2 * Copyright (C) 2001 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
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>
22 #define DM_MSG_PREFIX "table"
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)
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.
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.
38 * When the table is about to be destroyed, we wait for table->holders to
43 struct mapped_device *md;
49 unsigned int counts[MAX_DEPTH]; /* in nodes */
50 sector_t *index[MAX_DEPTH];
52 unsigned int num_targets;
53 unsigned int num_allocated;
55 struct dm_target *targets;
57 struct target_type *immutable_target_type;
58 unsigned integrity_supported:1;
62 * Indicates the rw permissions for the new logical
63 * device. This should be a combination of FMODE_READ
68 /* a list of devices used by this table */
69 struct list_head devices;
71 /* events get handed up using this callback */
72 void (*event_fn)(void *);
75 struct dm_md_mempools *mempools;
77 struct list_head target_callbacks;
81 * Similar to ceiling(log_size(n))
83 static unsigned int int_log(unsigned int n, unsigned int base)
88 n = dm_div_up(n, base);
96 * Calculate the index of the child node of the n'th node k'th key.
98 static inline unsigned int get_child(unsigned int n, unsigned int k)
100 return (n * CHILDREN_PER_NODE) + k;
104 * Return the n'th node of level l from table t.
106 static inline sector_t *get_node(struct dm_table *t,
107 unsigned int l, unsigned int n)
109 return t->index[l] + (n * KEYS_PER_NODE);
113 * Return the highest key that you could lookup from the n'th
114 * node on level l of the btree.
116 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
118 for (; l < t->depth - 1; l++)
119 n = get_child(n, CHILDREN_PER_NODE - 1);
121 if (n >= t->counts[l])
122 return (sector_t) - 1;
124 return get_node(t, l, n)[KEYS_PER_NODE - 1];
128 * Fills in a level of the btree based on the highs of the level
131 static int setup_btree_index(unsigned int l, struct dm_table *t)
136 for (n = 0U; n < t->counts[l]; n++) {
137 node = get_node(t, l, n);
139 for (k = 0U; k < KEYS_PER_NODE; k++)
140 node[k] = high(t, l + 1, get_child(n, k));
146 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
152 * Check that we're not going to overflow.
154 if (nmemb > (ULONG_MAX / elem_size))
157 size = nmemb * elem_size;
158 addr = vzalloc(size);
162 EXPORT_SYMBOL(dm_vcalloc);
165 * highs, and targets are managed as dynamic arrays during a
168 static int alloc_targets(struct dm_table *t, unsigned int num)
171 struct dm_target *n_targets;
172 int n = t->num_targets;
175 * Allocate both the target array and offset array at once.
176 * Append an empty entry to catch sectors beyond the end of
179 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
184 n_targets = (struct dm_target *) (n_highs + num);
187 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
188 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
191 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
194 t->num_allocated = num;
196 t->targets = n_targets;
201 int dm_table_create(struct dm_table **result, fmode_t mode,
202 unsigned num_targets, struct mapped_device *md)
204 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
209 INIT_LIST_HEAD(&t->devices);
210 INIT_LIST_HEAD(&t->target_callbacks);
211 atomic_set(&t->holders, 0);
214 num_targets = KEYS_PER_NODE;
216 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
218 if (alloc_targets(t, num_targets)) {
230 static void free_devices(struct list_head *devices)
232 struct list_head *tmp, *next;
234 list_for_each_safe(tmp, next, devices) {
235 struct dm_dev_internal *dd =
236 list_entry(tmp, struct dm_dev_internal, list);
237 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
243 void dm_table_destroy(struct dm_table *t)
250 while (atomic_read(&t->holders))
254 /* free the indexes */
256 vfree(t->index[t->depth - 2]);
258 /* free the targets */
259 for (i = 0; i < t->num_targets; i++) {
260 struct dm_target *tgt = t->targets + i;
265 dm_put_target_type(tgt->type);
270 /* free the device list */
271 free_devices(&t->devices);
273 dm_free_md_mempools(t->mempools);
278 void dm_table_get(struct dm_table *t)
280 atomic_inc(&t->holders);
282 EXPORT_SYMBOL(dm_table_get);
284 void dm_table_put(struct dm_table *t)
289 smp_mb__before_atomic_dec();
290 atomic_dec(&t->holders);
292 EXPORT_SYMBOL(dm_table_put);
295 * Checks to see if we need to extend highs or targets.
