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/lcm.h>
22 #include <linux/blk-mq.h>
23 #include <linux/mount.h>
24 #include <linux/dax.h>
26 #define DM_MSG_PREFIX "table"
28 #define NODE_SIZE L1_CACHE_BYTES
29 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
30 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
33 * Similar to ceiling(log_size(n))
35 static unsigned int int_log(unsigned int n, unsigned int base)
40 n = dm_div_up(n, base);
48 * Calculate the index of the child node of the n'th node k'th key.
50 static inline unsigned int get_child(unsigned int n, unsigned int k)
52 return (n * CHILDREN_PER_NODE) + k;
56 * Return the n'th node of level l from table t.
58 static inline sector_t *get_node(struct dm_table *t,
59 unsigned int l, unsigned int n)
61 return t->index[l] + (n * KEYS_PER_NODE);
65 * Return the highest key that you could lookup from the n'th
66 * node on level l of the btree.
68 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
70 for (; l < t->depth - 1; l++)
71 n = get_child(n, CHILDREN_PER_NODE - 1);
73 if (n >= t->counts[l])
74 return (sector_t) - 1;
76 return get_node(t, l, n)[KEYS_PER_NODE - 1];
80 * Fills in a level of the btree based on the highs of the level
83 static int setup_btree_index(unsigned int l, struct dm_table *t)
88 for (n = 0U; n < t->counts[l]; n++) {
89 node = get_node(t, l, n);
91 for (k = 0U; k < KEYS_PER_NODE; k++)
92 node[k] = high(t, l + 1, get_child(n, k));
98 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
104 * Check that we're not going to overflow.
106 if (nmemb > (ULONG_MAX / elem_size))
109 size = nmemb * elem_size;
110 addr = vzalloc(size);
114 EXPORT_SYMBOL(dm_vcalloc);
117 * highs, and targets are managed as dynamic arrays during a
120 static int alloc_targets(struct dm_table *t, unsigned int num)
123 struct dm_target *n_targets;
126 * Allocate both the target array and offset array at once.
128 n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
133 n_targets = (struct dm_target *) (n_highs + num);
135 memset(n_highs, -1, sizeof(*n_highs) * num);
138 t->num_allocated = num;
140 t->targets = n_targets;
145 int dm_table_create(struct dm_table **result, fmode_t mode,
146 unsigned num_targets, struct mapped_device *md)
148 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
153 INIT_LIST_HEAD(&t->devices);
156 num_targets = KEYS_PER_NODE;
158 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
165 if (alloc_targets(t, num_targets)) {
170 t->type = DM_TYPE_NONE;
177 static void free_devices(struct list_head *devices, struct mapped_device *md)
179 struct list_head *tmp, *next;
181 list_for_each_safe(tmp, next, devices) {
182 struct dm_dev_internal *dd =
183 list_entry(tmp, struct dm_dev_internal, list);
184 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
185 dm_device_name(md), dd->dm_dev->name);
186 dm_put_table_device(md, dd->dm_dev);
191 void dm_table_destroy(struct dm_table *t)
198 /* free the indexes */
200 vfree(t->index[t->depth - 2]);
202 /* free the targets */
203 for (i = 0; i < t->num_targets; i++) {
204 struct dm_target *tgt = t->targets + i;
209 dm_put_target_type(tgt->type);
214 /* free the device list */
215 free_devices(&t->devices, t->md);
217 dm_free_md_mempools(t->mempools);
223 * See if we've already got a device in the list.
225 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
227 struct dm_dev_internal *dd;
229 list_for_each_entry (dd, l, list)
230 if (dd->dm_dev->bdev->bd_dev == dev)
237 * If possible, this checks an area of a destination device is invalid.
239 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
240 sector_t start, sector_t len, void *data)
242 struct queue_limits *limits = data;
243 struct block_device *bdev = dev->bdev;
245 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
246 unsigned short logical_block_size_sectors =
247 limits->logical_block_size >> SECTOR_SHIFT;
248 char b[BDEVNAME_SIZE];
253 if ((start >= dev_size) || (start + len > dev_size)) {
254 DMWARN("%s: %s too small for target: "
255 "start=%llu, len=%llu, dev_size=%llu",
256 dm_device_name(ti->table->md), bdevname(bdev, b),
257 (unsigned long long)start,
258 (unsigned long long)len,
259 (unsigned long long)dev_size);
264 * If the target is mapped to zoned block device(s), check
265 * that the zones are not partially mapped.
