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>
22 #define DM_MSG_PREFIX "table"
25 #define NODE_SIZE L1_CACHE_BYTES
26 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
27 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
30 struct mapped_device *md;
35 unsigned int counts[MAX_DEPTH]; /* in nodes */
36 sector_t *index[MAX_DEPTH];
38 unsigned int num_targets;
39 unsigned int num_allocated;
41 struct dm_target *targets;
43 struct target_type *immutable_target_type;
44 unsigned integrity_supported:1;
48 * Indicates the rw permissions for the new logical
49 * device. This should be a combination of FMODE_READ
54 /* a list of devices used by this table */
55 struct list_head devices;
57 /* events get handed up using this callback */
58 void (*event_fn)(void *);
61 struct dm_md_mempools *mempools;
63 struct list_head target_callbacks;
67 * Similar to ceiling(log_size(n))
69 static unsigned int int_log(unsigned int n, unsigned int base)
74 n = dm_div_up(n, base);
82 * Calculate the index of the child node of the n'th node k'th key.
84 static inline unsigned int get_child(unsigned int n, unsigned int k)
86 return (n * CHILDREN_PER_NODE) + k;
90 * Return the n'th node of level l from table t.
92 static inline sector_t *get_node(struct dm_table *t,
93 unsigned int l, unsigned int n)
95 return t->index[l] + (n * KEYS_PER_NODE);
99 * Return the highest key that you could lookup from the n'th
100 * node on level l of the btree.
102 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
104 for (; l < t->depth - 1; l++)
105 n = get_child(n, CHILDREN_PER_NODE - 1);
107 if (n >= t->counts[l])
108 return (sector_t) - 1;
110 return get_node(t, l, n)[KEYS_PER_NODE - 1];
114 * Fills in a level of the btree based on the highs of the level
117 static int setup_btree_index(unsigned int l, struct dm_table *t)
122 for (n = 0U; n < t->counts[l]; n++) {
123 node = get_node(t, l, n);
125 for (k = 0U; k < KEYS_PER_NODE; k++)
126 node[k] = high(t, l + 1, get_child(n, k));
132 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
138 * Check that we're not going to overflow.
140 if (nmemb > (ULONG_MAX / elem_size))
143 size = nmemb * elem_size;
144 addr = vzalloc(size);
148 EXPORT_SYMBOL(dm_vcalloc);
151 * highs, and targets are managed as dynamic arrays during a
154 static int alloc_targets(struct dm_table *t, unsigned int num)
157 struct dm_target *n_targets;
160 * Allocate both the target array and offset array at once.
161 * Append an empty entry to catch sectors beyond the end of
164 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
169 n_targets = (struct dm_target *) (n_highs + num);
171 memset(n_highs, -1, sizeof(*n_highs) * num);
174 t->num_allocated = num;
176 t->targets = n_targets;
181 int dm_table_create(struct dm_table **result, fmode_t mode,
182 unsigned num_targets, struct mapped_device *md)
184 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
189 INIT_LIST_HEAD(&t->devices);
190 INIT_LIST_HEAD(&t->target_callbacks);
193 num_targets = KEYS_PER_NODE;
195 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
202 if (alloc_targets(t, num_targets)) {
213 static void free_devices(struct list_head *devices)
215 struct list_head *tmp, *next;
217 list_for_each_safe(tmp, next, devices) {
218 struct dm_dev_internal *dd =
219 list_entry(tmp, struct dm_dev_internal, list);
220 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
226 void dm_table_destroy(struct dm_table *t)
233 /* free the indexes */
235 vfree(t->index[t->depth - 2]);
237 /* free the targets */
238 for (i = 0; i < t->num_targets; i++) {
239 struct dm_target *tgt = t->targets + i;
244 dm_put_target_type(tgt->type);
249 /* free the device list */
250 free_devices(&t->devices);
252 dm_free_md_mempools(t->mempools);
258 * See if we've already got a device in the list.
260 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
262 struct dm_dev_internal *dd;
264 list_for_each_entry (dd, l, list)
265 if (dd->dm_dev.bdev->bd_dev == dev)
272 * Open a device so we can use it as a map destination.