297 static inline int check_space(struct dm_table *t)
299 if (t->num_targets >= t->num_allocated)
300 return alloc_targets(t, t->num_allocated * 2);
306 * See if we've already got a device in the list.
308 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
310 struct dm_dev_internal *dd;
312 list_for_each_entry (dd, l, list)
313 if (dd->dm_dev.bdev->bd_dev == dev)
320 * Open a device so we can use it as a map destination.
322 static int open_dev(struct dm_dev_internal *d, dev_t dev,
323 struct mapped_device *md)
325 static char *_claim_ptr = "I belong to device-mapper";
326 struct block_device *bdev;
330 BUG_ON(d->dm_dev.bdev);
332 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
334 return PTR_ERR(bdev);
336 r = bd_link_disk_holder(bdev, dm_disk(md));
338 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
342 d->dm_dev.bdev = bdev;
347 * Close a device that we've been using.
349 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
354 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
355 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
356 d->dm_dev.bdev = NULL;
360 * If possible, this checks an area of a destination device is invalid.
362 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
363 sector_t start, sector_t len, void *data)
365 struct request_queue *q;
366 struct queue_limits *limits = data;
367 struct block_device *bdev = dev->bdev;
369 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
370 unsigned short logical_block_size_sectors =
371 limits->logical_block_size >> SECTOR_SHIFT;
372 char b[BDEVNAME_SIZE];
375 * Some devices exist without request functions,
376 * such as loop devices not yet bound to backing files.
377 * Forbid the use of such devices.
379 q = bdev_get_queue(bdev);
380 if (!q || !q->make_request_fn) {
381 DMWARN("%s: %s is not yet initialised: "
382 "start=%llu, len=%llu, dev_size=%llu",
383 dm_device_name(ti->table->md), bdevname(bdev, b),
384 (unsigned long long)start,
385 (unsigned long long)len,
386 (unsigned long long)dev_size);
393 if ((start >= dev_size) || (start + len > dev_size)) {
394 DMWARN("%s: %s too small for target: "
395 "start=%llu, len=%llu, dev_size=%llu",
396 dm_device_name(ti->table->md), bdevname(bdev, b),
397 (unsigned long long)start,
398 (unsigned long long)len,
399 (unsigned long long)dev_size);
403 if (logical_block_size_sectors <= 1)
406 if (start & (logical_block_size_sectors - 1)) {
407 DMWARN("%s: start=%llu not aligned to h/w "
408 "logical block size %u of %s",
409 dm_device_name(ti->table->md),
410 (unsigned long long)start,
411 limits->logical_block_size, bdevname(bdev, b));
415 if (len & (logical_block_size_sectors - 1)) {
416 DMWARN("%s: len=%llu not aligned to h/w "
417 "logical block size %u of %s",
418 dm_device_name(ti->table->md),
419 (unsigned long long)len,
420 limits->logical_block_size, bdevname(bdev, b));
428 * This upgrades the mode on an already open dm_dev, being
429 * careful to leave things as they were if we fail to reopen the
430 * device and not to touch the existing bdev field in case
431 * it is accessed concurrently inside dm_table_any_congested().
433 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
434 struct mapped_device *md)
437 struct dm_dev_internal dd_new, dd_old;
439 dd_new = dd_old = *dd;
441 dd_new.dm_dev.mode = new_mode;
442 dd_new.dm_dev.bdev = NULL;
444 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
446 dd_new.dm_dev.mode &= ~FMODE_WRITE;
447 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
452 dd->dm_dev.mode = new_mode;
453 close_dev(&dd_old, md);
459 * Add a device to the list, or just increment the usage count if
460 * it's already present.
