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>
21 #include <linux/blk-mq.h>
22 #include <linux/mount.h>
24 #define DM_MSG_PREFIX "table"
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)
32 struct mapped_device *md;
37 unsigned int counts[MAX_DEPTH]; /* in nodes */
38 sector_t *index[MAX_DEPTH];
40 unsigned int num_targets;
41 unsigned int num_allocated;
43 struct dm_target *targets;
45 struct target_type *immutable_target_type;
47 bool integrity_supported:1;
52 * Indicates the rw permissions for the new logical
53 * device. This should be a combination of FMODE_READ
58 /* a list of devices used by this table */
59 struct list_head devices;
61 /* events get handed up using this callback */
62 void (*event_fn)(void *);
65 struct dm_md_mempools *mempools;
67 struct list_head target_callbacks;
71 * Similar to ceiling(log_size(n))
73 static unsigned int int_log(unsigned int n, unsigned int base)
78 n = dm_div_up(n, base);
86 * Calculate the index of the child node of the n'th node k'th key.
88 static inline unsigned int get_child(unsigned int n, unsigned int k)
90 return (n * CHILDREN_PER_NODE) + k;
94 * Return the n'th node of level l from table t.
96 static inline sector_t *get_node(struct dm_table *t,
97 unsigned int l, unsigned int n)
99 return t->index[l] + (n * KEYS_PER_NODE);
103 * Return the highest key that you could lookup from the n'th
104 * node on level l of the btree.
106 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
108 for (; l < t->depth - 1; l++)
109 n = get_child(n, CHILDREN_PER_NODE - 1);
111 if (n >= t->counts[l])
112 return (sector_t) - 1;
114 return get_node(t, l, n)[KEYS_PER_NODE - 1];
118 * Fills in a level of the btree based on the highs of the level
121 static int setup_btree_index(unsigned int l, struct dm_table *t)
126 for (n = 0U; n < t->counts[l]; n++) {
127 node = get_node(t, l, n);
129 for (k = 0U; k < KEYS_PER_NODE; k++)
130 node[k] = high(t, l + 1, get_child(n, k));
136 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
142 * Check that we're not going to overflow.
144 if (nmemb > (ULONG_MAX / elem_size))
147 size = nmemb * elem_size;
148 addr = vzalloc(size);
152 EXPORT_SYMBOL(dm_vcalloc);
155 * highs, and targets are managed as dynamic arrays during a
158 static int alloc_targets(struct dm_table *t, unsigned int num)
161 struct dm_target *n_targets;
164 * Allocate both the target array and offset array at once.
165 * Append an empty entry to catch sectors beyond the end of
168 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
173 n_targets = (struct dm_target *) (n_highs + num);
175 memset(n_highs, -1, sizeof(*n_highs) * num);
178 t->num_allocated = num;
180 t->targets = n_targets;
185 int dm_table_create(struct dm_table **result, fmode_t mode,
186 unsigned num_targets, struct mapped_device *md)
188 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
193 INIT_LIST_HEAD(&t->devices);
194 INIT_LIST_HEAD(&t->target_callbacks);
197 num_targets = KEYS_PER_NODE;
199 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
206 if (alloc_targets(t, num_targets)) {
211 t->type = DM_TYPE_NONE;
218 static void free_devices(struct list_head *devices, struct mapped_device *md)
220 struct list_head *tmp, *next;
222 list_for_each_safe(tmp, next, devices) {
223 struct dm_dev_internal *dd =
224 list_entry(tmp, struct dm_dev_internal, list);
225 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
226 dm_device_name(md), dd->dm_dev->name);
227 dm_put_table_device(md, dd->dm_dev);
232 void dm_table_destroy(struct dm_table *t)
239 /* free the indexes */
241 vfree(t->index[t->depth - 2]);
243 /* free the targets */
244 for (i = 0; i < t->num_targets; i++) {
245 struct dm_target *tgt = t->targets + i;
250 dm_put_target_type(tgt->type);
255 /* free the device list */
256 free_devices(&t->devices, t->md);
258 dm_free_md_mempools(t->mempools);
264 * See if we've already got a device in the list.
266 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
268 struct dm_dev_internal *dd;
270 list_for_each_entry (dd, l, list)
271 if (dd->dm_dev->bdev->bd_dev == dev)
278 * If possible, this checks an area of a destination device is invalid.
