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>
21 #include <linux/blk-mq.h>
22 #include <linux/mount.h>
24 #define DM_MSG_PREFIX "table"
27 #define NODE_SIZE L1_CACHE_BYTES
28 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
29 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
32 struct mapped_device *md;
33 enum dm_queue_mode type;
37 unsigned int counts[MAX_DEPTH]; /* in nodes */
38 sector_t *index[MAX_DEPTH];
40 unsigned int num_targets;
41 unsigned int num_allocated;
43 struct dm_target *targets;
45 struct target_type *immutable_target_type;
47 bool integrity_supported:1;
50 unsigned integrity_added:1;
53 * Indicates the rw permissions for the new logical
54 * device. This should be a combination of FMODE_READ
59 /* a list of devices used by this table */
60 struct list_head devices;
62 /* events get handed up using this callback */
63 void (*event_fn)(void *);
66 struct dm_md_mempools *mempools;
68 struct list_head target_callbacks;
72 * Similar to ceiling(log_size(n))
74 static unsigned int int_log(unsigned int n, unsigned int base)
79 n = dm_div_up(n, base);
87 * Calculate the index of the child node of the n'th node k'th key.
89 static inline unsigned int get_child(unsigned int n, unsigned int k)
91 return (n * CHILDREN_PER_NODE) + k;
95 * Return the n'th node of level l from table t.
97 static inline sector_t *get_node(struct dm_table *t,
98 unsigned int l, unsigned int n)
100 return t->index[l] + (n * KEYS_PER_NODE);
104 * Return the highest key that you could lookup from the n'th
105 * node on level l of the btree.
107 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
109 for (; l < t->depth - 1; l++)
110 n = get_child(n, CHILDREN_PER_NODE - 1);
112 if (n >= t->counts[l])
113 return (sector_t) - 1;
115 return get_node(t, l, n)[KEYS_PER_NODE - 1];
119 * Fills in a level of the btree based on the highs of the level
122 static int setup_btree_index(unsigned int l, struct dm_table *t)
127 for (n = 0U; n < t->counts[l]; n++) {
128 node = get_node(t, l, n);
130 for (k = 0U; k < KEYS_PER_NODE; k++)
131 node[k] = high(t, l + 1, get_child(n, k));
137 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
143 * Check that we're not going to overflow.
145 if (nmemb > (ULONG_MAX / elem_size))
148 size = nmemb * elem_size;
149 addr = vzalloc(size);
153 EXPORT_SYMBOL(dm_vcalloc);
156 * highs, and targets are managed as dynamic arrays during a
159 static int alloc_targets(struct dm_table *t, unsigned int num)
162 struct dm_target *n_targets;
165 * Allocate both the target array and offset array at once.
166 * Append an empty entry to catch sectors beyond the end of
169 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
174 n_targets = (struct dm_target *) (n_highs + num);
176 memset(n_highs, -1, sizeof(*n_highs) * num);
179 t->num_allocated = num;
181 t->targets = n_targets;
186 int dm_table_create(struct dm_table **result, fmode_t mode,
187 unsigned num_targets, struct mapped_device *md)
189 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
194 INIT_LIST_HEAD(&t->devices);
195 INIT_LIST_HEAD(&t->target_callbacks);
198 num_targets = KEYS_PER_NODE;
200 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
207 if (alloc_targets(t, num_targets)) {
212 t->type = DM_TYPE_NONE;
219 static void free_devices(struct list_head *devices, struct mapped_device *md)
221 struct list_head *tmp, *next;
223 list_for_each_safe(tmp, next, devices) {
224 struct dm_dev_internal *dd =
225 list_entry(tmp, struct dm_dev_internal, list);
226 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
227 dm_device_name(md), dd->dm_dev->name);
228 dm_put_table_device(md, dd->dm_dev);
233 void dm_table_destroy(struct dm_table *t)
240 /* free the indexes */
242 vfree(t->index[t->depth - 2]);
244 /* free the targets */
245 for (i = 0; i < t->num_targets; i++) {
246 struct dm_target *tgt = t->targets + i;
251 dm_put_target_type(tgt->type);
256 /* free the device list */
257 free_devices(&t->devices, t->md);
259 dm_free_md_mempools(t->mempools);
265 * See if we've already got a device in the list.
267 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
269 struct dm_dev_internal *dd;
271 list_for_each_entry (dd, l, list)
272 if (dd->dm_dev->bdev->bd_dev == dev)
279 * If possible, this checks an area of a destination device is invalid.
