2 * Copyright (C) 2001 Sistina Software (UK) Limited.
3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
5 * This file is released under the GPL.
10 #include <linux/module.h>
11 #include <linux/vmalloc.h>
12 #include <linux/blkdev.h>
13 #include <linux/namei.h>
14 #include <linux/ctype.h>
15 #include <linux/string.h>
16 #include <linux/slab.h>
17 #include <linux/interrupt.h>
18 #include <linux/mutex.h>
19 #include <linux/delay.h>
20 #include <linux/atomic.h>
21 #include <linux/blk-mq.h>
22 #include <linux/mount.h>
23 #include <linux/dax.h>
25 #define DM_MSG_PREFIX "table"
27 #define NODE_SIZE L1_CACHE_BYTES
28 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
29 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
32 * Similar to ceiling(log_size(n))
34 static unsigned int int_log(unsigned int n, unsigned int base)
39 n = dm_div_up(n, base);
47 * Calculate the index of the child node of the n'th node k'th key.
49 static inline unsigned int get_child(unsigned int n, unsigned int k)
51 return (n * CHILDREN_PER_NODE) + k;
55 * Return the n'th node of level l from table t.
57 static inline sector_t *get_node(struct dm_table *t,
58 unsigned int l, unsigned int n)
60 return t->index[l] + (n * KEYS_PER_NODE);
64 * Return the highest key that you could lookup from the n'th
65 * node on level l of the btree.
67 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
69 for (; l < t->depth - 1; l++)
70 n = get_child(n, CHILDREN_PER_NODE - 1);
72 if (n >= t->counts[l])
73 return (sector_t) - 1;
75 return get_node(t, l, n)[KEYS_PER_NODE - 1];
79 * Fills in a level of the btree based on the highs of the level
82 static int setup_btree_index(unsigned int l, struct dm_table *t)
87 for (n = 0U; n < t->counts[l]; n++) {
88 node = get_node(t, l, n);
90 for (k = 0U; k < KEYS_PER_NODE; k++)
91 node[k] = high(t, l + 1, get_child(n, k));
97 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
103 * Check that we're not going to overflow.
105 if (nmemb > (ULONG_MAX / elem_size))
108 size = nmemb * elem_size;
109 addr = vzalloc(size);
113 EXPORT_SYMBOL(dm_vcalloc);
116 * highs, and targets are managed as dynamic arrays during a
119 static int alloc_targets(struct dm_table *t, unsigned int num)
122 struct dm_target *n_targets;
125 * Allocate both the target array and offset array at once.
127 n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
132 n_targets = (struct dm_target *) (n_highs + num);
134 memset(n_highs, -1, sizeof(*n_highs) * num);
137 t->num_allocated = num;
139 t->targets = n_targets;
144 int dm_table_create(struct dm_table **result, fmode_t mode,
145 unsigned num_targets, struct mapped_device *md)
147 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
152 INIT_LIST_HEAD(&t->devices);
155 num_targets = KEYS_PER_NODE;
157 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
164 if (alloc_targets(t, num_targets)) {
169 t->type = DM_TYPE_NONE;
176 static void free_devices(struct list_head *devices, struct mapped_device *md)
178 struct list_head *tmp, *next;
180 list_for_each_safe(tmp, next, devices) {
181 struct dm_dev_internal *dd =
182 list_entry(tmp, struct dm_dev_internal, list);
183 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
184 dm_device_name(md), dd->dm_dev->name);
185 dm_put_table_device(md, dd->dm_dev);
190 static void dm_table_destroy_keyslot_manager(struct dm_table *t);
192 void dm_table_destroy(struct dm_table *t)
199 /* free the indexes */
201 vfree(t->index[t->depth - 2]);
203 /* free the targets */
204 for (i = 0; i < t->num_targets; i++) {
205 struct dm_target *tgt = t->targets + i;
210 dm_put_target_type(tgt->type);
215 /* free the device list */
216 free_devices(&t->devices, t->md);
218 dm_free_md_mempools(t->mempools);
220 dm_table_destroy_keyslot_manager(t);
226 * See if we've already got a device in the list.
228 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
230 struct dm_dev_internal *dd;
232 list_for_each_entry (dd, l, list)
233 if (dd->dm_dev->bdev->bd_dev == dev)
240 * If possible, this checks an area of a destination device is invalid.
242 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
243 sector_t start, sector_t len, void *data)
245 struct queue_limits *limits = data;
246 struct block_device *bdev = dev->bdev;
248 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
249 unsigned short logical_block_size_sectors =
250 limits->logical_block_size >> SECTOR_SHIFT;
251 char b[BDEVNAME_SIZE];
256 if ((start >= dev_size) || (start + len > dev_size)) {
257 DMWARN("%s: %s too small for target: "
258 "start=%llu, len=%llu, dev_size=%llu",
259 dm_device_name(ti->table->md), bdevname(bdev, b),
260 (unsigned long long)start,
261 (unsigned long long)len,
262 (unsigned long long)dev_size);
267 * If the target is mapped to zoned block device(s), check
268 * that the zones are not partially mapped.
270 if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
271 unsigned int zone_sectors = bdev_zone_sectors(bdev);
273 if (start & (zone_sectors - 1)) {
274 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
275 dm_device_name(ti->table->md),
276 (unsigned long long)start,
277 zone_sectors, bdevname(bdev, b));
282 * Note: The last zone of a zoned block device may be smaller
283 * than other zones. So for a target mapping the end of a
284 * zoned block device with such a zone, len would not be zone
285 * aligned. We do not allow such last smaller zone to be part
286 * of the mapping here to ensure that mappings with multiple
287 * devices do not end up with a smaller zone in the middle of
290 if (len & (zone_sectors - 1)) {
291 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
292 dm_device_name(ti->table->md),
293 (unsigned long long)len,
294 zone_sectors, bdevname(bdev, b));
299 if (logical_block_size_sectors <= 1)
302 if (start & (logical_block_size_sectors - 1)) {
303 DMWARN("%s: start=%llu not aligned to h/w "
304 "logical block size %u of %s",
305 dm_device_name(ti->table->md),
306 (unsigned long long)start,
307 limits->logical_block_size, bdevname(bdev, b));
311 if (len & (logical_block_size_sectors - 1)) {
312 DMWARN("%s: len=%llu not aligned to h/w "
313 "logical block size %u of %s",
314 dm_device_name(ti->table->md),
315 (unsigned long long)len,
316 limits->logical_block_size, bdevname(bdev, b));
324 * This upgrades the mode on an already open dm_dev, being
325 * careful to leave things as they were if we fail to reopen the
326 * device and not to touch the existing bdev field in case
327 * it is accessed concurrently.
