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);
448 dd->dm_dev.mode |= new_mode;
449 close_dev(&dd_old, md);
455 * Add a device to the list, or just increment the usage count if
456 * it's already present.
458 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
459 struct dm_dev **result)
462 dev_t uninitialized_var(dev);
463 struct dm_dev_internal *dd;
464 unsigned int major, minor;
465 struct dm_table *t = ti->table;
470 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
471 /* Extract the major/minor numbers */
472 dev = MKDEV(major, minor);
473 if (MAJOR(dev) != major || MINOR(dev) != minor)
476 /* convert the path to a device */
477 struct block_device *bdev = lookup_bdev(path);
480 return PTR_ERR(bdev);
485 dd = find_device(&t->devices, dev);
487 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
491 dd->dm_dev.mode = mode;
492 dd->dm_dev.bdev = NULL;
494 if ((r = open_dev(dd, dev, t->md))) {
499 format_dev_t(dd->dm_dev.name, dev);
501 atomic_set(&dd->count, 0);
502 list_add(&dd->list, &t->devices);
504 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
505 r = upgrade_mode(dd, mode, t->md);
509 atomic_inc(&dd->count);
511 *result = &dd->dm_dev;
514 EXPORT_SYMBOL(dm_get_device);
516 int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
517 sector_t start, sector_t len, void *data)
519 struct queue_limits *limits = data;
520 struct block_device *bdev = dev->bdev;
521 struct request_queue *q = bdev_get_queue(bdev);
522 char b[BDEVNAME_SIZE];
525 DMWARN("%s: Cannot set limits for nonexistent device %s",
526 dm_device_name(ti->table->md), bdevname(bdev, b));
530 if (bdev_stack_limits(limits, bdev, start) < 0)
531 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
532 "physical_block_size=%u, logical_block_size=%u, "
533 "alignment_offset=%u, start=%llu",
534 dm_device_name(ti->table->md), bdevname(bdev, b),
535 q->limits.physical_block_size,
536 q->limits.logical_block_size,
537 q->limits.alignment_offset,
538 (unsigned long long) start << SECTOR_SHIFT);
541 * Check if merge fn is supported.
542 * If not we'll force DM to use PAGE_SIZE or
543 * smaller I/O, just to be safe.
545 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
546 blk_limits_max_hw_sectors(limits,
547 (unsigned int) (PAGE_SIZE >> 9));
550 EXPORT_SYMBOL_GPL(dm_set_device_limits);
553 * Decrement a device's use count and remove it if necessary.
555 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
557 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
560 if (atomic_dec_and_test(&dd->count)) {
561 close_dev(dd, ti->table->md);
566 EXPORT_SYMBOL(dm_put_device);
569 * Checks to see if the target joins onto the end of the table.
571 static int adjoin(struct dm_table *table, struct dm_target *ti)
573 struct dm_target *prev;
575 if (!table->num_targets)
578 prev = &table->targets[table->num_targets - 1];
579 return (ti->begin == (prev->begin + prev->len));
583 * Used to dynamically allocate the arg array.
585 static char **realloc_argv(unsigned *array_size, char **old_argv)
590 new_size = *array_size ? *array_size * 2 : 64;
591 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
593 memcpy(argv, old_argv, *array_size * sizeof(*argv));
594 *array_size = new_size;
602 * Destructively splits up the argument list to pass to ctr.
604 int dm_split_args(int *argc, char ***argvp, char *input)
606 char *start, *end = input, *out, **argv = NULL;
607 unsigned array_size = 0;
616 argv = realloc_argv(&array_size, argv);
621 /* Skip whitespace */
622 start = skip_spaces(end);
625 break; /* success, we hit the end */
627 /* 'out' is used to remove any back-quotes */
630 /* Everything apart from '\0' can be quoted */
631 if (*end == '\\' && *(end + 1)) {
638 break; /* end of token */
643 /* have we already filled the array ? */
644 if ((*argc + 1) > array_size) {
645 argv = realloc_argv(&array_size, argv);
650 /* we know this is whitespace */
654 /* terminate the string and put it in the array */
665 * Impose necessary and sufficient conditions on a devices's table such
666 * that any incoming bio which respects its logical_block_size can be
667 * processed successfully. If it falls across the boundary between
668 * two or more targets, the size of each piece it gets split into must
669 * be compatible with the logical_block_size of the target processing it.