281 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
282 sector_t start, sector_t len, void *data)
284 struct request_queue *q;
285 struct queue_limits *limits = data;
286 struct block_device *bdev = dev->bdev;
288 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
289 unsigned short logical_block_size_sectors =
290 limits->logical_block_size >> SECTOR_SHIFT;
291 char b[BDEVNAME_SIZE];
294 * Some devices exist without request functions,
295 * such as loop devices not yet bound to backing files.
296 * Forbid the use of such devices.
298 q = bdev_get_queue(bdev);
299 if (!q || !q->make_request_fn) {
300 DMWARN("%s: %s is not yet initialised: "
301 "start=%llu, len=%llu, dev_size=%llu",
302 dm_device_name(ti->table->md), bdevname(bdev, b),
303 (unsigned long long)start,
304 (unsigned long long)len,
305 (unsigned long long)dev_size);
312 if ((start >= dev_size) || (start + len > dev_size)) {
313 DMWARN("%s: %s too small for target: "
314 "start=%llu, len=%llu, dev_size=%llu",
315 dm_device_name(ti->table->md), bdevname(bdev, b),
316 (unsigned long long)start,
317 (unsigned long long)len,
318 (unsigned long long)dev_size);
322 if (logical_block_size_sectors <= 1)
325 if (start & (logical_block_size_sectors - 1)) {
326 DMWARN("%s: start=%llu not aligned to h/w "
327 "logical block size %u of %s",
328 dm_device_name(ti->table->md),
329 (unsigned long long)start,
330 limits->logical_block_size, bdevname(bdev, b));
334 if (len & (logical_block_size_sectors - 1)) {
335 DMWARN("%s: len=%llu not aligned to h/w "
336 "logical block size %u of %s",
337 dm_device_name(ti->table->md),
338 (unsigned long long)len,
339 limits->logical_block_size, bdevname(bdev, b));
347 * This upgrades the mode on an already open dm_dev, being
348 * careful to leave things as they were if we fail to reopen the
349 * device and not to touch the existing bdev field in case
350 * it is accessed concurrently inside dm_table_any_congested().
352 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
353 struct mapped_device *md)
356 struct dm_dev *old_dev, *new_dev;
358 old_dev = dd->dm_dev;
360 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
361 dd->dm_dev->mode | new_mode, &new_dev);
365 dd->dm_dev = new_dev;
366 dm_put_table_device(md, old_dev);
372 * Convert the path to a device
374 dev_t dm_get_dev_t(const char *path)
377 struct block_device *bdev;
379 bdev = lookup_bdev(path);
381 dev = name_to_dev_t(path);
389 EXPORT_SYMBOL_GPL(dm_get_dev_t);
392 * Add a device to the list, or just increment the usage count if
393 * it's already present.
395 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
396 struct dm_dev **result)
400 struct dm_dev_internal *dd;
401 struct dm_table *t = ti->table;
405 dev = dm_get_dev_t(path);
409 dd = find_device(&t->devices, dev);
411 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
415 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
420 atomic_set(&dd->count, 0);
421 list_add(&dd->list, &t->devices);
423 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
424 r = upgrade_mode(dd, mode, t->md);
428 atomic_inc(&dd->count);
430 *result = dd->dm_dev;
433 EXPORT_SYMBOL(dm_get_device);
435 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
436 sector_t start, sector_t len, void *data)
438 struct queue_limits *limits = data;
439 struct block_device *bdev = dev->bdev;
440 struct request_queue *q = bdev_get_queue(bdev);
441 char b[BDEVNAME_SIZE];
444 DMWARN("%s: Cannot set limits for nonexistent device %s",
445 dm_device_name(ti->table->md), bdevname(bdev, b));
449 if (bdev_stack_limits(limits, bdev, start) < 0)
450 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
451 "physical_block_size=%u, logical_block_size=%u, "
452 "alignment_offset=%u, start=%llu",
453 dm_device_name(ti->table->md), bdevname(bdev, b),
454 q->limits.physical_block_size,
455 q->limits.logical_block_size,
456 q->limits.alignment_offset,
457 (unsigned long long) start << SECTOR_SHIFT);
463 * Decrement a device's use count and remove it if necessary.
465 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
468 struct list_head *devices = &ti->table->devices;
469 struct dm_dev_internal *dd;
471 list_for_each_entry(dd, devices, list) {
472 if (dd->dm_dev == d) {
478 DMWARN("%s: device %s not in table devices list",
479 dm_device_name(ti->table->md), d->name);
482 if (atomic_dec_and_test(&dd->count)) {
483 dm_put_table_device(ti->table->md, d);
488 EXPORT_SYMBOL(dm_put_device);
491 * Checks to see if the target joins onto the end of the table.
