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);
223 if (alloc_targets(t, num_targets)) {
234 static void free_devices(struct list_head *devices)
236 struct list_head *tmp, *next;
238 list_for_each_safe(tmp, next, devices) {
239 struct dm_dev_internal *dd =
240 list_entry(tmp, struct dm_dev_internal, list);
241 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
247 void dm_table_destroy(struct dm_table *t)
254 while (atomic_read(&t->holders))
258 /* free the indexes */
260 vfree(t->index[t->depth - 2]);
262 /* free the targets */
263 for (i = 0; i < t->num_targets; i++) {
264 struct dm_target *tgt = t->targets + i;
269 dm_put_target_type(tgt->type);
274 /* free the device list */
275 free_devices(&t->devices);
277 dm_free_md_mempools(t->mempools);
282 void dm_table_get(struct dm_table *t)
284 atomic_inc(&t->holders);
286 EXPORT_SYMBOL(dm_table_get);
288 void dm_table_put(struct dm_table *t)
293 smp_mb__before_atomic_dec();
294 atomic_dec(&t->holders);
296 EXPORT_SYMBOL(dm_table_put);
299 * Checks to see if we need to extend highs or targets.
301 static inline int check_space(struct dm_table *t)
303 if (t->num_targets >= t->num_allocated)
304 return alloc_targets(t, t->num_allocated * 2);
310 * See if we've already got a device in the list.
312 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
314 struct dm_dev_internal *dd;
316 list_for_each_entry (dd, l, list)
317 if (dd->dm_dev.bdev->bd_dev == dev)
324 * Open a device so we can use it as a map destination.
326 static int open_dev(struct dm_dev_internal *d, dev_t dev,
327 struct mapped_device *md)
329 static char *_claim_ptr = "I belong to device-mapper";
330 struct block_device *bdev;
334 BUG_ON(d->dm_dev.bdev);
336 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
338 return PTR_ERR(bdev);
340 r = bd_link_disk_holder(bdev, dm_disk(md));
342 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
346 d->dm_dev.bdev = bdev;
351 * Close a device that we've been using.
353 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
358 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
359 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
360 d->dm_dev.bdev = NULL;
364 * If possible, this checks an area of a destination device is invalid.
366 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
367 sector_t start, sector_t len, void *data)
369 struct request_queue *q;
370 struct queue_limits *limits = data;
371 struct block_device *bdev = dev->bdev;
373 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
374 unsigned short logical_block_size_sectors =
375 limits->logical_block_size >> SECTOR_SHIFT;
376 char b[BDEVNAME_SIZE];
379 * Some devices exist without request functions,
380 * such as loop devices not yet bound to backing files.
381 * Forbid the use of such devices.
383 q = bdev_get_queue(bdev);
384 if (!q || !q->make_request_fn) {
385 DMWARN("%s: %s is not yet initialised: "
386 "start=%llu, len=%llu, dev_size=%llu",
387 dm_device_name(ti->table->md), bdevname(bdev, b),
388 (unsigned long long)start,
389 (unsigned long long)len,
390 (unsigned long long)dev_size);
397 if ((start >= dev_size) || (start + len > dev_size)) {
398 DMWARN("%s: %s too small for target: "
399 "start=%llu, len=%llu, dev_size=%llu",
400 dm_device_name(ti->table->md), bdevname(bdev, b),
401 (unsigned long long)start,
402 (unsigned long long)len,
403 (unsigned long long)dev_size);
407 if (logical_block_size_sectors <= 1)
410 if (start & (logical_block_size_sectors - 1)) {
411 DMWARN("%s: start=%llu not aligned to h/w "
412 "logical block size %u of %s",
413 dm_device_name(ti->table->md),
414 (unsigned long long)start,
415 limits->logical_block_size, bdevname(bdev, b));
419 if (len & (logical_block_size_sectors - 1)) {
420 DMWARN("%s: len=%llu not aligned to h/w "
421 "logical block size %u of %s",
422 dm_device_name(ti->table->md),
423 (unsigned long long)len,
424 limits->logical_block_size, bdevname(bdev, b));
432 * This upgrades the mode on an already open dm_dev, being
433 * careful to leave things as they were if we fail to reopen the
434 * device and not to touch the existing bdev field in case
435 * it is accessed concurrently inside dm_table_any_congested().
