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)
352 if (lookup_bdev(path, &dev))
353 dev = name_to_dev_t(path);
356 EXPORT_SYMBOL_GPL(dm_get_dev_t);
359 * Add a device to the list, or just increment the usage count if
360 * it's already present.
362 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
363 struct dm_dev **result)
367 struct dm_dev_internal *dd;
368 struct dm_table *t = ti->table;
372 dev = dm_get_dev_t(path);
376 dd = find_device(&t->devices, dev);
378 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
382 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
387 refcount_set(&dd->count, 1);
388 list_add(&dd->list, &t->devices);
391 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
392 r = upgrade_mode(dd, mode, t->md);
396 refcount_inc(&dd->count);
398 *result = dd->dm_dev;
401 EXPORT_SYMBOL(dm_get_device);
403 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
404 sector_t start, sector_t len, void *data)
406 struct queue_limits *limits = data;
407 struct block_device *bdev = dev->bdev;
408 struct request_queue *q = bdev_get_queue(bdev);
409 char b[BDEVNAME_SIZE];
412 DMWARN("%s: Cannot set limits for nonexistent device %s",
413 dm_device_name(ti->table->md), bdevname(bdev, b));
417 if (blk_stack_limits(limits, &q->limits,
418 get_start_sect(bdev) + start) < 0)
419 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
420 "physical_block_size=%u, logical_block_size=%u, "
421 "alignment_offset=%u, start=%llu",
422 dm_device_name(ti->table->md), bdevname(bdev, b),
423 q->limits.physical_block_size,
424 q->limits.logical_block_size,
425 q->limits.alignment_offset,
426 (unsigned long long) start << SECTOR_SHIFT);
431 * Decrement a device's use count and remove it if necessary.
433 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
436 struct list_head *devices = &ti->table->devices;
437 struct dm_dev_internal *dd;
439 list_for_each_entry(dd, devices, list) {
440 if (dd->dm_dev == d) {
446 DMWARN("%s: device %s not in table devices list",
447 dm_device_name(ti->table->md), d->name);
450 if (refcount_dec_and_test(&dd->count)) {
451 dm_put_table_device(ti->table->md, d);
456 EXPORT_SYMBOL(dm_put_device);
459 * Checks to see if the target joins onto the end of the table.
461 static int adjoin(struct dm_table *table, struct dm_target *ti)
463 struct dm_target *prev;
465 if (!table->num_targets)
468 prev = &table->targets[table->num_targets - 1];
469 return (ti->begin == (prev->begin + prev->len));
473 * Used to dynamically allocate the arg array.
475 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
476 * process messages even if some device is suspended. These messages have a
477 * small fixed number of arguments.
479 * On the other hand, dm-switch needs to process bulk data using messages and
480 * excessive use of GFP_NOIO could cause trouble.
482 static char **realloc_argv(unsigned *size, char **old_argv)
489 new_size = *size * 2;
495 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
496 if (argv && old_argv) {
497 memcpy(argv, old_argv, *size * sizeof(*argv));
506 * Destructively splits up the argument list to pass to ctr.
508 int dm_split_args(int *argc, char ***argvp, char *input)
510 char *start, *end = input, *out, **argv = NULL;
511 unsigned array_size = 0;
520 argv = realloc_argv(&array_size, argv);
525 /* Skip whitespace */
526 start = skip_spaces(end);
529 break; /* success, we hit the end */
531 /* 'out' is used to remove any back-quotes */
534 /* Everything apart from '\0' can be quoted */
535 if (*end == '\\' && *(end + 1)) {
542 break; /* end of token */
547 /* have we already filled the array ? */
548 if ((*argc + 1) > array_size) {
549 argv = realloc_argv(&array_size, argv);
554 /* we know this is whitespace */
558 /* terminate the string and put it in the array */
569 * Impose necessary and sufficient conditions on a devices's table such
570 * that any incoming bio which respects its logical_block_size can be
571 * processed successfully. If it falls across the boundary between
572 * two or more targets, the size of each piece it gets split into must
573 * be compatible with the logical_block_size of the target processing it.
575 static int validate_hardware_logical_block_alignment(struct dm_table *table,
576 struct queue_limits *limits)
579 * This function uses arithmetic modulo the logical_block_size
580 * (in units of 512-byte sectors).
