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 struct mapped_device *md;
35 unsigned int counts[MAX_DEPTH]; /* in nodes */
36 sector_t *index[MAX_DEPTH];
38 unsigned int num_targets;
39 unsigned int num_allocated;
41 struct dm_target *targets;
43 struct target_type *immutable_target_type;
44 unsigned integrity_supported:1;
48 * Indicates the rw permissions for the new logical
49 * device. This should be a combination of FMODE_READ
54 /* a list of devices used by this table */
55 struct list_head devices;
57 /* events get handed up using this callback */
58 void (*event_fn)(void *);
61 struct dm_md_mempools *mempools;
63 struct list_head target_callbacks;
67 * Similar to ceiling(log_size(n))
69 static unsigned int int_log(unsigned int n, unsigned int base)
74 n = dm_div_up(n, base);
82 * Calculate the index of the child node of the n'th node k'th key.
84 static inline unsigned int get_child(unsigned int n, unsigned int k)
86 return (n * CHILDREN_PER_NODE) + k;
90 * Return the n'th node of level l from table t.
92 static inline sector_t *get_node(struct dm_table *t,
93 unsigned int l, unsigned int n)
95 return t->index[l] + (n * KEYS_PER_NODE);
99 * Return the highest key that you could lookup from the n'th
100 * node on level l of the btree.
102 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
104 for (; l < t->depth - 1; l++)
105 n = get_child(n, CHILDREN_PER_NODE - 1);
107 if (n >= t->counts[l])
108 return (sector_t) - 1;
110 return get_node(t, l, n)[KEYS_PER_NODE - 1];
114 * Fills in a level of the btree based on the highs of the level
117 static int setup_btree_index(unsigned int l, struct dm_table *t)
122 for (n = 0U; n < t->counts[l]; n++) {
123 node = get_node(t, l, n);
125 for (k = 0U; k < KEYS_PER_NODE; k++)
126 node[k] = high(t, l + 1, get_child(n, k));
132 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
138 * Check that we're not going to overflow.
140 if (nmemb > (ULONG_MAX / elem_size))
143 size = nmemb * elem_size;
144 addr = vzalloc(size);
148 EXPORT_SYMBOL(dm_vcalloc);
151 * highs, and targets are managed as dynamic arrays during a
154 static int alloc_targets(struct dm_table *t, unsigned int num)
157 struct dm_target *n_targets;
160 * Allocate both the target array and offset array at once.
161 * Append an empty entry to catch sectors beyond the end of
164 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
169 n_targets = (struct dm_target *) (n_highs + num);
171 memset(n_highs, -1, sizeof(*n_highs) * num);
174 t->num_allocated = num;
176 t->targets = n_targets;
181 int dm_table_create(struct dm_table **result, fmode_t mode,
182 unsigned num_targets, struct mapped_device *md)
184 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
189 INIT_LIST_HEAD(&t->devices);
190 INIT_LIST_HEAD(&t->target_callbacks);
193 num_targets = KEYS_PER_NODE;
195 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
202 if (alloc_targets(t, num_targets)) {
213 static void free_devices(struct list_head *devices, struct mapped_device *md)
215 struct list_head *tmp, *next;
217 list_for_each_safe(tmp, next, devices) {
218 struct dm_dev_internal *dd =
219 list_entry(tmp, struct dm_dev_internal, list);
220 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
221 dm_device_name(md), dd->dm_dev->name);
222 dm_put_table_device(md, dd->dm_dev);
227 void dm_table_destroy(struct dm_table *t)
234 /* free the indexes */
236 vfree(t->index[t->depth - 2]);
238 /* free the targets */
239 for (i = 0; i < t->num_targets; i++) {
240 struct dm_target *tgt = t->targets + i;
245 dm_put_target_type(tgt->type);
250 /* free the device list */
251 free_devices(&t->devices, t->md);
253 dm_free_md_mempools(t->mempools);
259 * See if we've already got a device in the list.
261 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
263 struct dm_dev_internal *dd;
265 list_for_each_entry (dd, l, list)
266 if (dd->dm_dev->bdev->bd_dev == dev)
273 * If possible, this checks an area of a destination device is invalid.
