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)) {
229 static void free_devices(struct list_head *devices)
231 struct list_head *tmp, *next;
233 list_for_each_safe(tmp, next, devices) {
234 struct dm_dev_internal *dd =
235 list_entry(tmp, struct dm_dev_internal, list);
236 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
242 void dm_table_destroy(struct dm_table *t)
249 while (atomic_read(&t->holders))
253 /* free the indexes */
255 vfree(t->index[t->depth - 2]);
257 /* free the targets */
258 for (i = 0; i < t->num_targets; i++) {
259 struct dm_target *tgt = t->targets + i;
264 dm_put_target_type(tgt->type);
269 /* free the device list */
270 free_devices(&t->devices);
272 dm_free_md_mempools(t->mempools);
277 void dm_table_get(struct dm_table *t)
279 atomic_inc(&t->holders);
281 EXPORT_SYMBOL(dm_table_get);
283 void dm_table_put(struct dm_table *t)
288 smp_mb__before_atomic_dec();
289 atomic_dec(&t->holders);
291 EXPORT_SYMBOL(dm_table_put);
294 * Checks to see if we need to extend highs or targets.
296 static inline int check_space(struct dm_table *t)
298 if (t->num_targets >= t->num_allocated)
299 return alloc_targets(t, t->num_allocated * 2);
305 * See if we've already got a device in the list.
307 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
309 struct dm_dev_internal *dd;
311 list_for_each_entry (dd, l, list)
312 if (dd->dm_dev.bdev->bd_dev == dev)
319 * Open a device so we can use it as a map destination.
321 static int open_dev(struct dm_dev_internal *d, dev_t dev,
322 struct mapped_device *md)
324 static char *_claim_ptr = "I belong to device-mapper";
325 struct block_device *bdev;
329 BUG_ON(d->dm_dev.bdev);
331 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
333 return PTR_ERR(bdev);
335 r = bd_link_disk_holder(bdev, dm_disk(md));
337 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
341 d->dm_dev.bdev = bdev;
346 * Close a device that we've been using.
348 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
353 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
354 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
355 d->dm_dev.bdev = NULL;
359 * If possible, this checks an area of a destination device is invalid.
361 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
362 sector_t start, sector_t len, void *data)
364 struct request_queue *q;
365 struct queue_limits *limits = data;
366 struct block_device *bdev = dev->bdev;
368 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
369 unsigned short logical_block_size_sectors =
370 limits->logical_block_size >> SECTOR_SHIFT;
371 char b[BDEVNAME_SIZE];
374 * Some devices exist without request functions,
375 * such as loop devices not yet bound to backing files.
376 * Forbid the use of such devices.
378 q = bdev_get_queue(bdev);
379 if (!q || !q->make_request_fn) {
380 DMWARN("%s: %s is not yet initialised: "
381 "start=%llu, len=%llu, dev_size=%llu",
382 dm_device_name(ti->table->md), bdevname(bdev, b),
383 (unsigned long long)start,
384 (unsigned long long)len,
385 (unsigned long long)dev_size);
392 if ((start >= dev_size) || (start + len > dev_size)) {
393 DMWARN("%s: %s too small for target: "
394 "start=%llu, len=%llu, dev_size=%llu",
395 dm_device_name(ti->table->md), bdevname(bdev, b),
396 (unsigned long long)start,
397 (unsigned long long)len,
398 (unsigned long long)dev_size);
402 if (logical_block_size_sectors <= 1)
405 if (start & (logical_block_size_sectors - 1)) {
406 DMWARN("%s: start=%llu not aligned to h/w "
407 "logical block size %u of %s",
408 dm_device_name(ti->table->md),
409 (unsigned long long)start,
410 limits->logical_block_size, bdevname(bdev, b));
414 if (len & (logical_block_size_sectors - 1)) {
415 DMWARN("%s: len=%llu not aligned to h/w "
416 "logical block size %u of %s",
417 dm_device_name(ti->table->md),
418 (unsigned long long)len,
419 limits->logical_block_size, bdevname(bdev, b));
427 * This upgrades the mode on an already open dm_dev, being
428 * careful to leave things as they were if we fail to reopen the
429 * device and not to touch the existing bdev field in case
430 * it is accessed concurrently inside dm_table_any_congested().
