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/mount.h>
15 #include <linux/ctype.h>
16 #include <linux/string.h>
17 #include <linux/slab.h>
18 #include <linux/interrupt.h>
19 #include <linux/mutex.h>
20 #include <linux/delay.h>
21 #include <linux/atomic.h>
22 #include <linux/blk-mq.h>
23 #include <linux/mount.h>
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;
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;
47 unsigned integrity_supported:1;
51 * Indicates the rw permissions for the new logical
52 * device. This should be a combination of FMODE_READ
57 /* a list of devices used by this table */
58 struct list_head devices;
60 /* events get handed up using this callback */
61 void (*event_fn)(void *);
64 struct dm_md_mempools *mempools;
66 struct list_head target_callbacks;
70 * Similar to ceiling(log_size(n))
72 static unsigned int int_log(unsigned int n, unsigned int base)
77 n = dm_div_up(n, base);
85 * Calculate the index of the child node of the n'th node k'th key.
87 static inline unsigned int get_child(unsigned int n, unsigned int k)
89 return (n * CHILDREN_PER_NODE) + k;
93 * Return the n'th node of level l from table t.
95 static inline sector_t *get_node(struct dm_table *t,
96 unsigned int l, unsigned int n)
98 return t->index[l] + (n * KEYS_PER_NODE);
102 * Return the highest key that you could lookup from the n'th
103 * node on level l of the btree.
105 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
107 for (; l < t->depth - 1; l++)
108 n = get_child(n, CHILDREN_PER_NODE - 1);
110 if (n >= t->counts[l])
111 return (sector_t) - 1;
113 return get_node(t, l, n)[KEYS_PER_NODE - 1];
117 * Fills in a level of the btree based on the highs of the level
120 static int setup_btree_index(unsigned int l, struct dm_table *t)
125 for (n = 0U; n < t->counts[l]; n++) {
126 node = get_node(t, l, n);
128 for (k = 0U; k < KEYS_PER_NODE; k++)
129 node[k] = high(t, l + 1, get_child(n, k));
135 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
141 * Check that we're not going to overflow.
143 if (nmemb > (ULONG_MAX / elem_size))
146 size = nmemb * elem_size;
147 addr = vzalloc(size);
151 EXPORT_SYMBOL(dm_vcalloc);
154 * highs, and targets are managed as dynamic arrays during a
157 static int alloc_targets(struct dm_table *t, unsigned int num)
160 struct dm_target *n_targets;
163 * Allocate both the target array and offset array at once.
164 * Append an empty entry to catch sectors beyond the end of
167 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
172 n_targets = (struct dm_target *) (n_highs + num);
174 memset(n_highs, -1, sizeof(*n_highs) * num);
177 t->num_allocated = num;
179 t->targets = n_targets;
184 int dm_table_create(struct dm_table **result, fmode_t mode,
185 unsigned num_targets, struct mapped_device *md)
187 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
192 INIT_LIST_HEAD(&t->devices);
193 INIT_LIST_HEAD(&t->target_callbacks);
196 num_targets = KEYS_PER_NODE;
198 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
205 if (alloc_targets(t, num_targets)) {
216 static void free_devices(struct list_head *devices, struct mapped_device *md)
218 struct list_head *tmp, *next;
220 list_for_each_safe(tmp, next, devices) {
221 struct dm_dev_internal *dd =
222 list_entry(tmp, struct dm_dev_internal, list);
223 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
224 dm_device_name(md), dd->dm_dev->name);
225 dm_put_table_device(md, dd->dm_dev);
230 void dm_table_destroy(struct dm_table *t)
237 /* free the indexes */
239 vfree(t->index[t->depth - 2]);
241 /* free the targets */
242 for (i = 0; i < t->num_targets; i++) {
243 struct dm_target *tgt = t->targets + i;
248 dm_put_target_type(tgt->type);
253 /* free the device list */
254 free_devices(&t->devices, t->md);
256 dm_free_md_mempools(t->mempools);
262 * See if we've already got a device in the list.
264 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
266 struct dm_dev_internal *dd;
268 list_for_each_entry (dd, l, list)
269 if (dd->dm_dev->bdev->bd_dev == dev)
276 * If possible, this checks an area of a destination device is invalid.
