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>
23 #define DM_MSG_PREFIX "table"
26 #define NODE_SIZE L1_CACHE_BYTES
27 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
28 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
31 struct mapped_device *md;
36 unsigned int counts[MAX_DEPTH]; /* in nodes */
37 sector_t *index[MAX_DEPTH];
39 unsigned int num_targets;
40 unsigned int num_allocated;
42 struct dm_target *targets;
44 struct target_type *immutable_target_type;
45 unsigned integrity_supported:1;
49 * Indicates the rw permissions for the new logical
50 * device. This should be a combination of FMODE_READ
55 /* a list of devices used by this table */
56 struct list_head devices;
58 /* events get handed up using this callback */
59 void (*event_fn)(void *);
62 struct dm_md_mempools *mempools;
64 struct list_head target_callbacks;
68 * Similar to ceiling(log_size(n))
70 static unsigned int int_log(unsigned int n, unsigned int base)
75 n = dm_div_up(n, base);
83 * Calculate the index of the child node of the n'th node k'th key.
85 static inline unsigned int get_child(unsigned int n, unsigned int k)
87 return (n * CHILDREN_PER_NODE) + k;
91 * Return the n'th node of level l from table t.
93 static inline sector_t *get_node(struct dm_table *t,
94 unsigned int l, unsigned int n)
96 return t->index[l] + (n * KEYS_PER_NODE);
100 * Return the highest key that you could lookup from the n'th
101 * node on level l of the btree.
103 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
105 for (; l < t->depth - 1; l++)
106 n = get_child(n, CHILDREN_PER_NODE - 1);
108 if (n >= t->counts[l])
109 return (sector_t) - 1;
111 return get_node(t, l, n)[KEYS_PER_NODE - 1];
115 * Fills in a level of the btree based on the highs of the level
118 static int setup_btree_index(unsigned int l, struct dm_table *t)
123 for (n = 0U; n < t->counts[l]; n++) {
124 node = get_node(t, l, n);
126 for (k = 0U; k < KEYS_PER_NODE; k++)
127 node[k] = high(t, l + 1, get_child(n, k));
133 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
139 * Check that we're not going to overflow.
141 if (nmemb > (ULONG_MAX / elem_size))
144 size = nmemb * elem_size;
145 addr = vzalloc(size);
149 EXPORT_SYMBOL(dm_vcalloc);
152 * highs, and targets are managed as dynamic arrays during a
155 static int alloc_targets(struct dm_table *t, unsigned int num)
158 struct dm_target *n_targets;
161 * Allocate both the target array and offset array at once.
162 * Append an empty entry to catch sectors beyond the end of
165 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
170 n_targets = (struct dm_target *) (n_highs + num);
172 memset(n_highs, -1, sizeof(*n_highs) * num);
175 t->num_allocated = num;
177 t->targets = n_targets;
182 int dm_table_create(struct dm_table **result, fmode_t mode,
183 unsigned num_targets, struct mapped_device *md)
185 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
190 INIT_LIST_HEAD(&t->devices);
191 INIT_LIST_HEAD(&t->target_callbacks);
194 num_targets = KEYS_PER_NODE;
196 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
203 if (alloc_targets(t, num_targets)) {
214 static void free_devices(struct list_head *devices, struct mapped_device *md)
216 struct list_head *tmp, *next;
218 list_for_each_safe(tmp, next, devices) {
219 struct dm_dev_internal *dd =
220 list_entry(tmp, struct dm_dev_internal, list);
221 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
222 dm_device_name(md), dd->dm_dev->name);
223 dm_put_table_device(md, dd->dm_dev);
228 void dm_table_destroy(struct dm_table *t)
235 /* free the indexes */
237 vfree(t->index[t->depth - 2]);
239 /* free the targets */
240 for (i = 0; i < t->num_targets; i++) {
241 struct dm_target *tgt = t->targets + i;
246 dm_put_target_type(tgt->type);
251 /* free the device list */
252 free_devices(&t->devices, t->md);
254 dm_free_md_mempools(t->mempools);
260 * See if we've already got a device in the list.