267 if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
268 unsigned int zone_sectors = bdev_zone_sectors(bdev);
270 if (start & (zone_sectors - 1)) {
271 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
272 dm_device_name(ti->table->md),
273 (unsigned long long)start,
274 zone_sectors, bdevname(bdev, b));
279 * Note: The last zone of a zoned block device may be smaller
280 * than other zones. So for a target mapping the end of a
281 * zoned block device with such a zone, len would not be zone
282 * aligned. We do not allow such last smaller zone to be part
283 * of the mapping here to ensure that mappings with multiple
284 * devices do not end up with a smaller zone in the middle of
287 if (len & (zone_sectors - 1)) {
288 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
289 dm_device_name(ti->table->md),
290 (unsigned long long)len,
291 zone_sectors, bdevname(bdev, b));
296 if (logical_block_size_sectors <= 1)
299 if (start & (logical_block_size_sectors - 1)) {
300 DMWARN("%s: start=%llu not aligned to h/w "
301 "logical block size %u of %s",
302 dm_device_name(ti->table->md),
303 (unsigned long long)start,
304 limits->logical_block_size, bdevname(bdev, b));
308 if (len & (logical_block_size_sectors - 1)) {
309 DMWARN("%s: len=%llu not aligned to h/w "
310 "logical block size %u of %s",
311 dm_device_name(ti->table->md),
312 (unsigned long long)len,
313 limits->logical_block_size, bdevname(bdev, b));
321 * This upgrades the mode on an already open dm_dev, being
322 * careful to leave things as they were if we fail to reopen the
323 * device and not to touch the existing bdev field in case
324 * it is accessed concurrently.
326 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
327 struct mapped_device *md)
330 struct dm_dev *old_dev, *new_dev;
332 old_dev = dd->dm_dev;
334 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
335 dd->dm_dev->mode | new_mode, &new_dev);
339 dd->dm_dev = new_dev;
340 dm_put_table_device(md, old_dev);
346 * Convert the path to a device
348 dev_t dm_get_dev_t(const char *path)
351 struct block_device *bdev;
353 bdev = lookup_bdev(path);
355 dev = name_to_dev_t(path);
363 EXPORT_SYMBOL_GPL(dm_get_dev_t);
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)
374 struct dm_dev_internal *dd;
375 struct dm_table *t = ti->table;
379 dev = dm_get_dev_t(path);
383 dd = find_device(&t->devices, dev);
385 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
389 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
394 refcount_set(&dd->count, 1);
395 list_add(&dd->list, &t->devices);
398 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
399 r = upgrade_mode(dd, mode, t->md);
403 refcount_inc(&dd->count);
405 *result = dd->dm_dev;
408 EXPORT_SYMBOL(dm_get_device);
410 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
411 sector_t start, sector_t len, void *data)
413 struct queue_limits *limits = data;
414 struct block_device *bdev = dev->bdev;
415 struct request_queue *q = bdev_get_queue(bdev);
416 char b[BDEVNAME_SIZE];
419 DMWARN("%s: Cannot set limits for nonexistent device %s",
420 dm_device_name(ti->table->md), bdevname(bdev, b));
424 if (blk_stack_limits(limits, &q->limits,
425 get_start_sect(bdev) + start) < 0)
426 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
427 "physical_block_size=%u, logical_block_size=%u, "
428 "alignment_offset=%u, start=%llu",
429 dm_device_name(ti->table->md), bdevname(bdev, b),
430 q->limits.physical_block_size,
431 q->limits.logical_block_size,
432 q->limits.alignment_offset,
433 (unsigned long long) start << SECTOR_SHIFT);
438 * Decrement a device's use count and remove it if necessary.
440 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
443 struct list_head *devices = &ti->table->devices;
444 struct dm_dev_internal *dd;
446 list_for_each_entry(dd, devices, list) {
447 if (dd->dm_dev == d) {
453 DMWARN("%s: device %s not in table devices list",
454 dm_device_name(ti->table->md), d->name);
457 if (refcount_dec_and_test(&dd->count)) {
458 dm_put_table_device(ti->table->md, d);
463 EXPORT_SYMBOL(dm_put_device);
466 * Checks to see if the target joins onto the end of the table.
468 static int adjoin(struct dm_table *table, struct dm_target *ti)
470 struct dm_target *prev;
472 if (!table->num_targets)
475 prev = &table->targets[table->num_targets - 1];
476 return (ti->begin == (prev->begin + prev->len));
480 * Used to dynamically allocate the arg array.
482 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
483 * process messages even if some device is suspended. These messages have a
484 * small fixed number of arguments.
486 * On the other hand, dm-switch needs to process bulk data using messages and
487 * excessive use of GFP_NOIO could cause trouble.
489 static char **realloc_argv(unsigned *size, char **old_argv)
496 new_size = *size * 2;
502 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
503 if (argv && old_argv) {
504 memcpy(argv, old_argv, *size * sizeof(*argv));
513 * Destructively splits up the argument list to pass to ctr.
515 int dm_split_args(int *argc, char ***argvp, char *input)
517 char *start, *end = input, *out, **argv = NULL;
518 unsigned array_size = 0;
527 argv = realloc_argv(&array_size, argv);
532 /* Skip whitespace */
533 start = skip_spaces(end);
536 break; /* success, we hit the end */
538 /* 'out' is used to remove any back-quotes */
541 /* Everything apart from '\0' can be quoted */
542 if (*end == '\\' && *(end + 1)) {
549 break; /* end of token */
554 /* have we already filled the array ? */
555 if ((*argc + 1) > array_size) {
556 argv = realloc_argv(&array_size, argv);
561 /* we know this is whitespace */
565 /* terminate the string and put it in the array */
576 * Impose necessary and sufficient conditions on a devices's table such
577 * that any incoming bio which respects its logical_block_size can be
578 * processed successfully. If it falls across the boundary between
579 * two or more targets, the size of each piece it gets split into must
580 * be compatible with the logical_block_size of the target processing it.
