ti->type->presuspend_undo(ti);
2101 break;
2102 case POSTSUSPEND:
2103 if (ti->type->postsuspend)
2104 ti->type->postsuspend(ti);
2105 break;
2106 }
2107 }
2108}
2109
2110void dm_table_presuspend_targets(struct dm_table *t)
2111{
2112 if (!t)
2113 return;
2114
2115 suspend_targets(t, mode: PRESUSPEND);
2116}
2117
2118void dm_table_presuspend_undo_targets(struct dm_table *t)
2119{
2120 if (!t)
2121 return;
2122
2123 suspend_targets(t, mode: PRESUSPEND_UNDO);
2124}
2125
2126void dm_table_postsuspend_targets(struct dm_table *t)
2127{
2128 if (!t)
2129 return;
2130
2131 suspend_targets(t, mode: POSTSUSPEND);
2132}
2133
2134int dm_table_resume_targets(struct dm_table *t)
2135{
2136 unsigned int i;
2137 int r = 0;
2138
2139 lockdep_assert_held(&t->md->suspend_lock);
2140
2141 for (i = 0; i < t->num_targets; i++) {
2142 struct dm_target *ti = dm_table_get_target(t, index: i);
2143
2144 if (!ti->type->preresume)
2145 continue;
2146
2147 r = ti->type->preresume(ti);
2148 if (r) {
2149 DMERR("%s: %s: preresume failed, error = %d",
2150 dm_device_name(t->md), ti->type->name, r);
2151 return r;
2152 }
2153 }
2154
2155 for (i = 0; i < t->num_targets; i++) {
2156 struct dm_target *ti = dm_table_get_target(t, index: i);
2157
2158 if (ti->type->resume)
2159 ti->type->resume(ti);
2160 }
2161
2162 return 0;
2163}
2164
2165struct mapped_device *dm_table_get_md(struct dm_table *t)
2166{
2167 return t->md;
2168}
2169EXPORT_SYMBOL(dm_table_get_md);
2170
2171const char *dm_table_device_name(struct dm_table *t)
2172{
2173 return dm_device_name(md: t->md);
2174}
2175EXPORT_SYMBOL_GPL(dm_table_device_name);
2176
2177void dm_table_run_md_queue_async(struct dm_table *t)
2178{
2179 if (!dm_table_request_based(t))
2180 return;
2181
2182 if (t->md->queue)
2183 blk_mq_run_hw_queues(q: t->md->queue, async: true);
2184}
2185EXPORT_SYMBOL(dm_table_run_md_queue_async);
2186
2187

source code of linux/drivers/md/dm-table.c