280 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
281 sector_t start, sector_t len, void *data)
283 struct request_queue *q;
284 struct queue_limits *limits = data;
285 struct block_device *bdev = dev->bdev;
287 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
288 unsigned short logical_block_size_sectors =
289 limits->logical_block_size >> SECTOR_SHIFT;
290 char b[BDEVNAME_SIZE];
293 * Some devices exist without request functions,
294 * such as loop devices not yet bound to backing files.
295 * Forbid the use of such devices.
297 q = bdev_get_queue(bdev);
298 if (!q || !q->make_request_fn) {
299 DMWARN("%s: %s is not yet initialised: "
300 "start=%llu, len=%llu, dev_size=%llu",
301 dm_device_name(ti->table->md), bdevname(bdev, b),
302 (unsigned long long)start,
303 (unsigned long long)len,
304 (unsigned long long)dev_size);
311 if ((start >= dev_size) || (start + len > dev_size)) {
312 DMWARN("%s: %s too small for target: "
313 "start=%llu, len=%llu, dev_size=%llu",
314 dm_device_name(ti->table->md), bdevname(bdev, b),
315 (unsigned long long)start,
316 (unsigned long long)len,
317 (unsigned long long)dev_size);
321 if (logical_block_size_sectors <= 1)
324 if (start & (logical_block_size_sectors - 1)) {
325 DMWARN("%s: start=%llu not aligned to h/w "
326 "logical block size %u of %s",
327 dm_device_name(ti->table->md),
328 (unsigned long long)start,
329 limits->logical_block_size, bdevname(bdev, b));
333 if (len & (logical_block_size_sectors - 1)) {
334 DMWARN("%s: len=%llu not aligned to h/w "
335 "logical block size %u of %s",
336 dm_device_name(ti->table->md),
337 (unsigned long long)len,
338 limits->logical_block_size, bdevname(bdev, b));
346 * This upgrades the mode on an already open dm_dev, being
347 * careful to leave things as they were if we fail to reopen the
348 * device and not to touch the existing bdev field in case
349 * it is accessed concurrently inside dm_table_any_congested().
351 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
352 struct mapped_device *md)
355 struct dm_dev *old_dev, *new_dev;
357 old_dev = dd->dm_dev;
359 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
360 dd->dm_dev->mode | new_mode, &new_dev);
364 dd->dm_dev = new_dev;
365 dm_put_table_device(md, old_dev);
371 * Convert the path to a device
373 dev_t dm_get_dev_t(const char *path)
375 dev_t uninitialized_var(dev);
376 struct block_device *bdev;
378 bdev = lookup_bdev(path);
380 dev = name_to_dev_t(path);
388 EXPORT_SYMBOL_GPL(dm_get_dev_t);
391 * Add a device to the list, or just increment the usage count if
392 * it's already present.
394 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
395 struct dm_dev **result)
399 struct dm_dev_internal *dd;
400 struct dm_table *t = ti->table;
404 dev = dm_get_dev_t(path);
408 dd = find_device(&t->devices, dev);
410 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
414 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
419 atomic_set(&dd->count, 0);
420 list_add(&dd->list, &t->devices);
422 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
423 r = upgrade_mode(dd, mode, t->md);
427 atomic_inc(&dd->count);
429 *result = dd->dm_dev;
432 EXPORT_SYMBOL(dm_get_device);
434 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
435 sector_t start, sector_t len, void *data)
437 struct queue_limits *limits = data;
438 struct block_device *bdev = dev->bdev;
439 struct request_queue *q = bdev_get_queue(bdev);
440 char b[BDEVNAME_SIZE];
443 DMWARN("%s: Cannot set limits for nonexistent device %s",
444 dm_device_name(ti->table->md), bdevname(bdev, b));
448 if (bdev_stack_limits(limits, bdev, start) < 0)
449 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
450 "physical_block_size=%u, logical_block_size=%u, "
451 "alignment_offset=%u, start=%llu",
452 dm_device_name(ti->table->md), bdevname(bdev, b),
453 q->limits.physical_block_size,
454 q->limits.logical_block_size,
455 q->limits.alignment_offset,
456 (unsigned long long) start << SECTOR_SHIFT);
462 * Decrement a device's use count and remove it if necessary.