274 static int open_dev(struct dm_dev_internal *d, dev_t dev,
275 struct mapped_device *md)
277 static char *_claim_ptr = "I belong to device-mapper";
278 struct block_device *bdev;
282 BUG_ON(d->dm_dev.bdev);
284 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
286 return PTR_ERR(bdev);
288 r = bd_link_disk_holder(bdev, dm_disk(md));
290 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
294 d->dm_dev.bdev = bdev;
299 * Close a device that we've been using.
301 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
306 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
307 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
308 d->dm_dev.bdev = NULL;
312 * If possible, this checks an area of a destination device is invalid.
314 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
315 sector_t start, sector_t len, void *data)
317 struct request_queue *q;
318 struct queue_limits *limits = data;
319 struct block_device *bdev = dev->bdev;
321 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
322 unsigned short logical_block_size_sectors =
323 limits->logical_block_size >> SECTOR_SHIFT;
324 char b[BDEVNAME_SIZE];
327 * Some devices exist without request functions,
328 * such as loop devices not yet bound to backing files.
329 * Forbid the use of such devices.
331 q = bdev_get_queue(bdev);
332 if (!q || !q->make_request_fn) {
333 DMWARN("%s: %s is not yet initialised: "
334 "start=%llu, len=%llu, dev_size=%llu",
335 dm_device_name(ti->table->md), bdevname(bdev, b),
336 (unsigned long long)start,
337 (unsigned long long)len,
338 (unsigned long long)dev_size);
345 if ((start >= dev_size) || (start + len > dev_size)) {
346 DMWARN("%s: %s too small for target: "
347 "start=%llu, len=%llu, dev_size=%llu",
348 dm_device_name(ti->table->md), bdevname(bdev, b),
349 (unsigned long long)start,
350 (unsigned long long)len,
351 (unsigned long long)dev_size);
355 if (logical_block_size_sectors <= 1)
358 if (start & (logical_block_size_sectors - 1)) {
359 DMWARN("%s: start=%llu not aligned to h/w "
360 "logical block size %u of %s",
361 dm_device_name(ti->table->md),
362 (unsigned long long)start,
363 limits->logical_block_size, bdevname(bdev, b));
367 if (len & (logical_block_size_sectors - 1)) {
368 DMWARN("%s: len=%llu not aligned to h/w "
369 "logical block size %u of %s",
370 dm_device_name(ti->table->md),
371 (unsigned long long)len,
372 limits->logical_block_size, bdevname(bdev, b));
380 * This upgrades the mode on an already open dm_dev, being
381 * careful to leave things as they were if we fail to reopen the
382 * device and not to touch the existing bdev field in case
383 * it is accessed concurrently inside dm_table_any_congested().
385 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
386 struct mapped_device *md)
389 struct dm_dev_internal dd_new, dd_old;
391 dd_new = dd_old = *dd;
393 dd_new.dm_dev.mode |= new_mode;
394 dd_new.dm_dev.bdev = NULL;
396 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
400 dd->dm_dev.mode |= new_mode;
401 close_dev(&dd_old, md);
407 * Add a device to the list, or just increment the usage count if
408 * it's already present.
410 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
411 struct dm_dev **result)
414 dev_t uninitialized_var(dev);
415 struct dm_dev_internal *dd;
416 unsigned int major, minor;
417 struct dm_table *t = ti->table;
422 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
423 /* Extract the major/minor numbers */
424 dev = MKDEV(major, minor);
425 if (MAJOR(dev) != major || MINOR(dev) != minor)
428 /* convert the path to a device */
429 struct block_device *bdev = lookup_bdev(path);
432 return PTR_ERR(bdev);
437 dd = find_device(&t->devices, dev);
439 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
443 dd->dm_dev.mode = mode;
444 dd->dm_dev.bdev = NULL;
446 if ((r = open_dev(dd, dev, t->md))) {
451 format_dev_t(dd->dm_dev.name, dev);
453 atomic_set(&dd->count, 0);
454 list_add(&dd->list, &t->devices);
456 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
457 r = upgrade_mode(dd, mode, t->md);
461 atomic_inc(&dd->count);
463 *result = &dd->dm_dev;
466 EXPORT_SYMBOL(dm_get_device);
468 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
469 sector_t start, sector_t len, void *data)
471 struct queue_limits *limits = data;
472 struct block_device *bdev = dev->bdev;
473 struct request_queue *q = bdev_get_queue(bdev);
474 char b[BDEVNAME_SIZE];
477 DMWARN("%s: Cannot set limits for nonexistent device %s",
478 dm_device_name(ti->table->md), bdevname(bdev, b));
482 if (bdev_stack_limits(limits, bdev, start) < 0)
483 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
484 "physical_block_size=%u, logical_block_size=%u, "
485 "alignment_offset=%u, start=%llu",
486 dm_device_name(ti->table->md), bdevname(bdev, b),
487 q->limits.physical_block_size,
488 q->limits.logical_block_size,
489 q->limits.alignment_offset,
490 (unsigned long long) start << SECTOR_SHIFT);
493 * Check if merge fn is supported.