582 unsigned short device_logical_block_size_sects =
583 limits->logical_block_size >> SECTOR_SHIFT;
586 * Offset of the start of the next table entry, mod logical_block_size.
588 unsigned short next_target_start = 0;
591 * Given an aligned bio that extends beyond the end of a
592 * target, how many sectors must the next target handle?
594 unsigned short remaining = 0;
596 struct dm_target *ti;
597 struct queue_limits ti_limits;
601 * Check each entry in the table in turn.
603 for (i = 0; i < dm_table_get_num_targets(table); i++) {
604 ti = dm_table_get_target(table, i);
606 blk_set_stacking_limits(&ti_limits);
608 /* combine all target devices' limits */
609 if (ti->type->iterate_devices)
610 ti->type->iterate_devices(ti, dm_set_device_limits,
614 * If the remaining sectors fall entirely within this
615 * table entry are they compatible with its logical_block_size?
617 if (remaining < ti->len &&
618 remaining & ((ti_limits.logical_block_size >>
623 (unsigned short) ((next_target_start + ti->len) &
624 (device_logical_block_size_sects - 1));
625 remaining = next_target_start ?
626 device_logical_block_size_sects - next_target_start : 0;
630 DMWARN("%s: table line %u (start sect %llu len %llu) "
631 "not aligned to h/w logical block size %u",
632 dm_device_name(table->md), i,
633 (unsigned long long) ti->begin,
634 (unsigned long long) ti->len,
635 limits->logical_block_size);
642 int dm_table_add_target(struct dm_table *t, const char *type,
643 sector_t start, sector_t len, char *params)
645 int r = -EINVAL, argc;
647 struct dm_target *tgt;
650 DMERR("%s: target type %s must appear alone in table",
651 dm_device_name(t->md), t->targets->type->name);
655 BUG_ON(t->num_targets >= t->num_allocated);
657 tgt = t->targets + t->num_targets;
658 memset(tgt, 0, sizeof(*tgt));
661 DMERR("%s: zero-length target", dm_device_name(t->md));
665 tgt->type = dm_get_target_type(type);
667 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
671 if (dm_target_needs_singleton(tgt->type)) {
672 if (t->num_targets) {
673 tgt->error = "singleton target type must appear alone in table";
679 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
680 tgt->error = "target type may not be included in a read-only table";
684 if (t->immutable_target_type) {
685 if (t->immutable_target_type != tgt->type) {
686 tgt->error = "immutable target type cannot be mixed with other target types";
689 } else if (dm_target_is_immutable(tgt->type)) {
690 if (t->num_targets) {
691 tgt->error = "immutable target type cannot be mixed with other target types";
694 t->immutable_target_type = tgt->type;
697 if (dm_target_has_integrity(tgt->type))
698 t->integrity_added = 1;
703 tgt->error = "Unknown error";
706 * Does this target adjoin the previous one ?
708 if (!adjoin(t, tgt)) {
709 tgt->error = "Gap in table";
713 r = dm_split_args(&argc, &argv, params);
715 tgt->error = "couldn't split parameters (insufficient memory)";
719 r = tgt->type->ctr(tgt, argc, argv);
724 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
726 if (!tgt->num_discard_bios && tgt->discards_supported)
727 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
728 dm_device_name(t->md), type);
733 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
734 dm_put_target_type(tgt->type);
739 * Target argument parsing helpers.
