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 unsigned integrity_supported:1;
58
59 /*
60 * Indicates the rw permissions for the new logical
61 * device. This should be a combination of FMODE_READ
62 * and FMODE_WRITE.
63 */
64 fmode_t mode;
65
66 /* a list of devices used by this table */
67 struct list_head devices;
68
69 /* events get handed up using this callback */
70 void (*event_fn)(void *);
71 void *event_context;
72
73 struct dm_md_mempools *mempools;
74
75 struct list_head target_callbacks;
76};
77
78/*
79 * Similar to ceiling(log_size(n))
80 */
81static unsigned int int_log(unsigned int n, unsigned int base)
82{
83 int result = 0;
84
85 while (n > 1) {
86 n = dm_div_up(n, base);
87 result++;
88 }
89
90 return result;
91}
92
93/*
94 * Calculate the index of the child node of the n'th node k'th key.
95 */
96static inline unsigned int get_child(unsigned int n, unsigned int k)
97{
98 return (n * CHILDREN_PER_NODE) + k;
99}
100
101/*
102 * Return the n'th node of level l from table t.
103 */
104static inline sector_t *get_node(struct dm_table *t,
105 unsigned int l, unsigned int n)
106{
107 return t->index[l] + (n * KEYS_PER_NODE);
108}
109
110/*
111 * Return the highest key that you could lookup from the n'th
112 * node on level l of the btree.
113 */
114static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
115{
116 for (; l < t->depth - 1; l++)
117 n = get_child(n, CHILDREN_PER_NODE - 1);
118
119 if (n >= t->counts[l])
120 return (sector_t) - 1;
121
122 return get_node(t, l, n)[KEYS_PER_NODE - 1];
123}
124
125/*
126 * Fills in a level of the btree based on the highs of the level
127 * below it.
128 */
129static int setup_btree_index(unsigned int l, struct dm_table *t)
130{
131 unsigned int n, k;
132 sector_t *node;
133
134 for (n = 0U; n < t->counts[l]; n++) {
135 node = get_node(t, l, n);
136
137 for (k = 0U; k < KEYS_PER_NODE; k++)
138 node[k] = high(t, l + 1, get_child(n, k));
139 }
140
141 return 0;
142}
143
144void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
145{
146 unsigned long size;
147 void *addr;
148
149 /*
150 * Check that we're not going to overflow.
151 */
152 if (nmemb > (ULONG_MAX / elem_size))
153 return NULL;
154
155 size = nmemb * elem_size;
156 addr = vzalloc(size);
157
158 return addr;
159}
160EXPORT_SYMBOL(dm_vcalloc);
161
162/*
163 * highs, and targets are managed as dynamic arrays during a
164 * table load.
165 */
166static int alloc_targets(struct dm_table *t, unsigned int num)
167{
168 sector_t *n_highs;
169 struct dm_target *n_targets;
170 int n = t->num_targets;
171
172 /*
173 * Allocate both the target array and offset array at once.
174 * Append an empty entry to catch sectors beyond the end of
175 * the device.
176 */
177 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
178 sizeof(sector_t));
179 if (!n_highs)
180 return -ENOMEM;
181
182 n_targets = (struct dm_target *) (n_highs + num);
183
184 if (n) {
185 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
186 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
187 }
188
189 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
190 vfree(t->highs);
191
192 t->num_allocated = num;
193 t->highs = n_highs;
194 t->targets = n_targets;
195
196 return 0;
197}
198
199int dm_table_create(struct dm_table **result, fmode_t mode,
200 unsigned num_targets, struct mapped_device *md)
201{
202 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
203
204 if (!t)
205 return -ENOMEM;
206
207 INIT_LIST_HEAD(&t->devices);
208 INIT_LIST_HEAD(&t->target_callbacks);
209 atomic_set(&t->holders, 0);
210
211 if (!num_targets)
212 num_targets = KEYS_PER_NODE;
213
214 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
215
216 if (alloc_targets(t, num_targets)) {
217 kfree(t);
218 t = NULL;
219 return -ENOMEM;
220 }
221
222 t->mode = mode;
223 t->md = md;
224 *result = t;
225 return 0;
226}
227
228static void free_devices(struct list_head *devices)
229{
230 struct list_head *tmp, *next;
231
232 list_for_each_safe(tmp, next, devices) {
233 struct dm_dev_internal *dd =
234 list_entry(tmp, struct dm_dev_internal, list);
235 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
236 dd->dm_dev.name);
237 kfree(dd);
238 }
239}
240
241void dm_table_destroy(struct dm_table *t)
242{
243 unsigned int i;
244
245 if (!t)
246 return;
247
248 while (atomic_read(&t->holders))
249 msleep(1);
250 smp_mb();
251
252 /* free the indexes */
253 if (t->depth >= 2)
254 vfree(t->index[t->depth - 2]);
255
256 /* free the targets */
257 for (i = 0; i < t->num_targets; i++) {
258 struct dm_target *tgt = t->targets + i;
259
260 if (tgt->type->dtr)
261 tgt->type->dtr(tgt);
262
263 dm_put_target_type(tgt->type);
264 }
265
266 vfree(t->highs);
267
268 /* free the device list */
269 if (t->devices.next != &t->devices)
270 free_devices(&t->devices);
271
272 dm_free_md_mempools(t->mempools);
273
274 kfree(t);
275}
276
277void dm_table_get(struct dm_table *t)
278{
279 atomic_inc(&t->holders);
280}
281EXPORT_SYMBOL(dm_table_get);
282
283void dm_table_put(struct dm_table *t)
284{
285 if (!t)
286 return;
287
288 smp_mb__before_atomic_dec();
289 atomic_dec(&t->holders);
290}
291EXPORT_SYMBOL(dm_table_put);
292
293/*
294 * Checks to see if we need to extend highs or targets.
295 */
296static inline int check_space(struct dm_table *t)
297{
298 if (t->num_targets >= t->num_allocated)
299 return alloc_targets(t, t->num_allocated * 2);
300
301 return 0;
302}
303
304/*
305 * See if we've already got a device in the list.
306 */
307static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
308{
309 struct dm_dev_internal *dd;
310
311 list_for_each_entry (dd, l, list)
312 if (dd->dm_dev.bdev->bd_dev == dev)
313 return dd;
314
315 return NULL;
316}
317
318/*
319 * Open a device so we can use it as a map destination.
320 */
321static int open_dev(struct dm_dev_internal *d, dev_t dev,
322 struct mapped_device *md)
323{
324 static char *_claim_ptr = "I belong to device-mapper";
325 struct block_device *bdev;
326
327 int r;
328
329 BUG_ON(d->dm_dev.bdev);
330
331 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
332 if (IS_ERR(bdev))
333 return PTR_ERR(bdev);
334
335 r = bd_link_disk_holder(bdev, dm_disk(md));
336 if (r) {
337 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
338 return r;
339 }
340
341 d->dm_dev.bdev = bdev;
342 return 0;
343}
344
345/*
346 * Close a device that we've been using.
347 */
348static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
349{
350 if (!d->dm_dev.bdev)
351 return;
352
353 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
354 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
355 d->dm_dev.bdev = NULL;
356}
357
358/*
359 * If possible, this checks an area of a destination device is invalid.
360 */
361static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
362 sector_t start, sector_t len, void *data)
363{
364 struct request_queue *q;
365 struct queue_limits *limits = data;
366 struct block_device *bdev = dev->bdev;
367 sector_t dev_size =
368 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
369 unsigned short logical_block_size_sectors =
370 limits->logical_block_size >> SECTOR_SHIFT;
371 char b[BDEVNAME_SIZE];
372
373 /*
374 * Some devices exist without request functions,
375 * such as loop devices not yet bound to backing files.
376 * Forbid the use of such devices.
377 */
378 q = bdev_get_queue(bdev);
379 if (!q || !q->make_request_fn) {
380 DMWARN("%s: %s is not yet initialised: "
381 "start=%llu, len=%llu, dev_size=%llu",
382 dm_device_name(ti->table->md), bdevname(bdev, b),
383 (unsigned long long)start,
384 (unsigned long long)len,
385 (unsigned long long)dev_size);
386 return 1;
387 }
388
389 if (!dev_size)
390 return 0;
391
392 if ((start >= dev_size) || (start + len > dev_size)) {
393 DMWARN("%s: %s too small for target: "
394 "start=%llu, len=%llu, dev_size=%llu",
395 dm_device_name(ti->table->md), bdevname(bdev, b),
396 (unsigned long long)start,
397 (unsigned long long)len,
398 (unsigned long long)dev_size);
399 return 1;
400 }
401
402 if (logical_block_size_sectors <= 1)
403 return 0;
404
405 if (start & (logical_block_size_sectors - 1)) {
406 DMWARN("%s: start=%llu not aligned to h/w "
407 "logical block size %u of %s",
408 dm_device_name(ti->table->md),
409 (unsigned long long)start,
410 limits->logical_block_size, bdevname(bdev, b));
411 return 1;
412 }
413
414 if (len & (logical_block_size_sectors - 1)) {
415 DMWARN("%s: len=%llu not aligned to h/w "
416 "logical block size %u of %s",
417 dm_device_name(ti->table->md),
418 (unsigned long long)len,
419 limits->logical_block_size, bdevname(bdev, b));
420 return 1;
421 }
422
423 return 0;
424}
425
426/*
427 * This upgrades the mode on an already open dm_dev, being
428 * careful to leave things as they were if we fail to reopen the
429 * device and not to touch the existing bdev field in case
430 * it is accessed concurrently inside dm_table_any_congested().
431 */
432static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
433 struct mapped_device *md)
434{
435 int r;
436 struct dm_dev_internal dd_new, dd_old;
437
438 dd_new = dd_old = *dd;
439
440 dd_new.dm_dev.mode |= new_mode;
441 dd_new.dm_dev.bdev = NULL;
442
443 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
444 if (r)
445 return r;
446
447 dd->dm_dev.mode |= new_mode;
448 close_dev(&dd_old, md);
449
450 return 0;
451}
452
453/*
454 * Add a device to the list, or just increment the usage count if
455 * it's already present.
456 */
457int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
458 struct dm_dev **result)
459{
460 int r;
461 dev_t uninitialized_var(dev);
462 struct dm_dev_internal *dd;
463 unsigned int major, minor;
464 struct dm_table *t = ti->table;
465
466 BUG_ON(!t);
467
468 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
469 /* Extract the major/minor numbers */
470 dev = MKDEV(major, minor);
471 if (MAJOR(dev) != major || MINOR(dev) != minor)
472 return -EOVERFLOW;
473 } else {
474 /* convert the path to a device */
475 struct block_device *bdev = lookup_bdev(path);
476
477 if (IS_ERR(bdev))
478 return PTR_ERR(bdev);
479 dev = bdev->bd_dev;
480 bdput(bdev);
481 }
482
483 dd = find_device(&t->devices, dev);
484 if (!dd) {
485 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
486 if (!dd)
487 return -ENOMEM;
488
489 dd->dm_dev.mode = mode;
490 dd->dm_dev.bdev = NULL;
491
492 if ((r = open_dev(dd, dev, t->md))) {
493 kfree(dd);
494 return r;
495 }
496
497 format_dev_t(dd->dm_dev.name, dev);
498
499 atomic_set(&dd->count, 0);
500 list_add(&dd->list, &t->devices);
501
502 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
503 r = upgrade_mode(dd, mode, t->md);
504 if (r)
505 return r;
506 }
507 atomic_inc(&dd->count);
508
509 *result = &dd->dm_dev;
510 return 0;
511}
512EXPORT_SYMBOL(dm_get_device);
513
514int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
515 sector_t start, sector_t len, void *data)
516{
517 struct queue_limits *limits = data;
518 struct block_device *bdev = dev->bdev;
519 struct request_queue *q = bdev_get_queue(bdev);
520 char b[BDEVNAME_SIZE];
521
522 if (unlikely(!q)) {
523 DMWARN("%s: Cannot set limits for nonexistent device %s",
524 dm_device_name(ti->table->md), bdevname(bdev, b));
525 return 0;
526 }
527
528 if (bdev_stack_limits(limits, bdev, start) < 0)
529 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
530 "physical_block_size=%u, logical_block_size=%u, "
531 "alignment_offset=%u, start=%llu",
532 dm_device_name(ti->table->md), bdevname(bdev, b),
533 q->limits.physical_block_size,
534 q->limits.logical_block_size,
535 q->limits.alignment_offset,
536 (unsigned long long) start << SECTOR_SHIFT);
537
538 /*
539 * Check if merge fn is supported.