494 * If not we'll force DM to use PAGE_SIZE or
495 * smaller I/O, just to be safe.
497 if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
498 blk_limits_max_hw_sectors(limits,
499 (unsigned int) (PAGE_SIZE >> 9));
504 * Decrement a device's use count and remove it if necessary.
506 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
508 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
511 if (atomic_dec_and_test(&dd->count)) {
512 close_dev(dd, ti->table->md);
517 EXPORT_SYMBOL(dm_put_device);
520 * Checks to see if the target joins onto the end of the table.
522 static int adjoin(struct dm_table *table, struct dm_target *ti)
524 struct dm_target *prev;
526 if (!table->num_targets)
529 prev = &table->targets[table->num_targets - 1];
530 return (ti->begin == (prev->begin + prev->len));
534 * Used to dynamically allocate the arg array.
536 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
537 * process messages even if some device is suspended. These messages have a
538 * small fixed number of arguments.
540 * On the other hand, dm-switch needs to process bulk data using messages and
541 * excessive use of GFP_NOIO could cause trouble.
543 static char **realloc_argv(unsigned *array_size, char **old_argv)
550 new_size = *array_size * 2;
556 argv = kmalloc(new_size * sizeof(*argv), gfp);
558 memcpy(argv, old_argv, *array_size * sizeof(*argv));
559 *array_size = new_size;
567 * Destructively splits up the argument list to pass to ctr.
569 int dm_split_args(int *argc, char ***argvp, char *input)
571 char *start, *end = input, *out, **argv = NULL;
572 unsigned array_size = 0;
581 argv = realloc_argv(&array_size, argv);
586 /* Skip whitespace */
587 start = skip_spaces(end);
590 break; /* success, we hit the end */
592 /* 'out' is used to remove any back-quotes */
595 /* Everything apart from '\0' can be quoted */
596 if (*end == '\\' && *(end + 1)) {
603 break; /* end of token */
608 /* have we already filled the array ? */
609 if ((*argc + 1) > array_size) {
610 argv = realloc_argv(&array_size, argv);
615 /* we know this is whitespace */
619 /* terminate the string and put it in the array */
630 * Impose necessary and sufficient conditions on a devices's table such
631 * that any incoming bio which respects its logical_block_size can be
632 * processed successfully. If it falls across the boundary between
633 * two or more targets, the size of each piece it gets split into must
634 * be compatible with the logical_block_size of the target processing it.
636 static int validate_hardware_logical_block_alignment(struct dm_table *table,
637 struct queue_limits *limits)
640 * This function uses arithmetic modulo the logical_block_size
641 * (in units of 512-byte sectors).
643 unsigned short device_logical_block_size_sects =
644 limits->logical_block_size >> SECTOR_SHIFT;
647 * Offset of the start of the next table entry, mod logical_block_size.
649 unsigned short next_target_start = 0;
652 * Given an aligned bio that extends beyond the end of a
653 * target, how many sectors must the next target handle?
655 unsigned short remaining = 0;
657 struct dm_target *uninitialized_var(ti);
658 struct queue_limits ti_limits;
662 * Check each entry in the table in turn.
664 while (i < dm_table_get_num_targets(table)) {
665 ti = dm_table_get_target(table, i++);
667 blk_set_stacking_limits(&ti_limits);
669 /* combine all target devices' limits */
670 if (ti->type->iterate_devices)
671 ti->type->iterate_devices(ti, dm_set_device_limits,
675 * If the remaining sectors fall entirely within this
676 * table entry are they compatible with its logical_block_size?