278 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
279 sector_t start, sector_t len, void *data)
281 struct request_queue *q;
282 struct queue_limits *limits = data;
283 struct block_device *bdev = dev->bdev;
285 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
286 unsigned short logical_block_size_sectors =
287 limits->logical_block_size >> SECTOR_SHIFT;
288 char b[BDEVNAME_SIZE];
291 * Some devices exist without request functions,
292 * such as loop devices not yet bound to backing files.
293 * Forbid the use of such devices.
295 q = bdev_get_queue(bdev);
296 if (!q || !q->make_request_fn) {
297 DMWARN("%s: %s is not yet initialised: "
298 "start=%llu, len=%llu, dev_size=%llu",
299 dm_device_name(ti->table->md), bdevname(bdev, b),
300 (unsigned long long)start,
301 (unsigned long long)len,
302 (unsigned long long)dev_size);
309 if ((start >= dev_size) || (start + len > dev_size)) {
310 DMWARN("%s: %s too small for target: "
311 "start=%llu, len=%llu, dev_size=%llu",
312 dm_device_name(ti->table->md), bdevname(bdev, b),
313 (unsigned long long)start,
314 (unsigned long long)len,
315 (unsigned long long)dev_size);
319 if (logical_block_size_sectors <= 1)
322 if (start & (logical_block_size_sectors - 1)) {
323 DMWARN("%s: start=%llu not aligned to h/w "
324 "logical block size %u of %s",
325 dm_device_name(ti->table->md),
326 (unsigned long long)start,
327 limits->logical_block_size, bdevname(bdev, b));
331 if (len & (logical_block_size_sectors - 1)) {
332 DMWARN("%s: len=%llu not aligned to h/w "
333 "logical block size %u of %s",
334 dm_device_name(ti->table->md),
335 (unsigned long long)len,
336 limits->logical_block_size, bdevname(bdev, b));
344 * This upgrades the mode on an already open dm_dev, being
345 * careful to leave things as they were if we fail to reopen the
346 * device and not to touch the existing bdev field in case
347 * it is accessed concurrently inside dm_table_any_congested().
349 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
350 struct mapped_device *md)
353 struct dm_dev *old_dev, *new_dev;
355 old_dev = dd->dm_dev;
357 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
358 dd->dm_dev->mode | new_mode, &new_dev);
362 dd->dm_dev = new_dev;
363 dm_put_table_device(md, old_dev);
369 * Convert the path to a device
371 dev_t dm_get_dev_t(const char *path)
373 dev_t uninitialized_var(dev);
374 struct block_device *bdev;
376 bdev = lookup_bdev(path);
378 dev = name_to_dev_t(path);
386 EXPORT_SYMBOL_GPL(dm_get_dev_t);
389 * Add a device to the list, or just increment the usage count if
390 * it's already present.
392 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
393 struct dm_dev **result)
397 struct dm_dev_internal *dd;
398 struct dm_table *t = ti->table;
402 dev = dm_get_dev_t(path);
406 dd = find_device(&t->devices, dev);
408 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
412 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
417 atomic_set(&dd->count, 0);
418 list_add(&dd->list, &t->devices);
420 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
421 r = upgrade_mode(dd, mode, t->md);
425 atomic_inc(&dd->count);
427 *result = dd->dm_dev;
430 EXPORT_SYMBOL(dm_get_device);
432 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
433 sector_t start, sector_t len, void *data)
435 struct queue_limits *limits = data;
436 struct block_device *bdev = dev->bdev;
437 struct request_queue *q = bdev_get_queue(bdev);
438 char b[BDEVNAME_SIZE];
441 DMWARN("%s: Cannot set limits for nonexistent device %s",
442 dm_device_name(ti->table->md), bdevname(bdev, b));
446 if (bdev_stack_limits(limits, bdev, start) < 0)
447 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
448 "physical_block_size=%u, logical_block_size=%u, "
449 "alignment_offset=%u, start=%llu",
450 dm_device_name(ti->table->md), bdevname(bdev, b),
451 q->limits.physical_block_size,
452 q->limits.logical_block_size,
453 q->limits.alignment_offset,
454 (unsigned long long) start << SECTOR_SHIFT);
460 * Decrement a device's use count and remove it if necessary.