462 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
463 struct dm_dev **result)
466 dev_t uninitialized_var(dev);
467 struct dm_dev_internal *dd;
468 unsigned int major, minor;
469 struct dm_table *t = ti->table;
474 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
475 /* Extract the major/minor numbers */
476 dev = MKDEV(major, minor);
477 if (MAJOR(dev) != major || MINOR(dev) != minor)
480 /* convert the path to a device */
481 struct block_device *bdev = lookup_bdev(path);
484 return PTR_ERR(bdev);
489 dd = find_device(&t->devices, dev);
491 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
495 dd->dm_dev.mode = mode;
496 dd->dm_dev.bdev = NULL;
498 r = open_dev(dd, dev, t->md);
500 dd->dm_dev.mode &= ~FMODE_WRITE;
501 r = open_dev(dd, dev, t->md);
508 if (dd->dm_dev.mode != mode)
509 t->mode = dd->dm_dev.mode;
511 format_dev_t(dd->dm_dev.name, dev);
513 atomic_set(&dd->count, 0);
514 list_add(&dd->list, &t->devices);
516 } else if (dd->dm_dev.mode != mode) {
517 r = upgrade_mode(dd, mode, t->md);
521 atomic_inc(&dd->count);
523 *result = &dd->dm_dev;
526 EXPORT_SYMBOL(dm_get_device);
528 int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
529 sector_t start, sector_t len, void *data)
531 struct queue_limits *limits = data;
532 struct block_device *bdev = dev->bdev;
533 struct request_queue *q = bdev_get_queue(bdev);
534 char b[BDEVNAME_SIZE];
537 DMWARN("%s: Cannot set limits for nonexistent device %s",
538 dm_device_name(ti->table->md), bdevname(bdev, b));
542 if (bdev_stack_limits(limits, bdev, start) < 0)
543 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
544 "physical_block_size=%u, logical_block_size=%u, "
545 "alignment_offset=%u, start=%llu",
546 dm_device_name(ti->table->md), bdevname(bdev, b),
547 q->limits.physical_block_size,
548 q->limits.logical_block_size,
549 q->limits.alignment_offset,
550 (unsigned long long) start << SECTOR_SHIFT);
553 * Check if merge fn is supported.
554 * If not we'll force DM to use PAGE_SIZE or
555 * smaller I/O, just to be safe.
557 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
558 blk_limits_max_hw_sectors(limits,
559 (unsigned int) (PAGE_SIZE >> 9));
562 EXPORT_SYMBOL_GPL(dm_set_device_limits);
565 * Decrement a device's use count and remove it if necessary.
567 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
569 struct dm_dev_internal *dd;
574 dd = container_of(d, struct dm_dev_internal, dm_dev);
575 if (atomic_dec_and_test(&dd->count)) {
576 close_dev(dd, ti->table->md);
581 EXPORT_SYMBOL(dm_put_device);
584 * Checks to see if the target joins onto the end of the table.
586 static int adjoin(struct dm_table *table, struct dm_target *ti)
588 struct dm_target *prev;
590 if (!table->num_targets)
593 prev = &table->targets[table->num_targets - 1];
594 return (ti->begin == (prev->begin + prev->len));
598 * Used to dynamically allocate the arg array.
600 static char **realloc_argv(unsigned *array_size, char **old_argv)
605 new_size = *array_size ? *array_size * 2 : 64;
606 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
608 memcpy(argv, old_argv, *array_size * sizeof(*argv));
609 *array_size = new_size;
617 * Destructively splits up the argument list to pass to ctr.