464 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
467 struct list_head *devices = &ti->table->devices;
468 struct dm_dev_internal *dd;
470 list_for_each_entry(dd, devices, list) {
471 if (dd->dm_dev == d) {
477 DMWARN("%s: device %s not in table devices list",
478 dm_device_name(ti->table->md), d->name);
481 if (atomic_dec_and_test(&dd->count)) {
482 dm_put_table_device(ti->table->md, d);
487 EXPORT_SYMBOL(dm_put_device);
490 * Checks to see if the target joins onto the end of the table.
492 static int adjoin(struct dm_table *table, struct dm_target *ti)
494 struct dm_target *prev;
496 if (!table->num_targets)
499 prev = &table->targets[table->num_targets - 1];
500 return (ti->begin == (prev->begin + prev->len));
504 * Used to dynamically allocate the arg array.
506 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
507 * process messages even if some device is suspended. These messages have a
508 * small fixed number of arguments.
510 * On the other hand, dm-switch needs to process bulk data using messages and
511 * excessive use of GFP_NOIO could cause trouble.
513 static char **realloc_argv(unsigned *array_size, char **old_argv)
520 new_size = *array_size * 2;
526 argv = kmalloc(new_size * sizeof(*argv), gfp);
528 memcpy(argv, old_argv, *array_size * sizeof(*argv));
529 *array_size = new_size;
537 * Destructively splits up the argument list to pass to ctr.
539 int dm_split_args(int *argc, char ***argvp, char *input)
541 char *start, *end = input, *out, **argv = NULL;
542 unsigned array_size = 0;
551 argv = realloc_argv(&array_size, argv);
556 /* Skip whitespace */
557 start = skip_spaces(end);
560 break; /* success, we hit the end */
562 /* 'out' is used to remove any back-quotes */
565 /* Everything apart from '\0' can be quoted */
566 if (*end == '\\' && *(end + 1)) {
573 break; /* end of token */
578 /* have we already filled the array ? */
579 if ((*argc + 1) > array_size) {
580 argv = realloc_argv(&array_size, argv);
585 /* we know this is whitespace */
589 /* terminate the string and put it in the array */
600 * Impose necessary and sufficient conditions on a devices's table such
601 * that any incoming bio which respects its logical_block_size can be
602 * processed successfully. If it falls across the boundary between
603 * two or more targets, the size of each piece it gets split into must
604 * be compatible with the logical_block_size of the target processing it.
606 static int validate_hardware_logical_block_alignment(struct dm_table *table,
607 struct queue_limits *limits)
610 * This function uses arithmetic modulo the logical_block_size
611 * (in units of 512-byte sectors).
613 unsigned short device_logical_block_size_sects =
614 limits->logical_block_size >> SECTOR_SHIFT;
617 * Offset of the start of the next table entry, mod logical_block_size.
619 unsigned short next_target_start = 0;
622 * Given an aligned bio that extends beyond the end of a
623 * target, how many sectors must the next target handle?
625 unsigned short remaining = 0;
627 struct dm_target *uninitialized_var(ti);
628 struct queue_limits ti_limits;
632 * Check each entry in the table in turn.
634 while (i < dm_table_get_num_targets(table)) {
635 ti = dm_table_get_target(table, i++);
637 blk_set_stacking_limits(&ti_limits);
639 /* combine all target devices' limits */
640 if (ti->type->iterate_devices)
641 ti->type->iterate_devices(ti, dm_set_device_limits,
645 * If the remaining sectors fall entirely within this
646 * table entry are they compatible with its logical_block_size?
648 if (remaining < ti->len &&
649 remaining & ((ti_limits.logical_block_size >>
654 (unsigned short) ((next_target_start + ti->len) &
655 (device_logical_block_size_sects - 1));
656 remaining = next_target_start ?