432 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
433 struct mapped_device *md)
436 struct dm_dev_internal dd_new, dd_old;
438 dd_new = dd_old = *dd;
440 dd_new.dm_dev.mode |= new_mode;
441 dd_new.dm_dev.bdev = NULL;
443 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
447 dd->dm_dev.mode |= new_mode;
448 close_dev(&dd_old, md);
454 * Add a device to the list, or just increment the usage count if
455 * it's already present.
457 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
458 struct dm_dev **result)
461 dev_t uninitialized_var(dev);
462 struct dm_dev_internal *dd;
463 unsigned int major, minor;
464 struct dm_table *t = ti->table;
469 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
470 /* Extract the major/minor numbers */
471 dev = MKDEV(major, minor);
472 if (MAJOR(dev) != major || MINOR(dev) != minor)
475 /* convert the path to a device */
476 struct block_device *bdev = lookup_bdev(path);
479 return PTR_ERR(bdev);
484 dd = find_device(&t->devices, dev);
486 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
490 dd->dm_dev.mode = mode;
491 dd->dm_dev.bdev = NULL;
493 if ((r = open_dev(dd, dev, t->md))) {
498 format_dev_t(dd->dm_dev.name, dev);
500 atomic_set(&dd->count, 0);
501 list_add(&dd->list, &t->devices);
503 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
504 r = upgrade_mode(dd, mode, t->md);
508 atomic_inc(&dd->count);
510 *result = &dd->dm_dev;
513 EXPORT_SYMBOL(dm_get_device);
515 int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
516 sector_t start, sector_t len, void *data)
518 struct queue_limits *limits = data;
519 struct block_device *bdev = dev->bdev;
520 struct request_queue *q = bdev_get_queue(bdev);
521 char b[BDEVNAME_SIZE];
524 DMWARN("%s: Cannot set limits for nonexistent device %s",
525 dm_device_name(ti->table->md), bdevname(bdev, b));
529 if (bdev_stack_limits(limits, bdev, start) < 0)
530 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
531 "physical_block_size=%u, logical_block_size=%u, "
532 "alignment_offset=%u, start=%llu",
533 dm_device_name(ti->table->md), bdevname(bdev, b),
534 q->limits.physical_block_size,
535 q->limits.logical_block_size,
536 q->limits.alignment_offset,
537 (unsigned long long) start << SECTOR_SHIFT);
540 * Check if merge fn is supported.
541 * If not we'll force DM to use PAGE_SIZE or
542 * smaller I/O, just to be safe.
544 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
545 blk_limits_max_hw_sectors(limits,
546 (unsigned int) (PAGE_SIZE >> 9));
549 EXPORT_SYMBOL_GPL(dm_set_device_limits);
552 * Decrement a device's use count and remove it if necessary.
554 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
556 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
559 if (atomic_dec_and_test(&dd->count)) {
560 close_dev(dd, ti->table->md);
565 EXPORT_SYMBOL(dm_put_device);
568 * Checks to see if the target joins onto the end of the table.
570 static int adjoin(struct dm_table *table, struct dm_target *ti)
572 struct dm_target *prev;
574 if (!table->num_targets)
577 prev = &table->targets[table->num_targets - 1];
578 return (ti->begin == (prev->begin + prev->len));
582 * Used to dynamically allocate the arg array.
584 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
585 * process messages even if some device is suspended. These messages have a
586 * small fixed number of arguments.
588 * On the other hand, dm-switch needs to process bulk data using messages and
589 * excessive use of GFP_NOIO could cause trouble.
591 static char **realloc_argv(unsigned *array_size, char **old_argv)
598 new_size = *array_size * 2;
604 argv = kmalloc(new_size * sizeof(*argv), gfp);
606 memcpy(argv, old_argv, *array_size * sizeof(*argv));
607 *array_size = new_size;
615 * Destructively splits up the argument list to pass to ctr.
617 int dm_split_args(int *argc, char ***argvp, char *input)
619 char *start, *end = input, *out, **argv = NULL;
620 unsigned array_size = 0;
629 argv = realloc_argv(&array_size, argv);
634 /* Skip whitespace */
635 start = skip_spaces(end);
638 break; /* success, we hit the end */
640 /* 'out' is used to remove any back-quotes */
643 /* Everything apart from '\0' can be quoted */
644 if (*end == '\\' && *(end + 1)) {
651 break; /* end of token */
656 /* have we already filled the array ? */
657 if ((*argc + 1) > array_size) {
658 argv = realloc_argv(&array_size, argv);
663 /* we know this is whitespace */
667 /* terminate the string and put it in the array */
678 * Impose necessary and sufficient conditions on a devices's table such
679 * that any incoming bio which respects its logical_block_size can be
680 * processed successfully. If it falls across the boundary between
681 * two or more targets, the size of each piece it gets split into must
682 * be compatible with the logical_block_size of the target processing it.