582 static int validate_hardware_logical_block_alignment(struct dm_table *table,
583 struct queue_limits *limits)
586 * This function uses arithmetic modulo the logical_block_size
587 * (in units of 512-byte sectors).
589 unsigned short device_logical_block_size_sects =
590 limits->logical_block_size >> SECTOR_SHIFT;
593 * Offset of the start of the next table entry, mod logical_block_size.
595 unsigned short next_target_start = 0;
598 * Given an aligned bio that extends beyond the end of a
599 * target, how many sectors must the next target handle?
601 unsigned short remaining = 0;
603 struct dm_target *ti;
604 struct queue_limits ti_limits;
608 * Check each entry in the table in turn.
610 for (i = 0; i < dm_table_get_num_targets(table); i++) {
611 ti = dm_table_get_target(table, i);
613 blk_set_stacking_limits(&ti_limits);
615 /* combine all target devices' limits */
616 if (ti->type->iterate_devices)
617 ti->type->iterate_devices(ti, dm_set_device_limits,
621 * If the remaining sectors fall entirely within this
622 * table entry are they compatible with its logical_block_size?
624 if (remaining < ti->len &&
625 remaining & ((ti_limits.logical_block_size >>
630 (unsigned short) ((next_target_start + ti->len) &
631 (device_logical_block_size_sects - 1));
632 remaining = next_target_start ?
633 device_logical_block_size_sects - next_target_start : 0;
637 DMWARN("%s: table line %u (start sect %llu len %llu) "
638 "not aligned to h/w logical block size %u",
639 dm_device_name(table->md), i,
640 (unsigned long long) ti->begin,
641 (unsigned long long) ti->len,
642 limits->logical_block_size);
649 int dm_table_add_target(struct dm_table *t, const char *type,
650 sector_t start, sector_t len, char *params)
652 int r = -EINVAL, argc;
654 struct dm_target *tgt;
657 DMERR("%s: target type %s must appear alone in table",
658 dm_device_name(t->md), t->targets->type->name);
662 BUG_ON(t->num_targets >= t->num_allocated);
664 tgt = t->targets + t->num_targets;
665 memset(tgt, 0, sizeof(*tgt));
668 DMERR("%s: zero-length target", dm_device_name(t->md));
672 tgt->type = dm_get_target_type(type);
674 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
678 if (dm_target_needs_singleton(tgt->type)) {
679 if (t->num_targets) {
680 tgt->error = "singleton target type must appear alone in table";
686 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
687 tgt->error = "target type may not be included in a read-only table";
691 if (t->immutable_target_type) {
692 if (t->immutable_target_type != tgt->type) {
693 tgt->error = "immutable target type cannot be mixed with other target types";
696 } else if (dm_target_is_immutable(tgt->type)) {
697 if (t->num_targets) {
698 tgt->error = "immutable target type cannot be mixed with other target types";
701 t->immutable_target_type = tgt->type;
704 if (dm_target_has_integrity(tgt->type))
705 t->integrity_added = 1;
710 tgt->error = "Unknown error";
713 * Does this target adjoin the previous one ?
715 if (!adjoin(t, tgt)) {
716 tgt->error = "Gap in table";
720 r = dm_split_args(&argc, &argv, params);
722 tgt->error = "couldn't split parameters (insufficient memory)";
726 r = tgt->type->ctr(tgt, argc, argv);
731 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
733 if (!tgt->num_discard_bios && tgt->discards_supported)
734 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
735 dm_device_name(t->md), type);
740 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
741 dm_put_target_type(tgt->type);
746 * Target argument parsing helpers.
