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>
23 #include <linux/dax.h>
25 #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 * Similar to ceiling(log_size(n))
34 static unsigned int int_log(unsigned int n, unsigned int base)
39 n = dm_div_up(n, base);
47 * Calculate the index of the child node of the n'th node k'th key.
49 static inline unsigned int get_child(unsigned int n, unsigned int k)
51 return (n * CHILDREN_PER_NODE) + k;
55 * Return the n'th node of level l from table t.
57 static inline sector_t *get_node(struct dm_table *t,
58 unsigned int l, unsigned int n)
60 return t->index[l] + (n * KEYS_PER_NODE);
64 * Return the highest key that you could lookup from the n'th
65 * node on level l of the btree.
67 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
69 for (; l < t->depth - 1; l++)
70 n = get_child(n, CHILDREN_PER_NODE - 1);
72 if (n >= t->counts[l])
73 return (sector_t) - 1;
75 return get_node(t, l, n)[KEYS_PER_NODE - 1];
79 * Fills in a level of the btree based on the highs of the level
82 static int setup_btree_index(unsigned int l, struct dm_table *t)
87 for (n = 0U; n < t->counts[l]; n++) {
88 node = get_node(t, l, n);
90 for (k = 0U; k < KEYS_PER_NODE; k++)
91 node[k] = high(t, l + 1, get_child(n, k));
97 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
103 * Check that we're not going to overflow.
105 if (nmemb > (ULONG_MAX / elem_size))
108 size = nmemb * elem_size;
109 addr = vzalloc(size);
113 EXPORT_SYMBOL(dm_vcalloc);
116 * highs, and targets are managed as dynamic arrays during a
119 static int alloc_targets(struct dm_table *t, unsigned int num)
122 struct dm_target *n_targets;
125 * Allocate both the target array and offset array at once.
127 n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
132 n_targets = (struct dm_target *) (n_highs + num);
134 memset(n_highs, -1, sizeof(*n_highs) * num);
137 t->num_allocated = num;
139 t->targets = n_targets;
144 int dm_table_create(struct dm_table **result, fmode_t mode,
145 unsigned num_targets, struct mapped_device *md)
147 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
152 INIT_LIST_HEAD(&t->devices);
155 num_targets = KEYS_PER_NODE;
157 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
164 if (alloc_targets(t, num_targets)) {
169 t->type = DM_TYPE_NONE;
176 static void free_devices(struct list_head *devices, struct mapped_device *md)
178 struct list_head *tmp, *next;
180 list_for_each_safe(tmp, next, devices) {
181 struct dm_dev_internal *dd =
182 list_entry(tmp, struct dm_dev_internal, list);
183 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
184 dm_device_name(md), dd->dm_dev->name);
185 dm_put_table_device(md, dd->dm_dev);
190 void dm_table_destroy(struct dm_table *t)
197 /* free the indexes */
199 vfree(t->index[t->depth - 2]);
201 /* free the targets */
202 for (i = 0; i < t->num_targets; i++) {
203 struct dm_target *tgt = t->targets + i;
208 dm_put_target_type(tgt->type);
213 /* free the device list */
214 free_devices(&t->devices, t->md);
216 dm_free_md_mempools(t->mempools);
222 * See if we've already got a device in the list.
224 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
226 struct dm_dev_internal *dd;
228 list_for_each_entry (dd, l, list)
229 if (dd->dm_dev->bdev->bd_dev == dev)
236 * If possible, this checks an area of a destination device is invalid.
238 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
239 sector_t start, sector_t len, void *data)
241 struct queue_limits *limits = data;
242 struct block_device *bdev = dev->bdev;
244 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
245 unsigned short logical_block_size_sectors =
246 limits->logical_block_size >> SECTOR_SHIFT;
247 char b[BDEVNAME_SIZE];
252 if ((start >= dev_size) || (start + len > dev_size)) {
253 DMWARN("%s: %s too small for target: "
254 "start=%llu, len=%llu, dev_size=%llu",
255 dm_device_name(ti->table->md), bdevname(bdev, b),
256 (unsigned long long)start,
257 (unsigned long long)len,
258 (unsigned long long)dev_size);
263 * If the target is mapped to zoned block device(s), check
264 * that the zones are not partially mapped.