275 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
276 sector_t start, sector_t len, void *data)
278 struct request_queue *q;
279 struct queue_limits *limits = data;
280 struct block_device *bdev = dev->bdev;
282 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
283 unsigned short logical_block_size_sectors =
284 limits->logical_block_size >> SECTOR_SHIFT;
285 char b[BDEVNAME_SIZE];
288 * Some devices exist without request functions,
289 * such as loop devices not yet bound to backing files.
290 * Forbid the use of such devices.
292 q = bdev_get_queue(bdev);
293 if (!q || !q->make_request_fn) {
294 DMWARN("%s: %s is not yet initialised: "
295 "start=%llu, len=%llu, dev_size=%llu",
296 dm_device_name(ti->table->md), bdevname(bdev, b),
297 (unsigned long long)start,
298 (unsigned long long)len,
299 (unsigned long long)dev_size);
306 if ((start >= dev_size) || (start + len > dev_size)) {
307 DMWARN("%s: %s too small for target: "
308 "start=%llu, len=%llu, dev_size=%llu",
309 dm_device_name(ti->table->md), bdevname(bdev, b),
310 (unsigned long long)start,
311 (unsigned long long)len,
312 (unsigned long long)dev_size);
316 if (logical_block_size_sectors <= 1)
319 if (start & (logical_block_size_sectors - 1)) {
320 DMWARN("%s: start=%llu not aligned to h/w "
321 "logical block size %u of %s",
322 dm_device_name(ti->table->md),
323 (unsigned long long)start,
324 limits->logical_block_size, bdevname(bdev, b));
328 if (len & (logical_block_size_sectors - 1)) {
329 DMWARN("%s: len=%llu not aligned to h/w "
330 "logical block size %u of %s",
331 dm_device_name(ti->table->md),
332 (unsigned long long)len,
333 limits->logical_block_size, bdevname(bdev, b));
341 * This upgrades the mode on an already open dm_dev, being
342 * careful to leave things as they were if we fail to reopen the
343 * device and not to touch the existing bdev field in case
344 * it is accessed concurrently inside dm_table_any_congested().
346 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
347 struct mapped_device *md)
350 struct dm_dev *old_dev, *new_dev;
352 old_dev = dd->dm_dev;
354 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
355 dd->dm_dev->mode | new_mode, &new_dev);
359 dd->dm_dev = new_dev;
360 dm_put_table_device(md, old_dev);
366 * Add a device to the list, or just increment the usage count if
367 * it's already present.
369 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
370 struct dm_dev **result)
373 dev_t uninitialized_var(dev);
374 struct dm_dev_internal *dd;
375 unsigned int major, minor;
376 struct dm_table *t = ti->table;
381 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
382 /* Extract the major/minor numbers */
383 dev = MKDEV(major, minor);
384 if (MAJOR(dev) != major || MINOR(dev) != minor)
387 /* convert the path to a device */
388 struct block_device *bdev = lookup_bdev(path);
391 return PTR_ERR(bdev);
396 dd = find_device(&t->devices, dev);
398 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
402 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
407 atomic_set(&dd->count, 0);
408 list_add(&dd->list, &t->devices);
410 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
411 r = upgrade_mode(dd, mode, t->md);
415 atomic_inc(&dd->count);
417 *result = dd->dm_dev;
420 EXPORT_SYMBOL(dm_get_device);
422 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
423 sector_t start, sector_t len, void *data)
425 struct queue_limits *limits = data;
426 struct block_device *bdev = dev->bdev;
427 struct request_queue *q = bdev_get_queue(bdev);
428 char b[BDEVNAME_SIZE];
431 DMWARN("%s: Cannot set limits for nonexistent device %s",
432 dm_device_name(ti->table->md), bdevname(bdev, b));
436 if (bdev_stack_limits(limits, bdev, start) < 0)
437 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
438 "physical_block_size=%u, logical_block_size=%u, "
439 "alignment_offset=%u, start=%llu",
440 dm_device_name(ti->table->md), bdevname(bdev, b),
441 q->limits.physical_block_size,
442 q->limits.logical_block_size,
443 q->limits.alignment_offset,
444 (unsigned long long) start << SECTOR_SHIFT);
447 * Check if merge fn is supported.
448 * If not we'll force DM to use PAGE_SIZE or
449 * smaller I/O, just to be safe.