437 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
438 struct mapped_device *md)
441 struct dm_dev_internal dd_new, dd_old;
443 dd_new = dd_old = *dd;
445 dd_new.dm_dev.mode |= new_mode;
446 dd_new.dm_dev.bdev = NULL;
448 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 if ((r = open_dev(dd, dev, t->md))) {
503 format_dev_t(dd->dm_dev.name, dev);
505 atomic_set(&dd->count, 0);
506 list_add(&dd->list, &t->devices);
508 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
509 r = upgrade_mode(dd, mode, t->md);
513 atomic_inc(&dd->count);
515 *result = &dd->dm_dev;
518 EXPORT_SYMBOL(dm_get_device);
520 int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
521 sector_t start, sector_t len, void *data)
523 struct queue_limits *limits = data;
524 struct block_device *bdev = dev->bdev;
525 struct request_queue *q = bdev_get_queue(bdev);
526 char b[BDEVNAME_SIZE];
529 DMWARN("%s: Cannot set limits for nonexistent device %s",
530 dm_device_name(ti->table->md), bdevname(bdev, b));
534 if (bdev_stack_limits(limits, bdev, start) < 0)
535 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
536 "physical_block_size=%u, logical_block_size=%u, "
537 "alignment_offset=%u, start=%llu",
538 dm_device_name(ti->table->md), bdevname(bdev, b),
539 q->limits.physical_block_size,
540 q->limits.logical_block_size,
541 q->limits.alignment_offset,
542 (unsigned long long) start << SECTOR_SHIFT);
545 * Check if merge fn is supported.
546 * If not we'll force DM to use PAGE_SIZE or
547 * smaller I/O, just to be safe.
549 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
550 blk_limits_max_hw_sectors(limits,
551 (unsigned int) (PAGE_SIZE >> 9));
554 EXPORT_SYMBOL_GPL(dm_set_device_limits);
557 * Decrement a device's use count and remove it if necessary.
559 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
561 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
564 if (atomic_dec_and_test(&dd->count)) {
565 close_dev(dd, ti->table->md);
570 EXPORT_SYMBOL(dm_put_device);
573 * Checks to see if the target joins onto the end of the table.
575 static int adjoin(struct dm_table *table, struct dm_target *ti)
577 struct dm_target *prev;
579 if (!table->num_targets)
582 prev = &table->targets[table->num_targets - 1];
583 return (ti->begin == (prev->begin + prev->len));
587 * Used to dynamically allocate the arg array.
589 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
590 * process messages even if some device is suspended. These messages have a
591 * small fixed number of arguments.
593 * On the other hand, dm-switch needs to process bulk data using messages and
594 * excessive use of GFP_NOIO could cause trouble.
596 static char **realloc_argv(unsigned *array_size, char **old_argv)
603 new_size = *array_size * 2;
609 argv = kmalloc(new_size * sizeof(*argv), gfp);
611 memcpy(argv, old_argv, *array_size * sizeof(*argv));
612 *array_size = new_size;
620 * Destructively splits up the argument list to pass to ctr.
622 int dm_split_args(int *argc, char ***argvp, char *input)
624 char *start, *end = input, *out, **argv = NULL;
625 unsigned array_size = 0;
634 argv = realloc_argv(&array_size, argv);
639 /* Skip whitespace */
640 start = skip_spaces(end);
643 break; /* success, we hit the end */
645 /* 'out' is used to remove any back-quotes */
648 /* Everything apart from '\0' can be quoted */
649 if (*end == '\\' && *(end + 1)) {
656 break; /* end of token */
661 /* have we already filled the array ? */
662 if ((*argc + 1) > array_size) {
663 argv = realloc_argv(&array_size, argv);
668 /* we know this is whitespace */
672 /* terminate the string and put it in the array */
683 * Impose necessary and sufficient conditions on a devices's table such
684 * that any incoming bio which respects its logical_block_size can be
685 * processed successfully. If it falls across the boundary between
686 * two or more targets, the size of each piece it gets split into must
687 * be compatible with the logical_block_size of the target processing it.
689 static int validate_hardware_logical_block_alignment(struct dm_table *table,
690 struct queue_limits *limits)
693 * This function uses arithmetic modulo the logical_block_size
694 * (in units of 512-byte sectors).
696 unsigned short device_logical_block_size_sects =
697 limits->logical_block_size >> SECTOR_SHIFT;
700 * Offset of the start of the next table entry, mod logical_block_size.
702 unsigned short next_target_start = 0;
705 * Given an aligned bio that extends beyond the end of a
706 * target, how many sectors must the next target handle?
708 unsigned short remaining = 0;
710 struct dm_target *uninitialized_var(ti);
711 struct queue_limits ti_limits;
715 * Check each entry in the table in turn.
717 while (i < dm_table_get_num_targets(table)) {
718 ti = dm_table_get_target(table, i++);
720 blk_set_stacking_limits(&ti_limits);
722 /* combine all target devices' limits */
723 if (ti->type->iterate_devices)
724 ti->type->iterate_devices(ti, dm_set_device_limits,
728 * If the remaining sectors fall entirely within this
729 * table entry are they compatible with its logical_block_size?
731 if (remaining < ti->len &&
732 remaining & ((ti_limits.logical_block_size >>
737 (unsigned short) ((next_target_start + ti->len) &
738 (device_logical_block_size_sects - 1));
739 remaining = next_target_start ?
740 device_logical_block_size_sects - next_target_start : 0;
744 DMWARN("%s: table line %u (start sect %llu len %llu) "
745 "not aligned to h/w logical block size %u",
746 dm_device_name(table->md), i,
747 (unsigned long long) ti->begin,
748 (unsigned long long) ti->len,
749 limits->logical_block_size);
756 int dm_table_add_target(struct dm_table *t, const char *type,
757 sector_t start, sector_t len, char *params)
759 int r = -EINVAL, argc;
761 struct dm_target *tgt;
764 DMERR("%s: target type %s must appear alone in table",
765 dm_device_name(t->md), t->targets->type->name);
769 if ((r = check_space(t)))
772 tgt = t->targets + t->num_targets;
773 memset(tgt, 0, sizeof(*tgt));
776 DMERR("%s: zero-length target", dm_device_name(t->md));
780 tgt->type = dm_get_target_type(type);
782 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
787 if (dm_target_needs_singleton(tgt->type)) {
788 if (t->num_targets) {
789 DMERR("%s: target type %s must appear alone in table",
790 dm_device_name(t->md), type);
796 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
797 DMERR("%s: target type %s may not be included in read-only tables",
798 dm_device_name(t->md), type);
802 if (t->immutable_target_type) {
803 if (t->immutable_target_type != tgt->type) {
804 DMERR("%s: immutable target type %s cannot be mixed with other target types",
805 dm_device_name(t->md), t->immutable_target_type->name);
808 } else if (dm_target_is_immutable(tgt->type)) {
809 if (t->num_targets) {
810 DMERR("%s: immutable target type %s cannot be mixed with other target types",
811 dm_device_name(t->md), tgt->type->name);
814 t->immutable_target_type = tgt->type;
820 tgt->error = "Unknown error";
823 * Does this target adjoin the previous one ?
825 if (!adjoin(t, tgt)) {
826 tgt->error = "Gap in table";
831 r = dm_split_args(&argc, &argv, params);
833 tgt->error = "couldn't split parameters (insufficient memory)";
837 r = tgt->type->ctr(tgt, argc, argv);
842 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
844 if (!tgt->num_discard_bios && tgt->discards_supported)
845 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
846 dm_device_name(t->md), type);
851 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
852 dm_put_target_type(tgt->type);
857 * Target argument parsing helpers.