684 static int validate_hardware_logical_block_alignment(struct dm_table *table,
685 struct queue_limits *limits)
688 * This function uses arithmetic modulo the logical_block_size
689 * (in units of 512-byte sectors).
691 unsigned short device_logical_block_size_sects =
692 limits->logical_block_size >> SECTOR_SHIFT;
695 * Offset of the start of the next table entry, mod logical_block_size.
697 unsigned short next_target_start = 0;
700 * Given an aligned bio that extends beyond the end of a
701 * target, how many sectors must the next target handle?
703 unsigned short remaining = 0;
705 struct dm_target *uninitialized_var(ti);
706 struct queue_limits ti_limits;
710 * Check each entry in the table in turn.
712 while (i < dm_table_get_num_targets(table)) {
713 ti = dm_table_get_target(table, i++);
715 blk_set_stacking_limits(&ti_limits);
717 /* combine all target devices' limits */
718 if (ti->type->iterate_devices)
719 ti->type->iterate_devices(ti, dm_set_device_limits,
723 * If the remaining sectors fall entirely within this
724 * table entry are they compatible with its logical_block_size?
726 if (remaining < ti->len &&
727 remaining & ((ti_limits.logical_block_size >>
732 (unsigned short) ((next_target_start + ti->len) &
733 (device_logical_block_size_sects - 1));
734 remaining = next_target_start ?
735 device_logical_block_size_sects - next_target_start : 0;
739 DMWARN("%s: table line %u (start sect %llu len %llu) "
740 "not aligned to h/w logical block size %u",
741 dm_device_name(table->md), i,
742 (unsigned long long) ti->begin,
743 (unsigned long long) ti->len,
744 limits->logical_block_size);
751 int dm_table_add_target(struct dm_table *t, const char *type,
752 sector_t start, sector_t len, char *params)
754 int r = -EINVAL, argc;
756 struct dm_target *tgt;
759 DMERR("%s: target type %s must appear alone in table",
760 dm_device_name(t->md), t->targets->type->name);
764 if ((r = check_space(t)))
767 tgt = t->targets + t->num_targets;
768 memset(tgt, 0, sizeof(*tgt));
771 DMERR("%s: zero-length target", dm_device_name(t->md));
775 tgt->type = dm_get_target_type(type);
777 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
782 if (dm_target_needs_singleton(tgt->type)) {
783 if (t->num_targets) {
784 DMERR("%s: target type %s must appear alone in table",
785 dm_device_name(t->md), type);
791 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
792 DMERR("%s: target type %s may not be included in read-only tables",
793 dm_device_name(t->md), type);
797 if (t->immutable_target_type) {
798 if (t->immutable_target_type != tgt->type) {
799 DMERR("%s: immutable target type %s cannot be mixed with other target types",
800 dm_device_name(t->md), t->immutable_target_type->name);
803 } else if (dm_target_is_immutable(tgt->type)) {
804 if (t->num_targets) {
805 DMERR("%s: immutable target type %s cannot be mixed with other target types",
806 dm_device_name(t->md), tgt->type->name);
809 t->immutable_target_type = tgt->type;
815 tgt->error = "Unknown error";
818 * Does this target adjoin the previous one ?
820 if (!adjoin(t, tgt)) {
821 tgt->error = "Gap in table";
826 r = dm_split_args(&argc, &argv, params);
828 tgt->error = "couldn't split parameters (insufficient memory)";
832 r = tgt->type->ctr(tgt, argc, argv);
837 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
839 if (!tgt->num_discard_bios && tgt->discards_supported)
840 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
841 dm_device_name(t->md), type);
846 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
847 dm_put_target_type(tgt->type);
852 * Target argument parsing helpers.