619 int dm_split_args(int *argc, char ***argvp, char *input)
621 char *start, *end = input, *out, **argv = NULL;
622 unsigned array_size = 0;
631 argv = realloc_argv(&array_size, argv);
636 /* Skip whitespace */
637 start = skip_spaces(end);
640 break; /* success, we hit the end */
642 /* 'out' is used to remove any back-quotes */
645 /* Everything apart from '\0' can be quoted */
646 if (*end == '\\' && *(end + 1)) {
653 break; /* end of token */
658 /* have we already filled the array ? */
659 if ((*argc + 1) > array_size) {
660 argv = realloc_argv(&array_size, argv);
665 /* we know this is whitespace */
669 /* terminate the string and put it in the array */
680 * Impose necessary and sufficient conditions on a devices's table such
681 * that any incoming bio which respects its logical_block_size can be
682 * processed successfully. If it falls across the boundary between
683 * two or more targets, the size of each piece it gets split into must
684 * be compatible with the logical_block_size of the target processing it.
686 static int validate_hardware_logical_block_alignment(struct dm_table *table,
687 struct queue_limits *limits)
690 * This function uses arithmetic modulo the logical_block_size
691 * (in units of 512-byte sectors).
693 unsigned short device_logical_block_size_sects =
694 limits->logical_block_size >> SECTOR_SHIFT;
697 * Offset of the start of the next table entry, mod logical_block_size.
699 unsigned short next_target_start = 0;
702 * Given an aligned bio that extends beyond the end of a
703 * target, how many sectors must the next target handle?
705 unsigned short remaining = 0;
707 struct dm_target *uninitialized_var(ti);
708 struct queue_limits ti_limits;
712 * Check each entry in the table in turn.
714 while (i < dm_table_get_num_targets(table)) {
715 ti = dm_table_get_target(table, i++);
717 blk_set_stacking_limits(&ti_limits);
719 /* combine all target devices' limits */
720 if (ti->type->iterate_devices)
721 ti->type->iterate_devices(ti, dm_set_device_limits,
725 * If the remaining sectors fall entirely within this
726 * table entry are they compatible with its logical_block_size?
728 if (remaining < ti->len &&
729 remaining & ((ti_limits.logical_block_size >>
734 (unsigned short) ((next_target_start + ti->len) &
735 (device_logical_block_size_sects - 1));
736 remaining = next_target_start ?
737 device_logical_block_size_sects - next_target_start : 0;
741 DMWARN("%s: table line %u (start sect %llu len %llu) "
742 "not aligned to h/w logical block size %u",
743 dm_device_name(table->md), i,
744 (unsigned long long) ti->begin,
745 (unsigned long long) ti->len,
746 limits->logical_block_size);
753 int dm_table_add_target(struct dm_table *t, const char *type,
754 sector_t start, sector_t len, char *params)
756 int r = -EINVAL, argc;
758 struct dm_target *tgt;
761 DMERR("%s: target type %s must appear alone in table",
762 dm_device_name(t->md), t->targets->type->name);
766 if ((r = check_space(t)))
769 tgt = t->targets + t->num_targets;
770 memset(tgt, 0, sizeof(*tgt));
773 DMERR("%s: zero-length target", dm_device_name(t->md));
777 tgt->type = dm_get_target_type(type);
779 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
784 if (dm_target_needs_singleton(tgt->type)) {
785 if (t->num_targets) {
786 DMERR("%s: target type %s must appear alone in table",
787 dm_device_name(t->md), type);
793 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
794 DMERR("%s: target type %s may not be included in read-only tables",
795 dm_device_name(t->md), type);
799 if (t->immutable_target_type) {
800 if (t->immutable_target_type != tgt->type) {
801 DMERR("%s: immutable target type %s cannot be mixed with other target types",
802 dm_device_name(t->md), t->immutable_target_type->name);
805 } else if (dm_target_is_immutable(tgt->type)) {
806 if (t->num_targets) {
807 DMERR("%s: immutable target type %s cannot be mixed with other target types",
808 dm_device_name(t->md), tgt->type->name);
811 t->immutable_target_type = tgt->type;
817 tgt->error = "Unknown error";
820 * Does this target adjoin the previous one ?
822 if (!adjoin(t, tgt)) {
823 tgt->error = "Gap in table";
828 r = dm_split_args(&argc, &argv, params);
830 tgt->error = "couldn't split parameters (insufficient memory)";
834 r = tgt->type->ctr(tgt, argc, argv);
839 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
841 if (!tgt->num_discard_requests && tgt->discards_supported)
842 DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
843 dm_device_name(t->md), type);
848 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
849 dm_put_target_type(tgt->type);
854 * Target argument parsing helpers.