859 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
860 unsigned *value, char **error, unsigned grouped)
862 const char *arg_str = dm_shift_arg(arg_set);
866 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
867 (*value < arg->min) ||
868 (*value > arg->max) ||
869 (grouped && arg_set->argc < *value)) {
877 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
878 unsigned *value, char **error)
880 return validate_next_arg(arg, arg_set, value, error, 0);
882 EXPORT_SYMBOL(dm_read_arg);
884 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
885 unsigned *value, char **error)
887 return validate_next_arg(arg, arg_set, value, error, 1);
889 EXPORT_SYMBOL(dm_read_arg_group);
891 const char *dm_shift_arg(struct dm_arg_set *as)
904 EXPORT_SYMBOL(dm_shift_arg);
906 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
908 BUG_ON(as->argc < num_args);
909 as->argc -= num_args;
910 as->argv += num_args;
912 EXPORT_SYMBOL(dm_consume_args);
914 static int dm_table_set_type(struct dm_table *t)
917 unsigned bio_based = 0, request_based = 0;
918 struct dm_target *tgt;
919 struct dm_dev_internal *dd;
920 struct list_head *devices;
922 for (i = 0; i < t->num_targets; i++) {
923 tgt = t->targets + i;
924 if (dm_target_request_based(tgt))
929 if (bio_based && request_based) {
930 DMWARN("Inconsistent table: different target types"
931 " can't be mixed up");
937 /* We must use this table as bio-based */
938 t->type = DM_TYPE_BIO_BASED;
942 BUG_ON(!request_based); /* No targets in this table */
944 /* Non-request-stackable devices can't be used for request-based dm */
945 devices = dm_table_get_devices(t);
946 list_for_each_entry(dd, devices, list) {
947 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
948 DMWARN("table load rejected: including"
949 " non-request-stackable devices");
955 * Request-based dm supports only tables that have a single target now.
956 * To support multiple targets, request splitting support is needed,
957 * and that needs lots of changes in the block-layer.
958 * (e.g. request completion process for partial completion.)
960 if (t->num_targets > 1) {
961 DMWARN("Request-based dm doesn't support multiple targets yet");
965 t->type = DM_TYPE_REQUEST_BASED;
970 unsigned dm_table_get_type(struct dm_table *t)
975 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
977 return t->immutable_target_type;
980 bool dm_table_request_based(struct dm_table *t)
982 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
985 int dm_table_alloc_md_mempools(struct dm_table *t)
987 unsigned type = dm_table_get_type(t);
988 unsigned per_bio_data_size = 0;
989 struct dm_target *tgt;
992 if (unlikely(type == DM_TYPE_NONE)) {
993 DMWARN("no table type is set, can't allocate mempools");
997 if (type == DM_TYPE_BIO_BASED)
998 for (i = 0; i < t->num_targets; i++) {
999 tgt = t->targets + i;
1000 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
1003 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported, per_bio_data_size);
1010 void dm_table_free_md_mempools(struct dm_table *t)
1012 dm_free_md_mempools(t->mempools);
1016 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1021 static int setup_indexes(struct dm_table *t)
1024 unsigned int total = 0;
1027 /* allocate the space for *all* the indexes */
1028 for (i = t->depth - 2; i >= 0; i--) {
1029 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1030 total += t->counts[i];
1033 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1037 /* set up internal nodes, bottom-up */
1038 for (i = t->depth - 2; i >= 0; i--) {
1039 t->index[i] = indexes;
1040 indexes += (KEYS_PER_NODE * t->counts[i]);
1041 setup_btree_index(i, t);
1048 * Builds the btree to index the map.
1050 static int dm_table_build_index(struct dm_table *t)
1053 unsigned int leaf_nodes;
1055 /* how many indexes will the btree have ? */
1056 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1057 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1059 /* leaf layer has already been set up */
1060 t->counts[t->depth - 1] = leaf_nodes;
1061 t->index[t->depth - 1] = t->highs;
1064 r = setup_indexes(t);
1070 * Get a disk whose integrity profile reflects the table's profile.
1071 * If %match_all is true, all devices' profiles must match.
1072 * If %match_all is false, all devices must at least have an
1073 * allocated integrity profile; but uninitialized is ok.
1074 * Returns NULL if integrity support was inconsistent or unavailable.
1076 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1079 struct list_head *devices = dm_table_get_devices(t);
1080 struct dm_dev_internal *dd = NULL;
1081 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1083 list_for_each_entry(dd, devices, list) {
1084 template_disk = dd->dm_dev.bdev->bd_disk;
1085 if (!blk_get_integrity(template_disk))
1087 if (!match_all && !blk_integrity_is_initialized(template_disk))
1088 continue; /* skip uninitialized profiles */
1089 else if (prev_disk &&
1090 blk_integrity_compare(prev_disk, template_disk) < 0)
1092 prev_disk = template_disk;
1095 return template_disk;
1099 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1100 dm_device_name(t->md),
1101 prev_disk->disk_name,
1102 template_disk->disk_name);
1107 * Register the mapped device for blk_integrity support if
1108 * the underlying devices have an integrity profile. But all devices
1109 * may not have matching profiles (checking all devices isn't reliable
1110 * during table load because this table may use other DM device(s) which
1111 * must be resumed before they will have an initialized integity profile).