451 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
452 blk_limits_max_hw_sectors(limits,
453 (unsigned int) (PAGE_SIZE >> 9));
458 * Decrement a device's use count and remove it if necessary.
460 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
463 struct list_head *devices = &ti->table->devices;
464 struct dm_dev_internal *dd;
466 list_for_each_entry(dd, devices, list) {
467 if (dd->dm_dev == d) {
473 DMWARN("%s: device %s not in table devices list",
474 dm_device_name(ti->table->md), d->name);
477 if (atomic_dec_and_test(&dd->count)) {
478 dm_put_table_device(ti->table->md, d);
483 EXPORT_SYMBOL(dm_put_device);
486 * Checks to see if the target joins onto the end of the table.
488 static int adjoin(struct dm_table *table, struct dm_target *ti)
490 struct dm_target *prev;
492 if (!table->num_targets)
495 prev = &table->targets[table->num_targets - 1];
496 return (ti->begin == (prev->begin + prev->len));
500 * Used to dynamically allocate the arg array.
502 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
503 * process messages even if some device is suspended. These messages have a
504 * small fixed number of arguments.
506 * On the other hand, dm-switch needs to process bulk data using messages and
507 * excessive use of GFP_NOIO could cause trouble.
509 static char **realloc_argv(unsigned *array_size, char **old_argv)
516 new_size = *array_size * 2;
522 argv = kmalloc(new_size * sizeof(*argv), gfp);
524 memcpy(argv, old_argv, *array_size * sizeof(*argv));
525 *array_size = new_size;
533 * Destructively splits up the argument list to pass to ctr.
535 int dm_split_args(int *argc, char ***argvp, char *input)
537 char *start, *end = input, *out, **argv = NULL;
538 unsigned array_size = 0;
547 argv = realloc_argv(&array_size, argv);
552 /* Skip whitespace */
553 start = skip_spaces(end);
556 break; /* success, we hit the end */
558 /* 'out' is used to remove any back-quotes */
561 /* Everything apart from '\0' can be quoted */
562 if (*end == '\\' && *(end + 1)) {
569 break; /* end of token */
574 /* have we already filled the array ? */
575 if ((*argc + 1) > array_size) {
576 argv = realloc_argv(&array_size, argv);
581 /* we know this is whitespace */
585 /* terminate the string and put it in the array */
596 * Impose necessary and sufficient conditions on a devices's table such
597 * that any incoming bio which respects its logical_block_size can be
598 * processed successfully. If it falls across the boundary between
599 * two or more targets, the size of each piece it gets split into must
600 * be compatible with the logical_block_size of the target processing it.
602 static int validate_hardware_logical_block_alignment(struct dm_table *table,
603 struct queue_limits *limits)
606 * This function uses arithmetic modulo the logical_block_size
607 * (in units of 512-byte sectors).
609 unsigned short device_logical_block_size_sects =
610 limits->logical_block_size >> SECTOR_SHIFT;
613 * Offset of the start of the next table entry, mod logical_block_size.
615 unsigned short next_target_start = 0;
618 * Given an aligned bio that extends beyond the end of a
619 * target, how many sectors must the next target handle?
621 unsigned short remaining = 0;
623 struct dm_target *uninitialized_var(ti);
624 struct queue_limits ti_limits;
628 * Check each entry in the table in turn.
630 while (i < dm_table_get_num_targets(table)) {
631 ti = dm_table_get_target(table, i++);
633 blk_set_stacking_limits(&ti_limits);
635 /* combine all target devices' limits */
636 if (ti->type->iterate_devices)
637 ti->type->iterate_devices(ti, dm_set_device_limits,
641 * If the remaining sectors fall entirely within this
642 * table entry are they compatible with its logical_block_size?
644 if (remaining < ti->len &&
645 remaining & ((ti_limits.logical_block_size >>
650 (unsigned short) ((next_target_start + ti->len) &
651 (device_logical_block_size_sects - 1));
652 remaining = next_target_start ?