540 * If not we'll force DM to use PAGE_SIZE or
541 * smaller I/O, just to be safe.
542 */
543 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
544 blk_limits_max_hw_sectors(limits,
545 (unsigned int) (PAGE_SIZE >> 9));
546 return 0;
547}
548EXPORT_SYMBOL_GPL(dm_set_device_limits);
549
550/*
551 * Decrement a device's use count and remove it if necessary.
552 */
553void dm_put_device(struct dm_target *ti, struct dm_dev *d)
554{
555 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
556 dm_dev);
557
558 if (atomic_dec_and_test(&dd->count)) {
559 close_dev(dd, ti->table->md);
560 list_del(&dd->list);
561 kfree(dd);
562 }
563}
564EXPORT_SYMBOL(dm_put_device);
565
566/*
567 * Checks to see if the target joins onto the end of the table.
568 */
569static int adjoin(struct dm_table *table, struct dm_target *ti)
570{
571 struct dm_target *prev;
572
573 if (!table->num_targets)
574 return !ti->begin;
575
576 prev = &table->targets[table->num_targets - 1];
577 return (ti->begin == (prev->begin + prev->len));
578}
579
580/*
581 * Used to dynamically allocate the arg array.
582 */
583static char **realloc_argv(unsigned *array_size, char **old_argv)
584{
585 char **argv;
586 unsigned new_size;
587
588 new_size = *array_size ? *array_size * 2 : 64;
589 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
590 if (argv) {
591 memcpy(argv, old_argv, *array_size * sizeof(*argv));
592 *array_size = new_size;
593 }
594
595 kfree(old_argv);
596 return argv;
597}
598
599/*
600 * Destructively splits up the argument list to pass to ctr.
601 */
602int dm_split_args(int *argc, char ***argvp, char *input)
603{
604 char *start, *end = input, *out, **argv = NULL;
605 unsigned array_size = 0;
606
607 *argc = 0;
608
609 if (!input) {
610 *argvp = NULL;
611 return 0;
612 }
613
614 argv = realloc_argv(&array_size, argv);
615 if (!argv)
616 return -ENOMEM;
617
618 while (1) {
619 /* Skip whitespace */
620 start = skip_spaces(end);
621
622 if (!*start)
623 break; /* success, we hit the end */
624
625 /* 'out' is used to remove any back-quotes */
626 end = out = start;
627 while (*end) {
628 /* Everything apart from '\0' can be quoted */
629 if (*end == '\\' && *(end + 1)) {
630 *out++ = *(end + 1);
631 end += 2;
632 continue;
633 }
634
635 if (isspace(*end))
636 break; /* end of token */
637
638 *out++ = *end++;
639 }
640
641 /* have we already filled the array ? */
642 if ((*argc + 1) > array_size) {
643 argv = realloc_argv(&array_size, argv);
644 if (!argv)
645 return -ENOMEM;
646 }
647
648 /* we know this is whitespace */
649 if (*end)
650 end++;
651
652 /* terminate the string and put it in the array */
653 *out = '\0';
654 argv[*argc] = start;
655 (*argc)++;
656 }
657
658 *argvp = argv;
659 return 0;
660}
661
662/*
663 * Impose necessary and sufficient conditions on a devices's table such
664 * that any incoming bio which respects its logical_block_size can be
665 * processed successfully. If it falls across the boundary between
666 * two or more targets, the size of each piece it gets split into must
667 * be compatible with the logical_block_size of the target processing it.
668 */
669static int validate_hardware_logical_block_alignment(struct dm_table *table,
670 struct queue_limits *limits)
671{
672 /*
673 * This function uses arithmetic modulo the logical_block_size
674 * (in units of 512-byte sectors).
675 */
676 unsigned short device_logical_block_size_sects =
677 limits->logical_block_size >> SECTOR_SHIFT;
678
679 /*
680 * Offset of the start of the next table entry, mod logical_block_size.
681 */
682 unsigned short next_target_start = 0;
683
684 /*
685 * Given an aligned bio that extends beyond the end of a
686 * target, how many sectors must the next target handle?
687 */
688 unsigned short remaining = 0;
689
690 struct dm_target *uninitialized_var(ti);
691 struct queue_limits ti_limits;
692 unsigned i = 0;
693
694 /*
695 * Check each entry in the table in turn.
696 */
697 while (i < dm_table_get_num_targets(table)) {
698 ti = dm_table_get_target(table, i++);
699
700 blk_set_default_limits(&ti_limits);
701
702 /* combine all target devices' limits */
703 if (ti->type->iterate_devices)
704 ti->type->iterate_devices(ti, dm_set_device_limits,
705 &ti_limits);
706
707 /*
708 * If the remaining sectors fall entirely within this
709 * table entry are they compatible with its logical_block_size?
710 */
711 if (remaining < ti->len &&
712 remaining & ((ti_limits.logical_block_size >>
713 SECTOR_SHIFT) - 1))
714 break; /* Error */
715
716 next_target_start =
717 (unsigned short) ((next_target_start + ti->len) &
718 (device_logical_block_size_sects - 1));
719 remaining = next_target_start ?
720 device_logical_block_size_sects - next_target_start : 0;
721 }
722
723 if (remaining) {
724 DMWARN("%s: table line %u (start sect %llu len %llu) "
725 "not aligned to h/w logical block size %u",
726 dm_device_name(table->md), i,
727 (unsigned long long) ti->begin,
728 (unsigned long long) ti->len,
729 limits->logical_block_size);
730 return -EINVAL;
731 }
732
733 return 0;
734}
735
736int dm_table_add_target(struct dm_table *t, const char *type,
737 sector_t start, sector_t len, char *params)
738{
739 int r = -EINVAL, argc;
740 char **argv;
741 struct dm_target *tgt;
742
743 if ((r = check_space(t)))
744 return r;
745
746 tgt = t->targets + t->num_targets;
747 memset(tgt, 0, sizeof(*tgt));
748
749 if (!len) {
750 DMERR("%s: zero-length target", dm_device_name(t->md));
751 return -EINVAL;
752 }
753
754 tgt->type = dm_get_target_type(type);
755 if (!tgt->type) {
756 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
757 type);
758 return -EINVAL;
759 }
760
761 tgt->table = t;
762 tgt->begin = start;
763 tgt->len = len;
764 tgt->error = "Unknown error";
765
766 /*
767 * Does this target adjoin the previous one ?
768 */
769 if (!adjoin(t, tgt)) {
770 tgt->error = "Gap in table";
771 r = -EINVAL;
772 goto bad;
773 }
774
775 r = dm_split_args(&argc, &argv, params);
776 if (r) {
777 tgt->error = "couldn't split parameters (insufficient memory)";
778 goto bad;
779 }
780
781 r = tgt->type->ctr(tgt, argc, argv);
782 kfree(argv);
783 if (r)
784 goto bad;
785
786 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
787
788 if (!tgt->num_discard_requests && tgt->discards_supported)
789 DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
790 dm_device_name(t->md), type);
791
792 return 0;
793
794 bad:
795 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
796 dm_put_target_type(tgt->type);
797 return r;
798}
799
800/*
801 * Target argument parsing helpers.
802 */
803static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
804 unsigned *value, char **error, unsigned grouped)
805{
806 const char *arg_str = dm_shift_arg(arg_set);
807
808 if (!arg_str ||
809 (sscanf(arg_str, "%u", value) != 1) ||
810 (*value < arg->min) ||
811 (*value > arg->max) ||
812 (grouped && arg_set->argc < *value)) {
813 *error = arg->error;
814 return -EINVAL;
815 }
816
817 return 0;
818}
819
820int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
821 unsigned *value, char **error)
822{
823 return validate_next_arg(arg, arg_set, value, error, 0);
824}
825EXPORT_SYMBOL(dm_read_arg);
826
827int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
828 unsigned *value, char **error)
829{
830 return validate_next_arg(arg, arg_set, value, error, 1);
831}
832EXPORT_SYMBOL(dm_read_arg_group);
833
834const char *dm_shift_arg(struct dm_arg_set *as)
835{
836 char *r;
837
838 if (as->argc) {
839 as->argc--;
840 r = *as->argv;
841 as->argv++;
842 return r;
843 }
844
845 return NULL;
846}
847EXPORT_SYMBOL(dm_shift_arg);
848
849void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
850{
851 BUG_ON(as->argc < num_args);
852 as->argc -= num_args;
853 as->argv += num_args;
854}
855EXPORT_SYMBOL(dm_consume_args);
856
857static int dm_table_set_type(struct dm_table *t)
858{
859 unsigned i;
860 unsigned bio_based = 0, request_based = 0;
861 struct dm_target *tgt;
862 struct dm_dev_internal *dd;
863 struct list_head *devices;
864
865 for (i = 0; i < t->num_targets; i++) {
866 tgt = t->targets + i;
867 if (dm_target_request_based(tgt))
868 request_based = 1;
869 else
870 bio_based = 1;
871
872 if (bio_based && request_based) {
873 DMWARN("Inconsistent table: different target types"
874 " can't be mixed up");
875 return -EINVAL;
876 }
877 }
878
879 if (bio_based) {
880 /* We must use this table as bio-based */
881 t->type = DM_TYPE_BIO_BASED;
882 return 0;
883 }
884
885 BUG_ON(!request_based); /* No targets in this table */
886
887 /* Non-request-stackable devices can't be used for request-based dm */
888 devices = dm_table_get_devices(t);
889 list_for_each_entry(dd, devices, list) {
890 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
891 DMWARN("table load rejected: including"
892 " non-request-stackable devices");
893 return -EINVAL;
894 }
895 }
896
897 /*
898 * Request-based dm supports only tables that have a single target now.
899 * To support multiple targets, request splitting support is needed,
900 * and that needs lots of changes in the block-layer.
901 * (e.g. request completion process for partial completion.)
902 */
903 if (t->num_targets > 1) {
904 DMWARN("Request-based dm doesn't support multiple targets yet");
905 return -EINVAL;
906 }
907
908 t->type = DM_TYPE_REQUEST_BASED;
909
910 return 0;
911}
912
913unsigned dm_table_get_type(struct dm_table *t)
914{
915 return t->type;
916}
917
918bool dm_table_request_based(struct dm_table *t)
919{
920 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
921}
922
923int dm_table_alloc_md_mempools(struct dm_table *t)
924{
925 unsigned type = dm_table_get_type(t);
926
927 if (unlikely(type == DM_TYPE_NONE)) {
928 DMWARN("no table type is set, can't allocate mempools");
929 return -EINVAL;
930 }
931
932 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
933 if (!t->mempools)
934 return -ENOMEM;
935
936 return 0;
937}
938
939void dm_table_free_md_mempools(struct dm_table *t)
940{
941 dm_free_md_mempools(t->mempools);
942 t->mempools = NULL;
943}
944
945struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
946{
947 return t->mempools;
948}
949
950static int setup_indexes(struct dm_table *t)
951{
952 int i;
953 unsigned int total = 0;
954 sector_t *indexes;
955
956 /* allocate the space for *all* the indexes */
957 for (i = t->depth - 2; i >= 0; i--) {
958 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
959 total += t->counts[i];
960 }
961
962 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
963 if (!indexes)
964 return -ENOMEM;
965
966 /* set up internal nodes, bottom-up */
967 for (i = t->depth - 2; i >= 0; i--) {
968 t->index[i] = indexes;
969 indexes += (KEYS_PER_NODE * t->counts[i]);
970 setup_btree_index(i, t);
971 }
972
973 return 0;
974}
975
976/*
977 * Builds the btree to index the map.
978 */
979static int dm_table_build_index(struct dm_table *t)
980{
981 int r = 0;
982 unsigned int leaf_nodes;
983
984 /* how many indexes will the btree have ? */
985 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
986 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
987
988 /* leaf layer has already been set up */
989 t->counts[t->depth - 1] = leaf_nodes;
990 t->index[t->depth - 1] = t->highs;
991
992 if (t->depth >= 2)
993 r = setup_indexes(t);
994
995 return r;
996}
997
998/*
999 * Get a disk whose integrity profile reflects the table's profile.
1000 * If %match_all is true, all devices' profiles must match.
1001 * If %match_all is false, all devices must at least have an
1002 * allocated integrity profile; but uninitialized is ok.
1003 * Returns NULL if integrity support was inconsistent or unavailable.
