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