462 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
465 struct list_head *devices = &ti->table->devices;
466 struct dm_dev_internal *dd;
468 list_for_each_entry(dd, devices, list) {
469 if (dd->dm_dev == d) {
475 DMWARN("%s: device %s not in table devices list",
476 dm_device_name(ti->table->md), d->name);
479 if (atomic_dec_and_test(&dd->count)) {
480 dm_put_table_device(ti->table->md, d);
485 EXPORT_SYMBOL(dm_put_device);
488 * Checks to see if the target joins onto the end of the table.
490 static int adjoin(struct dm_table *table, struct dm_target *ti)
492 struct dm_target *prev;
494 if (!table->num_targets)
497 prev = &table->targets[table->num_targets - 1];
498 return (ti->begin == (prev->begin + prev->len));
502 * Used to dynamically allocate the arg array.
504 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
505 * process messages even if some device is suspended. These messages have a
506 * small fixed number of arguments.
508 * On the other hand, dm-switch needs to process bulk data using messages and
509 * excessive use of GFP_NOIO could cause trouble.
511 static char **realloc_argv(unsigned *array_size, char **old_argv)
518 new_size = *array_size * 2;
524 argv = kmalloc(new_size * sizeof(*argv), gfp);
526 memcpy(argv, old_argv, *array_size * sizeof(*argv));
527 *array_size = new_size;
535 * Destructively splits up the argument list to pass to ctr.
537 int dm_split_args(int *argc, char ***argvp, char *input)
539 char *start, *end = input, *out, **argv = NULL;
540 unsigned array_size = 0;
549 argv = realloc_argv(&array_size, argv);
554 /* Skip whitespace */
555 start = skip_spaces(end);
558 break; /* success, we hit the end */
560 /* 'out' is used to remove any back-quotes */
563 /* Everything apart from '\0' can be quoted */
564 if (*end == '\\' && *(end + 1)) {
571 break; /* end of token */
576 /* have we already filled the array ? */
577 if ((*argc + 1) > array_size) {
578 argv = realloc_argv(&array_size, argv);
583 /* we know this is whitespace */
587 /* terminate the string and put it in the array */
598 * Impose necessary and sufficient conditions on a devices's table such
599 * that any incoming bio which respects its logical_block_size can be
600 * processed successfully. If it falls across the boundary between
601 * two or more targets, the size of each piece it gets split into must
602 * be compatible with the logical_block_size of the target processing it.
604 static int validate_hardware_logical_block_alignment(struct dm_table *table,
605 struct queue_limits *limits)
608 * This function uses arithmetic modulo the logical_block_size
609 * (in units of 512-byte sectors).
611 unsigned short device_logical_block_size_sects =
612 limits->logical_block_size >> SECTOR_SHIFT;
615 * Offset of the start of the next table entry, mod logical_block_size.
617 unsigned short next_target_start = 0;
620 * Given an aligned bio that extends beyond the end of a
621 * target, how many sectors must the next target handle?
623 unsigned short remaining = 0;
625 struct dm_target *uninitialized_var(ti);
626 struct queue_limits ti_limits;
630 * Check each entry in the table in turn.
632 while (i < dm_table_get_num_targets(table)) {
633 ti = dm_table_get_target(table, i++);
635 blk_set_stacking_limits(&ti_limits);
637 /* combine all target devices' limits */
638 if (ti->type->iterate_devices)
639 ti->type->iterate_devices(ti, dm_set_device_limits,
643 * If the remaining sectors fall entirely within this
644 * table entry are they compatible with its logical_block_size?
646 if (remaining < ti->len &&
647 remaining & ((ti_limits.logical_block_size >>
652 (unsigned short) ((next_target_start + ti->len) &
653 (device_logical_block_size_sects - 1));
654 remaining = next_target_start ?