1004 */
1005static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1006 bool match_all)
1007{
1008 struct list_head *devices = dm_table_get_devices(t);
1009 struct dm_dev_internal *dd = NULL;
1010 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1011
1012 list_for_each_entry(dd, devices, list) {
1013 template_disk = dd->dm_dev.bdev->bd_disk;
1014 if (!blk_get_integrity(template_disk))
1015 goto no_integrity;
1016 if (!match_all && !blk_integrity_is_initialized(template_disk))
1017 continue; /* skip uninitialized profiles */
1018 else if (prev_disk &&
1019 blk_integrity_compare(prev_disk, template_disk) < 0)
1020 goto no_integrity;
1021 prev_disk = template_disk;
1022 }
1023
1024 return template_disk;
1025
1026no_integrity:
1027 if (prev_disk)
1028 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1029 dm_device_name(t->md),
1030 prev_disk->disk_name,
1031 template_disk->disk_name);
1032 return NULL;
1033}
1034
1035/*
1036 * Register the mapped device for blk_integrity support if
1037 * the underlying devices have an integrity profile. But all devices
1038 * may not have matching profiles (checking all devices isn't reliable
1039 * during table load because this table may use other DM device(s) which
1040 * must be resumed before they will have an initialized integity profile).
1041 * Stacked DM devices force a 2 stage integrity profile validation:
1042 * 1 - during load, validate all initialized integrity profiles match
1043 * 2 - during resume, validate all integrity profiles match
1044 */
1045static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1046{
1047 struct gendisk *template_disk = NULL;
1048
1049 template_disk = dm_table_get_integrity_disk(t, false);
1050 if (!template_disk)
1051 return 0;
1052
1053 if (!blk_integrity_is_initialized(dm_disk(md))) {
1054 t->integrity_supported = 1;
1055 return blk_integrity_register(dm_disk(md), NULL);
1056 }
1057
1058 /*
1059 * If DM device already has an initalized integrity
1060 * profile the new profile should not conflict.
1061 */
1062 if (blk_integrity_is_initialized(template_disk) &&
1063 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1064 DMWARN("%s: conflict with existing integrity profile: "
1065 "%s profile mismatch",
1066 dm_device_name(t->md),
1067 template_disk->disk_name);
1068 return 1;
1069 }
1070
1071 /* Preserve existing initialized integrity profile */
1072 t->integrity_supported = 1;
1073 return 0;
1074}
1075
1076/*
1077 * Prepares the table for use by building the indices,
1078 * setting the type, and allocating mempools.
1079 */
1080int dm_table_complete(struct dm_table *t)
1081{
1082 int r;
1083
1084 r = dm_table_set_type(t);
1085 if (r) {
1086 DMERR("unable to set table type");
1087 return r;
1088 }
1089
1090 r = dm_table_build_index(t);
1091 if (r) {
1092 DMERR("unable to build btrees");
1093 return r;
1094 }
1095
1096 r = dm_table_prealloc_integrity(t, t->md);
1097 if (r) {
1098 DMERR("could not register integrity profile.");
1099 return r;
1100 }
1101
1102 r = dm_table_alloc_md_mempools(t);
1103 if (r)
1104 DMERR("unable to allocate mempools");
1105
1106 return r;
1107}
1108
1109static DEFINE_MUTEX(_event_lock);
1110void dm_table_event_callback(struct dm_table *t,
1111 void (*fn)(void *), void *context)
1112{
1113 mutex_lock(&_event_lock);
1114 t->event_fn = fn;
1115 t->event_context = context;
1116 mutex_unlock(&_event_lock);
1117}
1118
1119void dm_table_event(struct dm_table *t)
1120{
1121 /*
1122 * You can no longer call dm_table_event() from interrupt
1123 * context, use a bottom half instead.
1124 */
1125 BUG_ON(in_interrupt());
1126
1127 mutex_lock(&_event_lock);
1128 if (t->event_fn)
1129 t->event_fn(t->event_context);
1130 mutex_unlock(&_event_lock);
1131}
1132EXPORT_SYMBOL(dm_table_event);
1133
1134sector_t dm_table_get_size(struct dm_table *t)
1135{
1136 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1137}
1138EXPORT_SYMBOL(dm_table_get_size);
1139
1140struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1141{
1142 if (index >= t->num_targets)
1143 return NULL;
1144
1145 return t->targets + index;
1146}
1147
1148/*
1149 * Search the btree for the correct target.
1150 *
1151 * Caller should check returned pointer with dm_target_is_valid()
1152 * to trap I/O beyond end of device.
1153 */
1154struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1155{
1156 unsigned int l, n = 0, k = 0;
1157 sector_t *node;
1158
1159 for (l = 0; l < t->depth; l++) {
1160 n = get_child(n, k);
1161 node = get_node(t, l, n);
1162
1163 for (k = 0; k < KEYS_PER_NODE; k++)
1164 if (node[k] >= sector)
1165 break;
1166 }
1167
1168 return &t->targets[(KEYS_PER_NODE * n) + k];
1169}
1170
1171/*
1172 * Establish the new table's queue_limits and validate them.
1173 */
1174int dm_calculate_queue_limits(struct dm_table *table,
1175 struct queue_limits *limits)
1176{
1177 struct dm_target *uninitialized_var(ti);
1178 struct queue_limits ti_limits;
1179 unsigned i = 0;
1180
1181 blk_set_default_limits(limits);
1182
1183 while (i < dm_table_get_num_targets(table)) {
1184 blk_set_default_limits(&ti_limits);
1185
1186 ti = dm_table_get_target(table, i++);
1187
1188 if (!ti->type->iterate_devices)
1189 goto combine_limits;
1190
1191 /*
1192 * Combine queue limits of all the devices this target uses.
1193 */
1194 ti->type->iterate_devices(ti, dm_set_device_limits,
1195 &ti_limits);
1196
1197 /* Set I/O hints portion of queue limits */
1198 if (ti->type->io_hints)
1199 ti->type->io_hints(ti, &ti_limits);
1200
1201 /*
1202 * Check each device area is consistent with the target's
1203 * overall queue limits.
1204 */
1205 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1206 &ti_limits))
1207 return -EINVAL;
1208
1209combine_limits:
1210 /*
1211 * Merge this target's queue limits into the overall limits
1212 * for the table.
1213 */
1214 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1215 DMWARN("%s: adding target device "
1216 "(start sect %llu len %llu) "
1217 "caused an alignment inconsistency",
1218 dm_device_name(table->md),
1219 (unsigned long long) ti->begin,
1220 (unsigned long long) ti->len);
1221 }
1222
1223 return validate_hardware_logical_block_alignment(table, limits);
1224}
1225
1226/*
1227 * Set the integrity profile for this device if all devices used have
1228 * matching profiles. We're quite deep in the resume path but still
1229 * don't know if all devices (particularly DM devices this device
1230 * may be stacked on) have matching profiles. Even if the profiles
1231 * don't match we have no way to fail (to resume) at this point.
1232 */
1233static void dm_table_set_integrity(struct dm_table *t)
1234{
1235 struct gendisk *template_disk = NULL;
1236
1237 if (!blk_get_integrity(dm_disk(t->md)))
1238 return;
1239
1240 template_disk = dm_table_get_integrity_disk(t, true);
1241 if (template_disk)
1242 blk_integrity_register(dm_disk(t->md),
1243 blk_get_integrity(template_disk));
1244 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1245 DMWARN("%s: device no longer has a valid integrity profile",
1246 dm_device_name(t->md));
1247 else
1248 DMWARN("%s: unable to establish an integrity profile",
1249 dm_device_name(t->md));
1250}
1251
1252static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1253 sector_t start, sector_t len, void *data)
1254{
1255 unsigned flush = (*(unsigned *)data);
1256 struct request_queue *q = bdev_get_queue(dev->bdev);
1257
1258 return q && (q->flush_flags & flush);
1259}
1260
1261static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1262{
1263 struct dm_target *ti;
1264 unsigned i = 0;
1265
1266 /*
1267 * Require at least one underlying device to support flushes.
1268 * t->devices includes internal dm devices such as mirror logs
1269 * so we need to use iterate_devices here, which targets
1270 * supporting flushes must provide.
1271 */
1272 while (i < dm_table_get_num_targets(t)) {
1273 ti = dm_table_get_target(t, i++);
1274
1275 if (!ti->num_flush_requests)
1276 continue;
1277
1278 if (ti->type->iterate_devices &&
1279 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1280 return 1;
1281 }
1282
1283 return 0;
1284}
1285
1286static bool dm_table_discard_zeroes_data(struct dm_table *t)
1287{
1288 struct dm_target *ti;
1289 unsigned i = 0;
1290
1291 /* Ensure that all targets supports discard_zeroes_data. */
1292 while (i < dm_table_get_num_targets(t)) {
1293 ti = dm_table_get_target(t, i++);
1294
1295 if (ti->discard_zeroes_data_unsupported)
1296 return 0;
1297 }
1298
1299 return 1;
1300}
1301
1302void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1303 struct queue_limits *limits)
1304{
1305 unsigned flush = 0;
1306
1307 /*
1308 * Copy table's limits to the DM device's request_queue
1309 */
1310 q->limits = *limits;
1311
1312 if (!dm_table_supports_discards(t))
1313 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1314 else
1315 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1316
1317 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1318 flush |= REQ_FLUSH;
1319 if (dm_table_supports_flush(t, REQ_FUA))
1320 flush |= REQ_FUA;
1321 }
1322 blk_queue_flush(q, flush);
1323
1324 if (!dm_table_discard_zeroes_data(t))
1325 q->limits.discard_zeroes_data = 0;
1326
1327 dm_table_set_integrity(t);
1328
1329 /*
1330 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1331 * visible to other CPUs because, once the flag is set, incoming bios
1332 * are processed by request-based dm, which refers to the queue
1333 * settings.
1334 * Until the flag set, bios are passed to bio-based dm and queued to
1335 * md->deferred where queue settings are not needed yet.
1336 * Those bios are passed to request-based dm at the resume time.
1337 */
1338 smp_mb();
1339 if (dm_table_request_based(t))
1340 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1341}
1342
1343unsigned int dm_table_get_num_targets(struct dm_table *t)
1344{
1345 return t->num_targets;
1346}
1347
1348struct list_head *dm_table_get_devices(struct dm_table *t)
1349{
1350 return &t->devices;
1351}
1352
1353fmode_t dm_table_get_mode(struct dm_table *t)
1354{
1355 return t->mode;
1356}
1357EXPORT_SYMBOL(dm_table_get_mode);
1358
1359static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1360{
1361 int i = t->num_targets;
1362 struct dm_target *ti = t->targets;
1363
1364 while (i--) {
1365 if (postsuspend) {
1366 if (ti->type->postsuspend)
1367 ti->type->postsuspend(ti);
1368 } else if (ti->type->presuspend)
1369 ti->type->presuspend(ti);
1370
1371 ti++;
1372 }
1373}
1374
1375void dm_table_presuspend_targets(struct dm_table *t)
1376{
1377 if (!t)
1378 return;
1379
1380 suspend_targets(t, 0);
1381}
1382
1383void dm_table_postsuspend_targets(struct dm_table *t)
1384{
1385 if (!t)
1386 return;
1387
1388 suspend_targets(t, 1);
1389}
1390
1391int dm_table_resume_targets(struct dm_table *t)
1392{
1393 int i, r = 0;
1394
1395 for (i = 0; i < t->num_targets; i++) {
1396 struct dm_target *ti = t->targets + i;
1397
1398 if (!ti->type->preresume)
1399 continue;
1400
1401 r = ti->type->preresume(ti);
1402 if (r)
1403 return r;
1404 }
1405
1406 for (i = 0; i < t->num_targets; i++) {
1407 struct dm_target *ti = t->targets + i;
1408
1409 if (ti->type->resume)
1410 ti->type->resume(ti);
1411 }
1412
1413 return 0;
1414}
1415
1416void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1417{
1418 list_add(&cb->list, &t->target_callbacks);
1419}
1420EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1421
1422int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1423{
1424 struct dm_dev_internal *dd;
1425 struct list_head *devices = dm_table_get_devices(t);
1426 struct dm_target_callbacks *cb;
1427 int r = 0;
1428
1429 list_for_each_entry(dd, devices, list) {
1430 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1431 char b[BDEVNAME_SIZE];
1432
1433 if (likely(q))
1434 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1435 else
1436 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1437 dm_device_name(t->md),
1438 bdevname(dd->dm_dev.bdev, b));
1439 }
1440
1441 list_for_each_entry(cb, &t->target_callbacks, list)
1442 if (cb->congested_fn)
1443 r |= cb->congested_fn(cb, bdi_bits);
1444
1445 return r;
1446}
1447
1448int dm_table_any_busy_target(struct dm_table *t)
1449{
1450 unsigned i;
1451 struct dm_target *ti;
1452
1453 for (i = 0; i < t->num_targets; i++) {
1454 ti = t->targets + i;
1455 if (ti->type->busy && ti->type->busy(ti))
1456 return 1;
1457 }
1458
1459 return 0;
1460}
1461
1462struct mapped_device *dm_table_get_md(struct dm_table *t)
1463{
1464 return t->md;
1465}
1466EXPORT_SYMBOL(dm_table_get_md);
1467
1468static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1469 sector_t start, sector_t len, void *data)
1470{
1471 struct request_queue *q = bdev_get_queue(dev->bdev);
1472
1473 return q && blk_queue_discard(q);
1474}
1475
1476bool dm_table_supports_discards(struct dm_table *t)
1477{
1478 struct dm_target *ti;
1479 unsigned i = 0;
1480
1481 /*
1482 * Unless any target used by the table set discards_supported,
1483 * require at least one underlying device to support discards.
