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 unsigned int major, minor;
398 struct dm_dev_internal *dd;
399 struct dm_table *t = ti->table;
403 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
404 /* Extract the major/minor numbers */
405 dev = MKDEV(major, minor);
406 if (MAJOR(dev) != major || MINOR(dev) != minor)
409 dev = dm_get_dev_t(path);
414 dd = find_device(&t->devices, dev);
416 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
420 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
425 atomic_set(&dd->count, 0);
426 list_add(&dd->list, &t->devices);
428 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
429 r = upgrade_mode(dd, mode, t->md);
433 atomic_inc(&dd->count);
435 *result = dd->dm_dev;
438 EXPORT_SYMBOL(dm_get_device);
440 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
441 sector_t start, sector_t len, void *data)
443 struct queue_limits *limits = data;
444 struct block_device *bdev = dev->bdev;
445 struct request_queue *q = bdev_get_queue(bdev);
446 char b[BDEVNAME_SIZE];
449 DMWARN("%s: Cannot set limits for nonexistent device %s",
450 dm_device_name(ti->table->md), bdevname(bdev, b));
454 if (bdev_stack_limits(limits, bdev, start) < 0)
455 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
456 "physical_block_size=%u, logical_block_size=%u, "
457 "alignment_offset=%u, start=%llu",
458 dm_device_name(ti->table->md), bdevname(bdev, b),
459 q->limits.physical_block_size,
460 q->limits.logical_block_size,
461 q->limits.alignment_offset,
462 (unsigned long long) start << SECTOR_SHIFT);
468 * Decrement a device's use count and remove it if necessary.
470 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
473 struct list_head *devices = &ti->table->devices;
474 struct dm_dev_internal *dd;
476 list_for_each_entry(dd, devices, list) {
477 if (dd->dm_dev == d) {
483 DMWARN("%s: device %s not in table devices list",
484 dm_device_name(ti->table->md), d->name);
487 if (atomic_dec_and_test(&dd->count)) {
488 dm_put_table_device(ti->table->md, d);
493 EXPORT_SYMBOL(dm_put_device);
496 * Checks to see if the target joins onto the end of the table.
498 static int adjoin(struct dm_table *table, struct dm_target *ti)
500 struct dm_target *prev;
502 if (!table->num_targets)
505 prev = &table->targets[table->num_targets - 1];
506 return (ti->begin == (prev->begin + prev->len));
510 * Used to dynamically allocate the arg array.
512 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
513 * process messages even if some device is suspended. These messages have a
514 * small fixed number of arguments.
516 * On the other hand, dm-switch needs to process bulk data using messages and
517 * excessive use of GFP_NOIO could cause trouble.
519 static char **realloc_argv(unsigned *size, char **old_argv)
526 new_size = *size * 2;
532 argv = kmalloc(new_size * sizeof(*argv), gfp);
534 memcpy(argv, old_argv, *size * sizeof(*argv));
543 * Destructively splits up the argument list to pass to ctr.
545 int dm_split_args(int *argc, char ***argvp, char *input)
547 char *start, *end = input, *out, **argv = NULL;
548 unsigned array_size = 0;
557 argv = realloc_argv(&array_size, argv);
562 /* Skip whitespace */
563 start = skip_spaces(end);
566 break; /* success, we hit the end */
568 /* 'out' is used to remove any back-quotes */
571 /* Everything apart from '\0' can be quoted */
572 if (*end == '\\' && *(end + 1)) {
579 break; /* end of token */
584 /* have we already filled the array ? */
585 if ((*argc + 1) > array_size) {
586 argv = realloc_argv(&array_size, argv);
591 /* we know this is whitespace */
595 /* terminate the string and put it in the array */
606 * Impose necessary and sufficient conditions on a devices's table such
607 * that any incoming bio which respects its logical_block_size can be
608 * processed successfully. If it falls across the boundary between
609 * two or more targets, the size of each piece it gets split into must
610 * be compatible with the logical_block_size of the target processing it.
612 static int validate_hardware_logical_block_alignment(struct dm_table *table,
613 struct queue_limits *limits)
616 * This function uses arithmetic modulo the logical_block_size
617 * (in units of 512-byte sectors).