856 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
857 unsigned *value, char **error, unsigned grouped)
859 const char *arg_str = dm_shift_arg(arg_set);
863 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
864 (*value < arg->min) ||
865 (*value > arg->max) ||
866 (grouped && arg_set->argc < *value)) {
874 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
875 unsigned *value, char **error)
877 return validate_next_arg(arg, arg_set, value, error, 0);
879 EXPORT_SYMBOL(dm_read_arg);
881 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
882 unsigned *value, char **error)
884 return validate_next_arg(arg, arg_set, value, error, 1);
886 EXPORT_SYMBOL(dm_read_arg_group);
888 const char *dm_shift_arg(struct dm_arg_set *as)
901 EXPORT_SYMBOL(dm_shift_arg);
903 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
905 BUG_ON(as->argc < num_args);
906 as->argc -= num_args;
907 as->argv += num_args;
909 EXPORT_SYMBOL(dm_consume_args);
911 static int dm_table_set_type(struct dm_table *t)
914 unsigned bio_based = 0, request_based = 0;
915 struct dm_target *tgt;
916 struct dm_dev_internal *dd;
917 struct list_head *devices;
919 for (i = 0; i < t->num_targets; i++) {
920 tgt = t->targets + i;
921 if (dm_target_request_based(tgt))
926 if (bio_based && request_based) {
927 DMWARN("Inconsistent table: different target types"
928 " can't be mixed up");
934 /* We must use this table as bio-based */
935 t->type = DM_TYPE_BIO_BASED;
939 BUG_ON(!request_based); /* No targets in this table */
941 /* Non-request-stackable devices can't be used for request-based dm */
942 devices = dm_table_get_devices(t);
943 list_for_each_entry(dd, devices, list) {
944 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
945 DMWARN("table load rejected: including"
946 " non-request-stackable devices");
952 * Request-based dm supports only tables that have a single target now.
953 * To support multiple targets, request splitting support is needed,
954 * and that needs lots of changes in the block-layer.
955 * (e.g. request completion process for partial completion.)
957 if (t->num_targets > 1) {
958 DMWARN("Request-based dm doesn't support multiple targets yet");
962 t->type = DM_TYPE_REQUEST_BASED;
967 unsigned dm_table_get_type(struct dm_table *t)
972 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
974 return t->immutable_target_type;
977 bool dm_table_request_based(struct dm_table *t)
979 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
982 int dm_table_alloc_md_mempools(struct dm_table *t)
984 unsigned type = dm_table_get_type(t);
986 if (unlikely(type == DM_TYPE_NONE)) {
987 DMWARN("no table type is set, can't allocate mempools");
991 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
998 void dm_table_free_md_mempools(struct dm_table *t)
1000 dm_free_md_mempools(t->mempools);
1004 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1009 static int setup_indexes(struct dm_table *t)
1012 unsigned int total = 0;
1015 /* allocate the space for *all* the indexes */
1016 for (i = t->depth - 2; i >= 0; i--) {
1017 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1018 total += t->counts[i];
1021 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1025 /* set up internal nodes, bottom-up */
1026 for (i = t->depth - 2; i >= 0; i--) {
1027 t->index[i] = indexes;
1028 indexes += (KEYS_PER_NODE * t->counts[i]);
1029 setup_btree_index(i, t);
1036 * Builds the btree to index the map.
1038 static int dm_table_build_index(struct dm_table *t)
1041 unsigned int leaf_nodes;
1043 /* how many indexes will the btree have ? */
1044 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1045 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1047 /* leaf layer has already been set up */
1048 t->counts[t->depth - 1] = leaf_nodes;
1049 t->index[t->depth - 1] = t->highs;
1052 r = setup_indexes(t);
1058 * Get a disk whose integrity profile reflects the table's profile.