1484 * t->devices includes internal dm devices such as mirror logs
1485 * so we need to use iterate_devices here, which targets
1486 * supporting discard selectively must provide.
1487 */
1488 while (i < dm_table_get_num_targets(t)) {
1489 ti = dm_table_get_target(t, i++);
1490
1491 if (!ti->num_discard_requests)
1492 continue;
1493
1494 if (ti->discards_supported)
1495 return 1;
1496
1497 if (ti->type->iterate_devices &&
1498 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1499 return 1;
1500 }
1501
1502 return 0;
1503}
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-core.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#include <linux/blk-mq.h>
22#include <linux/mount.h>
23#include <linux/dax.h>
24
25#define DM_MSG_PREFIX "table"
26
27#define NODE_SIZE L1_CACHE_BYTES
28#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
29#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
30
31/*
32 * Similar to ceiling(log_size(n))
33 */
34static unsigned int int_log(unsigned int n, unsigned int base)
35{
36 int result = 0;
37
38 while (n > 1) {
39 n = dm_div_up(n, base);
40 result++;
41 }
42
43 return result;
44}
45
46/*
47 * Calculate the index of the child node of the n'th node k'th key.
48 */
49static inline unsigned int get_child(unsigned int n, unsigned int k)
50{
51 return (n * CHILDREN_PER_NODE) + k;
52}
53
54/*
55 * Return the n'th node of level l from table t.
56 */
57static inline sector_t *get_node(struct dm_table *t,
58 unsigned int l, unsigned int n)
59{
60 return t->index[l] + (n * KEYS_PER_NODE);
61}
62
63/*
64 * Return the highest key that you could lookup from the n'th
65 * node on level l of the btree.
66 */
67static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
68{
69 for (; l < t->depth - 1; l++)
70 n = get_child(n, CHILDREN_PER_NODE - 1);
71
72 if (n >= t->counts[l])
73 return (sector_t) - 1;
74
75 return get_node(t, l, n)[KEYS_PER_NODE - 1];
76}
77
78/*
79 * Fills in a level of the btree based on the highs of the level
80 * below it.
81 */
82static int setup_btree_index(unsigned int l, struct dm_table *t)
83{
84 unsigned int n, k;
85 sector_t *node;
86
87 for (n = 0U; n < t->counts[l]; n++) {
88 node = get_node(t, l, n);
89
90 for (k = 0U; k < KEYS_PER_NODE; k++)
91 node[k] = high(t, l + 1, get_child(n, k));
92 }
93
94 return 0;
95}
96
97/*
98 * highs, and targets are managed as dynamic arrays during a
99 * table load.
100 */
101static int alloc_targets(struct dm_table *t, unsigned int num)
102{
103 sector_t *n_highs;
104 struct dm_target *n_targets;
105
106 /*
107 * Allocate both the target array and offset array at once.
108 */
109 n_highs = kvcalloc(num, sizeof(struct dm_target) + sizeof(sector_t),
110 GFP_KERNEL);
111 if (!n_highs)
112 return -ENOMEM;
113
114 n_targets = (struct dm_target *) (n_highs + num);
115
116 memset(n_highs, -1, sizeof(*n_highs) * num);
117 kvfree(t->highs);
118
119 t->num_allocated = num;
120 t->highs = n_highs;
121 t->targets = n_targets;
122
123 return 0;
124}
125
126int dm_table_create(struct dm_table **result, fmode_t mode,
127 unsigned num_targets, struct mapped_device *md)
128{
129 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
130
131 if (!t)
132 return -ENOMEM;
133
134 INIT_LIST_HEAD(&t->devices);
135
136 if (!num_targets)
137 num_targets = KEYS_PER_NODE;
138
139 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
140
141 if (!num_targets) {
142 kfree(t);
143 return -ENOMEM;
144 }
145
146 if (alloc_targets(t, num_targets)) {
147 kfree(t);
148 return -ENOMEM;
149 }
150
151 t->type = DM_TYPE_NONE;
152 t->mode = mode;
153 t->md = md;
154 *result = t;
155 return 0;
156}
157
158static void free_devices(struct list_head *devices, struct mapped_device *md)
159{
160 struct list_head *tmp, *next;
161
162 list_for_each_safe(tmp, next, devices) {
163 struct dm_dev_internal *dd =
164 list_entry(tmp, struct dm_dev_internal, list);
165 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
166 dm_device_name(md), dd->dm_dev->name);
167 dm_put_table_device(md, dd->dm_dev);
168 kfree(dd);
169 }
170}
171
172static void dm_table_destroy_keyslot_manager(struct dm_table *t);
173
174void dm_table_destroy(struct dm_table *t)
175{
176 unsigned int i;
177
178 if (!t)
179 return;
180
181 /* free the indexes */
182 if (t->depth >= 2)
183 kvfree(t->index[t->depth - 2]);
184
185 /* free the targets */
186 for (i = 0; i < t->num_targets; i++) {
187 struct dm_target *tgt = t->targets + i;
188
189 if (tgt->type->dtr)
190 tgt->type->dtr(tgt);
191
192 dm_put_target_type(tgt->type);
193 }
194
195 kvfree(t->highs);
196
197 /* free the device list */
198 free_devices(&t->devices, t->md);
199
200 dm_free_md_mempools(t->mempools);
201
202 dm_table_destroy_keyslot_manager(t);
203
204 kfree(t);
205}
206
207/*
208 * See if we've already got a device in the list.
209 */
210static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
211{
212 struct dm_dev_internal *dd;
213
214 list_for_each_entry (dd, l, list)
215 if (dd->dm_dev->bdev->bd_dev == dev)
216 return dd;
217
218 return NULL;
219}
220
221/*
222 * If possible, this checks an area of a destination device is invalid.
223 */
224static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
225 sector_t start, sector_t len, void *data)
226{
227 struct queue_limits *limits = data;
228 struct block_device *bdev = dev->bdev;
229 sector_t dev_size =
230 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
231 unsigned short logical_block_size_sectors =
232 limits->logical_block_size >> SECTOR_SHIFT;
233 char b[BDEVNAME_SIZE];
234
235 if (!dev_size)
236 return 0;
237
238 if ((start >= dev_size) || (start + len > dev_size)) {
239 DMWARN("%s: %s too small for target: "
240 "start=%llu, len=%llu, dev_size=%llu",
241 dm_device_name(ti->table->md), bdevname(bdev, b),
242 (unsigned long long)start,
243 (unsigned long long)len,
244 (unsigned long long)dev_size);
245 return 1;
246 }
247
248 /*
249 * If the target is mapped to zoned block device(s), check
250 * that the zones are not partially mapped.
251 */
252 if (bdev_is_zoned(bdev)) {
253 unsigned int zone_sectors = bdev_zone_sectors(bdev);
254
255 if (start & (zone_sectors - 1)) {
256 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
257 dm_device_name(ti->table->md),
258 (unsigned long long)start,
259 zone_sectors, bdevname(bdev, b));
260 return 1;
261 }
262
263 /*
264 * Note: The last zone of a zoned block device may be smaller
265 * than other zones. So for a target mapping the end of a
266 * zoned block device with such a zone, len would not be zone
267 * aligned. We do not allow such last smaller zone to be part
268 * of the mapping here to ensure that mappings with multiple
269 * devices do not end up with a smaller zone in the middle of
270 * the sector range.
271 */
272 if (len & (zone_sectors - 1)) {
273 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
274 dm_device_name(ti->table->md),
275 (unsigned long long)len,
276 zone_sectors, bdevname(bdev, b));
277 return 1;
278 }
279 }
280
281 if (logical_block_size_sectors <= 1)
282 return 0;
283
284 if (start & (logical_block_size_sectors - 1)) {
285 DMWARN("%s: start=%llu not aligned to h/w "
286 "logical block size %u of %s",
287 dm_device_name(ti->table->md),
288 (unsigned long long)start,
289 limits->logical_block_size, bdevname(bdev, b));
290 return 1;
291 }
292
293 if (len & (logical_block_size_sectors - 1)) {
294 DMWARN("%s: len=%llu not aligned to h/w "
295 "logical block size %u of %s",
296 dm_device_name(ti->table->md),
297 (unsigned long long)len,
298 limits->logical_block_size, bdevname(bdev, b));
299 return 1;
300 }
301
302 return 0;
303}
304
305/*
306 * This upgrades the mode on an already open dm_dev, being
307 * careful to leave things as they were if we fail to reopen the
308 * device and not to touch the existing bdev field in case
309 * it is accessed concurrently.
310 */
311static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
312 struct mapped_device *md)
313{
314 int r;
315 struct dm_dev *old_dev, *new_dev;
316
317 old_dev = dd->dm_dev;
318
319 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
320 dd->dm_dev->mode | new_mode, &new_dev);
321 if (r)
322 return r;
323
324 dd->dm_dev = new_dev;
325 dm_put_table_device(md, old_dev);
326
327 return 0;
328}
329
330/*
331 * Convert the path to a device
332 */
333dev_t dm_get_dev_t(const char *path)
334{
335 dev_t dev;
336
337 if (lookup_bdev(path, &dev))
338 dev = name_to_dev_t(path);
339 return dev;
340}
341EXPORT_SYMBOL_GPL(dm_get_dev_t);
342
343/*
344 * Add a device to the list, or just increment the usage count if
345 * it's already present.
346 */
347int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
348 struct dm_dev **result)
349{
350 int r;
351 dev_t dev;
352 unsigned int major, minor;
353 char dummy;
354 struct dm_dev_internal *dd;
355 struct dm_table *t = ti->table;
356
357 BUG_ON(!t);
358
359 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
360 /* Extract the major/minor numbers */
361 dev = MKDEV(major, minor);
362 if (MAJOR(dev) != major || MINOR(dev) != minor)
363 return -EOVERFLOW;
364 } else {
365 dev = dm_get_dev_t(path);
366 if (!dev)
367 return -ENODEV;
368 }
369
370 dd = find_device(&t->devices, dev);
371 if (!dd) {
372 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
373 if (!dd)
374 return -ENOMEM;
375
376 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
377 kfree(dd);
378 return r;
379 }
380
381 refcount_set(&dd->count, 1);
382 list_add(&dd->list, &t->devices);
383 goto out;
384
385 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
386 r = upgrade_mode(dd, mode, t->md);
387 if (r)
388 return r;
389 }
390 refcount_inc(&dd->count);
391out:
392 *result = dd->dm_dev;
393 return 0;
394}
395EXPORT_SYMBOL(dm_get_device);
396
397static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
398 sector_t start, sector_t len, void *data)
399{
400 struct queue_limits *limits = data;
401 struct block_device *bdev = dev->bdev;
402 struct request_queue *q = bdev_get_queue(bdev);
403 char b[BDEVNAME_SIZE];
404
405 if (unlikely(!q)) {
406 DMWARN("%s: Cannot set limits for nonexistent device %s",
407 dm_device_name(ti->table->md), bdevname(bdev, b));
408 return 0;
409 }
410
411 if (blk_stack_limits(limits, &q->limits,
412 get_start_sect(bdev) + start) < 0)
413 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
414 "physical_block_size=%u, logical_block_size=%u, "
415 "alignment_offset=%u, start=%llu",
416 dm_device_name(ti->table->md), bdevname(bdev, b),
417 q->limits.physical_block_size,
418 q->limits.logical_block_size,
419 q->limits.alignment_offset,
420 (unsigned long long) start << SECTOR_SHIFT);
421 return 0;
422}
423
424/*
425 * Decrement a device's use count and remove it if necessary.