653 device_logical_block_size_sects - next_target_start : 0;
657 DMWARN("%s: table line %u (start sect %llu len %llu) "
658 "not aligned to h/w logical block size %u",
659 dm_device_name(table->md), i,
660 (unsigned long long) ti->begin,
661 (unsigned long long) ti->len,
662 limits->logical_block_size);
669 int dm_table_add_target(struct dm_table *t, const char *type,
670 sector_t start, sector_t len, char *params)
672 int r = -EINVAL, argc;
674 struct dm_target *tgt;
677 DMERR("%s: target type %s must appear alone in table",
678 dm_device_name(t->md), t->targets->type->name);
682 BUG_ON(t->num_targets >= t->num_allocated);
684 tgt = t->targets + t->num_targets;
685 memset(tgt, 0, sizeof(*tgt));
688 DMERR("%s: zero-length target", dm_device_name(t->md));
692 tgt->type = dm_get_target_type(type);
694 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
699 if (dm_target_needs_singleton(tgt->type)) {
700 if (t->num_targets) {
701 DMERR("%s: target type %s must appear alone in table",
702 dm_device_name(t->md), type);
708 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
709 DMERR("%s: target type %s may not be included in read-only tables",
710 dm_device_name(t->md), type);
714 if (t->immutable_target_type) {
715 if (t->immutable_target_type != tgt->type) {
716 DMERR("%s: immutable target type %s cannot be mixed with other target types",
717 dm_device_name(t->md), t->immutable_target_type->name);
720 } else if (dm_target_is_immutable(tgt->type)) {
721 if (t->num_targets) {
722 DMERR("%s: immutable target type %s cannot be mixed with other target types",
723 dm_device_name(t->md), tgt->type->name);
726 t->immutable_target_type = tgt->type;
732 tgt->error = "Unknown error";
735 * Does this target adjoin the previous one ?
737 if (!adjoin(t, tgt)) {
738 tgt->error = "Gap in table";
743 r = dm_split_args(&argc, &argv, params);
745 tgt->error = "couldn't split parameters (insufficient memory)";
749 r = tgt->type->ctr(tgt, argc, argv);
754 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
756 if (!tgt->num_discard_bios && tgt->discards_supported)
757 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
758 dm_device_name(t->md), type);
763 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
764 dm_put_target_type(tgt->type);
769 * Target argument parsing helpers.
771 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
772 unsigned *value, char **error, unsigned grouped)
774 const char *arg_str = dm_shift_arg(arg_set);
778 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
779 (*value < arg->min) ||
780 (*value > arg->max) ||
781 (grouped && arg_set->argc < *value)) {
789 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
790 unsigned *value, char **error)
792 return validate_next_arg(arg, arg_set, value, error, 0);
794 EXPORT_SYMBOL(dm_read_arg);
796 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
797 unsigned *value, char **error)
799 return validate_next_arg(arg, arg_set, value, error, 1);
801 EXPORT_SYMBOL(dm_read_arg_group);
803 const char *dm_shift_arg(struct dm_arg_set *as)
816 EXPORT_SYMBOL(dm_shift_arg);
818 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
820 BUG_ON(as->argc < num_args);
821 as->argc -= num_args;
822 as->argv += num_args;
824 EXPORT_SYMBOL(dm_consume_args);
826 static int dm_table_set_type(struct dm_table *t)
829 unsigned bio_based = 0, request_based = 0, hybrid = 0;
830 struct dm_target *tgt;
831 struct dm_dev_internal *dd;
832 struct list_head *devices;
833 unsigned live_md_type;
835 for (i = 0; i < t->num_targets; i++) {
836 tgt = t->targets + i;
837 if (dm_target_hybrid(tgt))
839 else if (dm_target_request_based(tgt))
844 if (bio_based && request_based) {
845 DMWARN("Inconsistent table: different target types"
846 " can't be mixed up");
851 if (hybrid && !bio_based && !request_based) {
853 * The targets can work either way.
854 * Determine the type from the live device.
855 * Default to bio-based if device is new.
857 live_md_type = dm_get_md_type(t->md);
858 if (live_md_type == DM_TYPE_REQUEST_BASED)
865 /* We must use this table as bio-based */
866 t->type = DM_TYPE_BIO_BASED;
870 BUG_ON(!request_based); /* No targets in this table */
872 /* Non-request-stackable devices can't be used for request-based dm */
873 devices = dm_table_get_devices(t);
874 list_for_each_entry(dd, devices, list) {
875 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev->bdev))) {
876 DMWARN("table load rejected: including"
877 " non-request-stackable devices");
883 * Request-based dm supports only tables that have a single target now.