854 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
855 unsigned *value, char **error, unsigned grouped)
857 const char *arg_str = dm_shift_arg(arg_set);
861 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
862 (*value < arg->min) ||
863 (*value > arg->max) ||
864 (grouped && arg_set->argc < *value)) {
872 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
873 unsigned *value, char **error)
875 return validate_next_arg(arg, arg_set, value, error, 0);
877 EXPORT_SYMBOL(dm_read_arg);
879 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
880 unsigned *value, char **error)
882 return validate_next_arg(arg, arg_set, value, error, 1);
884 EXPORT_SYMBOL(dm_read_arg_group);
886 const char *dm_shift_arg(struct dm_arg_set *as)
899 EXPORT_SYMBOL(dm_shift_arg);
901 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
903 BUG_ON(as->argc < num_args);
904 as->argc -= num_args;
905 as->argv += num_args;
907 EXPORT_SYMBOL(dm_consume_args);
909 static int dm_table_set_type(struct dm_table *t)
912 unsigned bio_based = 0, request_based = 0;
913 struct dm_target *tgt;
914 struct dm_dev_internal *dd;
915 struct list_head *devices;
917 for (i = 0; i < t->num_targets; i++) {
918 tgt = t->targets + i;
919 if (dm_target_request_based(tgt))
924 if (bio_based && request_based) {
925 DMWARN("Inconsistent table: different target types"
926 " can't be mixed up");
932 /* We must use this table as bio-based */
933 t->type = DM_TYPE_BIO_BASED;
937 BUG_ON(!request_based); /* No targets in this table */
939 /* Non-request-stackable devices can't be used for request-based dm */
940 devices = dm_table_get_devices(t);
941 list_for_each_entry(dd, devices, list) {
942 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
943 DMWARN("table load rejected: including"
944 " non-request-stackable devices");
950 * Request-based dm supports only tables that have a single target now.
951 * To support multiple targets, request splitting support is needed,
952 * and that needs lots of changes in the block-layer.
953 * (e.g. request completion process for partial completion.)
955 if (t->num_targets > 1) {
956 DMWARN("Request-based dm doesn't support multiple targets yet");
960 t->type = DM_TYPE_REQUEST_BASED;
965 unsigned dm_table_get_type(struct dm_table *t)
970 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
972 return t->immutable_target_type;
975 bool dm_table_request_based(struct dm_table *t)
977 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
980 int dm_table_alloc_md_mempools(struct dm_table *t)
982 unsigned type = dm_table_get_type(t);
983 unsigned per_bio_data_size = 0;
984 struct dm_target *tgt;
987 if (unlikely(type == DM_TYPE_NONE)) {
988 DMWARN("no table type is set, can't allocate mempools");
992 if (type == DM_TYPE_BIO_BASED)
993 for (i = 0; i < t->num_targets; i++) {
994 tgt = t->targets + i;
995 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
998 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported, per_bio_data_size);
1005 void dm_table_free_md_mempools(struct dm_table *t)
1007 dm_free_md_mempools(t->mempools);
1011 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1016 static int setup_indexes(struct dm_table *t)
1019 unsigned int total = 0;
1022 /* allocate the space for *all* the indexes */
1023 for (i = t->depth - 2; i >= 0; i--) {
1024 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1025 total += t->counts[i];
1028 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1032 /* set up internal nodes, bottom-up */
1033 for (i = t->depth - 2; i >= 0; i--) {
1034 t->index[i] = indexes;
1035 indexes += (KEYS_PER_NODE * t->counts[i]);
1036 setup_btree_index(i, t);
1043 * Builds the btree to index the map.
1045 static int dm_table_build_index(struct dm_table *t)
1048 unsigned int leaf_nodes;
1050 /* how many indexes will the btree have ? */
1051 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1052 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1054 /* leaf layer has already been set up */
1055 t->counts[t->depth - 1] = leaf_nodes;
1056 t->index[t->depth - 1] = t->highs;
1059 r = setup_indexes(t);
1065 * Get a disk whose integrity profile reflects the table's profile.
1066 * If %match_all is true, all devices' profiles must match.
1067 * If %match_all is false, all devices must at least have an
1068 * allocated integrity profile; but uninitialized is ok.
1069 * Returns NULL if integrity support was inconsistent or unavailable.