657 device_logical_block_size_sects - next_target_start : 0;
661 DMWARN("%s: table line %u (start sect %llu len %llu) "
662 "not aligned to h/w logical block size %u",
663 dm_device_name(table->md), i,
664 (unsigned long long) ti->begin,
665 (unsigned long long) ti->len,
666 limits->logical_block_size);
673 int dm_table_add_target(struct dm_table *t, const char *type,
674 sector_t start, sector_t len, char *params)
676 int r = -EINVAL, argc;
678 struct dm_target *tgt;
681 DMERR("%s: target type %s must appear alone in table",
682 dm_device_name(t->md), t->targets->type->name);
686 BUG_ON(t->num_targets >= t->num_allocated);
688 tgt = t->targets + t->num_targets;
689 memset(tgt, 0, sizeof(*tgt));
692 DMERR("%s: zero-length target", dm_device_name(t->md));
696 tgt->type = dm_get_target_type(type);
698 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
703 if (dm_target_needs_singleton(tgt->type)) {
704 if (t->num_targets) {
705 DMERR("%s: target type %s must appear alone in table",
706 dm_device_name(t->md), type);
712 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
713 DMERR("%s: target type %s may not be included in read-only tables",
714 dm_device_name(t->md), type);
718 if (t->immutable_target_type) {
719 if (t->immutable_target_type != tgt->type) {
720 DMERR("%s: immutable target type %s cannot be mixed with other target types",
721 dm_device_name(t->md), t->immutable_target_type->name);
724 } else if (dm_target_is_immutable(tgt->type)) {
725 if (t->num_targets) {
726 DMERR("%s: immutable target type %s cannot be mixed with other target types",
727 dm_device_name(t->md), tgt->type->name);
730 t->immutable_target_type = tgt->type;
736 tgt->error = "Unknown error";
739 * Does this target adjoin the previous one ?
741 if (!adjoin(t, tgt)) {
742 tgt->error = "Gap in table";
747 r = dm_split_args(&argc, &argv, params);
749 tgt->error = "couldn't split parameters (insufficient memory)";
753 r = tgt->type->ctr(tgt, argc, argv);
758 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
760 if (!tgt->num_discard_bios && tgt->discards_supported)
761 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
762 dm_device_name(t->md), type);
767 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
768 dm_put_target_type(tgt->type);
773 * Target argument parsing helpers.
775 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
776 unsigned *value, char **error, unsigned grouped)
778 const char *arg_str = dm_shift_arg(arg_set);
782 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
783 (*value < arg->min) ||
784 (*value > arg->max) ||
785 (grouped && arg_set->argc < *value)) {
793 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
794 unsigned *value, char **error)
796 return validate_next_arg(arg, arg_set, value, error, 0);
798 EXPORT_SYMBOL(dm_read_arg);
800 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
801 unsigned *value, char **error)
803 return validate_next_arg(arg, arg_set, value, error, 1);
805 EXPORT_SYMBOL(dm_read_arg_group);
807 const char *dm_shift_arg(struct dm_arg_set *as)
820 EXPORT_SYMBOL(dm_shift_arg);
822 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
824 BUG_ON(as->argc < num_args);
825 as->argc -= num_args;
826 as->argv += num_args;
828 EXPORT_SYMBOL(dm_consume_args);
830 static bool __table_type_request_based(unsigned table_type)
832 return (table_type == DM_TYPE_REQUEST_BASED ||
833 table_type == DM_TYPE_MQ_REQUEST_BASED);
836 void dm_table_set_type(struct dm_table *t, unsigned type)
840 EXPORT_SYMBOL_GPL(dm_table_set_type);
842 static int dm_table_determine_type(struct dm_table *t)
845 unsigned bio_based = 0, request_based = 0, hybrid = 0;
846 bool verify_blk_mq = false;
847 struct dm_target *tgt;
848 struct dm_dev_internal *dd;
849 struct list_head *devices = dm_table_get_devices(t);
850 unsigned live_md_type = dm_get_md_type(t->md);
852 if (t->type != DM_TYPE_NONE) {
853 /* target already set the table's type */
854 if (t->type == DM_TYPE_BIO_BASED)
856 goto verify_rq_based;
859 for (i = 0; i < t->num_targets; i++) {
860 tgt = t->targets + i;
861 if (dm_target_hybrid(tgt))
863 else if (dm_target_request_based(tgt))
868 if (bio_based && request_based) {
869 DMWARN("Inconsistent table: different target types"
870 " can't be mixed up");
875 if (hybrid && !bio_based && !request_based) {
877 * The targets can work either way.
878 * Determine the type from the live device.
879 * Default to bio-based if device is new.
881 if (__table_type_request_based(live_md_type))
888 /* We must use this table as bio-based */
889 t->type = DM_TYPE_BIO_BASED;
893 BUG_ON(!request_based); /* No targets in this table */
895 if (list_empty(devices) && __table_type_request_based(live_md_type)) {
896 /* inherit live MD type */
897 t->type = live_md_type;
902 * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
903 * having a compatible target use dm_table_set_type.
905 t->type = DM_TYPE_REQUEST_BASED;
909 * Request-based dm supports only tables that have a single target now.