329 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
330 struct mapped_device *md)
333 struct dm_dev *old_dev, *new_dev;
335 old_dev = dd->dm_dev;
337 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
338 dd->dm_dev->mode | new_mode, &new_dev);
342 dd->dm_dev = new_dev;
343 dm_put_table_device(md, old_dev);
349 * Convert the path to a device
351 dev_t dm_get_dev_t(const char *path)
355 if (lookup_bdev(path, &dev))
356 dev = name_to_dev_t(path);
359 EXPORT_SYMBOL_GPL(dm_get_dev_t);
362 * Add a device to the list, or just increment the usage count if
363 * it's already present.
365 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
366 struct dm_dev **result)
370 unsigned int major, minor;
372 struct dm_dev_internal *dd;
373 struct dm_table *t = ti->table;
377 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
378 /* Extract the major/minor numbers */
379 dev = MKDEV(major, minor);
380 if (MAJOR(dev) != major || MINOR(dev) != minor)
383 dev = dm_get_dev_t(path);
388 dd = find_device(&t->devices, dev);
390 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
394 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
399 refcount_set(&dd->count, 1);
400 list_add(&dd->list, &t->devices);
403 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
404 r = upgrade_mode(dd, mode, t->md);
408 refcount_inc(&dd->count);
410 *result = dd->dm_dev;
413 EXPORT_SYMBOL(dm_get_device);
415 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
416 sector_t start, sector_t len, void *data)
418 struct queue_limits *limits = data;
419 struct block_device *bdev = dev->bdev;
420 struct request_queue *q = bdev_get_queue(bdev);
421 char b[BDEVNAME_SIZE];
424 DMWARN("%s: Cannot set limits for nonexistent device %s",
425 dm_device_name(ti->table->md), bdevname(bdev, b));
429 if (blk_stack_limits(limits, &q->limits,
430 get_start_sect(bdev) + start) < 0)
431 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
432 "physical_block_size=%u, logical_block_size=%u, "
433 "alignment_offset=%u, start=%llu",
434 dm_device_name(ti->table->md), bdevname(bdev, b),
435 q->limits.physical_block_size,
436 q->limits.logical_block_size,
437 q->limits.alignment_offset,
438 (unsigned long long) start << SECTOR_SHIFT);
443 * Decrement a device's use count and remove it if necessary.
445 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
448 struct list_head *devices = &ti->table->devices;
449 struct dm_dev_internal *dd;
451 list_for_each_entry(dd, devices, list) {
452 if (dd->dm_dev == d) {
458 DMWARN("%s: device %s not in table devices list",
459 dm_device_name(ti->table->md), d->name);
462 if (refcount_dec_and_test(&dd->count)) {
463 dm_put_table_device(ti->table->md, d);
468 EXPORT_SYMBOL(dm_put_device);
471 * Checks to see if the target joins onto the end of the table.
473 static int adjoin(struct dm_table *table, struct dm_target *ti)
475 struct dm_target *prev;
477 if (!table->num_targets)
480 prev = &table->targets[table->num_targets - 1];
481 return (ti->begin == (prev->begin + prev->len));
485 * Used to dynamically allocate the arg array.
487 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
488 * process messages even if some device is suspended. These messages have a
489 * small fixed number of arguments.
491 * On the other hand, dm-switch needs to process bulk data using messages and
492 * excessive use of GFP_NOIO could cause trouble.
494 static char **realloc_argv(unsigned *size, char **old_argv)
501 new_size = *size * 2;
507 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
508 if (argv && old_argv) {
509 memcpy(argv, old_argv, *size * sizeof(*argv));
518 * Destructively splits up the argument list to pass to ctr.
520 int dm_split_args(int *argc, char ***argvp, char *input)
522 char *start, *end = input, *out, **argv = NULL;
523 unsigned array_size = 0;
532 argv = realloc_argv(&array_size, argv);
537 /* Skip whitespace */
538 start = skip_spaces(end);
541 break; /* success, we hit the end */
543 /* 'out' is used to remove any back-quotes */
546 /* Everything apart from '\0' can be quoted */
547 if (*end == '\\' && *(end + 1)) {
554 break; /* end of token */
559 /* have we already filled the array ? */
560 if ((*argc + 1) > array_size) {
561 argv = realloc_argv(&array_size, argv);
566 /* we know this is whitespace */
570 /* terminate the string and put it in the array */
581 * Impose necessary and sufficient conditions on a devices's table such
582 * that any incoming bio which respects its logical_block_size can be
583 * processed successfully. If it falls across the boundary between
584 * two or more targets, the size of each piece it gets split into must
585 * be compatible with the logical_block_size of the target processing it.
587 static int validate_hardware_logical_block_alignment(struct dm_table *table,
588 struct queue_limits *limits)
591 * This function uses arithmetic modulo the logical_block_size
592 * (in units of 512-byte sectors).
594 unsigned short device_logical_block_size_sects =
595 limits->logical_block_size >> SECTOR_SHIFT;
598 * Offset of the start of the next table entry, mod logical_block_size.
600 unsigned short next_target_start = 0;
603 * Given an aligned bio that extends beyond the end of a
604 * target, how many sectors must the next target handle?