748 static int validate_next_arg(const struct dm_arg *arg,
749 struct dm_arg_set *arg_set,
750 unsigned *value, char **error, unsigned grouped)
752 const char *arg_str = dm_shift_arg(arg_set);
756 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
757 (*value < arg->min) ||
758 (*value > arg->max) ||
759 (grouped && arg_set->argc < *value)) {
767 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
768 unsigned *value, char **error)
770 return validate_next_arg(arg, arg_set, value, error, 0);
772 EXPORT_SYMBOL(dm_read_arg);
774 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
775 unsigned *value, char **error)
777 return validate_next_arg(arg, arg_set, value, error, 1);
779 EXPORT_SYMBOL(dm_read_arg_group);
781 const char *dm_shift_arg(struct dm_arg_set *as)
794 EXPORT_SYMBOL(dm_shift_arg);
796 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
798 BUG_ON(as->argc < num_args);
799 as->argc -= num_args;
800 as->argv += num_args;
802 EXPORT_SYMBOL(dm_consume_args);
804 static bool __table_type_bio_based(enum dm_queue_mode table_type)
806 return (table_type == DM_TYPE_BIO_BASED ||
807 table_type == DM_TYPE_DAX_BIO_BASED);
810 static bool __table_type_request_based(enum dm_queue_mode table_type)
812 return table_type == DM_TYPE_REQUEST_BASED;
815 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
819 EXPORT_SYMBOL_GPL(dm_table_set_type);
821 /* validate the dax capability of the target device span */
822 int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
823 sector_t start, sector_t len, void *data)
825 int blocksize = *(int *) data, id;
828 id = dax_read_lock();
829 rc = dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
835 /* Check devices support synchronous DAX */
836 static int device_dax_synchronous(struct dm_target *ti, struct dm_dev *dev,
837 sector_t start, sector_t len, void *data)
839 return dev->dax_dev && dax_synchronous(dev->dax_dev);
842 bool dm_table_supports_dax(struct dm_table *t,
843 iterate_devices_callout_fn iterate_fn, int *blocksize)
845 struct dm_target *ti;
848 /* Ensure that all targets support DAX. */
849 for (i = 0; i < dm_table_get_num_targets(t); i++) {
850 ti = dm_table_get_target(t, i);
852 if (!ti->type->direct_access)
855 if (!ti->type->iterate_devices ||
856 !ti->type->iterate_devices(ti, iterate_fn, blocksize))
863 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
864 sector_t start, sector_t len, void *data)
866 struct block_device *bdev = dev->bdev;
867 struct request_queue *q = bdev_get_queue(bdev);
869 /* request-based cannot stack on partitions! */
870 if (bdev_is_partition(bdev))
873 return queue_is_mq(q);
876 static int dm_table_determine_type(struct dm_table *t)
879 unsigned bio_based = 0, request_based = 0, hybrid = 0;
880 struct dm_target *tgt;
881 struct list_head *devices = dm_table_get_devices(t);
882 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
883 int page_size = PAGE_SIZE;
885 if (t->type != DM_TYPE_NONE) {
886 /* target already set the table's type */
887 if (t->type == DM_TYPE_BIO_BASED) {
888 /* possibly upgrade to a variant of bio-based */
889 goto verify_bio_based;
891 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
892 goto verify_rq_based;
895 for (i = 0; i < t->num_targets; i++) {
896 tgt = t->targets + i;
897 if (dm_target_hybrid(tgt))
899 else if (dm_target_request_based(tgt))
904 if (bio_based && request_based) {
905 DMERR("Inconsistent table: different target types"
906 " can't be mixed up");
911 if (hybrid && !bio_based && !request_based) {
913 * The targets can work either way.
914 * Determine the type from the live device.
915 * Default to bio-based if device is new.
917 if (__table_type_request_based(live_md_type))
925 /* We must use this table as bio-based */
926 t->type = DM_TYPE_BIO_BASED;
927 if (dm_table_supports_dax(t, device_supports_dax, &page_size) ||
928 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
929 t->type = DM_TYPE_DAX_BIO_BASED;
934 BUG_ON(!request_based); /* No targets in this table */
936 t->type = DM_TYPE_REQUEST_BASED;
940 * Request-based dm supports only tables that have a single target now.
941 * To support multiple targets, request splitting support is needed,
942 * and that needs lots of changes in the block-layer.
943 * (e.g. request completion process for partial completion.)
945 if (t->num_targets > 1) {
946 DMERR("request-based DM doesn't support multiple targets");
950 if (list_empty(devices)) {
952 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
954 /* inherit live table's type */
956 t->type = live_table->type;
957 dm_put_live_table(t->md, srcu_idx);
961 tgt = dm_table_get_immutable_target(t);
963 DMERR("table load rejected: immutable target is required");
965 } else if (tgt->max_io_len) {
966 DMERR("table load rejected: immutable target that splits IO is not supported");
970 /* Non-request-stackable devices can't be used for request-based dm */
971 if (!tgt->type->iterate_devices ||
972 !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
973 DMERR("table load rejected: including non-request-stackable devices");
980 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
985 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
987 return t->immutable_target_type;
990 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
992 /* Immutable target is implicitly a singleton */
993 if (t->num_targets > 1 ||
994 !dm_target_is_immutable(t->targets[0].type))
1000 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1002 struct dm_target *ti;
1005 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1006 ti = dm_table_get_target(t, i);
1007 if (dm_target_is_wildcard(ti->type))
1014 bool dm_table_bio_based(struct dm_table *t)
1016 return __table_type_bio_based(dm_table_get_type(t));
1019 bool dm_table_request_based(struct dm_table *t)
1021 return __table_type_request_based(dm_table_get_type(t));
1024 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1026 enum dm_queue_mode type = dm_table_get_type(t);
1027 unsigned per_io_data_size = 0;
1028 unsigned min_pool_size = 0;
1029 struct dm_target *ti;
1032 if (unlikely(type == DM_TYPE_NONE)) {
1033 DMWARN("no table type is set, can't allocate mempools");
1037 if (__table_type_bio_based(type))
1038 for (i = 0; i < t->num_targets; i++) {
1039 ti = t->targets + i;
1040 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1041 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1044 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1045 per_io_data_size, min_pool_size);
1052 void dm_table_free_md_mempools(struct dm_table *t)
1054 dm_free_md_mempools(t->mempools);
1058 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1063 static int setup_indexes(struct dm_table *t)
1066 unsigned int total = 0;
1069 /* allocate the space for *all* the indexes */
1070 for (i = t->depth - 2; i >= 0; i--) {
1071 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1072 total += t->counts[i];
1075 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1079 /* set up internal nodes, bottom-up */
1080 for (i = t->depth - 2; i >= 0; i--) {
1081 t->index[i] = indexes;
1082 indexes += (KEYS_PER_NODE * t->counts[i]);
1083 setup_btree_index(i, t);
1090 * Builds the btree to index the map.