741 static int validate_next_arg(const struct dm_arg *arg,
742 struct dm_arg_set *arg_set,
743 unsigned *value, char **error, unsigned grouped)
745 const char *arg_str = dm_shift_arg(arg_set);
749 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
750 (*value < arg->min) ||
751 (*value > arg->max) ||
752 (grouped && arg_set->argc < *value)) {
760 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
761 unsigned *value, char **error)
763 return validate_next_arg(arg, arg_set, value, error, 0);
765 EXPORT_SYMBOL(dm_read_arg);
767 int dm_read_arg_group(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, 1);
772 EXPORT_SYMBOL(dm_read_arg_group);
774 const char *dm_shift_arg(struct dm_arg_set *as)
787 EXPORT_SYMBOL(dm_shift_arg);
789 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
791 BUG_ON(as->argc < num_args);
792 as->argc -= num_args;
793 as->argv += num_args;
795 EXPORT_SYMBOL(dm_consume_args);
797 static bool __table_type_bio_based(enum dm_queue_mode table_type)
799 return (table_type == DM_TYPE_BIO_BASED ||
800 table_type == DM_TYPE_DAX_BIO_BASED);
803 static bool __table_type_request_based(enum dm_queue_mode table_type)
805 return table_type == DM_TYPE_REQUEST_BASED;
808 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
812 EXPORT_SYMBOL_GPL(dm_table_set_type);
814 /* validate the dax capability of the target device span */
815 int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
816 sector_t start, sector_t len, void *data)
818 int blocksize = *(int *) data, id;
821 id = dax_read_lock();
822 rc = dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
828 /* Check devices support synchronous DAX */
829 static int device_dax_synchronous(struct dm_target *ti, struct dm_dev *dev,
830 sector_t start, sector_t len, void *data)
832 return dev->dax_dev && dax_synchronous(dev->dax_dev);
835 bool dm_table_supports_dax(struct dm_table *t,
836 iterate_devices_callout_fn iterate_fn, int *blocksize)
838 struct dm_target *ti;
841 /* Ensure that all targets support DAX. */
842 for (i = 0; i < dm_table_get_num_targets(t); i++) {
843 ti = dm_table_get_target(t, i);
845 if (!ti->type->direct_access)
848 if (!ti->type->iterate_devices ||
849 !ti->type->iterate_devices(ti, iterate_fn, blocksize))
856 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
857 sector_t start, sector_t len, void *data)
859 struct block_device *bdev = dev->bdev;
860 struct request_queue *q = bdev_get_queue(bdev);
862 /* request-based cannot stack on partitions! */
863 if (bdev_is_partition(bdev))
866 return queue_is_mq(q);
869 static int dm_table_determine_type(struct dm_table *t)
872 unsigned bio_based = 0, request_based = 0, hybrid = 0;
873 struct dm_target *tgt;
874 struct list_head *devices = dm_table_get_devices(t);
875 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
876 int page_size = PAGE_SIZE;
878 if (t->type != DM_TYPE_NONE) {
879 /* target already set the table's type */
880 if (t->type == DM_TYPE_BIO_BASED) {
881 /* possibly upgrade to a variant of bio-based */
882 goto verify_bio_based;
884 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
885 goto verify_rq_based;
888 for (i = 0; i < t->num_targets; i++) {
889 tgt = t->targets + i;
890 if (dm_target_hybrid(tgt))
892 else if (dm_target_request_based(tgt))
897 if (bio_based && request_based) {
898 DMERR("Inconsistent table: different target types"
899 " can't be mixed up");
904 if (hybrid && !bio_based && !request_based) {
906 * The targets can work either way.
907 * Determine the type from the live device.
908 * Default to bio-based if device is new.
910 if (__table_type_request_based(live_md_type))
918 /* We must use this table as bio-based */
919 t->type = DM_TYPE_BIO_BASED;
920 if (dm_table_supports_dax(t, device_supports_dax, &page_size) ||
921 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
922 t->type = DM_TYPE_DAX_BIO_BASED;
927 BUG_ON(!request_based); /* No targets in this table */
929 t->type = DM_TYPE_REQUEST_BASED;
933 * Request-based dm supports only tables that have a single target now.
934 * To support multiple targets, request splitting support is needed,
935 * and that needs lots of changes in the block-layer.
936 * (e.g. request completion process for partial completion.)