619 unsigned short device_logical_block_size_sects =
620 limits->logical_block_size >> SECTOR_SHIFT;
623 * Offset of the start of the next table entry, mod logical_block_size.
625 unsigned short next_target_start = 0;
628 * Given an aligned bio that extends beyond the end of a
629 * target, how many sectors must the next target handle?
631 unsigned short remaining = 0;
633 struct dm_target *uninitialized_var(ti);
634 struct queue_limits ti_limits;
638 * Check each entry in the table in turn.
640 while (i < dm_table_get_num_targets(table)) {
641 ti = dm_table_get_target(table, i++);
643 blk_set_stacking_limits(&ti_limits);
645 /* combine all target devices' limits */
646 if (ti->type->iterate_devices)
647 ti->type->iterate_devices(ti, dm_set_device_limits,
651 * If the remaining sectors fall entirely within this
652 * table entry are they compatible with its logical_block_size?
654 if (remaining < ti->len &&
655 remaining & ((ti_limits.logical_block_size >>
660 (unsigned short) ((next_target_start + ti->len) &
661 (device_logical_block_size_sects - 1));
662 remaining = next_target_start ?
663 device_logical_block_size_sects - next_target_start : 0;
667 DMWARN("%s: table line %u (start sect %llu len %llu) "
668 "not aligned to h/w logical block size %u",
669 dm_device_name(table->md), i,
670 (unsigned long long) ti->begin,
671 (unsigned long long) ti->len,
672 limits->logical_block_size);
679 int dm_table_add_target(struct dm_table *t, const char *type,
680 sector_t start, sector_t len, char *params)
682 int r = -EINVAL, argc;
684 struct dm_target *tgt;
687 DMERR("%s: target type %s must appear alone in table",
688 dm_device_name(t->md), t->targets->type->name);
692 BUG_ON(t->num_targets >= t->num_allocated);
694 tgt = t->targets + t->num_targets;
695 memset(tgt, 0, sizeof(*tgt));
698 DMERR("%s: zero-length target", dm_device_name(t->md));
702 tgt->type = dm_get_target_type(type);
704 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
709 if (dm_target_needs_singleton(tgt->type)) {
710 if (t->num_targets) {
711 DMERR("%s: target type %s must appear alone in table",
712 dm_device_name(t->md), type);
718 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
719 DMERR("%s: target type %s may not be included in read-only tables",
720 dm_device_name(t->md), type);
724 if (t->immutable_target_type) {
725 if (t->immutable_target_type != tgt->type) {
726 DMERR("%s: immutable target type %s cannot be mixed with other target types",
727 dm_device_name(t->md), t->immutable_target_type->name);
730 } else if (dm_target_is_immutable(tgt->type)) {
731 if (t->num_targets) {
732 DMERR("%s: immutable target type %s cannot be mixed with other target types",
733 dm_device_name(t->md), tgt->type->name);
736 t->immutable_target_type = tgt->type;
742 tgt->error = "Unknown error";
745 * Does this target adjoin the previous one ?
747 if (!adjoin(t, tgt)) {
748 tgt->error = "Gap in table";
753 r = dm_split_args(&argc, &argv, params);
755 tgt->error = "couldn't split parameters (insufficient memory)";
759 r = tgt->type->ctr(tgt, argc, argv);
764 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
766 if (!tgt->num_discard_bios && tgt->discards_supported)
767 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
768 dm_device_name(t->md), type);
773 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
774 dm_put_target_type(tgt->type);
779 * Target argument parsing helpers.
