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/blk-integrity.h>
14 #include <linux/namei.h>
15 #include <linux/ctype.h>
16 #include <linux/string.h>
17 #include <linux/slab.h>
18 #include <linux/interrupt.h>
19 #include <linux/mutex.h>
20 #include <linux/delay.h>
21 #include <linux/atomic.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));
99 * highs, and targets are managed as dynamic arrays during a
102 static int alloc_targets(struct dm_table *t, unsigned int num)
105 struct dm_target *n_targets;
108 * Allocate both the target array and offset array at once.
110 n_highs = kvcalloc(num, sizeof(struct dm_target) + sizeof(sector_t),
115 n_targets = (struct dm_target *) (n_highs + num);
117 memset(n_highs, -1, sizeof(*n_highs) * num);
120 t->num_allocated = num;
122 t->targets = n_targets;
127 int dm_table_create(struct dm_table **result, fmode_t mode,
128 unsigned num_targets, struct mapped_device *md)
130 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
135 INIT_LIST_HEAD(&t->devices);
138 num_targets = KEYS_PER_NODE;
140 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
147 if (alloc_targets(t, num_targets)) {
152 t->type = DM_TYPE_NONE;
159 static void free_devices(struct list_head *devices, struct mapped_device *md)
161 struct list_head *tmp, *next;
163 list_for_each_safe(tmp, next, devices) {
164 struct dm_dev_internal *dd =
165 list_entry(tmp, struct dm_dev_internal, list);
166 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
167 dm_device_name(md), dd->dm_dev->name);
168 dm_put_table_device(md, dd->dm_dev);
173 static void dm_table_destroy_crypto_profile(struct dm_table *t);
175 void dm_table_destroy(struct dm_table *t)
182 /* free the indexes */
184 kvfree(t->index[t->depth - 2]);
186 /* free the targets */
187 for (i = 0; i < t->num_targets; i++) {
188 struct dm_target *tgt = t->targets + i;
193 dm_put_target_type(tgt->type);
198 /* free the device list */
199 free_devices(&t->devices, t->md);
201 dm_free_md_mempools(t->mempools);
203 dm_table_destroy_crypto_profile(t);
209 * See if we've already got a device in the list.
211 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
213 struct dm_dev_internal *dd;
215 list_for_each_entry (dd, l, list)
216 if (dd->dm_dev->bdev->bd_dev == dev)
223 * If possible, this checks an area of a destination device is invalid.
225 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
226 sector_t start, sector_t len, void *data)
228 struct queue_limits *limits = data;
229 struct block_device *bdev = dev->bdev;
230 sector_t dev_size = bdev_nr_sectors(bdev);
231 unsigned short logical_block_size_sectors =
232 limits->logical_block_size >> SECTOR_SHIFT;
237 if ((start >= dev_size) || (start + len > dev_size)) {
238 DMWARN("%s: %pg too small for target: "
239 "start=%llu, len=%llu, dev_size=%llu",
240 dm_device_name(ti->table->md), bdev,
241 (unsigned long long)start,
242 (unsigned long long)len,
243 (unsigned long long)dev_size);
248 * If the target is mapped to zoned block device(s), check
249 * that the zones are not partially mapped.
251 if (bdev_is_zoned(bdev)) {
252 unsigned int zone_sectors = bdev_zone_sectors(bdev);
254 if (start & (zone_sectors - 1)) {
255 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %pg",
256 dm_device_name(ti->table->md),
257 (unsigned long long)start,
263 * Note: The last zone of a zoned block device may be smaller
264 * than other zones. So for a target mapping the end of a
265 * zoned block device with such a zone, len would not be zone
266 * aligned. We do not allow such last smaller zone to be part
267 * of the mapping here to ensure that mappings with multiple
268 * devices do not end up with a smaller zone in the middle of
271 if (len & (zone_sectors - 1)) {
272 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %pg",
273 dm_device_name(ti->table->md),
274 (unsigned long long)len,
280 if (logical_block_size_sectors <= 1)
283 if (start & (logical_block_size_sectors - 1)) {
284 DMWARN("%s: start=%llu not aligned to h/w "
285 "logical block size %u of %pg",
286 dm_device_name(ti->table->md),
287 (unsigned long long)start,
288 limits->logical_block_size, bdev);
292 if (len & (logical_block_size_sectors - 1)) {
293 DMWARN("%s: len=%llu not aligned to h/w "
294 "logical block size %u of %pg",
295 dm_device_name(ti->table->md),
296 (unsigned long long)len,
297 limits->logical_block_size, bdev);
305 * This upgrades the mode on an already open dm_dev, being
306 * careful to leave things as they were if we fail to reopen the
307 * device and not to touch the existing bdev field in case
308 * it is accessed concurrently.
310 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
311 struct mapped_device *md)
314 struct dm_dev *old_dev, *new_dev;
316 old_dev = dd->dm_dev;
318 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
319 dd->dm_dev->mode | new_mode, &new_dev);
323 dd->dm_dev = new_dev;
324 dm_put_table_device(md, old_dev);
330 * Convert the path to a device
332 dev_t dm_get_dev_t(const char *path)
336 if (lookup_bdev(path, &dev))
337 dev = name_to_dev_t(path);
340 EXPORT_SYMBOL_GPL(dm_get_dev_t);
343 * Add a device to the list, or just increment the usage count if
344 * it's already present.