671 static int validate_hardware_logical_block_alignment(struct dm_table *table,
672 struct queue_limits *limits)
675 * This function uses arithmetic modulo the logical_block_size
676 * (in units of 512-byte sectors).
678 unsigned short device_logical_block_size_sects =
679 limits->logical_block_size >> SECTOR_SHIFT;
682 * Offset of the start of the next table entry, mod logical_block_size.
684 unsigned short next_target_start = 0;
687 * Given an aligned bio that extends beyond the end of a
688 * target, how many sectors must the next target handle?
690 unsigned short remaining = 0;
692 struct dm_target *uninitialized_var(ti);
693 struct queue_limits ti_limits;
697 * Check each entry in the table in turn.
699 while (i < dm_table_get_num_targets(table)) {
700 ti = dm_table_get_target(table, i++);
702 blk_set_stacking_limits(&ti_limits);
704 /* combine all target devices' limits */
705 if (ti->type->iterate_devices)
706 ti->type->iterate_devices(ti, dm_set_device_limits,
710 * If the remaining sectors fall entirely within this
711 * table entry are they compatible with its logical_block_size?
713 if (remaining < ti->len &&
714 remaining & ((ti_limits.logical_block_size >>
719 (unsigned short) ((next_target_start + ti->len) &
720 (device_logical_block_size_sects - 1));
721 remaining = next_target_start ?
722 device_logical_block_size_sects - next_target_start : 0;
726 DMWARN("%s: table line %u (start sect %llu len %llu) "
727 "not aligned to h/w logical block size %u",
728 dm_device_name(table->md), i,
729 (unsigned long long) ti->begin,
730 (unsigned long long) ti->len,
731 limits->logical_block_size);
738 int dm_table_add_target(struct dm_table *t, const char *type,
739 sector_t start, sector_t len, char *params)
741 int r = -EINVAL, argc;
743 struct dm_target *tgt;
746 DMERR("%s: target type %s must appear alone in table",
747 dm_device_name(t->md), t->targets->type->name);
751 if ((r = check_space(t)))
754 tgt = t->targets + t->num_targets;
755 memset(tgt, 0, sizeof(*tgt));
758 DMERR("%s: zero-length target", dm_device_name(t->md));
762 tgt->type = dm_get_target_type(type);
764 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
769 if (dm_target_needs_singleton(tgt->type)) {
770 if (t->num_targets) {
771 DMERR("%s: target type %s must appear alone in table",
772 dm_device_name(t->md), type);
778 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
779 DMERR("%s: target type %s may not be included in read-only tables",
780 dm_device_name(t->md), type);
784 if (t->immutable_target_type) {
785 if (t->immutable_target_type != tgt->type) {
786 DMERR("%s: immutable target type %s cannot be mixed with other target types",
787 dm_device_name(t->md), t->immutable_target_type->name);
790 } else if (dm_target_is_immutable(tgt->type)) {
791 if (t->num_targets) {
792 DMERR("%s: immutable target type %s cannot be mixed with other target types",
793 dm_device_name(t->md), tgt->type->name);
796 t->immutable_target_type = tgt->type;
802 tgt->error = "Unknown error";
805 * Does this target adjoin the previous one ?
807 if (!adjoin(t, tgt)) {
808 tgt->error = "Gap in table";
813 r = dm_split_args(&argc, &argv, params);
815 tgt->error = "couldn't split parameters (insufficient memory)";
819 r = tgt->type->ctr(tgt, argc, argv);
824 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
826 if (!tgt->num_discard_requests && tgt->discards_supported)
827 DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
828 dm_device_name(t->md), type);
833 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
834 dm_put_target_type(tgt->type);
839 * Target argument parsing helpers.
