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>
23 #include <linux/dax.h>
25 #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 * Similar to ceiling(log_size(n))
34 static unsigned int int_log(unsigned int n, unsigned int base)
39 n = dm_div_up(n, base);
47 * Calculate the index of the child node of the n'th node k'th key.
49 static inline unsigned int get_child(unsigned int n, unsigned int k)
51 return (n * CHILDREN_PER_NODE) + k;
55 * Return the n'th node of level l from table t.
57 static inline sector_t *get_node(struct dm_table *t,
58 unsigned int l, unsigned int n)
60 return t->index[l] + (n * KEYS_PER_NODE);
64 * Return the highest key that you could lookup from the n'th
65 * node on level l of the btree.
67 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
69 for (; l < t->depth - 1; l++)
70 n = get_child(n, CHILDREN_PER_NODE - 1);
72 if (n >= t->counts[l])
73 return (sector_t) - 1;
75 return get_node(t, l, n)[KEYS_PER_NODE - 1];
79 * Fills in a level of the btree based on the highs of the level
82 static int setup_btree_index(unsigned int l, struct dm_table *t)
87 for (n = 0U; n < t->counts[l]; n++) {
88 node = get_node(t, l, n);
90 for (k = 0U; k < KEYS_PER_NODE; k++)
91 node[k] = high(t, l + 1, get_child(n, k));
98 * highs, and targets are managed as dynamic arrays during a
101 static int alloc_targets(struct dm_table *t, unsigned int num)
104 struct dm_target *n_targets;
107 * Allocate both the target array and offset array at once.
109 n_highs = kvcalloc(num, sizeof(struct dm_target) + sizeof(sector_t),
114 n_targets = (struct dm_target *) (n_highs + num);
116 memset(n_highs, -1, sizeof(*n_highs) * num);
119 t->num_allocated = num;
121 t->targets = n_targets;
126 int dm_table_create(struct dm_table **result, fmode_t mode,
127 unsigned num_targets, struct mapped_device *md)
129 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
134 INIT_LIST_HEAD(&t->devices);
137 num_targets = KEYS_PER_NODE;
139 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
146 if (alloc_targets(t, num_targets)) {
151 t->type = DM_TYPE_NONE;
158 static void free_devices(struct list_head *devices, struct mapped_device *md)
160 struct list_head *tmp, *next;
162 list_for_each_safe(tmp, next, devices) {
163 struct dm_dev_internal *dd =
164 list_entry(tmp, struct dm_dev_internal, list);
165 DMWARN("%s: dm_table_destroy: dm_put_device call missing for %s",
166 dm_device_name(md), dd->dm_dev->name);
167 dm_put_table_device(md, dd->dm_dev);
172 static void dm_table_destroy_keyslot_manager(struct dm_table *t);
174 void dm_table_destroy(struct dm_table *t)
181 /* free the indexes */
183 kvfree(t->index[t->depth - 2]);
185 /* free the targets */
186 for (i = 0; i < t->num_targets; i++) {
187 struct dm_target *tgt = t->targets + i;
192 dm_put_target_type(tgt->type);
197 /* free the device list */
198 free_devices(&t->devices, t->md);
200 dm_free_md_mempools(t->mempools);
202 dm_table_destroy_keyslot_manager(t);
208 * See if we've already got a device in the list.
210 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
212 struct dm_dev_internal *dd;
214 list_for_each_entry (dd, l, list)
215 if (dd->dm_dev->bdev->bd_dev == dev)
222 * If possible, this checks an area of a destination device is invalid.
224 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
225 sector_t start, sector_t len, void *data)
227 struct queue_limits *limits = data;
228 struct block_device *bdev = dev->bdev;
230 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
231 unsigned short logical_block_size_sectors =
232 limits->logical_block_size >> SECTOR_SHIFT;
233 char b[BDEVNAME_SIZE];
238 if ((start >= dev_size) || (start + len > dev_size)) {
239 DMWARN("%s: %s too small for target: "
240 "start=%llu, len=%llu, dev_size=%llu",
241 dm_device_name(ti->table->md), bdevname(bdev, b),
242 (unsigned long long)start,
243 (unsigned long long)len,
244 (unsigned long long)dev_size);
249 * If the target is mapped to zoned block device(s), check
250 * that the zones are not partially mapped.