781 static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
782 unsigned *value, char **error, unsigned grouped)
784 const char *arg_str = dm_shift_arg(arg_set);
788 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
789 (*value < arg->min) ||
790 (*value > arg->max) ||
791 (grouped && arg_set->argc < *value)) {
799 int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
800 unsigned *value, char **error)
802 return validate_next_arg(arg, arg_set, value, error, 0);
804 EXPORT_SYMBOL(dm_read_arg);
806 int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
807 unsigned *value, char **error)
809 return validate_next_arg(arg, arg_set, value, error, 1);
811 EXPORT_SYMBOL(dm_read_arg_group);
813 const char *dm_shift_arg(struct dm_arg_set *as)
826 EXPORT_SYMBOL(dm_shift_arg);
828 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
830 BUG_ON(as->argc < num_args);
831 as->argc -= num_args;
832 as->argv += num_args;
834 EXPORT_SYMBOL(dm_consume_args);
836 static bool __table_type_request_based(unsigned table_type)
838 return (table_type == DM_TYPE_REQUEST_BASED ||
839 table_type == DM_TYPE_MQ_REQUEST_BASED);
842 static int dm_table_set_type(struct dm_table *t)
845 unsigned bio_based = 0, request_based = 0, hybrid = 0;
846 bool use_blk_mq = false;
847 struct dm_target *tgt;
848 struct dm_dev_internal *dd;
849 struct list_head *devices;
850 unsigned live_md_type = dm_get_md_type(t->md);
852 for (i = 0; i < t->num_targets; i++) {
853 tgt = t->targets + i;
854 if (dm_target_hybrid(tgt))
856 else if (dm_target_request_based(tgt))
861 if (bio_based && request_based) {
862 DMWARN("Inconsistent table: different target types"
863 " can't be mixed up");
868 if (hybrid && !bio_based && !request_based) {
870 * The targets can work either way.
871 * Determine the type from the live device.
872 * Default to bio-based if device is new.
874 if (__table_type_request_based(live_md_type))
881 /* We must use this table as bio-based */
882 t->type = DM_TYPE_BIO_BASED;
886 BUG_ON(!request_based); /* No targets in this table */
889 * Request-based dm supports only tables that have a single target now.
890 * To support multiple targets, request splitting support is needed,
891 * and that needs lots of changes in the block-layer.
892 * (e.g. request completion process for partial completion.)
894 if (t->num_targets > 1) {
895 DMWARN("Request-based dm doesn't support multiple targets yet");
899 /* Non-request-stackable devices can't be used for request-based dm */
900 devices = dm_table_get_devices(t);
901 list_for_each_entry(dd, devices, list) {
902 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
904 if (!blk_queue_stackable(q)) {
905 DMERR("table load rejected: including"
906 " non-request-stackable devices");
915 /* verify _all_ devices in the table are blk-mq devices */
916 list_for_each_entry(dd, devices, list)
917 if (!bdev_get_queue(dd->dm_dev->bdev)->mq_ops) {
918 DMERR("table load rejected: not all devices"
919 " are blk-mq request-stackable");
922 t->type = DM_TYPE_MQ_REQUEST_BASED;
924 } else if (list_empty(devices) && __table_type_request_based(live_md_type)) {
925 /* inherit live MD type */
926 t->type = live_md_type;
929 t->type = DM_TYPE_REQUEST_BASED;
934 unsigned dm_table_get_type(struct dm_table *t)
939 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
941 return t->immutable_target_type;
944 bool dm_table_request_based(struct dm_table *t)
946 return __table_type_request_based(dm_table_get_type(t));
949 bool dm_table_mq_request_based(struct dm_table *t)
951 return dm_table_get_type(t) == DM_TYPE_MQ_REQUEST_BASED;
954 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
956 unsigned type = dm_table_get_type(t);
957 unsigned per_bio_data_size = 0;
958 struct dm_target *tgt;
961 if (unlikely(type == DM_TYPE_NONE)) {
962 DMWARN("no table type is set, can't allocate mempools");
966 if (type == DM_TYPE_BIO_BASED)
967 for (i = 0; i < t->num_targets; i++) {
968 tgt = t->targets + i;
969 per_bio_data_size = max(per_bio_data_size, tgt->per_bio_data_size);
972 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported, per_bio_data_size);
979 void dm_table_free_md_mempools(struct dm_table *t)
981 dm_free_md_mempools(t->mempools);
985 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
990 static int setup_indexes(struct dm_table *t)
993 unsigned int total = 0;
996 /* allocate the space for *all* the indexes */
997 for (i = t->depth - 2; i >= 0; i--) {
998 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
999 total += t->counts[i];
1002 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1006 /* set up internal nodes, bottom-up */
1007 for (i = t->depth - 2; i >= 0; i--) {
1008 t->index[i] = indexes;
1009 indexes += (KEYS_PER_NODE * t->counts[i]);
1010 setup_btree_index(i, t);
1017 * Builds the btree to index the map.
1019 static int dm_table_build_index(struct dm_table *t)
1022 unsigned int leaf_nodes;
1024 /* how many indexes will the btree have ? */
1025 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1026 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1028 /* leaf layer has already been set up */
1029 t->counts[t->depth - 1] = leaf_nodes;
1030 t->index[t->depth - 1] = t->highs;
1033 r = setup_indexes(t);
1038 static bool integrity_profile_exists(struct gendisk *disk)
1040 return !!blk_get_integrity(disk);
1044 * Get a disk whose integrity profile reflects the table's profile.
1045 * Returns NULL if integrity support was inconsistent or unavailable.
1047 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1049 struct list_head *devices = dm_table_get_devices(t);
1050 struct dm_dev_internal *dd = NULL;
1051 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1053 list_for_each_entry(dd, devices, list) {
1054 template_disk = dd->dm_dev->bdev->bd_disk;
1055 if (!integrity_profile_exists(template_disk))
1057 else if (prev_disk &&
1058 blk_integrity_compare(prev_disk, template_disk) < 0)
1060 prev_disk = template_disk;
1063 return template_disk;
1067 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1068 dm_device_name(t->md),
1069 prev_disk->disk_name,
1070 template_disk->disk_name);
1075 * Register the mapped device for blk_integrity support if the
1076 * underlying devices have an integrity profile. But all devices may
1077 * not have matching profiles (checking all devices isn't reliable
1078 * during table load because this table may use other DM device(s) which
1079 * must be resumed before they will have an initialized integity
1080 * profile). Consequently, stacked DM devices force a 2 stage integrity
1081 * profile validation: First pass during table load, final pass during
1084 static int dm_table_register_integrity(struct dm_table *t)
1086 struct mapped_device *md = t->md;
1087 struct gendisk *template_disk = NULL;
1089 template_disk = dm_table_get_integrity_disk(t);
1093 if (!integrity_profile_exists(dm_disk(md))) {
1094 t->integrity_supported = 1;
1096 * Register integrity profile during table load; we can do
1097 * this because the final profile must match during resume.
1099 blk_integrity_register(dm_disk(md),
1100 blk_get_integrity(template_disk));
1105 * If DM device already has an initialized integrity
1106 * profile the new profile should not conflict.
1108 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1109 DMWARN("%s: conflict with existing integrity profile: "
1110 "%s profile mismatch",
1111 dm_device_name(t->md),
1112 template_disk->disk_name);
1116 /* Preserve existing integrity profile */
1117 t->integrity_supported = 1;
1122 * Prepares the table for use by building the indices,
1123 * setting the type, and allocating mempools.
1125 int dm_table_complete(struct dm_table *t)
1129 r = dm_table_set_type(t);
1131 DMERR("unable to set table type");
1135 r = dm_table_build_index(t);
1137 DMERR("unable to build btrees");
1141 r = dm_table_register_integrity(t);
1143 DMERR("could not register integrity profile.");
1147 r = dm_table_alloc_md_mempools(t, t->md);
1149 DMERR("unable to allocate mempools");
1154 static DEFINE_MUTEX(_event_lock);
1155 void dm_table_event_callback(struct dm_table *t,
1156 void (*fn)(void *), void *context)
1158 mutex_lock(&_event_lock);
1160 t->event_context = context;
1161 mutex_unlock(&_event_lock);
1164 void dm_table_event(struct dm_table *t)
1166 mutex_lock(&_event_lock);
1168 t->event_fn(t->event_context);
1169 mutex_unlock(&_event_lock);
1171 EXPORT_SYMBOL(dm_table_event);
1173 inline sector_t dm_table_get_size(struct dm_table *t)
1175 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1177 EXPORT_SYMBOL(dm_table_get_size);
1179 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1181 if (index >= t->num_targets)
1184 return t->targets + index;
1188 * Search the btree for the correct target.
1190 * Caller should check returned pointer with dm_target_is_valid()
1191 * to trap I/O beyond end of device.
1193 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1195 unsigned int l, n = 0, k = 0;
1198 if (unlikely(sector >= dm_table_get_size(t)))
1199 return &t->targets[t->num_targets];
1201 for (l = 0; l < t->depth; l++) {
1202 n = get_child(n, k);
1203 node = get_node(t, l, n);
1205 for (k = 0; k < KEYS_PER_NODE; k++)
1206 if (node[k] >= sector)
1210 return &t->targets[(KEYS_PER_NODE * n) + k];
1214 * type->iterate_devices() should be called when the sanity check needs to
1215 * iterate and check all underlying data devices. iterate_devices() will
1216 * iterate all underlying data devices until it encounters a non-zero return
1217 * code, returned by whether the input iterate_devices_callout_fn, or
1218 * iterate_devices() itself internally.
1220 * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1221 * iterate multiple underlying devices internally, in which case a non-zero
1222 * return code returned by iterate_devices_callout_fn will stop the iteration
1225 * Cases requiring _any_ underlying device supporting some kind of attribute,
1226 * should use the iteration structure like dm_table_any_dev_attr(), or call
1227 * it directly. @func should handle semantics of positive examples, e.g.
1228 * capable of something.
1230 * Cases requiring _all_ underlying devices supporting some kind of attribute,
1231 * should use the iteration structure like dm_table_supports_nowait() or
1232 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1233 * uses an @anti_func that handle semantics of counter examples, e.g. not
1234 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func);
1236 static bool dm_table_any_dev_attr(struct dm_table *t,
1237 iterate_devices_callout_fn func)
1239 struct dm_target *ti;
1242 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1243 ti = dm_table_get_target(t, i);
1245 if (ti->type->iterate_devices &&
1246 ti->type->iterate_devices(ti, func, NULL))
1253 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1254 sector_t start, sector_t len, void *data)
1256 unsigned *num_devices = data;
1264 * Check whether a table has no data devices attached using each
1265 * target's iterate_devices method.
1266 * Returns false if the result is unknown because a target doesn't
1267 * support iterate_devices.
1269 bool dm_table_has_no_data_devices(struct dm_table *table)
1271 struct dm_target *uninitialized_var(ti);
1272 unsigned i = 0, num_devices = 0;
1274 while (i < dm_table_get_num_targets(table)) {
1275 ti = dm_table_get_target(table, i++);
1277 if (!ti->type->iterate_devices)
1280 ti->type->iterate_devices(ti, count_device, &num_devices);
1289 * Establish the new table's queue_limits and validate them.
1291 int dm_calculate_queue_limits(struct dm_table *table,
1292 struct queue_limits *limits)
1294 struct dm_target *uninitialized_var(ti);
1295 struct queue_limits ti_limits;
1298 blk_set_stacking_limits(limits);
1300 while (i < dm_table_get_num_targets(table)) {
1301 blk_set_stacking_limits(&ti_limits);
1303 ti = dm_table_get_target(table, i++);
1305 if (!ti->type->iterate_devices)
1306 goto combine_limits;
1309 * Combine queue limits of all the devices this target uses.
1311 ti->type->iterate_devices(ti, dm_set_device_limits,
1314 /* Set I/O hints portion of queue limits */
1315 if (ti->type->io_hints)
1316 ti->type->io_hints(ti, &ti_limits);
1319 * Check each device area is consistent with the target's
1320 * overall queue limits.
1322 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1328 * Merge this target's queue limits into the overall limits
1331 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1332 DMWARN("%s: adding target device "
1333 "(start sect %llu len %llu) "
1334 "caused an alignment inconsistency",
1335 dm_device_name(table->md),
1336 (unsigned long long) ti->begin,
1337 (unsigned long long) ti->len);
1340 return validate_hardware_logical_block_alignment(table, limits);
1344 * Verify that all devices have an integrity profile that matches the
1345 * DM device's registered integrity profile. If the profiles don't
1346 * match then unregister the DM device's integrity profile.
1348 static void dm_table_verify_integrity(struct dm_table *t)
1350 struct gendisk *template_disk = NULL;
1352 if (t->integrity_supported) {
1354 * Verify that the original integrity profile
1355 * matches all the devices in this table.