493 static int adjoin(struct dm_table *table, struct dm_target *ti)
495 struct dm_target *prev;
497 if (!table->num_targets)
500 prev = &table->targets[table->num_targets - 1];
501 return (ti->begin == (prev->begin + prev->len));
505 * Used to dynamically allocate the arg array.
507 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
508 * process messages even if some device is suspended. These messages have a
509 * small fixed number of arguments.
511 * On the other hand, dm-switch needs to process bulk data using messages and
512 * excessive use of GFP_NOIO could cause trouble.
514 static char **realloc_argv(unsigned *array_size, char **old_argv)
521 new_size = *array_size * 2;
527 argv = kmalloc(new_size * sizeof(*argv), gfp);
529 memcpy(argv, old_argv, *array_size * sizeof(*argv));
530 *array_size = new_size;
538 * Destructively splits up the argument list to pass to ctr.
540 int dm_split_args(int *argc, char ***argvp, char *input)
542 char *start, *end = input, *out, **argv = NULL;
543 unsigned array_size = 0;
552 argv = realloc_argv(&array_size, argv);
557 /* Skip whitespace */
558 start = skip_spaces(end);
561 break; /* success, we hit the end */
563 /* 'out' is used to remove any back-quotes */
566 /* Everything apart from '\0' can be quoted */
567 if (*end == '\\' && *(end + 1)) {
574 break; /* end of token */
579 /* have we already filled the array ? */
580 if ((*argc + 1) > array_size) {
581 argv = realloc_argv(&array_size, argv);
586 /* we know this is whitespace */
590 /* terminate the string and put it in the array */
601 * Impose necessary and sufficient conditions on a devices's table such
602 * that any incoming bio which respects its logical_block_size can be
603 * processed successfully. If it falls across the boundary between
604 * two or more targets, the size of each piece it gets split into must
605 * be compatible with the logical_block_size of the target processing it.
607 static int validate_hardware_logical_block_alignment(struct dm_table *table,
608 struct queue_limits *limits)
611 * This function uses arithmetic modulo the logical_block_size
612 * (in units of 512-byte sectors).
614 unsigned short device_logical_block_size_sects =
615 limits->logical_block_size >> SECTOR_SHIFT;
618 * Offset of the start of the next table entry, mod logical_block_size.
620 unsigned short next_target_start = 0;
623 * Given an aligned bio that extends beyond the end of a
624 * target, how many sectors must the next target handle?
626 unsigned short remaining = 0;
628 struct dm_target *uninitialized_var(ti);
629 struct queue_limits ti_limits;
633 * Check each entry in the table in turn.
635 for (i = 0; i < dm_table_get_num_targets(table); i++) {
636 ti = dm_table_get_target(table, i);
638 blk_set_stacking_limits(&ti_limits);
640 /* combine all target devices' limits */
641 if (ti->type->iterate_devices)
642 ti->type->iterate_devices(ti, dm_set_device_limits,
646 * If the remaining sectors fall entirely within this
647 * table entry are they compatible with its logical_block_size?
649 if (remaining < ti->len &&
650 remaining & ((ti_limits.logical_block_size >>
655 (unsigned short) ((next_target_start + ti->len) &
656 (device_logical_block_size_sects - 1));
657 remaining = next_target_start ?
658 device_logical_block_size_sects - next_target_start : 0;
662 DMWARN("%s: table line %u (start sect %llu len %llu) "
663 "not aligned to h/w logical block size %u",
664 dm_device_name(table->md), i,
665 (unsigned long long) ti->begin,
666 (unsigned long long) ti->len,
667 limits->logical_block_size);
674 int dm_table_add_target(struct dm_table *t, const char *type,
675 sector_t start, sector_t len, char *params)
677 int r = -EINVAL, argc;
679 struct dm_target *tgt;
682 DMERR("%s: target type %s must appear alone in table",
683 dm_device_name(t->md), t->targets->type->name);
687 BUG_ON(t->num_targets >= t->num_allocated);
689 tgt = t->targets + t->num_targets;
690 memset(tgt, 0, sizeof(*tgt));
693 DMERR("%s: zero-length target", dm_device_name(t->md));
697 tgt->type = dm_get_target_type(type);
699 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
703 if (dm_target_needs_singleton(tgt->type)) {
704 if (t->num_targets) {
705 tgt->error = "singleton target type must appear alone in table";
711 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
712 tgt->error = "target type may not be included in a read-only table";
716 if (t->immutable_target_type) {
717 if (t->immutable_target_type != tgt->type) {
718 tgt->error = "immutable target type cannot be mixed with other target types";
721 } else if (dm_target_is_immutable(tgt->type)) {
722 if (t->num_targets) {
723 tgt->error = "immutable target type cannot be mixed with other target types";
726 t->immutable_target_type = tgt->type;
729 if (dm_target_has_integrity(tgt->type))
730 t->integrity_added = 1;
735 tgt->error = "Unknown error";
738 * Does this target adjoin the previous one ?