1071 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1074 struct list_head *devices = dm_table_get_devices(t);
1075 struct dm_dev_internal *dd = NULL;
1076 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1078 list_for_each_entry(dd, devices, list) {
1079 template_disk = dd->dm_dev.bdev->bd_disk;
1080 if (!blk_get_integrity(template_disk))
1082 if (!match_all && !blk_integrity_is_initialized(template_disk))
1083 continue; /* skip uninitialized profiles */
1084 else if (prev_disk &&
1085 blk_integrity_compare(prev_disk, template_disk) < 0)
1087 prev_disk = template_disk;
1090 return template_disk;
1094 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1095 dm_device_name(t->md),
1096 prev_disk->disk_name,
1097 template_disk->disk_name);
1102 * Register the mapped device for blk_integrity support if
1103 * the underlying devices have an integrity profile. But all devices
1104 * may not have matching profiles (checking all devices isn't reliable
1105 * during table load because this table may use other DM device(s) which
1106 * must be resumed before they will have an initialized integity profile).
1107 * Stacked DM devices force a 2 stage integrity profile validation:
1108 * 1 - during load, validate all initialized integrity profiles match
1109 * 2 - during resume, validate all integrity profiles match
1111 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1113 struct gendisk *template_disk = NULL;
1115 template_disk = dm_table_get_integrity_disk(t, false);
1119 if (!blk_integrity_is_initialized(dm_disk(md))) {
1120 t->integrity_supported = 1;
1121 return blk_integrity_register(dm_disk(md), NULL);
1125 * If DM device already has an initalized integrity
1126 * profile the new profile should not conflict.
1128 if (blk_integrity_is_initialized(template_disk) &&
1129 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1130 DMWARN("%s: conflict with existing integrity profile: "
1131 "%s profile mismatch",
1132 dm_device_name(t->md),
1133 template_disk->disk_name);
1137 /* Preserve existing initialized integrity profile */
1138 t->integrity_supported = 1;
1143 * Prepares the table for use by building the indices,
1144 * setting the type, and allocating mempools.
1146 int dm_table_complete(struct dm_table *t)
1150 r = dm_table_set_type(t);
1152 DMERR("unable to set table type");
1156 r = dm_table_build_index(t);
1158 DMERR("unable to build btrees");
1162 r = dm_table_prealloc_integrity(t, t->md);
1164 DMERR("could not register integrity profile.");
1168 r = dm_table_alloc_md_mempools(t);
1170 DMERR("unable to allocate mempools");
1175 static DEFINE_MUTEX(_event_lock);
1176 void dm_table_event_callback(struct dm_table *t,
1177 void (*fn)(void *), void *context)
1179 mutex_lock(&_event_lock);
1181 t->event_context = context;
1182 mutex_unlock(&_event_lock);
1185 void dm_table_event(struct dm_table *t)
1188 * You can no longer call dm_table_event() from interrupt
1189 * context, use a bottom half instead.
1191 BUG_ON(in_interrupt());
1193 mutex_lock(&_event_lock);
1195 t->event_fn(t->event_context);
1196 mutex_unlock(&_event_lock);
1198 EXPORT_SYMBOL(dm_table_event);
1200 sector_t dm_table_get_size(struct dm_table *t)
1202 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1204 EXPORT_SYMBOL(dm_table_get_size);
1206 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1208 if (index >= t->num_targets)
1211 return t->targets + index;
1215 * Search the btree for the correct target.
1217 * Caller should check returned pointer with dm_target_is_valid()
1218 * to trap I/O beyond end of device.
1220 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1222 unsigned int l, n = 0, k = 0;
1225 for (l = 0; l < t->depth; l++) {
1226 n = get_child(n, k);
1227 node = get_node(t, l, n);
1229 for (k = 0; k < KEYS_PER_NODE; k++)
1230 if (node[k] >= sector)
1234 return &t->targets[(KEYS_PER_NODE * n) + k];
1237 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1238 sector_t start, sector_t len, void *data)
1240 unsigned *num_devices = data;
1248 * Check whether a table has no data devices attached using each
1249 * target's iterate_devices method.
1250 * Returns false if the result is unknown because a target doesn't
1251 * support iterate_devices.
1253 bool dm_table_has_no_data_devices(struct dm_table *table)
1255 struct dm_target *uninitialized_var(ti);
1256 unsigned i = 0, num_devices = 0;
1258 while (i < dm_table_get_num_targets(table)) {
1259 ti = dm_table_get_target(table, i++);
1261 if (!ti->type->iterate_devices)
1264 ti->type->iterate_devices(ti, count_device, &num_devices);
1273 * Establish the new table's queue_limits and validate them.