655 device_logical_block_size_sects - next_target_start : 0;
659 DMWARN("%s: table line %u (start sect %llu len %llu) "
660 "not aligned to h/w logical block size %u",
661 dm_device_name(table->md), i,
662 (unsigned long long) ti->begin,
663 (unsigned long long) ti->len,
664 limits->logical_block_size);
671 int dm_table_add_target(struct dm_table *t, const char *type,
672 sector_t start, sector_t len, char *params)
674 int r = -EINVAL, argc;
676 struct dm_target *tgt;
679 DMERR("%s: target type %s must appear alone in table",
680 dm_device_name(t->md), t->targets->type->name);
684 BUG_ON(t->num_targets >= t->num_allocated);
686 tgt = t->targets + t->num_targets;
687 memset(tgt, 0, sizeof(*tgt));
690 DMERR("%s: zero-length target", dm_device_name(t->md));
694 tgt->type = dm_get_target_type(type);
696 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
701 if (dm_target_needs_singleton(tgt->type)) {
702 if (t->num_targets) {
703 DMERR("%s: target type %s must appear alone in table",
704 dm_device_name(t->md), type);
710 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
711 DMERR("%s: target type %s may not be included in read-only tables",
712 dm_device_name(t->md), type);
716 if (t->immutable_target_type) {
717 if (t->immutable_target_type != tgt->type) {
718 DMERR("%s: immutable target type %s cannot be mixed with other target types",
719 dm_device_name(t->md), t->immutable_target_type->name);
722 } else if (dm_target_is_immutable(tgt->type)) {
723 if (t->num_targets) {
724 DMERR("%s: immutable target type %s cannot be mixed with other target types",
725 dm_device_name(t->md), tgt->type->name);
728 t->immutable_target_type = tgt->type;
734 tgt->error = "Unknown error";
737 * Does this target adjoin the previous one ?
739 if (!adjoin(t, tgt)) {
740 tgt->error = "Gap in table";
745 r = dm_split_args(&argc, &argv, params);
747 tgt->error = "couldn't split parameters (insufficient memory)";
751 r = tgt->type->ctr(tgt, argc, argv);
756 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
758 if (!tgt->num_discard_bios && tgt->discards_supported)
759 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
760 dm_device_name(t->md), type);
765 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
766 dm_put_target_type(tgt->type);
771 * Target argument parsing helpers.
773 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
774 unsigned *value, char **error, unsigned grouped)
776 const char *arg_str = dm_shift_arg(arg_set);
780 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
781 (*value < arg->min) ||
782 (*value > arg->max) ||
783 (grouped && arg_set->argc < *value)) {
791 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
792 unsigned *value, char **error)
794 return validate_next_arg(arg, arg_set, value, error, 0);
796 EXPORT_SYMBOL(dm_read_arg);
798 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
799 unsigned *value, char **error)
801 return validate_next_arg(arg, arg_set, value, error, 1);
803 EXPORT_SYMBOL(dm_read_arg_group);
805 const char *dm_shift_arg(struct dm_arg_set *as)
818 EXPORT_SYMBOL(dm_shift_arg);
820 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
822 BUG_ON(as->argc < num_args);
823 as->argc -= num_args;
824 as->argv += num_args;
826 EXPORT_SYMBOL(dm_consume_args);
828 static bool __table_type_request_based(unsigned table_type)
830 return (table_type == DM_TYPE_REQUEST_BASED ||
831 table_type == DM_TYPE_MQ_REQUEST_BASED);
834 static int dm_table_set_type(struct dm_table *t)
837 unsigned bio_based = 0, request_based = 0, hybrid = 0;
838 bool use_blk_mq = false;
839 struct dm_target *tgt;
840 struct dm_dev_internal *dd;
841 struct list_head *devices;
842 unsigned live_md_type = dm_get_md_type(t->md);
844 for (i = 0; i < t->num_targets; i++) {
845 tgt = t->targets + i;
846 if (dm_target_hybrid(tgt))
848 else if (dm_target_request_based(tgt))
853 if (bio_based && request_based) {
854 DMWARN("Inconsistent table: different target types"
855 " can't be mixed up");
860 if (hybrid && !bio_based && !request_based) {
862 * The targets can work either way.
863 * Determine the type from the live device.
864 * Default to bio-based if device is new.
866 if (__table_type_request_based(live_md_type))
873 /* We must use this table as bio-based */
874 t->type = DM_TYPE_BIO_BASED;
878 BUG_ON(!request_based); /* No targets in this table */
881 * Request-based dm supports only tables that have a single target now.
882 * To support multiple targets, request splitting support is needed,
883 * and that needs lots of changes in the block-layer.
884 * (e.g. request completion process for partial completion.)