1357 template_disk = dm_table_get_integrity_disk(t);
1358 if (template_disk &&
1359 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1363 if (integrity_profile_exists(dm_disk(t->md))) {
1364 DMWARN("%s: unable to establish an integrity profile",
1365 dm_device_name(t->md));
1366 blk_integrity_unregister(dm_disk(t->md));
1370 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1371 sector_t start, sector_t len, void *data)
1373 unsigned flush = (*(unsigned *)data);
1374 struct request_queue *q = bdev_get_queue(dev->bdev);
1376 return q && (q->flush_flags & flush);
1379 static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1381 struct dm_target *ti;
1385 * Require at least one underlying device to support flushes.
1386 * t->devices includes internal dm devices such as mirror logs
1387 * so we need to use iterate_devices here, which targets
1388 * supporting flushes must provide.
1390 while (i < dm_table_get_num_targets(t)) {
1391 ti = dm_table_get_target(t, i++);
1393 if (!ti->num_flush_bios)
1396 if (ti->flush_supported)
1399 if (ti->type->iterate_devices &&
1400 ti->type->iterate_devices(ti, device_flush_capable, &flush))
1407 static bool dm_table_discard_zeroes_data(struct dm_table *t)
1409 struct dm_target *ti;
1412 /* Ensure that all targets supports discard_zeroes_data. */
1413 while (i < dm_table_get_num_targets(t)) {
1414 ti = dm_table_get_target(t, i++);
1416 if (ti->discard_zeroes_data_unsupported)
1423 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1424 sector_t start, sector_t len, void *data)
1426 struct request_queue *q = bdev_get_queue(dev->bdev);
1428 return q && !blk_queue_nonrot(q);
1431 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1432 sector_t start, sector_t len, void *data)
1434 struct request_queue *q = bdev_get_queue(dev->bdev);
1436 return q && !blk_queue_add_random(q);
1439 static int queue_no_sg_merge(struct dm_target *ti, struct dm_dev *dev,
1440 sector_t start, sector_t len, void *data)
1442 struct request_queue *q = bdev_get_queue(dev->bdev);
1444 return q && test_bit(QUEUE_FLAG_NO_SG_MERGE, &q->queue_flags);
1447 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1448 sector_t start, sector_t len, void *data)
1450 struct request_queue *q = bdev_get_queue(dev->bdev);
1452 return q && !q->limits.max_write_same_sectors;
1455 static bool dm_table_supports_write_same(struct dm_table *t)
1457 struct dm_target *ti;
1460 while (i < dm_table_get_num_targets(t)) {
1461 ti = dm_table_get_target(t, i++);
1463 if (!ti->num_write_same_bios)
1466 if (!ti->type->iterate_devices ||
1467 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1474 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1475 sector_t start, sector_t len, void *data)
1477 struct request_queue *q = bdev_get_queue(dev->bdev);
1479 return q && blk_queue_discard(q);
1482 static bool dm_table_supports_discards(struct dm_table *t)
1484 struct dm_target *ti;
1488 * Unless any target used by the table set discards_supported,
1489 * require at least one underlying device to support discards.
1490 * t->devices includes internal dm devices such as mirror logs
1491 * so we need to use iterate_devices here, which targets
1492 * supporting discard selectively must provide.
1494 while (i < dm_table_get_num_targets(t)) {
1495 ti = dm_table_get_target(t, i++);
1497 if (!ti->num_discard_bios)
1500 if (ti->discards_supported)
1503 if (ti->type->iterate_devices &&
1504 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1511 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1512 struct queue_limits *limits)
1517 * Copy table's limits to the DM device's request_queue
1519 q->limits = *limits;
1521 if (!dm_table_supports_discards(t))
1522 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1524 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1526 if (dm_table_supports_flush(t, REQ_FLUSH)) {
1528 if (dm_table_supports_flush(t, REQ_FUA))
1531 blk_queue_flush(q, flush);
1533 if (!dm_table_discard_zeroes_data(t))
1534 q->limits.discard_zeroes_data = 0;
1536 /* Ensure that all underlying devices are non-rotational. */
1537 if (dm_table_any_dev_attr(t, device_is_rotational))
1538 queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1540 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1542 if (!dm_table_supports_write_same(t))
1543 q->limits.max_write_same_sectors = 0;
1545 if (dm_table_any_dev_attr(t, queue_no_sg_merge))
1546 queue_flag_set_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1548 queue_flag_clear_unlocked(QUEUE_FLAG_NO_SG_MERGE, q);
1550 dm_table_verify_integrity(t);
1553 * Determine whether or not this queue's I/O timings contribute
1554 * to the entropy pool, Only request-based targets use this.
