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>
24 #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 struct mapped_device *md;
37 unsigned int counts[MAX_DEPTH]; /* in nodes */
38 sector_t *index[MAX_DEPTH];
40 unsigned int num_targets;
41 unsigned int num_allocated;
43 struct dm_target *targets;
45 struct target_type *immutable_target_type;
46 unsigned integrity_supported:1;
50 * Indicates the rw permissions for the new logical
51 * device. This should be a combination of FMODE_READ
56 /* a list of devices used by this table */
57 struct list_head devices;
59 /* events get handed up using this callback */
60 void (*event_fn)(void *);
63 struct dm_md_mempools *mempools;
65 struct list_head target_callbacks;
69 * Similar to ceiling(log_size(n))
71 static unsigned int int_log(unsigned int n, unsigned int base)
76 n = dm_div_up(n, base);
84 * Calculate the index of the child node of the n'th node k'th key.
86 static inline unsigned int get_child(unsigned int n, unsigned int k)
88 return (n * CHILDREN_PER_NODE) + k;
92 * Return the n'th node of level l from table t.
94 static inline sector_t *get_node(struct dm_table *t,
95 unsigned int l, unsigned int n)
97 return t->index[l] + (n * KEYS_PER_NODE);
101 * Return the highest key that you could lookup from the n'th
102 * node on level l of the btree.
104 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
106 for (; l < t->depth - 1; l++)
107 n = get_child(n, CHILDREN_PER_NODE - 1);
109 if (n >= t->counts[l])
110 return (sector_t) - 1;
112 return get_node(t, l, n)[KEYS_PER_NODE - 1];
116 * Fills in a level of the btree based on the highs of the level
119 static int setup_btree_index(unsigned int l, struct dm_table *t)
124 for (n = 0U; n < t->counts[l]; n++) {
125 node = get_node(t, l, n);
127 for (k = 0U; k < KEYS_PER_NODE; k++)
128 node[k] = high(t, l + 1, get_child(n, k));
134 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
140 * Check that we're not going to overflow.
142 if (nmemb > (ULONG_MAX / elem_size))
145 size = nmemb * elem_size;
146 addr = vzalloc(size);
150 EXPORT_SYMBOL(dm_vcalloc);
153 * highs, and targets are managed as dynamic arrays during a
156 static int alloc_targets(struct dm_table *t, unsigned int num)
159 struct dm_target *n_targets;
162 * Allocate both the target array and offset array at once.
163 * Append an empty entry to catch sectors beyond the end of
166 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
171 n_targets = (struct dm_target *) (n_highs + num);
173 memset(n_highs, -1, sizeof(*n_highs) * num);
176 t->num_allocated = num;
178 t->targets = n_targets;
183 int dm_table_create(struct dm_table **result, fmode_t mode,
184 unsigned num_targets, struct mapped_device *md)
186 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
191 INIT_LIST_HEAD(&t->devices);
192 INIT_LIST_HEAD(&t->target_callbacks);
195 num_targets = KEYS_PER_NODE;
197 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
204 if (alloc_targets(t, num_targets)) {
215 static void free_devices(struct list_head *devices, struct mapped_device *md)
217 struct list_head *tmp, *next;
219 list_for_each_safe(tmp, next, devices) {
220 struct dm_dev_internal *dd =
221 list_entry(tmp, struct dm_dev_internal, list);
222 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
223 dm_device_name(md), dd->dm_dev->name);
224 dm_put_table_device(md, dd->dm_dev);
229 void dm_table_destroy(struct dm_table *t)
236 /* free the indexes */
238 vfree(t->index[t->depth - 2]);
240 /* free the targets */
241 for (i = 0; i < t->num_targets; i++) {
242 struct dm_target *tgt = t->targets + i;
247 dm_put_target_type(tgt->type);
252 /* free the device list */
253 free_devices(&t->devices, t->md);
255 dm_free_md_mempools(t->mempools);
261 * See if we've already got a device in the list.
263 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
265 struct dm_dev_internal *dd;
267 list_for_each_entry (dd, l, list)
268 if (dd->dm_dev->bdev->bd_dev == dev)
275 * If possible, this checks an area of a destination device is invalid.