1112 * Stacked DM devices force a 2 stage integrity profile validation:
1113 * 1 - during load, validate all initialized integrity profiles match
1114 * 2 - during resume, validate all integrity profiles match
1116 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1118 struct gendisk *template_disk = NULL;
1120 template_disk = dm_table_get_integrity_disk(t, false);
1124 if (!blk_integrity_is_initialized(dm_disk(md))) {
1125 t->integrity_supported = 1;
1126 return blk_integrity_register(dm_disk(md), NULL);
1130 * If DM device already has an initalized integrity
1131 * profile the new profile should not conflict.
1133 if (blk_integrity_is_initialized(template_disk) &&
1134 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1135 DMWARN("%s: conflict with existing integrity profile: "
1136 "%s profile mismatch",
1137 dm_device_name(t->md),
1138 template_disk->disk_name);
1142 /* Preserve existing initialized integrity profile */
1143 t->integrity_supported = 1;
1148 * Prepares the table for use by building the indices,
1149 * setting the type, and allocating mempools.
1151 int dm_table_complete(struct dm_table *t)
1155 r = dm_table_set_type(t);
1157 DMERR("unable to set table type");
1161 r = dm_table_build_index(t);
1163 DMERR("unable to build btrees");
1167 r = dm_table_prealloc_integrity(t, t->md);
1169 DMERR("could not register integrity profile.");
1173 r = dm_table_alloc_md_mempools(t);
1175 DMERR("unable to allocate mempools");
1180 static DEFINE_MUTEX(_event_lock);
1181 void dm_table_event_callback(struct dm_table *t,
1182 void (*fn)(void *), void *context)
1184 mutex_lock(&_event_lock);
1186 t->event_context = context;
1187 mutex_unlock(&_event_lock);
1190 void dm_table_event(struct dm_table *t)
1193 * You can no longer call dm_table_event() from interrupt
1194 * context, use a bottom half instead.
1196 BUG_ON(in_interrupt());
1198 mutex_lock(&_event_lock);
1200 t->event_fn(t->event_context);
1201 mutex_unlock(&_event_lock);
1203 EXPORT_SYMBOL(dm_table_event);
1205 sector_t dm_table_get_size(struct dm_table *t)
1207 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1209 EXPORT_SYMBOL(dm_table_get_size);
1211 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1213 if (index >= t->num_targets)
1216 return t->targets + index;
1220 * Search the btree for the correct target.
1222 * Caller should check returned pointer with dm_target_is_valid()
1223 * to trap I/O beyond end of device.
1225 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1227 unsigned int l, n = 0, k = 0;
1230 for (l = 0; l < t->depth; l++) {
1231 n = get_child(n, k);
1232 node = get_node(t, l, n);
1234 for (k = 0; k < KEYS_PER_NODE; k++)
1235 if (node[k] >= sector)
1239 return &t->targets[(KEYS_PER_NODE * n) + k];
1242 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1243 sector_t start, sector_t len, void *data)
1245 unsigned *num_devices = data;
1253 * Check whether a table has no data devices attached using each
1254 * target's iterate_devices method.
1255 * Returns false if the result is unknown because a target doesn't
1256 * support iterate_devices.
1258 bool dm_table_has_no_data_devices(struct dm_table *table)
1260 struct dm_target *uninitialized_var(ti);
1261 unsigned i = 0, num_devices = 0;
1263 while (i < dm_table_get_num_targets(table)) {
1264 ti = dm_table_get_target(table, i++);
1266 if (!ti->type->iterate_devices)
1269 ti->type->iterate_devices(ti, count_device, &num_devices);
1278 * Establish the new table's queue_limits and validate them.
1280 int dm_calculate_queue_limits(struct dm_table *table,
1281 struct queue_limits *limits)
1283 struct dm_target *uninitialized_var(ti);
1284 struct queue_limits ti_limits;
1287 blk_set_stacking_limits(limits);
1289 while (i < dm_table_get_num_targets(table)) {
1290 blk_set_stacking_limits(&ti_limits);
1292 ti = dm_table_get_target(table, i++);
1294 if (!ti->type->iterate_devices)
1295 goto combine_limits;
1298 * Combine queue limits of all the devices this target uses.