606 unsigned short remaining = 0;
608 struct dm_target *ti;
609 struct queue_limits ti_limits;
613 * Check each entry in the table in turn.
615 for (i = 0; i < dm_table_get_num_targets(table); i++) {
616 ti = dm_table_get_target(table, i);
618 blk_set_stacking_limits(&ti_limits);
620 /* combine all target devices' limits */
621 if (ti->type->iterate_devices)
622 ti->type->iterate_devices(ti, dm_set_device_limits,
626 * If the remaining sectors fall entirely within this
627 * table entry are they compatible with its logical_block_size?
629 if (remaining < ti->len &&
630 remaining & ((ti_limits.logical_block_size >>
635 (unsigned short) ((next_target_start + ti->len) &
636 (device_logical_block_size_sects - 1));
637 remaining = next_target_start ?
638 device_logical_block_size_sects - next_target_start : 0;
642 DMWARN("%s: table line %u (start sect %llu len %llu) "
643 "not aligned to h/w logical block size %u",
644 dm_device_name(table->md), i,
645 (unsigned long long) ti->begin,
646 (unsigned long long) ti->len,
647 limits->logical_block_size);
654 int dm_table_add_target(struct dm_table *t, const char *type,
655 sector_t start, sector_t len, char *params)
657 int r = -EINVAL, argc;
659 struct dm_target *tgt;
662 DMERR("%s: target type %s must appear alone in table",
663 dm_device_name(t->md), t->targets->type->name);
667 BUG_ON(t->num_targets >= t->num_allocated);
669 tgt = t->targets + t->num_targets;
670 memset(tgt, 0, sizeof(*tgt));
673 DMERR("%s: zero-length target", dm_device_name(t->md));
677 tgt->type = dm_get_target_type(type);
679 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
683 if (dm_target_needs_singleton(tgt->type)) {
684 if (t->num_targets) {
685 tgt->error = "singleton target type must appear alone in table";
691 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
692 tgt->error = "target type may not be included in a read-only table";
696 if (t->immutable_target_type) {
697 if (t->immutable_target_type != tgt->type) {
698 tgt->error = "immutable target type cannot be mixed with other target types";
701 } else if (dm_target_is_immutable(tgt->type)) {
702 if (t->num_targets) {
703 tgt->error = "immutable target type cannot be mixed with other target types";
706 t->immutable_target_type = tgt->type;
709 if (dm_target_has_integrity(tgt->type))
710 t->integrity_added = 1;
715 tgt->error = "Unknown error";
718 * Does this target adjoin the previous one ?
720 if (!adjoin(t, tgt)) {
721 tgt->error = "Gap in table";
725 r = dm_split_args(&argc, &argv, params);
727 tgt->error = "couldn't split parameters (insufficient memory)";
731 r = tgt->type->ctr(tgt, argc, argv);
736 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
738 if (!tgt->num_discard_bios && tgt->discards_supported)
739 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
740 dm_device_name(t->md), type);
745 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
746 dm_put_target_type(tgt->type);
751 * Target argument parsing helpers.
753 static int validate_next_arg(const struct dm_arg *arg,
754 struct dm_arg_set *arg_set,
755 unsigned *value, char **error, unsigned grouped)
757 const char *arg_str = dm_shift_arg(arg_set);
761 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
762 (*value < arg->min) ||
763 (*value > arg->max) ||
764 (grouped && arg_set->argc < *value)) {
772 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
773 unsigned *value, char **error)
775 return validate_next_arg(arg, arg_set, value, error, 0);
777 EXPORT_SYMBOL(dm_read_arg);
779 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
780 unsigned *value, char **error)
782 return validate_next_arg(arg, arg_set, value, error, 1);
784 EXPORT_SYMBOL(dm_read_arg_group);
786 const char *dm_shift_arg(struct dm_arg_set *as)
799 EXPORT_SYMBOL(dm_shift_arg);
801 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
803 BUG_ON(as->argc < num_args);
804 as->argc -= num_args;
805 as->argv += num_args;
807 EXPORT_SYMBOL(dm_consume_args);
809 static bool __table_type_bio_based(enum dm_queue_mode table_type)
811 return (table_type == DM_TYPE_BIO_BASED ||
812 table_type == DM_TYPE_DAX_BIO_BASED);
815 static bool __table_type_request_based(enum dm_queue_mode table_type)
817 return table_type == DM_TYPE_REQUEST_BASED;
820 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
824 EXPORT_SYMBOL_GPL(dm_table_set_type);
826 /* validate the dax capability of the target device span */
827 int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
828 sector_t start, sector_t len, void *data)
830 int blocksize = *(int *) data, id;
833 id = dax_read_lock();
834 rc = !dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
840 /* Check devices support synchronous DAX */
841 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
842 sector_t start, sector_t len, void *data)
844 return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
847 bool dm_table_supports_dax(struct dm_table *t,
848 iterate_devices_callout_fn iterate_fn, int *blocksize)
850 struct dm_target *ti;
853 /* Ensure that all targets support DAX. */
854 for (i = 0; i < dm_table_get_num_targets(t); i++) {
855 ti = dm_table_get_target(t, i);
857 if (!ti->type->direct_access)
860 if (!ti->type->iterate_devices ||
861 ti->type->iterate_devices(ti, iterate_fn, blocksize))
868 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
869 sector_t start, sector_t len, void *data)
871 struct block_device *bdev = dev->bdev;
872 struct request_queue *q = bdev_get_queue(bdev);
874 /* request-based cannot stack on partitions! */
875 if (bdev_is_partition(bdev))
878 return queue_is_mq(q);
881 static int dm_table_determine_type(struct dm_table *t)
884 unsigned bio_based = 0, request_based = 0, hybrid = 0;
885 struct dm_target *tgt;
886 struct list_head *devices = dm_table_get_devices(t);
887 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
888 int page_size = PAGE_SIZE;
890 if (t->type != DM_TYPE_NONE) {
891 /* target already set the table's type */
892 if (t->type == DM_TYPE_BIO_BASED) {
893 /* possibly upgrade to a variant of bio-based */
894 goto verify_bio_based;
896 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
897 goto verify_rq_based;
900 for (i = 0; i < t->num_targets; i++) {
901 tgt = t->targets + i;
902 if (dm_target_hybrid(tgt))
904 else if (dm_target_request_based(tgt))
909 if (bio_based && request_based) {
910 DMERR("Inconsistent table: different target types"
911 " can't be mixed up");
916 if (hybrid && !bio_based && !request_based) {
918 * The targets can work either way.
919 * Determine the type from the live device.
920 * Default to bio-based if device is new.
922 if (__table_type_request_based(live_md_type))
930 /* We must use this table as bio-based */
931 t->type = DM_TYPE_BIO_BASED;
932 if (dm_table_supports_dax(t, device_not_dax_capable, &page_size) ||
933 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
934 t->type = DM_TYPE_DAX_BIO_BASED;
939 BUG_ON(!request_based); /* No targets in this table */
941 t->type = DM_TYPE_REQUEST_BASED;
945 * Request-based dm supports only tables that have a single target now.
946 * To support multiple targets, request splitting support is needed,
947 * and that needs lots of changes in the block-layer.
948 * (e.g. request completion process for partial completion.)
950 if (t->num_targets > 1) {
951 DMERR("request-based DM doesn't support multiple targets");
955 if (list_empty(devices)) {
957 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
959 /* inherit live table's type */
961 t->type = live_table->type;
962 dm_put_live_table(t->md, srcu_idx);
966 tgt = dm_table_get_immutable_target(t);
968 DMERR("table load rejected: immutable target is required");
970 } else if (tgt->max_io_len) {
971 DMERR("table load rejected: immutable target that splits IO is not supported");
975 /* Non-request-stackable devices can't be used for request-based dm */
976 if (!tgt->type->iterate_devices ||
977 !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
978 DMERR("table load rejected: including non-request-stackable devices");
985 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
990 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
992 return t->immutable_target_type;
995 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
997 /* Immutable target is implicitly a singleton */
998 if (t->num_targets > 1 ||
999 !dm_target_is_immutable(t->targets[0].type))
1005 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1007 struct dm_target *ti;
1010 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1011 ti = dm_table_get_target(t, i);
1012 if (dm_target_is_wildcard(ti->type))
1019 bool dm_table_bio_based(struct dm_table *t)
1021 return __table_type_bio_based(dm_table_get_type(t));
1024 bool dm_table_request_based(struct dm_table *t)
1026 return __table_type_request_based(dm_table_get_type(t));
1029 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1031 enum dm_queue_mode type = dm_table_get_type(t);
1032 unsigned per_io_data_size = 0;
1033 unsigned min_pool_size = 0;
1034 struct dm_target *ti;
1037 if (unlikely(type == DM_TYPE_NONE)) {
1038 DMWARN("no table type is set, can't allocate mempools");
1042 if (__table_type_bio_based(type))
1043 for (i = 0; i < t->num_targets; i++) {
1044 ti = t->targets + i;
1045 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1046 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1049 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1050 per_io_data_size, min_pool_size);
1057 void dm_table_free_md_mempools(struct dm_table *t)
1059 dm_free_md_mempools(t->mempools);
1063 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1068 static int setup_indexes(struct dm_table *t)
1071 unsigned int total = 0;
1074 /* allocate the space for *all* the indexes */
1075 for (i = t->depth - 2; i >= 0; i--) {
1076 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1077 total += t->counts[i];
1080 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1084 /* set up internal nodes, bottom-up */
1085 for (i = t->depth - 2; i >= 0; i--) {
1086 t->index[i] = indexes;
1087 indexes += (KEYS_PER_NODE * t->counts[i]);
1088 setup_btree_index(i, t);
1095 * Builds the btree to index the map.
1097 static int dm_table_build_index(struct dm_table *t)
1100 unsigned int leaf_nodes;
1102 /* how many indexes will the btree have ? */
1103 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1104 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1106 /* leaf layer has already been set up */
1107 t->counts[t->depth - 1] = leaf_nodes;
1108 t->index[t->depth - 1] = t->highs;
1111 r = setup_indexes(t);
1116 static bool integrity_profile_exists(struct gendisk *disk)
1118 return !!blk_get_integrity(disk);
1122 * Get a disk whose integrity profile reflects the table's profile.
1123 * Returns NULL if integrity support was inconsistent or unavailable.
1125 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1127 struct list_head *devices = dm_table_get_devices(t);
1128 struct dm_dev_internal *dd = NULL;
1129 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1132 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1133 struct dm_target *ti = dm_table_get_target(t, i);
1134 if (!dm_target_passes_integrity(ti->type))
1138 list_for_each_entry(dd, devices, list) {
1139 template_disk = dd->dm_dev->bdev->bd_disk;
1140 if (!integrity_profile_exists(template_disk))
1142 else if (prev_disk &&
1143 blk_integrity_compare(prev_disk, template_disk) < 0)
1145 prev_disk = template_disk;
1148 return template_disk;
1152 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1153 dm_device_name(t->md),
1154 prev_disk->disk_name,
1155 template_disk->disk_name);
1160 * Register the mapped device for blk_integrity support if the
1161 * underlying devices have an integrity profile. But all devices may
1162 * not have matching profiles (checking all devices isn't reliable
1163 * during table load because this table may use other DM device(s) which
1164 * must be resumed before they will have an initialized integity
1165 * profile). Consequently, stacked DM devices force a 2 stage integrity
1166 * profile validation: First pass during table load, final pass during
1169 static int dm_table_register_integrity(struct dm_table *t)
1171 struct mapped_device *md = t->md;
1172 struct gendisk *template_disk = NULL;
1174 /* If target handles integrity itself do not register it here. */
1175 if (t->integrity_added)
1178 template_disk = dm_table_get_integrity_disk(t);
1182 if (!integrity_profile_exists(dm_disk(md))) {
1183 t->integrity_supported = true;
1185 * Register integrity profile during table load; we can do
1186 * this because the final profile must match during resume.
1188 blk_integrity_register(dm_disk(md),
1189 blk_get_integrity(template_disk));
1194 * If DM device already has an initialized integrity
1195 * profile the new profile should not conflict.
1197 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1198 DMWARN("%s: conflict with existing integrity profile: "
1199 "%s profile mismatch",
1200 dm_device_name(t->md),
1201 template_disk->disk_name);
1205 /* Preserve existing integrity profile */
1206 t->integrity_supported = true;
1210 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1212 struct dm_keyslot_manager {
1213 struct blk_keyslot_manager ksm;
1214 struct mapped_device *md;
1217 struct dm_keyslot_evict_args {
1218 const struct blk_crypto_key *key;
1222 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1223 sector_t start, sector_t len, void *data)
1225 struct dm_keyslot_evict_args *args = data;
1228 err = blk_crypto_evict_key(bdev_get_queue(dev->bdev), args->key);
1231 /* Always try to evict the key from all devices. */
1236 * When an inline encryption key is evicted from a device-mapper device, evict
1237 * it from all the underlying devices.
1239 static int dm_keyslot_evict(struct blk_keyslot_manager *ksm,
1240 const struct blk_crypto_key *key, unsigned int slot)
1242 struct dm_keyslot_manager *dksm = container_of(ksm,
1243 struct dm_keyslot_manager,
1245 struct mapped_device *md = dksm->md;
1246 struct dm_keyslot_evict_args args = { key };
1250 struct dm_target *ti;
1252 t = dm_get_live_table(md, &srcu_idx);
1255 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1256 ti = dm_table_get_target(t, i);
1257 if (!ti->type->iterate_devices)
1259 ti->type->iterate_devices(ti, dm_keyslot_evict_callback, &args);
1261 dm_put_live_table(md, srcu_idx);
1265 static struct blk_ksm_ll_ops dm_ksm_ll_ops = {
1266 .keyslot_evict = dm_keyslot_evict,
1269 static int device_intersect_crypto_modes(struct dm_target *ti,
1270 struct dm_dev *dev, sector_t start,
1271 sector_t len, void *data)
1273 struct blk_keyslot_manager *parent = data;
1274 struct blk_keyslot_manager *child = bdev_get_queue(dev->bdev)->ksm;
1276 blk_ksm_intersect_modes(parent, child);
1280 void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm)
1282 struct dm_keyslot_manager *dksm = container_of(ksm,
1283 struct dm_keyslot_manager,
1289 blk_ksm_destroy(ksm);
1293 static void dm_table_destroy_keyslot_manager(struct dm_table *t)
1295 dm_destroy_keyslot_manager(t->ksm);
1300 * Constructs and initializes t->ksm with a keyslot manager that
1301 * represents the common set of crypto capabilities of the devices
1302 * described by the dm_table. However, if the constructed keyslot
1303 * manager does not support a superset of the crypto capabilities
1304 * supported by the current keyslot manager of the mapped_device,
1305 * it returns an error instead, since we don't support restricting
1306 * crypto capabilities on table changes. Finally, if the constructed
1307 * keyslot manager doesn't actually support any crypto modes at all,
1308 * it just returns NULL.
1310 static int dm_table_construct_keyslot_manager(struct dm_table *t)
1312 struct dm_keyslot_manager *dksm;
1313 struct blk_keyslot_manager *ksm;
1314 struct dm_target *ti;
1316 bool ksm_is_empty = true;
1318 dksm = kmalloc(sizeof(*dksm), GFP_KERNEL);
1324 blk_ksm_init_passthrough(ksm);
1325 ksm->ksm_ll_ops = dm_ksm_ll_ops;
1326 ksm->max_dun_bytes_supported = UINT_MAX;
1327 memset(ksm->crypto_modes_supported, 0xFF,
1328 sizeof(ksm->crypto_modes_supported));
1330 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1331 ti = dm_table_get_target(t, i);
1333 if (!dm_target_passes_crypto(ti->type)) {
1334 blk_ksm_intersect_modes(ksm, NULL);
1337 if (!ti->type->iterate_devices)
1339 ti->type->iterate_devices(ti, device_intersect_crypto_modes,
1343 if (t->md->queue && !blk_ksm_is_superset(ksm, t->md->queue->ksm)) {
1344 DMWARN("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1345 dm_destroy_keyslot_manager(ksm);
1350 * If the new KSM doesn't actually support any crypto modes, we may as
1351 * well represent it with a NULL ksm.
1353 ksm_is_empty = true;
1354 for (i = 0; i < ARRAY_SIZE(ksm->crypto_modes_supported); i++) {
1355 if (ksm->crypto_modes_supported[i]) {
1356 ksm_is_empty = false;
1362 dm_destroy_keyslot_manager(ksm);
1367 * t->ksm is only set temporarily while the table is being set
1368 * up, and it gets set to NULL after the capabilities have
1369 * been transferred to the request_queue.
1376 static void dm_update_keyslot_manager(struct request_queue *q,
1382 /* Make the ksm less restrictive */
1384 blk_ksm_register(t->ksm, q);
1386 blk_ksm_update_capabilities(q->ksm, t->ksm);
1387 dm_destroy_keyslot_manager(t->ksm);
1392 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1394 static int dm_table_construct_keyslot_manager(struct dm_table *t)
1399 void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm)
1403 static void dm_table_destroy_keyslot_manager(struct dm_table *t)
1407 static void dm_update_keyslot_manager(struct request_queue *q,
1412 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1415 * Prepares the table for use by building the indices,
1416 * setting the type, and allocating mempools.
1418 int dm_table_complete(struct dm_table *t)
1422 r = dm_table_determine_type(t);
1424 DMERR("unable to determine table type");
1428 r = dm_table_build_index(t);
1430 DMERR("unable to build btrees");
1434 r = dm_table_register_integrity(t);
1436 DMERR("could not register integrity profile.");
1440 r = dm_table_construct_keyslot_manager(t);
1442 DMERR("could not construct keyslot manager.");
1446 r = dm_table_alloc_md_mempools(t, t->md);
1448 DMERR("unable to allocate mempools");
1453 static DEFINE_MUTEX(_event_lock);
1454 void dm_table_event_callback(struct dm_table *t,
1455 void (*fn)(void *), void *context)
1457 mutex_lock(&_event_lock);
1459 t->event_context = context;
1460 mutex_unlock(&_event_lock);
1463 void dm_table_event(struct dm_table *t)
1465 mutex_lock(&_event_lock);
1467 t->event_fn(t->event_context);
1468 mutex_unlock(&_event_lock);
1470 EXPORT_SYMBOL(dm_table_event);
1472 inline sector_t dm_table_get_size(struct dm_table *t)
1474 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1476 EXPORT_SYMBOL(dm_table_get_size);
1478 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1480 if (index >= t->num_targets)
1483 return t->targets + index;
1487 * Search the btree for the correct target.
1489 * Caller should check returned pointer for NULL
1490 * to trap I/O beyond end of device.
1492 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1494 unsigned int l, n = 0, k = 0;
1497 if (unlikely(sector >= dm_table_get_size(t)))
1500 for (l = 0; l < t->depth; l++) {
1501 n = get_child(n, k);
1502 node = get_node(t, l, n);
1504 for (k = 0; k < KEYS_PER_NODE; k++)
1505 if (node[k] >= sector)
1509 return &t->targets[(KEYS_PER_NODE * n) + k];
1513 * type->iterate_devices() should be called when the sanity check needs to
1514 * iterate and check all underlying data devices. iterate_devices() will
1515 * iterate all underlying data devices until it encounters a non-zero return
1516 * code, returned by whether the input iterate_devices_callout_fn, or
1517 * iterate_devices() itself internally.
1519 * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1520 * iterate multiple underlying devices internally, in which case a non-zero
1521 * return code returned by iterate_devices_callout_fn will stop the iteration
1524 * Cases requiring _any_ underlying device supporting some kind of attribute,
1525 * should use the iteration structure like dm_table_any_dev_attr(), or call
1526 * it directly. @func should handle semantics of positive examples, e.g.
1527 * capable of something.
1529 * Cases requiring _all_ underlying devices supporting some kind of attribute,
1530 * should use the iteration structure like dm_table_supports_nowait() or
1531 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1532 * uses an @anti_func that handle semantics of counter examples, e.g. not
1533 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1535 static bool dm_table_any_dev_attr(struct dm_table *t,
1536 iterate_devices_callout_fn func, void *data)
1538 struct dm_target *ti;
1541 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1542 ti = dm_table_get_target(t, i);
1544 if (ti->type->iterate_devices &&
1545 ti->type->iterate_devices(ti, func, data))
1552 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1553 sector_t start, sector_t len, void *data)
1555 unsigned *num_devices = data;
1563 * Check whether a table has no data devices attached using each
1564 * target's iterate_devices method.
1565 * Returns false if the result is unknown because a target doesn't
1566 * support iterate_devices.
1568 bool dm_table_has_no_data_devices(struct dm_table *table)
1570 struct dm_target *ti;
1571 unsigned i, num_devices;
1573 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1574 ti = dm_table_get_target(table, i);
1576 if (!ti->type->iterate_devices)
1580 ti->type->iterate_devices(ti, count_device, &num_devices);
1588 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1589 sector_t start, sector_t len, void *data)
1591 struct request_queue *q = bdev_get_queue(dev->bdev);
1592 enum blk_zoned_model *zoned_model = data;
1594 return blk_queue_zoned_model(q) != *zoned_model;
1598 * Check the device zoned model based on the target feature flag. If the target
1599 * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1600 * also accepted but all devices must have the same zoned model. If the target
1601 * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1602 * zoned model with all zoned devices having the same zone size.
1604 static bool dm_table_supports_zoned_model(struct dm_table *t,
1605 enum blk_zoned_model zoned_model)
1607 struct dm_target *ti;
1610 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1611 ti = dm_table_get_target(t, i);
1613 if (dm_target_supports_zoned_hm(ti->type)) {
1614 if (!ti->type->iterate_devices ||
1615 ti->type->iterate_devices(ti, device_not_zoned_model,
1618 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1619 if (zoned_model == BLK_ZONED_HM)
1627 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1628 sector_t start, sector_t len, void *data)
1630 struct request_queue *q = bdev_get_queue(dev->bdev);
1631 unsigned int *zone_sectors = data;
1633 if (!blk_queue_is_zoned(q))
1636 return blk_queue_zone_sectors(q) != *zone_sectors;
1640 * Check consistency of zoned model and zone sectors across all targets. For
1641 * zone sectors, if the destination device is a zoned block device, it shall
1642 * have the specified zone_sectors.
1644 static int validate_hardware_zoned_model(struct dm_table *table,
1645 enum blk_zoned_model zoned_model,
1646 unsigned int zone_sectors)
1648 if (zoned_model == BLK_ZONED_NONE)
1651 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1652 DMERR("%s: zoned model is not consistent across all devices",
1653 dm_device_name(table->md));
1657 /* Check zone size validity and compatibility */
1658 if (!zone_sectors || !is_power_of_2(zone_sectors))
1661 if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1662 DMERR("%s: zone sectors is not consistent across all zoned devices",
1663 dm_device_name(table->md));
1671 * Establish the new table's queue_limits and validate them.
1673 int dm_calculate_queue_limits(struct dm_table *table,
1674 struct queue_limits *limits)
1676 struct dm_target *ti;
1677 struct queue_limits ti_limits;
1679 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1680 unsigned int zone_sectors = 0;
1682 blk_set_stacking_limits(limits);
1684 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1685 blk_set_stacking_limits(&ti_limits);
1687 ti = dm_table_get_target(table, i);
1689 if (!ti->type->iterate_devices)
1690 goto combine_limits;
1693 * Combine queue limits of all the devices this target uses.
1695 ti->type->iterate_devices(ti, dm_set_device_limits,
1698 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1700 * After stacking all limits, validate all devices
1701 * in table support this zoned model and zone sectors.
1703 zoned_model = ti_limits.zoned;
1704 zone_sectors = ti_limits.chunk_sectors;
1707 /* Set I/O hints portion of queue limits */
1708 if (ti->type->io_hints)
1709 ti->type->io_hints(ti, &ti_limits);
1712 * Check each device area is consistent with the target's
1713 * overall queue limits.
1715 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1721 * Merge this target's queue limits into the overall limits
1724 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1725 DMWARN("%s: adding target device "
1726 "(start sect %llu len %llu) "
1727 "caused an alignment inconsistency",
1728 dm_device_name(table->md),
1729 (unsigned long long) ti->begin,
1730 (unsigned long long) ti->len);
1734 * Verify that the zoned model and zone sectors, as determined before
1735 * any .io_hints override, are the same across all devices in the table.
1736 * - this is especially relevant if .io_hints is emulating a disk-managed
1737 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1740 if (limits->zoned != BLK_ZONED_NONE) {
1742 * ...IF the above limits stacking determined a zoned model
1743 * validate that all of the table's devices conform to it.
1745 zoned_model = limits->zoned;
1746 zone_sectors = limits->chunk_sectors;
1748 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1751 return validate_hardware_logical_block_alignment(table, limits);
1755 * Verify that all devices have an integrity profile that matches the
1756 * DM device's registered integrity profile. If the profiles don't
1757 * match then unregister the DM device's integrity profile.
1759 static void dm_table_verify_integrity(struct dm_table *t)
1761 struct gendisk *template_disk = NULL;
1763 if (t->integrity_added)
1766 if (t->integrity_supported) {
1768 * Verify that the original integrity profile
1769 * matches all the devices in this table.
1771 template_disk = dm_table_get_integrity_disk(t);
1772 if (template_disk &&
1773 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1777 if (integrity_profile_exists(dm_disk(t->md))) {
1778 DMWARN("%s: unable to establish an integrity profile",
1779 dm_device_name(t->md));
1780 blk_integrity_unregister(dm_disk(t->md));
1784 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1785 sector_t start, sector_t len, void *data)
1787 unsigned long flush = (unsigned long) data;
1788 struct request_queue *q = bdev_get_queue(dev->bdev);
1790 return (q->queue_flags & flush);
1793 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1795 struct dm_target *ti;
1799 * Require at least one underlying device to support flushes.
1800 * t->devices includes internal dm devices such as mirror logs
1801 * so we need to use iterate_devices here, which targets
1802 * supporting flushes must provide.
1804 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1805 ti = dm_table_get_target(t, i);
1807 if (!ti->num_flush_bios)
1810 if (ti->flush_supported)
1813 if (ti->type->iterate_devices &&
1814 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1821 static int device_dax_write_cache_enabled(struct dm_target *ti,
1822 struct dm_dev *dev, sector_t start,
1823 sector_t len, void *data)
1825 struct dax_device *dax_dev = dev->dax_dev;
1830 if (dax_write_cache_enabled(dax_dev))
1835 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1836 sector_t start, sector_t len, void *data)
1838 struct request_queue *q = bdev_get_queue(dev->bdev);
1840 return !blk_queue_nonrot(q);
1843 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1844 sector_t start, sector_t len, void *data)
1846 struct request_queue *q = bdev_get_queue(dev->bdev);
1848 return !blk_queue_add_random(q);
1851 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1852 sector_t start, sector_t len, void *data)
1854 struct request_queue *q = bdev_get_queue(dev->bdev);
1856 return !q->limits.max_write_same_sectors;
1859 static bool dm_table_supports_write_same(struct dm_table *t)
1861 struct dm_target *ti;
1864 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1865 ti = dm_table_get_target(t, i);
1867 if (!ti->num_write_same_bios)
1870 if (!ti->type->iterate_devices ||
1871 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1878 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1879 sector_t start, sector_t len, void *data)
1881 struct request_queue *q = bdev_get_queue(dev->bdev);
1883 return !q->limits.max_write_zeroes_sectors;
1886 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1888 struct dm_target *ti;
1891 while (i < dm_table_get_num_targets(t)) {
1892 ti = dm_table_get_target(t, i++);
1894 if (!ti->num_write_zeroes_bios)
1897 if (!ti->type->iterate_devices ||
1898 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1905 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1906 sector_t start, sector_t len, void *data)
1908 struct request_queue *q = bdev_get_queue(dev->bdev);
1910 return !blk_queue_nowait(q);
1913 static bool dm_table_supports_nowait(struct dm_table *t)
1915 struct dm_target *ti;
1918 while (i < dm_table_get_num_targets(t)) {
1919 ti = dm_table_get_target(t, i++);
1921 if (!dm_target_supports_nowait(ti->type))
1924 if (!ti->type->iterate_devices ||
1925 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1932 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1933 sector_t start, sector_t len, void *data)
1935 struct request_queue *q = bdev_get_queue(dev->bdev);
1937 return !blk_queue_discard(q);
1940 static bool dm_table_supports_discards(struct dm_table *t)
1942 struct dm_target *ti;
1945 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1946 ti = dm_table_get_target(t, i);
1948 if (!ti->num_discard_bios)
1952 * Either the target provides discard support (as implied by setting
1953 * 'discards_supported') or it relies on _all_ data devices having
1956 if (!ti->discards_supported &&
1957 (!ti->type->iterate_devices ||
1958 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1965 static int device_not_secure_erase_capable(struct dm_target *ti,
1966 struct dm_dev *dev, sector_t start,
1967 sector_t len, void *data)
1969 struct request_queue *q = bdev_get_queue(dev->bdev);
1971 return !blk_queue_secure_erase(q);
1974 static bool dm_table_supports_secure_erase(struct dm_table *t)
1976 struct dm_target *ti;
1979 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1980 ti = dm_table_get_target(t, i);
1982 if (!ti->num_secure_erase_bios)
1985 if (!ti->type->iterate_devices ||
1986 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1993 static int device_requires_stable_pages(struct dm_target *ti,
1994 struct dm_dev *dev, sector_t start,
1995 sector_t len, void *data)
1997 struct request_queue *q = bdev_get_queue(dev->bdev);
1999 return blk_queue_stable_writes(q);
2002 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
2003 struct queue_limits *limits)
2005 bool wc = false, fua = false;
2006 int page_size = PAGE_SIZE;
2009 * Copy table's limits to the DM device's request_queue
2011 q->limits = *limits;
2013 if (dm_table_supports_nowait(t))
2014 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
2016 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
2018 if (!dm_table_supports_discards(t)) {
2019 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
2020 /* Must also clear discard limits... */
2021 q->limits.max_discard_sectors = 0;
2022 q->limits.max_hw_discard_sectors = 0;
2023 q->limits.discard_granularity = 0;
2024 q->limits.discard_alignment = 0;
2025 q->limits.discard_misaligned = 0;
2027 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
2029 if (dm_table_supports_secure_erase(t))
2030 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
2032 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
2034 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
2037 blk_queue_write_cache(q, wc, fua);
2039 if (dm_table_supports_dax(t, device_not_dax_capable, &page_size)) {
2040 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
2041 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable, NULL))
2042 set_dax_synchronous(t->md->dax_dev);
2045 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
2047 if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
2048 dax_write_cache(t->md->dax_dev, true);
2050 /* Ensure that all underlying devices are non-rotational. */
2051 if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
2052 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2054 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2056 if (!dm_table_supports_write_same(t))
2057 q->limits.max_write_same_sectors = 0;
2058 if (!dm_table_supports_write_zeroes(t))
2059 q->limits.max_write_zeroes_sectors = 0;
2061 dm_table_verify_integrity(t);
2064 * Some devices don't use blk_integrity but still want stable pages
2065 * because they do their own checksumming.
2066 * If any underlying device requires stable pages, a table must require
2067 * them as well. Only targets that support iterate_devices are considered:
2068 * don't want error, zero, etc to require stable pages.
2070 if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
2071 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2073 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2076 * Determine whether or not this queue's I/O timings contribute
2077 * to the entropy pool, Only request-based targets use this.
2078 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2081 if (blk_queue_add_random(q) &&
2082 dm_table_any_dev_attr(t, device_is_not_random, NULL))
2083 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2086 * For a zoned target, the number of zones should be updated for the
2087 * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
2088 * target, this is all that is needed.
2090 #ifdef CONFIG_BLK_DEV_ZONED
2091 if (blk_queue_is_zoned(q)) {
2092 WARN_ON_ONCE(queue_is_mq(q));
2093 q->nr_zones = blkdev_nr_zones(t->md->disk);
2097 dm_update_keyslot_manager(q, t);
2098 blk_queue_update_readahead(q);
2101 unsigned int dm_table_get_num_targets(struct dm_table *t)
2103 return t->num_targets;
2106 struct list_head *dm_table_get_devices(struct dm_table *t)
2111 fmode_t dm_table_get_mode(struct dm_table *t)
2115 EXPORT_SYMBOL(dm_table_get_mode);
2123 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2125 int i = t->num_targets;
2126 struct dm_target *ti = t->targets;
2128 lockdep_assert_held(&t->md->suspend_lock);
2133 if (ti->type->presuspend)
2134 ti->type->presuspend(ti);
2136 case PRESUSPEND_UNDO:
2137 if (ti->type->presuspend_undo)
2138 ti->type->presuspend_undo(ti);
2141 if (ti->type->postsuspend)
2142 ti->type->postsuspend(ti);
2149 void dm_table_presuspend_targets(struct dm_table *t)
2154 suspend_targets(t, PRESUSPEND);
2157 void dm_table_presuspend_undo_targets(struct dm_table *t)
2162 suspend_targets(t, PRESUSPEND_UNDO);
2165 void dm_table_postsuspend_targets(struct dm_table *t)
2170 suspend_targets(t, POSTSUSPEND);
2173 int dm_table_resume_targets(struct dm_table *t)
2177 lockdep_assert_held(&t->md->suspend_lock);
2179 for (i = 0; i < t->num_targets; i++) {
2180 struct dm_target *ti = t->targets + i;
2182 if (!ti->type->preresume)
2185 r = ti->type->preresume(ti);
2187 DMERR("%s: %s: preresume failed, error = %d",
2188 dm_device_name(t->md), ti->type->name, r);
2193 for (i = 0; i < t->num_targets; i++) {
2194 struct dm_target *ti = t->targets + i;
2196 if (ti->type->resume)
2197 ti->type->resume(ti);
2203 struct mapped_device *dm_table_get_md(struct dm_table *t)
2207 EXPORT_SYMBOL(dm_table_get_md);
2209 const char *dm_table_device_name(struct dm_table *t)
2211 return dm_device_name(t->md);
2213 EXPORT_SYMBOL_GPL(dm_table_device_name);
2215 void dm_table_run_md_queue_async(struct dm_table *t)
2217 if (!dm_table_request_based(t))
2221 blk_mq_run_hw_queues(t->md->queue, true);
2223 EXPORT_SYMBOL(dm_table_run_md_queue_async);