740 if (!adjoin(t, tgt)) {
741 tgt->error = "Gap in table";
745 r = dm_split_args(&argc, &argv, params);
747 tgt->error = "couldn't split parameters (insufficient memory)";
751 r = tgt->type->ctr(tgt, argc, argv);
756 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
758 if (!tgt->num_discard_bios && tgt->discards_supported)
759 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
760 dm_device_name(t->md), type);
765 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
766 dm_put_target_type(tgt->type);
771 * Target argument parsing helpers.
773 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
774 unsigned *value, char **error, unsigned grouped)
776 const char *arg_str = dm_shift_arg(arg_set);
780 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
781 (*value < arg->min) ||
782 (*value > arg->max) ||
783 (grouped && arg_set->argc < *value)) {
791 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
792 unsigned *value, char **error)
794 return validate_next_arg(arg, arg_set, value, error, 0);
796 EXPORT_SYMBOL(dm_read_arg);
798 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
799 unsigned *value, char **error)
801 return validate_next_arg(arg, arg_set, value, error, 1);
803 EXPORT_SYMBOL(dm_read_arg_group);
805 const char *dm_shift_arg(struct dm_arg_set *as)
818 EXPORT_SYMBOL(dm_shift_arg);
820 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
822 BUG_ON(as->argc < num_args);
823 as->argc -= num_args;
824 as->argv += num_args;
826 EXPORT_SYMBOL(dm_consume_args);
828 static bool __table_type_bio_based(enum dm_queue_mode table_type)
830 return (table_type == DM_TYPE_BIO_BASED ||
831 table_type == DM_TYPE_DAX_BIO_BASED);
834 static bool __table_type_request_based(enum dm_queue_mode table_type)
836 return (table_type == DM_TYPE_REQUEST_BASED ||
837 table_type == DM_TYPE_MQ_REQUEST_BASED);
840 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
844 EXPORT_SYMBOL_GPL(dm_table_set_type);
846 static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
847 sector_t start, sector_t len, void *data)
849 struct request_queue *q = bdev_get_queue(dev->bdev);
851 return q && blk_queue_dax(q);
854 static bool dm_table_supports_dax(struct dm_table *t)
856 struct dm_target *ti;
859 /* Ensure that all targets support DAX. */
860 for (i = 0; i < dm_table_get_num_targets(t); i++) {
861 ti = dm_table_get_target(t, i);
863 if (!ti->type->direct_access)
866 if (!ti->type->iterate_devices ||
867 !ti->type->iterate_devices(ti, device_supports_dax, NULL))
874 static int dm_table_determine_type(struct dm_table *t)
877 unsigned bio_based = 0, request_based = 0, hybrid = 0;
878 unsigned sq_count = 0, mq_count = 0;
879 struct dm_target *tgt;
880 struct dm_dev_internal *dd;
881 struct list_head *devices = dm_table_get_devices(t);
882 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
884 if (t->type != DM_TYPE_NONE) {
885 /* target already set the table's type */
886 if (t->type == DM_TYPE_BIO_BASED)
888 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
889 goto verify_rq_based;
892 for (i = 0; i < t->num_targets; i++) {
893 tgt = t->targets + i;
894 if (dm_target_hybrid(tgt))
896 else if (dm_target_request_based(tgt))
901 if (bio_based && request_based) {
902 DMWARN("Inconsistent table: different target types"
903 " can't be mixed up");
908 if (hybrid && !bio_based && !request_based) {
910 * The targets can work either way.
911 * Determine the type from the live device.
912 * Default to bio-based if device is new.
914 if (__table_type_request_based(live_md_type))
921 /* We must use this table as bio-based */
922 t->type = DM_TYPE_BIO_BASED;
923 if (dm_table_supports_dax(t) ||
924 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
925 t->type = DM_TYPE_DAX_BIO_BASED;
929 BUG_ON(!request_based); /* No targets in this table */
932 * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
933 * having a compatible target use dm_table_set_type.