346 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
347 struct dm_dev **result)
351 unsigned int major, minor;
353 struct dm_dev_internal *dd;
354 struct dm_table *t = ti->table;
358 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
359 /* Extract the major/minor numbers */
360 dev = MKDEV(major, minor);
361 if (MAJOR(dev) != major || MINOR(dev) != minor)
364 dev = dm_get_dev_t(path);
369 dd = find_device(&t->devices, dev);
371 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
375 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
380 refcount_set(&dd->count, 1);
381 list_add(&dd->list, &t->devices);
384 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
385 r = upgrade_mode(dd, mode, t->md);
389 refcount_inc(&dd->count);
391 *result = dd->dm_dev;
394 EXPORT_SYMBOL(dm_get_device);
396 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
397 sector_t start, sector_t len, void *data)
399 struct queue_limits *limits = data;
400 struct block_device *bdev = dev->bdev;
401 struct request_queue *q = bdev_get_queue(bdev);
404 DMWARN("%s: Cannot set limits for nonexistent device %pg",
405 dm_device_name(ti->table->md), bdev);
409 if (blk_stack_limits(limits, &q->limits,
410 get_start_sect(bdev) + start) < 0)
411 DMWARN("%s: adding target device %pg caused an alignment inconsistency: "
412 "physical_block_size=%u, logical_block_size=%u, "
413 "alignment_offset=%u, start=%llu",
414 dm_device_name(ti->table->md), bdev,
415 q->limits.physical_block_size,
416 q->limits.logical_block_size,
417 q->limits.alignment_offset,
418 (unsigned long long) start << SECTOR_SHIFT);
423 * Decrement a device's use count and remove it if necessary.
425 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
428 struct list_head *devices = &ti->table->devices;
429 struct dm_dev_internal *dd;
431 list_for_each_entry(dd, devices, list) {
432 if (dd->dm_dev == d) {
438 DMWARN("%s: device %s not in table devices list",
439 dm_device_name(ti->table->md), d->name);
442 if (refcount_dec_and_test(&dd->count)) {
443 dm_put_table_device(ti->table->md, d);
448 EXPORT_SYMBOL(dm_put_device);
451 * Checks to see if the target joins onto the end of the table.
453 static int adjoin(struct dm_table *table, struct dm_target *ti)
455 struct dm_target *prev;
457 if (!table->num_targets)
460 prev = &table->targets[table->num_targets - 1];
461 return (ti->begin == (prev->begin + prev->len));
465 * Used to dynamically allocate the arg array.
467 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
468 * process messages even if some device is suspended. These messages have a
469 * small fixed number of arguments.
471 * On the other hand, dm-switch needs to process bulk data using messages and
472 * excessive use of GFP_NOIO could cause trouble.
474 static char **realloc_argv(unsigned *size, char **old_argv)
481 new_size = *size * 2;
487 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
488 if (argv && old_argv) {
489 memcpy(argv, old_argv, *size * sizeof(*argv));
498 * Destructively splits up the argument list to pass to ctr.
500 int dm_split_args(int *argc, char ***argvp, char *input)
502 char *start, *end = input, *out, **argv = NULL;
503 unsigned array_size = 0;
512 argv = realloc_argv(&array_size, argv);
517 /* Skip whitespace */
518 start = skip_spaces(end);
521 break; /* success, we hit the end */
523 /* 'out' is used to remove any back-quotes */
526 /* Everything apart from '\0' can be quoted */
527 if (*end == '\\' && *(end + 1)) {
534 break; /* end of token */
539 /* have we already filled the array ? */
540 if ((*argc + 1) > array_size) {
541 argv = realloc_argv(&array_size, argv);
546 /* we know this is whitespace */
550 /* terminate the string and put it in the array */
561 * Impose necessary and sufficient conditions on a devices's table such
562 * that any incoming bio which respects its logical_block_size can be
563 * processed successfully. If it falls across the boundary between
564 * two or more targets, the size of each piece it gets split into must
565 * be compatible with the logical_block_size of the target processing it.
567 static int validate_hardware_logical_block_alignment(struct dm_table *table,
568 struct queue_limits *limits)
571 * This function uses arithmetic modulo the logical_block_size
572 * (in units of 512-byte sectors).
574 unsigned short device_logical_block_size_sects =
575 limits->logical_block_size >> SECTOR_SHIFT;
578 * Offset of the start of the next table entry, mod logical_block_size.
580 unsigned short next_target_start = 0;
583 * Given an aligned bio that extends beyond the end of a
584 * target, how many sectors must the next target handle?
586 unsigned short remaining = 0;
588 struct dm_target *ti;
589 struct queue_limits ti_limits;
593 * Check each entry in the table in turn.
595 for (i = 0; i < dm_table_get_num_targets(table); i++) {
596 ti = dm_table_get_target(table, i);
598 blk_set_stacking_limits(&ti_limits);
600 /* combine all target devices' limits */
601 if (ti->type->iterate_devices)
602 ti->type->iterate_devices(ti, dm_set_device_limits,
606 * If the remaining sectors fall entirely within this
607 * table entry are they compatible with its logical_block_size?
609 if (remaining < ti->len &&
610 remaining & ((ti_limits.logical_block_size >>
615 (unsigned short) ((next_target_start + ti->len) &
616 (device_logical_block_size_sects - 1));
617 remaining = next_target_start ?