252 if (bdev_zoned_model(bdev) != BLK_ZONED_NONE) {
253 unsigned int zone_sectors = bdev_zone_sectors(bdev);
255 if (start & (zone_sectors - 1)) {
256 DMWARN("%s: start=%llu not aligned to h/w zone size %u of %s",
257 dm_device_name(ti->table->md),
258 (unsigned long long)start,
259 zone_sectors, bdevname(bdev, b));
264 * Note: The last zone of a zoned block device may be smaller
265 * than other zones. So for a target mapping the end of a
266 * zoned block device with such a zone, len would not be zone
267 * aligned. We do not allow such last smaller zone to be part
268 * of the mapping here to ensure that mappings with multiple
269 * devices do not end up with a smaller zone in the middle of
272 if (len & (zone_sectors - 1)) {
273 DMWARN("%s: len=%llu not aligned to h/w zone size %u of %s",
274 dm_device_name(ti->table->md),
275 (unsigned long long)len,
276 zone_sectors, bdevname(bdev, b));
281 if (logical_block_size_sectors <= 1)
284 if (start & (logical_block_size_sectors - 1)) {
285 DMWARN("%s: start=%llu not aligned to h/w "
286 "logical block size %u of %s",
287 dm_device_name(ti->table->md),
288 (unsigned long long)start,
289 limits->logical_block_size, bdevname(bdev, b));
293 if (len & (logical_block_size_sectors - 1)) {
294 DMWARN("%s: len=%llu not aligned to h/w "
295 "logical block size %u of %s",
296 dm_device_name(ti->table->md),
297 (unsigned long long)len,
298 limits->logical_block_size, bdevname(bdev, b));
306 * This upgrades the mode on an already open dm_dev, being
307 * careful to leave things as they were if we fail to reopen the
308 * device and not to touch the existing bdev field in case
309 * it is accessed concurrently.
311 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
312 struct mapped_device *md)
315 struct dm_dev *old_dev, *new_dev;
317 old_dev = dd->dm_dev;
319 r = dm_get_table_device(md, dd->dm_dev->bdev->bd_dev,
320 dd->dm_dev->mode | new_mode, &new_dev);
324 dd->dm_dev = new_dev;
325 dm_put_table_device(md, old_dev);
331 * Convert the path to a device
333 dev_t dm_get_dev_t(const char *path)
337 if (lookup_bdev(path, &dev))
338 dev = name_to_dev_t(path);
341 EXPORT_SYMBOL_GPL(dm_get_dev_t);
344 * Add a device to the list, or just increment the usage count if
345 * it's already present.
347 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
348 struct dm_dev **result)
352 unsigned int major, minor;
354 struct dm_dev_internal *dd;
355 struct dm_table *t = ti->table;
359 if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
360 /* Extract the major/minor numbers */
361 dev = MKDEV(major, minor);
362 if (MAJOR(dev) != major || MINOR(dev) != minor)
365 dev = dm_get_dev_t(path);
370 dd = find_device(&t->devices, dev);
372 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
376 if ((r = dm_get_table_device(t->md, dev, mode, &dd->dm_dev))) {
381 refcount_set(&dd->count, 1);
382 list_add(&dd->list, &t->devices);
385 } else if (dd->dm_dev->mode != (mode | dd->dm_dev->mode)) {
386 r = upgrade_mode(dd, mode, t->md);
390 refcount_inc(&dd->count);
392 *result = dd->dm_dev;
395 EXPORT_SYMBOL(dm_get_device);
397 static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
398 sector_t start, sector_t len, void *data)
400 struct queue_limits *limits = data;
401 struct block_device *bdev = dev->bdev;
402 struct request_queue *q = bdev_get_queue(bdev);
403 char b[BDEVNAME_SIZE];
406 DMWARN("%s: Cannot set limits for nonexistent device %s",
407 dm_device_name(ti->table->md), bdevname(bdev, b));
411 if (blk_stack_limits(limits, &q->limits,
412 get_start_sect(bdev) + start) < 0)
413 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
414 "physical_block_size=%u, logical_block_size=%u, "
415 "alignment_offset=%u, start=%llu",
416 dm_device_name(ti->table->md), bdevname(bdev, b),
417 q->limits.physical_block_size,
418 q->limits.logical_block_size,
419 q->limits.alignment_offset,
420 (unsigned long long) start << SECTOR_SHIFT);
425 * Decrement a device's use count and remove it if necessary.
427 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
430 struct list_head *devices = &ti->table->devices;
431 struct dm_dev_internal *dd;
433 list_for_each_entry(dd, devices, list) {
434 if (dd->dm_dev == d) {
440 DMWARN("%s: device %s not in table devices list",
441 dm_device_name(ti->table->md), d->name);
444 if (refcount_dec_and_test(&dd->count)) {
445 dm_put_table_device(ti->table->md, d);
450 EXPORT_SYMBOL(dm_put_device);
453 * Checks to see if the target joins onto the end of the table.
455 static int adjoin(struct dm_table *table, struct dm_target *ti)
457 struct dm_target *prev;
459 if (!table->num_targets)
462 prev = &table->targets[table->num_targets - 1];
463 return (ti->begin == (prev->begin + prev->len));
467 * Used to dynamically allocate the arg array.
469 * We do first allocation with GFP_NOIO because dm-mpath and dm-thin must
470 * process messages even if some device is suspended. These messages have a
471 * small fixed number of arguments.
473 * On the other hand, dm-switch needs to process bulk data using messages and
474 * excessive use of GFP_NOIO could cause trouble.
476 static char **realloc_argv(unsigned *size, char **old_argv)
483 new_size = *size * 2;
489 argv = kmalloc_array(new_size, sizeof(*argv), gfp);
490 if (argv && old_argv) {
491 memcpy(argv, old_argv, *size * sizeof(*argv));
500 * Destructively splits up the argument list to pass to ctr.