935 t->type = DM_TYPE_REQUEST_BASED;
939 * Request-based dm supports only tables that have a single target now.
940 * To support multiple targets, request splitting support is needed,
941 * and that needs lots of changes in the block-layer.
942 * (e.g. request completion process for partial completion.)
944 if (t->num_targets > 1) {
945 DMWARN("Request-based dm doesn't support multiple targets yet");
949 if (list_empty(devices)) {
951 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
953 /* inherit live table's type and all_blk_mq */
955 t->type = live_table->type;
956 t->all_blk_mq = live_table->all_blk_mq;
958 dm_put_live_table(t->md, srcu_idx);
962 /* Non-request-stackable devices can't be used for request-based dm */
963 list_for_each_entry(dd, devices, list) {
964 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
966 if (!blk_queue_stackable(q)) {
967 DMERR("table load rejected: including"
968 " non-request-stackable devices");
977 if (sq_count && mq_count) {
978 DMERR("table load rejected: not all devices are blk-mq request-stackable");
981 t->all_blk_mq = mq_count > 0;
983 if (t->type == DM_TYPE_MQ_REQUEST_BASED && !t->all_blk_mq) {
984 DMERR("table load rejected: all devices are not blk-mq request-stackable");
991 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
996 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
998 return t->immutable_target_type;
1001 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1003 /* Immutable target is implicitly a singleton */
1004 if (t->num_targets > 1 ||
1005 !dm_target_is_immutable(t->targets[0].type))
1011 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1013 struct dm_target *ti;
1016 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1017 ti = dm_table_get_target(t, i);
1018 if (dm_target_is_wildcard(ti->type))
1025 bool dm_table_bio_based(struct dm_table *t)
1027 return __table_type_bio_based(dm_table_get_type(t));
1030 bool dm_table_request_based(struct dm_table *t)
1032 return __table_type_request_based(dm_table_get_type(t));
1035 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1037 return t->all_blk_mq;
1040 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1042 enum dm_queue_mode type = dm_table_get_type(t);
1043 unsigned per_io_data_size = 0;
1044 struct dm_target *tgt;
1047 if (unlikely(type == DM_TYPE_NONE)) {
1048 DMWARN("no table type is set, can't allocate mempools");
1052 if (__table_type_bio_based(type))
1053 for (i = 0; i < t->num_targets; i++) {
1054 tgt = t->targets + i;
1055 per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1058 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1065 void dm_table_free_md_mempools(struct dm_table *t)
1067 dm_free_md_mempools(t->mempools);
1071 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1076 static int setup_indexes(struct dm_table *t)
1079 unsigned int total = 0;
1082 /* allocate the space for *all* the indexes */
1083 for (i = t->depth - 2; i >= 0; i--) {
1084 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1085 total += t->counts[i];
1088 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1092 /* set up internal nodes, bottom-up */
1093 for (i = t->depth - 2; i >= 0; i--) {
1094 t->index[i] = indexes;
1095 indexes += (KEYS_PER_NODE * t->counts[i]);
1096 setup_btree_index(i, t);
1103 * Builds the btree to index the map.
1105 static int dm_table_build_index(struct dm_table *t)
1108 unsigned int leaf_nodes;
1110 /* how many indexes will the btree have ? */
1111 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1112 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1114 /* leaf layer has already been set up */
1115 t->counts[t->depth - 1] = leaf_nodes;
1116 t->index[t->depth - 1] = t->highs;
1119 r = setup_indexes(t);
1124 static bool integrity_profile_exists(struct gendisk *disk)
1126 return !!blk_get_integrity(disk);
1130 * Get a disk whose integrity profile reflects the table's profile.
1131 * Returns NULL if integrity support was inconsistent or unavailable.
1133 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1135 struct list_head *devices = dm_table_get_devices(t);
1136 struct dm_dev_internal *dd = NULL;
1137 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1140 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1141 struct dm_target *ti = dm_table_get_target(t, i);
1142 if (!dm_target_passes_integrity(ti->type))
1146 list_for_each_entry(dd, devices, list) {
1147 template_disk = dd->dm_dev->bdev->bd_disk;
1148 if (!integrity_profile_exists(template_disk))
1150 else if (prev_disk &&
1151 blk_integrity_compare(prev_disk, template_disk) < 0)
1153 prev_disk = template_disk;
1156 return template_disk;
1160 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1161 dm_device_name(t->md),
1162 prev_disk->disk_name,
1163 template_disk->disk_name);
1168 * Register the mapped device for blk_integrity support if the
1169 * underlying devices have an integrity profile. But all devices may
1170 * not have matching profiles (checking all devices isn't reliable
1171 * during table load because this table may use other DM device(s) which
1172 * must be resumed before they will have an initialized integity
1173 * profile). Consequently, stacked DM devices force a 2 stage integrity
1174 * profile validation: First pass during table load, final pass during
1177 static int dm_table_register_integrity(struct dm_table *t)
1179 struct mapped_device *md = t->md;
1180 struct gendisk *template_disk = NULL;
1182 /* If target handles integrity itself do not register it here. */
1183 if (t->integrity_added)
1186 template_disk = dm_table_get_integrity_disk(t);
1190 if (!integrity_profile_exists(dm_disk(md))) {
1191 t->integrity_supported = true;
1193 * Register integrity profile during table load; we can do
1194 * this because the final profile must match during resume.
1196 blk_integrity_register(dm_disk(md),
1197 blk_get_integrity(template_disk));
1202 * If DM device already has an initialized integrity
1203 * profile the new profile should not conflict.
1205 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1206 DMWARN("%s: conflict with existing integrity profile: "
1207 "%s profile mismatch",
1208 dm_device_name(t->md),
1209 template_disk->disk_name);
1213 /* Preserve existing integrity profile */
1214 t->integrity_supported = true;
1219 * Prepares the table for use by building the indices,
1220 * setting the type, and allocating mempools.
1222 int dm_table_complete(struct dm_table *t)
1226 r = dm_table_determine_type(t);
1228 DMERR("unable to determine table type");
1232 r = dm_table_build_index(t);
1234 DMERR("unable to build btrees");
1238 r = dm_table_register_integrity(t);
1240 DMERR("could not register integrity profile.");
1244 r = dm_table_alloc_md_mempools(t, t->md);
1246 DMERR("unable to allocate mempools");
1251 static DEFINE_MUTEX(_event_lock);
1252 void dm_table_event_callback(struct dm_table *t,
1253 void (*fn)(void *), void *context)
1255 mutex_lock(&_event_lock);
1257 t->event_context = context;
1258 mutex_unlock(&_event_lock);
1261 void dm_table_event(struct dm_table *t)
1264 * You can no longer call dm_table_event() from interrupt
1265 * context, use a bottom half instead.
1267 BUG_ON(in_interrupt());
1269 mutex_lock(&_event_lock);
1271 t->event_fn(t->event_context);
1272 mutex_unlock(&_event_lock);
1274 EXPORT_SYMBOL(dm_table_event);
1276 sector_t dm_table_get_size(struct dm_table *t)
1278 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1280 EXPORT_SYMBOL(dm_table_get_size);
1282 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1284 if (index >= t->num_targets)
1287 return t->targets + index;
1291 * Search the btree for the correct target.
1293 * Caller should check returned pointer with dm_target_is_valid()
1294 * to trap I/O beyond end of device.
1296 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1298 unsigned int l, n = 0, k = 0;
1301 for (l = 0; l < t->depth; l++) {
1302 n = get_child(n, k);
1303 node = get_node(t, l, n);
1305 for (k = 0; k < KEYS_PER_NODE; k++)
1306 if (node[k] >= sector)
1310 return &t->targets[(KEYS_PER_NODE * n) + k];
1313 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1314 sector_t start, sector_t len, void *data)
1316 unsigned *num_devices = data;
1324 * Check whether a table has no data devices attached using each
1325 * target's iterate_devices method.
1326 * Returns false if the result is unknown because a target doesn't
1327 * support iterate_devices.
1329 bool dm_table_has_no_data_devices(struct dm_table *table)
1331 struct dm_target *ti;
1332 unsigned i, num_devices;
1334 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1335 ti = dm_table_get_target(table, i);
1337 if (!ti->type->iterate_devices)
1341 ti->type->iterate_devices(ti, count_device, &num_devices);
1350 * Establish the new table's queue_limits and validate them.
1352 int dm_calculate_queue_limits(struct dm_table *table,
1353 struct queue_limits *limits)
1355 struct dm_target *ti;
1356 struct queue_limits ti_limits;
1359 blk_set_stacking_limits(limits);
1361 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1362 blk_set_stacking_limits(&ti_limits);
1364 ti = dm_table_get_target(table, i);
1366 if (!ti->type->iterate_devices)
1367 goto combine_limits;
1370 * Combine queue limits of all the devices this target uses.
1372 ti->type->iterate_devices(ti, dm_set_device_limits,
1375 /* Set I/O hints portion of queue limits */
1376 if (ti->type->io_hints)
1377 ti->type->io_hints(ti, &ti_limits);
1380 * Check each device area is consistent with the target's
1381 * overall queue limits.