262 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
264 struct dm_dev_internal *dd;
266 list_for_each_entry (dd, l, list)
267 if (dd->dm_dev->bdev->bd_dev == dev)
274 * If possible, this checks an area of a destination device is invalid.
276 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
277 sector_t start, sector_t len, void *data)
279 struct request_queue *q;
280 struct queue_limits *limits = data;
281 struct block_device *bdev = dev->bdev;
283 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
284 unsigned short logical_block_size_sectors =
285 limits->logical_block_size >> SECTOR_SHIFT;
286 char b[BDEVNAME_SIZE];
289 * Some devices exist without request functions,
290 * such as loop devices not yet bound to backing files.
291 * Forbid the use of such devices.
293 q = bdev_get_queue(bdev);
294 if (!q || !q->make_request_fn) {
295 DMWARN("%s: %s is not yet initialised: "
296 "start=%llu, len=%llu, dev_size=%llu",
297 dm_device_name(ti->table->md), bdevname(bdev, b),
298 (unsigned long long)start,
299 (unsigned long long)len,
300 (unsigned long long)dev_size);
307 if ((start >= dev_size) || (start + len > dev_size)) {
308 DMWARN("%s: %s too small for target: "
309 "start=%llu, len=%llu, dev_size=%llu",
310 dm_device_name(ti->table->md), bdevname(bdev, b),
311 (unsigned long long)start,
312 (unsigned long long)len,
313 (unsigned long long)dev_size);
317 if (logical_block_size_sectors <= 1)
320 if (start & (logical_block_size_sectors - 1)) {
321 DMWARN("%s: start=%llu not aligned to h/w "
322 "logical block size %u of %s",
323 dm_device_name(ti->table->md),
324 (unsigned long long)start,
325 limits->logical_block_size, bdevname(bdev, b));
329 if (len & (logical_block_size_sectors - 1)) {
330 DMWARN("%s: len=%llu not aligned to h/w "
331 "logical block size %u of %s",
332 dm_device_name(ti->table->md),
333 (unsigned long long)len,
334 limits->logical_block_size, bdevname(bdev, b));
342 * This upgrades the mode on an already open dm_dev, being
343 * careful to leave things as they were if we fail to reopen the
344 * device and not to touch the existing bdev field in case
345 * it is accessed concurrently inside dm_table_any_congested().
347 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
348 struct mapped_device *md)
351 struct dm_dev *old_dev, *new_dev;
353 old_dev = dd->dm_dev;
355 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
356 dd->dm_dev->mode | new_mode, &new_dev);
360 dd->dm_dev = new_dev;
361 dm_put_table_device(md, old_dev);
367 * Add a device to the list, or just increment the usage count if
368 * it's already present.
370 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
371 struct dm_dev **result)
374 dev_t uninitialized_var(dev);
375 struct dm_dev_internal *dd;
376 unsigned int major, minor;
377 struct dm_table *t = ti->table;
382 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
383 /* Extract the major/minor numbers */
384 dev = MKDEV(major, minor);
385 if (MAJOR(dev) != major || MINOR(dev) != minor)
388 /* convert the path to a device */
389 struct block_device *bdev = lookup_bdev(path);
392 return PTR_ERR(bdev);
397 dd = find_device(&t->devices, dev);
399 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
403 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
408 atomic_set(&dd->count, 0);
409 list_add(&dd->list, &t->devices);
411 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
412 r = upgrade_mode(dd, mode, t->md);
416 atomic_inc(&dd->count);
418 *result = dd->dm_dev;
421 EXPORT_SYMBOL(dm_get_device);
423 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
424 sector_t start, sector_t len, void *data)
426 struct queue_limits *limits = data;
427 struct block_device *bdev = dev->bdev;
428 struct request_queue *q = bdev_get_queue(bdev);
429 char b[BDEVNAME_SIZE];
432 DMWARN("%s: Cannot set limits for nonexistent device %s",
433 dm_device_name(ti->table->md), bdevname(bdev, b));
437 if (bdev_stack_limits(limits, bdev, start) < 0)
438 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
439 "physical_block_size=%u, logical_block_size=%u, "
440 "alignment_offset=%u, start=%llu",
441 dm_device_name(ti->table->md), bdevname(bdev, b),
442 q->limits.physical_block_size,
443 q->limits.logical_block_size,
444 q->limits.alignment_offset,
445 (unsigned long long) start << SECTOR_SHIFT);
448 * Check if merge fn is supported.