266 if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
267 unsigned int zone_sectors = bdev_zone_sectors(bdev);
269 if (start & (zone_sectors - 1)) {
270 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
271 dm_device_name(ti->table->md),
272 (unsigned long long)start,
273 zone_sectors, bdevname(bdev, b));
278 * Note: The last zone of a zoned block device may be smaller
279 * than other zones. So for a target mapping the end of a
280 * zoned block device with such a zone, len would not be zone
281 * aligned. We do not allow such last smaller zone to be part
282 * of the mapping here to ensure that mappings with multiple
283 * devices do not end up with a smaller zone in the middle of
286 if (len & (zone_sectors - 1)) {
287 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
288 dm_device_name(ti->table->md),
289 (unsigned long long)len,
290 zone_sectors, bdevname(bdev, b));
295 if (logical_block_size_sectors <= 1)
298 if (start & (logical_block_size_sectors - 1)) {
299 DMWARN("%s: start=%llu not aligned to h/w "
300 "logical block size %u of %s",
301 dm_device_name(ti->table->md),
302 (unsigned long long)start,
303 limits->logical_block_size, bdevname(bdev, b));
307 if (len & (logical_block_size_sectors - 1)) {
308 DMWARN("%s: len=%llu not aligned to h/w "
309 "logical block size %u of %s",
310 dm_device_name(ti->table->md),
311 (unsigned long long)len,
312 limits->logical_block_size, bdevname(bdev, b));
320 * This upgrades the mode on an already open dm_dev, being
321 * careful to leave things as they were if we fail to reopen the
322 * device and not to touch the existing bdev field in case
323 * it is accessed concurrently.
325 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
326 struct mapped_device *md)
329 struct dm_dev *old_dev, *new_dev;
331 old_dev = dd->dm_dev;
333 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
334 dd->dm_dev->mode | new_mode, &new_dev);
338 dd->dm_dev = new_dev;
339 dm_put_table_device(md, old_dev);
345 * Convert the path to a device
347 dev_t dm_get_dev_t(const char *path)
350 struct block_device *bdev;
352 bdev = lookup_bdev(path);
354 dev = name_to_dev_t(path);
362 EXPORT_SYMBOL_GPL(dm_get_dev_t);
365 * Add a device to the list, or just increment the usage count if
366 * it's already present.
368 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
369 struct dm_dev **result)
373 struct dm_dev_internal *dd;
374 struct dm_table *t = ti->table;
378 dev = dm_get_dev_t(path);
382 dd = find_device(&t->devices, dev);
384 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
388 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
393 refcount_set(&dd->count, 1);
394 list_add(&dd->list, &t->devices);
397 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
398 r = upgrade_mode(dd, mode, t->md);
402 refcount_inc(&dd->count);
404 *result = dd->dm_dev;
407 EXPORT_SYMBOL(dm_get_device);
409 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
410 sector_t start, sector_t len, void *data)
412 struct queue_limits *limits = data;
413 struct block_device *bdev = dev->bdev;
414 struct request_queue *q = bdev_get_queue(bdev);
415 char b[BDEVNAME_SIZE];
418 DMWARN("%s: Cannot set limits for nonexistent device %s",
419 dm_device_name(ti->table->md), bdevname(bdev, b));
423 if (blk_stack_limits(limits, &q->limits,
424 get_start_sect(bdev) + start) < 0)
425 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
426 "physical_block_size=%u, logical_block_size=%u, "
427 "alignment_offset=%u, start=%llu",
428 dm_device_name(ti->table->md), bdevname(bdev, b),
429 q->limits.physical_block_size,
430 q->limits.logical_block_size,
431 q->limits.alignment_offset,
432 (unsigned long long) start << SECTOR_SHIFT);
437 * Decrement a device's use count and remove it if necessary.
439 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
442 struct list_head *devices = &ti->table->devices;
443 struct dm_dev_internal *dd;
445 list_for_each_entry(dd, devices, list) {
446 if (dd->dm_dev == d) {
452 DMWARN("%s: device %s not in table devices list",
453 dm_device_name(ti->table->md), d->name);
456 if (refcount_dec_and_test(&dd->count)) {
457 dm_put_table_device(ti->table->md, d);
462 EXPORT_SYMBOL(dm_put_device);
465 * Checks to see if the target joins onto the end of the table.
467 static int adjoin(struct dm_table *table, struct dm_target *ti)
469 struct dm_target *prev;
471 if (!table->num_targets)
474 prev = &table->targets[table->num_targets - 1];
475 return (ti->begin == (prev->begin + prev->len));
479 * Used to dynamically allocate the arg array.
481 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
482 * process messages even if some device is suspended. These messages have a
483 * small fixed number of arguments.
485 * On the other hand, dm-switch needs to process bulk data using messages and
486 * excessive use of GFP_NOIO could cause trouble.
488 static char **realloc_argv(unsigned *size, char **old_argv)
495 new_size = *size * 2;
501 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
502 if (argv && old_argv) {
503 memcpy(argv, old_argv, *size * sizeof(*argv));
512 * Destructively splits up the argument list to pass to ctr.
514 int dm_split_args(int *argc, char ***argvp, char *input)
516 char *start, *end = input, *out, **argv = NULL;
517 unsigned array_size = 0;
526 argv = realloc_argv(&array_size, argv);
531 /* Skip whitespace */
532 start = skip_spaces(end);
535 break; /* success, we hit the end */
537 /* 'out' is used to remove any back-quotes */
540 /* Everything apart from '\0' can be quoted */
541 if (*end == '\\' && *(end + 1)) {
548 break; /* end of token */
553 /* have we already filled the array ? */
554 if ((*argc + 1) > array_size) {
555 argv = realloc_argv(&array_size, argv);
560 /* we know this is whitespace */
564 /* terminate the string and put it in the array */
575 * Impose necessary and sufficient conditions on a devices's table such
576 * that any incoming bio which respects its logical_block_size can be
577 * processed successfully. If it falls across the boundary between
578 * two or more targets, the size of each piece it gets split into must
579 * be compatible with the logical_block_size of the target processing it.