1092 static int dm_table_build_index(struct dm_table *t)
1095 unsigned int leaf_nodes;
1097 /* how many indexes will the btree have ? */
1098 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1099 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1101 /* leaf layer has already been set up */
1102 t->counts[t->depth - 1] = leaf_nodes;
1103 t->index[t->depth - 1] = t->highs;
1106 r = setup_indexes(t);
1111 static bool integrity_profile_exists(struct gendisk *disk)
1113 return !!blk_get_integrity(disk);
1117 * Get a disk whose integrity profile reflects the table's profile.
1118 * Returns NULL if integrity support was inconsistent or unavailable.
1120 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1122 struct list_head *devices = dm_table_get_devices(t);
1123 struct dm_dev_internal *dd = NULL;
1124 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1127 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1128 struct dm_target *ti = dm_table_get_target(t, i);
1129 if (!dm_target_passes_integrity(ti->type))
1133 list_for_each_entry(dd, devices, list) {
1134 template_disk = dd->dm_dev->bdev->bd_disk;
1135 if (!integrity_profile_exists(template_disk))
1137 else if (prev_disk &&
1138 blk_integrity_compare(prev_disk, template_disk) < 0)
1140 prev_disk = template_disk;
1143 return template_disk;
1147 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1148 dm_device_name(t->md),
1149 prev_disk->disk_name,
1150 template_disk->disk_name);
1155 * Register the mapped device for blk_integrity support if the
1156 * underlying devices have an integrity profile. But all devices may
1157 * not have matching profiles (checking all devices isn't reliable
1158 * during table load because this table may use other DM device(s) which
1159 * must be resumed before they will have an initialized integity
1160 * profile). Consequently, stacked DM devices force a 2 stage integrity
1161 * profile validation: First pass during table load, final pass during
1164 static int dm_table_register_integrity(struct dm_table *t)
1166 struct mapped_device *md = t->md;
1167 struct gendisk *template_disk = NULL;
1169 /* If target handles integrity itself do not register it here. */
1170 if (t->integrity_added)
1173 template_disk = dm_table_get_integrity_disk(t);
1177 if (!integrity_profile_exists(dm_disk(md))) {
1178 t->integrity_supported = true;
1180 * Register integrity profile during table load; we can do
1181 * this because the final profile must match during resume.
1183 blk_integrity_register(dm_disk(md),
1184 blk_get_integrity(template_disk));
1189 * If DM device already has an initialized integrity
1190 * profile the new profile should not conflict.
1192 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1193 DMWARN("%s: conflict with existing integrity profile: "
1194 "%s profile mismatch",
1195 dm_device_name(t->md),
1196 template_disk->disk_name);
1200 /* Preserve existing integrity profile */
1201 t->integrity_supported = true;
1206 * Prepares the table for use by building the indices,
1207 * setting the type, and allocating mempools.
1209 int dm_table_complete(struct dm_table *t)
1213 r = dm_table_determine_type(t);
1215 DMERR("unable to determine table type");
1219 r = dm_table_build_index(t);
1221 DMERR("unable to build btrees");
1225 r = dm_table_register_integrity(t);
1227 DMERR("could not register integrity profile.");
1231 r = dm_table_alloc_md_mempools(t, t->md);
1233 DMERR("unable to allocate mempools");
1238 static DEFINE_MUTEX(_event_lock);
1239 void dm_table_event_callback(struct dm_table *t,
1240 void (*fn)(void *), void *context)
1242 mutex_lock(&_event_lock);
1244 t->event_context = context;
1245 mutex_unlock(&_event_lock);
1248 void dm_table_event(struct dm_table *t)
1251 * You can no longer call dm_table_event() from interrupt
1252 * context, use a bottom half instead.
1254 BUG_ON(in_interrupt());
1256 mutex_lock(&_event_lock);
1258 t->event_fn(t->event_context);
1259 mutex_unlock(&_event_lock);
1261 EXPORT_SYMBOL(dm_table_event);
1263 inline sector_t dm_table_get_size(struct dm_table *t)
1265 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1267 EXPORT_SYMBOL(dm_table_get_size);
1269 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1271 if (index >= t->num_targets)
1274 return t->targets + index;
1278 * Search the btree for the correct target.
1280 * Caller should check returned pointer for NULL
1281 * to trap I/O beyond end of device.