1275 int dm_calculate_queue_limits(struct dm_table *table,
1276 struct queue_limits *limits)
1278 struct dm_target *uninitialized_var(ti);
1279 struct queue_limits ti_limits;
1282 blk_set_stacking_limits(limits);
1284 while (i < dm_table_get_num_targets(table)) {
1285 blk_set_stacking_limits(&ti_limits);
1287 ti = dm_table_get_target(table, i++);
1289 if (!ti->type->iterate_devices)
1290 goto combine_limits;
1293 * Combine queue limits of all the devices this target uses.
1295 ti->type->iterate_devices(ti, dm_set_device_limits,
1298 /* Set I/O hints portion of queue limits */
1299 if (ti->type->io_hints)
1300 ti->type->io_hints(ti, &ti_limits);
1303 * Check each device area is consistent with the target's
1304 * overall queue limits.
1306 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1312 * Merge this target's queue limits into the overall limits
1315 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1316 DMWARN("%s: adding target device "
1317 "(start sect %llu len %llu) "
1318 "caused an alignment inconsistency",
1319 dm_device_name(table->md),
1320 (unsigned long long) ti->begin,
1321 (unsigned long long) ti->len);
1324 return validate_hardware_logical_block_alignment(table, limits);
1328 * Set the integrity profile for this device if all devices used have
1329 * matching profiles. We're quite deep in the resume path but still
1330 * don't know if all devices (particularly DM devices this device
1331 * may be stacked on) have matching profiles. Even if the profiles
1332 * don't match we have no way to fail (to resume) at this point.
1334 static void dm_table_set_integrity(struct dm_table *t)
1336 struct gendisk *template_disk = NULL;
1338 if (!blk_get_integrity(dm_disk(t->md)))
1341 template_disk = dm_table_get_integrity_disk(t, true);
1343 blk_integrity_register(dm_disk(t->md),
1344 blk_get_integrity(template_disk));
1345 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1346 DMWARN("%s: device no longer has a valid integrity profile",
1347 dm_device_name(t->md));
1349 DMWARN("%s: unable to establish an integrity profile",
1350 dm_device_name(t->md));
1353 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1354 sector_t start, sector_t len, void *data)
1356 unsigned flush = (*(unsigned *)data);
1357 struct request_queue *q = bdev_get_queue(dev->bdev);
1359 return q && (q->flush_flags & flush);
1362 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1364 struct dm_target *ti;
1368 * Require at least one underlying device to support flushes.
1369 * t->devices includes internal dm devices such as mirror logs
1370 * so we need to use iterate_devices here, which targets
1371 * supporting flushes must provide.
1373 while (i < dm_table_get_num_targets(t)) {
1374 ti = dm_table_get_target(t, i++);
1376 if (!ti->num_flush_bios)
1379 if (ti->flush_supported)
1382 if (ti->type->iterate_devices &&
1383 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1390 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1392 struct dm_target *ti;
1395 /* Ensure that all targets supports discard_zeroes_data. */
1396 while (i < dm_table_get_num_targets(t)) {
1397 ti = dm_table_get_target(t, i++);
1399 if (ti->discard_zeroes_data_unsupported)
1406 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1407 sector_t start, sector_t len, void *data)
1409 struct request_queue *q = bdev_get_queue(dev->bdev);
1411 return q && blk_queue_nonrot(q);
1414 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1415 sector_t start, sector_t len, void *data)
1417 struct request_queue *q = bdev_get_queue(dev->bdev);
1419 return q && !blk_queue_add_random(q);
1422 static bool dm_table_all_devices_attribute(struct dm_table *t,
1423 iterate_devices_callout_fn func)
1425 struct dm_target *ti;
1428 while (i < dm_table_get_num_targets(t)) {
1429 ti = dm_table_get_target(t, i++);
1431 if (!ti->type->iterate_devices ||
1432 !ti->type->iterate_devices(ti, func, NULL))
1439 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1440 sector_t start, sector_t len, void *data)
1442 struct request_queue *q = bdev_get_queue(dev->bdev);
1444 return q && !q->limits.max_write_same_sectors;
1447 static bool dm_table_supports_write_same(struct dm_table *t)
1449 struct dm_target *ti;
1452 while (i < dm_table_get_num_targets(t)) {
1453 ti = dm_table_get_target(t, i++);
1455 if (!ti->num_write_same_bios)
1458 if (!ti->type->iterate_devices ||
1459 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1466 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1467 struct queue_limits *limits)
1472 * Copy table's limits to the DM device's request_queue
1474 q->limits = *limits;
1476 if (!dm_table_supports_discards(t))
1477 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1479 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1481 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1483 if (dm_table_supports_flush(t, REQ_FUA))
1486 blk_queue_flush(q, flush);
1488 if (!dm_table_discard_zeroes_data(t))
1489 q->limits.discard_zeroes_data = 0;
1491 /* Ensure that all underlying devices are non-rotational. */
1492 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1493 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1495 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1497 if (!dm_table_supports_write_same(t))
1498 q->limits.max_write_same_sectors = 0;
1500 dm_table_set_integrity(t);
1503 * Determine whether or not this queue's I/O timings contribute
1504 * to the entropy pool, Only request-based targets use this.