618 device_logical_block_size_sects - next_target_start : 0;
622 DMWARN("%s: table line %u (start sect %llu len %llu) "
623 "not aligned to h/w logical block size %u",
624 dm_device_name(table->md), i,
625 (unsigned long long) ti->begin,
626 (unsigned long long) ti->len,
627 limits->logical_block_size);
634 int dm_table_add_target(struct dm_table *t, const char *type,
635 sector_t start, sector_t len, char *params)
637 int r = -EINVAL, argc;
639 struct dm_target *tgt;
642 DMERR("%s: target type %s must appear alone in table",
643 dm_device_name(t->md), t->targets->type->name);
647 BUG_ON(t->num_targets >= t->num_allocated);
649 tgt = t->targets + t->num_targets;
650 memset(tgt, 0, sizeof(*tgt));
653 DMERR("%s: zero-length target", dm_device_name(t->md));
657 tgt->type = dm_get_target_type(type);
659 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
663 if (dm_target_needs_singleton(tgt->type)) {
664 if (t->num_targets) {
665 tgt->error = "singleton target type must appear alone in table";
671 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
672 tgt->error = "target type may not be included in a read-only table";
676 if (t->immutable_target_type) {
677 if (t->immutable_target_type != tgt->type) {
678 tgt->error = "immutable target type cannot be mixed with other target types";
681 } else if (dm_target_is_immutable(tgt->type)) {
682 if (t->num_targets) {
683 tgt->error = "immutable target type cannot be mixed with other target types";
686 t->immutable_target_type = tgt->type;
689 if (dm_target_has_integrity(tgt->type))
690 t->integrity_added = 1;
695 tgt->error = "Unknown error";
698 * Does this target adjoin the previous one ?
700 if (!adjoin(t, tgt)) {
701 tgt->error = "Gap in table";
705 r = dm_split_args(&argc, &argv, params);
707 tgt->error = "couldn't split parameters";
711 r = tgt->type->ctr(tgt, argc, argv);
716 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
718 if (!tgt->num_discard_bios && tgt->discards_supported)
719 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
720 dm_device_name(t->md), type);
722 if (tgt->limit_swap_bios && !static_key_enabled(&swap_bios_enabled.key))
723 static_branch_enable(&swap_bios_enabled);
728 DMERR("%s: %s: %s (%pe)", dm_device_name(t->md), type, tgt->error, ERR_PTR(r));
729 dm_put_target_type(tgt->type);
734 * Target argument parsing helpers.
736 static int validate_next_arg(const struct dm_arg *arg,
737 struct dm_arg_set *arg_set,
738 unsigned *value, char **error, unsigned grouped)
740 const char *arg_str = dm_shift_arg(arg_set);
744 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
745 (*value < arg->min) ||
746 (*value > arg->max) ||
747 (grouped && arg_set->argc < *value)) {
755 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
756 unsigned *value, char **error)
758 return validate_next_arg(arg, arg_set, value, error, 0);
760 EXPORT_SYMBOL(dm_read_arg);
762 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
763 unsigned *value, char **error)
765 return validate_next_arg(arg, arg_set, value, error, 1);
767 EXPORT_SYMBOL(dm_read_arg_group);
769 const char *dm_shift_arg(struct dm_arg_set *as)
782 EXPORT_SYMBOL(dm_shift_arg);
784 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
786 BUG_ON(as->argc < num_args);
787 as->argc -= num_args;
788 as->argv += num_args;
790 EXPORT_SYMBOL(dm_consume_args);
792 static bool __table_type_bio_based(enum dm_queue_mode table_type)
794 return (table_type == DM_TYPE_BIO_BASED ||
795 table_type == DM_TYPE_DAX_BIO_BASED);
798 static bool __table_type_request_based(enum dm_queue_mode table_type)
800 return table_type == DM_TYPE_REQUEST_BASED;
803 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
807 EXPORT_SYMBOL_GPL(dm_table_set_type);
809 /* validate the dax capability of the target device span */
810 static int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
811 sector_t start, sector_t len, void *data)
816 DMDEBUG("%pg: error: dax unsupported by block device", dev->bdev);
820 /* Check devices support synchronous DAX */
821 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
822 sector_t start, sector_t len, void *data)
824 return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
827 static bool dm_table_supports_dax(struct dm_table *t,
828 iterate_devices_callout_fn iterate_fn)
830 struct dm_target *ti;
833 /* Ensure that all targets support DAX. */
834 for (i = 0; i < dm_table_get_num_targets(t); i++) {
835 ti = dm_table_get_target(t, i);
837 if (!ti->type->direct_access)
840 if (!ti->type->iterate_devices ||
841 ti->type->iterate_devices(ti, iterate_fn, NULL))
848 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
849 sector_t start, sector_t len, void *data)
851 struct block_device *bdev = dev->bdev;
852 struct request_queue *q = bdev_get_queue(bdev);
854 /* request-based cannot stack on partitions! */
855 if (bdev_is_partition(bdev))
858 return queue_is_mq(q);
861 static int dm_table_determine_type(struct dm_table *t)
864 unsigned bio_based = 0, request_based = 0, hybrid = 0;
865 struct dm_target *tgt;
866 struct list_head *devices = dm_table_get_devices(t);
867 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
869 if (t->type != DM_TYPE_NONE) {
870 /* target already set the table's type */
871 if (t->type == DM_TYPE_BIO_BASED) {
872 /* possibly upgrade to a variant of bio-based */
873 goto verify_bio_based;
875 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
876 goto verify_rq_based;
879 for (i = 0; i < t->num_targets; i++) {
880 tgt = t->targets + i;
881 if (dm_target_hybrid(tgt))
883 else if (dm_target_request_based(tgt))
888 if (bio_based && request_based) {
889 DMERR("Inconsistent table: different target types"
890 " can't be mixed up");
895 if (hybrid && !bio_based && !request_based) {
897 * The targets can work either way.
898 * Determine the type from the live device.
899 * Default to bio-based if device is new.
901 if (__table_type_request_based(live_md_type))
909 /* We must use this table as bio-based */
910 t->type = DM_TYPE_BIO_BASED;
911 if (dm_table_supports_dax(t, device_not_dax_capable) ||
912 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
913 t->type = DM_TYPE_DAX_BIO_BASED;
918 BUG_ON(!request_based); /* No targets in this table */
920 t->type = DM_TYPE_REQUEST_BASED;
924 * Request-based dm supports only tables that have a single target now.