841 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
842 unsigned *value, char **error, unsigned grouped)
844 const char *arg_str = dm_shift_arg(arg_set);
848 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
849 (*value < arg->min) ||
850 (*value > arg->max) ||
851 (grouped && arg_set->argc < *value)) {
859 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
860 unsigned *value, char **error)
862 return validate_next_arg(arg, arg_set, value, error, 0);
864 EXPORT_SYMBOL(dm_read_arg);
866 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
867 unsigned *value, char **error)
869 return validate_next_arg(arg, arg_set, value, error, 1);
871 EXPORT_SYMBOL(dm_read_arg_group);
873 const char *dm_shift_arg(struct dm_arg_set *as)
886 EXPORT_SYMBOL(dm_shift_arg);
888 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
890 BUG_ON(as->argc < num_args);
891 as->argc -= num_args;
892 as->argv += num_args;
894 EXPORT_SYMBOL(dm_consume_args);
896 static int dm_table_set_type(struct dm_table *t)
899 unsigned bio_based = 0, request_based = 0;
900 struct dm_target *tgt;
901 struct dm_dev_internal *dd;
902 struct list_head *devices;
904 for (i = 0; i < t->num_targets; i++) {
905 tgt = t->targets + i;
906 if (dm_target_request_based(tgt))
911 if (bio_based && request_based) {
912 DMWARN("Inconsistent table: different target types"
913 " can't be mixed up");
919 /* We must use this table as bio-based */
920 t->type = DM_TYPE_BIO_BASED;
924 BUG_ON(!request_based); /* No targets in this table */
926 /* Non-request-stackable devices can't be used for request-based dm */
927 devices = dm_table_get_devices(t);
928 list_for_each_entry(dd, devices, list) {
929 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
930 DMWARN("table load rejected: including"
931 " non-request-stackable devices");
937 * Request-based dm supports only tables that have a single target now.
938 * To support multiple targets, request splitting support is needed,
939 * and that needs lots of changes in the block-layer.
940 * (e.g. request completion process for partial completion.)
942 if (t->num_targets > 1) {
943 DMWARN("Request-based dm doesn't support multiple targets yet");
947 t->type = DM_TYPE_REQUEST_BASED;
952 unsigned dm_table_get_type(struct dm_table *t)
957 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
959 return t->immutable_target_type;
962 bool dm_table_request_based(struct dm_table *t)
964 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
967 int dm_table_alloc_md_mempools(struct dm_table *t)
969 unsigned type = dm_table_get_type(t);
970 unsigned per_bio_data_size = 0;
971 struct dm_target *tgt;
974 if (unlikely(type == DM_TYPE_NONE)) {
975 DMWARN("no table type is set, can't allocate mempools");
979 if (type == DM_TYPE_BIO_BASED)
980 for (i = 0; i < t->num_targets; i++) {
981 tgt = t->targets + i;
982 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
985 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported, per_bio_data_size);
992 void dm_table_free_md_mempools(struct dm_table *t)
994 dm_free_md_mempools(t->mempools);
998 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1003 static int setup_indexes(struct dm_table *t)
1006 unsigned int total = 0;
1009 /* allocate the space for *all* the indexes */
1010 for (i = t->depth - 2; i >= 0; i--) {
1011 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1012 total += t->counts[i];
1015 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1019 /* set up internal nodes, bottom-up */
1020 for (i = t->depth - 2; i >= 0; i--) {
1021 t->index[i] = indexes;
1022 indexes += (KEYS_PER_NODE * t->counts[i]);
1023 setup_btree_index(i, t);
1030 * Builds the btree to index the map.
1032 static int dm_table_build_index(struct dm_table *t)
1035 unsigned int leaf_nodes;
1037 /* how many indexes will the btree have ? */
1038 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1039 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1041 /* leaf layer has already been set up */
1042 t->counts[t->depth - 1] = leaf_nodes;
1043 t->index[t->depth - 1] = t->highs;
1046 r = setup_indexes(t);
1052 * Get a disk whose integrity profile reflects the table's profile.
1053 * If %match_all is true, all devices' profiles must match.
1054 * If %match_all is false, all devices must at least have an
1055 * allocated integrity profile; but uninitialized is ok.
1056 * Returns NULL if integrity support was inconsistent or unavailable.
1058 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1061 struct list_head *devices = dm_table_get_devices(t);
1062 struct dm_dev_internal *dd = NULL;
1063 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1065 list_for_each_entry(dd, devices, list) {
1066 template_disk = dd->dm_dev.bdev->bd_disk;
1067 if (!blk_get_integrity(template_disk))
1069 if (!match_all && !blk_integrity_is_initialized(template_disk))
1070 continue; /* skip uninitialized profiles */
1071 else if (prev_disk &&
1072 blk_integrity_compare(prev_disk, template_disk) < 0)
1074 prev_disk = template_disk;
1077 return template_disk;
1081 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1082 dm_device_name(t->md),
1083 prev_disk->disk_name,
1084 template_disk->disk_name);
1089 * Register the mapped device for blk_integrity support if
1090 * the underlying devices have an integrity profile. But all devices
1091 * may not have matching profiles (checking all devices isn't reliable
1092 * during table load because this table may use other DM device(s) which
1093 * must be resumed before they will have an initialized integity profile).