1505 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1508 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1509 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1512 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1513 * visible to other CPUs because, once the flag is set, incoming bios
1514 * are processed by request-based dm, which refers to the queue
1516 * Until the flag set, bios are passed to bio-based dm and queued to
1517 * md->deferred where queue settings are not needed yet.
1518 * Those bios are passed to request-based dm at the resume time.
1521 if (dm_table_request_based(t))
1522 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1525 unsigned int dm_table_get_num_targets(struct dm_table *t)
1527 return t->num_targets;
1530 struct list_head *dm_table_get_devices(struct dm_table *t)
1535 fmode_t dm_table_get_mode(struct dm_table *t)
1539 EXPORT_SYMBOL(dm_table_get_mode);
1541 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1543 int i = t->num_targets;
1544 struct dm_target *ti = t->targets;
1548 if (ti->type->postsuspend)
1549 ti->type->postsuspend(ti);
1550 } else if (ti->type->presuspend)
1551 ti->type->presuspend(ti);
1557 void dm_table_presuspend_targets(struct dm_table *t)
1562 suspend_targets(t, 0);
1565 void dm_table_postsuspend_targets(struct dm_table *t)
1570 suspend_targets(t, 1);
1573 int dm_table_resume_targets(struct dm_table *t)
1577 for (i = 0; i < t->num_targets; i++) {
1578 struct dm_target *ti = t->targets + i;
1580 if (!ti->type->preresume)
1583 r = ti->type->preresume(ti);
1588 for (i = 0; i < t->num_targets; i++) {
1589 struct dm_target *ti = t->targets + i;
1591 if (ti->type->resume)
1592 ti->type->resume(ti);
1598 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1600 list_add(&cb->list, &t->target_callbacks);
1602 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1604 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1606 struct dm_dev_internal *dd;
1607 struct list_head *devices = dm_table_get_devices(t);
1608 struct dm_target_callbacks *cb;
1611 list_for_each_entry(dd, devices, list) {
1612 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1613 char b[BDEVNAME_SIZE];
1616 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1618 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1619 dm_device_name(t->md),
1620 bdevname(dd->dm_dev.bdev, b));
1623 list_for_each_entry(cb, &t->target_callbacks, list)
1624 if (cb->congested_fn)
1625 r |= cb->congested_fn(cb, bdi_bits);
1630 int dm_table_any_busy_target(struct dm_table *t)
1633 struct dm_target *ti;
1635 for (i = 0; i < t->num_targets; i++) {
1636 ti = t->targets + i;
1637 if (ti->type->busy && ti->type->busy(ti))
1644 struct mapped_device *dm_table_get_md(struct dm_table *t)
1648 EXPORT_SYMBOL(dm_table_get_md);
1650 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1651 sector_t start, sector_t len, void *data)
1653 struct request_queue *q = bdev_get_queue(dev->bdev);
1655 return q && blk_queue_discard(q);
1658 bool dm_table_supports_discards(struct dm_table *t)
1660 struct dm_target *ti;
1664 * Unless any target used by the table set discards_supported,
1665 * require at least one underlying device to support discards.
1666 * t->devices includes internal dm devices such as mirror logs
1667 * so we need to use iterate_devices here, which targets
1668 * supporting discard selectively must provide.
1670 while (i < dm_table_get_num_targets(t)) {
1671 ti = dm_table_get_target(t, i++);
1673 if (!ti->num_discard_bios)
1676 if (ti->discards_supported)
1679 if (ti->type->iterate_devices &&
1680 ti->type->iterate_devices(ti, device_discard_capable, NULL))