925 * To support multiple targets, request splitting support is needed,
926 * and that needs lots of changes in the block-layer.
927 * (e.g. request completion process for partial completion.)
929 if (t->num_targets > 1) {
930 DMERR("request-based DM doesn't support multiple targets");
934 if (list_empty(devices)) {
936 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
938 /* inherit live table's type */
940 t->type = live_table->type;
941 dm_put_live_table(t->md, srcu_idx);
945 tgt = dm_table_get_immutable_target(t);
947 DMERR("table load rejected: immutable target is required");
949 } else if (tgt->max_io_len) {
950 DMERR("table load rejected: immutable target that splits IO is not supported");
954 /* Non-request-stackable devices can't be used for request-based dm */
955 if (!tgt->type->iterate_devices ||
956 !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
957 DMERR("table load rejected: including non-request-stackable devices");
964 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
969 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
971 return t->immutable_target_type;
974 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
976 /* Immutable target is implicitly a singleton */
977 if (t->num_targets > 1 ||
978 !dm_target_is_immutable(t->targets[0].type))
984 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
986 struct dm_target *ti;
989 for (i = 0; i < dm_table_get_num_targets(t); i++) {
990 ti = dm_table_get_target(t, i);
991 if (dm_target_is_wildcard(ti->type))
998 bool dm_table_bio_based(struct dm_table *t)
1000 return __table_type_bio_based(dm_table_get_type(t));
1003 bool dm_table_request_based(struct dm_table *t)
1005 return __table_type_request_based(dm_table_get_type(t));
1008 static bool dm_table_supports_poll(struct dm_table *t);
1010 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1012 enum dm_queue_mode type = dm_table_get_type(t);
1013 unsigned per_io_data_size = 0;
1014 unsigned min_pool_size = 0;
1015 struct dm_target *ti;
1017 bool poll_supported = false;
1019 if (unlikely(type == DM_TYPE_NONE)) {
1020 DMWARN("no table type is set, can't allocate mempools");
1024 if (__table_type_bio_based(type)) {
1025 for (i = 0; i < t->num_targets; i++) {
1026 ti = t->targets + i;
1027 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1028 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1030 poll_supported = dm_table_supports_poll(t);
1033 t->mempools = dm_alloc_md_mempools(md, type, per_io_data_size, min_pool_size,
1034 t->integrity_supported, poll_supported);
1041 void dm_table_free_md_mempools(struct dm_table *t)
1043 dm_free_md_mempools(t->mempools);
1047 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1052 static int setup_indexes(struct dm_table *t)
1055 unsigned int total = 0;
1058 /* allocate the space for *all* the indexes */
1059 for (i = t->depth - 2; i >= 0; i--) {
1060 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1061 total += t->counts[i];
1064 indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1068 /* set up internal nodes, bottom-up */
1069 for (i = t->depth - 2; i >= 0; i--) {
1070 t->index[i] = indexes;
1071 indexes += (KEYS_PER_NODE * t->counts[i]);
1072 setup_btree_index(i, t);
1079 * Builds the btree to index the map.
1081 static int dm_table_build_index(struct dm_table *t)
1084 unsigned int leaf_nodes;
1086 /* how many indexes will the btree have ? */
1087 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1088 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1090 /* leaf layer has already been set up */
1091 t->counts[t->depth - 1] = leaf_nodes;
1092 t->index[t->depth - 1] = t->highs;
1095 r = setup_indexes(t);
1100 static bool integrity_profile_exists(struct gendisk *disk)
1102 return !!blk_get_integrity(disk);
1106 * Get a disk whose integrity profile reflects the table's profile.
1107 * Returns NULL if integrity support was inconsistent or unavailable.
1109 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1111 struct list_head *devices = dm_table_get_devices(t);
1112 struct dm_dev_internal *dd = NULL;
1113 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1116 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1117 struct dm_target *ti = dm_table_get_target(t, i);
1118 if (!dm_target_passes_integrity(ti->type))
1122 list_for_each_entry(dd, devices, list) {
1123 template_disk = dd->dm_dev->bdev->bd_disk;
1124 if (!integrity_profile_exists(template_disk))
1126 else if (prev_disk &&
1127 blk_integrity_compare(prev_disk, template_disk) < 0)
1129 prev_disk = template_disk;
1132 return template_disk;
1136 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1137 dm_device_name(t->md),
1138 prev_disk->disk_name,
1139 template_disk->disk_name);
1144 * Register the mapped device for blk_integrity support if the
1145 * underlying devices have an integrity profile. But all devices may
1146 * not have matching profiles (checking all devices isn't reliable
1147 * during table load because this table may use other DM device(s) which
1148 * must be resumed before they will have an initialized integity
1149 * profile). Consequently, stacked DM devices force a 2 stage integrity
1150 * profile validation: First pass during table load, final pass during
1153 static int dm_table_register_integrity(struct dm_table *t)
1155 struct mapped_device *md = t->md;
1156 struct gendisk *template_disk = NULL;
1158 /* If target handles integrity itself do not register it here. */
1159 if (t->integrity_added)
1162 template_disk = dm_table_get_integrity_disk(t);
1166 if (!integrity_profile_exists(dm_disk(md))) {
1167 t->integrity_supported = true;
1169 * Register integrity profile during table load; we can do
1170 * this because the final profile must match during resume.
1172 blk_integrity_register(dm_disk(md),
1173 blk_get_integrity(template_disk));
1178 * If DM device already has an initialized integrity
1179 * profile the new profile should not conflict.