1283 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1285 unsigned int l, n = 0, k = 0;
1288 if (unlikely(sector >= dm_table_get_size(t)))
1291 for (l = 0; l < t->depth; l++) {
1292 n = get_child(n, k);
1293 node = get_node(t, l, n);
1295 for (k = 0; k < KEYS_PER_NODE; k++)
1296 if (node[k] >= sector)
1300 return &t->targets[(KEYS_PER_NODE * n) + k];
1303 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1304 sector_t start, sector_t len, void *data)
1306 unsigned *num_devices = data;
1314 * Check whether a table has no data devices attached using each
1315 * target's iterate_devices method.
1316 * Returns false if the result is unknown because a target doesn't
1317 * support iterate_devices.
1319 bool dm_table_has_no_data_devices(struct dm_table *table)
1321 struct dm_target *ti;
1322 unsigned i, num_devices;
1324 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1325 ti = dm_table_get_target(table, i);
1327 if (!ti->type->iterate_devices)
1331 ti->type->iterate_devices(ti, count_device, &num_devices);
1339 static int device_is_zoned_model(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);
1343 enum blk_zoned_model *zoned_model = data;
1345 return q && blk_queue_zoned_model(q) == *zoned_model;
1348 static bool dm_table_supports_zoned_model(struct dm_table *t,
1349 enum blk_zoned_model zoned_model)
1351 struct dm_target *ti;
1354 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1355 ti = dm_table_get_target(t, i);
1357 if (zoned_model == BLK_ZONED_HM &&
1358 !dm_target_supports_zoned_hm(ti->type))
1361 if (!ti->type->iterate_devices ||
1362 !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1369 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1370 sector_t start, sector_t len, void *data)
1372 struct request_queue *q = bdev_get_queue(dev->bdev);
1373 unsigned int *zone_sectors = data;
1375 return q && blk_queue_zone_sectors(q) == *zone_sectors;
1378 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1379 unsigned int zone_sectors)
1381 struct dm_target *ti;
1384 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1385 ti = dm_table_get_target(t, i);
1387 if (!ti->type->iterate_devices ||
1388 !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1395 static int validate_hardware_zoned_model(struct dm_table *table,
1396 enum blk_zoned_model zoned_model,
1397 unsigned int zone_sectors)
1399 if (zoned_model == BLK_ZONED_NONE)
1402 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1403 DMERR("%s: zoned model is not consistent across all devices",
1404 dm_device_name(table->md));
1408 /* Check zone size validity and compatibility */
1409 if (!zone_sectors || !is_power_of_2(zone_sectors))
1412 if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1413 DMERR("%s: zone sectors is not consistent across all devices",
1414 dm_device_name(table->md));
1422 * Establish the new table's queue_limits and validate them.
1424 int dm_calculate_queue_limits(struct dm_table *table,
1425 struct queue_limits *limits)
1427 struct dm_target *ti;
1428 struct queue_limits ti_limits;
1430 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1431 unsigned int zone_sectors = 0;
1433 blk_set_stacking_limits(limits);
1435 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1436 blk_set_stacking_limits(&ti_limits);
1438 ti = dm_table_get_target(table, i);
1440 if (!ti->type->iterate_devices)
1441 goto combine_limits;
1444 * Combine queue limits of all the devices this target uses.
1446 ti->type->iterate_devices(ti, dm_set_device_limits,
1449 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1451 * After stacking all limits, validate all devices
1452 * in table support this zoned model and zone sectors.
1454 zoned_model = ti_limits.zoned;
1455 zone_sectors = ti_limits.chunk_sectors;
1458 /* Stack chunk_sectors if target-specific splitting is required */
1460 ti_limits.chunk_sectors = lcm_not_zero(ti->max_io_len,
1461 ti_limits.chunk_sectors);
1462 /* Set I/O hints portion of queue limits */
1463 if (ti->type->io_hints)
1464 ti->type->io_hints(ti, &ti_limits);
1467 * Check each device area is consistent with the target's
1468 * overall queue limits.
1470 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1476 * Merge this target's queue limits into the overall limits
1479 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1480 DMWARN("%s: adding target device "
1481 "(start sect %llu len %llu) "
1482 "caused an alignment inconsistency",
1483 dm_device_name(table->md),
1484 (unsigned long long) ti->begin,
1485 (unsigned long long) ti->len);
1489 * Verify that the zoned model and zone sectors, as determined before
1490 * any .io_hints override, are the same across all devices in the table.
1491 * - this is especially relevant if .io_hints is emulating a disk-managed
1492 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1495 if (limits->zoned != BLK_ZONED_NONE) {
1497 * ...IF the above limits stacking determined a zoned model
1498 * validate that all of the table's devices conform to it.
1500 zoned_model = limits->zoned;
1501 zone_sectors = limits->chunk_sectors;
1503 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1506 return validate_hardware_logical_block_alignment(table, limits);
1510 * Verify that all devices have an integrity profile that matches the
1511 * DM device's registered integrity profile. If the profiles don't
1512 * match then unregister the DM device's integrity profile.
1514 static void dm_table_verify_integrity(struct dm_table *t)
1516 struct gendisk *template_disk = NULL;
1518 if (t->integrity_added)
1521 if (t->integrity_supported) {
1523 * Verify that the original integrity profile
1524 * matches all the devices in this table.