886 if (t->num_targets > 1) {
887 DMWARN("Request-based dm doesn't support multiple targets yet");
891 /* Non-request-stackable devices can't be used for request-based dm */
892 devices = dm_table_get_devices(t);
893 list_for_each_entry(dd, devices, list) {
894 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
896 if (!blk_queue_stackable(q)) {
897 DMERR("table load rejected: including"
898 " non-request-stackable devices");
907 /* verify _all_ devices in the table are blk-mq devices */
908 list_for_each_entry(dd, devices, list)
909 if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
910 DMERR("table load rejected: not all devices"
911 " are blk-mq request-stackable");
914 t->type = DM_TYPE_MQ_REQUEST_BASED;
916 } else if (list_empty(devices) && __table_type_request_based(live_md_type)) {
917 /* inherit live MD type */
918 t->type = live_md_type;
921 t->type = DM_TYPE_REQUEST_BASED;
926 unsigned dm_table_get_type(struct dm_table *t)
931 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
933 return t->immutable_target_type;
936 bool dm_table_request_based(struct dm_table *t)
938 return __table_type_request_based(dm_table_get_type(t));
941 bool dm_table_mq_request_based(struct dm_table *t)
943 return dm_table_get_type(t) == DM_TYPE_MQ_REQUEST_BASED;
946 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
948 unsigned type = dm_table_get_type(t);
949 unsigned per_bio_data_size = 0;
950 struct dm_target *tgt;
953 if (unlikely(type == DM_TYPE_NONE)) {
954 DMWARN("no table type is set, can't allocate mempools");
958 if (type == DM_TYPE_BIO_BASED)
959 for (i = 0; i < t->num_targets; i++) {
960 tgt = t->targets + i;
961 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
964 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_bio_data_size);
971 void dm_table_free_md_mempools(struct dm_table *t)
973 dm_free_md_mempools(t->mempools);
977 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
982 static int setup_indexes(struct dm_table *t)
985 unsigned int total = 0;
988 /* allocate the space for *all* the indexes */
989 for (i = t->depth - 2; i >= 0; i--) {
990 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
991 total += t->counts[i];
994 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
998 /* set up internal nodes, bottom-up */
999 for (i = t->depth - 2; i >= 0; i--) {
1000 t->index[i] = indexes;
1001 indexes += (KEYS_PER_NODE * t->counts[i]);
1002 setup_btree_index(i, t);
1009 * Builds the btree to index the map.
1011 static int dm_table_build_index(struct dm_table *t)
1014 unsigned int leaf_nodes;
1016 /* how many indexes will the btree have ? */
1017 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1018 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1020 /* leaf layer has already been set up */
1021 t->counts[t->depth - 1] = leaf_nodes;
1022 t->index[t->depth - 1] = t->highs;
1025 r = setup_indexes(t);
1030 static bool integrity_profile_exists(struct gendisk *disk)
1032 return !!blk_get_integrity(disk);
1036 * Get a disk whose integrity profile reflects the table's profile.
1037 * Returns NULL if integrity support was inconsistent or unavailable.
1039 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1041 struct list_head *devices = dm_table_get_devices(t);
1042 struct dm_dev_internal *dd = NULL;
1043 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1045 list_for_each_entry(dd, devices, list) {
1046 template_disk = dd->dm_dev->bdev->bd_disk;
1047 if (!integrity_profile_exists(template_disk))
1049 else if (prev_disk &&
1050 blk_integrity_compare(prev_disk, template_disk) < 0)
1052 prev_disk = template_disk;
1055 return template_disk;
1059 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1060 dm_device_name(t->md),
1061 prev_disk->disk_name,
1062 template_disk->disk_name);
1067 * Register the mapped device for blk_integrity support if the
1068 * underlying devices have an integrity profile. But all devices may
1069 * not have matching profiles (checking all devices isn't reliable
1070 * during table load because this table may use other DM device(s) which
1071 * must be resumed before they will have an initialized integity
1072 * profile). Consequently, stacked DM devices force a 2 stage integrity
1073 * profile validation: First pass during table load, final pass during
1076 static int dm_table_register_integrity(struct dm_table *t)
1078 struct mapped_device *md = t->md;
1079 struct gendisk *template_disk = NULL;
1081 template_disk = dm_table_get_integrity_disk(t);
1085 if (!integrity_profile_exists(dm_disk(md))) {
1086 t->integrity_supported = 1;
1088 * Register integrity profile during table load; we can do
1089 * this because the final profile must match during resume.