449 * If not we'll force DM to use PAGE_SIZE or
450 * smaller I/O, just to be safe.
452 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
453 blk_limits_max_hw_sectors(limits,
454 (unsigned int) (PAGE_SIZE >> 9));
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 int dm_table_set_type(struct dm_table *t)
830 unsigned bio_based = 0, request_based = 0, hybrid = 0;
831 bool use_blk_mq = false;
832 struct dm_target *tgt;
833 struct dm_dev_internal *dd;
834 struct list_head *devices;
835 unsigned live_md_type = dm_get_md_type(t->md);
837 for (i = 0; i < t->num_targets; i++) {
838 tgt = t->targets + i;
839 if (dm_target_hybrid(tgt))
841 else if (dm_target_request_based(tgt))
846 if (bio_based && request_based) {
847 DMWARN("Inconsistent table: different target types"
848 " can't be mixed up");
853 if (hybrid && !bio_based && !request_based) {
855 * The targets can work either way.
856 * Determine the type from the live device.
857 * Default to bio-based if device is new.
859 if (live_md_type == DM_TYPE_REQUEST_BASED ||
860 live_md_type == DM_TYPE_MQ_REQUEST_BASED)
867 /* We must use this table as bio-based */
868 t->type = DM_TYPE_BIO_BASED;
872 BUG_ON(!request_based); /* No targets in this table */
875 * Request-based dm supports only tables that have a single target now.
876 * To support multiple targets, request splitting support is needed,
877 * and that needs lots of changes in the block-layer.
878 * (e.g. request completion process for partial completion.)
880 if (t->num_targets > 1) {
881 DMWARN("Request-based dm doesn't support multiple targets yet");
885 /* Non-request-stackable devices can't be used for request-based dm */
886 devices = dm_table_get_devices(t);
887 list_for_each_entry(dd, devices, list) {
888 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
890 if (!blk_queue_stackable(q)) {
891 DMERR("table load rejected: including"
892 " non-request-stackable devices");
901 /* verify _all_ devices in the table are blk-mq devices */
902 list_for_each_entry(dd, devices, list)
903 if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
904 DMERR("table load rejected: not all devices"
905 " are blk-mq request-stackable");
908 t->type = DM_TYPE_MQ_REQUEST_BASED;
910 } else if (hybrid && list_empty(devices) && live_md_type != DM_TYPE_NONE) {
911 /* inherit live MD type */
912 t->type = live_md_type;
915 t->type = DM_TYPE_REQUEST_BASED;
920 unsigned dm_table_get_type(struct dm_table *t)
925 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
927 return t->immutable_target_type;
930 bool dm_table_request_based(struct dm_table *t)
932 unsigned table_type = dm_table_get_type(t);
934 return (table_type == DM_TYPE_REQUEST_BASED ||
935 table_type == DM_TYPE_MQ_REQUEST_BASED);
938 bool dm_table_mq_request_based(struct dm_table *t)
940 return dm_table_get_type(t) == DM_TYPE_MQ_REQUEST_BASED;
943 static int dm_table_alloc_md_mempools(struct dm_table *t)
945 unsigned type = dm_table_get_type(t);
946 unsigned per_bio_data_size = 0;
947 struct dm_target *tgt;
950 if (unlikely(type == DM_TYPE_NONE)) {
951 DMWARN("no table type is set, can't allocate mempools");
955 if (type == DM_TYPE_BIO_BASED)
956 for (i = 0; i < t->num_targets; i++) {
957 tgt = t->targets + i;
958 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
961 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported, per_bio_data_size);
968 void dm_table_free_md_mempools(struct dm_table *t)
970 dm_free_md_mempools(t->mempools);
974 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
979 static int setup_indexes(struct dm_table *t)
982 unsigned int total = 0;
985 /* allocate the space for *all* the indexes */
986 for (i = t->depth - 2; i >= 0; i--) {
987 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
988 total += t->counts[i];
991 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
995 /* set up internal nodes, bottom-up */
996 for (i = t->depth - 2; i >= 0; i--) {
997 t->index[i] = indexes;
998 indexes += (KEYS_PER_NODE * t->counts[i]);
999 setup_btree_index(i, t);
1006 * Builds the btree to index the map.