277 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
278 sector_t start, sector_t len, void *data)
280 struct request_queue *q;
281 struct queue_limits *limits = data;
282 struct block_device *bdev = dev->bdev;
284 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
285 unsigned short logical_block_size_sectors =
286 limits->logical_block_size >> SECTOR_SHIFT;
287 char b[BDEVNAME_SIZE];
290 * Some devices exist without request functions,
291 * such as loop devices not yet bound to backing files.
292 * Forbid the use of such devices.
294 q = bdev_get_queue(bdev);
295 if (!q || !q->make_request_fn) {
296 DMWARN("%s: %s is not yet initialised: "
297 "start=%llu, len=%llu, dev_size=%llu",
298 dm_device_name(ti->table->md), bdevname(bdev, b),
299 (unsigned long long)start,
300 (unsigned long long)len,
301 (unsigned long long)dev_size);
308 if ((start >= dev_size) || (start + len > dev_size)) {
309 DMWARN("%s: %s too small for target: "
310 "start=%llu, len=%llu, dev_size=%llu",
311 dm_device_name(ti->table->md), bdevname(bdev, b),
312 (unsigned long long)start,
313 (unsigned long long)len,
314 (unsigned long long)dev_size);
318 if (logical_block_size_sectors <= 1)
321 if (start & (logical_block_size_sectors - 1)) {
322 DMWARN("%s: start=%llu not aligned to h/w "
323 "logical block size %u of %s",
324 dm_device_name(ti->table->md),
325 (unsigned long long)start,
326 limits->logical_block_size, bdevname(bdev, b));
330 if (len & (logical_block_size_sectors - 1)) {
331 DMWARN("%s: len=%llu not aligned to h/w "
332 "logical block size %u of %s",
333 dm_device_name(ti->table->md),
334 (unsigned long long)len,
335 limits->logical_block_size, bdevname(bdev, b));
343 * This upgrades the mode on an already open dm_dev, being
344 * careful to leave things as they were if we fail to reopen the
345 * device and not to touch the existing bdev field in case
346 * it is accessed concurrently inside dm_table_any_congested().
348 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
349 struct mapped_device *md)
352 struct dm_dev *old_dev, *new_dev;
354 old_dev = dd->dm_dev;
356 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
357 dd->dm_dev->mode | new_mode, &new_dev);
361 dd->dm_dev = new_dev;
362 dm_put_table_device(md, old_dev);
368 * Convert the path to a device
370 dev_t dm_get_dev_t(const char *path)
372 dev_t uninitialized_var(dev);
373 struct block_device *bdev;
375 bdev = lookup_bdev(path);
377 dev = name_to_dev_t(path);
385 EXPORT_SYMBOL_GPL(dm_get_dev_t);
388 * Add a device to the list, or just increment the usage count if
389 * it's already present.
391 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
392 struct dm_dev **result)
396 struct dm_dev_internal *dd;
397 struct dm_table *t = ti->table;
401 dev = dm_get_dev_t(path);
405 dd = find_device(&t->devices, dev);
407 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
411 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
416 atomic_set(&dd->count, 0);
417 list_add(&dd->list, &t->devices);
419 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
420 r = upgrade_mode(dd, mode, t->md);
424 atomic_inc(&dd->count);
426 *result = dd->dm_dev;
429 EXPORT_SYMBOL(dm_get_device);
431 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
432 sector_t start, sector_t len, void *data)
434 struct queue_limits *limits = data;
435 struct block_device *bdev = dev->bdev;
436 struct request_queue *q = bdev_get_queue(bdev);
437 char b[BDEVNAME_SIZE];
440 DMWARN("%s: Cannot set limits for nonexistent device %s",
441 dm_device_name(ti->table->md), bdevname(bdev, b));
445 if (bdev_stack_limits(limits, bdev, start) < 0)
446 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
447 "physical_block_size=%u, logical_block_size=%u, "
448 "alignment_offset=%u, start=%llu",
449 dm_device_name(ti->table->md), bdevname(bdev, b),
450 q->limits.physical_block_size,
451 q->limits.logical_block_size,
452 q->limits.alignment_offset,
453 (unsigned long long) start << SECTOR_SHIFT);
459 * Decrement a device's use count and remove it if necessary.