910 * To support multiple targets, request splitting support is needed,
911 * and that needs lots of changes in the block-layer.
912 * (e.g. request completion process for partial completion.)
914 if (t->num_targets > 1) {
915 DMWARN("Request-based dm doesn't support multiple targets yet");
919 /* Non-request-stackable devices can't be used for request-based dm */
920 list_for_each_entry(dd, devices, list) {
921 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
923 if (!blk_queue_stackable(q)) {
924 DMERR("table load rejected: including"
925 " non-request-stackable devices");
930 verify_blk_mq = true;
934 /* verify _all_ devices in the table are blk-mq devices */
935 list_for_each_entry(dd, devices, list)
936 if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
937 DMERR("table load rejected: not all devices"
938 " are blk-mq request-stackable");
942 t->all_blk_mq = true;
948 unsigned dm_table_get_type(struct dm_table *t)
953 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
955 return t->immutable_target_type;
958 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
960 /* Immutable target is implicitly a singleton */
961 if (t->num_targets > 1 ||
962 !dm_target_is_immutable(t->targets[0].type))
968 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
970 struct dm_target *uninitialized_var(ti);
973 while (i < dm_table_get_num_targets(t)) {
974 ti = dm_table_get_target(t, i++);
975 if (dm_target_is_wildcard(ti->type))
982 bool dm_table_request_based(struct dm_table *t)
984 return __table_type_request_based(dm_table_get_type(t));
987 bool dm_table_all_blk_mq_devices(struct dm_table *t)
989 return t->all_blk_mq;
992 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
994 unsigned type = dm_table_get_type(t);
995 unsigned per_io_data_size = 0;
996 struct dm_target *tgt;
999 if (unlikely(type == DM_TYPE_NONE)) {
1000 DMWARN("no table type is set, can't allocate mempools");
1004 if (type == DM_TYPE_BIO_BASED)
1005 for (i = 0; i < t->num_targets; i++) {
1006 tgt = t->targets + i;
1007 per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1010 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1017 void dm_table_free_md_mempools(struct dm_table *t)
1019 dm_free_md_mempools(t->mempools);
1023 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1028 static int setup_indexes(struct dm_table *t)
1031 unsigned int total = 0;
1034 /* allocate the space for *all* the indexes */
1035 for (i = t->depth - 2; i >= 0; i--) {
1036 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1037 total += t->counts[i];
1040 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1044 /* set up internal nodes, bottom-up */
1045 for (i = t->depth - 2; i >= 0; i--) {
1046 t->index[i] = indexes;
1047 indexes += (KEYS_PER_NODE * t->counts[i]);
1048 setup_btree_index(i, t);
1055 * Builds the btree to index the map.
1057 static int dm_table_build_index(struct dm_table *t)
1060 unsigned int leaf_nodes;
1062 /* how many indexes will the btree have ? */
1063 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1064 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1066 /* leaf layer has already been set up */
1067 t->counts[t->depth - 1] = leaf_nodes;
1068 t->index[t->depth - 1] = t->highs;
1071 r = setup_indexes(t);
1076 static bool integrity_profile_exists(struct gendisk *disk)
1078 return !!blk_get_integrity(disk);
1082 * Get a disk whose integrity profile reflects the table's profile.
1083 * Returns NULL if integrity support was inconsistent or unavailable.
1085 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1087 struct list_head *devices = dm_table_get_devices(t);
1088 struct dm_dev_internal *dd = NULL;
1089 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1091 list_for_each_entry(dd, devices, list) {
1092 template_disk = dd->dm_dev->bdev->bd_disk;
1093 if (!integrity_profile_exists(template_disk))
1095 else if (prev_disk &&
1096 blk_integrity_compare(prev_disk, template_disk) < 0)
1098 prev_disk = template_disk;
1101 return template_disk;
1105 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1106 dm_device_name(t->md),
1107 prev_disk->disk_name,
1108 template_disk->disk_name);
1113 * Register the mapped device for blk_integrity support if the
1114 * underlying devices have an integrity profile. But all devices may
1115 * not have matching profiles (checking all devices isn't reliable
1116 * during table load because this table may use other DM device(s) which
1117 * must be resumed before they will have an initialized integity
1118 * profile). Consequently, stacked DM devices force a 2 stage integrity
1119 * profile validation: First pass during table load, final pass during
1122 static int dm_table_register_integrity(struct dm_table *t)
1124 struct mapped_device *md = t->md;
1125 struct gendisk *template_disk = NULL;
1127 template_disk = dm_table_get_integrity_disk(t);
1131 if (!integrity_profile_exists(dm_disk(md))) {
1132 t->integrity_supported = true;
1134 * Register integrity profile during table load; we can do
1135 * this because the final profile must match during resume.