1059 * If %match_all is true, all devices' profiles must match.
1060 * If %match_all is false, all devices must at least have an
1061 * allocated integrity profile; but uninitialized is ok.
1062 * Returns NULL if integrity support was inconsistent or unavailable.
1064 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1067 struct list_head *devices = dm_table_get_devices(t);
1068 struct dm_dev_internal *dd = NULL;
1069 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1071 list_for_each_entry(dd, devices, list) {
1072 template_disk = dd->dm_dev.bdev->bd_disk;
1073 if (!blk_get_integrity(template_disk))
1075 if (!match_all && !blk_integrity_is_initialized(template_disk))
1076 continue; /* skip uninitialized profiles */
1077 else if (prev_disk &&
1078 blk_integrity_compare(prev_disk, template_disk) < 0)
1080 prev_disk = template_disk;
1083 return template_disk;
1087 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1088 dm_device_name(t->md),
1089 prev_disk->disk_name,
1090 template_disk->disk_name);
1095 * Register the mapped device for blk_integrity support if
1096 * the underlying devices have an integrity profile. But all devices
1097 * may not have matching profiles (checking all devices isn't reliable
1098 * during table load because this table may use other DM device(s) which
1099 * must be resumed before they will have an initialized integity profile).
1100 * Stacked DM devices force a 2 stage integrity profile validation:
1101 * 1 - during load, validate all initialized integrity profiles match
1102 * 2 - during resume, validate all integrity profiles match
1104 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1106 struct gendisk *template_disk = NULL;
1108 template_disk = dm_table_get_integrity_disk(t, false);
1112 if (!blk_integrity_is_initialized(dm_disk(md))) {
1113 t->integrity_supported = 1;
1114 return blk_integrity_register(dm_disk(md), NULL);
1118 * If DM device already has an initalized integrity
1119 * profile the new profile should not conflict.
1121 if (blk_integrity_is_initialized(template_disk) &&
1122 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1123 DMWARN("%s: conflict with existing integrity profile: "
1124 "%s profile mismatch",
1125 dm_device_name(t->md),
1126 template_disk->disk_name);
1130 /* Preserve existing initialized integrity profile */
1131 t->integrity_supported = 1;
1136 * Prepares the table for use by building the indices,
1137 * setting the type, and allocating mempools.
1139 int dm_table_complete(struct dm_table *t)
1143 r = dm_table_set_type(t);
1145 DMERR("unable to set table type");
1149 r = dm_table_build_index(t);
1151 DMERR("unable to build btrees");
1155 r = dm_table_prealloc_integrity(t, t->md);
1157 DMERR("could not register integrity profile.");
1161 r = dm_table_alloc_md_mempools(t);
1163 DMERR("unable to allocate mempools");
1168 static DEFINE_MUTEX(_event_lock);
1169 void dm_table_event_callback(struct dm_table *t,
1170 void (*fn)(void *), void *context)
1172 mutex_lock(&_event_lock);
1174 t->event_context = context;
1175 mutex_unlock(&_event_lock);
1178 void dm_table_event(struct dm_table *t)
1181 * You can no longer call dm_table_event() from interrupt
1182 * context, use a bottom half instead.
1184 BUG_ON(in_interrupt());
1186 mutex_lock(&_event_lock);
1188 t->event_fn(t->event_context);
1189 mutex_unlock(&_event_lock);
1191 EXPORT_SYMBOL(dm_table_event);
1193 sector_t dm_table_get_size(struct dm_table *t)
1195 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1197 EXPORT_SYMBOL(dm_table_get_size);
1199 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1201 if (index >= t->num_targets)
1204 return t->targets + index;
1208 * Search the btree for the correct target.
1210 * Caller should check returned pointer with dm_target_is_valid()
1211 * to trap I/O beyond end of device.
1213 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1215 unsigned int l, n = 0, k = 0;
1218 for (l = 0; l < t->depth; l++) {
1219 n = get_child(n, k);
1220 node = get_node(t, l, n);
1222 for (k = 0; k < KEYS_PER_NODE; k++)
1223 if (node[k] >= sector)
1227 return &t->targets[(KEYS_PER_NODE * n) + k];
1231 * Establish the new table's queue_limits and validate them.