461 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
464 struct list_head *devices = &ti->table->devices;
465 struct dm_dev_internal *dd;
467 list_for_each_entry(dd, devices, list) {
468 if (dd->dm_dev == d) {
474 DMWARN("%s: device %s not in table devices list",
475 dm_device_name(ti->table->md), d->name);
478 if (atomic_dec_and_test(&dd->count)) {
479 dm_put_table_device(ti->table->md, d);
484 EXPORT_SYMBOL(dm_put_device);
487 * Checks to see if the target joins onto the end of the table.
489 static int adjoin(struct dm_table *table, struct dm_target *ti)
491 struct dm_target *prev;
493 if (!table->num_targets)
496 prev = &table->targets[table->num_targets - 1];
497 return (ti->begin == (prev->begin + prev->len));
501 * Used to dynamically allocate the arg array.
503 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
504 * process messages even if some device is suspended. These messages have a
505 * small fixed number of arguments.
507 * On the other hand, dm-switch needs to process bulk data using messages and
508 * excessive use of GFP_NOIO could cause trouble.
510 static char **realloc_argv(unsigned *array_size, char **old_argv)
517 new_size = *array_size * 2;
523 argv = kmalloc(new_size * sizeof(*argv), gfp);
525 memcpy(argv, old_argv, *array_size * sizeof(*argv));
526 *array_size = new_size;
534 * Destructively splits up the argument list to pass to ctr.
536 int dm_split_args(int *argc, char ***argvp, char *input)
538 char *start, *end = input, *out, **argv = NULL;
539 unsigned array_size = 0;
548 argv = realloc_argv(&array_size, argv);
553 /* Skip whitespace */
554 start = skip_spaces(end);
557 break; /* success, we hit the end */
559 /* 'out' is used to remove any back-quotes */
562 /* Everything apart from '\0' can be quoted */
563 if (*end == '\\' && *(end + 1)) {
570 break; /* end of token */
575 /* have we already filled the array ? */
576 if ((*argc + 1) > array_size) {
577 argv = realloc_argv(&array_size, argv);
582 /* we know this is whitespace */
586 /* terminate the string and put it in the array */
597 * Impose necessary and sufficient conditions on a devices's table such
598 * that any incoming bio which respects its logical_block_size can be
599 * processed successfully. If it falls across the boundary between
600 * two or more targets, the size of each piece it gets split into must
601 * be compatible with the logical_block_size of the target processing it.
603 static int validate_hardware_logical_block_alignment(struct dm_table *table,
604 struct queue_limits *limits)
607 * This function uses arithmetic modulo the logical_block_size
608 * (in units of 512-byte sectors).
610 unsigned short device_logical_block_size_sects =
611 limits->logical_block_size >> SECTOR_SHIFT;
614 * Offset of the start of the next table entry, mod logical_block_size.
616 unsigned short next_target_start = 0;
619 * Given an aligned bio that extends beyond the end of a
620 * target, how many sectors must the next target handle?
622 unsigned short remaining = 0;
624 struct dm_target *uninitialized_var(ti);
625 struct queue_limits ti_limits;
629 * Check each entry in the table in turn.
631 while (i < dm_table_get_num_targets(table)) {
632 ti = dm_table_get_target(table, i++);
634 blk_set_stacking_limits(&ti_limits);
636 /* combine all target devices' limits */
637 if (ti->type->iterate_devices)
638 ti->type->iterate_devices(ti, dm_set_device_limits,
642 * If the remaining sectors fall entirely within this
643 * table entry are they compatible with its logical_block_size?
645 if (remaining < ti->len &&
646 remaining & ((ti_limits.logical_block_size >>
651 (unsigned short) ((next_target_start + ti->len) &
652 (device_logical_block_size_sects - 1));
653 remaining = next_target_start ?