1137 blk_integrity_register(dm_disk(md),
1138 blk_get_integrity(template_disk));
1143 * If DM device already has an initialized integrity
1144 * profile the new profile should not conflict.
1146 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1147 DMWARN("%s: conflict with existing integrity profile: "
1148 "%s profile mismatch",
1149 dm_device_name(t->md),
1150 template_disk->disk_name);
1154 /* Preserve existing integrity profile */
1155 t->integrity_supported = true;
1160 * Prepares the table for use by building the indices,
1161 * setting the type, and allocating mempools.
1163 int dm_table_complete(struct dm_table *t)
1167 r = dm_table_determine_type(t);
1169 DMERR("unable to determine table type");
1173 r = dm_table_build_index(t);
1175 DMERR("unable to build btrees");
1179 r = dm_table_register_integrity(t);
1181 DMERR("could not register integrity profile.");
1185 r = dm_table_alloc_md_mempools(t, t->md);
1187 DMERR("unable to allocate mempools");
1192 static DEFINE_MUTEX(_event_lock);
1193 void dm_table_event_callback(struct dm_table *t,
1194 void (*fn)(void *), void *context)
1196 mutex_lock(&_event_lock);
1198 t->event_context = context;
1199 mutex_unlock(&_event_lock);
1202 void dm_table_event(struct dm_table *t)
1205 * You can no longer call dm_table_event() from interrupt
1206 * context, use a bottom half instead.
1208 BUG_ON(in_interrupt());
1210 mutex_lock(&_event_lock);
1212 t->event_fn(t->event_context);
1213 mutex_unlock(&_event_lock);
1215 EXPORT_SYMBOL(dm_table_event);
1217 sector_t dm_table_get_size(struct dm_table *t)
1219 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1221 EXPORT_SYMBOL(dm_table_get_size);
1223 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1225 if (index >= t->num_targets)
1228 return t->targets + index;
1232 * Search the btree for the correct target.
1234 * Caller should check returned pointer with dm_target_is_valid()
1235 * to trap I/O beyond end of device.
1237 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1239 unsigned int l, n = 0, k = 0;
1242 for (l = 0; l < t->depth; l++) {
1243 n = get_child(n, k);
1244 node = get_node(t, l, n);
1246 for (k = 0; k < KEYS_PER_NODE; k++)
1247 if (node[k] >= sector)
1251 return &t->targets[(KEYS_PER_NODE * n) + k];
1254 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1255 sector_t start, sector_t len, void *data)
1257 unsigned *num_devices = data;
1265 * Check whether a table has no data devices attached using each
1266 * target's iterate_devices method.
1267 * Returns false if the result is unknown because a target doesn't
1268 * support iterate_devices.
1270 bool dm_table_has_no_data_devices(struct dm_table *table)
1272 struct dm_target *uninitialized_var(ti);
1273 unsigned i = 0, num_devices = 0;
1275 while (i < dm_table_get_num_targets(table)) {
1276 ti = dm_table_get_target(table, i++);
1278 if (!ti->type->iterate_devices)
1281 ti->type->iterate_devices(ti, count_device, &num_devices);
1290 * Establish the new table's queue_limits and validate them.
1292 int dm_calculate_queue_limits(struct dm_table *table,
1293 struct queue_limits *limits)
1295 struct dm_target *uninitialized_var(ti);
1296 struct queue_limits ti_limits;
1299 blk_set_stacking_limits(limits);
1301 while (i < dm_table_get_num_targets(table)) {
1302 blk_set_stacking_limits(&ti_limits);
1304 ti = dm_table_get_target(table, i++);
1306 if (!ti->type->iterate_devices)
1307 goto combine_limits;
1310 * Combine queue limits of all the devices this target uses.
1312 ti->type->iterate_devices(ti, dm_set_device_limits,
1315 /* Set I/O hints portion of queue limits */
1316 if (ti->type->io_hints)
1317 ti->type->io_hints(ti, &ti_limits);
1320 * Check each device area is consistent with the target's
1321 * overall queue limits.