938 if (t->num_targets > 1) {
939 DMERR("request-based DM doesn't support multiple targets");
943 if (list_empty(devices)) {
945 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
947 /* inherit live table's type */
949 t->type = live_table->type;
950 dm_put_live_table(t->md, srcu_idx);
954 tgt = dm_table_get_immutable_target(t);
956 DMERR("table load rejected: immutable target is required");
958 } else if (tgt->max_io_len) {
959 DMERR("table load rejected: immutable target that splits IO is not supported");
963 /* Non-request-stackable devices can't be used for request-based dm */
964 if (!tgt->type->iterate_devices ||
965 !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
966 DMERR("table load rejected: including non-request-stackable devices");
973 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
978 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
980 return t->immutable_target_type;
983 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
985 /* Immutable target is implicitly a singleton */
986 if (t->num_targets > 1 ||
987 !dm_target_is_immutable(t->targets[0].type))
993 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
995 struct dm_target *ti;
998 for (i = 0; i < dm_table_get_num_targets(t); i++) {
999 ti = dm_table_get_target(t, i);
1000 if (dm_target_is_wildcard(ti->type))
1007 bool dm_table_bio_based(struct dm_table *t)
1009 return __table_type_bio_based(dm_table_get_type(t));
1012 bool dm_table_request_based(struct dm_table *t)
1014 return __table_type_request_based(dm_table_get_type(t));
1017 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1019 enum dm_queue_mode type = dm_table_get_type(t);
1020 unsigned per_io_data_size = 0;
1021 unsigned min_pool_size = 0;
1022 struct dm_target *ti;
1025 if (unlikely(type == DM_TYPE_NONE)) {
1026 DMWARN("no table type is set, can't allocate mempools");
1030 if (__table_type_bio_based(type))
1031 for (i = 0; i < t->num_targets; i++) {
1032 ti = t->targets + i;
1033 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1034 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1037 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1038 per_io_data_size, min_pool_size);
1045 void dm_table_free_md_mempools(struct dm_table *t)
1047 dm_free_md_mempools(t->mempools);
1051 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1056 static int setup_indexes(struct dm_table *t)
1059 unsigned int total = 0;
1062 /* allocate the space for *all* the indexes */
1063 for (i = t->depth - 2; i >= 0; i--) {
1064 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1065 total += t->counts[i];
1068 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1072 /* set up internal nodes, bottom-up */
1073 for (i = t->depth - 2; i >= 0; i--) {
1074 t->index[i] = indexes;
1075 indexes += (KEYS_PER_NODE * t->counts[i]);
1076 setup_btree_index(i, t);
1083 * Builds the btree to index the map.
1085 static int dm_table_build_index(struct dm_table *t)
1088 unsigned int leaf_nodes;
1090 /* how many indexes will the btree have ? */
1091 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1092 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1094 /* leaf layer has already been set up */
1095 t->counts[t->depth - 1] = leaf_nodes;
1096 t->index[t->depth - 1] = t->highs;
1099 r = setup_indexes(t);
1104 static bool integrity_profile_exists(struct gendisk *disk)
1106 return !!blk_get_integrity(disk);
1110 * Get a disk whose integrity profile reflects the table's profile.
1111 * Returns NULL if integrity support was inconsistent or unavailable.
1113 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1115 struct list_head *devices = dm_table_get_devices(t);
1116 struct dm_dev_internal *dd = NULL;
1117 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1120 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1121 struct dm_target *ti = dm_table_get_target(t, i);
1122 if (!dm_target_passes_integrity(ti->type))
1126 list_for_each_entry(dd, devices, list) {
1127 template_disk = dd->dm_dev->bdev->bd_disk;
1128 if (!integrity_profile_exists(template_disk))
1130 else if (prev_disk &&
1131 blk_integrity_compare(prev_disk, template_disk) < 0)
1133 prev_disk = template_disk;
1136 return template_disk;
1140 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1141 dm_device_name(t->md),
1142 prev_disk->disk_name,
1143 template_disk->disk_name);
1148 * Register the mapped device for blk_integrity support if the
1149 * underlying devices have an integrity profile. But all devices may
1150 * not have matching profiles (checking all devices isn't reliable
1151 * during table load because this table may use other DM device(s) which
1152 * must be resumed before they will have an initialized integity
1153 * profile). Consequently, stacked DM devices force a 2 stage integrity
1154 * profile validation: First pass during table load, final pass during
1157 static int dm_table_register_integrity(struct dm_table *t)
1159 struct mapped_device *md = t->md;
1160 struct gendisk *template_disk = NULL;
1162 /* If target handles integrity itself do not register it here. */
1163 if (t->integrity_added)
1166 template_disk = dm_table_get_integrity_disk(t);
1170 if (!integrity_profile_exists(dm_disk(md))) {
1171 t->integrity_supported = true;
1173 * Register integrity profile during table load; we can do
1174 * this because the final profile must match during resume.
1176 blk_integrity_register(dm_disk(md),
1177 blk_get_integrity(template_disk));
1182 * If DM device already has an initialized integrity
1183 * profile the new profile should not conflict.
1185 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1186 DMWARN("%s: conflict with existing integrity profile: "
1187 "%s profile mismatch",
1188 dm_device_name(t->md),
1189 template_disk->disk_name);
1193 /* Preserve existing integrity profile */
1194 t->integrity_supported = true;
1199 * Prepares the table for use by building the indices,
1200 * setting the type, and allocating mempools.