1233 int dm_calculate_queue_limits(struct dm_table *table,
1234 struct queue_limits *limits)
1236 struct dm_target *uninitialized_var(ti);
1237 struct queue_limits ti_limits;
1240 blk_set_stacking_limits(limits);
1242 while (i < dm_table_get_num_targets(table)) {
1243 blk_set_stacking_limits(&ti_limits);
1245 ti = dm_table_get_target(table, i++);
1247 if (!ti->type->iterate_devices)
1248 goto combine_limits;
1251 * Combine queue limits of all the devices this target uses.
1253 ti->type->iterate_devices(ti, dm_set_device_limits,
1256 /* Set I/O hints portion of queue limits */
1257 if (ti->type->io_hints)
1258 ti->type->io_hints(ti, &ti_limits);
1261 * Check each device area is consistent with the target's
1262 * overall queue limits.
1264 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1270 * Merge this target's queue limits into the overall limits
1273 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1274 DMWARN("%s: adding target device "
1275 "(start sect %llu len %llu) "
1276 "caused an alignment inconsistency",
1277 dm_device_name(table->md),
1278 (unsigned long long) ti->begin,
1279 (unsigned long long) ti->len);
1282 return validate_hardware_logical_block_alignment(table, limits);
1286 * Set the integrity profile for this device if all devices used have
1287 * matching profiles. We're quite deep in the resume path but still
1288 * don't know if all devices (particularly DM devices this device
1289 * may be stacked on) have matching profiles. Even if the profiles
1290 * don't match we have no way to fail (to resume) at this point.
1292 static void dm_table_set_integrity(struct dm_table *t)
1294 struct gendisk *template_disk = NULL;
1296 if (!blk_get_integrity(dm_disk(t->md)))
1299 template_disk = dm_table_get_integrity_disk(t, true);
1301 blk_integrity_register(dm_disk(t->md),
1302 blk_get_integrity(template_disk));
1303 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1304 DMWARN("%s: device no longer has a valid integrity profile",
1305 dm_device_name(t->md));
1307 DMWARN("%s: unable to establish an integrity profile",
1308 dm_device_name(t->md));
1311 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1312 sector_t start, sector_t len, void *data)
1314 unsigned flush = (*(unsigned *)data);
1315 struct request_queue *q = bdev_get_queue(dev->bdev);
1317 return q && (q->flush_flags & flush);
1320 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1322 struct dm_target *ti;
1326 * Require at least one underlying device to support flushes.
1327 * t->devices includes internal dm devices such as mirror logs
1328 * so we need to use iterate_devices here, which targets
1329 * supporting flushes must provide.
1331 while (i < dm_table_get_num_targets(t)) {
1332 ti = dm_table_get_target(t, i++);
1334 if (!ti->num_flush_requests)
1337 if (ti->type->iterate_devices &&
1338 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1345 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1347 struct dm_target *ti;
1350 /* Ensure that all targets supports discard_zeroes_data. */
1351 while (i < dm_table_get_num_targets(t)) {
1352 ti = dm_table_get_target(t, i++);
1354 if (ti->discard_zeroes_data_unsupported)
1361 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1362 sector_t start, sector_t len, void *data)
1364 struct request_queue *q = bdev_get_queue(dev->bdev);
1366 return q && blk_queue_nonrot(q);
1369 static bool dm_table_is_nonrot(struct dm_table *t)
1371 struct dm_target *ti;
1374 /* Ensure that all underlying device are non-rotational. */
1375 while (i < dm_table_get_num_targets(t)) {
1376 ti = dm_table_get_target(t, i++);
1378 if (!ti->type->iterate_devices ||
1379 !ti->type->iterate_devices(ti, device_is_nonrot, NULL))
1386 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1387 struct queue_limits *limits)
1392 * Copy table's limits to the DM device's request_queue
1394 q->limits = *limits;
1396 if (!dm_table_supports_discards(t))
1397 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1399 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1401 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1403 if (dm_table_supports_flush(t, REQ_FUA))
1406 blk_queue_flush(q, flush);
1408 if (!dm_table_discard_zeroes_data(t))
1409 q->limits.discard_zeroes_data = 0;
1411 if (dm_table_is_nonrot(t))
1412 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1414 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1416 dm_table_set_integrity(t);
1419 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1420 * visible to other CPUs because, once the flag is set, incoming bios
1421 * are processed by request-based dm, which refers to the queue
1423 * Until the flag set, bios are passed to bio-based dm and queued to
1424 * md->deferred where queue settings are not needed yet.