1323 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1329 * Merge this target's queue limits into the overall limits
1332 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1333 DMWARN("%s: adding target device "
1334 "(start sect %llu len %llu) "
1335 "caused an alignment inconsistency",
1336 dm_device_name(table->md),
1337 (unsigned long long) ti->begin,
1338 (unsigned long long) ti->len);
1341 return validate_hardware_logical_block_alignment(table, limits);
1345 * Verify that all devices have an integrity profile that matches the
1346 * DM device's registered integrity profile. If the profiles don't
1347 * match then unregister the DM device's integrity profile.
1349 static void dm_table_verify_integrity(struct dm_table *t)
1351 struct gendisk *template_disk = NULL;
1353 if (t->integrity_supported) {
1355 * Verify that the original integrity profile
1356 * matches all the devices in this table.
1358 template_disk = dm_table_get_integrity_disk(t);
1359 if (template_disk &&
1360 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1364 if (integrity_profile_exists(dm_disk(t->md))) {
1365 DMWARN("%s: unable to establish an integrity profile",
1366 dm_device_name(t->md));
1367 blk_integrity_unregister(dm_disk(t->md));
1371 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1372 sector_t start, sector_t len, void *data)
1374 unsigned long flush = (unsigned long) data;
1375 struct request_queue *q = bdev_get_queue(dev->bdev);
1377 return q && (q->queue_flags & flush);
1380 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1382 struct dm_target *ti;
1386 * Require at least one underlying device to support flushes.
1387 * t->devices includes internal dm devices such as mirror logs
1388 * so we need to use iterate_devices here, which targets
1389 * supporting flushes must provide.
1391 while (i < dm_table_get_num_targets(t)) {
1392 ti = dm_table_get_target(t, i++);
1394 if (!ti->num_flush_bios)
1397 if (ti->flush_supported)
1400 if (ti->type->iterate_devices &&
1401 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1408 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1410 struct dm_target *ti;
1413 /* Ensure that all targets supports discard_zeroes_data. */
1414 while (i < dm_table_get_num_targets(t)) {
1415 ti = dm_table_get_target(t, i++);
1417 if (ti->discard_zeroes_data_unsupported)
1424 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1425 sector_t start, sector_t len, void *data)
1427 struct request_queue *q = bdev_get_queue(dev->bdev);
1429 return q && blk_queue_nonrot(q);
1432 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1433 sector_t start, sector_t len, void *data)
1435 struct request_queue *q = bdev_get_queue(dev->bdev);
1437 return q && !blk_queue_add_random(q);
1440 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1441 sector_t start, sector_t len, void *data)
1443 struct request_queue *q = bdev_get_queue(dev->bdev);
1445 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1448 static bool dm_table_all_devices_attribute(struct dm_table *t,
1449 iterate_devices_callout_fn func)
1451 struct dm_target *ti;
1454 while (i < dm_table_get_num_targets(t)) {
1455 ti = dm_table_get_target(t, i++);
1457 if (!ti->type->iterate_devices ||
1458 !ti->type->iterate_devices(ti, func, NULL))
1465 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1466 sector_t start, sector_t len, void *data)
1468 struct request_queue *q = bdev_get_queue(dev->bdev);
1470 return q && !q->limits.max_write_same_sectors;
1473 static bool dm_table_supports_write_same(struct dm_table *t)
1475 struct dm_target *ti;
1478 while (i < dm_table_get_num_targets(t)) {
1479 ti = dm_table_get_target(t, i++);
1481 if (!ti->num_write_same_bios)
1484 if (!ti->type->iterate_devices ||
1485 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1492 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1493 sector_t start, sector_t len, void *data)
1495 struct request_queue *q = bdev_get_queue(dev->bdev);
1497 return q && blk_queue_discard(q);
1500 static bool dm_table_supports_discards(struct dm_table *t)
1502 struct dm_target *ti;
1506 * Unless any target used by the table set discards_supported,
1507 * require at least one underlying device to support discards.
1508 * t->devices includes internal dm devices such as mirror logs
1509 * so we need to use iterate_devices here, which targets
1510 * supporting discard selectively must provide.