426 */
427void dm_put_device(struct dm_target *ti, struct dm_dev *d)
428{
429 int found = 0;
430 struct list_head *devices = &ti->table->devices;
431 struct dm_dev_internal *dd;
432
433 list_for_each_entry(dd, devices, list) {
434 if (dd->dm_dev == d) {
435 found = 1;
436 break;
437 }
438 }
439 if (!found) {
440 DMWARN("%s: device %s not in table devices list",
441 dm_device_name(ti->table->md), d->name);
442 return;
443 }
444 if (refcount_dec_and_test(&dd->count)) {
445 dm_put_table_device(ti->table->md, d);
446 list_del(&dd->list);
447 kfree(dd);
448 }
449}
450EXPORT_SYMBOL(dm_put_device);
451
452/*
453 * Checks to see if the target joins onto the end of the table.
454 */
455static int adjoin(struct dm_table *table, struct dm_target *ti)
456{
457 struct dm_target *prev;
458
459 if (!table->num_targets)
460 return !ti->begin;
461
462 prev = &table->targets[table->num_targets - 1];
463 return (ti->begin == (prev->begin + prev->len));
464}
465
466/*
467 * Used to dynamically allocate the arg array.
468 *
469 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
470 * process messages even if some device is suspended. These messages have a
471 * small fixed number of arguments.
472 *
473 * On the other hand, dm-switch needs to process bulk data using messages and
474 * excessive use of GFP_NOIO could cause trouble.
475 */
476static char **realloc_argv(unsigned *size, char **old_argv)
477{
478 char **argv;
479 unsigned new_size;
480 gfp_t gfp;
481
482 if (*size) {
483 new_size = *size * 2;
484 gfp = GFP_KERNEL;
485 } else {
486 new_size = 8;
487 gfp = GFP_NOIO;
488 }
489 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
490 if (argv && old_argv) {
491 memcpy(argv, old_argv, *size * sizeof(*argv));
492 *size = new_size;
493 }
494
495 kfree(old_argv);
496 return argv;
497}
498
499/*
500 * Destructively splits up the argument list to pass to ctr.
501 */
502int dm_split_args(int *argc, char ***argvp, char *input)
503{
504 char *start, *end = input, *out, **argv = NULL;
505 unsigned array_size = 0;
506
507 *argc = 0;
508
509 if (!input) {
510 *argvp = NULL;
511 return 0;
512 }
513
514 argv = realloc_argv(&array_size, argv);
515 if (!argv)
516 return -ENOMEM;
517
518 while (1) {
519 /* Skip whitespace */
520 start = skip_spaces(end);
521
522 if (!*start)
523 break; /* success, we hit the end */
524
525 /* 'out' is used to remove any back-quotes */
526 end = out = start;
527 while (*end) {
528 /* Everything apart from '\0' can be quoted */
529 if (*end == '\\' && *(end + 1)) {
530 *out++ = *(end + 1);
531 end += 2;
532 continue;
533 }
534
535 if (isspace(*end))
536 break; /* end of token */
537
538 *out++ = *end++;
539 }
540
541 /* have we already filled the array ? */
542 if ((*argc + 1) > array_size) {
543 argv = realloc_argv(&array_size, argv);
544 if (!argv)
545 return -ENOMEM;
546 }
547
548 /* we know this is whitespace */
549 if (*end)
550 end++;
551
552 /* terminate the string and put it in the array */
553 *out = '\0';
554 argv[*argc] = start;
555 (*argc)++;
556 }
557
558 *argvp = argv;
559 return 0;
560}
561
562/*
563 * Impose necessary and sufficient conditions on a devices's table such
564 * that any incoming bio which respects its logical_block_size can be
565 * processed successfully. If it falls across the boundary between
566 * two or more targets, the size of each piece it gets split into must
567 * be compatible with the logical_block_size of the target processing it.
568 */
569static int validate_hardware_logical_block_alignment(struct dm_table *table,
570 struct queue_limits *limits)
571{
572 /*
573 * This function uses arithmetic modulo the logical_block_size
574 * (in units of 512-byte sectors).
575 */
576 unsigned short device_logical_block_size_sects =
577 limits->logical_block_size >> SECTOR_SHIFT;
578
579 /*
580 * Offset of the start of the next table entry, mod logical_block_size.
581 */
582 unsigned short next_target_start = 0;
583
584 /*
585 * Given an aligned bio that extends beyond the end of a
586 * target, how many sectors must the next target handle?
587 */
588 unsigned short remaining = 0;
589
590 struct dm_target *ti;
591 struct queue_limits ti_limits;
592 unsigned i;
593
594 /*
595 * Check each entry in the table in turn.
596 */
597 for (i = 0; i < dm_table_get_num_targets(table); i++) {
598 ti = dm_table_get_target(table, i);
599
600 blk_set_stacking_limits(&ti_limits);
601
602 /* combine all target devices' limits */
603 if (ti->type->iterate_devices)
604 ti->type->iterate_devices(ti, dm_set_device_limits,
605 &ti_limits);
606
607 /*
608 * If the remaining sectors fall entirely within this
609 * table entry are they compatible with its logical_block_size?
610 */
611 if (remaining < ti->len &&
612 remaining & ((ti_limits.logical_block_size >>
613 SECTOR_SHIFT) - 1))
614 break; /* Error */
615
616 next_target_start =
617 (unsigned short) ((next_target_start + ti->len) &
618 (device_logical_block_size_sects - 1));
619 remaining = next_target_start ?
620 device_logical_block_size_sects - next_target_start : 0;
621 }
622
623 if (remaining) {
624 DMWARN("%s: table line %u (start sect %llu len %llu) "
625 "not aligned to h/w logical block size %u",
626 dm_device_name(table->md), i,
627 (unsigned long long) ti->begin,
628 (unsigned long long) ti->len,
629 limits->logical_block_size);
630 return -EINVAL;
631 }
632
633 return 0;
634}
635
636int dm_table_add_target(struct dm_table *t, const char *type,
637 sector_t start, sector_t len, char *params)
638{
639 int r = -EINVAL, argc;
640 char **argv;
641 struct dm_target *tgt;
642
643 if (t->singleton) {
644 DMERR("%s: target type %s must appear alone in table",
645 dm_device_name(t->md), t->targets->type->name);
646 return -EINVAL;
647 }
648
649 BUG_ON(t->num_targets >= t->num_allocated);
650
651 tgt = t->targets + t->num_targets;
652 memset(tgt, 0, sizeof(*tgt));
653
654 if (!len) {
655 DMERR("%s: zero-length target", dm_device_name(t->md));
656 return -EINVAL;
657 }
658
659 tgt->type = dm_get_target_type(type);
660 if (!tgt->type) {
661 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
662 return -EINVAL;
663 }
664
665 if (dm_target_needs_singleton(tgt->type)) {
666 if (t->num_targets) {
667 tgt->error = "singleton target type must appear alone in table";
668 goto bad;
669 }
670 t->singleton = true;
671 }
672
673 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
674 tgt->error = "target type may not be included in a read-only table";
675 goto bad;
676 }
677
678 if (t->immutable_target_type) {
679 if (t->immutable_target_type != tgt->type) {
680 tgt->error = "immutable target type cannot be mixed with other target types";
681 goto bad;
682 }
683 } else if (dm_target_is_immutable(tgt->type)) {
684 if (t->num_targets) {
685 tgt->error = "immutable target type cannot be mixed with other target types";
686 goto bad;
687 }
688 t->immutable_target_type = tgt->type;
689 }
690
691 if (dm_target_has_integrity(tgt->type))
692 t->integrity_added = 1;
693
694 tgt->table = t;
695 tgt->begin = start;
696 tgt->len = len;
697 tgt->error = "Unknown error";
698
699 /*
700 * Does this target adjoin the previous one ?
701 */
702 if (!adjoin(t, tgt)) {
703 tgt->error = "Gap in table";
704 goto bad;
705 }
706
707 r = dm_split_args(&argc, &argv, params);
708 if (r) {
709 tgt->error = "couldn't split parameters (insufficient memory)";
710 goto bad;
711 }
712
713 r = tgt->type->ctr(tgt, argc, argv);
714 kfree(argv);
715 if (r)
716 goto bad;
717
718 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
719
720 if (!tgt->num_discard_bios && tgt->discards_supported)
721 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
722 dm_device_name(t->md), type);
723
724 return 0;
725
726 bad:
727 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
728 dm_put_target_type(tgt->type);
729 return r;
730}
731
732/*
733 * Target argument parsing helpers.
734 */
735static int validate_next_arg(const struct dm_arg *arg,
736 struct dm_arg_set *arg_set,
737 unsigned *value, char **error, unsigned grouped)
738{
739 const char *arg_str = dm_shift_arg(arg_set);
740 char dummy;
741
742 if (!arg_str ||
743 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
744 (*value < arg->min) ||
745 (*value > arg->max) ||
746 (grouped && arg_set->argc < *value)) {
747 *error = arg->error;
748 return -EINVAL;
749 }
750
751 return 0;
752}
753
754int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
755 unsigned *value, char **error)
756{
757 return validate_next_arg(arg, arg_set, value, error, 0);
758}
759EXPORT_SYMBOL(dm_read_arg);
760
761int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
762 unsigned *value, char **error)
763{
764 return validate_next_arg(arg, arg_set, value, error, 1);
765}
766EXPORT_SYMBOL(dm_read_arg_group);
767
768const char *dm_shift_arg(struct dm_arg_set *as)
769{
770 char *r;
771
772 if (as->argc) {
773 as->argc--;
774 r = *as->argv;
775 as->argv++;
776 return r;
777 }
778
779 return NULL;
780}
781EXPORT_SYMBOL(dm_shift_arg);
782
783void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
784{
785 BUG_ON(as->argc < num_args);
786 as->argc -= num_args;
787 as->argv += num_args;
788}
789EXPORT_SYMBOL(dm_consume_args);
790
791static bool __table_type_bio_based(enum dm_queue_mode table_type)
792{
793 return (table_type == DM_TYPE_BIO_BASED ||
794 table_type == DM_TYPE_DAX_BIO_BASED);
795}
796
797static bool __table_type_request_based(enum dm_queue_mode table_type)
798{
799 return table_type == DM_TYPE_REQUEST_BASED;
800}
801
802void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
803{
804 t->type = type;
805}
806EXPORT_SYMBOL_GPL(dm_table_set_type);
807
808/* validate the dax capability of the target device span */
809int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
810 sector_t start, sector_t len, void *data)
811{
812 int blocksize = *(int *) data, id;
813 bool rc;
814
815 id = dax_read_lock();
816 rc = !dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
817 dax_read_unlock(id);
818
819 return rc;
820}
821
822/* Check devices support synchronous DAX */
823static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
824 sector_t start, sector_t len, void *data)
825{
826 return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
827}
828
829bool dm_table_supports_dax(struct dm_table *t,
830 iterate_devices_callout_fn iterate_fn, int *blocksize)
831{
832 struct dm_target *ti;
833 unsigned i;
834
835 /* Ensure that all targets support DAX. */
836 for (i = 0; i < dm_table_get_num_targets(t); i++) {
837 ti = dm_table_get_target(t, i);
838
839 if (!ti->type->direct_access)
840 return false;
841
842 if (!ti->type->iterate_devices ||
843 ti->type->iterate_devices(ti, iterate_fn, blocksize))
844 return false;
845 }
846
847 return true;
848}
849
850static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
851 sector_t start, sector_t len, void *data)
852{
853 struct block_device *bdev = dev->bdev;
854 struct request_queue *q = bdev_get_queue(bdev);
855
856 /* request-based cannot stack on partitions! */
857 if (bdev_is_partition(bdev))
858 return false;
859
860 return queue_is_mq(q);
861}
862
863static int dm_table_determine_type(struct dm_table *t)
864{
865 unsigned i;
866 unsigned bio_based = 0, request_based = 0, hybrid = 0;
867 struct dm_target *tgt;
868 struct list_head *devices = dm_table_get_devices(t);
869 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
870 int page_size = PAGE_SIZE;
871
872 if (t->type != DM_TYPE_NONE) {
873 /* target already set the table's type */
874 if (t->type == DM_TYPE_BIO_BASED) {
875 /* possibly upgrade to a variant of bio-based */
876 goto verify_bio_based;
877 }
878 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
879 goto verify_rq_based;
880 }
881
882 for (i = 0; i < t->num_targets; i++) {
883 tgt = t->targets + i;
884 if (dm_target_hybrid(tgt))
885 hybrid = 1;
886 else if (dm_target_request_based(tgt))
887 request_based = 1;
888 else
889 bio_based = 1;
890
891 if (bio_based && request_based) {
892 DMERR("Inconsistent table: different target types"
893 " can't be mixed up");
894 return -EINVAL;
895 }
896 }
897
898 if (hybrid && !bio_based && !request_based) {
899 /*
900 * The targets can work either way.