1425 * Those bios are passed to request-based dm at the resume time.
1428 if (dm_table_request_based(t))
1429 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1432 unsigned int dm_table_get_num_targets(struct dm_table *t)
1434 return t->num_targets;
1437 struct list_head *dm_table_get_devices(struct dm_table *t)
1442 fmode_t dm_table_get_mode(struct dm_table *t)
1446 EXPORT_SYMBOL(dm_table_get_mode);
1448 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1450 int i = t->num_targets;
1451 struct dm_target *ti = t->targets;
1455 if (ti->type->postsuspend)
1456 ti->type->postsuspend(ti);
1457 } else if (ti->type->presuspend)
1458 ti->type->presuspend(ti);
1464 void dm_table_presuspend_targets(struct dm_table *t)
1469 suspend_targets(t, 0);
1472 void dm_table_postsuspend_targets(struct dm_table *t)
1477 suspend_targets(t, 1);
1480 int dm_table_resume_targets(struct dm_table *t)
1484 for (i = 0; i < t->num_targets; i++) {
1485 struct dm_target *ti = t->targets + i;
1487 if (!ti->type->preresume)
1490 r = ti->type->preresume(ti);
1495 for (i = 0; i < t->num_targets; i++) {
1496 struct dm_target *ti = t->targets + i;
1498 if (ti->type->resume)
1499 ti->type->resume(ti);
1505 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1507 list_add(&cb->list, &t->target_callbacks);
1509 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1511 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1513 struct dm_dev_internal *dd;
1514 struct list_head *devices = dm_table_get_devices(t);
1515 struct dm_target_callbacks *cb;
1518 list_for_each_entry(dd, devices, list) {
1519 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1520 char b[BDEVNAME_SIZE];
1523 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1525 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1526 dm_device_name(t->md),
1527 bdevname(dd->dm_dev.bdev, b));
1530 list_for_each_entry(cb, &t->target_callbacks, list)
1531 if (cb->congested_fn)
1532 r |= cb->congested_fn(cb, bdi_bits);
1537 int dm_table_any_busy_target(struct dm_table *t)
1540 struct dm_target *ti;
1542 for (i = 0; i < t->num_targets; i++) {
1543 ti = t->targets + i;
1544 if (ti->type->busy && ti->type->busy(ti))
1551 struct mapped_device *dm_table_get_md(struct dm_table *t)
1555 EXPORT_SYMBOL(dm_table_get_md);
1557 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1558 sector_t start, sector_t len, void *data)
1560 struct request_queue *q = bdev_get_queue(dev->bdev);
1562 return q && blk_queue_discard(q);
1565 bool dm_table_supports_discards(struct dm_table *t)
1567 struct dm_target *ti;
1571 * Unless any target used by the table set discards_supported,
1572 * require at least one underlying device to support discards.
1573 * t->devices includes internal dm devices such as mirror logs
1574 * so we need to use iterate_devices here, which targets
1575 * supporting discard selectively must provide.
1577 while (i < dm_table_get_num_targets(t)) {
1578 ti = dm_table_get_target(t, i++);
1580 if (!ti->num_discard_requests)
1583 if (ti->discards_supported)
1586 if (ti->type->iterate_devices &&
1587 ti->type->iterate_devices(ti, device_discard_capable, NULL))