884 * To support multiple targets, request splitting support is needed,
885 * and that needs lots of changes in the block-layer.
886 * (e.g. request completion process for partial completion.)
888 if (t->num_targets > 1) {
889 DMWARN("Request-based dm doesn't support multiple targets yet");
893 t->type = DM_TYPE_REQUEST_BASED;
898 unsigned dm_table_get_type(struct dm_table *t)
903 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
905 return t->immutable_target_type;
908 bool dm_table_request_based(struct dm_table *t)
910 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
913 static int dm_table_alloc_md_mempools(struct dm_table *t)
915 unsigned type = dm_table_get_type(t);
916 unsigned per_bio_data_size = 0;
917 struct dm_target *tgt;
920 if (unlikely(type == DM_TYPE_NONE)) {
921 DMWARN("no table type is set, can't allocate mempools");
925 if (type == DM_TYPE_BIO_BASED)
926 for (i = 0; i < t->num_targets; i++) {
927 tgt = t->targets + i;
928 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
931 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported, per_bio_data_size);
938 void dm_table_free_md_mempools(struct dm_table *t)
940 dm_free_md_mempools(t->mempools);
944 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
949 static int setup_indexes(struct dm_table *t)
952 unsigned int total = 0;
955 /* allocate the space for *all* the indexes */
956 for (i = t->depth - 2; i >= 0; i--) {
957 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
958 total += t->counts[i];
961 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
965 /* set up internal nodes, bottom-up */
966 for (i = t->depth - 2; i >= 0; i--) {
967 t->index[i] = indexes;
968 indexes += (KEYS_PER_NODE * t->counts[i]);
969 setup_btree_index(i, t);
976 * Builds the btree to index the map.
978 static int dm_table_build_index(struct dm_table *t)
981 unsigned int leaf_nodes;
983 /* how many indexes will the btree have ? */
984 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
985 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
987 /* leaf layer has already been set up */
988 t->counts[t->depth - 1] = leaf_nodes;
989 t->index[t->depth - 1] = t->highs;
992 r = setup_indexes(t);
998 * Get a disk whose integrity profile reflects the table's profile.
999 * If %match_all is true, all devices' profiles must match.
1000 * If %match_all is false, all devices must at least have an
1001 * allocated integrity profile; but uninitialized is ok.
1002 * Returns NULL if integrity support was inconsistent or unavailable.
1004 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1007 struct list_head *devices = dm_table_get_devices(t);
1008 struct dm_dev_internal *dd = NULL;
1009 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1011 list_for_each_entry(dd, devices, list) {
1012 template_disk = dd->dm_dev->bdev->bd_disk;
1013 if (!blk_get_integrity(template_disk))
1015 if (!match_all && !blk_integrity_is_initialized(template_disk))
1016 continue; /* skip uninitialized profiles */
1017 else if (prev_disk &&
1018 blk_integrity_compare(prev_disk, template_disk) < 0)
1020 prev_disk = template_disk;
1023 return template_disk;
1027 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1028 dm_device_name(t->md),
1029 prev_disk->disk_name,
1030 template_disk->disk_name);
1035 * Register the mapped device for blk_integrity support if
1036 * the underlying devices have an integrity profile. But all devices
1037 * may not have matching profiles (checking all devices isn't reliable
1038 * during table load because this table may use other DM device(s) which
1039 * must be resumed before they will have an initialized integity profile).
1040 * Stacked DM devices force a 2 stage integrity profile validation:
1041 * 1 - during load, validate all initialized integrity profiles match
1042 * 2 - during resume, validate all integrity profiles match
1044 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1046 struct gendisk *template_disk = NULL;
1048 template_disk = dm_table_get_integrity_disk(t, false);
1052 if (!blk_integrity_is_initialized(dm_disk(md))) {
1053 t->integrity_supported = 1;
1054 return blk_integrity_register(dm_disk(md), NULL);
1058 * If DM device already has an initalized integrity
1059 * profile the new profile should not conflict.
1061 if (blk_integrity_is_initialized(template_disk) &&
1062 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1063 DMWARN("%s: conflict with existing integrity profile: "
1064 "%s profile mismatch",
1065 dm_device_name(t->md),
1066 template_disk->disk_name);
1070 /* Preserve existing initialized integrity profile */
1071 t->integrity_supported = 1;
1076 * Prepares the table for use by building the indices,
1077 * setting the type, and allocating mempools.
1079 int dm_table_complete(struct dm_table *t)
1083 r = dm_table_set_type(t);
1085 DMERR("unable to set table type");
1089 r = dm_table_build_index(t);
1091 DMERR("unable to build btrees");
1095 r = dm_table_prealloc_integrity(t, t->md);
1097 DMERR("could not register integrity profile.");
1101 r = dm_table_alloc_md_mempools(t);
1103 DMERR("unable to allocate mempools");
1108 static DEFINE_MUTEX(_event_lock);
1109 void dm_table_event_callback(struct dm_table *t,
1110 void (*fn)(void *), void *context)
1112 mutex_lock(&_event_lock);
1114 t->event_context = context;
1115 mutex_unlock(&_event_lock);
1118 void dm_table_event(struct dm_table *t)
1121 * You can no longer call dm_table_event() from interrupt
1122 * context, use a bottom half instead.
1124 BUG_ON(in_interrupt());
1126 mutex_lock(&_event_lock);
1128 t->event_fn(t->event_context);
1129 mutex_unlock(&_event_lock);
1131 EXPORT_SYMBOL(dm_table_event);
1133 sector_t dm_table_get_size(struct dm_table *t)
1135 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1137 EXPORT_SYMBOL(dm_table_get_size);
1139 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1141 if (index >= t->num_targets)
1144 return t->targets + index;
1148 * Search the btree for the correct target.
1150 * Caller should check returned pointer with dm_target_is_valid()
1151 * to trap I/O beyond end of device.
1153 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1155 unsigned int l, n = 0, k = 0;
1158 for (l = 0; l < t->depth; l++) {
1159 n = get_child(n, k);
1160 node = get_node(t, l, n);
1162 for (k = 0; k < KEYS_PER_NODE; k++)
1163 if (node[k] >= sector)
1167 return &t->targets[(KEYS_PER_NODE * n) + k];
1170 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1171 sector_t start, sector_t len, void *data)
1173 unsigned *num_devices = data;
1181 * Check whether a table has no data devices attached using each
1182 * target's iterate_devices method.
1183 * Returns false if the result is unknown because a target doesn't
1184 * support iterate_devices.
1186 bool dm_table_has_no_data_devices(struct dm_table *table)
1188 struct dm_target *uninitialized_var(ti);
1189 unsigned i = 0, num_devices = 0;
1191 while (i < dm_table_get_num_targets(table)) {
1192 ti = dm_table_get_target(table, i++);
1194 if (!ti->type->iterate_devices)
1197 ti->type->iterate_devices(ti, count_device, &num_devices);
1206 * Establish the new table's queue_limits and validate them.
1208 int dm_calculate_queue_limits(struct dm_table *table,
1209 struct queue_limits *limits)
1211 struct dm_target *uninitialized_var(ti);
1212 struct queue_limits ti_limits;
1215 blk_set_stacking_limits(limits);
1217 while (i < dm_table_get_num_targets(table)) {
1218 blk_set_stacking_limits(&ti_limits);
1220 ti = dm_table_get_target(table, i++);
1222 if (!ti->type->iterate_devices)
1223 goto combine_limits;
1226 * Combine queue limits of all the devices this target uses.
1228 ti->type->iterate_devices(ti, dm_set_device_limits,
1231 /* Set I/O hints portion of queue limits */
1232 if (ti->type->io_hints)
1233 ti->type->io_hints(ti, &ti_limits);
1236 * Check each device area is consistent with the target's
1237 * overall queue limits.
1239 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1245 * Merge this target's queue limits into the overall limits
1248 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1249 DMWARN("%s: adding target device "
1250 "(start sect %llu len %llu) "
1251 "caused an alignment inconsistency",
1252 dm_device_name(table->md),
1253 (unsigned long long) ti->begin,
1254 (unsigned long long) ti->len);
1257 return validate_hardware_logical_block_alignment(table, limits);
1261 * Set the integrity profile for this device if all devices used have
1262 * matching profiles. We're quite deep in the resume path but still
1263 * don't know if all devices (particularly DM devices this device
1264 * may be stacked on) have matching profiles. Even if the profiles
1265 * don't match we have no way to fail (to resume) at this point.