654 device_logical_block_size_sects - next_target_start : 0;
658 DMWARN("%s: table line %u (start sect %llu len %llu) "
659 "not aligned to h/w logical block size %u",
660 dm_device_name(table->md), i,
661 (unsigned long long) ti->begin,
662 (unsigned long long) ti->len,
663 limits->logical_block_size);
670 int dm_table_add_target(struct dm_table *t, const char *type,
671 sector_t start, sector_t len, char *params)
673 int r = -EINVAL, argc;
675 struct dm_target *tgt;
678 DMERR("%s: target type %s must appear alone in table",
679 dm_device_name(t->md), t->targets->type->name);
683 BUG_ON(t->num_targets >= t->num_allocated);
685 tgt = t->targets + t->num_targets;
686 memset(tgt, 0, sizeof(*tgt));
689 DMERR("%s: zero-length target", dm_device_name(t->md));
693 tgt->type = dm_get_target_type(type);
695 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
700 if (dm_target_needs_singleton(tgt->type)) {
701 if (t->num_targets) {
702 DMERR("%s: target type %s must appear alone in table",
703 dm_device_name(t->md), type);
709 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
710 DMERR("%s: target type %s may not be included in read-only tables",
711 dm_device_name(t->md), type);
715 if (t->immutable_target_type) {
716 if (t->immutable_target_type != tgt->type) {
717 DMERR("%s: immutable target type %s cannot be mixed with other target types",
718 dm_device_name(t->md), t->immutable_target_type->name);
721 } else if (dm_target_is_immutable(tgt->type)) {
722 if (t->num_targets) {
723 DMERR("%s: immutable target type %s cannot be mixed with other target types",
724 dm_device_name(t->md), tgt->type->name);
727 t->immutable_target_type = tgt->type;
733 tgt->error = "Unknown error";
736 * Does this target adjoin the previous one ?
738 if (!adjoin(t, tgt)) {
739 tgt->error = "Gap in table";
744 r = dm_split_args(&argc, &argv, params);
746 tgt->error = "couldn't split parameters (insufficient memory)";
750 r = tgt->type->ctr(tgt, argc, argv);
755 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
757 if (!tgt->num_discard_bios && tgt->discards_supported)
758 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
759 dm_device_name(t->md), type);
764 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
765 dm_put_target_type(tgt->type);
770 * Target argument parsing helpers.
772 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
773 unsigned *value, char **error, unsigned grouped)
775 const char *arg_str = dm_shift_arg(arg_set);
779 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
780 (*value < arg->min) ||
781 (*value > arg->max) ||
782 (grouped && arg_set->argc < *value)) {
790 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
791 unsigned *value, char **error)
793 return validate_next_arg(arg, arg_set, value, error, 0);
795 EXPORT_SYMBOL(dm_read_arg);
797 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
798 unsigned *value, char **error)
800 return validate_next_arg(arg, arg_set, value, error, 1);
802 EXPORT_SYMBOL(dm_read_arg_group);
804 const char *dm_shift_arg(struct dm_arg_set *as)
817 EXPORT_SYMBOL(dm_shift_arg);
819 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
821 BUG_ON(as->argc < num_args);
822 as->argc -= num_args;
823 as->argv += num_args;
825 EXPORT_SYMBOL(dm_consume_args);
827 static bool __table_type_request_based(unsigned table_type)
829 return (table_type == DM_TYPE_REQUEST_BASED ||
830 table_type == DM_TYPE_MQ_REQUEST_BASED);
833 static int dm_table_set_type(struct dm_table *t)
836 unsigned bio_based = 0, request_based = 0, hybrid = 0;
837 bool use_blk_mq = false;
838 struct dm_target *tgt;
839 struct dm_dev_internal *dd;
840 struct list_head *devices;
841 unsigned live_md_type = dm_get_md_type(t->md);
843 for (i = 0; i < t->num_targets; i++) {
844 tgt = t->targets + i;
845 if (dm_target_hybrid(tgt))
847 else if (dm_target_request_based(tgt))
852 if (bio_based && request_based) {
853 DMWARN("Inconsistent table: different target types"
854 " can't be mixed up");
859 if (hybrid && !bio_based && !request_based) {
861 * The targets can work either way.
862 * Determine the type from the live device.
863 * Default to bio-based if device is new.
865 if (__table_type_request_based(live_md_type))
872 /* We must use this table as bio-based */
873 t->type = DM_TYPE_BIO_BASED;
877 BUG_ON(!request_based); /* No targets in this table */
880 * Request-based dm supports only tables that have a single target now.
881 * To support multiple targets, request splitting support is needed,
882 * and that needs lots of changes in the block-layer.
883 * (e.g. request completion process for partial completion.)
885 if (t->num_targets > 1) {
886 DMWARN("Request-based dm doesn't support multiple targets yet");
890 /* Non-request-stackable devices can't be used for request-based dm */
891 devices = dm_table_get_devices(t);
892 list_for_each_entry(dd, devices, list) {
893 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
895 if (!blk_queue_stackable(q)) {
896 DMERR("table load rejected: including"
897 " non-request-stackable devices");
906 /* verify _all_ devices in the table are blk-mq devices */
907 list_for_each_entry(dd, devices, list)
908 if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
909 DMERR("table load rejected: not all devices"
910 " are blk-mq request-stackable");
913 t->type = DM_TYPE_MQ_REQUEST_BASED;
915 } else if (list_empty(devices) && __table_type_request_based(live_md_type)) {
916 /* inherit live MD type */
917 t->type = live_md_type;
920 t->type = DM_TYPE_REQUEST_BASED;
925 unsigned dm_table_get_type(struct dm_table *t)
930 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
932 return t->immutable_target_type;
935 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
937 /* Immutable target is implicitly a singleton */
938 if (t->num_targets > 1 ||
939 !dm_target_is_immutable(t->targets[0].type))
945 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
947 struct dm_target *uninitialized_var(ti);
950 while (i < dm_table_get_num_targets(t)) {
951 ti = dm_table_get_target(t, i++);
952 if (dm_target_is_wildcard(ti->type))
959 bool dm_table_request_based(struct dm_table *t)
961 return __table_type_request_based(dm_table_get_type(t));
964 bool dm_table_mq_request_based(struct dm_table *t)
966 return dm_table_get_type(t) == DM_TYPE_MQ_REQUEST_BASED;
969 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
971 unsigned type = dm_table_get_type(t);
972 unsigned per_io_data_size = 0;
973 struct dm_target *tgt;
976 if (unlikely(type == DM_TYPE_NONE)) {
977 DMWARN("no table type is set, can't allocate mempools");
981 if (type == DM_TYPE_BIO_BASED)
982 for (i = 0; i < t->num_targets; i++) {
983 tgt = t->targets + i;
984 per_io_data_size = max(per_io_data_size, tgt->per_io_data_size);
987 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_io_data_size);
994 void dm_table_free_md_mempools(struct dm_table *t)
996 dm_free_md_mempools(t->mempools);
1000 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1005 static int setup_indexes(struct dm_table *t)
1008 unsigned int total = 0;
1011 /* allocate the space for *all* the indexes */
1012 for (i = t->depth - 2; i >= 0; i--) {
1013 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1014 total += t->counts[i];
1017 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1021 /* set up internal nodes, bottom-up */
1022 for (i = t->depth - 2; i >= 0; i--) {
1023 t->index[i] = indexes;
1024 indexes += (KEYS_PER_NODE * t->counts[i]);
1025 setup_btree_index(i, t);
1032 * Builds the btree to index the map.
1034 static int dm_table_build_index(struct dm_table *t)
1037 unsigned int leaf_nodes;
1039 /* how many indexes will the btree have ? */
1040 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1041 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1043 /* leaf layer has already been set up */
1044 t->counts[t->depth - 1] = leaf_nodes;
1045 t->index[t->depth - 1] = t->highs;
1048 r = setup_indexes(t);
1053 static bool integrity_profile_exists(struct gendisk *disk)
1055 return !!blk_get_integrity(disk);
1059 * Get a disk whose integrity profile reflects the table's profile.
1060 * Returns NULL if integrity support was inconsistent or unavailable.
1062 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1064 struct list_head *devices = dm_table_get_devices(t);
1065 struct dm_dev_internal *dd = NULL;
1066 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1068 list_for_each_entry(dd, devices, list) {
1069 template_disk = dd->dm_dev->bdev->bd_disk;
1070 if (!integrity_profile_exists(template_disk))
1072 else if (prev_disk &&
1073 blk_integrity_compare(prev_disk, template_disk) < 0)
1075 prev_disk = template_disk;
1078 return template_disk;
1082 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1083 dm_device_name(t->md),
1084 prev_disk->disk_name,
1085 template_disk->disk_name);
1090 * Register the mapped device for blk_integrity support if the
1091 * underlying devices have an integrity profile. But all devices may
1092 * not have matching profiles (checking all devices isn't reliable
1093 * during table load because this table may use other DM device(s) which
1094 * must be resumed before they will have an initialized integity
1095 * profile). Consequently, stacked DM devices force a 2 stage integrity
1096 * profile validation: First pass during table load, final pass during
1099 static int dm_table_register_integrity(struct dm_table *t)
1101 struct mapped_device *md = t->md;
1102 struct gendisk *template_disk = NULL;
1104 template_disk = dm_table_get_integrity_disk(t);
1108 if (!integrity_profile_exists(dm_disk(md))) {
1109 t->integrity_supported = 1;
1111 * Register integrity profile during table load; we can do
1112 * this because the final profile must match during resume.
1114 blk_integrity_register(dm_disk(md),
1115 blk_get_integrity(template_disk));
1120 * If DM device already has an initialized integrity
1121 * profile the new profile should not conflict.
1123 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1124 DMWARN("%s: conflict with existing integrity profile: "
1125 "%s profile mismatch",
1126 dm_device_name(t->md),
1127 template_disk->disk_name);
1131 /* Preserve existing integrity profile */
1132 t->integrity_supported = 1;
1137 * Prepares the table for use by building the indices,
1138 * setting the type, and allocating mempools.
1140 int dm_table_complete(struct dm_table *t)
1144 r = dm_table_set_type(t);
1146 DMERR("unable to set table type");
1150 r = dm_table_build_index(t);
1152 DMERR("unable to build btrees");
1156 r = dm_table_register_integrity(t);
1158 DMERR("could not register integrity profile.");
1162 r = dm_table_alloc_md_mempools(t, t->md);
1164 DMERR("unable to allocate mempools");
1169 static DEFINE_MUTEX(_event_lock);
1170 void dm_table_event_callback(struct dm_table *t,
1171 void (*fn)(void *), void *context)
1173 mutex_lock(&_event_lock);
1175 t->event_context = context;
1176 mutex_unlock(&_event_lock);
1179 void dm_table_event(struct dm_table *t)
1182 * You can no longer call dm_table_event() from interrupt
1183 * context, use a bottom half instead.
1185 BUG_ON(in_interrupt());
1187 mutex_lock(&_event_lock);
1189 t->event_fn(t->event_context);
1190 mutex_unlock(&_event_lock);
1192 EXPORT_SYMBOL(dm_table_event);
1194 sector_t dm_table_get_size(struct dm_table *t)
1196 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1198 EXPORT_SYMBOL(dm_table_get_size);
1200 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1202 if (index >= t->num_targets)
1205 return t->targets + index;
1209 * Search the btree for the correct target.
1211 * Caller should check returned pointer with dm_target_is_valid()
1212 * to trap I/O beyond end of device.
1214 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1216 unsigned int l, n = 0, k = 0;
1219 for (l = 0; l < t->depth; l++) {
1220 n = get_child(n, k);
1221 node = get_node(t, l, n);
1223 for (k = 0; k < KEYS_PER_NODE; k++)
1224 if (node[k] >= sector)
1228 return &t->targets[(KEYS_PER_NODE * n) + k];
1231 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1232 sector_t start, sector_t len, void *data)
1234 unsigned *num_devices = data;
1242 * Check whether a table has no data devices attached using each
1243 * target's iterate_devices method.
1244 * Returns false if the result is unknown because a target doesn't
1245 * support iterate_devices.
1247 bool dm_table_has_no_data_devices(struct dm_table *table)
1249 struct dm_target *uninitialized_var(ti);
1250 unsigned i = 0, num_devices = 0;
1252 while (i < dm_table_get_num_targets(table)) {
1253 ti = dm_table_get_target(table, i++);
1255 if (!ti->type->iterate_devices)
1258 ti->type->iterate_devices(ti, count_device, &num_devices);
1267 * Establish the new table's queue_limits and validate them.
1269 int dm_calculate_queue_limits(struct dm_table *table,
1270 struct queue_limits *limits)
1272 struct dm_target *uninitialized_var(ti);
1273 struct queue_limits ti_limits;
1276 blk_set_stacking_limits(limits);
1278 while (i < dm_table_get_num_targets(table)) {
1279 blk_set_stacking_limits(&ti_limits);
1281 ti = dm_table_get_target(table, i++);
1283 if (!ti->type->iterate_devices)
1284 goto combine_limits;
1287 * Combine queue limits of all the devices this target uses.
1289 ti->type->iterate_devices(ti, dm_set_device_limits,
1292 /* Set I/O hints portion of queue limits */
1293 if (ti->type->io_hints)
1294 ti->type->io_hints(ti, &ti_limits);
1297 * Check each device area is consistent with the target's
1298 * overall queue limits.
1300 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1306 * Merge this target's queue limits into the overall limits
1309 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1310 DMWARN("%s: adding target device "
1311 "(start sect %llu len %llu) "
1312 "caused an alignment inconsistency",
1313 dm_device_name(table->md),
1314 (unsigned long long) ti->begin,
1315 (unsigned long long) ti->len);
1318 return validate_hardware_logical_block_alignment(table, limits);
1322 * Verify that all devices have an integrity profile that matches the
1323 * DM device's registered integrity profile. If the profiles don't
1324 * match then unregister the DM device's integrity profile.
1326 static void dm_table_verify_integrity(struct dm_table *t)
1328 struct gendisk *template_disk = NULL;
1330 if (t->integrity_supported) {
1332 * Verify that the original integrity profile
1333 * matches all the devices in this table.
1335 template_disk = dm_table_get_integrity_disk(t);
1336 if (template_disk &&
1337 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1341 if (integrity_profile_exists(dm_disk(t->md))) {
1342 DMWARN("%s: unable to establish an integrity profile",
1343 dm_device_name(t->md));
1344 blk_integrity_unregister(dm_disk(t->md));
1348 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1349 sector_t start, sector_t len, void *data)
1351 unsigned flush = (*(unsigned *)data);
1352 struct request_queue *q = bdev_get_queue(dev->bdev);
1354 return q && (q->flush_flags & flush);
1357 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1359 struct dm_target *ti;
1363 * Require at least one underlying device to support flushes.
1364 * t->devices includes internal dm devices such as mirror logs
1365 * so we need to use iterate_devices here, which targets
1366 * supporting flushes must provide.
1368 while (i < dm_table_get_num_targets(t)) {
1369 ti = dm_table_get_target(t, i++);
1371 if (!ti->num_flush_bios)
1374 if (ti->flush_supported)
1377 if (ti->type->iterate_devices &&
1378 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1385 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1387 struct dm_target *ti;
1390 /* Ensure that all targets supports discard_zeroes_data. */
1391 while (i < dm_table_get_num_targets(t)) {
1392 ti = dm_table_get_target(t, i++);
1394 if (ti->discard_zeroes_data_unsupported)
1401 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1402 sector_t start, sector_t len, void *data)
1404 struct request_queue *q = bdev_get_queue(dev->bdev);
1406 return q && blk_queue_nonrot(q);
1409 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1410 sector_t start, sector_t len, void *data)
1412 struct request_queue *q = bdev_get_queue(dev->bdev);
1414 return q && !blk_queue_add_random(q);
1417 static int queue_supports_sg_merge(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);
1422 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1425 static bool dm_table_all_devices_attribute(struct dm_table *t,
1426 iterate_devices_callout_fn func)
1428 struct dm_target *ti;
1431 while (i < dm_table_get_num_targets(t)) {
1432 ti = dm_table_get_target(t, i++);
1434 if (!ti->type->iterate_devices ||
1435 !ti->type->iterate_devices(ti, func, NULL))
1442 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1443 sector_t start, sector_t len, void *data)
1445 struct request_queue *q = bdev_get_queue(dev->bdev);
1447 return q && !q->limits.max_write_same_sectors;
1450 static bool dm_table_supports_write_same(struct dm_table *t)
1452 struct dm_target *ti;
1455 while (i < dm_table_get_num_targets(t)) {
1456 ti = dm_table_get_target(t, i++);
1458 if (!ti->num_write_same_bios)
1461 if (!ti->type->iterate_devices ||
1462 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1469 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1470 sector_t start, sector_t len, void *data)
1472 struct request_queue *q = bdev_get_queue(dev->bdev);
1474 return q && blk_queue_discard(q);
1477 static bool dm_table_supports_discards(struct dm_table *t)
1479 struct dm_target *ti;
1483 * Unless any target used by the table set discards_supported,
1484 * require at least one underlying device to support discards.
1485 * t->devices includes internal dm devices such as mirror logs
1486 * so we need to use iterate_devices here, which targets
1487 * supporting discard selectively must provide.
1489 while (i < dm_table_get_num_targets(t)) {
1490 ti = dm_table_get_target(t, i++);
1492 if (!ti->num_discard_bios)
1495 if (ti->discards_supported)
1498 if (ti->type->iterate_devices &&
1499 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1506 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1507 struct queue_limits *limits)
1512 * Copy table's limits to the DM device's request_queue
1514 q->limits = *limits;
1516 if (!dm_table_supports_discards(t))
1517 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1519 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1521 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1523 if (dm_table_supports_flush(t, REQ_FUA))
1526 blk_queue_flush(q, flush);
1528 if (!dm_table_discard_zeroes_data(t))
1529 q->limits.discard_zeroes_data = 0;
1531 /* Ensure that all underlying devices are non-rotational. */
1532 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1533 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1535 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1537 if (!dm_table_supports_write_same(t))
1538 q->limits.max_write_same_sectors = 0;
1540 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1541 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1543 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1545 dm_table_verify_integrity(t);
1548 * Determine whether or not this queue's I/O timings contribute
1549 * to the entropy pool, Only request-based targets use this.
1550 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1553 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1554 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1557 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1558 * visible to other CPUs because, once the flag is set, incoming bios
1559 * are processed by request-based dm, which refers to the queue
1561 * Until the flag set, bios are passed to bio-based dm and queued to
1562 * md->deferred where queue settings are not needed yet.
1563 * Those bios are passed to request-based dm at the resume time.
1566 if (dm_table_request_based(t))
1567 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1570 unsigned int dm_table_get_num_targets(struct dm_table *t)
1572 return t->num_targets;
1575 struct list_head *dm_table_get_devices(struct dm_table *t)
1580 fmode_t dm_table_get_mode(struct dm_table *t)
1584 EXPORT_SYMBOL(dm_table_get_mode);
1592 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1594 int i = t->num_targets;
1595 struct dm_target *ti = t->targets;
1600 if (ti->type->presuspend)
1601 ti->type->presuspend(ti);
1603 case PRESUSPEND_UNDO:
1604 if (ti->type->presuspend_undo)
1605 ti->type->presuspend_undo(ti);
1608 if (ti->type->postsuspend)
1609 ti->type->postsuspend(ti);
1616 void dm_table_presuspend_targets(struct dm_table *t)
1621 suspend_targets(t, PRESUSPEND);
1624 void dm_table_presuspend_undo_targets(struct dm_table *t)
1629 suspend_targets(t, PRESUSPEND_UNDO);
1632 void dm_table_postsuspend_targets(struct dm_table *t)
1637 suspend_targets(t, POSTSUSPEND);
1640 int dm_table_resume_targets(struct dm_table *t)
1644 for (i = 0; i < t->num_targets; i++) {
1645 struct dm_target *ti = t->targets + i;
1647 if (!ti->type->preresume)
1650 r = ti->type->preresume(ti);
1652 DMERR("%s: %s: preresume failed, error = %d",
1653 dm_device_name(t->md), ti->type->name, r);
1658 for (i = 0; i < t->num_targets; i++) {
1659 struct dm_target *ti = t->targets + i;
1661 if (ti->type->resume)
1662 ti->type->resume(ti);
1668 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1670 list_add(&cb->list, &t->target_callbacks);
1672 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1674 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1676 struct dm_dev_internal *dd;
1677 struct list_head *devices = dm_table_get_devices(t);
1678 struct dm_target_callbacks *cb;
1681 list_for_each_entry(dd, devices, list) {
1682 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1683 char b[BDEVNAME_SIZE];
1686 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1688 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1689 dm_device_name(t->md),
1690 bdevname(dd->dm_dev->bdev, b));
1693 list_for_each_entry(cb, &t->target_callbacks, list)
1694 if (cb->congested_fn)
1695 r |= cb->congested_fn(cb, bdi_bits);
1700 struct mapped_device *dm_table_get_md(struct dm_table *t)
1704 EXPORT_SYMBOL(dm_table_get_md);
1706 void dm_table_run_md_queue_async(struct dm_table *t)
1708 struct mapped_device *md;
1709 struct request_queue *queue;
1710 unsigned long flags;
1712 if (!dm_table_request_based(t))
1715 md = dm_table_get_md(t);
1716 queue = dm_get_md_queue(md);
1719 blk_mq_run_hw_queues(queue, true);
1721 spin_lock_irqsave(queue->queue_lock, flags);
1722 blk_run_queue_async(queue);
1723 spin_unlock_irqrestore(queue->queue_lock, flags);
1727 EXPORT_SYMBOL(dm_table_run_md_queue_async);