1091 blk_integrity_register(dm_disk(md),
1092 blk_get_integrity(template_disk));
1097 * If DM device already has an initialized integrity
1098 * profile the new profile should not conflict.
1100 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1101 DMWARN("%s: conflict with existing integrity profile: "
1102 "%s profile mismatch",
1103 dm_device_name(t->md),
1104 template_disk->disk_name);
1108 /* Preserve existing integrity profile */
1109 t->integrity_supported = 1;
1114 * Prepares the table for use by building the indices,
1115 * setting the type, and allocating mempools.
1117 int dm_table_complete(struct dm_table *t)
1121 r = dm_table_set_type(t);
1123 DMERR("unable to set table type");
1127 r = dm_table_build_index(t);
1129 DMERR("unable to build btrees");
1133 r = dm_table_register_integrity(t);
1135 DMERR("could not register integrity profile.");
1139 r = dm_table_alloc_md_mempools(t, t->md);
1141 DMERR("unable to allocate mempools");
1146 static DEFINE_MUTEX(_event_lock);
1147 void dm_table_event_callback(struct dm_table *t,
1148 void (*fn)(void *), void *context)
1150 mutex_lock(&_event_lock);
1152 t->event_context = context;
1153 mutex_unlock(&_event_lock);
1156 void dm_table_event(struct dm_table *t)
1159 * You can no longer call dm_table_event() from interrupt
1160 * context, use a bottom half instead.
1162 BUG_ON(in_interrupt());
1164 mutex_lock(&_event_lock);
1166 t->event_fn(t->event_context);
1167 mutex_unlock(&_event_lock);
1169 EXPORT_SYMBOL(dm_table_event);
1171 sector_t dm_table_get_size(struct dm_table *t)
1173 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1175 EXPORT_SYMBOL(dm_table_get_size);
1177 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1179 if (index >= t->num_targets)
1182 return t->targets + index;
1186 * Search the btree for the correct target.
1188 * Caller should check returned pointer with dm_target_is_valid()
1189 * to trap I/O beyond end of device.
1191 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1193 unsigned int l, n = 0, k = 0;
1196 for (l = 0; l < t->depth; l++) {
1197 n = get_child(n, k);
1198 node = get_node(t, l, n);
1200 for (k = 0; k < KEYS_PER_NODE; k++)
1201 if (node[k] >= sector)
1205 return &t->targets[(KEYS_PER_NODE * n) + k];
1208 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1209 sector_t start, sector_t len, void *data)
1211 unsigned *num_devices = data;
1219 * Check whether a table has no data devices attached using each
1220 * target's iterate_devices method.
1221 * Returns false if the result is unknown because a target doesn't
1222 * support iterate_devices.
1224 bool dm_table_has_no_data_devices(struct dm_table *table)
1226 struct dm_target *uninitialized_var(ti);
1227 unsigned i = 0, num_devices = 0;
1229 while (i < dm_table_get_num_targets(table)) {
1230 ti = dm_table_get_target(table, i++);
1232 if (!ti->type->iterate_devices)
1235 ti->type->iterate_devices(ti, count_device, &num_devices);
1244 * Establish the new table's queue_limits and validate them.
1246 int dm_calculate_queue_limits(struct dm_table *table,
1247 struct queue_limits *limits)
1249 struct dm_target *uninitialized_var(ti);
1250 struct queue_limits ti_limits;
1253 blk_set_stacking_limits(limits);
1255 while (i < dm_table_get_num_targets(table)) {
1256 blk_set_stacking_limits(&ti_limits);
1258 ti = dm_table_get_target(table, i++);
1260 if (!ti->type->iterate_devices)
1261 goto combine_limits;
1264 * Combine queue limits of all the devices this target uses.
1266 ti->type->iterate_devices(ti, dm_set_device_limits,
1269 /* Set I/O hints portion of queue limits */
1270 if (ti->type->io_hints)
1271 ti->type->io_hints(ti, &ti_limits);
1274 * Check each device area is consistent with the target's
1275 * overall queue limits.