1181 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1182 DMWARN("%s: conflict with existing integrity profile: "
1183 "%s profile mismatch",
1184 dm_device_name(t->md),
1185 template_disk->disk_name);
1189 /* Preserve existing integrity profile */
1190 t->integrity_supported = true;
1194 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1196 struct dm_crypto_profile {
1197 struct blk_crypto_profile profile;
1198 struct mapped_device *md;
1201 struct dm_keyslot_evict_args {
1202 const struct blk_crypto_key *key;
1206 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1207 sector_t start, sector_t len, void *data)
1209 struct dm_keyslot_evict_args *args = data;
1212 err = blk_crypto_evict_key(bdev_get_queue(dev->bdev), args->key);
1215 /* Always try to evict the key from all devices. */
1220 * When an inline encryption key is evicted from a device-mapper device, evict
1221 * it from all the underlying devices.
1223 static int dm_keyslot_evict(struct blk_crypto_profile *profile,
1224 const struct blk_crypto_key *key, unsigned int slot)
1226 struct mapped_device *md =
1227 container_of(profile, struct dm_crypto_profile, profile)->md;
1228 struct dm_keyslot_evict_args args = { key };
1232 struct dm_target *ti;
1234 t = dm_get_live_table(md, &srcu_idx);
1237 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1238 ti = dm_table_get_target(t, i);
1239 if (!ti->type->iterate_devices)
1241 ti->type->iterate_devices(ti, dm_keyslot_evict_callback, &args);
1243 dm_put_live_table(md, srcu_idx);
1248 device_intersect_crypto_capabilities(struct dm_target *ti, struct dm_dev *dev,
1249 sector_t start, sector_t len, void *data)
1251 struct blk_crypto_profile *parent = data;
1252 struct blk_crypto_profile *child =
1253 bdev_get_queue(dev->bdev)->crypto_profile;
1255 blk_crypto_intersect_capabilities(parent, child);
1259 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1261 struct dm_crypto_profile *dmcp = container_of(profile,
1262 struct dm_crypto_profile,
1268 blk_crypto_profile_destroy(profile);
1272 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1274 dm_destroy_crypto_profile(t->crypto_profile);
1275 t->crypto_profile = NULL;
1279 * Constructs and initializes t->crypto_profile with a crypto profile that
1280 * represents the common set of crypto capabilities of the devices described by
1281 * the dm_table. However, if the constructed crypto profile doesn't support all
1282 * crypto capabilities that are supported by the current mapped_device, it
1283 * returns an error instead, since we don't support removing crypto capabilities
1284 * on table changes. Finally, if the constructed crypto profile is "empty" (has
1285 * no crypto capabilities at all), it just sets t->crypto_profile to NULL.
1287 static int dm_table_construct_crypto_profile(struct dm_table *t)
1289 struct dm_crypto_profile *dmcp;
1290 struct blk_crypto_profile *profile;
1291 struct dm_target *ti;
1293 bool empty_profile = true;
1295 dmcp = kmalloc(sizeof(*dmcp), GFP_KERNEL);
1300 profile = &dmcp->profile;
1301 blk_crypto_profile_init(profile, 0);
1302 profile->ll_ops.keyslot_evict = dm_keyslot_evict;
1303 profile->max_dun_bytes_supported = UINT_MAX;
1304 memset(profile->modes_supported, 0xFF,
1305 sizeof(profile->modes_supported));
1307 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1308 ti = dm_table_get_target(t, i);
1310 if (!dm_target_passes_crypto(ti->type)) {
1311 blk_crypto_intersect_capabilities(profile, NULL);
1314 if (!ti->type->iterate_devices)
1316 ti->type->iterate_devices(ti,
1317 device_intersect_crypto_capabilities,
1322 !blk_crypto_has_capabilities(profile,
1323 t->md->queue->crypto_profile)) {
1324 DMWARN("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1325 dm_destroy_crypto_profile(profile);
1330 * If the new profile doesn't actually support any crypto capabilities,
1331 * we may as well represent it with a NULL profile.
1333 for (i = 0; i < ARRAY_SIZE(profile->modes_supported); i++) {
1334 if (profile->modes_supported[i]) {
1335 empty_profile = false;
1340 if (empty_profile) {
1341 dm_destroy_crypto_profile(profile);
1346 * t->crypto_profile is only set temporarily while the table is being
1347 * set up, and it gets set to NULL after the profile has been
1348 * transferred to the request_queue.
1350 t->crypto_profile = profile;
1355 static void dm_update_crypto_profile(struct request_queue *q,
1358 if (!t->crypto_profile)
1361 /* Make the crypto profile less restrictive. */
1362 if (!q->crypto_profile) {
1363 blk_crypto_register(t->crypto_profile, q);
1365 blk_crypto_update_capabilities(q->crypto_profile,
1367 dm_destroy_crypto_profile(t->crypto_profile);
1369 t->crypto_profile = NULL;
1372 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1374 static int dm_table_construct_crypto_profile(struct dm_table *t)
1379 void dm_destroy_crypto_profile(struct blk_crypto_profile *profile)
1383 static void dm_table_destroy_crypto_profile(struct dm_table *t)
1387 static void dm_update_crypto_profile(struct request_queue *q,
1392 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1395 * Prepares the table for use by building the indices,
1396 * setting the type, and allocating mempools.
1398 int dm_table_complete(struct dm_table *t)
1402 r = dm_table_determine_type(t);
1404 DMERR("unable to determine table type");
1408 r = dm_table_build_index(t);
1410 DMERR("unable to build btrees");
1414 r = dm_table_register_integrity(t);
1416 DMERR("could not register integrity profile.");
1420 r = dm_table_construct_crypto_profile(t);
1422 DMERR("could not construct crypto profile.");
1426 r = dm_table_alloc_md_mempools(t, t->md);
1428 DMERR("unable to allocate mempools");
1433 static DEFINE_MUTEX(_event_lock);
1434 void dm_table_event_callback(struct dm_table *t,
1435 void (*fn)(void *), void *context)
1437 mutex_lock(&_event_lock);
1439 t->event_context = context;
1440 mutex_unlock(&_event_lock);
1443 void dm_table_event(struct dm_table *t)
1445 mutex_lock(&_event_lock);
1447 t->event_fn(t->event_context);
1448 mutex_unlock(&_event_lock);
1450 EXPORT_SYMBOL(dm_table_event);
1452 inline sector_t dm_table_get_size(struct dm_table *t)
1454 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1456 EXPORT_SYMBOL(dm_table_get_size);
1458 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1460 if (index >= t->num_targets)
1463 return t->targets + index;
1467 * Search the btree for the correct target.
1469 * Caller should check returned pointer for NULL
1470 * to trap I/O beyond end of device.
1472 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1474 unsigned int l, n = 0, k = 0;
1477 if (unlikely(sector >= dm_table_get_size(t)))
1480 for (l = 0; l < t->depth; l++) {
1481 n = get_child(n, k);
1482 node = get_node(t, l, n);
1484 for (k = 0; k < KEYS_PER_NODE; k++)
1485 if (node[k] >= sector)
1489 return &t->targets[(KEYS_PER_NODE * n) + k];
1492 static int device_not_poll_capable(struct dm_target *ti, struct dm_dev *dev,
1493 sector_t start, sector_t len, void *data)
1495 struct request_queue *q = bdev_get_queue(dev->bdev);
1497 return !test_bit(QUEUE_FLAG_POLL, &q->queue_flags);
1501 * type->iterate_devices() should be called when the sanity check needs to
1502 * iterate and check all underlying data devices. iterate_devices() will
1503 * iterate all underlying data devices until it encounters a non-zero return
1504 * code, returned by whether the input iterate_devices_callout_fn, or
1505 * iterate_devices() itself internally.
1507 * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1508 * iterate multiple underlying devices internally, in which case a non-zero
1509 * return code returned by iterate_devices_callout_fn will stop the iteration
1512 * Cases requiring _any_ underlying device supporting some kind of attribute,
1513 * should use the iteration structure like dm_table_any_dev_attr(), or call
1514 * it directly. @func should handle semantics of positive examples, e.g.
1515 * capable of something.
1517 * Cases requiring _all_ underlying devices supporting some kind of attribute,
1518 * should use the iteration structure like dm_table_supports_nowait() or
1519 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1520 * uses an @anti_func that handle semantics of counter examples, e.g. not
1521 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1523 static bool dm_table_any_dev_attr(struct dm_table *t,
1524 iterate_devices_callout_fn func, void *data)
1526 struct dm_target *ti;
1529 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1530 ti = dm_table_get_target(t, i);
1532 if (ti->type->iterate_devices &&
1533 ti->type->iterate_devices(ti, func, data))
1540 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1541 sector_t start, sector_t len, void *data)
1543 unsigned *num_devices = data;
1550 static bool dm_table_supports_poll(struct dm_table *t)
1552 struct dm_target *ti;
1555 while (i < dm_table_get_num_targets(t)) {
1556 ti = dm_table_get_target(t, i++);
1558 if (!ti->type->iterate_devices ||
1559 ti->type->iterate_devices(ti, device_not_poll_capable, NULL))
1567 * Check whether a table has no data devices attached using each
1568 * target's iterate_devices method.
1569 * Returns false if the result is unknown because a target doesn't
1570 * support iterate_devices.
1572 bool dm_table_has_no_data_devices(struct dm_table *table)
1574 struct dm_target *ti;
1575 unsigned i, num_devices;
1577 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1578 ti = dm_table_get_target(table, i);
1580 if (!ti->type->iterate_devices)
1584 ti->type->iterate_devices(ti, count_device, &num_devices);
1592 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1593 sector_t start, sector_t len, void *data)
1595 struct request_queue *q = bdev_get_queue(dev->bdev);
1596 enum blk_zoned_model *zoned_model = data;
1598 return blk_queue_zoned_model(q) != *zoned_model;
1602 * Check the device zoned model based on the target feature flag. If the target
1603 * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1604 * also accepted but all devices must have the same zoned model. If the target
1605 * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1606 * zoned model with all zoned devices having the same zone size.
1608 static bool dm_table_supports_zoned_model(struct dm_table *t,
1609 enum blk_zoned_model zoned_model)
1611 struct dm_target *ti;
1614 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1615 ti = dm_table_get_target(t, i);
1617 if (dm_target_supports_zoned_hm(ti->type)) {
1618 if (!ti->type->iterate_devices ||
1619 ti->type->iterate_devices(ti, device_not_zoned_model,
1622 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1623 if (zoned_model == BLK_ZONED_HM)
1631 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1632 sector_t start, sector_t len, void *data)
1634 struct request_queue *q = bdev_get_queue(dev->bdev);
1635 unsigned int *zone_sectors = data;
1637 if (!blk_queue_is_zoned(q))
1640 return blk_queue_zone_sectors(q) != *zone_sectors;
1644 * Check consistency of zoned model and zone sectors across all targets. For
1645 * zone sectors, if the destination device is a zoned block device, it shall
1646 * have the specified zone_sectors.
1648 static int validate_hardware_zoned_model(struct dm_table *table,
1649 enum blk_zoned_model zoned_model,
1650 unsigned int zone_sectors)
1652 if (zoned_model == BLK_ZONED_NONE)
1655 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1656 DMERR("%s: zoned model is not consistent across all devices",
1657 dm_device_name(table->md));
1661 /* Check zone size validity and compatibility */
1662 if (!zone_sectors || !is_power_of_2(zone_sectors))
1665 if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1666 DMERR("%s: zone sectors is not consistent across all zoned devices",
1667 dm_device_name(table->md));
1675 * Establish the new table's queue_limits and validate them.