1094 * Stacked DM devices force a 2 stage integrity profile validation:
1095 * 1 - during load, validate all initialized integrity profiles match
1096 * 2 - during resume, validate all integrity profiles match
1098 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1100 struct gendisk *template_disk = NULL;
1102 template_disk = dm_table_get_integrity_disk(t, false);
1106 if (!blk_integrity_is_initialized(dm_disk(md))) {
1107 t->integrity_supported = 1;
1108 return blk_integrity_register(dm_disk(md), NULL);
1112 * If DM device already has an initalized integrity
1113 * profile the new profile should not conflict.
1115 if (blk_integrity_is_initialized(template_disk) &&
1116 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1117 DMWARN("%s: conflict with existing integrity profile: "
1118 "%s profile mismatch",
1119 dm_device_name(t->md),
1120 template_disk->disk_name);
1124 /* Preserve existing initialized integrity profile */
1125 t->integrity_supported = 1;
1130 * Prepares the table for use by building the indices,
1131 * setting the type, and allocating mempools.
1133 int dm_table_complete(struct dm_table *t)
1137 r = dm_table_set_type(t);
1139 DMERR("unable to set table type");
1143 r = dm_table_build_index(t);
1145 DMERR("unable to build btrees");
1149 r = dm_table_prealloc_integrity(t, t->md);
1151 DMERR("could not register integrity profile.");
1155 r = dm_table_alloc_md_mempools(t);
1157 DMERR("unable to allocate mempools");
1162 static DEFINE_MUTEX(_event_lock);
1163 void dm_table_event_callback(struct dm_table *t,
1164 void (*fn)(void *), void *context)
1166 mutex_lock(&_event_lock);
1168 t->event_context = context;
1169 mutex_unlock(&_event_lock);
1172 void dm_table_event(struct dm_table *t)
1175 * You can no longer call dm_table_event() from interrupt
1176 * context, use a bottom half instead.
1178 BUG_ON(in_interrupt());
1180 mutex_lock(&_event_lock);
1182 t->event_fn(t->event_context);
1183 mutex_unlock(&_event_lock);
1185 EXPORT_SYMBOL(dm_table_event);
1187 sector_t dm_table_get_size(struct dm_table *t)
1189 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1191 EXPORT_SYMBOL(dm_table_get_size);
1193 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1195 if (index >= t->num_targets)
1198 return t->targets + index;
1202 * Search the btree for the correct target.
1204 * Caller should check returned pointer with dm_target_is_valid()
1205 * to trap I/O beyond end of device.
1207 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1209 unsigned int l, n = 0, k = 0;
1212 for (l = 0; l < t->depth; l++) {
1213 n = get_child(n, k);
1214 node = get_node(t, l, n);
1216 for (k = 0; k < KEYS_PER_NODE; k++)
1217 if (node[k] >= sector)
1221 return &t->targets[(KEYS_PER_NODE * n) + k];
1224 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1225 sector_t start, sector_t len, void *data)
1227 unsigned *num_devices = data;
1235 * Check whether a table has no data devices attached using each
1236 * target's iterate_devices method.
1237 * Returns false if the result is unknown because a target doesn't
1238 * support iterate_devices.
1240 bool dm_table_has_no_data_devices(struct dm_table *table)
1242 struct dm_target *uninitialized_var(ti);
1243 unsigned i = 0, num_devices = 0;
1245 while (i < dm_table_get_num_targets(table)) {
1246 ti = dm_table_get_target(table, i++);
1248 if (!ti->type->iterate_devices)
1251 ti->type->iterate_devices(ti, count_device, &num_devices);
1260 * Establish the new table's queue_limits and validate them.