1008 static int dm_table_build_index(struct dm_table *t)
1011 unsigned int leaf_nodes;
1013 /* how many indexes will the btree have ? */
1014 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1015 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1017 /* leaf layer has already been set up */
1018 t->counts[t->depth - 1] = leaf_nodes;
1019 t->index[t->depth - 1] = t->highs;
1022 r = setup_indexes(t);
1028 * Get a disk whose integrity profile reflects the table's profile.
1029 * If %match_all is true, all devices' profiles must match.
1030 * If %match_all is false, all devices must at least have an
1031 * allocated integrity profile; but uninitialized is ok.
1032 * Returns NULL if integrity support was inconsistent or unavailable.
1034 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1037 struct list_head *devices = dm_table_get_devices(t);
1038 struct dm_dev_internal *dd = NULL;
1039 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1041 list_for_each_entry(dd, devices, list) {
1042 template_disk = dd->dm_dev->bdev->bd_disk;
1043 if (!blk_get_integrity(template_disk))
1045 if (!match_all && !blk_integrity_is_initialized(template_disk))
1046 continue; /* skip uninitialized profiles */
1047 else if (prev_disk &&
1048 blk_integrity_compare(prev_disk, template_disk) < 0)
1050 prev_disk = template_disk;
1053 return template_disk;
1057 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1058 dm_device_name(t->md),
1059 prev_disk->disk_name,
1060 template_disk->disk_name);
1065 * Register the mapped device for blk_integrity support if
1066 * the underlying devices have an integrity profile. But all devices
1067 * may not have matching profiles (checking all devices isn't reliable
1068 * during table load because this table may use other DM device(s) which
1069 * must be resumed before they will have an initialized integity profile).
1070 * Stacked DM devices force a 2 stage integrity profile validation:
1071 * 1 - during load, validate all initialized integrity profiles match
1072 * 2 - during resume, validate all integrity profiles match
1074 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1076 struct gendisk *template_disk = NULL;
1078 template_disk = dm_table_get_integrity_disk(t, false);
1082 if (!blk_integrity_is_initialized(dm_disk(md))) {
1083 t->integrity_supported = 1;
1084 return blk_integrity_register(dm_disk(md), NULL);
1088 * If DM device already has an initalized integrity
1089 * profile the new profile should not conflict.
1091 if (blk_integrity_is_initialized(template_disk) &&
1092 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1093 DMWARN("%s: conflict with existing integrity profile: "
1094 "%s profile mismatch",
1095 dm_device_name(t->md),
1096 template_disk->disk_name);
1100 /* Preserve existing initialized integrity profile */
1101 t->integrity_supported = 1;
1106 * Prepares the table for use by building the indices,
1107 * setting the type, and allocating mempools.
1109 int dm_table_complete(struct dm_table *t)
1113 r = dm_table_set_type(t);
1115 DMERR("unable to set table type");
1119 r = dm_table_build_index(t);
1121 DMERR("unable to build btrees");
1125 r = dm_table_prealloc_integrity(t, t->md);
1127 DMERR("could not register integrity profile.");
1131 r = dm_table_alloc_md_mempools(t);
1133 DMERR("unable to allocate mempools");
1138 static DEFINE_MUTEX(_event_lock);
1139 void dm_table_event_callback(struct dm_table *t,
1140 void (*fn)(void *), void *context)
1142 mutex_lock(&_event_lock);
1144 t->event_context = context;
1145 mutex_unlock(&_event_lock);
1148 void dm_table_event(struct dm_table *t)
1151 * You can no longer call dm_table_event() from interrupt
1152 * context, use a bottom half instead.
1154 BUG_ON(in_interrupt());
1156 mutex_lock(&_event_lock);
1158 t->event_fn(t->event_context);
1159 mutex_unlock(&_event_lock);
1161 EXPORT_SYMBOL(dm_table_event);
1163 sector_t dm_table_get_size(struct dm_table *t)
1165 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1167 EXPORT_SYMBOL(dm_table_get_size);
1169 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1171 if (index >= t->num_targets)
1174 return t->targets + index;
1178 * Search the btree for the correct target.
1180 * Caller should check returned pointer with dm_target_is_valid()
1181 * to trap I/O beyond end of device.
1183 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1185 unsigned int l, n = 0, k = 0;
1188 for (l = 0; l < t->depth; l++) {
1189 n = get_child(n, k);
1190 node = get_node(t, l, n);
1192 for (k = 0; k < KEYS_PER_NODE; k++)
1193 if (node[k] >= sector)
1197 return &t->targets[(KEYS_PER_NODE * n) + k];
1200 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1201 sector_t start, sector_t len, void *data)
1203 unsigned *num_devices = data;
1211 * Check whether a table has no data devices attached using each
1212 * target's iterate_devices method.
1213 * Returns false if the result is unknown because a target doesn't
1214 * support iterate_devices.
1216 bool dm_table_has_no_data_devices(struct dm_table *table)
1218 struct dm_target *uninitialized_var(ti);
1219 unsigned i = 0, num_devices = 0;
1221 while (i < dm_table_get_num_targets(table)) {
1222 ti = dm_table_get_target(table, i++);
1224 if (!ti->type->iterate_devices)
1227 ti->type->iterate_devices(ti, count_device, &num_devices);
1236 * Establish the new table's queue_limits and validate them.
1238 int dm_calculate_queue_limits(struct dm_table *table,
1239 struct queue_limits *limits)
1241 struct dm_target *uninitialized_var(ti);
1242 struct queue_limits ti_limits;
1245 blk_set_stacking_limits(limits);
1247 while (i < dm_table_get_num_targets(table)) {
1248 blk_set_stacking_limits(&ti_limits);
1250 ti = dm_table_get_target(table, i++);
1252 if (!ti->type->iterate_devices)
1253 goto combine_limits;
1256 * Combine queue limits of all the devices this target uses.
1258 ti->type->iterate_devices(ti, dm_set_device_limits,
1261 /* Set I/O hints portion of queue limits */
1262 if (ti->type->io_hints)
1263 ti->type->io_hints(ti, &ti_limits);
1266 * Check each device area is consistent with the target's
1267 * overall queue limits.
1269 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1275 * Merge this target's queue limits into the overall limits
1278 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1279 DMWARN("%s: adding target device "
1280 "(start sect %llu len %llu) "
1281 "caused an alignment inconsistency",
1282 dm_device_name(table->md),
1283 (unsigned long long) ti->begin,
1284 (unsigned long long) ti->len);
1287 return validate_hardware_logical_block_alignment(table, limits);
1291 * Set the integrity profile for this device if all devices used have
1292 * matching profiles. We're quite deep in the resume path but still
1293 * don't know if all devices (particularly DM devices this device
1294 * may be stacked on) have matching profiles. Even if the profiles
1295 * don't match we have no way to fail (to resume) at this point.
1297 static void dm_table_set_integrity(struct dm_table *t)
1299 struct gendisk *template_disk = NULL;
1301 if (!blk_get_integrity(dm_disk(t->md)))
1304 template_disk = dm_table_get_integrity_disk(t, true);
1306 blk_integrity_register(dm_disk(t->md),
1307 blk_get_integrity(template_disk));
1308 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1309 DMWARN("%s: device no longer has a valid integrity profile",
1310 dm_device_name(t->md));
1312 DMWARN("%s: unable to establish an integrity profile",
1313 dm_device_name(t->md));
1316 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1317 sector_t start, sector_t len, void *data)
1319 unsigned flush = (*(unsigned *)data);
1320 struct request_queue *q = bdev_get_queue(dev->bdev);
1322 return q && (q->flush_flags & flush);
1325 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1327 struct dm_target *ti;
1331 * Require at least one underlying device to support flushes.