1267 static void dm_table_set_integrity(struct dm_table *t)
1269 struct gendisk *template_disk = NULL;
1271 if (!blk_get_integrity(dm_disk(t->md)))
1274 template_disk = dm_table_get_integrity_disk(t, true);
1276 blk_integrity_register(dm_disk(t->md),
1277 blk_get_integrity(template_disk));
1278 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1279 DMWARN("%s: device no longer has a valid integrity profile",
1280 dm_device_name(t->md));
1282 DMWARN("%s: unable to establish an integrity profile",
1283 dm_device_name(t->md));
1286 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1287 sector_t start, sector_t len, void *data)
1289 unsigned flush = (*(unsigned *)data);
1290 struct request_queue *q = bdev_get_queue(dev->bdev);
1292 return q && (q->flush_flags & flush);
1295 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1297 struct dm_target *ti;
1301 * Require at least one underlying device to support flushes.
1302 * t->devices includes internal dm devices such as mirror logs
1303 * so we need to use iterate_devices here, which targets
1304 * supporting flushes must provide.
1306 while (i < dm_table_get_num_targets(t)) {
1307 ti = dm_table_get_target(t, i++);
1309 if (!ti->num_flush_bios)
1312 if (ti->flush_supported)
1315 if (ti->type->iterate_devices &&
1316 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1323 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1325 struct dm_target *ti;
1328 /* Ensure that all targets supports discard_zeroes_data. */
1329 while (i < dm_table_get_num_targets(t)) {
1330 ti = dm_table_get_target(t, i++);
1332 if (ti->discard_zeroes_data_unsupported)
1339 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1340 sector_t start, sector_t len, void *data)
1342 struct request_queue *q = bdev_get_queue(dev->bdev);
1344 return q && blk_queue_nonrot(q);
1347 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1348 sector_t start, sector_t len, void *data)
1350 struct request_queue *q = bdev_get_queue(dev->bdev);
1352 return q && !blk_queue_add_random(q);
1355 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1356 sector_t start, sector_t len, void *data)
1358 struct request_queue *q = bdev_get_queue(dev->bdev);
1360 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1363 static bool dm_table_all_devices_attribute(struct dm_table *t,
1364 iterate_devices_callout_fn func)
1366 struct dm_target *ti;
1369 while (i < dm_table_get_num_targets(t)) {
1370 ti = dm_table_get_target(t, i++);
1372 if (!ti->type->iterate_devices ||
1373 !ti->type->iterate_devices(ti, func, NULL))
1380 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1381 sector_t start, sector_t len, void *data)
1383 struct request_queue *q = bdev_get_queue(dev->bdev);
1385 return q && !q->limits.max_write_same_sectors;
1388 static bool dm_table_supports_write_same(struct dm_table *t)
1390 struct dm_target *ti;
1393 while (i < dm_table_get_num_targets(t)) {
1394 ti = dm_table_get_target(t, i++);
1396 if (!ti->num_write_same_bios)
1399 if (!ti->type->iterate_devices ||
1400 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1407 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1408 sector_t start, sector_t len, void *data)
1410 struct request_queue *q = bdev_get_queue(dev->bdev);
1412 return q && blk_queue_discard(q);
1415 static bool dm_table_supports_discards(struct dm_table *t)
1417 struct dm_target *ti;
1421 * Unless any target used by the table set discards_supported,
1422 * require at least one underlying device to support discards.
1423 * t->devices includes internal dm devices such as mirror logs
1424 * so we need to use iterate_devices here, which targets
1425 * supporting discard selectively must provide.
1427 while (i < dm_table_get_num_targets(t)) {
1428 ti = dm_table_get_target(t, i++);
1430 if (!ti->num_discard_bios)
1433 if (ti->discards_supported)
1436 if (ti->type->iterate_devices &&
1437 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1444 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1445 struct queue_limits *limits)
1450 * Copy table's limits to the DM device's request_queue
1452 q->limits = *limits;
1454 if (!dm_table_supports_discards(t))
1455 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1457 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1459 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1461 if (dm_table_supports_flush(t, REQ_FUA))
1464 blk_queue_flush(q, flush);
1466 if (!dm_table_discard_zeroes_data(t))
1467 q->limits.discard_zeroes_data = 0;
1469 /* Ensure that all underlying devices are non-rotational. */
1470 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1471 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1473 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1475 if (!dm_table_supports_write_same(t))
1476 q->limits.max_write_same_sectors = 0;
1478 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1479 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1481 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1483 dm_table_set_integrity(t);
1486 * Determine whether or not this queue's I/O timings contribute
1487 * to the entropy pool, Only request-based targets use this.