1555 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1558 if (blk_queue_add_random(q) && dm_table_any_dev_attr(t, device_is_not_random))
1559 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1562 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1563 * visible to other CPUs because, once the flag is set, incoming bios
1564 * are processed by request-based dm, which refers to the queue
1566 * Until the flag set, bios are passed to bio-based dm and queued to
1567 * md->deferred where queue settings are not needed yet.
1568 * Those bios are passed to request-based dm at the resume time.
1571 if (dm_table_request_based(t))
1572 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1575 unsigned int dm_table_get_num_targets(struct dm_table *t)
1577 return t->num_targets;
1580 struct list_head *dm_table_get_devices(struct dm_table *t)
1585 fmode_t dm_table_get_mode(struct dm_table *t)
1589 EXPORT_SYMBOL(dm_table_get_mode);
1597 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
1599 int i = t->num_targets;
1600 struct dm_target *ti = t->targets;
1605 if (ti->type->presuspend)
1606 ti->type->presuspend(ti);
1608 case PRESUSPEND_UNDO:
1609 if (ti->type->presuspend_undo)
1610 ti->type->presuspend_undo(ti);
1613 if (ti->type->postsuspend)
1614 ti->type->postsuspend(ti);
1621 void dm_table_presuspend_targets(struct dm_table *t)
1626 suspend_targets(t, PRESUSPEND);
1629 void dm_table_presuspend_undo_targets(struct dm_table *t)
1634 suspend_targets(t, PRESUSPEND_UNDO);
1637 void dm_table_postsuspend_targets(struct dm_table *t)
1642 suspend_targets(t, POSTSUSPEND);
1645 int dm_table_resume_targets(struct dm_table *t)
1649 for (i = 0; i < t->num_targets; i++) {
1650 struct dm_target *ti = t->targets + i;
1652 if (!ti->type->preresume)
1655 r = ti->type->preresume(ti);
1657 DMERR("%s: %s: preresume failed, error = %d",
1658 dm_device_name(t->md), ti->type->name, r);
1663 for (i = 0; i < t->num_targets; i++) {
1664 struct dm_target *ti = t->targets + i;
1666 if (ti->type->resume)
1667 ti->type->resume(ti);
1673 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1675 list_add(&cb->list, &t->target_callbacks);
1677 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1679 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1681 struct dm_dev_internal *dd;
1682 struct list_head *devices = dm_table_get_devices(t);
1683 struct dm_target_callbacks *cb;
1686 list_for_each_entry(dd, devices, list) {
1687 struct request_queue *q = bdev_get_queue(dd->dm_dev->bdev);
1688 char b[BDEVNAME_SIZE];
1691 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1693 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1694 dm_device_name(t->md),
1695 bdevname(dd->dm_dev->bdev, b));
1698 list_for_each_entry(cb, &t->target_callbacks, list)
1699 if (cb->congested_fn)
1700 r |= cb->congested_fn(cb, bdi_bits);
1705 struct mapped_device *dm_table_get_md(struct dm_table *t)
1709 EXPORT_SYMBOL(dm_table_get_md);
1711 void dm_table_run_md_queue_async(struct dm_table *t)
1713 struct mapped_device *md;
1714 struct request_queue *queue;
1715 unsigned long flags;
1717 if (!dm_table_request_based(t))
1720 md = dm_table_get_md(t);
1721 queue = dm_get_md_queue(md);
1724 blk_mq_run_hw_queues(queue, true);
1726 spin_lock_irqsave(queue->queue_lock, flags);
1727 blk_run_queue_async(queue);
1728 spin_unlock_irqrestore(queue->queue_lock, flags);
1732 EXPORT_SYMBOL(dm_table_run_md_queue_async);