1332 * t->devices includes internal dm devices such as mirror logs
1333 * so we need to use iterate_devices here, which targets
1334 * supporting flushes must provide.
1336 while (i < dm_table_get_num_targets(t)) {
1337 ti = dm_table_get_target(t, i++);
1339 if (!ti->num_flush_bios)
1342 if (ti->flush_supported)
1345 if (ti->type->iterate_devices &&
1346 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1353 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1355 struct dm_target *ti;
1358 /* Ensure that all targets supports discard_zeroes_data. */
1359 while (i < dm_table_get_num_targets(t)) {
1360 ti = dm_table_get_target(t, i++);
1362 if (ti->discard_zeroes_data_unsupported)
1369 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1370 sector_t start, sector_t len, void *data)
1372 struct request_queue *q = bdev_get_queue(dev->bdev);
1374 return q && blk_queue_nonrot(q);
1377 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1378 sector_t start, sector_t len, void *data)
1380 struct request_queue *q = bdev_get_queue(dev->bdev);
1382 return q && !blk_queue_add_random(q);
1385 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1386 sector_t start, sector_t len, void *data)
1388 struct request_queue *q = bdev_get_queue(dev->bdev);
1390 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1393 static int queue_supports_sg_gaps(struct dm_target *ti, struct dm_dev *dev,
1394 sector_t start, sector_t len, void *data)
1396 struct request_queue *q = bdev_get_queue(dev->bdev);
1398 return q && !test_bit(QUEUE_FLAG_SG_GAPS, &q->queue_flags);
1401 static bool dm_table_all_devices_attribute(struct dm_table *t,
1402 iterate_devices_callout_fn func)
1404 struct dm_target *ti;
1407 while (i < dm_table_get_num_targets(t)) {
1408 ti = dm_table_get_target(t, i++);
1410 if (!ti->type->iterate_devices ||
1411 !ti->type->iterate_devices(ti, func, NULL))
1418 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1419 sector_t start, sector_t len, void *data)
1421 struct request_queue *q = bdev_get_queue(dev->bdev);
1423 return q && !q->limits.max_write_same_sectors;
1426 static bool dm_table_supports_write_same(struct dm_table *t)
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->num_write_same_bios)
1437 if (!ti->type->iterate_devices ||
1438 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1445 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1446 sector_t start, sector_t len, void *data)
1448 struct request_queue *q = bdev_get_queue(dev->bdev);
1450 return q && blk_queue_discard(q);
1453 static bool dm_table_supports_discards(struct dm_table *t)
1455 struct dm_target *ti;
1459 * Unless any target used by the table set discards_supported,
1460 * require at least one underlying device to support discards.
1461 * t->devices includes internal dm devices such as mirror logs
1462 * so we need to use iterate_devices here, which targets
1463 * supporting discard selectively must provide.
1465 while (i < dm_table_get_num_targets(t)) {
1466 ti = dm_table_get_target(t, i++);
1468 if (!ti->num_discard_bios)
1471 if (ti->discards_supported)
1474 if (ti->type->iterate_devices &&
1475 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1482 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1483 struct queue_limits *limits)
1488 * Copy table's limits to the DM device's request_queue
1490 q->limits = *limits;
1492 if (!dm_table_supports_discards(t))
1493 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1495 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1497 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1499 if (dm_table_supports_flush(t, REQ_FUA))
1502 blk_queue_flush(q, flush);
1504 if (!dm_table_discard_zeroes_data(t))
1505 q->limits.discard_zeroes_data = 0;
1507 /* Ensure that all underlying devices are non-rotational. */
1508 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1509 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1511 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1513 if (!dm_table_supports_write_same(t))
1514 q->limits.max_write_same_sectors = 0;
1516 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1517 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1519 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1521 if (dm_table_all_devices_attribute(t, queue_supports_sg_gaps))
1522 queue_flag_clear_unlocked(QUEUE_FLAG_SG_GAPS, q);
1524 queue_flag_set_unlocked(QUEUE_FLAG_SG_GAPS, q);
1526 dm_table_set_integrity(t);
1529 * Determine whether or not this queue's I/O timings contribute
1530 * to the entropy pool, Only request-based targets use this.