581 static int validate_hardware_logical_block_alignment(struct dm_table *table,
582 struct queue_limits *limits)
585 * This function uses arithmetic modulo the logical_block_size
586 * (in units of 512-byte sectors).
588 unsigned short device_logical_block_size_sects =
589 limits->logical_block_size >> SECTOR_SHIFT;
592 * Offset of the start of the next table entry, mod logical_block_size.
594 unsigned short next_target_start = 0;
597 * Given an aligned bio that extends beyond the end of a
598 * target, how many sectors must the next target handle?
600 unsigned short remaining = 0;
602 struct dm_target *ti;
603 struct queue_limits ti_limits;
607 * Check each entry in the table in turn.
609 for (i = 0; i < dm_table_get_num_targets(table); i++) {
610 ti = dm_table_get_target(table, i);
612 blk_set_stacking_limits(&ti_limits);
614 /* combine all target devices' limits */
615 if (ti->type->iterate_devices)
616 ti->type->iterate_devices(ti, dm_set_device_limits,
620 * If the remaining sectors fall entirely within this
621 * table entry are they compatible with its logical_block_size?
623 if (remaining < ti->len &&
624 remaining & ((ti_limits.logical_block_size >>
629 (unsigned short) ((next_target_start + ti->len) &
630 (device_logical_block_size_sects - 1));
631 remaining = next_target_start ?
632 device_logical_block_size_sects - next_target_start : 0;
636 DMWARN("%s: table line %u (start sect %llu len %llu) "
637 "not aligned to h/w logical block size %u",
638 dm_device_name(table->md), i,
639 (unsigned long long) ti->begin,
640 (unsigned long long) ti->len,
641 limits->logical_block_size);
648 int dm_table_add_target(struct dm_table *t, const char *type,
649 sector_t start, sector_t len, char *params)
651 int r = -EINVAL, argc;
653 struct dm_target *tgt;
656 DMERR("%s: target type %s must appear alone in table",
657 dm_device_name(t->md), t->targets->type->name);
661 BUG_ON(t->num_targets >= t->num_allocated);
663 tgt = t->targets + t->num_targets;
664 memset(tgt, 0, sizeof(*tgt));
667 DMERR("%s: zero-length target", dm_device_name(t->md));
671 tgt->type = dm_get_target_type(type);
673 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
677 if (dm_target_needs_singleton(tgt->type)) {
678 if (t->num_targets) {
679 tgt->error = "singleton target type must appear alone in table";
685 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
686 tgt->error = "target type may not be included in a read-only table";
690 if (t->immutable_target_type) {
691 if (t->immutable_target_type != tgt->type) {
692 tgt->error = "immutable target type cannot be mixed with other target types";
695 } else if (dm_target_is_immutable(tgt->type)) {
696 if (t->num_targets) {
697 tgt->error = "immutable target type cannot be mixed with other target types";
700 t->immutable_target_type = tgt->type;
703 if (dm_target_has_integrity(tgt->type))
704 t->integrity_added = 1;
709 tgt->error = "Unknown error";
712 * Does this target adjoin the previous one ?
714 if (!adjoin(t, tgt)) {
715 tgt->error = "Gap in table";
719 r = dm_split_args(&argc, &argv, params);
721 tgt->error = "couldn't split parameters (insufficient memory)";
725 r = tgt->type->ctr(tgt, argc, argv);
730 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
732 if (!tgt->num_discard_bios && tgt->discards_supported)
733 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
734 dm_device_name(t->md), type);
739 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
740 dm_put_target_type(tgt->type);
745 * Target argument parsing helpers.
747 static int validate_next_arg(const struct dm_arg *arg,
748 struct dm_arg_set *arg_set,
749 unsigned *value, char **error, unsigned grouped)
751 const char *arg_str = dm_shift_arg(arg_set);
755 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
756 (*value < arg->min) ||
757 (*value > arg->max) ||
758 (grouped && arg_set->argc < *value)) {
766 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
767 unsigned *value, char **error)
769 return validate_next_arg(arg, arg_set, value, error, 0);
771 EXPORT_SYMBOL(dm_read_arg);
773 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
774 unsigned *value, char **error)
776 return validate_next_arg(arg, arg_set, value, error, 1);
778 EXPORT_SYMBOL(dm_read_arg_group);
780 const char *dm_shift_arg(struct dm_arg_set *as)
793 EXPORT_SYMBOL(dm_shift_arg);
795 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
797 BUG_ON(as->argc < num_args);
798 as->argc -= num_args;
799 as->argv += num_args;
801 EXPORT_SYMBOL(dm_consume_args);
803 static bool __table_type_bio_based(enum dm_queue_mode table_type)
805 return (table_type == DM_TYPE_BIO_BASED ||
806 table_type == DM_TYPE_DAX_BIO_BASED);
809 static bool __table_type_request_based(enum dm_queue_mode table_type)
811 return table_type == DM_TYPE_REQUEST_BASED;
814 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
818 EXPORT_SYMBOL_GPL(dm_table_set_type);
820 /* validate the dax capability of the target device span */
821 int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
822 sector_t start, sector_t len, void *data)
824 int blocksize = *(int *) data, id;
827 id = dax_read_lock();
828 rc = dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
834 /* Check devices support synchronous DAX */
835 static int device_dax_synchronous(struct dm_target *ti, struct dm_dev *dev,
836 sector_t start, sector_t len, void *data)
838 return dev->dax_dev && dax_synchronous(dev->dax_dev);
841 bool dm_table_supports_dax(struct dm_table *t,
842 iterate_devices_callout_fn iterate_fn, int *blocksize)
844 struct dm_target *ti;
847 /* Ensure that all targets support DAX. */
848 for (i = 0; i < dm_table_get_num_targets(t); i++) {
849 ti = dm_table_get_target(t, i);
851 if (!ti->type->direct_access)
854 if (!ti->type->iterate_devices ||
855 !ti->type->iterate_devices(ti, iterate_fn, blocksize))
862 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
863 sector_t start, sector_t len, void *data)
865 struct block_device *bdev = dev->bdev;
866 struct request_queue *q = bdev_get_queue(bdev);
868 /* request-based cannot stack on partitions! */
869 if (bdev_is_partition(bdev))
872 return queue_is_mq(q);
875 static int dm_table_determine_type(struct dm_table *t)
878 unsigned bio_based = 0, request_based = 0, hybrid = 0;
879 struct dm_target *tgt;
880 struct list_head *devices = dm_table_get_devices(t);
881 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
882 int page_size = PAGE_SIZE;
884 if (t->type != DM_TYPE_NONE) {
885 /* target already set the table's type */
886 if (t->type == DM_TYPE_BIO_BASED) {
887 /* possibly upgrade to a variant of bio-based */
888 goto verify_bio_based;
890 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
891 goto verify_rq_based;
894 for (i = 0; i < t->num_targets; i++) {
895 tgt = t->targets + i;
896 if (dm_target_hybrid(tgt))
898 else if (dm_target_request_based(tgt))
903 if (bio_based && request_based) {
904 DMERR("Inconsistent table: different target types"
905 " can't be mixed up");
910 if (hybrid && !bio_based && !request_based) {
912 * The targets can work either way.
913 * Determine the type from the live device.
914 * Default to bio-based if device is new.
916 if (__table_type_request_based(live_md_type))
924 /* We must use this table as bio-based */
925 t->type = DM_TYPE_BIO_BASED;
926 if (dm_table_supports_dax(t, device_supports_dax, &page_size) ||
927 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
928 t->type = DM_TYPE_DAX_BIO_BASED;
933 BUG_ON(!request_based); /* No targets in this table */
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 DMERR("request-based DM doesn't support multiple targets");
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 */
955 t->type = live_table->type;
956 dm_put_live_table(t->md, srcu_idx);
960 tgt = dm_table_get_immutable_target(t);
962 DMERR("table load rejected: immutable target is required");
964 } else if (tgt->max_io_len) {
965 DMERR("table load rejected: immutable target that splits IO is not supported");
969 /* Non-request-stackable devices can't be used for request-based dm */
970 if (!tgt->type->iterate_devices ||
971 !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
972 DMERR("table load rejected: including non-request-stackable devices");
979 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
984 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
986 return t->immutable_target_type;
989 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
991 /* Immutable target is implicitly a singleton */
992 if (t->num_targets > 1 ||
993 !dm_target_is_immutable(t->targets[0].type))
999 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1001 struct dm_target *ti;
1004 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1005 ti = dm_table_get_target(t, i);
1006 if (dm_target_is_wildcard(ti->type))
1013 bool dm_table_bio_based(struct dm_table *t)
1015 return __table_type_bio_based(dm_table_get_type(t));
1018 bool dm_table_request_based(struct dm_table *t)
1020 return __table_type_request_based(dm_table_get_type(t));
1023 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1025 enum dm_queue_mode type = dm_table_get_type(t);
1026 unsigned per_io_data_size = 0;
1027 unsigned min_pool_size = 0;
1028 struct dm_target *ti;
1031 if (unlikely(type == DM_TYPE_NONE)) {
1032 DMWARN("no table type is set, can't allocate mempools");
1036 if (__table_type_bio_based(type))
1037 for (i = 0; i < t->num_targets; i++) {
1038 ti = t->targets + i;
1039 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1040 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1043 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1044 per_io_data_size, min_pool_size);
1051 void dm_table_free_md_mempools(struct dm_table *t)
1053 dm_free_md_mempools(t->mempools);
1057 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1062 static int setup_indexes(struct dm_table *t)
1065 unsigned int total = 0;
1068 /* allocate the space for *all* the indexes */
1069 for (i = t->depth - 2; i >= 0; i--) {
1070 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1071 total += t->counts[i];
1074 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1078 /* set up internal nodes, bottom-up */
1079 for (i = t->depth - 2; i >= 0; i--) {
1080 t->index[i] = indexes;
1081 indexes += (KEYS_PER_NODE * t->counts[i]);
1082 setup_btree_index(i, t);
1089 * Builds the btree to index the map.
1091 static int dm_table_build_index(struct dm_table *t)
1094 unsigned int leaf_nodes;
1096 /* how many indexes will the btree have ? */
1097 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1098 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1100 /* leaf layer has already been set up */
1101 t->counts[t->depth - 1] = leaf_nodes;
1102 t->index[t->depth - 1] = t->highs;
1105 r = setup_indexes(t);
1110 static bool integrity_profile_exists(struct gendisk *disk)
1112 return !!blk_get_integrity(disk);
1116 * Get a disk whose integrity profile reflects the table's profile.
1117 * Returns NULL if integrity support was inconsistent or unavailable.
1119 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1121 struct list_head *devices = dm_table_get_devices(t);
1122 struct dm_dev_internal *dd = NULL;
1123 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1126 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1127 struct dm_target *ti = dm_table_get_target(t, i);
1128 if (!dm_target_passes_integrity(ti->type))
1132 list_for_each_entry(dd, devices, list) {
1133 template_disk = dd->dm_dev->bdev->bd_disk;
1134 if (!integrity_profile_exists(template_disk))
1136 else if (prev_disk &&
1137 blk_integrity_compare(prev_disk, template_disk) < 0)
1139 prev_disk = template_disk;
1142 return template_disk;
1146 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1147 dm_device_name(t->md),
1148 prev_disk->disk_name,
1149 template_disk->disk_name);
1154 * Register the mapped device for blk_integrity support if the
1155 * underlying devices have an integrity profile. But all devices may
1156 * not have matching profiles (checking all devices isn't reliable
1157 * during table load because this table may use other DM device(s) which
1158 * must be resumed before they will have an initialized integity
1159 * profile). Consequently, stacked DM devices force a 2 stage integrity
1160 * profile validation: First pass during table load, final pass during
1163 static int dm_table_register_integrity(struct dm_table *t)
1165 struct mapped_device *md = t->md;
1166 struct gendisk *template_disk = NULL;
1168 /* If target handles integrity itself do not register it here. */
1169 if (t->integrity_added)
1172 template_disk = dm_table_get_integrity_disk(t);
1176 if (!integrity_profile_exists(dm_disk(md))) {
1177 t->integrity_supported = true;
1179 * Register integrity profile during table load; we can do
1180 * this because the final profile must match during resume.
1182 blk_integrity_register(dm_disk(md),
1183 blk_get_integrity(template_disk));
1188 * If DM device already has an initialized integrity
1189 * profile the new profile should not conflict.
1191 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1192 DMWARN("%s: conflict with existing integrity profile: "
1193 "%s profile mismatch",
1194 dm_device_name(t->md),
1195 template_disk->disk_name);
1199 /* Preserve existing integrity profile */
1200 t->integrity_supported = true;
1205 * Prepares the table for use by building the indices,
1206 * setting the type, and allocating mempools.
1208 int dm_table_complete(struct dm_table *t)
1212 r = dm_table_determine_type(t);
1214 DMERR("unable to determine table type");
1218 r = dm_table_build_index(t);
1220 DMERR("unable to build btrees");
1224 r = dm_table_register_integrity(t);
1226 DMERR("could not register integrity profile.");
1230 r = dm_table_alloc_md_mempools(t, t->md);
1232 DMERR("unable to allocate mempools");
1237 static DEFINE_MUTEX(_event_lock);
1238 void dm_table_event_callback(struct dm_table *t,
1239 void (*fn)(void *), void *context)
1241 mutex_lock(&_event_lock);
1243 t->event_context = context;
1244 mutex_unlock(&_event_lock);
1247 void dm_table_event(struct dm_table *t)
1249 mutex_lock(&_event_lock);
1251 t->event_fn(t->event_context);
1252 mutex_unlock(&_event_lock);
1254 EXPORT_SYMBOL(dm_table_event);
1256 inline sector_t dm_table_get_size(struct dm_table *t)
1258 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1260 EXPORT_SYMBOL(dm_table_get_size);
1262 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1264 if (index >= t->num_targets)
1267 return t->targets + index;
1271 * Search the btree for the correct target.
1273 * Caller should check returned pointer for NULL
1274 * to trap I/O beyond end of device.
1276 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1278 unsigned int l, n = 0, k = 0;
1281 if (unlikely(sector >= dm_table_get_size(t)))
1284 for (l = 0; l < t->depth; l++) {
1285 n = get_child(n, k);
1286 node = get_node(t, l, n);
1288 for (k = 0; k < KEYS_PER_NODE; k++)
1289 if (node[k] >= sector)
1293 return &t->targets[(KEYS_PER_NODE * n) + k];
1296 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1297 sector_t start, sector_t len, void *data)
1299 unsigned *num_devices = data;
1307 * Check whether a table has no data devices attached using each
1308 * target's iterate_devices method.
1309 * Returns false if the result is unknown because a target doesn't
1310 * support iterate_devices.
1312 bool dm_table_has_no_data_devices(struct dm_table *table)
1314 struct dm_target *ti;
1315 unsigned i, num_devices;
1317 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1318 ti = dm_table_get_target(table, i);
1320 if (!ti->type->iterate_devices)
1324 ti->type->iterate_devices(ti, count_device, &num_devices);
1332 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1333 sector_t start, sector_t len, void *data)
1335 struct request_queue *q = bdev_get_queue(dev->bdev);
1336 enum blk_zoned_model *zoned_model = data;
1338 return q && blk_queue_zoned_model(q) == *zoned_model;
1341 static bool dm_table_supports_zoned_model(struct dm_table *t,
1342 enum blk_zoned_model zoned_model)
1344 struct dm_target *ti;
1347 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1348 ti = dm_table_get_target(t, i);
1350 if (zoned_model == BLK_ZONED_HM &&
1351 !dm_target_supports_zoned_hm(ti->type))
1354 if (!ti->type->iterate_devices ||
1355 !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1362 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1363 sector_t start, sector_t len, void *data)
1365 struct request_queue *q = bdev_get_queue(dev->bdev);
1366 unsigned int *zone_sectors = data;
1368 return q && blk_queue_zone_sectors(q) == *zone_sectors;
1371 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1372 unsigned int zone_sectors)
1374 struct dm_target *ti;
1377 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1378 ti = dm_table_get_target(t, i);
1380 if (!ti->type->iterate_devices ||
1381 !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1388 static int validate_hardware_zoned_model(struct dm_table *table,
1389 enum blk_zoned_model zoned_model,
1390 unsigned int zone_sectors)
1392 if (zoned_model == BLK_ZONED_NONE)
1395 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1396 DMERR("%s: zoned model is not consistent across all devices",
1397 dm_device_name(table->md));
1401 /* Check zone size validity and compatibility */
1402 if (!zone_sectors || !is_power_of_2(zone_sectors))
1405 if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1406 DMERR("%s: zone sectors is not consistent across all devices",
1407 dm_device_name(table->md));
1415 * Establish the new table's queue_limits and validate them.
1417 int dm_calculate_queue_limits(struct dm_table *table,
1418 struct queue_limits *limits)
1420 struct dm_target *ti;
1421 struct queue_limits ti_limits;
1423 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1424 unsigned int zone_sectors = 0;
1426 blk_set_stacking_limits(limits);
1428 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1429 blk_set_stacking_limits(&ti_limits);
1431 ti = dm_table_get_target(table, i);
1433 if (!ti->type->iterate_devices)
1434 goto combine_limits;
1437 * Combine queue limits of all the devices this target uses.
1439 ti->type->iterate_devices(ti, dm_set_device_limits,
1442 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1444 * After stacking all limits, validate all devices
1445 * in table support this zoned model and zone sectors.
1447 zoned_model = ti_limits.zoned;
1448 zone_sectors = ti_limits.chunk_sectors;
1451 /* Set I/O hints portion of queue limits */
1452 if (ti->type->io_hints)
1453 ti->type->io_hints(ti, &ti_limits);
1456 * Check each device area is consistent with the target's
1457 * overall queue limits.
1459 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1465 * Merge this target's queue limits into the overall limits
1468 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1469 DMWARN("%s: adding target device "
1470 "(start sect %llu len %llu) "
1471 "caused an alignment inconsistency",
1472 dm_device_name(table->md),
1473 (unsigned long long) ti->begin,
1474 (unsigned long long) ti->len);
1478 * Verify that the zoned model and zone sectors, as determined before
1479 * any .io_hints override, are the same across all devices in the table.
1480 * - this is especially relevant if .io_hints is emulating a disk-managed
1481 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1484 if (limits->zoned != BLK_ZONED_NONE) {
1486 * ...IF the above limits stacking determined a zoned model
1487 * validate that all of the table's devices conform to it.
1489 zoned_model = limits->zoned;
1490 zone_sectors = limits->chunk_sectors;
1492 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1495 return validate_hardware_logical_block_alignment(table, limits);
1499 * Verify that all devices have an integrity profile that matches the
1500 * DM device's registered integrity profile. If the profiles don't
1501 * match then unregister the DM device's integrity profile.
1503 static void dm_table_verify_integrity(struct dm_table *t)
1505 struct gendisk *template_disk = NULL;
1507 if (t->integrity_added)
1510 if (t->integrity_supported) {
1512 * Verify that the original integrity profile
1513 * matches all the devices in this table.
