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)) {
235 static void free_devices(struct list_head *devices)
237 struct list_head *tmp, *next;
239 list_for_each_safe(tmp, next, devices) {
240 struct dm_dev_internal *dd =
241 list_entry(tmp, struct dm_dev_internal, list);
242 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
248 void dm_table_destroy(struct dm_table *t)
255 while (atomic_read(&t->holders))
259 /* free the indexes */
261 vfree(t->index[t->depth - 2]);
263 /* free the targets */
264 for (i = 0; i < t->num_targets; i++) {
265 struct dm_target *tgt = t->targets + i;
270 dm_put_target_type(tgt->type);
275 /* free the device list */
276 if (t->devices.next != &t->devices)
277 free_devices(&t->devices);
279 dm_free_md_mempools(t->mempools);
284 void dm_table_get(struct dm_table *t)
286 atomic_inc(&t->holders);
288 EXPORT_SYMBOL(dm_table_get);
290 void dm_table_put(struct dm_table *t)
295 smp_mb__before_atomic_dec();
296 atomic_dec(&t->holders);
298 EXPORT_SYMBOL(dm_table_put);
301 * Checks to see if we need to extend highs or targets.
303 static inline int check_space(struct dm_table *t)
305 if (t->num_targets >= t->num_allocated)
306 return alloc_targets(t, t->num_allocated * 2);
312 * See if we've already got a device in the list.
314 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
316 struct dm_dev_internal *dd;
318 list_for_each_entry (dd, l, list)
319 if (dd->dm_dev.bdev->bd_dev == dev)
326 * Open a device so we can use it as a map destination.
328 static int open_dev(struct dm_dev_internal *d, dev_t dev,
329 struct mapped_device *md)
331 static char *_claim_ptr = "I belong to device-mapper";
332 struct block_device *bdev;
336 BUG_ON(d->dm_dev.bdev);
338 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
340 return PTR_ERR(bdev);
342 r = bd_link_disk_holder(bdev, dm_disk(md));
344 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
348 d->dm_dev.bdev = bdev;
353 * Close a device that we've been using.
355 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
360 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
361 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
362 d->dm_dev.bdev = NULL;
366 * If possible, this checks an area of a destination device is invalid.
368 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
369 sector_t start, sector_t len, void *data)
371 struct request_queue *q;
372 struct queue_limits *limits = data;
373 struct block_device *bdev = dev->bdev;
375 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
376 unsigned short logical_block_size_sectors =
377 limits->logical_block_size >> SECTOR_SHIFT;
378 char b[BDEVNAME_SIZE];
381 * Some devices exist without request functions,
382 * such as loop devices not yet bound to backing files.
383 * Forbid the use of such devices.
385 q = bdev_get_queue(bdev);
386 if (!q || !q->make_request_fn) {
387 DMWARN("%s: %s is not yet initialised: "
388 "start=%llu, len=%llu, dev_size=%llu",
389 dm_device_name(ti->table->md), bdevname(bdev, b),
390 (unsigned long long)start,
391 (unsigned long long)len,
392 (unsigned long long)dev_size);
399 if ((start >= dev_size) || (start + len > dev_size)) {
400 DMWARN("%s: %s too small for target: "
401 "start=%llu, len=%llu, dev_size=%llu",
402 dm_device_name(ti->table->md), bdevname(bdev, b),
403 (unsigned long long)start,
404 (unsigned long long)len,
405 (unsigned long long)dev_size);
409 if (logical_block_size_sectors <= 1)
412 if (start & (logical_block_size_sectors - 1)) {
413 DMWARN("%s: start=%llu not aligned to h/w "
414 "logical block size %u of %s",
415 dm_device_name(ti->table->md),
416 (unsigned long long)start,
417 limits->logical_block_size, bdevname(bdev, b));
421 if (len & (logical_block_size_sectors - 1)) {
422 DMWARN("%s: len=%llu not aligned to h/w "
423 "logical block size %u of %s",
424 dm_device_name(ti->table->md),
425 (unsigned long long)len,
426 limits->logical_block_size, bdevname(bdev, b));
434 * This upgrades the mode on an already open dm_dev, being
435 * careful to leave things as they were if we fail to reopen the
436 * device and not to touch the existing bdev field in case
437 * it is accessed concurrently inside dm_table_any_congested().