1202 int dm_table_complete(struct dm_table *t)
1206 r = dm_table_determine_type(t);
1208 DMERR("unable to determine table type");
1212 r = dm_table_build_index(t);
1214 DMERR("unable to build btrees");
1218 r = dm_table_register_integrity(t);
1220 DMERR("could not register integrity profile.");
1224 r = dm_table_alloc_md_mempools(t, t->md);
1226 DMERR("unable to allocate mempools");
1231 static DEFINE_MUTEX(_event_lock);
1232 void dm_table_event_callback(struct dm_table *t,
1233 void (*fn)(void *), void *context)
1235 mutex_lock(&_event_lock);
1237 t->event_context = context;
1238 mutex_unlock(&_event_lock);
1241 void dm_table_event(struct dm_table *t)
1244 * You can no longer call dm_table_event() from interrupt
1245 * context, use a bottom half instead.
1247 BUG_ON(in_interrupt());
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 /* Stack chunk_sectors if target-specific splitting is required */
1453 ti_limits.chunk_sectors = lcm_not_zero(ti->max_io_len,
1454 ti_limits.chunk_sectors);
1455 /* Set I/O hints portion of queue limits */
1456 if (ti->type->io_hints)
1457 ti->type->io_hints(ti, &ti_limits);
1460 * Check each device area is consistent with the target's
1461 * overall queue limits.
1463 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1469 * Merge this target's queue limits into the overall limits
1472 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1473 DMWARN("%s: adding target device "
1474 "(start sect %llu len %llu) "
1475 "caused an alignment inconsistency",
1476 dm_device_name(table->md),
1477 (unsigned long long) ti->begin,
1478 (unsigned long long) ti->len);
1482 * Verify that the zoned model and zone sectors, as determined before
1483 * any .io_hints override, are the same across all devices in the table.
1484 * - this is especially relevant if .io_hints is emulating a disk-managed
1485 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1488 if (limits->zoned != BLK_ZONED_NONE) {
1490 * ...IF the above limits stacking determined a zoned model
1491 * validate that all of the table's devices conform to it.
1493 zoned_model = limits->zoned;
1494 zone_sectors = limits->chunk_sectors;
1496 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1499 return validate_hardware_logical_block_alignment(table, limits);
1503 * Verify that all devices have an integrity profile that matches the
1504 * DM device's registered integrity profile. If the profiles don't
1505 * match then unregister the DM device's integrity profile.
1507 static void dm_table_verify_integrity(struct dm_table *t)
1509 struct gendisk *template_disk = NULL;
1511 if (t->integrity_added)
1514 if (t->integrity_supported) {
1516 * Verify that the original integrity profile
1517 * matches all the devices in this table.
1519 template_disk = dm_table_get_integrity_disk(t);
1520 if (template_disk &&
1521 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1525 if (integrity_profile_exists(dm_disk(t->md))) {
1526 DMWARN("%s: unable to establish an integrity profile",
1527 dm_device_name(t->md));
1528 blk_integrity_unregister(dm_disk(t->md));
1532 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1533 sector_t start, sector_t len, void *data)
1535 unsigned long flush = (unsigned long) data;
1536 struct request_queue *q = bdev_get_queue(dev->bdev);
1538 return q && (q->queue_flags & flush);
1541 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1543 struct dm_target *ti;
1547 * Require at least one underlying device to support flushes.
1548 * t->devices includes internal dm devices such as mirror logs
1549 * so we need to use iterate_devices here, which targets
1550 * supporting flushes must provide.