1300 ti->type->iterate_devices(ti, dm_set_device_limits,
1303 /* Set I/O hints portion of queue limits */
1304 if (ti->type->io_hints)
1305 ti->type->io_hints(ti, &ti_limits);
1308 * Check each device area is consistent with the target's
1309 * overall queue limits.
1311 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1317 * Merge this target's queue limits into the overall limits
1320 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1321 DMWARN("%s: adding target device "
1322 "(start sect %llu len %llu) "
1323 "caused an alignment inconsistency",
1324 dm_device_name(table->md),
1325 (unsigned long long) ti->begin,
1326 (unsigned long long) ti->len);
1329 return validate_hardware_logical_block_alignment(table, limits);
1333 * Set the integrity profile for this device if all devices used have
1334 * matching profiles. We're quite deep in the resume path but still
1335 * don't know if all devices (particularly DM devices this device
1336 * may be stacked on) have matching profiles. Even if the profiles
1337 * don't match we have no way to fail (to resume) at this point.
1339 static void dm_table_set_integrity(struct dm_table *t)
1341 struct gendisk *template_disk = NULL;
1343 if (!blk_get_integrity(dm_disk(t->md)))
1346 template_disk = dm_table_get_integrity_disk(t, true);
1348 blk_integrity_register(dm_disk(t->md),
1349 blk_get_integrity(template_disk));
1350 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1351 DMWARN("%s: device no longer has a valid integrity profile",
1352 dm_device_name(t->md));
1354 DMWARN("%s: unable to establish an integrity profile",
1355 dm_device_name(t->md));
1358 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1359 sector_t start, sector_t len, void *data)
1361 unsigned flush = (*(unsigned *)data);
1362 struct request_queue *q = bdev_get_queue(dev->bdev);
1364 return q && (q->flush_flags & flush);
1367 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1369 struct dm_target *ti;
1373 * Require at least one underlying device to support flushes.
1374 * t->devices includes internal dm devices such as mirror logs
1375 * so we need to use iterate_devices here, which targets
1376 * supporting flushes must provide.
1378 while (i < dm_table_get_num_targets(t)) {
1379 ti = dm_table_get_target(t, i++);
1381 if (!ti->num_flush_bios)
1384 if (ti->flush_supported)
1387 if (ti->type->iterate_devices &&
1388 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1395 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1397 struct dm_target *ti;
1400 /* Ensure that all targets supports discard_zeroes_data. */
1401 while (i < dm_table_get_num_targets(t)) {
1402 ti = dm_table_get_target(t, i++);
1404 if (ti->discard_zeroes_data_unsupported)
1411 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1412 sector_t start, sector_t len, void *data)
1414 struct request_queue *q = bdev_get_queue(dev->bdev);
1416 return q && blk_queue_nonrot(q);
1419 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1420 sector_t start, sector_t len, void *data)
1422 struct request_queue *q = bdev_get_queue(dev->bdev);
1424 return q && !blk_queue_add_random(q);
1427 static bool dm_table_all_devices_attribute(struct dm_table *t,
1428 iterate_devices_callout_fn func)
1430 struct dm_target *ti;
1433 while (i < dm_table_get_num_targets(t)) {
1434 ti = dm_table_get_target(t, i++);
1436 if (!ti->type->iterate_devices ||
1437 !ti->type->iterate_devices(ti, func, NULL))
1444 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1445 sector_t start, sector_t len, void *data)
1447 struct request_queue *q = bdev_get_queue(dev->bdev);
1449 return q && !q->limits.max_write_same_sectors;
1452 static bool dm_table_supports_write_same(struct dm_table *t)
1454 struct dm_target *ti;
1457 while (i < dm_table_get_num_targets(t)) {
1458 ti = dm_table_get_target(t, i++);
1460 if (!ti->num_write_same_bios)
1463 if (!ti->type->iterate_devices ||
1464 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1471 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1472 struct queue_limits *limits)
1477 * Copy table's limits to the DM device's request_queue
1479 q->limits = *limits;
1481 if (!dm_table_supports_discards(t))
1482 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1484 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1486 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1488 if (dm_table_supports_flush(t, REQ_FUA))
1491 blk_queue_flush(q, flush);
1493 if (!dm_table_discard_zeroes_data(t))
1494 q->limits.discard_zeroes_data = 0;
1496 /* Ensure that all underlying devices are non-rotational. */
1497 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1498 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1500 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1502 if (!dm_table_supports_write_same(t))
1503 q->limits.max_write_same_sectors = 0;
1505 dm_table_set_integrity(t);
1508 * Determine whether or not this queue's I/O timings contribute
1509 * to the entropy pool, Only request-based targets use this.