1515 template_disk = dm_table_get_integrity_disk(t);
1516 if (template_disk &&
1517 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1521 if (integrity_profile_exists(dm_disk(t->md))) {
1522 DMWARN("%s: unable to establish an integrity profile",
1523 dm_device_name(t->md));
1524 blk_integrity_unregister(dm_disk(t->md));
1528 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1529 sector_t start, sector_t len, void *data)
1531 unsigned long flush = (unsigned long) data;
1532 struct request_queue *q = bdev_get_queue(dev->bdev);
1534 return q && (q->queue_flags & flush);
1537 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1539 struct dm_target *ti;
1543 * Require at least one underlying device to support flushes.
1544 * t->devices includes internal dm devices such as mirror logs
1545 * so we need to use iterate_devices here, which targets
1546 * supporting flushes must provide.
1548 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1549 ti = dm_table_get_target(t, i);
1551 if (!ti->num_flush_bios)
1554 if (ti->flush_supported)
1557 if (ti->type->iterate_devices &&
1558 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1565 static int device_dax_write_cache_enabled(struct dm_target *ti,
1566 struct dm_dev *dev, sector_t start,
1567 sector_t len, void *data)
1569 struct dax_device *dax_dev = dev->dax_dev;
1574 if (dax_write_cache_enabled(dax_dev))
1579 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1581 struct dm_target *ti;
1584 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1585 ti = dm_table_get_target(t, i);
1587 if (ti->type->iterate_devices &&
1588 ti->type->iterate_devices(ti,
1589 device_dax_write_cache_enabled, NULL))
1596 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1597 sector_t start, sector_t len, void *data)
1599 struct request_queue *q = bdev_get_queue(dev->bdev);
1601 return q && blk_queue_nonrot(q);
1604 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1605 sector_t start, sector_t len, void *data)
1607 struct request_queue *q = bdev_get_queue(dev->bdev);
1609 return q && !blk_queue_add_random(q);
1612 static bool dm_table_all_devices_attribute(struct dm_table *t,
1613 iterate_devices_callout_fn func)
1615 struct dm_target *ti;
1618 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1619 ti = dm_table_get_target(t, i);
1621 if (!ti->type->iterate_devices ||
1622 !ti->type->iterate_devices(ti, func, NULL))
1629 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1630 sector_t start, sector_t len, void *data)
1632 struct request_queue *q = bdev_get_queue(dev->bdev);
1634 return q && !q->limits.max_write_same_sectors;
1637 static bool dm_table_supports_write_same(struct dm_table *t)
1639 struct dm_target *ti;
1642 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1643 ti = dm_table_get_target(t, i);
1645 if (!ti->num_write_same_bios)
1648 if (!ti->type->iterate_devices ||
1649 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1656 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1657 sector_t start, sector_t len, void *data)
1659 struct request_queue *q = bdev_get_queue(dev->bdev);
1661 return q && !q->limits.max_write_zeroes_sectors;
1664 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1666 struct dm_target *ti;
1669 while (i < dm_table_get_num_targets(t)) {
1670 ti = dm_table_get_target(t, i++);
1672 if (!ti->num_write_zeroes_bios)
1675 if (!ti->type->iterate_devices ||
1676 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1683 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1684 sector_t start, sector_t len, void *data)
1686 struct request_queue *q = bdev_get_queue(dev->bdev);
1688 return q && !blk_queue_nowait(q);
1691 static bool dm_table_supports_nowait(struct dm_table *t)
1693 struct dm_target *ti;
1696 while (i < dm_table_get_num_targets(t)) {
1697 ti = dm_table_get_target(t, i++);
1699 if (!dm_target_supports_nowait(ti->type))
1702 if (!ti->type->iterate_devices ||
1703 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1710 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1711 sector_t start, sector_t len, void *data)
1713 struct request_queue *q = bdev_get_queue(dev->bdev);
1715 return q && !blk_queue_discard(q);
1718 static bool dm_table_supports_discards(struct dm_table *t)
1720 struct dm_target *ti;
1723 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1724 ti = dm_table_get_target(t, i);
1726 if (!ti->num_discard_bios)
1730 * Either the target provides discard support (as implied by setting
1731 * 'discards_supported') or it relies on _all_ data devices having
1734 if (!ti->discards_supported &&
1735 (!ti->type->iterate_devices ||
1736 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1743 static int device_not_secure_erase_capable(struct dm_target *ti,
1744 struct dm_dev *dev, sector_t start,
1745 sector_t len, void *data)
1747 struct request_queue *q = bdev_get_queue(dev->bdev);
1749 return q && !blk_queue_secure_erase(q);
1752 static bool dm_table_supports_secure_erase(struct dm_table *t)
1754 struct dm_target *ti;
1757 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1758 ti = dm_table_get_target(t, i);
1760 if (!ti->num_secure_erase_bios)
1763 if (!ti->type->iterate_devices ||
1764 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1771 static int device_requires_stable_pages(struct dm_target *ti,
1772 struct dm_dev *dev, sector_t start,
1773 sector_t len, void *data)
1775 struct request_queue *q = bdev_get_queue(dev->bdev);
1777 return q && blk_queue_stable_writes(q);
1781 * If any underlying device requires stable pages, a table must require
1782 * them as well. Only targets that support iterate_devices are considered:
1783 * don't want error, zero, etc to require stable pages.