1677 int dm_calculate_queue_limits(struct dm_table *table,
1678 struct queue_limits *limits)
1680 struct dm_target *ti;
1681 struct queue_limits ti_limits;
1683 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1684 unsigned int zone_sectors = 0;
1686 blk_set_stacking_limits(limits);
1688 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1689 blk_set_stacking_limits(&ti_limits);
1691 ti = dm_table_get_target(table, i);
1693 if (!ti->type->iterate_devices)
1694 goto combine_limits;
1697 * Combine queue limits of all the devices this target uses.
1699 ti->type->iterate_devices(ti, dm_set_device_limits,
1702 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1704 * After stacking all limits, validate all devices
1705 * in table support this zoned model and zone sectors.
1707 zoned_model = ti_limits.zoned;
1708 zone_sectors = ti_limits.chunk_sectors;
1711 /* Set I/O hints portion of queue limits */
1712 if (ti->type->io_hints)
1713 ti->type->io_hints(ti, &ti_limits);
1716 * Check each device area is consistent with the target's
1717 * overall queue limits.
1719 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1725 * Merge this target's queue limits into the overall limits
1728 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1729 DMWARN("%s: adding target device "
1730 "(start sect %llu len %llu) "
1731 "caused an alignment inconsistency",
1732 dm_device_name(table->md),
1733 (unsigned long long) ti->begin,
1734 (unsigned long long) ti->len);
1738 * Verify that the zoned model and zone sectors, as determined before
1739 * any .io_hints override, are the same across all devices in the table.
1740 * - this is especially relevant if .io_hints is emulating a disk-managed
1741 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1744 if (limits->zoned != BLK_ZONED_NONE) {
1746 * ...IF the above limits stacking determined a zoned model
1747 * validate that all of the table's devices conform to it.
1749 zoned_model = limits->zoned;
1750 zone_sectors = limits->chunk_sectors;
1752 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1755 return validate_hardware_logical_block_alignment(table, limits);
1759 * Verify that all devices have an integrity profile that matches the
1760 * DM device's registered integrity profile. If the profiles don't
1761 * match then unregister the DM device's integrity profile.
1763 static void dm_table_verify_integrity(struct dm_table *t)
1765 struct gendisk *template_disk = NULL;
1767 if (t->integrity_added)
1770 if (t->integrity_supported) {
1772 * Verify that the original integrity profile
1773 * matches all the devices in this table.
1775 template_disk = dm_table_get_integrity_disk(t);
1776 if (template_disk &&
1777 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1781 if (integrity_profile_exists(dm_disk(t->md))) {
1782 DMWARN("%s: unable to establish an integrity profile",
1783 dm_device_name(t->md));
1784 blk_integrity_unregister(dm_disk(t->md));
1788 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1789 sector_t start, sector_t len, void *data)
1791 unsigned long flush = (unsigned long) data;
1792 struct request_queue *q = bdev_get_queue(dev->bdev);
1794 return (q->queue_flags & flush);
1797 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1799 struct dm_target *ti;
1803 * Require at least one underlying device to support flushes.
1804 * t->devices includes internal dm devices such as mirror logs
1805 * so we need to use iterate_devices here, which targets
1806 * supporting flushes must provide.
1808 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1809 ti = dm_table_get_target(t, i);
1811 if (!ti->num_flush_bios)
1814 if (ti->flush_supported)
1817 if (ti->type->iterate_devices &&
1818 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1825 static int device_dax_write_cache_enabled(struct dm_target *ti,
1826 struct dm_dev *dev, sector_t start,
1827 sector_t len, void *data)
1829 struct dax_device *dax_dev = dev->dax_dev;
1834 if (dax_write_cache_enabled(dax_dev))
1839 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1840 sector_t start, sector_t len, void *data)
1842 return !bdev_nonrot(dev->bdev);
1845 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1846 sector_t start, sector_t len, void *data)
1848 struct request_queue *q = bdev_get_queue(dev->bdev);
1850 return !blk_queue_add_random(q);
1853 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1854 sector_t start, sector_t len, void *data)
1856 struct request_queue *q = bdev_get_queue(dev->bdev);
1858 return !q->limits.max_write_zeroes_sectors;
1861 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1863 struct dm_target *ti;
1866 while (i < dm_table_get_num_targets(t)) {
1867 ti = dm_table_get_target(t, i++);
1869 if (!ti->num_write_zeroes_bios)
1872 if (!ti->type->iterate_devices ||
1873 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1880 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1881 sector_t start, sector_t len, void *data)
1883 struct request_queue *q = bdev_get_queue(dev->bdev);
1885 return !blk_queue_nowait(q);
1888 static bool dm_table_supports_nowait(struct dm_table *t)
1890 struct dm_target *ti;
1893 while (i < dm_table_get_num_targets(t)) {
1894 ti = dm_table_get_target(t, i++);
1896 if (!dm_target_supports_nowait(ti->type))
1899 if (!ti->type->iterate_devices ||
1900 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1907 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1908 sector_t start, sector_t len, void *data)
1910 return !bdev_max_discard_sectors(dev->bdev);
1913 static bool dm_table_supports_discards(struct dm_table *t)
1915 struct dm_target *ti;
1918 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1919 ti = dm_table_get_target(t, i);
1921 if (!ti->num_discard_bios)
1925 * Either the target provides discard support (as implied by setting
1926 * 'discards_supported') or it relies on _all_ data devices having
1929 if (!ti->discards_supported &&