1510 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1513 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1514 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1517 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1518 * visible to other CPUs because, once the flag is set, incoming bios
1519 * are processed by request-based dm, which refers to the queue
1521 * Until the flag set, bios are passed to bio-based dm and queued to
1522 * md->deferred where queue settings are not needed yet.
1523 * Those bios are passed to request-based dm at the resume time.
1526 if (dm_table_request_based(t))
1527 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1530 unsigned int dm_table_get_num_targets(struct dm_table *t)
1532 return t->num_targets;
1535 struct list_head *dm_table_get_devices(struct dm_table *t)
1540 fmode_t dm_table_get_mode(struct dm_table *t)
1544 EXPORT_SYMBOL(dm_table_get_mode);
1546 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1548 int i = t->num_targets;
1549 struct dm_target *ti = t->targets;
1553 if (ti->type->postsuspend)
1554 ti->type->postsuspend(ti);
1555 } else if (ti->type->presuspend)
1556 ti->type->presuspend(ti);
1562 void dm_table_presuspend_targets(struct dm_table *t)
1567 suspend_targets(t, 0);
1570 void dm_table_postsuspend_targets(struct dm_table *t)
1575 suspend_targets(t, 1);
1578 int dm_table_resume_targets(struct dm_table *t)
1582 for (i = 0; i < t->num_targets; i++) {
1583 struct dm_target *ti = t->targets + i;
1585 if (!ti->type->preresume)
1588 r = ti->type->preresume(ti);
1593 for (i = 0; i < t->num_targets; i++) {
1594 struct dm_target *ti = t->targets + i;
1596 if (ti->type->resume)
1597 ti->type->resume(ti);
1603 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1605 list_add(&cb->list, &t->target_callbacks);
1607 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1609 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1611 struct dm_dev_internal *dd;
1612 struct list_head *devices = dm_table_get_devices(t);
1613 struct dm_target_callbacks *cb;
1616 list_for_each_entry(dd, devices, list) {
1617 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1618 char b[BDEVNAME_SIZE];
1621 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1623 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1624 dm_device_name(t->md),
1625 bdevname(dd->dm_dev.bdev, b));
1628 list_for_each_entry(cb, &t->target_callbacks, list)
1629 if (cb->congested_fn)
1630 r |= cb->congested_fn(cb, bdi_bits);
1635 int dm_table_any_busy_target(struct dm_table *t)
1638 struct dm_target *ti;
1640 for (i = 0; i < t->num_targets; i++) {
1641 ti = t->targets + i;
1642 if (ti->type->busy && ti->type->busy(ti))
1649 struct mapped_device *dm_table_get_md(struct dm_table *t)
1653 EXPORT_SYMBOL(dm_table_get_md);
1655 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1656 sector_t start, sector_t len, void *data)
1658 struct request_queue *q = bdev_get_queue(dev->bdev);
1660 return q && blk_queue_discard(q);
1663 bool dm_table_supports_discards(struct dm_table *t)
1665 struct dm_target *ti;
1669 * Unless any target used by the table set discards_supported,
1670 * require at least one underlying device to support discards.
1671 * t->devices includes internal dm devices such as mirror logs
1672 * so we need to use iterate_devices here, which targets
1673 * supporting discard selectively must provide.
1675 while (i < dm_table_get_num_targets(t)) {
1676 ti = dm_table_get_target(t, i++);
1678 if (!ti->num_discard_bios)
1681 if (ti->discards_supported)
1684 if (ti->type->iterate_devices &&
1685 ti->type->iterate_devices(ti, device_discard_capable, NULL))