901 * Determine the type from the live device.
902 * Default to bio-based if device is new.
903 */
904 if (__table_type_request_based(live_md_type))
905 request_based = 1;
906 else
907 bio_based = 1;
908 }
909
910 if (bio_based) {
911verify_bio_based:
912 /* We must use this table as bio-based */
913 t->type = DM_TYPE_BIO_BASED;
914 if (dm_table_supports_dax(t, device_not_dax_capable, &page_size) ||
915 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
916 t->type = DM_TYPE_DAX_BIO_BASED;
917 }
918 return 0;
919 }
920
921 BUG_ON(!request_based); /* No targets in this table */
922
923 t->type = DM_TYPE_REQUEST_BASED;
924
925verify_rq_based:
926 /*
927 * Request-based dm supports only tables that have a single target now.
928 * To support multiple targets, request splitting support is needed,
929 * and that needs lots of changes in the block-layer.
930 * (e.g. request completion process for partial completion.)
931 */
932 if (t->num_targets > 1) {
933 DMERR("request-based DM doesn't support multiple targets");
934 return -EINVAL;
935 }
936
937 if (list_empty(devices)) {
938 int srcu_idx;
939 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
940
941 /* inherit live table's type */
942 if (live_table)
943 t->type = live_table->type;
944 dm_put_live_table(t->md, srcu_idx);
945 return 0;
946 }
947
948 tgt = dm_table_get_immutable_target(t);
949 if (!tgt) {
950 DMERR("table load rejected: immutable target is required");
951 return -EINVAL;
952 } else if (tgt->max_io_len) {
953 DMERR("table load rejected: immutable target that splits IO is not supported");
954 return -EINVAL;
955 }
956
957 /* Non-request-stackable devices can't be used for request-based dm */
958 if (!tgt->type->iterate_devices ||
959 !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
960 DMERR("table load rejected: including non-request-stackable devices");
961 return -EINVAL;
962 }
963
964 return 0;
965}
966
967enum dm_queue_mode dm_table_get_type(struct dm_table *t)
968{
969 return t->type;
970}
971
972struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
973{
974 return t->immutable_target_type;
975}
976
977struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
978{
979 /* Immutable target is implicitly a singleton */
980 if (t->num_targets > 1 ||
981 !dm_target_is_immutable(t->targets[0].type))
982 return NULL;
983
984 return t->targets;
985}
986
987struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
988{
989 struct dm_target *ti;
990 unsigned i;
991
992 for (i = 0; i < dm_table_get_num_targets(t); i++) {
993 ti = dm_table_get_target(t, i);
994 if (dm_target_is_wildcard(ti->type))
995 return ti;
996 }
997
998 return NULL;
999}
1000
1001bool dm_table_bio_based(struct dm_table *t)
1002{
1003 return __table_type_bio_based(dm_table_get_type(t));
1004}
1005
1006bool dm_table_request_based(struct dm_table *t)
1007{
1008 return __table_type_request_based(dm_table_get_type(t));
1009}
1010
1011static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1012{
1013 enum dm_queue_mode type = dm_table_get_type(t);
1014 unsigned per_io_data_size = 0;
1015 unsigned min_pool_size = 0;
1016 struct dm_target *ti;
1017 unsigned i;
1018
1019 if (unlikely(type == DM_TYPE_NONE)) {
1020 DMWARN("no table type is set, can't allocate mempools");
1021 return -EINVAL;
1022 }
1023
1024 if (__table_type_bio_based(type))
1025 for (i = 0; i < t->num_targets; i++) {
1026 ti = t->targets + i;
1027 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1028 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1029 }
1030
1031 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1032 per_io_data_size, min_pool_size);
1033 if (!t->mempools)
1034 return -ENOMEM;
1035
1036 return 0;
1037}
1038
1039void dm_table_free_md_mempools(struct dm_table *t)
1040{
1041 dm_free_md_mempools(t->mempools);
1042 t->mempools = NULL;
1043}
1044
1045struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1046{
1047 return t->mempools;
1048}
1049
1050static int setup_indexes(struct dm_table *t)
1051{
1052 int i;
1053 unsigned int total = 0;
1054 sector_t *indexes;
1055
1056 /* allocate the space for *all* the indexes */
1057 for (i = t->depth - 2; i >= 0; i--) {
1058 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1059 total += t->counts[i];
1060 }
1061
1062 indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1063 if (!indexes)
1064 return -ENOMEM;
1065
1066 /* set up internal nodes, bottom-up */
1067 for (i = t->depth - 2; i >= 0; i--) {
1068 t->index[i] = indexes;
1069 indexes += (KEYS_PER_NODE * t->counts[i]);
1070 setup_btree_index(i, t);
1071 }
1072
1073 return 0;
1074}
1075
1076/*
1077 * Builds the btree to index the map.
1078 */
1079static int dm_table_build_index(struct dm_table *t)
1080{
1081 int r = 0;
1082 unsigned int leaf_nodes;
1083
1084 /* how many indexes will the btree have ? */
1085 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1086 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1087
1088 /* leaf layer has already been set up */
1089 t->counts[t->depth - 1] = leaf_nodes;
1090 t->index[t->depth - 1] = t->highs;
1091
1092 if (t->depth >= 2)
1093 r = setup_indexes(t);
1094
1095 return r;
1096}
1097
1098static bool integrity_profile_exists(struct gendisk *disk)
1099{
1100 return !!blk_get_integrity(disk);
1101}
1102
1103/*
1104 * Get a disk whose integrity profile reflects the table's profile.
1105 * Returns NULL if integrity support was inconsistent or unavailable.
1106 */
1107static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1108{
1109 struct list_head *devices = dm_table_get_devices(t);
1110 struct dm_dev_internal *dd = NULL;
1111 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1112 unsigned i;
1113
1114 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1115 struct dm_target *ti = dm_table_get_target(t, i);
1116 if (!dm_target_passes_integrity(ti->type))
1117 goto no_integrity;
1118 }
1119
1120 list_for_each_entry(dd, devices, list) {
1121 template_disk = dd->dm_dev->bdev->bd_disk;
1122 if (!integrity_profile_exists(template_disk))
1123 goto no_integrity;
1124 else if (prev_disk &&
1125 blk_integrity_compare(prev_disk, template_disk) < 0)
1126 goto no_integrity;
1127 prev_disk = template_disk;
1128 }
1129
1130 return template_disk;
1131
1132no_integrity:
1133 if (prev_disk)
1134 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1135 dm_device_name(t->md),
1136 prev_disk->disk_name,
1137 template_disk->disk_name);
1138 return NULL;
1139}
1140
1141/*
1142 * Register the mapped device for blk_integrity support if the
1143 * underlying devices have an integrity profile. But all devices may
1144 * not have matching profiles (checking all devices isn't reliable
1145 * during table load because this table may use other DM device(s) which
1146 * must be resumed before they will have an initialized integity
1147 * profile). Consequently, stacked DM devices force a 2 stage integrity
1148 * profile validation: First pass during table load, final pass during
1149 * resume.
1150 */
1151static int dm_table_register_integrity(struct dm_table *t)
1152{
1153 struct mapped_device *md = t->md;
1154 struct gendisk *template_disk = NULL;
1155
1156 /* If target handles integrity itself do not register it here. */
1157 if (t->integrity_added)
1158 return 0;
1159
1160 template_disk = dm_table_get_integrity_disk(t);
1161 if (!template_disk)
1162 return 0;
1163
1164 if (!integrity_profile_exists(dm_disk(md))) {
1165 t->integrity_supported = true;
1166 /*
1167 * Register integrity profile during table load; we can do
1168 * this because the final profile must match during resume.
1169 */
1170 blk_integrity_register(dm_disk(md),
1171 blk_get_integrity(template_disk));
1172 return 0;
1173 }
1174
1175 /*
1176 * If DM device already has an initialized integrity
1177 * profile the new profile should not conflict.
1178 */
1179 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1180 DMWARN("%s: conflict with existing integrity profile: "
1181 "%s profile mismatch",
1182 dm_device_name(t->md),
1183 template_disk->disk_name);
1184 return 1;
1185 }
1186
1187 /* Preserve existing integrity profile */
1188 t->integrity_supported = true;
1189 return 0;
1190}
1191
1192#ifdef CONFIG_BLK_INLINE_ENCRYPTION
1193
1194struct dm_keyslot_manager {
1195 struct blk_keyslot_manager ksm;
1196 struct mapped_device *md;
1197};
1198
1199struct dm_keyslot_evict_args {
1200 const struct blk_crypto_key *key;
1201 int err;
1202};
1203
1204static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1205 sector_t start, sector_t len, void *data)
1206{
1207 struct dm_keyslot_evict_args *args = data;
1208 int err;
1209
1210 err = blk_crypto_evict_key(bdev_get_queue(dev->bdev), args->key);
1211 if (!args->err)
1212 args->err = err;
1213 /* Always try to evict the key from all devices. */
1214 return 0;
1215}
1216
1217/*
1218 * When an inline encryption key is evicted from a device-mapper device, evict
1219 * it from all the underlying devices.
1220 */
1221static int dm_keyslot_evict(struct blk_keyslot_manager *ksm,
1222 const struct blk_crypto_key *key, unsigned int slot)
1223{
1224 struct dm_keyslot_manager *dksm = container_of(ksm,
1225 struct dm_keyslot_manager,
1226 ksm);
1227 struct mapped_device *md = dksm->md;
1228 struct dm_keyslot_evict_args args = { key };
1229 struct dm_table *t;
1230 int srcu_idx;
1231 int i;
1232 struct dm_target *ti;
1233
1234 t = dm_get_live_table(md, &srcu_idx);
1235 if (!t)
1236 return 0;
1237 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1238 ti = dm_table_get_target(t, i);
1239 if (!ti->type->iterate_devices)
1240 continue;
1241 ti->type->iterate_devices(ti, dm_keyslot_evict_callback, &args);
1242 }
1243 dm_put_live_table(md, srcu_idx);
1244 return args.err;
1245}
1246
1247static const struct blk_ksm_ll_ops dm_ksm_ll_ops = {
1248 .keyslot_evict = dm_keyslot_evict,
1249};
1250
1251static int device_intersect_crypto_modes(struct dm_target *ti,
1252 struct dm_dev *dev, sector_t start,
1253 sector_t len, void *data)
1254{
1255 struct blk_keyslot_manager *parent = data;
1256 struct blk_keyslot_manager *child = bdev_get_queue(dev->bdev)->ksm;
1257
1258 blk_ksm_intersect_modes(parent, child);
1259 return 0;
1260}
1261
1262void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm)
1263{
1264 struct dm_keyslot_manager *dksm = container_of(ksm,
1265 struct dm_keyslot_manager,
1266 ksm);
1267
1268 if (!ksm)
1269 return;
1270
1271 blk_ksm_destroy(ksm);
1272 kfree(dksm);
1273}
1274
1275static void dm_table_destroy_keyslot_manager(struct dm_table *t)
1276{
1277 dm_destroy_keyslot_manager(t->ksm);
1278 t->ksm = NULL;
1279}
1280
1281/*
1282 * Constructs and initializes t->ksm with a keyslot manager that
1283 * represents the common set of crypto capabilities of the devices
1284 * described by the dm_table. However, if the constructed keyslot
1285 * manager does not support a superset of the crypto capabilities
1286 * supported by the current keyslot manager of the mapped_device,
1287 * it returns an error instead, since we don't support restricting
1288 * crypto capabilities on table changes. Finally, if the constructed
1289 * keyslot manager doesn't actually support any crypto modes at all,
1290 * it just returns NULL.
1291 */
1292static int dm_table_construct_keyslot_manager(struct dm_table *t)
1293{
1294 struct dm_keyslot_manager *dksm;
1295 struct blk_keyslot_manager *ksm;
1296 struct dm_target *ti;
1297 unsigned int i;
1298 bool ksm_is_empty = true;
1299
1300 dksm = kmalloc(sizeof(*dksm), GFP_KERNEL);
1301 if (!dksm)
1302 return -ENOMEM;
1303 dksm->md = t->md;
1304
1305 ksm = &dksm->ksm;
1306 blk_ksm_init_passthrough(ksm);
1307 ksm->ksm_ll_ops = dm_ksm_ll_ops;
1308 ksm->max_dun_bytes_supported = UINT_MAX;
1309 memset(ksm->crypto_modes_supported, 0xFF,
1310 sizeof(ksm->crypto_modes_supported));
1311
1312 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1313 ti = dm_table_get_target(t, i);
1314
1315 if (!dm_target_passes_crypto(ti->type)) {
1316 blk_ksm_intersect_modes(ksm, NULL);
1317 break;
1318 }
1319 if (!ti->type->iterate_devices)
1320 continue;
1321 ti->type->iterate_devices(ti, device_intersect_crypto_modes,
1322 ksm);
1323 }
1324
1325 if (t->md->queue && !blk_ksm_is_superset(ksm, t->md->queue->ksm)) {
1326 DMWARN("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1327 dm_destroy_keyslot_manager(ksm);
1328 return -EINVAL;
1329 }
1330
1331 /*
1332 * If the new KSM doesn't actually support any crypto modes, we may as
1333 * well represent it with a NULL ksm.