1512 while (i < dm_table_get_num_targets(t)) {
1513 ti = dm_table_get_target(t, i++);
1515 if (!ti->num_discard_bios)
1518 if (ti->discards_supported)
1521 if (ti->type->iterate_devices &&
1522 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1529 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1530 struct queue_limits *limits)
1532 bool wc = false, fua = false;
1535 * Copy table's limits to the DM device's request_queue
1537 q->limits = *limits;
1539 if (!dm_table_supports_discards(t))
1540 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1542 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1544 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1546 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1549 blk_queue_write_cache(q, wc, fua);
1551 if (!dm_table_discard_zeroes_data(t))
1552 q->limits.discard_zeroes_data = 0;
1554 /* Ensure that all underlying devices are non-rotational. */
1555 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1556 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1558 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1560 if (!dm_table_supports_write_same(t))
1561 q->limits.max_write_same_sectors = 0;
1563 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1564 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1566 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1568 dm_table_verify_integrity(t);
1571 * Determine whether or not this queue's I/O timings contribute
1572 * to the entropy pool, Only request-based targets use this.
1573 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1576 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1577 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1580 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1581 * visible to other CPUs because, once the flag is set, incoming bios
1582 * are processed by request-based dm, which refers to the queue
1584 * Until the flag set, bios are passed to bio-based dm and queued to
1585 * md->deferred where queue settings are not needed yet.
1586 * Those bios are passed to request-based dm at the resume time.
1589 if (dm_table_request_based(t))
1590 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1593 unsigned int dm_table_get_num_targets(struct dm_table *t)
1595 return t->num_targets;
1598 struct list_head *dm_table_get_devices(struct dm_table *t)
1603 fmode_t dm_table_get_mode(struct dm_table *t)
1607 EXPORT_SYMBOL(dm_table_get_mode);
1615 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1617 int i = t->num_targets;
1618 struct dm_target *ti = t->targets;
1623 if (ti->type->presuspend)
1624 ti->type->presuspend(ti);
1626 case PRESUSPEND_UNDO:
1627 if (ti->type->presuspend_undo)
1628 ti->type->presuspend_undo(ti);
1631 if (ti->type->postsuspend)
1632 ti->type->postsuspend(ti);
1639 void dm_table_presuspend_targets(struct dm_table *t)
1644 suspend_targets(t, PRESUSPEND);
1647 void dm_table_presuspend_undo_targets(struct dm_table *t)
1652 suspend_targets(t, PRESUSPEND_UNDO);
1655 void dm_table_postsuspend_targets(struct dm_table *t)
1660 suspend_targets(t, POSTSUSPEND);
1663 int dm_table_resume_targets(struct dm_table *t)
1667 for (i = 0; i < t->num_targets; i++) {
1668 struct dm_target *ti = t->targets + i;
1670 if (!ti->type->preresume)
1673 r = ti->type->preresume(ti);
1675 DMERR("%s: %s: preresume failed, error = %d",
1676 dm_device_name(t->md), ti->type->name, r);
1681 for (i = 0; i < t->num_targets; i++) {
1682 struct dm_target *ti = t->targets + i;
1684 if (ti->type->resume)
1685 ti->type->resume(ti);
1691 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1693 list_add(&cb->list, &t->target_callbacks);
1695 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1697 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1699 struct dm_dev_internal *dd;
1700 struct list_head *devices = dm_table_get_devices(t);
1701 struct dm_target_callbacks *cb;
1704 list_for_each_entry(dd, devices, list) {
1705 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1706 char b[BDEVNAME_SIZE];
1709 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1711 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1712 dm_device_name(t->md),
1713 bdevname(dd->dm_dev->bdev, b));
1716 list_for_each_entry(cb, &t->target_callbacks, list)
1717 if (cb->congested_fn)
1718 r |= cb->congested_fn(cb, bdi_bits);
1723 struct mapped_device *dm_table_get_md(struct dm_table *t)
1727 EXPORT_SYMBOL(dm_table_get_md);
1729 void dm_table_run_md_queue_async(struct dm_table *t)
1731 struct mapped_device *md;
1732 struct request_queue *queue;
1733 unsigned long flags;
1735 if (!dm_table_request_based(t))
1738 md = dm_table_get_md(t);
1739 queue = dm_get_md_queue(md);
1742 blk_mq_run_hw_queues(queue, true);
1744 spin_lock_irqsave(queue->queue_lock, flags);
1745 blk_run_queue_async(queue);
1746 spin_unlock_irqrestore(queue->queue_lock, flags);
1750 EXPORT_SYMBOL(dm_table_run_md_queue_async);