1526 template_disk = dm_table_get_integrity_disk(t);
1527 if (template_disk &&
1528 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1532 if (integrity_profile_exists(dm_disk(t->md))) {
1533 DMWARN("%s: unable to establish an integrity profile",
1534 dm_device_name(t->md));
1535 blk_integrity_unregister(dm_disk(t->md));
1539 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1540 sector_t start, sector_t len, void *data)
1542 unsigned long flush = (unsigned long) data;
1543 struct request_queue *q = bdev_get_queue(dev->bdev);
1545 return q && (q->queue_flags & flush);
1548 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1550 struct dm_target *ti;
1554 * Require at least one underlying device to support flushes.
1555 * t->devices includes internal dm devices such as mirror logs
1556 * so we need to use iterate_devices here, which targets
1557 * supporting flushes must provide.
1559 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1560 ti = dm_table_get_target(t, i);
1562 if (!ti->num_flush_bios)
1565 if (ti->flush_supported)
1568 if (ti->type->iterate_devices &&
1569 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1576 static int device_dax_write_cache_enabled(struct dm_target *ti,
1577 struct dm_dev *dev, sector_t start,
1578 sector_t len, void *data)
1580 struct dax_device *dax_dev = dev->dax_dev;
1585 if (dax_write_cache_enabled(dax_dev))
1590 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1592 struct dm_target *ti;
1595 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1596 ti = dm_table_get_target(t, i);
1598 if (ti->type->iterate_devices &&
1599 ti->type->iterate_devices(ti,
1600 device_dax_write_cache_enabled, NULL))
1607 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1608 sector_t start, sector_t len, void *data)
1610 struct request_queue *q = bdev_get_queue(dev->bdev);
1612 return q && blk_queue_nonrot(q);
1615 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1616 sector_t start, sector_t len, void *data)
1618 struct request_queue *q = bdev_get_queue(dev->bdev);
1620 return q && !blk_queue_add_random(q);
1623 static bool dm_table_all_devices_attribute(struct dm_table *t,
1624 iterate_devices_callout_fn func)
1626 struct dm_target *ti;
1629 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1630 ti = dm_table_get_target(t, i);
1632 if (!ti->type->iterate_devices ||
1633 !ti->type->iterate_devices(ti, func, NULL))
1640 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1641 sector_t start, sector_t len, void *data)
1643 struct request_queue *q = bdev_get_queue(dev->bdev);
1645 return q && !q->limits.max_write_same_sectors;
1648 static bool dm_table_supports_write_same(struct dm_table *t)
1650 struct dm_target *ti;
1653 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1654 ti = dm_table_get_target(t, i);
1656 if (!ti->num_write_same_bios)
1659 if (!ti->type->iterate_devices ||
1660 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1667 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1668 sector_t start, sector_t len, void *data)
1670 struct request_queue *q = bdev_get_queue(dev->bdev);
1672 return q && !q->limits.max_write_zeroes_sectors;
1675 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1677 struct dm_target *ti;
1680 while (i < dm_table_get_num_targets(t)) {
1681 ti = dm_table_get_target(t, i++);
1683 if (!ti->num_write_zeroes_bios)
1686 if (!ti->type->iterate_devices ||
1687 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1694 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1695 sector_t start, sector_t len, void *data)
1697 struct request_queue *q = bdev_get_queue(dev->bdev);
1699 return q && !blk_queue_nowait(q);
1702 static bool dm_table_supports_nowait(struct dm_table *t)
1704 struct dm_target *ti;
1707 while (i < dm_table_get_num_targets(t)) {
1708 ti = dm_table_get_target(t, i++);
1710 if (!dm_target_supports_nowait(ti->type))
1713 if (!ti->type->iterate_devices ||
1714 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1721 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1722 sector_t start, sector_t len, void *data)
1724 struct request_queue *q = bdev_get_queue(dev->bdev);
1726 return q && !blk_queue_discard(q);
1729 static bool dm_table_supports_discards(struct dm_table *t)
1731 struct dm_target *ti;
1734 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1735 ti = dm_table_get_target(t, i);
1737 if (!ti->num_discard_bios)
1741 * Either the target provides discard support (as implied by setting
1742 * 'discards_supported') or it relies on _all_ data devices having
1745 if (!ti->discards_supported &&
1746 (!ti->type->iterate_devices ||
1747 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1754 static int device_not_secure_erase_capable(struct dm_target *ti,
1755 struct dm_dev *dev, sector_t start,
1756 sector_t len, void *data)
1758 struct request_queue *q = bdev_get_queue(dev->bdev);
1760 return q && !blk_queue_secure_erase(q);
1763 static bool dm_table_supports_secure_erase(struct dm_table *t)
1765 struct dm_target *ti;
1768 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1769 ti = dm_table_get_target(t, i);
1771 if (!ti->num_secure_erase_bios)
1774 if (!ti->type->iterate_devices ||
1775 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1782 static int device_requires_stable_pages(struct dm_target *ti,
1783 struct dm_dev *dev, sector_t start,
1784 sector_t len, void *data)
1786 struct request_queue *q = bdev_get_queue(dev->bdev);
1788 return q && blk_queue_stable_writes(q);
1792 * If any underlying device requires stable pages, a table must require
1793 * them as well. Only targets that support iterate_devices are considered:
1794 * don't want error, zero, etc to require stable pages.