1277 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1283 * Merge this target's queue limits into the overall limits
1286 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1287 DMWARN("%s: adding target device "
1288 "(start sect %llu len %llu) "
1289 "caused an alignment inconsistency",
1290 dm_device_name(table->md),
1291 (unsigned long long) ti->begin,
1292 (unsigned long long) ti->len);
1295 return validate_hardware_logical_block_alignment(table, limits);
1299 * Verify that all devices have an integrity profile that matches the
1300 * DM device's registered integrity profile. If the profiles don't
1301 * match then unregister the DM device's integrity profile.
1303 static void dm_table_verify_integrity(struct dm_table *t)
1305 struct gendisk *template_disk = NULL;
1307 if (t->integrity_supported) {
1309 * Verify that the original integrity profile
1310 * matches all the devices in this table.
1312 template_disk = dm_table_get_integrity_disk(t);
1313 if (template_disk &&
1314 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1318 if (integrity_profile_exists(dm_disk(t->md))) {
1319 DMWARN("%s: unable to establish an integrity profile",
1320 dm_device_name(t->md));
1321 blk_integrity_unregister(dm_disk(t->md));
1325 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1326 sector_t start, sector_t len, void *data)
1328 unsigned flush = (*(unsigned *)data);
1329 struct request_queue *q = bdev_get_queue(dev->bdev);
1331 return q && (q->flush_flags & flush);
1334 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1336 struct dm_target *ti;
1340 * Require at least one underlying device to support flushes.
1341 * t->devices includes internal dm devices such as mirror logs
1342 * so we need to use iterate_devices here, which targets
1343 * supporting flushes must provide.
1345 while (i < dm_table_get_num_targets(t)) {
1346 ti = dm_table_get_target(t, i++);
1348 if (!ti->num_flush_bios)
1351 if (ti->flush_supported)
1354 if (ti->type->iterate_devices &&
1355 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1362 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1364 struct dm_target *ti;
1367 /* Ensure that all targets supports discard_zeroes_data. */
1368 while (i < dm_table_get_num_targets(t)) {
1369 ti = dm_table_get_target(t, i++);
1371 if (ti->discard_zeroes_data_unsupported)
1378 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1379 sector_t start, sector_t len, void *data)
1381 struct request_queue *q = bdev_get_queue(dev->bdev);
1383 return q && blk_queue_nonrot(q);
1386 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1387 sector_t start, sector_t len, void *data)
1389 struct request_queue *q = bdev_get_queue(dev->bdev);
1391 return q && !blk_queue_add_random(q);
1394 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1395 sector_t start, sector_t len, void *data)
1397 struct request_queue *q = bdev_get_queue(dev->bdev);
1399 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1402 static bool dm_table_all_devices_attribute(struct dm_table *t,
1403 iterate_devices_callout_fn func)
1405 struct dm_target *ti;
1408 while (i < dm_table_get_num_targets(t)) {
1409 ti = dm_table_get_target(t, i++);
1411 if (!ti->type->iterate_devices ||
1412 !ti->type->iterate_devices(ti, func, NULL))
1419 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1420 sector_t start, sector_t len, void *data)
1422 struct request_queue *q = bdev_get_queue(dev->bdev);
1424 return q && !q->limits.max_write_same_sectors;
1427 static bool dm_table_supports_write_same(struct dm_table *t)
1429 struct dm_target *ti;
1432 while (i < dm_table_get_num_targets(t)) {
1433 ti = dm_table_get_target(t, i++);
1435 if (!ti->num_write_same_bios)
1438 if (!ti->type->iterate_devices ||
1439 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1446 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1447 sector_t start, sector_t len, void *data)
1449 struct request_queue *q = bdev_get_queue(dev->bdev);
1451 return q && blk_queue_discard(q);
1454 static bool dm_table_supports_discards(struct dm_table *t)
1456 struct dm_target *ti;
1460 * Unless any target used by the table set discards_supported,
1461 * require at least one underlying device to support discards.
1462 * t->devices includes internal dm devices such as mirror logs
1463 * so we need to use iterate_devices here, which targets
1464 * supporting discard selectively must provide.