1334 */
1335 ksm_is_empty = true;
1336 for (i = 0; i < ARRAY_SIZE(ksm->crypto_modes_supported); i++) {
1337 if (ksm->crypto_modes_supported[i]) {
1338 ksm_is_empty = false;
1339 break;
1340 }
1341 }
1342
1343 if (ksm_is_empty) {
1344 dm_destroy_keyslot_manager(ksm);
1345 ksm = NULL;
1346 }
1347
1348 /*
1349 * t->ksm is only set temporarily while the table is being set
1350 * up, and it gets set to NULL after the capabilities have
1351 * been transferred to the request_queue.
1352 */
1353 t->ksm = ksm;
1354
1355 return 0;
1356}
1357
1358static void dm_update_keyslot_manager(struct request_queue *q,
1359 struct dm_table *t)
1360{
1361 if (!t->ksm)
1362 return;
1363
1364 /* Make the ksm less restrictive */
1365 if (!q->ksm) {
1366 blk_ksm_register(t->ksm, q);
1367 } else {
1368 blk_ksm_update_capabilities(q->ksm, t->ksm);
1369 dm_destroy_keyslot_manager(t->ksm);
1370 }
1371 t->ksm = NULL;
1372}
1373
1374#else /* CONFIG_BLK_INLINE_ENCRYPTION */
1375
1376static int dm_table_construct_keyslot_manager(struct dm_table *t)
1377{
1378 return 0;
1379}
1380
1381void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm)
1382{
1383}
1384
1385static void dm_table_destroy_keyslot_manager(struct dm_table *t)
1386{
1387}
1388
1389static void dm_update_keyslot_manager(struct request_queue *q,
1390 struct dm_table *t)
1391{
1392}
1393
1394#endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1395
1396/*
1397 * Prepares the table for use by building the indices,
1398 * setting the type, and allocating mempools.
1399 */
1400int dm_table_complete(struct dm_table *t)
1401{
1402 int r;
1403
1404 r = dm_table_determine_type(t);
1405 if (r) {
1406 DMERR("unable to determine table type");
1407 return r;
1408 }
1409
1410 r = dm_table_build_index(t);
1411 if (r) {
1412 DMERR("unable to build btrees");
1413 return r;
1414 }
1415
1416 r = dm_table_register_integrity(t);
1417 if (r) {
1418 DMERR("could not register integrity profile.");
1419 return r;
1420 }
1421
1422 r = dm_table_construct_keyslot_manager(t);
1423 if (r) {
1424 DMERR("could not construct keyslot manager.");
1425 return r;
1426 }
1427
1428 r = dm_table_alloc_md_mempools(t, t->md);
1429 if (r)
1430 DMERR("unable to allocate mempools");
1431
1432 return r;
1433}
1434
1435static DEFINE_MUTEX(_event_lock);
1436void dm_table_event_callback(struct dm_table *t,
1437 void (*fn)(void *), void *context)
1438{
1439 mutex_lock(&_event_lock);
1440 t->event_fn = fn;
1441 t->event_context = context;
1442 mutex_unlock(&_event_lock);
1443}
1444
1445void dm_table_event(struct dm_table *t)
1446{
1447 mutex_lock(&_event_lock);
1448 if (t->event_fn)
1449 t->event_fn(t->event_context);
1450 mutex_unlock(&_event_lock);
1451}
1452EXPORT_SYMBOL(dm_table_event);
1453
1454inline sector_t dm_table_get_size(struct dm_table *t)
1455{
1456 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1457}
1458EXPORT_SYMBOL(dm_table_get_size);
1459
1460struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1461{
1462 if (index >= t->num_targets)
1463 return NULL;
1464
1465 return t->targets + index;
1466}
1467
1468/*
1469 * Search the btree for the correct target.
1470 *
1471 * Caller should check returned pointer for NULL
1472 * to trap I/O beyond end of device.
1473 */
1474struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1475{
1476 unsigned int l, n = 0, k = 0;
1477 sector_t *node;
1478
1479 if (unlikely(sector >= dm_table_get_size(t)))
1480 return NULL;
1481
1482 for (l = 0; l < t->depth; l++) {
1483 n = get_child(n, k);
1484 node = get_node(t, l, n);
1485
1486 for (k = 0; k < KEYS_PER_NODE; k++)
1487 if (node[k] >= sector)
1488 break;
1489 }
1490
1491 return &t->targets[(KEYS_PER_NODE * n) + k];
1492}
1493
1494/*
1495 * type->iterate_devices() should be called when the sanity check needs to
1496 * iterate and check all underlying data devices. iterate_devices() will
1497 * iterate all underlying data devices until it encounters a non-zero return
1498 * code, returned by whether the input iterate_devices_callout_fn, or
1499 * iterate_devices() itself internally.
1500 *
1501 * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1502 * iterate multiple underlying devices internally, in which case a non-zero
1503 * return code returned by iterate_devices_callout_fn will stop the iteration
1504 * in advance.
1505 *
1506 * Cases requiring _any_ underlying device supporting some kind of attribute,
1507 * should use the iteration structure like dm_table_any_dev_attr(), or call
1508 * it directly. @func should handle semantics of positive examples, e.g.
1509 * capable of something.
1510 *
1511 * Cases requiring _all_ underlying devices supporting some kind of attribute,
1512 * should use the iteration structure like dm_table_supports_nowait() or
1513 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1514 * uses an @anti_func that handle semantics of counter examples, e.g. not
1515 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1516 */
1517static bool dm_table_any_dev_attr(struct dm_table *t,
1518 iterate_devices_callout_fn func, void *data)
1519{
1520 struct dm_target *ti;
1521 unsigned int i;
1522
1523 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1524 ti = dm_table_get_target(t, i);
1525
1526 if (ti->type->iterate_devices &&
1527 ti->type->iterate_devices(ti, func, data))
1528 return true;
1529 }
1530
1531 return false;
1532}
1533
1534static int count_device(struct dm_target *ti, struct dm_dev *dev,
1535 sector_t start, sector_t len, void *data)
1536{
1537 unsigned *num_devices = data;
1538
1539 (*num_devices)++;
1540
1541 return 0;
1542}
1543
1544/*
1545 * Check whether a table has no data devices attached using each
1546 * target's iterate_devices method.
1547 * Returns false if the result is unknown because a target doesn't
1548 * support iterate_devices.
1549 */
1550bool dm_table_has_no_data_devices(struct dm_table *table)
1551{
1552 struct dm_target *ti;
1553 unsigned i, num_devices;
1554
1555 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1556 ti = dm_table_get_target(table, i);
1557
1558 if (!ti->type->iterate_devices)
1559 return false;
1560
1561 num_devices = 0;
1562 ti->type->iterate_devices(ti, count_device, &num_devices);
1563 if (num_devices)
1564 return false;
1565 }
1566
1567 return true;
1568}
1569
1570static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1571 sector_t start, sector_t len, void *data)
1572{
1573 struct request_queue *q = bdev_get_queue(dev->bdev);
1574 enum blk_zoned_model *zoned_model = data;
1575
1576 return blk_queue_zoned_model(q) != *zoned_model;
1577}
1578
1579/*
1580 * Check the device zoned model based on the target feature flag. If the target
1581 * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1582 * also accepted but all devices must have the same zoned model. If the target
1583 * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1584 * zoned model with all zoned devices having the same zone size.
1585 */
1586static bool dm_table_supports_zoned_model(struct dm_table *t,
1587 enum blk_zoned_model zoned_model)
1588{
1589 struct dm_target *ti;
1590 unsigned i;
1591
1592 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1593 ti = dm_table_get_target(t, i);
1594
1595 if (dm_target_supports_zoned_hm(ti->type)) {
1596 if (!ti->type->iterate_devices ||
1597 ti->type->iterate_devices(ti, device_not_zoned_model,
1598 &zoned_model))
1599 return false;
1600 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1601 if (zoned_model == BLK_ZONED_HM)
1602 return false;
1603 }
1604 }
1605
1606 return true;
1607}
1608
1609static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1610 sector_t start, sector_t len, void *data)
1611{
1612 struct request_queue *q = bdev_get_queue(dev->bdev);
1613 unsigned int *zone_sectors = data;
1614
1615 if (!blk_queue_is_zoned(q))
1616 return 0;
1617
1618 return blk_queue_zone_sectors(q) != *zone_sectors;
1619}
1620
1621/*
1622 * Check consistency of zoned model and zone sectors across all targets. For
1623 * zone sectors, if the destination device is a zoned block device, it shall
1624 * have the specified zone_sectors.
1625 */
1626static int validate_hardware_zoned_model(struct dm_table *table,
1627 enum blk_zoned_model zoned_model,
1628 unsigned int zone_sectors)
1629{
1630 if (zoned_model == BLK_ZONED_NONE)
1631 return 0;
1632
1633 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1634 DMERR("%s: zoned model is not consistent across all devices",
1635 dm_device_name(table->md));
1636 return -EINVAL;
1637 }
1638
1639 /* Check zone size validity and compatibility */
1640 if (!zone_sectors || !is_power_of_2(zone_sectors))
1641 return -EINVAL;
1642
1643 if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1644 DMERR("%s: zone sectors is not consistent across all zoned devices",
1645 dm_device_name(table->md));
1646 return -EINVAL;
1647 }
1648
1649 return 0;
1650}
1651
1652/*
1653 * Establish the new table's queue_limits and validate them.
1654 */
1655int dm_calculate_queue_limits(struct dm_table *table,
1656 struct queue_limits *limits)
1657{
1658 struct dm_target *ti;
1659 struct queue_limits ti_limits;
1660 unsigned i;
1661 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1662 unsigned int zone_sectors = 0;
1663
1664 blk_set_stacking_limits(limits);
1665
1666 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1667 blk_set_stacking_limits(&ti_limits);
1668
1669 ti = dm_table_get_target(table, i);
1670
1671 if (!ti->type->iterate_devices)
1672 goto combine_limits;
1673
1674 /*
1675 * Combine queue limits of all the devices this target uses.
1676 */
1677 ti->type->iterate_devices(ti, dm_set_device_limits,
1678 &ti_limits);
1679
1680 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1681 /*
1682 * After stacking all limits, validate all devices
1683 * in table support this zoned model and zone sectors.
1684 */
1685 zoned_model = ti_limits.zoned;
1686 zone_sectors = ti_limits.chunk_sectors;
1687 }
1688
1689 /* Set I/O hints portion of queue limits */
1690 if (ti->type->io_hints)
1691 ti->type->io_hints(ti, &ti_limits);
1692
1693 /*
1694 * Check each device area is consistent with the target's
1695 * overall queue limits.
1696 */
1697 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1698 &ti_limits))
1699 return -EINVAL;
1700
1701combine_limits:
1702 /*
1703 * Merge this target's queue limits into the overall limits
1704 * for the table.
1705 */
1706 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1707 DMWARN("%s: adding target device "
1708 "(start sect %llu len %llu) "
1709 "caused an alignment inconsistency",
1710 dm_device_name(table->md),
1711 (unsigned long long) ti->begin,
1712 (unsigned long long) ti->len);
1713 }
1714
1715 /*
1716 * Verify that the zoned model and zone sectors, as determined before
1717 * any .io_hints override, are the same across all devices in the table.
1718 * - this is especially relevant if .io_hints is emulating a disk-managed
1719 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1720 * BUT...
1721 */
1722 if (limits->zoned != BLK_ZONED_NONE) {
1723 /*
1724 * ...IF the above limits stacking determined a zoned model
1725 * validate that all of the table's devices conform to it.
1726 */
1727 zoned_model = limits->zoned;
1728 zone_sectors = limits->chunk_sectors;
1729 }
1730 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1731 return -EINVAL;
1732
1733 return validate_hardware_logical_block_alignment(table, limits);
1734}
1735
1736/*
1737 * Verify that all devices have an integrity profile that matches the
1738 * DM device's registered integrity profile. If the profiles don't
1739 * match then unregister the DM device's integrity profile.