1383 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1389 * Merge this target's queue limits into the overall limits
1392 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1393 DMWARN("%s: adding target device "
1394 "(start sect %llu len %llu) "
1395 "caused an alignment inconsistency",
1396 dm_device_name(table->md),
1397 (unsigned long long) ti->begin,
1398 (unsigned long long) ti->len);
1401 return validate_hardware_logical_block_alignment(table, limits);
1405 * Verify that all devices have an integrity profile that matches the
1406 * DM device's registered integrity profile. If the profiles don't
1407 * match then unregister the DM device's integrity profile.
1409 static void dm_table_verify_integrity(struct dm_table *t)
1411 struct gendisk *template_disk = NULL;
1413 if (t->integrity_added)
1416 if (t->integrity_supported) {
1418 * Verify that the original integrity profile
1419 * matches all the devices in this table.
1421 template_disk = dm_table_get_integrity_disk(t);
1422 if (template_disk &&
1423 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1427 if (integrity_profile_exists(dm_disk(t->md))) {
1428 DMWARN("%s: unable to establish an integrity profile",
1429 dm_device_name(t->md));
1430 blk_integrity_unregister(dm_disk(t->md));
1434 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1435 sector_t start, sector_t len, void *data)
1437 unsigned long flush = (unsigned long) data;
1438 struct request_queue *q = bdev_get_queue(dev->bdev);
1440 return q && (q->queue_flags & flush);
1443 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1445 struct dm_target *ti;
1449 * Require at least one underlying device to support flushes.
1450 * t->devices includes internal dm devices such as mirror logs
1451 * so we need to use iterate_devices here, which targets
1452 * supporting flushes must provide.
1454 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1455 ti = dm_table_get_target(t, i);
1457 if (!ti->num_flush_bios)
1460 if (ti->flush_supported)
1463 if (ti->type->iterate_devices &&
1464 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1471 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1472 sector_t start, sector_t len, void *data)
1474 struct request_queue *q = bdev_get_queue(dev->bdev);
1476 return q && blk_queue_nonrot(q);
1479 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1480 sector_t start, sector_t len, void *data)
1482 struct request_queue *q = bdev_get_queue(dev->bdev);
1484 return q && !blk_queue_add_random(q);
1487 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1488 sector_t start, sector_t len, void *data)
1490 struct request_queue *q = bdev_get_queue(dev->bdev);
1492 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1495 static bool dm_table_all_devices_attribute(struct dm_table *t,
1496 iterate_devices_callout_fn func)
1498 struct dm_target *ti;
1501 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1502 ti = dm_table_get_target(t, i);
1504 if (!ti->type->iterate_devices ||
1505 !ti->type->iterate_devices(ti, func, NULL))
1512 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1513 sector_t start, sector_t len, void *data)
1515 struct request_queue *q = bdev_get_queue(dev->bdev);
1517 return q && !q->limits.max_write_same_sectors;
1520 static bool dm_table_supports_write_same(struct dm_table *t)
1522 struct dm_target *ti;
1525 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1526 ti = dm_table_get_target(t, i);
1528 if (!ti->num_write_same_bios)
1531 if (!ti->type->iterate_devices ||
1532 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1539 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1540 sector_t start, sector_t len, void *data)
1542 struct request_queue *q = bdev_get_queue(dev->bdev);
1544 return q && !q->limits.max_write_zeroes_sectors;
1547 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1549 struct dm_target *ti;
1552 while (i < dm_table_get_num_targets(t)) {
1553 ti = dm_table_get_target(t, i++);
1555 if (!ti->num_write_zeroes_bios)
1558 if (!ti->type->iterate_devices ||
1559 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1567 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1568 sector_t start, sector_t len, void *data)
1570 struct request_queue *q = bdev_get_queue(dev->bdev);
1572 return q && blk_queue_discard(q);
1575 static bool dm_table_supports_discards(struct dm_table *t)
1577 struct dm_target *ti;
1581 * Unless any target used by the table set discards_supported,
1582 * require at least one underlying device to support discards.
1583 * t->devices includes internal dm devices such as mirror logs
1584 * so we need to use iterate_devices here, which targets
1585 * supporting discard selectively must provide.