439 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
440 struct mapped_device *md)
443 struct dm_dev_internal dd_new, dd_old;
445 dd_new = dd_old = *dd;
447 dd_new.dm_dev.mode |= new_mode;
448 dd_new.dm_dev.bdev = NULL;
450 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
454 dd->dm_dev.mode |= new_mode;
455 close_dev(&dd_old, md);
461 * Add a device to the list, or just increment the usage count if
462 * it's already present.
464 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
465 struct dm_dev **result)
468 dev_t uninitialized_var(dev);
469 struct dm_dev_internal *dd;
470 unsigned int major, minor;
471 struct dm_table *t = ti->table;
475 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
476 /* Extract the major/minor numbers */
477 dev = MKDEV(major, minor);
478 if (MAJOR(dev) != major || MINOR(dev) != minor)
481 /* convert the path to a device */
482 struct block_device *bdev = lookup_bdev(path);
485 return PTR_ERR(bdev);
490 dd = find_device(&t->devices, dev);
492 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
496 dd->dm_dev.mode = mode;
497 dd->dm_dev.bdev = NULL;
499 if ((r = open_dev(dd, dev, t->md))) {
504 format_dev_t(dd->dm_dev.name, dev);
506 atomic_set(&dd->count, 0);
507 list_add(&dd->list, &t->devices);
509 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
510 r = upgrade_mode(dd, mode, t->md);
514 atomic_inc(&dd->count);
516 *result = &dd->dm_dev;
519 EXPORT_SYMBOL(dm_get_device);
521 int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
522 sector_t start, sector_t len, void *data)
524 struct queue_limits *limits = data;
525 struct block_device *bdev = dev->bdev;
526 struct request_queue *q = bdev_get_queue(bdev);
527 char b[BDEVNAME_SIZE];
530 DMWARN("%s: Cannot set limits for nonexistent device %s",
531 dm_device_name(ti->table->md), bdevname(bdev, b));
535 if (bdev_stack_limits(limits, bdev, start) < 0)
536 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
537 "physical_block_size=%u, logical_block_size=%u, "
538 "alignment_offset=%u, start=%llu",
539 dm_device_name(ti->table->md), bdevname(bdev, b),
540 q->limits.physical_block_size,
541 q->limits.logical_block_size,
542 q->limits.alignment_offset,
543 (unsigned long long) start << SECTOR_SHIFT);
546 * Check if merge fn is supported.
547 * If not we'll force DM to use PAGE_SIZE or
548 * smaller I/O, just to be safe.
550 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
551 blk_limits_max_hw_sectors(limits,
552 (unsigned int) (PAGE_SIZE >> 9));
555 EXPORT_SYMBOL_GPL(dm_set_device_limits);
558 * Decrement a device's use count and remove it if necessary.
560 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
562 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
565 if (atomic_dec_and_test(&dd->count)) {
566 close_dev(dd, ti->table->md);
571 EXPORT_SYMBOL(dm_put_device);
574 * Checks to see if the target joins onto the end of the table.
576 static int adjoin(struct dm_table *table, struct dm_target *ti)
578 struct dm_target *prev;
580 if (!table->num_targets)
583 prev = &table->targets[table->num_targets - 1];
584 return (ti->begin == (prev->begin + prev->len));
588 * Used to dynamically allocate the arg array.
590 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
591 * process messages even if some device is suspended. These messages have a
592 * small fixed number of arguments.
594 * On the other hand, dm-switch needs to process bulk data using messages and
595 * excessive use of GFP_NOIO could cause trouble.
597 static char **realloc_argv(unsigned *array_size, char **old_argv)
604 new_size = *array_size * 2;
610 argv = kmalloc(new_size * sizeof(*argv), gfp);
612 memcpy(argv, old_argv, *array_size * sizeof(*argv));
613 *array_size = new_size;
621 * Destructively splits up the argument list to pass to ctr.
623 int dm_split_args(int *argc, char ***argvp, char *input)
625 char *start, *end = input, *out, **argv = NULL;
626 unsigned array_size = 0;
635 argv = realloc_argv(&array_size, argv);
640 /* Skip whitespace */
641 start = skip_spaces(end);
644 break; /* success, we hit the end */
646 /* 'out' is used to remove any back-quotes */
649 /* Everything apart from '\0' can be quoted */
650 if (*end == '\\' && *(end + 1)) {
657 break; /* end of token */
662 /* have we already filled the array ? */
663 if ((*argc + 1) > array_size) {
664 argv = realloc_argv(&array_size, argv);
669 /* we know this is whitespace */
673 /* terminate the string and put it in the array */
684 * Impose necessary and sufficient conditions on a devices's table such
685 * that any incoming bio which respects its logical_block_size can be
686 * processed successfully. If it falls across the boundary between
687 * two or more targets, the size of each piece it gets split into must
688 * be compatible with the logical_block_size of the target processing it.
690 static int validate_hardware_logical_block_alignment(struct dm_table *table,
691 struct queue_limits *limits)
694 * This function uses arithmetic modulo the logical_block_size
695 * (in units of 512-byte sectors).
697 unsigned short device_logical_block_size_sects =
698 limits->logical_block_size >> SECTOR_SHIFT;
701 * Offset of the start of the next table entry, mod logical_block_size.
703 unsigned short next_target_start = 0;
706 * Given an aligned bio that extends beyond the end of a
707 * target, how many sectors must the next target handle?
709 unsigned short remaining = 0;
711 struct dm_target *uninitialized_var(ti);
712 struct queue_limits ti_limits;
716 * Check each entry in the table in turn.
718 while (i < dm_table_get_num_targets(table)) {
719 ti = dm_table_get_target(table, i++);
721 blk_set_default_limits(&ti_limits);
723 /* combine all target devices' limits */
724 if (ti->type->iterate_devices)
725 ti->type->iterate_devices(ti, dm_set_device_limits,
729 * If the remaining sectors fall entirely within this
730 * table entry are they compatible with its logical_block_size?
732 if (remaining < ti->len &&
733 remaining & ((ti_limits.logical_block_size >>
738 (unsigned short) ((next_target_start + ti->len) &
739 (device_logical_block_size_sects - 1));
740 remaining = next_target_start ?
741 device_logical_block_size_sects - next_target_start : 0;
745 DMWARN("%s: table line %u (start sect %llu len %llu) "
746 "not aligned to h/w logical block size %u",
747 dm_device_name(table->md), i,
748 (unsigned long long) ti->begin,
749 (unsigned long long) ti->len,
750 limits->logical_block_size);
757 int dm_table_add_target(struct dm_table *t, const char *type,
758 sector_t start, sector_t len, char *params)
760 int r = -EINVAL, argc;
762 struct dm_target *tgt;
765 DMERR("%s: target type %s must appear alone in table",
766 dm_device_name(t->md), t->targets->type->name);
770 if ((r = check_space(t)))
773 tgt = t->targets + t->num_targets;
774 memset(tgt, 0, sizeof(*tgt));
777 DMERR("%s: zero-length target", dm_device_name(t->md));
781 tgt->type = dm_get_target_type(type);
783 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
788 if (dm_target_needs_singleton(tgt->type)) {
789 if (t->num_targets) {
790 DMERR("%s: target type %s must appear alone in table",
791 dm_device_name(t->md), type);
797 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
798 DMERR("%s: target type %s may not be included in read-only tables",
799 dm_device_name(t->md), type);
803 if (t->immutable_target_type) {
804 if (t->immutable_target_type != tgt->type) {
805 DMERR("%s: immutable target type %s cannot be mixed with other target types",
806 dm_device_name(t->md), t->immutable_target_type->name);
809 } else if (dm_target_is_immutable(tgt->type)) {
810 if (t->num_targets) {
811 DMERR("%s: immutable target type %s cannot be mixed with other target types",
812 dm_device_name(t->md), tgt->type->name);
815 t->immutable_target_type = tgt->type;
821 tgt->error = "Unknown error";
824 * Does this target adjoin the previous one ?
826 if (!adjoin(t, tgt)) {
827 tgt->error = "Gap in table";
832 r = dm_split_args(&argc, &argv, params);
834 tgt->error = "couldn't split parameters (insufficient memory)";
838 r = tgt->type->ctr(tgt, argc, argv);
843 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
845 if (!tgt->num_discard_requests && tgt->discards_supported)
846 DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
847 dm_device_name(t->md), type);
852 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
853 dm_put_target_type(tgt->type);
858 * Target argument parsing helpers.
860 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
861 unsigned *value, char **error, unsigned grouped)
863 const char *arg_str = dm_shift_arg(arg_set);
866 (sscanf(arg_str, "%u", value) != 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);
989 if (unlikely(type == DM_TYPE_NONE)) {
990 DMWARN("no table type is set, can't allocate mempools");
994 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
1001 void dm_table_free_md_mempools(struct dm_table *t)
1003 dm_free_md_mempools(t->mempools);
1007 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1012 static int setup_indexes(struct dm_table *t)
1015 unsigned int total = 0;
1018 /* allocate the space for *all* the indexes */
1019 for (i = t->depth - 2; i >= 0; i--) {
1020 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1021 total += t->counts[i];
1024 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1028 /* set up internal nodes, bottom-up */
1029 for (i = t->depth - 2; i >= 0; i--) {
1030 t->index[i] = indexes;
1031 indexes += (KEYS_PER_NODE * t->counts[i]);
1032 setup_btree_index(i, t);
1039 * Builds the btree to index the map.
1041 static int dm_table_build_index(struct dm_table *t)
1044 unsigned int leaf_nodes;
1046 /* how many indexes will the btree have ? */
1047 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1048 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1050 /* leaf layer has already been set up */
1051 t->counts[t->depth - 1] = leaf_nodes;
1052 t->index[t->depth - 1] = t->highs;
1055 r = setup_indexes(t);
1061 * Get a disk whose integrity profile reflects the table's profile.
1062 * If %match_all is true, all devices' profiles must match.
1063 * If %match_all is false, all devices must at least have an
1064 * allocated integrity profile; but uninitialized is ok.
1065 * Returns NULL if integrity support was inconsistent or unavailable.
1067 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1070 struct list_head *devices = dm_table_get_devices(t);
1071 struct dm_dev_internal *dd = NULL;
1072 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1074 list_for_each_entry(dd, devices, list) {
1075 template_disk = dd->dm_dev.bdev->bd_disk;
1076 if (!blk_get_integrity(template_disk))
1078 if (!match_all && !blk_integrity_is_initialized(template_disk))
1079 continue; /* skip uninitialized profiles */
1080 else if (prev_disk &&
1081 blk_integrity_compare(prev_disk, template_disk) < 0)
1083 prev_disk = template_disk;
1086 return template_disk;
1090 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1091 dm_device_name(t->md),
1092 prev_disk->disk_name,
1093 template_disk->disk_name);
1098 * Register the mapped device for blk_integrity support if
1099 * the underlying devices have an integrity profile. But all devices
1100 * may not have matching profiles (checking all devices isn't reliable
1101 * during table load because this table may use other DM device(s) which
1102 * must be resumed before they will have an initialized integity profile).
1103 * Stacked DM devices force a 2 stage integrity profile validation:
1104 * 1 - during load, validate all initialized integrity profiles match
1105 * 2 - during resume, validate all integrity profiles match
1107 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1109 struct gendisk *template_disk = NULL;
1111 template_disk = dm_table_get_integrity_disk(t, false);
1115 if (!blk_integrity_is_initialized(dm_disk(md))) {
1116 t->integrity_supported = 1;
1117 return blk_integrity_register(dm_disk(md), NULL);
1121 * If DM device already has an initalized integrity
1122 * profile the new profile should not conflict.
1124 if (blk_integrity_is_initialized(template_disk) &&
1125 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1126 DMWARN("%s: conflict with existing integrity profile: "
1127 "%s profile mismatch",
1128 dm_device_name(t->md),
1129 template_disk->disk_name);
1133 /* Preserve existing initialized integrity profile */
1134 t->integrity_supported = 1;
1139 * Prepares the table for use by building the indices,
1140 * setting the type, and allocating mempools.
1142 int dm_table_complete(struct dm_table *t)
1146 r = dm_table_set_type(t);
1148 DMERR("unable to set table type");
1152 r = dm_table_build_index(t);
1154 DMERR("unable to build btrees");
1158 r = dm_table_prealloc_integrity(t, t->md);
1160 DMERR("could not register integrity profile.");
1164 r = dm_table_alloc_md_mempools(t);
1166 DMERR("unable to allocate mempools");
1171 static DEFINE_MUTEX(_event_lock);
1172 void dm_table_event_callback(struct dm_table *t,
1173 void (*fn)(void *), void *context)
1175 mutex_lock(&_event_lock);
1177 t->event_context = context;
1178 mutex_unlock(&_event_lock);
1181 void dm_table_event(struct dm_table *t)
1184 * You can no longer call dm_table_event() from interrupt
1185 * context, use a bottom half instead.
1187 BUG_ON(in_interrupt());
1189 mutex_lock(&_event_lock);
1191 t->event_fn(t->event_context);
1192 mutex_unlock(&_event_lock);
1194 EXPORT_SYMBOL(dm_table_event);
1196 sector_t dm_table_get_size(struct dm_table *t)
1198 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1200 EXPORT_SYMBOL(dm_table_get_size);
1202 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1204 if (index >= t->num_targets)
1207 return t->targets + index;
1211 * Search the btree for the correct target.
1213 * Caller should check returned pointer with dm_target_is_valid()
1214 * to trap I/O beyond end of device.
1216 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1218 unsigned int l, n = 0, k = 0;
1221 for (l = 0; l < t->depth; l++) {
1222 n = get_child(n, k);
1223 node = get_node(t, l, n);
1225 for (k = 0; k < KEYS_PER_NODE; k++)
1226 if (node[k] >= sector)
1230 return &t->targets[(KEYS_PER_NODE * n) + k];
1234 * Establish the new table's queue_limits and validate them.
1236 int dm_calculate_queue_limits(struct dm_table *table,
1237 struct queue_limits *limits)
1239 struct dm_target *uninitialized_var(ti);
1240 struct queue_limits ti_limits;
1243 blk_set_default_limits(limits);
1245 while (i < dm_table_get_num_targets(table)) {
1246 blk_set_default_limits(&ti_limits);
1248 ti = dm_table_get_target(table, i++);
1250 if (!ti->type->iterate_devices)
1251 goto combine_limits;
1254 * Combine queue limits of all the devices this target uses.
1256 ti->type->iterate_devices(ti, dm_set_device_limits,
1259 /* Set I/O hints portion of queue limits */
1260 if (ti->type->io_hints)
1261 ti->type->io_hints(ti, &ti_limits);
1264 * Check each device area is consistent with the target's
1265 * overall queue limits.
1267 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1273 * Merge this target's queue limits into the overall limits
1276 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1277 DMWARN("%s: adding target device "
1278 "(start sect %llu len %llu) "
1279 "caused an alignment inconsistency",
1280 dm_device_name(table->md),
1281 (unsigned long long) ti->begin,
1282 (unsigned long long) ti->len);
1285 return validate_hardware_logical_block_alignment(table, limits);
1289 * Set the integrity profile for this device if all devices used have
1290 * matching profiles. We're quite deep in the resume path but still
1291 * don't know if all devices (particularly DM devices this device
1292 * may be stacked on) have matching profiles. Even if the profiles
1293 * don't match we have no way to fail (to resume) at this point.
1295 static void dm_table_set_integrity(struct dm_table *t)
1297 struct gendisk *template_disk = NULL;
1299 if (!blk_get_integrity(dm_disk(t->md)))
1302 template_disk = dm_table_get_integrity_disk(t, true);
1304 blk_integrity_register(dm_disk(t->md),
1305 blk_get_integrity(template_disk));
1306 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1307 DMWARN("%s: device no longer has a valid integrity profile",
1308 dm_device_name(t->md));
1310 DMWARN("%s: unable to establish an integrity profile",
1311 dm_device_name(t->md));
1314 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1315 sector_t start, sector_t len, void *data)
1317 unsigned flush = (*(unsigned *)data);
1318 struct request_queue *q = bdev_get_queue(dev->bdev);
1320 return q && (q->flush_flags & flush);
1323 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1325 struct dm_target *ti;
1329 * Require at least one underlying device to support flushes.
1330 * t->devices includes internal dm devices such as mirror logs
1331 * so we need to use iterate_devices here, which targets
1332 * supporting flushes must provide.
1334 while (i < dm_table_get_num_targets(t)) {
1335 ti = dm_table_get_target(t, i++);
1337 if (!ti->num_flush_requests)
1340 if (ti->type->iterate_devices &&
1341 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1348 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1350 struct dm_target *ti;
1353 /* Ensure that all targets supports discard_zeroes_data. */
1354 while (i < dm_table_get_num_targets(t)) {
1355 ti = dm_table_get_target(t, i++);
1357 if (ti->discard_zeroes_data_unsupported)
1364 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1365 sector_t start, sector_t len, void *data)
1367 struct request_queue *q = bdev_get_queue(dev->bdev);
1369 return q && blk_queue_nonrot(q);
1372 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1373 sector_t start, sector_t len, void *data)
1375 struct request_queue *q = bdev_get_queue(dev->bdev);
1377 return q && !blk_queue_add_random(q);
1380 static bool dm_table_all_devices_attribute(struct dm_table *t,
1381 iterate_devices_callout_fn func)
1383 struct dm_target *ti;
1386 while (i < dm_table_get_num_targets(t)) {
1387 ti = dm_table_get_target(t, i++);
1389 if (!ti->type->iterate_devices ||
1390 !ti->type->iterate_devices(ti, func, NULL))
1397 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1398 struct queue_limits *limits)
1403 * Copy table's limits to the DM device's request_queue
1405 q->limits = *limits;
1407 if (!dm_table_supports_discards(t))
1408 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1410 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1412 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1414 if (dm_table_supports_flush(t, REQ_FUA))
1417 blk_queue_flush(q, flush);
1419 if (!dm_table_discard_zeroes_data(t))
1420 q->limits.discard_zeroes_data = 0;
1422 /* Ensure that all underlying devices are non-rotational. */
1423 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1424 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1426 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1428 dm_table_set_integrity(t);
1431 * Determine whether or not this queue's I/O timings contribute
1432 * to the entropy pool, Only request-based targets use this.
1433 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1436 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1437 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1440 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1441 * visible to other CPUs because, once the flag is set, incoming bios
1442 * are processed by request-based dm, which refers to the queue
1444 * Until the flag set, bios are passed to bio-based dm and queued to
1445 * md->deferred where queue settings are not needed yet.
1446 * Those bios are passed to request-based dm at the resume time.
1449 if (dm_table_request_based(t))
1450 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1453 unsigned int dm_table_get_num_targets(struct dm_table *t)
1455 return t->num_targets;
1458 struct list_head *dm_table_get_devices(struct dm_table *t)
1463 fmode_t dm_table_get_mode(struct dm_table *t)
1467 EXPORT_SYMBOL(dm_table_get_mode);
1469 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1471 int i = t->num_targets;
1472 struct dm_target *ti = t->targets;
1476 if (ti->type->postsuspend)
1477 ti->type->postsuspend(ti);
1478 } else if (ti->type->presuspend)
1479 ti->type->presuspend(ti);
1485 void dm_table_presuspend_targets(struct dm_table *t)
1490 suspend_targets(t, 0);
1493 void dm_table_postsuspend_targets(struct dm_table *t)
1498 suspend_targets(t, 1);
1501 int dm_table_resume_targets(struct dm_table *t)
1505 for (i = 0; i < t->num_targets; i++) {
1506 struct dm_target *ti = t->targets + i;
1508 if (!ti->type->preresume)
1511 r = ti->type->preresume(ti);
1516 for (i = 0; i < t->num_targets; i++) {
1517 struct dm_target *ti = t->targets + i;
1519 if (ti->type->resume)
1520 ti->type->resume(ti);
1526 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1528 list_add(&cb->list, &t->target_callbacks);
1530 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1532 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1534 struct dm_dev_internal *dd;
1535 struct list_head *devices = dm_table_get_devices(t);
1536 struct dm_target_callbacks *cb;
1539 list_for_each_entry(dd, devices, list) {
1540 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1541 char b[BDEVNAME_SIZE];
1544 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1546 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1547 dm_device_name(t->md),
1548 bdevname(dd->dm_dev.bdev, b));
1551 list_for_each_entry(cb, &t->target_callbacks, list)
1552 if (cb->congested_fn)
1553 r |= cb->congested_fn(cb, bdi_bits);
1558 int dm_table_any_busy_target(struct dm_table *t)
1561 struct dm_target *ti;
1563 for (i = 0; i < t->num_targets; i++) {
1564 ti = t->targets + i;
1565 if (ti->type->busy && ti->type->busy(ti))
1572 struct mapped_device *dm_table_get_md(struct dm_table *t)
1576 EXPORT_SYMBOL(dm_table_get_md);
1578 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1579 sector_t start, sector_t len, void *data)
1581 struct request_queue *q = bdev_get_queue(dev->bdev);
1583 return q && blk_queue_discard(q);
1586 bool dm_table_supports_discards(struct dm_table *t)
1588 struct dm_target *ti;
1592 * Unless any target used by the table set discards_supported,
1593 * require at least one underlying device to support discards.
1594 * t->devices includes internal dm devices such as mirror logs
1595 * so we need to use iterate_devices here, which targets
1596 * supporting discard selectively must provide.
1598 while (i < dm_table_get_num_targets(t)) {
1599 ti = dm_table_get_target(t, i++);
1601 if (!ti->num_discard_requests)
1604 if (ti->discards_supported)
1607 if (ti->type->iterate_devices &&
1608 ti->type->iterate_devices(ti, device_discard_capable, NULL))