1796 static bool dm_table_requires_stable_pages(struct dm_table *t)
1798 struct dm_target *ti;
1801 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1802 ti = dm_table_get_target(t, i);
1804 if (ti->type->iterate_devices &&
1805 ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1812 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1813 struct queue_limits *limits)
1815 bool wc = false, fua = false;
1816 int page_size = PAGE_SIZE;
1819 * Copy table's limits to the DM device's request_queue
1821 q->limits = *limits;
1823 if (dm_table_supports_nowait(t))
1824 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1826 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1828 if (!dm_table_supports_discards(t)) {
1829 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1830 /* Must also clear discard limits... */
1831 q->limits.max_discard_sectors = 0;
1832 q->limits.max_hw_discard_sectors = 0;
1833 q->limits.discard_granularity = 0;
1834 q->limits.discard_alignment = 0;
1835 q->limits.discard_misaligned = 0;
1837 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1839 if (dm_table_supports_secure_erase(t))
1840 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1842 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1844 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1847 blk_queue_write_cache(q, wc, fua);
1849 if (dm_table_supports_dax(t, device_supports_dax, &page_size)) {
1850 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1851 if (dm_table_supports_dax(t, device_dax_synchronous, NULL))
1852 set_dax_synchronous(t->md->dax_dev);
1855 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1857 if (dm_table_supports_dax_write_cache(t))
1858 dax_write_cache(t->md->dax_dev, true);
1860 /* Ensure that all underlying devices are non-rotational. */
1861 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1862 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1864 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1866 if (!dm_table_supports_write_same(t))
1867 q->limits.max_write_same_sectors = 0;
1868 if (!dm_table_supports_write_zeroes(t))
1869 q->limits.max_write_zeroes_sectors = 0;
1871 dm_table_verify_integrity(t);
1874 * Some devices don't use blk_integrity but still want stable pages
1875 * because they do their own checksumming.
1877 if (dm_table_requires_stable_pages(t))
1878 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
1880 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
1883 * Determine whether or not this queue's I/O timings contribute
1884 * to the entropy pool, Only request-based targets use this.
1885 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1888 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1889 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1892 * For a zoned target, the number of zones should be updated for the
1893 * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1894 * target, this is all that is needed.
1896 #ifdef CONFIG_BLK_DEV_ZONED
1897 if (blk_queue_is_zoned(q)) {
1898 WARN_ON_ONCE(queue_is_mq(q));
1899 q->nr_zones = blkdev_nr_zones(t->md->disk);
1903 blk_queue_update_readahead(q);
1906 unsigned int dm_table_get_num_targets(struct dm_table *t)
1908 return t->num_targets;
1911 struct list_head *dm_table_get_devices(struct dm_table *t)
1916 fmode_t dm_table_get_mode(struct dm_table *t)
1920 EXPORT_SYMBOL(dm_table_get_mode);
1928 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1930 int i = t->num_targets;
1931 struct dm_target *ti = t->targets;
1933 lockdep_assert_held(&t->md->suspend_lock);
1938 if (ti->type->presuspend)
1939 ti->type->presuspend(ti);
1941 case PRESUSPEND_UNDO:
1942 if (ti->type->presuspend_undo)
1943 ti->type->presuspend_undo(ti);
1946 if (ti->type->postsuspend)
1947 ti->type->postsuspend(ti);
1954 void dm_table_presuspend_targets(struct dm_table *t)
1959 suspend_targets(t, PRESUSPEND);
1962 void dm_table_presuspend_undo_targets(struct dm_table *t)
1967 suspend_targets(t, PRESUSPEND_UNDO);
1970 void dm_table_postsuspend_targets(struct dm_table *t)
1975 suspend_targets(t, POSTSUSPEND);
1978 int dm_table_resume_targets(struct dm_table *t)
1982 lockdep_assert_held(&t->md->suspend_lock);
1984 for (i = 0; i < t->num_targets; i++) {
1985 struct dm_target *ti = t->targets + i;
1987 if (!ti->type->preresume)
1990 r = ti->type->preresume(ti);
1992 DMERR("%s: %s: preresume failed, error = %d",
1993 dm_device_name(t->md), ti->type->name, r);
1998 for (i = 0; i < t->num_targets; i++) {
1999 struct dm_target *ti = t->targets + i;
2001 if (ti->type->resume)
2002 ti->type->resume(ti);
2008 struct mapped_device *dm_table_get_md(struct dm_table *t)
2012 EXPORT_SYMBOL(dm_table_get_md);
2014 const char *dm_table_device_name(struct dm_table *t)
2016 return dm_device_name(t->md);
2018 EXPORT_SYMBOL_GPL(dm_table_device_name);
2020 void dm_table_run_md_queue_async(struct dm_table *t)
2022 if (!dm_table_request_based(t))
2026 blk_mq_run_hw_queues(t->md->queue, true);
2028 EXPORT_SYMBOL(dm_table_run_md_queue_async);