1262 int dm_calculate_queue_limits(struct dm_table *table,
1263 struct queue_limits *limits)
1265 struct dm_target *uninitialized_var(ti);
1266 struct queue_limits ti_limits;
1269 blk_set_stacking_limits(limits);
1271 while (i < dm_table_get_num_targets(table)) {
1272 blk_set_stacking_limits(&ti_limits);
1274 ti = dm_table_get_target(table, i++);
1276 if (!ti->type->iterate_devices)
1277 goto combine_limits;
1280 * Combine queue limits of all the devices this target uses.
1282 ti->type->iterate_devices(ti, dm_set_device_limits,
1285 /* Set I/O hints portion of queue limits */
1286 if (ti->type->io_hints)
1287 ti->type->io_hints(ti, &ti_limits);
1290 * Check each device area is consistent with the target's
1291 * overall queue limits.
1293 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1299 * Merge this target's queue limits into the overall limits
1302 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1303 DMWARN("%s: adding target device "
1304 "(start sect %llu len %llu) "
1305 "caused an alignment inconsistency",
1306 dm_device_name(table->md),
1307 (unsigned long long) ti->begin,
1308 (unsigned long long) ti->len);
1311 return validate_hardware_logical_block_alignment(table, limits);
1315 * Set the integrity profile for this device if all devices used have
1316 * matching profiles. We're quite deep in the resume path but still
1317 * don't know if all devices (particularly DM devices this device
1318 * may be stacked on) have matching profiles. Even if the profiles
1319 * don't match we have no way to fail (to resume) at this point.
1321 static void dm_table_set_integrity(struct dm_table *t)
1323 struct gendisk *template_disk = NULL;
1325 if (!blk_get_integrity(dm_disk(t->md)))
1328 template_disk = dm_table_get_integrity_disk(t, true);
1330 blk_integrity_register(dm_disk(t->md),
1331 blk_get_integrity(template_disk));
1332 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1333 DMWARN("%s: device no longer has a valid integrity profile",
1334 dm_device_name(t->md));
1336 DMWARN("%s: unable to establish an integrity profile",
1337 dm_device_name(t->md));
1340 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1341 sector_t start, sector_t len, void *data)
1343 unsigned flush = (*(unsigned *)data);
1344 struct request_queue *q = bdev_get_queue(dev->bdev);
1346 return q && (q->flush_flags & flush);
1349 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1351 struct dm_target *ti;
1355 * Require at least one underlying device to support flushes.
1356 * t->devices includes internal dm devices such as mirror logs
1357 * so we need to use iterate_devices here, which targets
1358 * supporting flushes must provide.
1360 while (i < dm_table_get_num_targets(t)) {
1361 ti = dm_table_get_target(t, i++);
1363 if (!ti->num_flush_requests)
1366 if (ti->flush_supported)
1369 if (ti->type->iterate_devices &&
1370 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1377 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1379 struct dm_target *ti;
1382 /* Ensure that all targets supports discard_zeroes_data. */
1383 while (i < dm_table_get_num_targets(t)) {
1384 ti = dm_table_get_target(t, i++);
1386 if (ti->discard_zeroes_data_unsupported)
1393 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1394 sector_t start, sector_t len, void *data)
1396 struct request_queue *q = bdev_get_queue(dev->bdev);
1398 return q && blk_queue_nonrot(q);
1401 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1402 sector_t start, sector_t len, void *data)
1404 struct request_queue *q = bdev_get_queue(dev->bdev);
1406 return q && !blk_queue_add_random(q);
1409 static bool dm_table_all_devices_attribute(struct dm_table *t,
1410 iterate_devices_callout_fn func)
1412 struct dm_target *ti;
1415 while (i < dm_table_get_num_targets(t)) {
1416 ti = dm_table_get_target(t, i++);
1418 if (!ti->type->iterate_devices ||
1419 !ti->type->iterate_devices(ti, func, NULL))
1426 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1427 sector_t start, sector_t len, void *data)
1429 struct request_queue *q = bdev_get_queue(dev->bdev);
1431 return q && !q->limits.max_write_same_sectors;
1434 static bool dm_table_supports_write_same(struct dm_table *t)
1436 struct dm_target *ti;
1439 while (i < dm_table_get_num_targets(t)) {
1440 ti = dm_table_get_target(t, i++);
1442 if (!ti->num_write_same_requests)
1445 if (!ti->type->iterate_devices ||
1446 !ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1453 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1454 struct queue_limits *limits)
1459 * Copy table's limits to the DM device's request_queue
1461 q->limits = *limits;
1463 if (!dm_table_supports_discards(t))
1464 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1466 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1468 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1470 if (dm_table_supports_flush(t, REQ_FUA))
1473 blk_queue_flush(q, flush);
1475 if (!dm_table_discard_zeroes_data(t))
1476 q->limits.discard_zeroes_data = 0;
1478 /* Ensure that all underlying devices are non-rotational. */
1479 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1480 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1482 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1484 if (!dm_table_supports_write_same(t))
1485 q->limits.max_write_same_sectors = 0;
1487 dm_table_set_integrity(t);
1490 * Determine whether or not this queue's I/O timings contribute
1491 * to the entropy pool, Only request-based targets use this.