1488 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1491 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1492 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1495 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1496 * visible to other CPUs because, once the flag is set, incoming bios
1497 * are processed by request-based dm, which refers to the queue
1499 * Until the flag set, bios are passed to bio-based dm and queued to
1500 * md->deferred where queue settings are not needed yet.
1501 * Those bios are passed to request-based dm at the resume time.
1504 if (dm_table_request_based(t))
1505 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1508 unsigned int dm_table_get_num_targets(struct dm_table *t)
1510 return t->num_targets;
1513 struct list_head *dm_table_get_devices(struct dm_table *t)
1518 fmode_t dm_table_get_mode(struct dm_table *t)
1522 EXPORT_SYMBOL(dm_table_get_mode);
1530 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1532 int i = t->num_targets;
1533 struct dm_target *ti = t->targets;
1538 if (ti->type->presuspend)
1539 ti->type->presuspend(ti);
1541 case PRESUSPEND_UNDO:
1542 if (ti->type->presuspend_undo)
1543 ti->type->presuspend_undo(ti);
1546 if (ti->type->postsuspend)
1547 ti->type->postsuspend(ti);
1554 void dm_table_presuspend_targets(struct dm_table *t)
1559 suspend_targets(t, PRESUSPEND);
1562 void dm_table_presuspend_undo_targets(struct dm_table *t)
1567 suspend_targets(t, PRESUSPEND_UNDO);
1570 void dm_table_postsuspend_targets(struct dm_table *t)
1575 suspend_targets(t, POSTSUSPEND);
1578 int dm_table_resume_targets(struct dm_table *t)
1582 for (i = 0; i < t->num_targets; i++) {
1583 struct dm_target *ti = t->targets + i;
1585 if (!ti->type->preresume)
1588 r = ti->type->preresume(ti);
1590 DMERR("%s: %s: preresume failed, error = %d",
1591 dm_device_name(t->md), ti->type->name, r);
1596 for (i = 0; i < t->num_targets; i++) {
1597 struct dm_target *ti = t->targets + i;
1599 if (ti->type->resume)
1600 ti->type->resume(ti);
1606 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1608 list_add(&cb->list, &t->target_callbacks);
1610 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1612 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1614 struct dm_dev_internal *dd;
1615 struct list_head *devices = dm_table_get_devices(t);
1616 struct dm_target_callbacks *cb;
1619 list_for_each_entry(dd, devices, list) {
1620 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1621 char b[BDEVNAME_SIZE];
1624 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1626 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1627 dm_device_name(t->md),
1628 bdevname(dd->dm_dev->bdev, b));
1631 list_for_each_entry(cb, &t->target_callbacks, list)
1632 if (cb->congested_fn)
1633 r |= cb->congested_fn(cb, bdi_bits);
1638 int dm_table_any_busy_target(struct dm_table *t)
1641 struct dm_target *ti;
1643 for (i = 0; i < t->num_targets; i++) {
1644 ti = t->targets + i;
1645 if (ti->type->busy && ti->type->busy(ti))
1652 struct mapped_device *dm_table_get_md(struct dm_table *t)
1656 EXPORT_SYMBOL(dm_table_get_md);
1658 void dm_table_run_md_queue_async(struct dm_table *t)
1660 struct mapped_device *md;
1661 struct request_queue *queue;
1662 unsigned long flags;
1664 if (!dm_table_request_based(t))
1667 md = dm_table_get_md(t);
1668 queue = dm_get_md_queue(md);
1670 spin_lock_irqsave(queue->queue_lock, flags);
1671 blk_run_queue_async(queue);
1672 spin_unlock_irqrestore(queue->queue_lock, flags);
1675 EXPORT_SYMBOL(dm_table_run_md_queue_async);