1587 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1588 ti = dm_table_get_target(t, i);
1590 if (!ti->num_discard_bios)
1593 if (ti->discards_supported)
1596 if (ti->type->iterate_devices &&
1597 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1604 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1605 struct queue_limits *limits)
1607 bool wc = false, fua = false;
1610 * Copy table's limits to the DM device's request_queue
1612 q->limits = *limits;
1614 if (!dm_table_supports_discards(t))
1615 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1617 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1619 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1621 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1624 blk_queue_write_cache(q, wc, fua);
1626 /* Ensure that all underlying devices are non-rotational. */
1627 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1628 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1630 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1632 if (!dm_table_supports_write_same(t))
1633 q->limits.max_write_same_sectors = 0;
1634 if (!dm_table_supports_write_zeroes(t))
1635 q->limits.max_write_zeroes_sectors = 0;
1637 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1638 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1640 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1642 dm_table_verify_integrity(t);
1645 * Determine whether or not this queue's I/O timings contribute
1646 * to the entropy pool, Only request-based targets use this.
1647 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1650 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1651 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1654 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1655 * visible to other CPUs because, once the flag is set, incoming bios
1656 * are processed by request-based dm, which refers to the queue
1658 * Until the flag set, bios are passed to bio-based dm and queued to
1659 * md->deferred where queue settings are not needed yet.
1660 * Those bios are passed to request-based dm at the resume time.
1663 if (dm_table_request_based(t))
1664 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1667 unsigned int dm_table_get_num_targets(struct dm_table *t)
1669 return t->num_targets;
1672 struct list_head *dm_table_get_devices(struct dm_table *t)
1677 fmode_t dm_table_get_mode(struct dm_table *t)
1681 EXPORT_SYMBOL(dm_table_get_mode);
1689 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1691 int i = t->num_targets;
1692 struct dm_target *ti = t->targets;
1694 lockdep_assert_held(&t->md->suspend_lock);
1699 if (ti->type->presuspend)
1700 ti->type->presuspend(ti);
1702 case PRESUSPEND_UNDO:
1703 if (ti->type->presuspend_undo)
1704 ti->type->presuspend_undo(ti);
1707 if (ti->type->postsuspend)
1708 ti->type->postsuspend(ti);
1715 void dm_table_presuspend_targets(struct dm_table *t)
1720 suspend_targets(t, PRESUSPEND);
1723 void dm_table_presuspend_undo_targets(struct dm_table *t)
1728 suspend_targets(t, PRESUSPEND_UNDO);
1731 void dm_table_postsuspend_targets(struct dm_table *t)
1736 suspend_targets(t, POSTSUSPEND);
1739 int dm_table_resume_targets(struct dm_table *t)
1743 lockdep_assert_held(&t->md->suspend_lock);
1745 for (i = 0; i < t->num_targets; i++) {
1746 struct dm_target *ti = t->targets + i;
1748 if (!ti->type->preresume)
1751 r = ti->type->preresume(ti);
1753 DMERR("%s: %s: preresume failed, error = %d",
1754 dm_device_name(t->md), ti->type->name, r);
1759 for (i = 0; i < t->num_targets; i++) {
1760 struct dm_target *ti = t->targets + i;
1762 if (ti->type->resume)
1763 ti->type->resume(ti);
1769 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1771 list_add(&cb->list, &t->target_callbacks);
1773 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1775 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1777 struct dm_dev_internal *dd;
1778 struct list_head *devices = dm_table_get_devices(t);
1779 struct dm_target_callbacks *cb;
1782 list_for_each_entry(dd, devices, list) {
1783 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1784 char b[BDEVNAME_SIZE];
1787 r |= bdi_congested(q->backing_dev_info, bdi_bits);
1789 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1790 dm_device_name(t->md),
1791 bdevname(dd->dm_dev->bdev, b));
1794 list_for_each_entry(cb, &t->target_callbacks, list)
1795 if (cb->congested_fn)
1796 r |= cb->congested_fn(cb, bdi_bits);
1801 struct mapped_device *dm_table_get_md(struct dm_table *t)
1805 EXPORT_SYMBOL(dm_table_get_md);
1807 void dm_table_run_md_queue_async(struct dm_table *t)
1809 struct mapped_device *md;
1810 struct request_queue *queue;
1811 unsigned long flags;
1813 if (!dm_table_request_based(t))
1816 md = dm_table_get_md(t);
1817 queue = dm_get_md_queue(md);
1820 blk_mq_run_hw_queues(queue, true);
1822 spin_lock_irqsave(queue->queue_lock, flags);
1823 blk_run_queue_async(queue);
1824 spin_unlock_irqrestore(queue->queue_lock, flags);
1828 EXPORT_SYMBOL(dm_table_run_md_queue_async);