1740 */
1741static void dm_table_verify_integrity(struct dm_table *t)
1742{
1743 struct gendisk *template_disk = NULL;
1744
1745 if (t->integrity_added)
1746 return;
1747
1748 if (t->integrity_supported) {
1749 /*
1750 * Verify that the original integrity profile
1751 * matches all the devices in this table.
1752 */
1753 template_disk = dm_table_get_integrity_disk(t);
1754 if (template_disk &&
1755 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1756 return;
1757 }
1758
1759 if (integrity_profile_exists(dm_disk(t->md))) {
1760 DMWARN("%s: unable to establish an integrity profile",
1761 dm_device_name(t->md));
1762 blk_integrity_unregister(dm_disk(t->md));
1763 }
1764}
1765
1766static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1767 sector_t start, sector_t len, void *data)
1768{
1769 unsigned long flush = (unsigned long) data;
1770 struct request_queue *q = bdev_get_queue(dev->bdev);
1771
1772 return (q->queue_flags & flush);
1773}
1774
1775static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1776{
1777 struct dm_target *ti;
1778 unsigned i;
1779
1780 /*
1781 * Require at least one underlying device to support flushes.
1782 * t->devices includes internal dm devices such as mirror logs
1783 * so we need to use iterate_devices here, which targets
1784 * supporting flushes must provide.
1785 */
1786 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1787 ti = dm_table_get_target(t, i);
1788
1789 if (!ti->num_flush_bios)
1790 continue;
1791
1792 if (ti->flush_supported)
1793 return true;
1794
1795 if (ti->type->iterate_devices &&
1796 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1797 return true;
1798 }
1799
1800 return false;
1801}
1802
1803static int device_dax_write_cache_enabled(struct dm_target *ti,
1804 struct dm_dev *dev, sector_t start,
1805 sector_t len, void *data)
1806{
1807 struct dax_device *dax_dev = dev->dax_dev;
1808
1809 if (!dax_dev)
1810 return false;
1811
1812 if (dax_write_cache_enabled(dax_dev))
1813 return true;
1814 return false;
1815}
1816
1817static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1818 sector_t start, sector_t len, void *data)
1819{
1820 struct request_queue *q = bdev_get_queue(dev->bdev);
1821
1822 return !blk_queue_nonrot(q);
1823}
1824
1825static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1826 sector_t start, sector_t len, void *data)
1827{
1828 struct request_queue *q = bdev_get_queue(dev->bdev);
1829
1830 return !blk_queue_add_random(q);
1831}
1832
1833static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1834 sector_t start, sector_t len, void *data)
1835{
1836 struct request_queue *q = bdev_get_queue(dev->bdev);
1837
1838 return !q->limits.max_write_same_sectors;
1839}
1840
1841static bool dm_table_supports_write_same(struct dm_table *t)
1842{
1843 struct dm_target *ti;
1844 unsigned i;
1845
1846 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1847 ti = dm_table_get_target(t, i);
1848
1849 if (!ti->num_write_same_bios)
1850 return false;
1851
1852 if (!ti->type->iterate_devices ||
1853 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1854 return false;
1855 }
1856
1857 return true;
1858}
1859
1860static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1861 sector_t start, sector_t len, void *data)
1862{
1863 struct request_queue *q = bdev_get_queue(dev->bdev);
1864
1865 return !q->limits.max_write_zeroes_sectors;
1866}
1867
1868static bool dm_table_supports_write_zeroes(struct dm_table *t)
1869{
1870 struct dm_target *ti;
1871 unsigned i = 0;
1872
1873 while (i < dm_table_get_num_targets(t)) {
1874 ti = dm_table_get_target(t, i++);
1875
1876 if (!ti->num_write_zeroes_bios)
1877 return false;
1878
1879 if (!ti->type->iterate_devices ||
1880 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1881 return false;
1882 }
1883
1884 return true;
1885}
1886
1887static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1888 sector_t start, sector_t len, void *data)
1889{
1890 struct request_queue *q = bdev_get_queue(dev->bdev);
1891
1892 return !blk_queue_nowait(q);
1893}
1894
1895static bool dm_table_supports_nowait(struct dm_table *t)
1896{
1897 struct dm_target *ti;
1898 unsigned i = 0;
1899
1900 while (i < dm_table_get_num_targets(t)) {
1901 ti = dm_table_get_target(t, i++);
1902
1903 if (!dm_target_supports_nowait(ti->type))
1904 return false;
1905
1906 if (!ti->type->iterate_devices ||
1907 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1908 return false;
1909 }
1910
1911 return true;
1912}
1913
1914static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1915 sector_t start, sector_t len, void *data)
1916{
1917 struct request_queue *q = bdev_get_queue(dev->bdev);
1918
1919 return !blk_queue_discard(q);
1920}
1921
1922static bool dm_table_supports_discards(struct dm_table *t)
1923{
1924 struct dm_target *ti;
1925 unsigned i;
1926
1927 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1928 ti = dm_table_get_target(t, i);
1929
1930 if (!ti->num_discard_bios)
1931 return false;
1932
1933 /*
1934 * Either the target provides discard support (as implied by setting
1935 * 'discards_supported') or it relies on _all_ data devices having
1936 * discard support.
1937 */
1938 if (!ti->discards_supported &&
1939 (!ti->type->iterate_devices ||
1940 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1941 return false;
1942 }
1943
1944 return true;
1945}
1946
1947static int device_not_secure_erase_capable(struct dm_target *ti,
1948 struct dm_dev *dev, sector_t start,
1949 sector_t len, void *data)
1950{
1951 struct request_queue *q = bdev_get_queue(dev->bdev);
1952
1953 return !blk_queue_secure_erase(q);
1954}
1955
1956static bool dm_table_supports_secure_erase(struct dm_table *t)
1957{
1958 struct dm_target *ti;
1959 unsigned int i;
1960
1961 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1962 ti = dm_table_get_target(t, i);
1963
1964 if (!ti->num_secure_erase_bios)
1965 return false;
1966
1967 if (!ti->type->iterate_devices ||
1968 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1969 return false;
1970 }
1971
1972 return true;
1973}
1974
1975static int device_requires_stable_pages(struct dm_target *ti,
1976 struct dm_dev *dev, sector_t start,
1977 sector_t len, void *data)
1978{
1979 struct request_queue *q = bdev_get_queue(dev->bdev);
1980
1981 return blk_queue_stable_writes(q);
1982}
1983
1984int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1985 struct queue_limits *limits)
1986{
1987 bool wc = false, fua = false;
1988 int page_size = PAGE_SIZE;
1989 int r;
1990
1991 /*
1992 * Copy table's limits to the DM device's request_queue
1993 */
1994 q->limits = *limits;
1995
1996 if (dm_table_supports_nowait(t))
1997 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1998 else
1999 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
2000
2001 if (!dm_table_supports_discards(t)) {
2002 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
2003 /* Must also clear discard limits... */
2004 q->limits.max_discard_sectors = 0;
2005 q->limits.max_hw_discard_sectors = 0;
2006 q->limits.discard_granularity = 0;
2007 q->limits.discard_alignment = 0;
2008 q->limits.discard_misaligned = 0;
2009 } else
2010 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
2011
2012 if (dm_table_supports_secure_erase(t))
2013 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
2014
2015 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
2016 wc = true;
2017 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
2018 fua = true;
2019 }
2020 blk_queue_write_cache(q, wc, fua);
2021
2022 if (dm_table_supports_dax(t, device_not_dax_capable, &page_size)) {
2023 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
2024 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable, NULL))
2025 set_dax_synchronous(t->md->dax_dev);
2026 }
2027 else
2028 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
2029
2030 if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
2031 dax_write_cache(t->md->dax_dev, true);
2032
2033 /* Ensure that all underlying devices are non-rotational. */
2034 if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
2035 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2036 else
2037 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2038
2039 if (!dm_table_supports_write_same(t))
2040 q->limits.max_write_same_sectors = 0;
2041 if (!dm_table_supports_write_zeroes(t))
2042 q->limits.max_write_zeroes_sectors = 0;
2043
2044 dm_table_verify_integrity(t);
2045
2046 /*
2047 * Some devices don't use blk_integrity but still want stable pages
2048 * because they do their own checksumming.
2049 * If any underlying device requires stable pages, a table must require
2050 * them as well. Only targets that support iterate_devices are considered:
2051 * don't want error, zero, etc to require stable pages.
2052 */
2053 if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
2054 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2055 else
2056 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2057
2058 /*
2059 * Determine whether or not this queue's I/O timings contribute
2060 * to the entropy pool, Only request-based targets use this.
2061 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2062 * have it set.
2063 */
2064 if (blk_queue_add_random(q) &&
2065 dm_table_any_dev_attr(t, device_is_not_random, NULL))
2066 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2067
2068 /*
2069 * For a zoned target, setup the zones related queue attributes
2070 * and resources necessary for zone append emulation if necessary.
2071 */
2072 if (blk_queue_is_zoned(q)) {
2073 r = dm_set_zones_restrictions(t, q);
2074 if (r)
2075 return r;
2076 }
2077
2078 dm_update_keyslot_manager(q, t);
2079 blk_queue_update_readahead(q);
2080
2081 return 0;
2082}
2083
2084unsigned int dm_table_get_num_targets(struct dm_table *t)
2085{
2086 return t->num_targets;
2087}
2088
2089struct list_head *dm_table_get_devices(struct dm_table *t)
2090{
2091 return &t->devices;
2092}
2093
2094fmode_t dm_table_get_mode(struct dm_table *t)
2095{
2096 return t->mode;
2097}
2098EXPORT_SYMBOL(dm_table_get_mode);
2099
2100enum suspend_mode {
2101 PRESUSPEND,
2102 PRESUSPEND_UNDO,
2103 POSTSUSPEND,
2104};
2105
2106static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2107{
2108 int i = t->num_targets;
2109 struct dm_target *ti = t->targets;
2110
2111 lockdep_assert_held(&t->md->suspend_lock);
2112
2113 while (i--) {
2114 switch (mode) {
2115 case PRESUSPEND:
2116 if (ti->type->presuspend)
2117 ti->type->presuspend(ti);
2118 break;
2119 case PRESUSPEND_UNDO:
2120 if (ti->type->presuspend_undo)
2121 ti->type->presuspend_undo(ti);
2122 break;
2123 case POSTSUSPEND:
2124 if (ti->type->postsuspend)
2125 ti->type->postsuspend(ti);
2126 break;
2127 }
2128 ti++;
2129 }
2130}
2131
2132void dm_table_presuspend_targets(struct dm_table *t)
2133{
2134 if (!t)
2135 return;
2136
2137 suspend_targets(t, PRESUSPEND);
2138}
2139
2140void dm_table_presuspend_undo_targets(struct dm_table *t)
2141{
2142 if (!t)
2143 return;
2144
2145 suspend_targets(t, PRESUSPEND_UNDO);
2146}
2147
2148void dm_table_postsuspend_targets(struct dm_table *t)
2149{
2150 if (!t)
2151 return;
2152
2153 suspend_targets(t, POSTSUSPEND);
2154}
2155
2156int dm_table_resume_targets(struct dm_table *t)
2157{
2158 int i, r = 0;
2159
2160 lockdep_assert_held(&t->md->suspend_lock);
2161
2162 for (i = 0; i < t->num_targets; i++) {
2163 struct dm_target *ti = t->targets + i;
2164
2165 if (!ti->type->preresume)
2166 continue;
2167
2168 r = ti->type->preresume(ti);
2169 if (r) {
2170 DMERR("%s: %s: preresume failed, error = %d",
2171 dm_device_name(t->md), ti->type->name, r);
2172 return r;
2173 }
2174 }
2175
2176 for (i = 0; i < t->num_targets; i++) {
2177 struct dm_target *ti = t->targets + i;
2178
2179 if (ti->type->resume)
2180 ti->type->resume(ti);
2181 }
2182
2183 return 0;
2184}
2185
2186struct mapped_device *dm_table_get_md(struct dm_table *t)
2187{
2188 return t->md;
2189}
2190EXPORT_SYMBOL(dm_table_get_md);
2191
2192const char *dm_table_device_name(struct dm_table *t)
2193{
2194 return dm_device_name(t->md);
2195}
2196EXPORT_SYMBOL_GPL(dm_table_device_name);
2197
2198void dm_table_run_md_queue_async(struct dm_table *t)
2199{
2200 if (!dm_table_request_based(t))
2201 return;
2202
2203 if (t->md->queue)
2204 blk_mq_run_hw_queues(t->md->queue, true);
2205}
2206EXPORT_SYMBOL(dm_table_run_md_queue_async);
2207