678 if (remaining < ti->len &&
679 remaining & ((ti_limits.logical_block_size >>
684 (unsigned short) ((next_target_start + ti->len) &
685 (device_logical_block_size_sects - 1));
686 remaining = next_target_start ?
687 device_logical_block_size_sects - next_target_start : 0;
691 DMWARN("%s: table line %u (start sect %llu len %llu) "
692 "not aligned to h/w logical block size %u",
693 dm_device_name(table->md), i,
694 (unsigned long long) ti->begin,
695 (unsigned long long) ti->len,
696 limits->logical_block_size);
703 int dm_table_add_target(struct dm_table *t, const char *type,
704 sector_t start, sector_t len, char *params)
706 int r = -EINVAL, argc;
708 struct dm_target *tgt;
711 DMERR("%s: target type %s must appear alone in table",
712 dm_device_name(t->md), t->targets->type->name);
716 BUG_ON(t->num_targets >= t->num_allocated);
718 tgt = t->targets + t->num_targets;
719 memset(tgt, 0, sizeof(*tgt));
722 DMERR("%s: zero-length target", dm_device_name(t->md));
726 tgt->type = dm_get_target_type(type);
728 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
733 if (dm_target_needs_singleton(tgt->type)) {
734 if (t->num_targets) {
735 DMERR("%s: target type %s must appear alone in table",
736 dm_device_name(t->md), type);
742 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
743 DMERR("%s: target type %s may not be included in read-only tables",
744 dm_device_name(t->md), type);
748 if (t->immutable_target_type) {
749 if (t->immutable_target_type != tgt->type) {
750 DMERR("%s: immutable target type %s cannot be mixed with other target types",
751 dm_device_name(t->md), t->immutable_target_type->name);
754 } else if (dm_target_is_immutable(tgt->type)) {
755 if (t->num_targets) {
756 DMERR("%s: immutable target type %s cannot be mixed with other target types",
757 dm_device_name(t->md), tgt->type->name);
760 t->immutable_target_type = tgt->type;
766 tgt->error = "Unknown error";
769 * Does this target adjoin the previous one ?
771 if (!adjoin(t, tgt)) {
772 tgt->error = "Gap in table";
777 r = dm_split_args(&argc, &argv, params);
779 tgt->error = "couldn't split parameters (insufficient memory)";
783 r = tgt->type->ctr(tgt, argc, argv);
788 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
790 if (!tgt->num_discard_bios && tgt->discards_supported)
791 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
792 dm_device_name(t->md), type);
797 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
798 dm_put_target_type(tgt->type);
803 * Target argument parsing helpers.
805 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
806 unsigned *value, char **error, unsigned grouped)
808 const char *arg_str = dm_shift_arg(arg_set);
812 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
813 (*value < arg->min) ||
814 (*value > arg->max) ||
815 (grouped && arg_set->argc < *value)) {
823 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
824 unsigned *value, char **error)
826 return validate_next_arg(arg, arg_set, value, error, 0);
828 EXPORT_SYMBOL(dm_read_arg);
830 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
831 unsigned *value, char **error)
833 return validate_next_arg(arg, arg_set, value, error, 1);
835 EXPORT_SYMBOL(dm_read_arg_group);
837 const char *dm_shift_arg(struct dm_arg_set *as)
850 EXPORT_SYMBOL(dm_shift_arg);
852 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
854 BUG_ON(as->argc < num_args);
855 as->argc -= num_args;
856 as->argv += num_args;
858 EXPORT_SYMBOL(dm_consume_args);
860 static int dm_table_set_type(struct dm_table *t)
863 unsigned bio_based = 0, request_based = 0, hybrid = 0;
864 struct dm_target *tgt;
865 struct dm_dev_internal *dd;
866 struct list_head *devices;
867 unsigned live_md_type;
869 for (i = 0; i < t->num_targets; i++) {
870 tgt = t->targets + i;
871 if (dm_target_hybrid(tgt))
873 else if (dm_target_request_based(tgt))
878 if (bio_based && request_based) {
879 DMWARN("Inconsistent table: different target types"
880 " can't be mixed up");
885 if (hybrid && !bio_based && !request_based) {
887 * The targets can work either way.
888 * Determine the type from the live device.
889 * Default to bio-based if device is new.