1492 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1495 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1496 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1499 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1500 * visible to other CPUs because, once the flag is set, incoming bios
1501 * are processed by request-based dm, which refers to the queue
1503 * Until the flag set, bios are passed to bio-based dm and queued to
1504 * md->deferred where queue settings are not needed yet.
1505 * Those bios are passed to request-based dm at the resume time.
1508 if (dm_table_request_based(t))
1509 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1512 unsigned int dm_table_get_num_targets(struct dm_table *t)
1514 return t->num_targets;
1517 struct list_head *dm_table_get_devices(struct dm_table *t)
1522 fmode_t dm_table_get_mode(struct dm_table *t)
1526 EXPORT_SYMBOL(dm_table_get_mode);
1528 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1530 int i = t->num_targets;
1531 struct dm_target *ti = t->targets;
1535 if (ti->type->postsuspend)
1536 ti->type->postsuspend(ti);
1537 } else if (ti->type->presuspend)
1538 ti->type->presuspend(ti);
1544 void dm_table_presuspend_targets(struct dm_table *t)
1549 suspend_targets(t, 0);
1552 void dm_table_postsuspend_targets(struct dm_table *t)
1557 suspend_targets(t, 1);
1560 int dm_table_resume_targets(struct dm_table *t)
1564 for (i = 0; i < t->num_targets; i++) {
1565 struct dm_target *ti = t->targets + i;
1567 if (!ti->type->preresume)
1570 r = ti->type->preresume(ti);
1575 for (i = 0; i < t->num_targets; i++) {
1576 struct dm_target *ti = t->targets + i;
1578 if (ti->type->resume)
1579 ti->type->resume(ti);
1585 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1587 list_add(&cb->list, &t->target_callbacks);
1589 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1591 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1593 struct dm_dev_internal *dd;
1594 struct list_head *devices = dm_table_get_devices(t);
1595 struct dm_target_callbacks *cb;
1598 list_for_each_entry(dd, devices, list) {
1599 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1600 char b[BDEVNAME_SIZE];
1603 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1605 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1606 dm_device_name(t->md),
1607 bdevname(dd->dm_dev.bdev, b));
1610 list_for_each_entry(cb, &t->target_callbacks, list)
1611 if (cb->congested_fn)
1612 r |= cb->congested_fn(cb, bdi_bits);
1617 int dm_table_any_busy_target(struct dm_table *t)
1620 struct dm_target *ti;
1622 for (i = 0; i < t->num_targets; i++) {
1623 ti = t->targets + i;
1624 if (ti->type->busy && ti->type->busy(ti))
1631 struct mapped_device *dm_table_get_md(struct dm_table *t)
1635 EXPORT_SYMBOL(dm_table_get_md);
1637 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1638 sector_t start, sector_t len, void *data)
1640 struct request_queue *q = bdev_get_queue(dev->bdev);
1642 return q && blk_queue_discard(q);
1645 bool dm_table_supports_discards(struct dm_table *t)
1647 struct dm_target *ti;
1651 * Unless any target used by the table set discards_supported,
1652 * require at least one underlying device to support discards.
1653 * t->devices includes internal dm devices such as mirror logs
1654 * so we need to use iterate_devices here, which targets
1655 * supporting discard selectively must provide.
1657 while (i < dm_table_get_num_targets(t)) {
1658 ti = dm_table_get_target(t, i++);
1660 if (!ti->num_discard_requests)
1663 if (ti->discards_supported)
1666 if (ti->type->iterate_devices &&
1667 ti->type->iterate_devices(ti, device_discard_capable, NULL))