1552 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1553 ti = dm_table_get_target(t, i);
1555 if (!ti->num_flush_bios)
1558 if (ti->flush_supported)
1561 if (ti->type->iterate_devices &&
1562 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1569 static int device_dax_write_cache_enabled(struct dm_target *ti,
1570 struct dm_dev *dev, sector_t start,
1571 sector_t len, void *data)
1573 struct dax_device *dax_dev = dev->dax_dev;
1578 if (dax_write_cache_enabled(dax_dev))
1583 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1585 struct dm_target *ti;
1588 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1589 ti = dm_table_get_target(t, i);
1591 if (ti->type->iterate_devices &&
1592 ti->type->iterate_devices(ti,
1593 device_dax_write_cache_enabled, NULL))
1600 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1601 sector_t start, sector_t len, void *data)
1603 struct request_queue *q = bdev_get_queue(dev->bdev);
1605 return q && blk_queue_nonrot(q);
1608 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1609 sector_t start, sector_t len, void *data)
1611 struct request_queue *q = bdev_get_queue(dev->bdev);
1613 return q && !blk_queue_add_random(q);
1616 static bool dm_table_all_devices_attribute(struct dm_table *t,
1617 iterate_devices_callout_fn func)
1619 struct dm_target *ti;
1622 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1623 ti = dm_table_get_target(t, i);
1625 if (!ti->type->iterate_devices ||
1626 !ti->type->iterate_devices(ti, func, NULL))
1633 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1634 sector_t start, sector_t len, void *data)
1636 struct request_queue *q = bdev_get_queue(dev->bdev);
1638 return q && !q->limits.max_write_same_sectors;
1641 static bool dm_table_supports_write_same(struct dm_table *t)
1643 struct dm_target *ti;
1646 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1647 ti = dm_table_get_target(t, i);
1649 if (!ti->num_write_same_bios)
1652 if (!ti->type->iterate_devices ||
1653 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1660 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1661 sector_t start, sector_t len, void *data)
1663 struct request_queue *q = bdev_get_queue(dev->bdev);
1665 return q && !q->limits.max_write_zeroes_sectors;
1668 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1670 struct dm_target *ti;
1673 while (i < dm_table_get_num_targets(t)) {
1674 ti = dm_table_get_target(t, i++);
1676 if (!ti->num_write_zeroes_bios)
1679 if (!ti->type->iterate_devices ||
1680 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1687 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1688 sector_t start, sector_t len, void *data)
1690 struct request_queue *q = bdev_get_queue(dev->bdev);
1692 return q && !blk_queue_nowait(q);
1695 static bool dm_table_supports_nowait(struct dm_table *t)
1697 struct dm_target *ti;
1700 while (i < dm_table_get_num_targets(t)) {
1701 ti = dm_table_get_target(t, i++);
1703 if (!dm_target_supports_nowait(ti->type))
1706 if (!ti->type->iterate_devices ||
1707 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1714 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1715 sector_t start, sector_t len, void *data)
1717 struct request_queue *q = bdev_get_queue(dev->bdev);
1719 return q && !blk_queue_discard(q);
1722 static bool dm_table_supports_discards(struct dm_table *t)
1724 struct dm_target *ti;
1727 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1728 ti = dm_table_get_target(t, i);
1730 if (!ti->num_discard_bios)
1734 * Either the target provides discard support (as implied by setting
1735 * 'discards_supported') or it relies on _all_ data devices having
1738 if (!ti->discards_supported &&
1739 (!ti->type->iterate_devices ||
1740 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1747 static int device_not_secure_erase_capable(struct dm_target *ti,
1748 struct dm_dev *dev, sector_t start,
1749 sector_t len, void *data)
1751 struct request_queue *q = bdev_get_queue(dev->bdev);
1753 return q && !blk_queue_secure_erase(q);
1756 static bool dm_table_supports_secure_erase(struct dm_table *t)
1758 struct dm_target *ti;
1761 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1762 ti = dm_table_get_target(t, i);
1764 if (!ti->num_secure_erase_bios)
1767 if (!ti->type->iterate_devices ||
1768 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1775 static int device_requires_stable_pages(struct dm_target *ti,
1776 struct dm_dev *dev, sector_t start,
1777 sector_t len, void *data)
1779 struct request_queue *q = bdev_get_queue(dev->bdev);
1781 return q && blk_queue_stable_writes(q);
1785 * If any underlying device requires stable pages, a table must require
1786 * them as well. Only targets that support iterate_devices are considered:
1787 * don't want error, zero, etc to require stable pages.