1785 static bool dm_table_requires_stable_pages(struct dm_table *t)
1787 struct dm_target *ti;
1790 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1791 ti = dm_table_get_target(t, i);
1793 if (ti->type->iterate_devices &&
1794 ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1801 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1802 struct queue_limits *limits)
1804 bool wc = false, fua = false;
1805 int page_size = PAGE_SIZE;
1808 * Copy table's limits to the DM device's request_queue
1810 q->limits = *limits;
1812 if (dm_table_supports_nowait(t))
1813 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1815 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1817 if (!dm_table_supports_discards(t)) {
1818 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1819 /* Must also clear discard limits... */
1820 q->limits.max_discard_sectors = 0;
1821 q->limits.max_hw_discard_sectors = 0;
1822 q->limits.discard_granularity = 0;
1823 q->limits.discard_alignment = 0;
1824 q->limits.discard_misaligned = 0;
1826 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1828 if (dm_table_supports_secure_erase(t))
1829 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1831 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1833 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1836 blk_queue_write_cache(q, wc, fua);
1838 if (dm_table_supports_dax(t, device_supports_dax, &page_size)) {
1839 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1840 if (dm_table_supports_dax(t, device_dax_synchronous, NULL))
1841 set_dax_synchronous(t->md->dax_dev);
1844 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1846 if (dm_table_supports_dax_write_cache(t))
1847 dax_write_cache(t->md->dax_dev, true);
1849 /* Ensure that all underlying devices are non-rotational. */
1850 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1851 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1853 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1855 if (!dm_table_supports_write_same(t))
1856 q->limits.max_write_same_sectors = 0;
1857 if (!dm_table_supports_write_zeroes(t))
1858 q->limits.max_write_zeroes_sectors = 0;
1860 dm_table_verify_integrity(t);
1863 * Some devices don't use blk_integrity but still want stable pages
1864 * because they do their own checksumming.
1866 if (dm_table_requires_stable_pages(t))
1867 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
1869 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
1872 * Determine whether or not this queue's I/O timings contribute
1873 * to the entropy pool, Only request-based targets use this.
1874 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1877 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1878 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1881 * For a zoned target, the number of zones should be updated for the
1882 * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1883 * target, this is all that is needed.
1885 #ifdef CONFIG_BLK_DEV_ZONED
1886 if (blk_queue_is_zoned(q)) {
1887 WARN_ON_ONCE(queue_is_mq(q));
1888 q->nr_zones = blkdev_nr_zones(t->md->disk);
1892 blk_queue_update_readahead(q);
1895 unsigned int dm_table_get_num_targets(struct dm_table *t)
1897 return t->num_targets;
1900 struct list_head *dm_table_get_devices(struct dm_table *t)
1905 fmode_t dm_table_get_mode(struct dm_table *t)
1909 EXPORT_SYMBOL(dm_table_get_mode);
1917 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1919 int i = t->num_targets;
1920 struct dm_target *ti = t->targets;
1922 lockdep_assert_held(&t->md->suspend_lock);
1927 if (ti->type->presuspend)
1928 ti->type->presuspend(ti);
1930 case PRESUSPEND_UNDO:
1931 if (ti->type->presuspend_undo)
1932 ti->type->presuspend_undo(ti);
1935 if (ti->type->postsuspend)
1936 ti->type->postsuspend(ti);
1943 void dm_table_presuspend_targets(struct dm_table *t)
1948 suspend_targets(t, PRESUSPEND);
1951 void dm_table_presuspend_undo_targets(struct dm_table *t)
1956 suspend_targets(t, PRESUSPEND_UNDO);
1959 void dm_table_postsuspend_targets(struct dm_table *t)
1964 suspend_targets(t, POSTSUSPEND);
1967 int dm_table_resume_targets(struct dm_table *t)
1971 lockdep_assert_held(&t->md->suspend_lock);
1973 for (i = 0; i < t->num_targets; i++) {
1974 struct dm_target *ti = t->targets + i;
1976 if (!ti->type->preresume)
1979 r = ti->type->preresume(ti);
1981 DMERR("%s: %s: preresume failed, error = %d",
1982 dm_device_name(t->md), ti->type->name, r);
1987 for (i = 0; i < t->num_targets; i++) {
1988 struct dm_target *ti = t->targets + i;
1990 if (ti->type->resume)
1991 ti->type->resume(ti);
1997 struct mapped_device *dm_table_get_md(struct dm_table *t)
2001 EXPORT_SYMBOL(dm_table_get_md);
2003 const char *dm_table_device_name(struct dm_table *t)
2005 return dm_device_name(t->md);
2007 EXPORT_SYMBOL_GPL(dm_table_device_name);
2009 void dm_table_run_md_queue_async(struct dm_table *t)
2011 if (!dm_table_request_based(t))
2015 blk_mq_run_hw_queues(t->md->queue, true);
2017 EXPORT_SYMBOL(dm_table_run_md_queue_async);