891 live_md_type = dm_get_md_type(t->md);
892 if (live_md_type == DM_TYPE_REQUEST_BASED)
899 /* We must use this table as bio-based */
900 t->type = DM_TYPE_BIO_BASED;
904 BUG_ON(!request_based); /* No targets in this table */
906 /* Non-request-stackable devices can't be used for request-based dm */
907 devices = dm_table_get_devices(t);
908 list_for_each_entry(dd, devices, list) {
909 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
910 DMWARN("table load rejected: including"
911 " non-request-stackable devices");
917 * Request-based dm supports only tables that have a single target now.
918 * To support multiple targets, request splitting support is needed,
919 * and that needs lots of changes in the block-layer.
920 * (e.g. request completion process for partial completion.)
922 if (t->num_targets > 1) {
923 DMWARN("Request-based dm doesn't support multiple targets yet");
927 t->type = DM_TYPE_REQUEST_BASED;
932 unsigned dm_table_get_type(struct dm_table *t)
937 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
939 return t->immutable_target_type;
942 bool dm_table_request_based(struct dm_table *t)
944 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
947 static int dm_table_alloc_md_mempools(struct dm_table *t)
949 unsigned type = dm_table_get_type(t);
950 unsigned per_bio_data_size = 0;
951 struct dm_target *tgt;
954 if (unlikely(type == DM_TYPE_NONE)) {
955 DMWARN("no table type is set, can't allocate mempools");
959 if (type == DM_TYPE_BIO_BASED)
960 for (i = 0; i < t->num_targets; i++) {
961 tgt = t->targets + i;
962 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
965 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported, per_bio_data_size);
972 void dm_table_free_md_mempools(struct dm_table *t)
974 dm_free_md_mempools(t->mempools);
978 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
983 static int setup_indexes(struct dm_table *t)
986 unsigned int total = 0;
989 /* allocate the space for *all* the indexes */
990 for (i = t->depth - 2; i >= 0; i--) {
991 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
992 total += t->counts[i];
995 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
999 /* set up internal nodes, bottom-up */
1000 for (i = t->depth - 2; i >= 0; i--) {
1001 t->index[i] = indexes;
1002 indexes += (KEYS_PER_NODE * t->counts[i]);
1003 setup_btree_index(i, t);
1010 * Builds the btree to index the map.
1012 static int dm_table_build_index(struct dm_table *t)
1015 unsigned int leaf_nodes;
1017 /* how many indexes will the btree have ? */
1018 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1019 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1021 /* leaf layer has already been set up */
1022 t->counts[t->depth - 1] = leaf_nodes;
1023 t->index[t->depth - 1] = t->highs;
1026 r = setup_indexes(t);
1032 * Get a disk whose integrity profile reflects the table's profile.
1033 * If %match_all is true, all devices' profiles must match.
1034 * If %match_all is false, all devices must at least have an
1035 * allocated integrity profile; but uninitialized is ok.
1036 * Returns NULL if integrity support was inconsistent or unavailable.
1038 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 (!blk_get_integrity(template_disk))
1049 if (!match_all && !blk_integrity_is_initialized(template_disk))
1050 continue; /* skip uninitialized profiles */
1051 else if (prev_disk &&
1052 blk_integrity_compare(prev_disk, template_disk) < 0)
1054 prev_disk = template_disk;
1057 return template_disk;
1061 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1062 dm_device_name(t->md),
1063 prev_disk->disk_name,
1064 template_disk->disk_name);
1069 * Register the mapped device for blk_integrity support if
1070 * the underlying devices have an integrity profile. But all devices
1071 * may not have matching profiles (checking all devices isn't reliable
1072 * during table load because this table may use other DM device(s) which
1073 * must be resumed before they will have an initialized integity profile).
1074 * Stacked DM devices force a 2 stage integrity profile validation:
1075 * 1 - during load, validate all initialized integrity profiles match
1076 * 2 - during resume, validate all integrity profiles match
1078 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1080 struct gendisk *template_disk = NULL;
1082 template_disk = dm_table_get_integrity_disk(t, false);
1086 if (!blk_integrity_is_initialized(dm_disk(md))) {
1087 t->integrity_supported = 1;
1088 return blk_integrity_register(dm_disk(md), NULL);
1092 * If DM device already has an initalized integrity
1093 * profile the new profile should not conflict.
1095 if (blk_integrity_is_initialized(template_disk) &&
1096 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1097 DMWARN("%s: conflict with existing integrity profile: "
1098 "%s profile mismatch",
1099 dm_device_name(t->md),
1100 template_disk->disk_name);
1104 /* Preserve existing initialized integrity profile */
1105 t->integrity_supported = 1;
1110 * Prepares the table for use by building the indices,
1111 * setting the type, and allocating mempools.
1113 int dm_table_complete(struct dm_table *t)
1117 r = dm_table_set_type(t);
1119 DMERR("unable to set table type");
1123 r = dm_table_build_index(t);
1125 DMERR("unable to build btrees");
1129 r = dm_table_prealloc_integrity(t, t->md);
1131 DMERR("could not register integrity profile.");
1135 r = dm_table_alloc_md_mempools(t);
1137 DMERR("unable to allocate mempools");
1142 static DEFINE_MUTEX(_event_lock);
1143 void dm_table_event_callback(struct dm_table *t,
1144 void (*fn)(void *), void *context)
1146 mutex_lock(&_event_lock);
1148 t->event_context = context;
1149 mutex_unlock(&_event_lock);
1152 void dm_table_event(struct dm_table *t)
1155 * You can no longer call dm_table_event() from interrupt
1156 * context, use a bottom half instead.
1158 BUG_ON(in_interrupt());
1160 mutex_lock(&_event_lock);
1162 t->event_fn(t->event_context);
1163 mutex_unlock(&_event_lock);
1165 EXPORT_SYMBOL(dm_table_event);
1167 sector_t dm_table_get_size(struct dm_table *t)
1169 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1171 EXPORT_SYMBOL(dm_table_get_size);
1173 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1175 if (index >= t->num_targets)
1178 return t->targets + index;
1182 * Search the btree for the correct target.
1184 * Caller should check returned pointer with dm_target_is_valid()
1185 * to trap I/O beyond end of device.
1187 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1189 unsigned int l, n = 0, k = 0;
1192 for (l = 0; l < t->depth; l++) {
1193 n = get_child(n, k);
1194 node = get_node(t, l, n);
1196 for (k = 0; k < KEYS_PER_NODE; k++)
1197 if (node[k] >= sector)
1201 return &t->targets[(KEYS_PER_NODE * n) + k];
1204 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1205 sector_t start, sector_t len, void *data)
1207 unsigned *num_devices = data;
1215 * Check whether a table has no data devices attached using each
1216 * target's iterate_devices method.
1217 * Returns false if the result is unknown because a target doesn't
1218 * support iterate_devices.
1220 bool dm_table_has_no_data_devices(struct dm_table *table)
1222 struct dm_target *uninitialized_var(ti);
1223 unsigned i = 0, num_devices = 0;
1225 while (i < dm_table_get_num_targets(table)) {
1226 ti = dm_table_get_target(table, i++);
1228 if (!ti->type->iterate_devices)
1231 ti->type->iterate_devices(ti, count_device, &num_devices);
1240 * Establish the new table's queue_limits and validate them.
1242 int dm_calculate_queue_limits(struct dm_table *table,
1243 struct queue_limits *limits)
1245 struct dm_target *uninitialized_var(ti);
1246 struct queue_limits ti_limits;
1249 blk_set_stacking_limits(limits);
1251 while (i < dm_table_get_num_targets(table)) {
1252 blk_set_stacking_limits(&ti_limits);
1254 ti = dm_table_get_target(table, i++);
1256 if (!ti->type->iterate_devices)
1257 goto combine_limits;
1260 * Combine queue limits of all the devices this target uses.
1262 ti->type->iterate_devices(ti, dm_set_device_limits,
1265 /* Set I/O hints portion of queue limits */
1266 if (ti->type->io_hints)
1267 ti->type->io_hints(ti, &ti_limits);
1270 * Check each device area is consistent with the target's
1271 * overall queue limits.
1273 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1279 * Merge this target's queue limits into the overall limits
1282 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1283 DMWARN("%s: adding target device "
1284 "(start sect %llu len %llu) "
1285 "caused an alignment inconsistency",
1286 dm_device_name(table->md),
1287 (unsigned long long) ti->begin,
1288 (unsigned long long) ti->len);
1291 return validate_hardware_logical_block_alignment(table, limits);
1295 * Set the integrity profile for this device if all devices used have
1296 * matching profiles. We're quite deep in the resume path but still
1297 * don't know if all devices (particularly DM devices this device
1298 * may be stacked on) have matching profiles. Even if the profiles
1299 * don't match we have no way to fail (to resume) at this point.
1301 static void dm_table_set_integrity(struct dm_table *t)
1303 struct gendisk *template_disk = NULL;
1305 if (!blk_get_integrity(dm_disk(t->md)))
1308 template_disk = dm_table_get_integrity_disk(t, true);
1310 blk_integrity_register(dm_disk(t->md),
1311 blk_get_integrity(template_disk));
1312 else if (blk_integrity_is_initialized(dm_disk(t->md)))
1313 DMWARN("%s: device no longer has a valid integrity profile",
1314 dm_device_name(t->md));
1316 DMWARN("%s: unable to establish an integrity profile",
1317 dm_device_name(t->md));
1320 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1321 sector_t start, sector_t len, void *data)
1323 unsigned flush = (*(unsigned *)data);
1324 struct request_queue *q = bdev_get_queue(dev->bdev);
1326 return q && (q->flush_flags & flush);
1329 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1331 struct dm_target *ti;
1335 * Require at least one underlying device to support flushes.
1336 * t->devices includes internal dm devices such as mirror logs
1337 * so we need to use iterate_devices here, which targets
1338 * supporting flushes must provide.
1340 while (i < dm_table_get_num_targets(t)) {
1341 ti = dm_table_get_target(t, i++);
1343 if (!ti->num_flush_bios)
1346 if (ti->flush_supported)
1349 if (ti->type->iterate_devices &&
1350 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1357 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1359 struct dm_target *ti;
1362 /* Ensure that all targets supports discard_zeroes_data. */
1363 while (i < dm_table_get_num_targets(t)) {
1364 ti = dm_table_get_target(t, i++);
1366 if (ti->discard_zeroes_data_unsupported)
1373 static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1374 sector_t start, sector_t len, void *data)
1376 struct request_queue *q = bdev_get_queue(dev->bdev);
1378 return q && blk_queue_nonrot(q);
1381 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1382 sector_t start, sector_t len, void *data)
1384 struct request_queue *q = bdev_get_queue(dev->bdev);
1386 return q && !blk_queue_add_random(q);
1389 static int queue_supports_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1390 sector_t start, sector_t len, void *data)
1392 struct request_queue *q = bdev_get_queue(dev->bdev);
1394 return q && !test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1397 static bool dm_table_all_devices_attribute(struct dm_table *t,
1398 iterate_devices_callout_fn func)
1400 struct dm_target *ti;
1403 while (i < dm_table_get_num_targets(t)) {
1404 ti = dm_table_get_target(t, i++);
1406 if (!ti->type->iterate_devices ||
1407 !ti->type->iterate_devices(ti, func, NULL))
1414 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1415 sector_t start, sector_t len, void *data)
1417 struct request_queue *q = bdev_get_queue(dev->bdev);
1419 return q && !q->limits.max_write_same_sectors;
1422 static bool dm_table_supports_write_same(struct dm_table *t)
1424 struct dm_target *ti;
1427 while (i < dm_table_get_num_targets(t)) {
1428 ti = dm_table_get_target(t, i++);
1430 if (!ti->num_write_same_bios)
1433 if (!ti->type->iterate_devices ||
1434 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1441 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1442 sector_t start, sector_t len, void *data)
1444 struct request_queue *q = bdev_get_queue(dev->bdev);
1446 return q && blk_queue_discard(q);
1449 static bool dm_table_supports_discards(struct dm_table *t)
1451 struct dm_target *ti;
1455 * Unless any target used by the table set discards_supported,
1456 * require at least one underlying device to support discards.
1457 * t->devices includes internal dm devices such as mirror logs
1458 * so we need to use iterate_devices here, which targets
1459 * supporting discard selectively must provide.
1461 while (i < dm_table_get_num_targets(t)) {
1462 ti = dm_table_get_target(t, i++);
1464 if (!ti->num_discard_bios)
1467 if (ti->discards_supported)
1470 if (ti->type->iterate_devices &&
1471 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1478 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1479 struct queue_limits *limits)
1484 * Copy table's limits to the DM device's request_queue
1486 q->limits = *limits;
1488 if (!dm_table_supports_discards(t))
1489 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1491 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1493 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1495 if (dm_table_supports_flush(t, REQ_FUA))
1498 blk_queue_flush(q, flush);
1500 if (!dm_table_discard_zeroes_data(t))
1501 q->limits.discard_zeroes_data = 0;
1503 /* Ensure that all underlying devices are non-rotational. */
1504 if (dm_table_all_devices_attribute(t, device_is_nonrot))
1505 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1507 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1509 if (!dm_table_supports_write_same(t))
1510 q->limits.max_write_same_sectors = 0;
1512 if (dm_table_all_devices_attribute(t, queue_supports_sg_merge))
1513 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1515 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1517 dm_table_set_integrity(t);
1520 * Determine whether or not this queue's I/O timings contribute
1521 * to the entropy pool, Only request-based targets use this.
1522 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1525 if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1526 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1529 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1530 * visible to other CPUs because, once the flag is set, incoming bios
1531 * are processed by request-based dm, which refers to the queue
1533 * Until the flag set, bios are passed to bio-based dm and queued to
1534 * md->deferred where queue settings are not needed yet.
1535 * Those bios are passed to request-based dm at the resume time.
1538 if (dm_table_request_based(t))
1539 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1542 unsigned int dm_table_get_num_targets(struct dm_table *t)
1544 return t->num_targets;
1547 struct list_head *dm_table_get_devices(struct dm_table *t)
1552 fmode_t dm_table_get_mode(struct dm_table *t)
1556 EXPORT_SYMBOL(dm_table_get_mode);
1558 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1560 int i = t->num_targets;
1561 struct dm_target *ti = t->targets;
1565 if (ti->type->postsuspend)
1566 ti->type->postsuspend(ti);
1567 } else if (ti->type->presuspend)
1568 ti->type->presuspend(ti);
1574 void dm_table_presuspend_targets(struct dm_table *t)
1579 suspend_targets(t, 0);
1582 void dm_table_postsuspend_targets(struct dm_table *t)
1587 suspend_targets(t, 1);
1590 int dm_table_resume_targets(struct dm_table *t)
1594 for (i = 0; i < t->num_targets; i++) {
1595 struct dm_target *ti = t->targets + i;
1597 if (!ti->type->preresume)
1600 r = ti->type->preresume(ti);
1602 DMERR("%s: %s: preresume failed, error = %d",
1603 dm_device_name(t->md), ti->type->name, r);
1608 for (i = 0; i < t->num_targets; i++) {
1609 struct dm_target *ti = t->targets + i;
1611 if (ti->type->resume)
1612 ti->type->resume(ti);
1618 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1620 list_add(&cb->list, &t->target_callbacks);
1622 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1624 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1626 struct dm_dev_internal *dd;
1627 struct list_head *devices = dm_table_get_devices(t);
1628 struct dm_target_callbacks *cb;
1631 list_for_each_entry(dd, devices, list) {
1632 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1633 char b[BDEVNAME_SIZE];
1636 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1638 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1639 dm_device_name(t->md),
1640 bdevname(dd->dm_dev.bdev, b));
1643 list_for_each_entry(cb, &t->target_callbacks, list)
1644 if (cb->congested_fn)
1645 r |= cb->congested_fn(cb, bdi_bits);
1650 int dm_table_any_busy_target(struct dm_table *t)
1653 struct dm_target *ti;
1655 for (i = 0; i < t->num_targets; i++) {
1656 ti = t->targets + i;
1657 if (ti->type->busy && ti->type->busy(ti))
1664 struct mapped_device *dm_table_get_md(struct dm_table *t)
1668 EXPORT_SYMBOL(dm_table_get_md);
1670 void dm_table_run_md_queue_async(struct dm_table *t)
1672 struct mapped_device *md;
1673 struct request_queue *queue;
1674 unsigned long flags;
1676 if (!dm_table_request_based(t))
1679 md = dm_table_get_md(t);
1680 queue = dm_get_md_queue(md);
1682 spin_lock_irqsave(queue->queue_lock, flags);
1683 blk_run_queue_async(queue);
1684 spin_unlock_irqrestore(queue->queue_lock, flags);
1687 EXPORT_SYMBOL(dm_table_run_md_queue_async);