1789 static bool dm_table_requires_stable_pages(struct dm_table *t)
1791 struct dm_target *ti;
1794 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1795 ti = dm_table_get_target(t, i);
1797 if (ti->type->iterate_devices &&
1798 ti->type->iterate_devices(ti, device_requires_stable_pages, NULL))
1805 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1806 struct queue_limits *limits)
1808 bool wc = false, fua = false;
1809 int page_size = PAGE_SIZE;
1812 * Copy table's limits to the DM device's request_queue
1814 q->limits = *limits;
1816 if (dm_table_supports_nowait(t))
1817 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1819 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1821 if (!dm_table_supports_discards(t)) {
1822 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
1823 /* Must also clear discard limits... */
1824 q->limits.max_discard_sectors = 0;
1825 q->limits.max_hw_discard_sectors = 0;
1826 q->limits.discard_granularity = 0;
1827 q->limits.discard_alignment = 0;
1828 q->limits.discard_misaligned = 0;
1830 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
1832 if (dm_table_supports_secure_erase(t))
1833 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
1835 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1837 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1840 blk_queue_write_cache(q, wc, fua);
1842 if (dm_table_supports_dax(t, device_supports_dax, &page_size)) {
1843 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
1844 if (dm_table_supports_dax(t, device_dax_synchronous, NULL))
1845 set_dax_synchronous(t->md->dax_dev);
1848 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
1850 if (dm_table_supports_dax_write_cache(t))
1851 dax_write_cache(t->md->dax_dev, true);
1853 /* Ensure that all underlying devices are non-rotational. */
1854 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1855 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
1857 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
1859 if (!dm_table_supports_write_same(t))
1860 q->limits.max_write_same_sectors = 0;
1861 if (!dm_table_supports_write_zeroes(t))
1862 q->limits.max_write_zeroes_sectors = 0;
1864 dm_table_verify_integrity(t);
1867 * Some devices don't use blk_integrity but still want stable pages
1868 * because they do their own checksumming.
1870 if (dm_table_requires_stable_pages(t))
1871 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
1873 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
1876 * Determine whether or not this queue's I/O timings contribute
1877 * to the entropy pool, Only request-based targets use this.
1878 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1881 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1882 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
1885 * For a zoned target, the number of zones should be updated for the
1886 * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
1887 * target, this is all that is needed.
1889 #ifdef CONFIG_BLK_DEV_ZONED
1890 if (blk_queue_is_zoned(q)) {
1891 WARN_ON_ONCE(queue_is_mq(q));
1892 q->nr_zones = blkdev_nr_zones(t->md->disk);
1896 blk_queue_update_readahead(q);
1899 unsigned int dm_table_get_num_targets(struct dm_table *t)
1901 return t->num_targets;
1904 struct list_head *dm_table_get_devices(struct dm_table *t)
1909 fmode_t dm_table_get_mode(struct dm_table *t)
1913 EXPORT_SYMBOL(dm_table_get_mode);
1921 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1923 int i = t->num_targets;
1924 struct dm_target *ti = t->targets;
1926 lockdep_assert_held(&t->md->suspend_lock);
1931 if (ti->type->presuspend)
1932 ti->type->presuspend(ti);
1934 case PRESUSPEND_UNDO:
1935 if (ti->type->presuspend_undo)
1936 ti->type->presuspend_undo(ti);
1939 if (ti->type->postsuspend)
1940 ti->type->postsuspend(ti);
1947 void dm_table_presuspend_targets(struct dm_table *t)
1952 suspend_targets(t, PRESUSPEND);
1955 void dm_table_presuspend_undo_targets(struct dm_table *t)
1960 suspend_targets(t, PRESUSPEND_UNDO);
1963 void dm_table_postsuspend_targets(struct dm_table *t)
1968 suspend_targets(t, POSTSUSPEND);
1971 int dm_table_resume_targets(struct dm_table *t)
1975 lockdep_assert_held(&t->md->suspend_lock);
1977 for (i = 0; i < t->num_targets; i++) {
1978 struct dm_target *ti = t->targets + i;
1980 if (!ti->type->preresume)
1983 r = ti->type->preresume(ti);
1985 DMERR("%s: %s: preresume failed, error = %d",
1986 dm_device_name(t->md), ti->type->name, r);
1991 for (i = 0; i < t->num_targets; i++) {
1992 struct dm_target *ti = t->targets + i;
1994 if (ti->type->resume)
1995 ti->type->resume(ti);
2001 struct mapped_device *dm_table_get_md(struct dm_table *t)
2005 EXPORT_SYMBOL(dm_table_get_md);
2007 const char *dm_table_device_name(struct dm_table *t)
2009 return dm_device_name(t->md);
2011 EXPORT_SYMBOL_GPL(dm_table_device_name);
2013 void dm_table_run_md_queue_async(struct dm_table *t)
2015 if (!dm_table_request_based(t))
2019 blk_mq_run_hw_queues(t->md->queue, true);
2021 EXPORT_SYMBOL(dm_table_run_md_queue_async);