1466 while (i < dm_table_get_num_targets(t)) {
1467 ti = dm_table_get_target(t, i++);
1469 if (!ti->num_discard_bios)
1472 if (ti->discards_supported)
1475 if (ti->type->iterate_devices &&
1476 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1483 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1484 struct queue_limits *limits)
1489 * Copy table's limits to the DM device's request_queue
1491 q->limits = *limits;
1493 if (!dm_table_supports_discards(t))
1494 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1496 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1498 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1500 if (dm_table_supports_flush(t, REQ_FUA))
1503 blk_queue_flush(q, flush);
1505 if (!dm_table_discard_zeroes_data(t))
1506 q->limits.discard_zeroes_data = 0;
1508 /* Ensure that all underlying devices are non-rotational. */
1509 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1510 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1512 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1514 if (!dm_table_supports_write_same(t))
1515 q->limits.max_write_same_sectors = 0;
1517 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1518 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1520 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1522 dm_table_verify_integrity(t);
1525 * Determine whether or not this queue's I/O timings contribute
1526 * to the entropy pool, Only request-based targets use this.
1527 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1530 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1531 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1534 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1535 * visible to other CPUs because, once the flag is set, incoming bios
1536 * are processed by request-based dm, which refers to the queue
1538 * Until the flag set, bios are passed to bio-based dm and queued to
1539 * md->deferred where queue settings are not needed yet.
1540 * Those bios are passed to request-based dm at the resume time.
1543 if (dm_table_request_based(t))
1544 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1547 unsigned int dm_table_get_num_targets(struct dm_table *t)
1549 return t->num_targets;
1552 struct list_head *dm_table_get_devices(struct dm_table *t)
1557 fmode_t dm_table_get_mode(struct dm_table *t)
1561 EXPORT_SYMBOL(dm_table_get_mode);
1569 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1571 int i = t->num_targets;
1572 struct dm_target *ti = t->targets;
1577 if (ti->type->presuspend)
1578 ti->type->presuspend(ti);
1580 case PRESUSPEND_UNDO:
1581 if (ti->type->presuspend_undo)
1582 ti->type->presuspend_undo(ti);
1585 if (ti->type->postsuspend)
1586 ti->type->postsuspend(ti);
1593 void dm_table_presuspend_targets(struct dm_table *t)
1598 suspend_targets(t, PRESUSPEND);
1601 void dm_table_presuspend_undo_targets(struct dm_table *t)
1606 suspend_targets(t, PRESUSPEND_UNDO);
1609 void dm_table_postsuspend_targets(struct dm_table *t)
1614 suspend_targets(t, POSTSUSPEND);
1617 int dm_table_resume_targets(struct dm_table *t)
1621 for (i = 0; i < t->num_targets; i++) {
1622 struct dm_target *ti = t->targets + i;
1624 if (!ti->type->preresume)
1627 r = ti->type->preresume(ti);
1629 DMERR("%s: %s: preresume failed, error = %d",
1630 dm_device_name(t->md), ti->type->name, r);
1635 for (i = 0; i < t->num_targets; i++) {
1636 struct dm_target *ti = t->targets + i;
1638 if (ti->type->resume)
1639 ti->type->resume(ti);
1645 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1647 list_add(&cb->list, &t->target_callbacks);
1649 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1651 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1653 struct dm_dev_internal *dd;
1654 struct list_head *devices = dm_table_get_devices(t);
1655 struct dm_target_callbacks *cb;
1658 list_for_each_entry(dd, devices, list) {
1659 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1660 char b[BDEVNAME_SIZE];
1663 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1665 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1666 dm_device_name(t->md),
1667 bdevname(dd->dm_dev->bdev, b));
1670 list_for_each_entry(cb, &t->target_callbacks, list)
1671 if (cb->congested_fn)
1672 r |= cb->congested_fn(cb, bdi_bits);
1677 struct mapped_device *dm_table_get_md(struct dm_table *t)
1681 EXPORT_SYMBOL(dm_table_get_md);
1683 void dm_table_run_md_queue_async(struct dm_table *t)
1685 struct mapped_device *md;
1686 struct request_queue *queue;
1687 unsigned long flags;
1689 if (!dm_table_request_based(t))
1692 md = dm_table_get_md(t);
1693 queue = dm_get_md_queue(md);
1696 blk_mq_run_hw_queues(queue, true);
1698 spin_lock_irqsave(queue->queue_lock, flags);
1699 blk_run_queue_async(queue);
1700 spin_unlock_irqrestore(queue->queue_lock, flags);
1704 EXPORT_SYMBOL(dm_table_run_md_queue_async);