1531 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1534 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1535 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1538 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1539 * visible to other CPUs because, once the flag is set, incoming bios
1540 * are processed by request-based dm, which refers to the queue
1542 * Until the flag set, bios are passed to bio-based dm and queued to
1543 * md->deferred where queue settings are not needed yet.
1544 * Those bios are passed to request-based dm at the resume time.
1547 if (dm_table_request_based(t))
1548 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1551 unsigned int dm_table_get_num_targets(struct dm_table *t)
1553 return t->num_targets;
1556 struct list_head *dm_table_get_devices(struct dm_table *t)
1561 fmode_t dm_table_get_mode(struct dm_table *t)
1565 EXPORT_SYMBOL(dm_table_get_mode);
1573 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1575 int i = t->num_targets;
1576 struct dm_target *ti = t->targets;
1581 if (ti->type->presuspend)
1582 ti->type->presuspend(ti);
1584 case PRESUSPEND_UNDO:
1585 if (ti->type->presuspend_undo)
1586 ti->type->presuspend_undo(ti);
1589 if (ti->type->postsuspend)
1590 ti->type->postsuspend(ti);
1597 void dm_table_presuspend_targets(struct dm_table *t)
1602 suspend_targets(t, PRESUSPEND);
1605 void dm_table_presuspend_undo_targets(struct dm_table *t)
1610 suspend_targets(t, PRESUSPEND_UNDO);
1613 void dm_table_postsuspend_targets(struct dm_table *t)
1618 suspend_targets(t, POSTSUSPEND);
1621 int dm_table_resume_targets(struct dm_table *t)
1625 for (i = 0; i < t->num_targets; i++) {
1626 struct dm_target *ti = t->targets + i;
1628 if (!ti->type->preresume)
1631 r = ti->type->preresume(ti);
1633 DMERR("%s: %s: preresume failed, error = %d",
1634 dm_device_name(t->md), ti->type->name, r);
1639 for (i = 0; i < t->num_targets; i++) {
1640 struct dm_target *ti = t->targets + i;
1642 if (ti->type->resume)
1643 ti->type->resume(ti);
1649 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1651 list_add(&cb->list, &t->target_callbacks);
1653 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1655 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1657 struct dm_dev_internal *dd;
1658 struct list_head *devices = dm_table_get_devices(t);
1659 struct dm_target_callbacks *cb;
1662 list_for_each_entry(dd, devices, list) {
1663 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1664 char b[BDEVNAME_SIZE];
1667 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1669 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1670 dm_device_name(t->md),
1671 bdevname(dd->dm_dev->bdev, b));
1674 list_for_each_entry(cb, &t->target_callbacks, list)
1675 if (cb->congested_fn)
1676 r |= cb->congested_fn(cb, bdi_bits);
1681 struct mapped_device *dm_table_get_md(struct dm_table *t)
1685 EXPORT_SYMBOL(dm_table_get_md);
1687 void dm_table_run_md_queue_async(struct dm_table *t)
1689 struct mapped_device *md;
1690 struct request_queue *queue;
1691 unsigned long flags;
1693 if (!dm_table_request_based(t))
1696 md = dm_table_get_md(t);
1697 queue = dm_get_md_queue(md);
1700 blk_mq_run_hw_queues(queue, true);
1702 spin_lock_irqsave(queue->queue_lock, flags);
1703 blk_run_queue_async(queue);
1704 spin_unlock_irqrestore(queue->queue_lock, flags);
1708 EXPORT_SYMBOL(dm_table_run_md_queue_async);