502 int dm_split_args(int *argc, char ***argvp, char *input)
504 char *start, *end = input, *out, **argv = NULL;
505 unsigned array_size = 0;
514 argv = realloc_argv(&array_size, argv);
519 /* Skip whitespace */
520 start = skip_spaces(end);
523 break; /* success, we hit the end */
525 /* 'out' is used to remove any back-quotes */
528 /* Everything apart from '\0' can be quoted */
529 if (*end == '\\' && *(end + 1)) {
536 break; /* end of token */
541 /* have we already filled the array ? */
542 if ((*argc + 1) > array_size) {
543 argv = realloc_argv(&array_size, argv);
548 /* we know this is whitespace */
552 /* terminate the string and put it in the array */
563 * Impose necessary and sufficient conditions on a devices's table such
564 * that any incoming bio which respects its logical_block_size can be
565 * processed successfully. If it falls across the boundary between
566 * two or more targets, the size of each piece it gets split into must
567 * be compatible with the logical_block_size of the target processing it.
569 static int validate_hardware_logical_block_alignment(struct dm_table *table,
570 struct queue_limits *limits)
573 * This function uses arithmetic modulo the logical_block_size
574 * (in units of 512-byte sectors).
576 unsigned short device_logical_block_size_sects =
577 limits->logical_block_size >> SECTOR_SHIFT;
580 * Offset of the start of the next table entry, mod logical_block_size.
582 unsigned short next_target_start = 0;
585 * Given an aligned bio that extends beyond the end of a
586 * target, how many sectors must the next target handle?
588 unsigned short remaining = 0;
590 struct dm_target *ti;
591 struct queue_limits ti_limits;
595 * Check each entry in the table in turn.
597 for (i = 0; i < dm_table_get_num_targets(table); i++) {
598 ti = dm_table_get_target(table, i);
600 blk_set_stacking_limits(&ti_limits);
602 /* combine all target devices' limits */
603 if (ti->type->iterate_devices)
604 ti->type->iterate_devices(ti, dm_set_device_limits,
608 * If the remaining sectors fall entirely within this
609 * table entry are they compatible with its logical_block_size?
611 if (remaining < ti->len &&
612 remaining & ((ti_limits.logical_block_size >>
617 (unsigned short) ((next_target_start + ti->len) &
618 (device_logical_block_size_sects - 1));
619 remaining = next_target_start ?
620 device_logical_block_size_sects - next_target_start : 0;
624 DMWARN("%s: table line %u (start sect %llu len %llu) "
625 "not aligned to h/w logical block size %u",
626 dm_device_name(table->md), i,
627 (unsigned long long) ti->begin,
628 (unsigned long long) ti->len,
629 limits->logical_block_size);
636 int dm_table_add_target(struct dm_table *t, const char *type,
637 sector_t start, sector_t len, char *params)
639 int r = -EINVAL, argc;
641 struct dm_target *tgt;
644 DMERR("%s: target type %s must appear alone in table",
645 dm_device_name(t->md), t->targets->type->name);
649 BUG_ON(t->num_targets >= t->num_allocated);
651 tgt = t->targets + t->num_targets;
652 memset(tgt, 0, sizeof(*tgt));
655 DMERR("%s: zero-length target", dm_device_name(t->md));
659 tgt->type = dm_get_target_type(type);
661 DMERR("%s: %s: unknown target type", dm_device_name(t->md), type);
665 if (dm_target_needs_singleton(tgt->type)) {
666 if (t->num_targets) {
667 tgt->error = "singleton target type must appear alone in table";
673 if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
674 tgt->error = "target type may not be included in a read-only table";
678 if (t->immutable_target_type) {
679 if (t->immutable_target_type != tgt->type) {
680 tgt->error = "immutable target type cannot be mixed with other target types";
683 } else if (dm_target_is_immutable(tgt->type)) {
684 if (t->num_targets) {
685 tgt->error = "immutable target type cannot be mixed with other target types";
688 t->immutable_target_type = tgt->type;
691 if (dm_target_has_integrity(tgt->type))
692 t->integrity_added = 1;
697 tgt->error = "Unknown error";
700 * Does this target adjoin the previous one ?
702 if (!adjoin(t, tgt)) {
703 tgt->error = "Gap in table";
707 r = dm_split_args(&argc, &argv, params);
709 tgt->error = "couldn't split parameters (insufficient memory)";
713 r = tgt->type->ctr(tgt, argc, argv);
718 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
720 if (!tgt->num_discard_bios && tgt->discards_supported)
721 DMWARN("%s: %s: ignoring discards_supported because num_discard_bios is zero.",
722 dm_device_name(t->md), type);
727 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
728 dm_put_target_type(tgt->type);
733 * Target argument parsing helpers.
735 static int validate_next_arg(const struct dm_arg *arg,
736 struct dm_arg_set *arg_set,
737 unsigned *value, char **error, unsigned grouped)
739 const char *arg_str = dm_shift_arg(arg_set);
743 (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
744 (*value < arg->min) ||
745 (*value > arg->max) ||
746 (grouped && arg_set->argc < *value)) {
754 int dm_read_arg(const struct dm_arg *arg, struct dm_arg_set *arg_set,
755 unsigned *value, char **error)
757 return validate_next_arg(arg, arg_set, value, error, 0);
759 EXPORT_SYMBOL(dm_read_arg);
761 int dm_read_arg_group(const struct dm_arg *arg, struct dm_arg_set *arg_set,
762 unsigned *value, char **error)
764 return validate_next_arg(arg, arg_set, value, error, 1);
766 EXPORT_SYMBOL(dm_read_arg_group);
768 const char *dm_shift_arg(struct dm_arg_set *as)
781 EXPORT_SYMBOL(dm_shift_arg);
783 void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
785 BUG_ON(as->argc < num_args);
786 as->argc -= num_args;
787 as->argv += num_args;
789 EXPORT_SYMBOL(dm_consume_args);
791 static bool __table_type_bio_based(enum dm_queue_mode table_type)
793 return (table_type == DM_TYPE_BIO_BASED ||
794 table_type == DM_TYPE_DAX_BIO_BASED);
797 static bool __table_type_request_based(enum dm_queue_mode table_type)
799 return table_type == DM_TYPE_REQUEST_BASED;
802 void dm_table_set_type(struct dm_table *t, enum dm_queue_mode type)
806 EXPORT_SYMBOL_GPL(dm_table_set_type);
808 /* validate the dax capability of the target device span */
809 int device_not_dax_capable(struct dm_target *ti, struct dm_dev *dev,
810 sector_t start, sector_t len, void *data)
812 int blocksize = *(int *) data, id;
815 id = dax_read_lock();
816 rc = !dax_supported(dev->dax_dev, dev->bdev, blocksize, start, len);
822 /* Check devices support synchronous DAX */
823 static int device_not_dax_synchronous_capable(struct dm_target *ti, struct dm_dev *dev,
824 sector_t start, sector_t len, void *data)
826 return !dev->dax_dev || !dax_synchronous(dev->dax_dev);
829 bool dm_table_supports_dax(struct dm_table *t,
830 iterate_devices_callout_fn iterate_fn, int *blocksize)
832 struct dm_target *ti;
835 /* Ensure that all targets support DAX. */
836 for (i = 0; i < dm_table_get_num_targets(t); i++) {
837 ti = dm_table_get_target(t, i);
839 if (!ti->type->direct_access)
842 if (!ti->type->iterate_devices ||
843 ti->type->iterate_devices(ti, iterate_fn, blocksize))
850 static int device_is_rq_stackable(struct dm_target *ti, struct dm_dev *dev,
851 sector_t start, sector_t len, void *data)
853 struct block_device *bdev = dev->bdev;
854 struct request_queue *q = bdev_get_queue(bdev);
856 /* request-based cannot stack on partitions! */
857 if (bdev_is_partition(bdev))
860 return queue_is_mq(q);
863 static int dm_table_determine_type(struct dm_table *t)
866 unsigned bio_based = 0, request_based = 0, hybrid = 0;
867 struct dm_target *tgt;
868 struct list_head *devices = dm_table_get_devices(t);
869 enum dm_queue_mode live_md_type = dm_get_md_type(t->md);
870 int page_size = PAGE_SIZE;
872 if (t->type != DM_TYPE_NONE) {
873 /* target already set the table's type */
874 if (t->type == DM_TYPE_BIO_BASED) {
875 /* possibly upgrade to a variant of bio-based */
876 goto verify_bio_based;
878 BUG_ON(t->type == DM_TYPE_DAX_BIO_BASED);
879 goto verify_rq_based;
882 for (i = 0; i < t->num_targets; i++) {
883 tgt = t->targets + i;
884 if (dm_target_hybrid(tgt))
886 else if (dm_target_request_based(tgt))
891 if (bio_based && request_based) {
892 DMERR("Inconsistent table: different target types"
893 " can't be mixed up");
898 if (hybrid && !bio_based && !request_based) {
900 * The targets can work either way.
901 * Determine the type from the live device.
902 * Default to bio-based if device is new.
904 if (__table_type_request_based(live_md_type))
912 /* We must use this table as bio-based */
913 t->type = DM_TYPE_BIO_BASED;
914 if (dm_table_supports_dax(t, device_not_dax_capable, &page_size) ||
915 (list_empty(devices) && live_md_type == DM_TYPE_DAX_BIO_BASED)) {
916 t->type = DM_TYPE_DAX_BIO_BASED;
921 BUG_ON(!request_based); /* No targets in this table */
923 t->type = DM_TYPE_REQUEST_BASED;
927 * Request-based dm supports only tables that have a single target now.
928 * To support multiple targets, request splitting support is needed,
929 * and that needs lots of changes in the block-layer.
930 * (e.g. request completion process for partial completion.)
932 if (t->num_targets > 1) {
933 DMERR("request-based DM doesn't support multiple targets");
937 if (list_empty(devices)) {
939 struct dm_table *live_table = dm_get_live_table(t->md, &srcu_idx);
941 /* inherit live table's type */
943 t->type = live_table->type;
944 dm_put_live_table(t->md, srcu_idx);
948 tgt = dm_table_get_immutable_target(t);
950 DMERR("table load rejected: immutable target is required");
952 } else if (tgt->max_io_len) {
953 DMERR("table load rejected: immutable target that splits IO is not supported");
957 /* Non-request-stackable devices can't be used for request-based dm */
958 if (!tgt->type->iterate_devices ||
959 !tgt->type->iterate_devices(tgt, device_is_rq_stackable, NULL)) {
960 DMERR("table load rejected: including non-request-stackable devices");
967 enum dm_queue_mode dm_table_get_type(struct dm_table *t)
972 struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
974 return t->immutable_target_type;
977 struct dm_target *dm_table_get_immutable_target(struct dm_table *t)
979 /* Immutable target is implicitly a singleton */
980 if (t->num_targets > 1 ||
981 !dm_target_is_immutable(t->targets[0].type))
987 struct dm_target *dm_table_get_wildcard_target(struct dm_table *t)
989 struct dm_target *ti;
992 for (i = 0; i < dm_table_get_num_targets(t); i++) {
993 ti = dm_table_get_target(t, i);
994 if (dm_target_is_wildcard(ti->type))
1001 bool dm_table_bio_based(struct dm_table *t)
1003 return __table_type_bio_based(dm_table_get_type(t));
1006 bool dm_table_request_based(struct dm_table *t)
1008 return __table_type_request_based(dm_table_get_type(t));
1011 static int dm_table_alloc_md_mempools(struct dm_table *t, struct mapped_device *md)
1013 enum dm_queue_mode type = dm_table_get_type(t);
1014 unsigned per_io_data_size = 0;
1015 unsigned min_pool_size = 0;
1016 struct dm_target *ti;
1019 if (unlikely(type == DM_TYPE_NONE)) {
1020 DMWARN("no table type is set, can't allocate mempools");
1024 if (__table_type_bio_based(type))
1025 for (i = 0; i < t->num_targets; i++) {
1026 ti = t->targets + i;
1027 per_io_data_size = max(per_io_data_size, ti->per_io_data_size);
1028 min_pool_size = max(min_pool_size, ti->num_flush_bios);
1031 t->mempools = dm_alloc_md_mempools(md, type, t->integrity_supported,
1032 per_io_data_size, min_pool_size);
1039 void dm_table_free_md_mempools(struct dm_table *t)
1041 dm_free_md_mempools(t->mempools);
1045 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
1050 static int setup_indexes(struct dm_table *t)
1053 unsigned int total = 0;
1056 /* allocate the space for *all* the indexes */
1057 for (i = t->depth - 2; i >= 0; i--) {
1058 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1059 total += t->counts[i];
1062 indexes = kvcalloc(total, NODE_SIZE, GFP_KERNEL);
1066 /* set up internal nodes, bottom-up */
1067 for (i = t->depth - 2; i >= 0; i--) {
1068 t->index[i] = indexes;
1069 indexes += (KEYS_PER_NODE * t->counts[i]);
1070 setup_btree_index(i, t);
1077 * Builds the btree to index the map.
1079 static int dm_table_build_index(struct dm_table *t)
1082 unsigned int leaf_nodes;
1084 /* how many indexes will the btree have ? */
1085 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1086 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1088 /* leaf layer has already been set up */
1089 t->counts[t->depth - 1] = leaf_nodes;
1090 t->index[t->depth - 1] = t->highs;
1093 r = setup_indexes(t);
1098 static bool integrity_profile_exists(struct gendisk *disk)
1100 return !!blk_get_integrity(disk);
1104 * Get a disk whose integrity profile reflects the table's profile.
1105 * Returns NULL if integrity support was inconsistent or unavailable.
1107 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t)
1109 struct list_head *devices = dm_table_get_devices(t);
1110 struct dm_dev_internal *dd = NULL;
1111 struct gendisk *prev_disk = NULL, *template_disk = NULL;
1114 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1115 struct dm_target *ti = dm_table_get_target(t, i);
1116 if (!dm_target_passes_integrity(ti->type))
1120 list_for_each_entry(dd, devices, list) {
1121 template_disk = dd->dm_dev->bdev->bd_disk;
1122 if (!integrity_profile_exists(template_disk))
1124 else if (prev_disk &&
1125 blk_integrity_compare(prev_disk, template_disk) < 0)
1127 prev_disk = template_disk;
1130 return template_disk;
1134 DMWARN("%s: integrity not set: %s and %s profile mismatch",
1135 dm_device_name(t->md),
1136 prev_disk->disk_name,
1137 template_disk->disk_name);
1142 * Register the mapped device for blk_integrity support if the
1143 * underlying devices have an integrity profile. But all devices may
1144 * not have matching profiles (checking all devices isn't reliable
1145 * during table load because this table may use other DM device(s) which
1146 * must be resumed before they will have an initialized integity
1147 * profile). Consequently, stacked DM devices force a 2 stage integrity
1148 * profile validation: First pass during table load, final pass during
1151 static int dm_table_register_integrity(struct dm_table *t)
1153 struct mapped_device *md = t->md;
1154 struct gendisk *template_disk = NULL;
1156 /* If target handles integrity itself do not register it here. */
1157 if (t->integrity_added)
1160 template_disk = dm_table_get_integrity_disk(t);
1164 if (!integrity_profile_exists(dm_disk(md))) {
1165 t->integrity_supported = true;
1167 * Register integrity profile during table load; we can do
1168 * this because the final profile must match during resume.
1170 blk_integrity_register(dm_disk(md),
1171 blk_get_integrity(template_disk));
1176 * If DM device already has an initialized integrity
1177 * profile the new profile should not conflict.
1179 if (blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1180 DMWARN("%s: conflict with existing integrity profile: "
1181 "%s profile mismatch",
1182 dm_device_name(t->md),
1183 template_disk->disk_name);
1187 /* Preserve existing integrity profile */
1188 t->integrity_supported = true;
1192 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
1194 struct dm_keyslot_manager {
1195 struct blk_keyslot_manager ksm;
1196 struct mapped_device *md;
1199 struct dm_keyslot_evict_args {
1200 const struct blk_crypto_key *key;
1204 static int dm_keyslot_evict_callback(struct dm_target *ti, struct dm_dev *dev,
1205 sector_t start, sector_t len, void *data)
1207 struct dm_keyslot_evict_args *args = data;
1210 err = blk_crypto_evict_key(bdev_get_queue(dev->bdev), args->key);
1213 /* Always try to evict the key from all devices. */
1218 * When an inline encryption key is evicted from a device-mapper device, evict
1219 * it from all the underlying devices.
1221 static int dm_keyslot_evict(struct blk_keyslot_manager *ksm,
1222 const struct blk_crypto_key *key, unsigned int slot)
1224 struct dm_keyslot_manager *dksm = container_of(ksm,
1225 struct dm_keyslot_manager,
1227 struct mapped_device *md = dksm->md;
1228 struct dm_keyslot_evict_args args = { key };
1232 struct dm_target *ti;
1234 t = dm_get_live_table(md, &srcu_idx);
1237 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1238 ti = dm_table_get_target(t, i);
1239 if (!ti->type->iterate_devices)
1241 ti->type->iterate_devices(ti, dm_keyslot_evict_callback, &args);
1243 dm_put_live_table(md, srcu_idx);
1247 static struct blk_ksm_ll_ops dm_ksm_ll_ops = {
1248 .keyslot_evict = dm_keyslot_evict,
1251 static int device_intersect_crypto_modes(struct dm_target *ti,
1252 struct dm_dev *dev, sector_t start,
1253 sector_t len, void *data)
1255 struct blk_keyslot_manager *parent = data;
1256 struct blk_keyslot_manager *child = bdev_get_queue(dev->bdev)->ksm;
1258 blk_ksm_intersect_modes(parent, child);
1262 void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm)
1264 struct dm_keyslot_manager *dksm = container_of(ksm,
1265 struct dm_keyslot_manager,
1271 blk_ksm_destroy(ksm);
1275 static void dm_table_destroy_keyslot_manager(struct dm_table *t)
1277 dm_destroy_keyslot_manager(t->ksm);
1282 * Constructs and initializes t->ksm with a keyslot manager that
1283 * represents the common set of crypto capabilities of the devices
1284 * described by the dm_table. However, if the constructed keyslot
1285 * manager does not support a superset of the crypto capabilities
1286 * supported by the current keyslot manager of the mapped_device,
1287 * it returns an error instead, since we don't support restricting
1288 * crypto capabilities on table changes. Finally, if the constructed
1289 * keyslot manager doesn't actually support any crypto modes at all,
1290 * it just returns NULL.
1292 static int dm_table_construct_keyslot_manager(struct dm_table *t)
1294 struct dm_keyslot_manager *dksm;
1295 struct blk_keyslot_manager *ksm;
1296 struct dm_target *ti;
1298 bool ksm_is_empty = true;
1300 dksm = kmalloc(sizeof(*dksm), GFP_KERNEL);
1306 blk_ksm_init_passthrough(ksm);
1307 ksm->ksm_ll_ops = dm_ksm_ll_ops;
1308 ksm->max_dun_bytes_supported = UINT_MAX;
1309 memset(ksm->crypto_modes_supported, 0xFF,
1310 sizeof(ksm->crypto_modes_supported));
1312 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1313 ti = dm_table_get_target(t, i);
1315 if (!dm_target_passes_crypto(ti->type)) {
1316 blk_ksm_intersect_modes(ksm, NULL);
1319 if (!ti->type->iterate_devices)
1321 ti->type->iterate_devices(ti, device_intersect_crypto_modes,
1325 if (t->md->queue && !blk_ksm_is_superset(ksm, t->md->queue->ksm)) {
1326 DMWARN("Inline encryption capabilities of new DM table were more restrictive than the old table's. This is not supported!");
1327 dm_destroy_keyslot_manager(ksm);
1332 * If the new KSM doesn't actually support any crypto modes, we may as
1333 * well represent it with a NULL ksm.
1335 ksm_is_empty = true;
1336 for (i = 0; i < ARRAY_SIZE(ksm->crypto_modes_supported); i++) {
1337 if (ksm->crypto_modes_supported[i]) {
1338 ksm_is_empty = false;
1344 dm_destroy_keyslot_manager(ksm);
1349 * t->ksm is only set temporarily while the table is being set
1350 * up, and it gets set to NULL after the capabilities have
1351 * been transferred to the request_queue.
1358 static void dm_update_keyslot_manager(struct request_queue *q,
1364 /* Make the ksm less restrictive */
1366 blk_ksm_register(t->ksm, q);
1368 blk_ksm_update_capabilities(q->ksm, t->ksm);
1369 dm_destroy_keyslot_manager(t->ksm);
1374 #else /* CONFIG_BLK_INLINE_ENCRYPTION */
1376 static int dm_table_construct_keyslot_manager(struct dm_table *t)
1381 void dm_destroy_keyslot_manager(struct blk_keyslot_manager *ksm)
1385 static void dm_table_destroy_keyslot_manager(struct dm_table *t)
1389 static void dm_update_keyslot_manager(struct request_queue *q,
1394 #endif /* !CONFIG_BLK_INLINE_ENCRYPTION */
1397 * Prepares the table for use by building the indices,
1398 * setting the type, and allocating mempools.
1400 int dm_table_complete(struct dm_table *t)
1404 r = dm_table_determine_type(t);
1406 DMERR("unable to determine table type");
1410 r = dm_table_build_index(t);
1412 DMERR("unable to build btrees");
1416 r = dm_table_register_integrity(t);
1418 DMERR("could not register integrity profile.");
1422 r = dm_table_construct_keyslot_manager(t);
1424 DMERR("could not construct keyslot manager.");
1428 r = dm_table_alloc_md_mempools(t, t->md);
1430 DMERR("unable to allocate mempools");
1435 static DEFINE_MUTEX(_event_lock);
1436 void dm_table_event_callback(struct dm_table *t,
1437 void (*fn)(void *), void *context)
1439 mutex_lock(&_event_lock);
1441 t->event_context = context;
1442 mutex_unlock(&_event_lock);
1445 void dm_table_event(struct dm_table *t)
1447 mutex_lock(&_event_lock);
1449 t->event_fn(t->event_context);
1450 mutex_unlock(&_event_lock);
1452 EXPORT_SYMBOL(dm_table_event);
1454 inline sector_t dm_table_get_size(struct dm_table *t)
1456 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1458 EXPORT_SYMBOL(dm_table_get_size);
1460 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1462 if (index >= t->num_targets)
1465 return t->targets + index;
1469 * Search the btree for the correct target.
1471 * Caller should check returned pointer for NULL
1472 * to trap I/O beyond end of device.
1474 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1476 unsigned int l, n = 0, k = 0;
1479 if (unlikely(sector >= dm_table_get_size(t)))
1482 for (l = 0; l < t->depth; l++) {
1483 n = get_child(n, k);
1484 node = get_node(t, l, n);
1486 for (k = 0; k < KEYS_PER_NODE; k++)
1487 if (node[k] >= sector)
1491 return &t->targets[(KEYS_PER_NODE * n) + k];
1495 * type->iterate_devices() should be called when the sanity check needs to
1496 * iterate and check all underlying data devices. iterate_devices() will
1497 * iterate all underlying data devices until it encounters a non-zero return
1498 * code, returned by whether the input iterate_devices_callout_fn, or
1499 * iterate_devices() itself internally.
1501 * For some target type (e.g. dm-stripe), one call of iterate_devices() may
1502 * iterate multiple underlying devices internally, in which case a non-zero
1503 * return code returned by iterate_devices_callout_fn will stop the iteration
1506 * Cases requiring _any_ underlying device supporting some kind of attribute,
1507 * should use the iteration structure like dm_table_any_dev_attr(), or call
1508 * it directly. @func should handle semantics of positive examples, e.g.
1509 * capable of something.
1511 * Cases requiring _all_ underlying devices supporting some kind of attribute,
1512 * should use the iteration structure like dm_table_supports_nowait() or
1513 * dm_table_supports_discards(). Or introduce dm_table_all_devs_attr() that
1514 * uses an @anti_func that handle semantics of counter examples, e.g. not
1515 * capable of something. So: return !dm_table_any_dev_attr(t, anti_func, data);
1517 static bool dm_table_any_dev_attr(struct dm_table *t,
1518 iterate_devices_callout_fn func, void *data)
1520 struct dm_target *ti;
1523 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1524 ti = dm_table_get_target(t, i);
1526 if (ti->type->iterate_devices &&
1527 ti->type->iterate_devices(ti, func, data))
1534 static int count_device(struct dm_target *ti, struct dm_dev *dev,
1535 sector_t start, sector_t len, void *data)
1537 unsigned *num_devices = data;
1545 * Check whether a table has no data devices attached using each
1546 * target's iterate_devices method.
1547 * Returns false if the result is unknown because a target doesn't
1548 * support iterate_devices.
1550 bool dm_table_has_no_data_devices(struct dm_table *table)
1552 struct dm_target *ti;
1553 unsigned i, num_devices;
1555 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1556 ti = dm_table_get_target(table, i);
1558 if (!ti->type->iterate_devices)
1562 ti->type->iterate_devices(ti, count_device, &num_devices);
1570 static int device_not_zoned_model(struct dm_target *ti, struct dm_dev *dev,
1571 sector_t start, sector_t len, void *data)
1573 struct request_queue *q = bdev_get_queue(dev->bdev);
1574 enum blk_zoned_model *zoned_model = data;
1576 return blk_queue_zoned_model(q) != *zoned_model;
1580 * Check the device zoned model based on the target feature flag. If the target
1581 * has the DM_TARGET_ZONED_HM feature flag set, host-managed zoned devices are
1582 * also accepted but all devices must have the same zoned model. If the target
1583 * has the DM_TARGET_MIXED_ZONED_MODEL feature set, the devices can have any
1584 * zoned model with all zoned devices having the same zone size.
1586 static bool dm_table_supports_zoned_model(struct dm_table *t,
1587 enum blk_zoned_model zoned_model)
1589 struct dm_target *ti;
1592 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1593 ti = dm_table_get_target(t, i);
1595 if (dm_target_supports_zoned_hm(ti->type)) {
1596 if (!ti->type->iterate_devices ||
1597 ti->type->iterate_devices(ti, device_not_zoned_model,
1600 } else if (!dm_target_supports_mixed_zoned_model(ti->type)) {
1601 if (zoned_model == BLK_ZONED_HM)
1609 static int device_not_matches_zone_sectors(struct dm_target *ti, struct dm_dev *dev,
1610 sector_t start, sector_t len, void *data)
1612 struct request_queue *q = bdev_get_queue(dev->bdev);
1613 unsigned int *zone_sectors = data;
1615 if (!blk_queue_is_zoned(q))
1618 return blk_queue_zone_sectors(q) != *zone_sectors;
1622 * Check consistency of zoned model and zone sectors across all targets. For
1623 * zone sectors, if the destination device is a zoned block device, it shall
1624 * have the specified zone_sectors.
1626 static int validate_hardware_zoned_model(struct dm_table *table,
1627 enum blk_zoned_model zoned_model,
1628 unsigned int zone_sectors)
1630 if (zoned_model == BLK_ZONED_NONE)
1633 if (!dm_table_supports_zoned_model(table, zoned_model)) {
1634 DMERR("%s: zoned model is not consistent across all devices",
1635 dm_device_name(table->md));
1639 /* Check zone size validity and compatibility */
1640 if (!zone_sectors || !is_power_of_2(zone_sectors))
1643 if (dm_table_any_dev_attr(table, device_not_matches_zone_sectors, &zone_sectors)) {
1644 DMERR("%s: zone sectors is not consistent across all zoned devices",
1645 dm_device_name(table->md));
1653 * Establish the new table's queue_limits and validate them.
1655 int dm_calculate_queue_limits(struct dm_table *table,
1656 struct queue_limits *limits)
1658 struct dm_target *ti;
1659 struct queue_limits ti_limits;
1661 enum blk_zoned_model zoned_model = BLK_ZONED_NONE;
1662 unsigned int zone_sectors = 0;
1664 blk_set_stacking_limits(limits);
1666 for (i = 0; i < dm_table_get_num_targets(table); i++) {
1667 blk_set_stacking_limits(&ti_limits);
1669 ti = dm_table_get_target(table, i);
1671 if (!ti->type->iterate_devices)
1672 goto combine_limits;
1675 * Combine queue limits of all the devices this target uses.
1677 ti->type->iterate_devices(ti, dm_set_device_limits,
1680 if (zoned_model == BLK_ZONED_NONE && ti_limits.zoned != BLK_ZONED_NONE) {
1682 * After stacking all limits, validate all devices
1683 * in table support this zoned model and zone sectors.
1685 zoned_model = ti_limits.zoned;
1686 zone_sectors = ti_limits.chunk_sectors;
1689 /* Set I/O hints portion of queue limits */
1690 if (ti->type->io_hints)
1691 ti->type->io_hints(ti, &ti_limits);
1694 * Check each device area is consistent with the target's
1695 * overall queue limits.
1697 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1703 * Merge this target's queue limits into the overall limits
1706 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1707 DMWARN("%s: adding target device "
1708 "(start sect %llu len %llu) "
1709 "caused an alignment inconsistency",
1710 dm_device_name(table->md),
1711 (unsigned long long) ti->begin,
1712 (unsigned long long) ti->len);
1716 * Verify that the zoned model and zone sectors, as determined before
1717 * any .io_hints override, are the same across all devices in the table.
1718 * - this is especially relevant if .io_hints is emulating a disk-managed
1719 * zoned model (aka BLK_ZONED_NONE) on host-managed zoned block devices.
1722 if (limits->zoned != BLK_ZONED_NONE) {
1724 * ...IF the above limits stacking determined a zoned model
1725 * validate that all of the table's devices conform to it.
1727 zoned_model = limits->zoned;
1728 zone_sectors = limits->chunk_sectors;
1730 if (validate_hardware_zoned_model(table, zoned_model, zone_sectors))
1733 return validate_hardware_logical_block_alignment(table, limits);
1737 * