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"
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 struct mapped_device *md;
34 enum dm_queue_mode type;
38 unsigned int counts[MAX_DEPTH]; /* in nodes */
39 sector_t *index[MAX_DEPTH];
41 unsigned int num_targets;
42 unsigned int num_allocated;
44 struct dm_target *targets;
46 struct target_type *immutable_target_type;
48 bool integrity_supported:1;
51 unsigned integrity_added:1;
54 * Indicates the rw permissions for the new logical
55 * device. This should be a combination of FMODE_READ
60 /* a list of devices used by this table */
61 struct list_head devices;
63 /* events get handed up using this callback */
64 void (*event_fn)(void *);
67 struct dm_md_mempools *mempools;
69 struct list_head target_callbacks;
73 * Similar to ceiling(log_size(n))
75 static unsigned int int_log(unsigned int n, unsigned int base)
80 n = dm_div_up(n, base);
88 * Calculate the index of the child node of the n'th node k'th key.
90 static inline unsigned int get_child(unsigned int n, unsigned int k)
92 return (n * CHILDREN_PER_NODE) + k;
96 * Return the n'th node of level l from table t.
98 static inline sector_t *get_node(struct dm_table *t,
99 unsigned int l, unsigned int n)
101 return t->index[l] + (n * KEYS_PER_NODE);
105 * Return the highest key that you could lookup from the n'th
106 * node on level l of the btree.
108 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
110 for (; l < t->depth - 1; l++)
111 n = get_child(n, CHILDREN_PER_NODE - 1);
113 if (n >= t->counts[l])
114 return (sector_t) - 1;
116 return get_node(t, l, n)[KEYS_PER_NODE - 1];
120 * Fills in a level of the btree based on the highs of the level
123 static int setup_btree_index(unsigned int l, struct dm_table *t)
128 for (n = 0U; n < t->counts[l]; n++) {
129 node = get_node(t, l, n);
131 for (k = 0U; k < KEYS_PER_NODE; k++)
132 node[k] = high(t, l + 1, get_child(n, k));
138 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
144 * Check that we're not going to overflow.
146 if (nmemb > (ULONG_MAX / elem_size))
149 size = nmemb * elem_size;
150 addr = vzalloc(size);
154 EXPORT_SYMBOL(dm_vcalloc);
157 * highs, and targets are managed as dynamic arrays during a
160 static int alloc_targets(struct dm_table *t, unsigned int num)
163 struct dm_target *n_targets;
166 * Allocate both the target array and offset array at once.
167 * Append an empty entry to catch sectors beyond the end of
170 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
175 n_targets = (struct dm_target *) (n_highs + num);
177 memset(n_highs, -1, sizeof(*n_highs) * num);
180 t->num_allocated = num;
182 t->targets = n_targets;
187 int dm_table_create(struct dm_table **result, fmode_t mode,
188 unsigned num_targets, struct mapped_device *md)
190 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
195 INIT_LIST_HEAD(&t->devices);
196 INIT_LIST_HEAD(&t->target_callbacks);
199 num_targets = KEYS_PER_NODE;
201 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
208 if (alloc_targets(t, num_targets)) {
213 t->type = DM_TYPE_NONE;
220 static void free_devices(struct list_head *devices, struct mapped_device *md)
222 struct list_head *tmp, *next;
224 list_for_each_safe(tmp, next, devices) {
225 struct dm_dev_internal *dd =
226 list_entry(tmp, struct dm_dev_internal, list);
227 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
228 dm_device_name(md), dd->dm_dev->name);
229 dm_put_table_device(md, dd->dm_dev);
234 void dm_table_destroy(struct dm_table *t)
241 /* free the indexes */
243 vfree(t->index[t->depth - 2]);
245 /* free the targets */
246 for (i = 0; i < t->num_targets; i++) {
247 struct dm_target *tgt = t->targets + i;
252 dm_put_target_type(tgt->type);
257 /* free the device list */
258 free_devices(&t->devices, t->md);
260 dm_free_md_mempools(t->mempools);
266 * See if we've already got a device in the list.
268 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
270 struct dm_dev_internal *dd;
272 list_for_each_entry (dd, l, list)
273 if (dd->dm_dev->bdev->bd_dev == dev)
280 * If possible, this checks an area of a destination device is invalid.
282 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
283 sector_t start, sector_t len, void *data)
285 struct request_queue *q;
286 struct queue_limits *limits = data;
287 struct block_device *bdev = dev->bdev;
289 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
290 unsigned short logical_block_size_sectors =
291 limits->logical_block_size >> SECTOR_SHIFT;
292 char b[BDEVNAME_SIZE];
295 * Some devices exist without request functions,
296 * such as loop devices not yet bound to backing files.
297 * Forbid the use of such devices.
299 q = bdev_get_queue(bdev);
300 if (!q || !q->make_request_fn) {
301 DMWARN("%s: %s is not yet initialised: "
302 "start=%llu, len=%llu, dev_size=%llu",
303 dm_device_name(ti->table->md), bdevname(bdev, b),
304 (unsigned long long)start,
305 (unsigned long long)len,
306 (unsigned long long)dev_size);
313 if ((start >= dev_size) || (start + len > dev_size)) {
314 DMWARN("%s: %s too small for target: "
315 "start=%llu, len=%llu, dev_size=%llu",
316 dm_device_name(ti->table->md), bdevname(bdev, b),
317 (unsigned long long)start,
318 (unsigned long long)len,
319 (unsigned long long)dev_size);
324 * If the target is mapped to zoned block device(s), check
325 * that the zones are not partially mapped.
327 if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
328 unsigned int zone_sectors = bdev_zone_sectors(bdev);
330 if (start & (zone_sectors - 1)) {
331 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
332 dm_device_name(ti->table->md),
333 (unsigned long long)start,
334 zone_sectors, bdevname(bdev, b));
339 * Note: The last zone of a zoned block device may be smaller
340 * than other zones. So for a target mapping the end of a
341 * zoned block device with such a zone, len would not be zone
342 * aligned. We do not allow such last smaller zone to be part
343 * of the mapping here to ensure that mappings with multiple
344 * devices do not end up with a smaller zone in the middle of
347 if (len & (zone_sectors - 1)) {
348 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
349 dm_device_name(ti->table->md),
350 (unsigned long long)len,
351 zone_sectors, bdevname(bdev, b));
356 if (logical_block_size_sectors <= 1)
359 if (start & (logical_block_size_sectors - 1)) {
360 DMWARN("%s: start=%llu not aligned to h/w "
361 "logical block size %u of %s",
362 dm_device_name(ti->table->md),
363 (unsigned long long)start,
364 limits->logical_block_size, bdevname(bdev, b));
368 if (len & (logical_block_size_sectors - 1)) {
369 DMWARN("%s: len=%llu not aligned to h/w "
370 "logical block size %u of %s",
371 dm_device_name(ti->table->md),
372 (unsigned long long)len,
373 limits->logical_block_size, bdevname(bdev, b));
381 * This upgrades the mode on an already open dm_dev, being
382 * careful to leave things as they were if we fail to reopen the
383 * device and not to touch the existing bdev field in case
384 * it is accessed concurrently inside dm_table_any_congested().
386 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
387 struct mapped_device *md)
390 struct dm_dev *old_dev, *new_dev;
392 old_dev = dd->dm_dev;
394 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
395 dd->dm_dev->mode | new_mode, &new_dev);
399 dd->dm_dev = new_dev;
400 dm_put_table_device(md, old_dev);
406 * Convert the path to a device
408 dev_t dm_get_dev_t(const char *path)
411 struct block_device *bdev;
413 bdev = lookup_bdev(path);
415 dev = name_to_dev_t(path);
423 EXPORT_SYMBOL_GPL(dm_get_dev_t);
426 * Add a device to the list, or just increment the usage count if
427 * it's already present.
429 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
430 struct dm_dev **result)
434 struct dm_dev_internal *dd;
435 struct dm_table *t = ti->table;
439 dev = dm_get_dev_t(path);
443 dd = find_device(&t->devices, dev);
445 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
449 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
454 refcount_set(&dd->count, 1);
455 list_add(&dd->list, &t->devices);
458 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
459 r = upgrade_mode(dd, mode, t->md);
463 refcount_inc(&dd->count);
465 *result = dd->dm_dev;
468 EXPORT_SYMBOL(dm_get_device);
470 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
471 sector_t start, sector_t len, void *data)
473 struct queue_limits *limits = data;
474 struct block_device *bdev = dev->bdev;
475 struct request_queue *q = bdev_get_queue(bdev);
476 char b[BDEVNAME_SIZE];
479 DMWARN("%s: Cannot set limits for nonexistent device %s",
480 dm_device_name(ti->table->md), bdevname(bdev, b));
484 if (bdev_stack_limits(limits, bdev, start) < 0)
485 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
486 "physical_block_size=%u, logical_block_size=%u, "
487 "alignment_offset=%u, start=%llu",
488 dm_device_name(ti->table->md), bdevname(bdev, b),
489 q->limits.physical_block_size,
490 q->limits.logical_block_size,
491 q->limits.alignment_offset,
492 (unsigned long long) start << SECTOR_SHIFT);
494 limits->zoned = blk_queue_zoned_model(q);
500 * Decrement a device's use count and remove it if necessary.
502 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
505 struct list_head *devices = &ti->table->devices;
506 struct dm_dev_internal *dd;
508 list_for_each_entry(dd, devices, list) {
509 if (dd->dm_dev == d) {
515 DMWARN("%s: device %s not in table devices list",
516 dm_device_name(ti->table->md), d->name);
519 if (refcount_dec_and_test(&dd->count)) {
520 dm_put_table_device(ti->table->md, d);
525 EXPORT_SYMBOL(dm_put_device);
528 * Checks to see if the target joins onto the end of the table.
530 static int adjoin(struct dm_table *table, struct dm_target *ti)
532 struct dm_target *prev;
534 if (!table->num_targets)
537 prev = &table->targets[table->num_targets - 1];
538 return (ti->begin == (prev->begin + prev->len));
542 * Used to dynamically allocate the arg array.
544 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
545 * process messages even if some device is suspended. These messages have a
546 * small fixed number of arguments.
548 * On the other hand, dm-switch needs to process bulk data using messages and
549 * excessive use of GFP_NOIO could cause trouble.
551 static char **realloc_argv(unsigned *array_size, char **old_argv)
558 new_size = *array_size * 2;
564 argv = kmalloc(new_size * sizeof(*argv), gfp);
566 memcpy(argv, old_argv, *array_size * sizeof(*argv));
567 *array_size = new_size;
575 * Destructively splits up the argument list to pass to ctr.
577 int dm_split_args(int *argc, char ***argvp, char *input)
579 char *start, *end = input, *out, **argv = NULL;
580 unsigned array_size = 0;
589 argv = realloc_argv(&array_size, argv);
594 /* Skip whitespace */
595 start = skip_spaces(end);
598 break; /* success, we hit the end */
600 /* 'out' is used to remove any back-quotes */
603 /* Everything apart from '\0' can be quoted */
604 if (*end == '\\' && *(end + 1)) {
611 break; /* end of token */
616 /* have we already filled the array ? */
617 if ((*argc + 1) > array_size) {
618 argv = realloc_argv(&array_size, argv);
623 /* we know this is whitespace */
627 /* terminate the string and put it in the array */
638 * Impose necessary and sufficient conditions on a devices's table such
639 * that any incoming bio which respects its logical_block_size can be
640 * processed successfully. If it falls across the boundary between
641 * two or more targets, the size of each piece it gets split into must
642 * be compatible with the logical_block_size of the target processing it.
644 static int validate_hardware_logical_block_alignment(struct dm_table *table,
645 struct queue_limits *limits)
648 * This function uses arithmetic modulo the logical_block_size
649 * (in units of 512-byte sectors).
651 unsigned short device_logical_block_size_sects =
652 limits->logical_block_size >> SECTOR_SHIFT;
655 * Offset of the start of the next table entry, mod logical_block_size.
657 unsigned short next_target_start = 0;
660 * Given an aligned bio that extends beyond the end of a
661 * target, how many sectors must the next target handle?
663 unsigned short remaining = 0;
665 struct dm_target *uninitialized_var(ti);
666 struct queue_limits ti_limits;
670 * Check each entry in the table in turn.
672 for (i = 0; i < dm_table_get_num_targets(table); i++) {
673 ti = dm_table_get_target(table, i);
675 blk_set_stacking_limits(&ti_limits);
677 /* combine all target devices' limits */
678 if (ti->type->iterate_devices)
679 ti->type->iterate_devices(ti, dm_set_device_limits,
683 * If the remaining sectors fall entirely within this
684 * table entry are they compatible with its logical_block_size?
686 if (remaining < ti->len &&
687 remaining & ((ti_limits.logical_block_size >>
692 (unsigned short) ((next_target_start + ti->len) &
693 (device_logical_block_size_sects - 1));
694 remaining = next_target_start ?
695 device_logical_block_size_sects - next_target_start : 0;
699 DMWARN("%s: table line %u (start sect %llu len %llu) "
700 "not aligned to h/w logical block size %u",
701 dm_device_name(table->md), i,
702 (unsigned long long) ti->begin,
703 (unsigned long long) ti->len,
704 limits->logical_block_size);
711 int dm_table_add_target(struct dm_table *t, const char *type,
712 sector_t start, sector_t len, char *params)
714 int r = -EINVAL, argc;
716 struct dm_target *tgt;
719 DMERR("%s: target type %s must appear alone in table",
720 dm_device_name(t->md), t->targets->type->name);
724 BUG_ON(t->num_targets >= t->num_allocated);
726 tgt = t->targets + t->num_targets;
727 memset(tgt, 0, sizeof(*tgt));
730 DMERR("%s: zero-length target", dm_device_name(t->md));
734 tgt->type = dm_get_target_type(type);
736 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
740 if (dm_target_needs_singleton(tgt->type)) {
741 if (t->num_targets) {
742 tgt->error = "singleton target type must appear alone in table";
748 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
749 tgt->error = "target type may not be included in a read-only table";
753 if (t->immutable_target_type) {
754 if (t->immutable_target_type != tgt->type) {
755 tgt->error = "immutable target type cannot be mixed with other target types";
758 } else if (dm_target_is_immutable(tgt->type)) {
759 if (t->num_targets) {
760 tgt->error = "immutable target type cannot be mixed with other target types";
763 t->immutable_target_type = tgt->type;
766 if (dm_target_has_integrity(tgt->type))
767 t->integrity_added = 1;
772 tgt->error = "Unknown error";
775 * Does this target adjoin the previous one ?
777 if (!adjoin(t, tgt)) {
778 tgt->error = "Gap in table";
782 r = dm_split_args(&argc, &argv, params);
784 tgt->error = "couldn't split parameters (insufficient memory)";
788 r = tgt->type->ctr(tgt, argc, argv);
793 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
795 if (!tgt->num_discard_bios && tgt->discards_supported)
796 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
797 dm_device_name(t->md), type);
802 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
803 dm_put_target_type(tgt->type);
808 * Target argument parsing helpers.
810 static int validate_next_arg(const struct dm_arg *arg,
811 struct dm_arg_set *arg_set,
812 unsigned *value, char **error, unsigned grouped)
814 const char *arg_str = dm_shift_arg(arg_set);
818 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
819 (*value < arg->min) ||
820 (*value > arg->max) ||
821 (grouped && arg_set->argc < *value)) {
829 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
830 unsigned *value, char **error)
832 return validate_next_arg(arg, arg_set, value, error, 0);
834 EXPORT_SYMBOL(dm_read_arg);
836 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
837 unsigned *value, char **error)
839 return validate_next_arg(arg, arg_set, value, error, 1);
841 EXPORT_SYMBOL(dm_read_arg_group);
843 const char *dm_shift_arg(struct dm_arg_set *as)
856 EXPORT_SYMBOL(dm_shift_arg);
858 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
860 BUG_ON(as->argc < num_args);
861 as->argc -= num_args;
862 as->argv += num_args;
864 EXPORT_SYMBOL(dm_consume_args);
866 static bool __table_type_bio_based(enum dm_queue_mode table_type)
868 return (table_type == DM_TYPE_BIO_BASED ||
869 table_type == DM_TYPE_DAX_BIO_BASED);
872 static bool __table_type_request_based(enum dm_queue_mode table_type)
874 return (table_type == DM_TYPE_REQUEST_BASED ||
875 table_type == DM_TYPE_MQ_REQUEST_BASED);
878 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
882 EXPORT_SYMBOL_GPL(dm_table_set_type);
884 static int device_supports_dax(struct dm_target *ti, struct dm_dev *dev,
885 sector_t start, sector_t len, void *data)
887 struct request_queue *q = bdev_get_queue(dev->bdev);
889 return q && blk_queue_dax(q);
892 static bool dm_table_supports_dax(struct dm_table *t)
894 struct dm_target *ti;
897 /* Ensure that all targets support DAX. */
898 for (i = 0; i < dm_table_get_num_targets(t); i++) {
899 ti = dm_table_get_target(t, i);
901 if (!ti->type->direct_access)
904 if (!ti->type->iterate_devices ||
905 !ti->type->iterate_devices(ti, device_supports_dax, NULL))
912 static int dm_table_determine_type(struct dm_table *t)
915 unsigned bio_based = 0, request_based = 0, hybrid = 0;
916 unsigned sq_count = 0, mq_count = 0;
917 struct dm_target *tgt;
918 struct dm_dev_internal *dd;
919 struct list_head *devices = dm_table_get_devices(t);
920 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
922 if (t->type != DM_TYPE_NONE) {
923 /* target already set the table's type */
924 if (t->type == DM_TYPE_BIO_BASED)
926 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
927 goto verify_rq_based;
930 for (i = 0; i < t->num_targets; i++) {
931 tgt = t->targets + i;
932 if (dm_target_hybrid(tgt))
934 else if (dm_target_request_based(tgt))
939 if (bio_based && request_based) {
940 DMWARN("Inconsistent table: different target types"
941 " can't be mixed up");
946 if (hybrid && !bio_based && !request_based) {
948 * The targets can work either way.
949 * Determine the type from the live device.
950 * Default to bio-based if device is new.
952 if (__table_type_request_based(live_md_type))
959 /* We must use this table as bio-based */
960 t->type = DM_TYPE_BIO_BASED;
961 if (dm_table_supports_dax(t) ||
962 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED))
963 t->type = DM_TYPE_DAX_BIO_BASED;
967 BUG_ON(!request_based); /* No targets in this table */
970 * The only way to establish DM_TYPE_MQ_REQUEST_BASED is by
971 * having a compatible target use dm_table_set_type.
973 t->type = DM_TYPE_REQUEST_BASED;
977 * Request-based dm supports only tables that have a single target now.
978 * To support multiple targets, request splitting support is needed,
979 * and that needs lots of changes in the block-layer.
980 * (e.g. request completion process for partial completion.)
982 if (t->num_targets > 1) {
983 DMWARN("Request-based dm doesn't support multiple targets yet");
987 if (list_empty(devices)) {
989 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
991 /* inherit live table's type and all_blk_mq */
993 t->type = live_table->type;
994 t->all_blk_mq = live_table->all_blk_mq;
996 dm_put_live_table(t->md, srcu_idx);
1000 /* Non-request-stackable devices can't be used for request-based dm */
1001 list_for_each_entry(dd, devices, list) {
1002 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1004 if (!queue_is_rq_based(q)) {
1005 DMERR("table load rejected: including"
1006 " non-request-stackable devices");
1015 if (sq_count && mq_count) {
1016 DMERR("table load rejected: not all devices are blk-mq request-stackable");
1019 t->all_blk_mq = mq_count > 0;
1021 if (t->type == DM_TYPE_MQ_REQUEST_BASED && !t->all_blk_mq) {
1022 DMERR("table load rejected: all devices are not blk-mq request-stackable");
1029 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
1034 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
1036 return t->immutable_target_type;
1039 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
1041 /* Immutable target is implicitly a singleton */
1042 if (t->num_targets > 1 ||
1043 !dm_target_is_immutable(t->targets[0].type))
1049 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
1051 struct dm_target *ti;
1054 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1055 ti = dm_table_get_target(t, i);
1056 if (dm_target_is_wildcard(ti->type))
1063 bool dm_table_bio_based(struct dm_table *t)
1065 return __table_type_bio_based(dm_table_get_type(t));
1068 bool dm_table_request_based(struct dm_table *t)
1070 return __table_type_request_based(dm_table_get_type(t));
1073 bool dm_table_all_blk_mq_devices(struct dm_table *t)
1075 return t->all_blk_mq;
1078 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1080 enum dm_queue_mode type = dm_table_get_type(t);
1081 unsigned per_io_data_size = 0;
1082 struct dm_target *tgt;
1085 if (unlikely(type == DM_TYPE_NONE)) {
1086 DMWARN("no table type is set, can't allocate mempools");
1090 if (__table_type_bio_based(type))
1091 for (i = 0; i < t->num_targets; i++) {
1092 tgt = t->targets + i;
1093 per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
1096 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
1103 void dm_table_free_md_mempools(struct dm_table *t)
1105 dm_free_md_mempools(t->mempools);
1109 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1114 static int setup_indexes(struct dm_table *t)
1117 unsigned int total = 0;
1120 /* allocate the space for *all* the indexes */
1121 for (i = t->depth - 2; i >= 0; i--) {
1122 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1123 total += t->counts[i];
1126 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1130 /* set up internal nodes, bottom-up */
1131 for (i = t->depth - 2; i >= 0; i--) {
1132 t->index[i] = indexes;
1133 indexes += (KEYS_PER_NODE * t->counts[i]);
1134 setup_btree_index(i, t);
1141 * Builds the btree to index the map.
1143 static int dm_table_build_index(struct dm_table *t)
1146 unsigned int leaf_nodes;
1148 /* how many indexes will the btree have ? */
1149 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1150 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1152 /* leaf layer has already been set up */
1153 t->counts[t->depth - 1] = leaf_nodes;
1154 t->index[t->depth - 1] = t->highs;
1157 r = setup_indexes(t);
1162 static bool integrity_profile_exists(struct gendisk *disk)
1164 return !!blk_get_integrity(disk);
1168 * Get a disk whose integrity profile reflects the table's profile.
1169 * Returns NULL if integrity support was inconsistent or unavailable.
1171 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1173 struct list_head *devices = dm_table_get_devices(t);
1174 struct dm_dev_internal *dd = NULL;
1175 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1178 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1179 struct dm_target *ti = dm_table_get_target(t, i);
1180 if (!dm_target_passes_integrity(ti->type))
1184 list_for_each_entry(dd, devices, list) {
1185 template_disk = dd->dm_dev->bdev->bd_disk;
1186 if (!integrity_profile_exists(template_disk))
1188 else if (prev_disk &&
1189 blk_integrity_compare(prev_disk, template_disk) < 0)
1191 prev_disk = template_disk;
1194 return template_disk;
1198 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1199 dm_device_name(t->md),
1200 prev_disk->disk_name,
1201 template_disk->disk_name);
1206 * Register the mapped device for blk_integrity support if the
1207 * underlying devices have an integrity profile. But all devices may
1208 * not have matching profiles (checking all devices isn't reliable
1209 * during table load because this table may use other DM device(s) which
1210 * must be resumed before they will have an initialized integity
1211 * profile). Consequently, stacked DM devices force a 2 stage integrity
1212 * profile validation: First pass during table load, final pass during
1215 static int dm_table_register_integrity(struct dm_table *t)
1217 struct mapped_device *md = t->md;
1218 struct gendisk *template_disk = NULL;
1220 /* If target handles integrity itself do not register it here. */
1221 if (t->integrity_added)
1224 template_disk = dm_table_get_integrity_disk(t);
1228 if (!integrity_profile_exists(dm_disk(md))) {
1229 t->integrity_supported = true;
1231 * Register integrity profile during table load; we can do
1232 * this because the final profile must match during resume.
1234 blk_integrity_register(dm_disk(md),
1235 blk_get_integrity(template_disk));
1240 * If DM device already has an initialized integrity
1241 * profile the new profile should not conflict.
1243 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1244 DMWARN("%s: conflict with existing integrity profile: "
1245 "%s profile mismatch",
1246 dm_device_name(t->md),
1247 template_disk->disk_name);
1251 /* Preserve existing integrity profile */
1252 t->integrity_supported = true;
1257 * Prepares the table for use by building the indices,
1258 * setting the type, and allocating mempools.
1260 int dm_table_complete(struct dm_table *t)
1264 r = dm_table_determine_type(t);
1266 DMERR("unable to determine table type");
1270 r = dm_table_build_index(t);
1272 DMERR("unable to build btrees");
1276 r = dm_table_register_integrity(t);
1278 DMERR("could not register integrity profile.");
1282 r = dm_table_alloc_md_mempools(t, t->md);
1284 DMERR("unable to allocate mempools");
1289 static DEFINE_MUTEX(_event_lock);
1290 void dm_table_event_callback(struct dm_table *t,
1291 void (*fn)(void *), void *context)
1293 mutex_lock(&_event_lock);
1295 t->event_context = context;
1296 mutex_unlock(&_event_lock);
1299 void dm_table_event(struct dm_table *t)
1302 * You can no longer call dm_table_event() from interrupt
1303 * context, use a bottom half instead.
1305 BUG_ON(in_interrupt());
1307 mutex_lock(&_event_lock);
1309 t->event_fn(t->event_context);
1310 mutex_unlock(&_event_lock);
1312 EXPORT_SYMBOL(dm_table_event);
1314 sector_t dm_table_get_size(struct dm_table *t)
1316 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1318 EXPORT_SYMBOL(dm_table_get_size);
1320 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1322 if (index >= t->num_targets)
1325 return t->targets + index;
1329 * Search the btree for the correct target.
1331 * Caller should check returned pointer with dm_target_is_valid()
1332 * to trap I/O beyond end of device.
1334 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1336 unsigned int l, n = 0, k = 0;
1339 for (l = 0; l < t->depth; l++) {
1340 n = get_child(n, k);
1341 node = get_node(t, l, n);
1343 for (k = 0; k < KEYS_PER_NODE; k++)
1344 if (node[k] >= sector)
1348 return &t->targets[(KEYS_PER_NODE * n) + k];
1351 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1352 sector_t start, sector_t len, void *data)
1354 unsigned *num_devices = data;
1362 * Check whether a table has no data devices attached using each
1363 * target's iterate_devices method.
1364 * Returns false if the result is unknown because a target doesn't
1365 * support iterate_devices.
1367 bool dm_table_has_no_data_devices(struct dm_table *table)
1369 struct dm_target *ti;
1370 unsigned i, num_devices;
1372 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1373 ti = dm_table_get_target(table, i);
1375 if (!ti->type->iterate_devices)
1379 ti->type->iterate_devices(ti, count_device, &num_devices);
1387 static int device_is_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1388 sector_t start, sector_t len, void *data)
1390 struct request_queue *q = bdev_get_queue(dev->bdev);
1391 enum blk_zoned_model *zoned_model = data;
1393 return q && blk_queue_zoned_model(q) == *zoned_model;
1396 static bool dm_table_supports_zoned_model(struct dm_table *t,
1397 enum blk_zoned_model zoned_model)
1399 struct dm_target *ti;
1402 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1403 ti = dm_table_get_target(t, i);
1405 if (zoned_model == BLK_ZONED_HM &&
1406 !dm_target_supports_zoned_hm(ti->type))
1409 if (!ti->type->iterate_devices ||
1410 !ti->type->iterate_devices(ti, device_is_zoned_model, &zoned_model))
1417 static int device_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1418 sector_t start, sector_t len, void *data)
1420 struct request_queue *q = bdev_get_queue(dev->bdev);
1421 unsigned int *zone_sectors = data;
1423 return q && blk_queue_zone_sectors(q) == *zone_sectors;
1426 static bool dm_table_matches_zone_sectors(struct dm_table *t,
1427 unsigned int zone_sectors)
1429 struct dm_target *ti;
1432 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1433 ti = dm_table_get_target(t, i);
1435 if (!ti->type->iterate_devices ||
1436 !ti->type->iterate_devices(ti, device_matches_zone_sectors, &zone_sectors))
1443 static int validate_hardware_zoned_model(struct dm_table *table,
1444 enum blk_zoned_model zoned_model,
1445 unsigned int zone_sectors)
1447 if (zoned_model == BLK_ZONED_NONE)
1450 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1451 DMERR("%s: zoned model is not consistent across all devices",
1452 dm_device_name(table->md));
1456 /* Check zone size validity and compatibility */
1457 if (!zone_sectors || !is_power_of_2(zone_sectors))
1460 if (!dm_table_matches_zone_sectors(table, zone_sectors)) {
1461 DMERR("%s: zone sectors is not consistent across all devices",
1462 dm_device_name(table->md));
1470 * Establish the new table's queue_limits and validate them.
1472 int dm_calculate_queue_limits(struct dm_table *table,
1473 struct queue_limits *limits)
1475 struct dm_target *ti;
1476 struct queue_limits ti_limits;
1478 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1479 unsigned int zone_sectors = 0;
1481 blk_set_stacking_limits(limits);
1483 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1484 blk_set_stacking_limits(&ti_limits);
1486 ti = dm_table_get_target(table, i);
1488 if (!ti->type->iterate_devices)
1489 goto combine_limits;
1492 * Combine queue limits of all the devices this target uses.
1494 ti->type->iterate_devices(ti, dm_set_device_limits,
1497 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1499 * After stacking all limits, validate all devices
1500 * in table support this zoned model and zone sectors.
1502 zoned_model = ti_limits.zoned;
1503 zone_sectors = ti_limits.chunk_sectors;
1506 /* Set I/O hints portion of queue limits */
1507 if (ti->type->io_hints)
1508 ti->type->io_hints(ti, &ti_limits);
1511 * Check each device area is consistent with the target's
1512 * overall queue limits.
1514 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1520 * Merge this target's queue limits into the overall limits
1523 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1524 DMWARN("%s: adding target device "
1525 "(start sect %llu len %llu) "
1526 "caused an alignment inconsistency",
1527 dm_device_name(table->md),
1528 (unsigned long long) ti->begin,
1529 (unsigned long long) ti->len);
1532 * FIXME: this should likely be moved to blk_stack_limits(), would
1533 * also eliminate limits->zoned stacking hack in dm_set_device_limits()
1535 if (limits->zoned == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1537 * By default, the stacked limits zoned model is set to
1538 * BLK_ZONED_NONE in blk_set_stacking_limits(). Update
1539 * this model using the first target model reported
1540 * that is not BLK_ZONED_NONE. This will be either the
1541 * first target device zoned model or the model reported
1542 * by the target .io_hints.
1544 limits->zoned = ti_limits.zoned;
1549 * Verify that the zoned model and zone sectors, as determined before
1550 * any .io_hints override, are the same across all devices in the table.
1551 * - this is especially relevant if .io_hints is emulating a disk-managed
1552 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1555 if (limits->zoned != BLK_ZONED_NONE) {
1557 * ...IF the above limits stacking determined a zoned model
1558 * validate that all of the table's devices conform to it.
1560 zoned_model = limits->zoned;
1561 zone_sectors = limits->chunk_sectors;
1563 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1566 return validate_hardware_logical_block_alignment(table, limits);
1570 * Verify that all devices have an integrity profile that matches the
1571 * DM device's registered integrity profile. If the profiles don't
1572 * match then unregister the DM device's integrity profile.
1574 static void dm_table_verify_integrity(struct dm_table *t)
1576 struct gendisk *template_disk = NULL;
1578 if (t->integrity_added)
1581 if (t->integrity_supported) {
1583 * Verify that the original integrity profile
1584 * matches all the devices in this table.
1586 template_disk = dm_table_get_integrity_disk(t);
1587 if (template_disk &&
1588 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1592 if (integrity_profile_exists(dm_disk(t->md))) {
1593 DMWARN("%s: unable to establish an integrity profile",
1594 dm_device_name(t->md));
1595 blk_integrity_unregister(dm_disk(t->md));
1599 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1600 sector_t start, sector_t len, void *data)
1602 unsigned long flush = (unsigned long) data;
1603 struct request_queue *q = bdev_get_queue(dev->bdev);
1605 return q && (q->queue_flags & flush);
1608 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1610 struct dm_target *ti;
1614 * Require at least one underlying device to support flushes.
1615 * t->devices includes internal dm devices such as mirror logs
1616 * so we need to use iterate_devices here, which targets
1617 * supporting flushes must provide.
1619 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1620 ti = dm_table_get_target(t, i);
1622 if (!ti->num_flush_bios)
1625 if (ti->flush_supported)
1628 if (ti->type->iterate_devices &&
1629 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1636 static int device_dax_write_cache_enabled(struct dm_target *ti,
1637 struct dm_dev *dev, sector_t start,
1638 sector_t len, void *data)
1640 struct dax_device *dax_dev = dev->dax_dev;
1645 if (dax_write_cache_enabled(dax_dev))
1650 static int dm_table_supports_dax_write_cache(struct dm_table *t)
1652 struct dm_target *ti;
1655 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1656 ti = dm_table_get_target(t, i);
1658 if (ti->type->iterate_devices &&
1659 ti->type->iterate_devices(ti,
1660 device_dax_write_cache_enabled, NULL))
1667 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1668 sector_t start, sector_t len, void *data)
1670 struct request_queue *q = bdev_get_queue(dev->bdev);
1672 return q && blk_queue_nonrot(q);
1675 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1676 sector_t start, sector_t len, void *data)
1678 struct request_queue *q = bdev_get_queue(dev->bdev);
1680 return q && !blk_queue_add_random(q);
1683 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1684 sector_t start, sector_t len, void *data)
1686 struct request_queue *q = bdev_get_queue(dev->bdev);
1688 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1691 static bool dm_table_all_devices_attribute(struct dm_table *t,
1692 iterate_devices_callout_fn func)
1694 struct dm_target *ti;
1697 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1698 ti = dm_table_get_target(t, i);
1700 if (!ti->type->iterate_devices ||
1701 !ti->type->iterate_devices(ti, func, NULL))
1708 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1709 sector_t start, sector_t len, void *data)
1711 struct request_queue *q = bdev_get_queue(dev->bdev);
1713 return q && !q->limits.max_write_same_sectors;
1716 static bool dm_table_supports_write_same(struct dm_table *t)
1718 struct dm_target *ti;
1721 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1722 ti = dm_table_get_target(t, i);
1724 if (!ti->num_write_same_bios)
1727 if (!ti->type->iterate_devices ||
1728 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1735 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1736 sector_t start, sector_t len, void *data)
1738 struct request_queue *q = bdev_get_queue(dev->bdev);
1740 return q && !q->limits.max_write_zeroes_sectors;
1743 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1745 struct dm_target *ti;
1748 while (i < dm_table_get_num_targets(t)) {
1749 ti = dm_table_get_target(t, i++);
1751 if (!ti->num_write_zeroes_bios)
1754 if (!ti->type->iterate_devices ||
1755 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1762 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1763 sector_t start, sector_t len, void *data)
1765 struct request_queue *q = bdev_get_queue(dev->bdev);
1767 return q && !blk_queue_discard(q);
1770 static bool dm_table_supports_discards(struct dm_table *t)
1772 struct dm_target *ti;
1775 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1776 ti = dm_table_get_target(t, i);
1778 if (!ti->num_discard_bios)
1782 * Either the target provides discard support (as implied by setting
1783 * 'discards_supported') or it relies on _all_ data devices having
1786 if (!ti->discards_supported &&
1787 (!ti->type->iterate_devices ||
1788 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1795 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1796 struct queue_limits *limits)
1798 bool wc = false, fua = false;
1801 * Copy table's limits to the DM device's request_queue
1803 q->limits = *limits;
1805 if (!dm_table_supports_discards(t)) {
1806 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1807 /* Must also clear discard limits... */
1808 q->limits.max_discard_sectors = 0;
1809 q->limits.max_hw_discard_sectors = 0;
1810 q->limits.discard_granularity = 0;
1811 q->limits.discard_alignment = 0;
1812 q->limits.discard_misaligned = 0;
1814 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1816 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
1818 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
1821 blk_queue_write_cache(q, wc, fua);
1823 if (dm_table_supports_dax_write_cache(t))
1824 dax_write_cache(t->md->dax_dev, true);
1826 /* Ensure that all underlying devices are non-rotational. */
1827 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1828 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1830 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1832 if (!dm_table_supports_write_same(t))
1833 q->limits.max_write_same_sectors = 0;
1834 if (!dm_table_supports_write_zeroes(t))
1835 q->limits.max_write_zeroes_sectors = 0;
1837 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1838 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1840 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1842 dm_table_verify_integrity(t);
1845 * Determine whether or not this queue's I/O timings contribute
1846 * to the entropy pool, Only request-based targets use this.
1847 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1850 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1851 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1854 unsigned int dm_table_get_num_targets(struct dm_table *t)
1856 return t->num_targets;
1859 struct list_head *dm_table_get_devices(struct dm_table *t)
1864 fmode_t dm_table_get_mode(struct dm_table *t)
1868 EXPORT_SYMBOL(dm_table_get_mode);
1876 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1878 int i = t->num_targets;
1879 struct dm_target *ti = t->targets;
1881 lockdep_assert_held(&t->md->suspend_lock);
1886 if (ti->type->presuspend)
1887 ti->type->presuspend(ti);
1889 case PRESUSPEND_UNDO:
1890 if (ti->type->presuspend_undo)
1891 ti->type->presuspend_undo(ti);
1894 if (ti->type->postsuspend)
1895 ti->type->postsuspend(ti);
1902 void dm_table_presuspend_targets(struct dm_table *t)
1907 suspend_targets(t, PRESUSPEND);
1910 void dm_table_presuspend_undo_targets(struct dm_table *t)
1915 suspend_targets(t, PRESUSPEND_UNDO);
1918 void dm_table_postsuspend_targets(struct dm_table *t)
1923 suspend_targets(t, POSTSUSPEND);
1926 int dm_table_resume_targets(struct dm_table *t)
1930 lockdep_assert_held(&t->md->suspend_lock);
1932 for (i = 0; i < t->num_targets; i++) {
1933 struct dm_target *ti = t->targets + i;
1935 if (!ti->type->preresume)
1938 r = ti->type->preresume(ti);
1940 DMERR("%s: %s: preresume failed, error = %d",
1941 dm_device_name(t->md), ti->type->name, r);
1946 for (i = 0; i < t->num_targets; i++) {
1947 struct dm_target *ti = t->targets + i;
1949 if (ti->type->resume)
1950 ti->type->resume(ti);
1956 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1958 list_add(&cb->list, &t->target_callbacks);
1960 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1962 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1964 struct dm_dev_internal *dd;
1965 struct list_head *devices = dm_table_get_devices(t);
1966 struct dm_target_callbacks *cb;
1969 list_for_each_entry(dd, devices, list) {
1970 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1971 char b[BDEVNAME_SIZE];
1974 r |= bdi_congested(q->backing_dev_info, bdi_bits);
1976 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1977 dm_device_name(t->md),
1978 bdevname(dd->dm_dev->bdev, b));
1981 list_for_each_entry(cb, &t->target_callbacks, list)
1982 if (cb->congested_fn)
1983 r |= cb->congested_fn(cb, bdi_bits);
1988 struct mapped_device *dm_table_get_md(struct dm_table *t)
1992 EXPORT_SYMBOL(dm_table_get_md);
1994 void dm_table_run_md_queue_async(struct dm_table *t)
1996 struct mapped_device *md;
1997 struct request_queue *queue;
1998 unsigned long flags;
2000 if (!dm_table_request_based(t))
2003 md = dm_table_get_md(t);
2004 queue = dm_get_md_queue(md);
2007 blk_mq_run_hw_queues(queue, true);
2009 spin_lock_irqsave(queue->queue_lock, flags);
2010 blk_run_queue_async(queue);
2011 spin_unlock_irqrestore(queue->queue_lock, flags);
2015 EXPORT_SYMBOL(dm_table_run_md_queue_async);