1930 (!ti->type->iterate_devices ||
1931 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1938 static int device_not_secure_erase_capable(struct dm_target *ti,
1939 struct dm_dev *dev, sector_t start,
1940 sector_t len, void *data)
1942 return !bdev_max_secure_erase_sectors(dev->bdev);
1945 static bool dm_table_supports_secure_erase(struct dm_table *t)
1947 struct dm_target *ti;
1950 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1951 ti = dm_table_get_target(t, i);
1953 if (!ti->num_secure_erase_bios)
1956 if (!ti->type->iterate_devices ||
1957 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1964 static int device_requires_stable_pages(struct dm_target *ti,
1965 struct dm_dev *dev, sector_t start,
1966 sector_t len, void *data)
1968 return bdev_stable_writes(dev->bdev);
1971 int dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1972 struct queue_limits *limits)
1974 bool wc = false, fua = false;
1978 * Copy table's limits to the DM device's request_queue
1980 q->limits = *limits;
1982 if (dm_table_supports_nowait(t))
1983 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1985 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
1987 if (!dm_table_supports_discards(t)) {
1988 q->limits.max_discard_sectors = 0;
1989 q->limits.max_hw_discard_sectors = 0;
1990 q->limits.discard_granularity = 0;
1991 q->limits.discard_alignment = 0;
1992 q->limits.discard_misaligned = 0;
1995 if (!dm_table_supports_secure_erase(t))
1996 q->limits.max_secure_erase_sectors = 0;
1998 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
2000 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
2003 blk_queue_write_cache(q, wc, fua);
2005 if (dm_table_supports_dax(t, device_not_dax_capable)) {
2006 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
2007 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable))
2008 set_dax_synchronous(t->md->dax_dev);
2011 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
2013 if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
2014 dax_write_cache(t->md->dax_dev, true);
2016 /* Ensure that all underlying devices are non-rotational. */
2017 if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
2018 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2020 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2022 if (!dm_table_supports_write_zeroes(t))
2023 q->limits.max_write_zeroes_sectors = 0;
2025 dm_table_verify_integrity(t);
2028 * Some devices don't use blk_integrity but still want stable pages
2029 * because they do their own checksumming.
2030 * If any underlying device requires stable pages, a table must require
2031 * them as well. Only targets that support iterate_devices are considered:
2032 * don't want error, zero, etc to require stable pages.
2034 if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
2035 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2037 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2040 * Determine whether or not this queue's I/O timings contribute
2041 * to the entropy pool, Only request-based targets use this.
2042 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2045 if (blk_queue_add_random(q) &&
2046 dm_table_any_dev_attr(t, device_is_not_random, NULL))
2047 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2050 * For a zoned target, setup the zones related queue attributes
2051 * and resources necessary for zone append emulation if necessary.
2053 if (blk_queue_is_zoned(q)) {
2054 r = dm_set_zones_restrictions(t, q);
2057 if (!static_key_enabled(&zoned_enabled.key))
2058 static_branch_enable(&zoned_enabled);
2061 dm_update_crypto_profile(q, t);
2062 disk_update_readahead(t->md->disk);
2065 * Check for request-based device is left to
2066 * dm_mq_init_request_queue()->blk_mq_init_allocated_queue().
2068 * For bio-based device, only set QUEUE_FLAG_POLL when all
2069 * underlying devices supporting polling.
2071 if (__table_type_bio_based(t->type)) {
2072 if (dm_table_supports_poll(t))
2073 blk_queue_flag_set(QUEUE_FLAG_POLL, q);
2075 blk_queue_flag_clear(QUEUE_FLAG_POLL, q);
2081 unsigned int dm_table_get_num_targets(struct dm_table *t)
2083 return t->num_targets;
2086 struct list_head *dm_table_get_devices(struct dm_table *t)
2091 fmode_t dm_table_get_mode(struct dm_table *t)
2095 EXPORT_SYMBOL(dm_table_get_mode);
2103 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2105 int i = t->num_targets;
2106 struct dm_target *ti = t->targets;
2108 lockdep_assert_held(&t->md->suspend_lock);
2113 if (ti->type->presuspend)
2114 ti->type->presuspend(ti);
2116 case PRESUSPEND_UNDO:
2117 if (ti->type->presuspend_undo)
2118 ti->type->presuspend_undo(ti);
2121 if (ti->type->postsuspend)
2122 ti->type->postsuspend(ti);
2129 void dm_table_presuspend_targets(struct dm_table *t)
2134 suspend_targets(t, PRESUSPEND);
2137 void dm_table_presuspend_undo_targets(struct dm_table *t)
2142 suspend_targets(t, PRESUSPEND_UNDO);
2145 void dm_table_postsuspend_targets(struct dm_table *t)
2150 suspend_targets(t, POSTSUSPEND);
2153 int dm_table_resume_targets(struct dm_table *t)
2157 lockdep_assert_held(&t->md->suspend_lock);
2159 for (i = 0; i < t->num_targets; i++) {
2160 struct dm_target *ti = t->targets + i;
2162 if (!ti->type->preresume)
2165 r = ti->type->preresume(ti);
2167 DMERR("%s: %s: preresume failed, error = %d",
2168 dm_device_name(t->md), ti->type->name, r);
2173 for (i = 0; i < t->num_targets; i++) {
2174 struct dm_target *ti = t->targets + i;
2176 if (ti->type->resume)
2177 ti->type->resume(ti);
2183 struct mapped_device *dm_table_get_md(struct dm_table *t)
2187 EXPORT_SYMBOL(dm_table_get_md);
2189 const char *dm_table_device_name(struct dm_table *t)
2191 return dm_device_name(t->md);
2193 EXPORT_SYMBOL_GPL(dm_table_device_name);
2195 void dm_table_run_md_queue_async(struct dm_table *t)
2197 if (!dm_table_request_based(t))
2201 blk_mq_run_hw_queues(t->md->queue, true);
2203 EXPORT_SYMBOL(dm_table_run_md_queue_async);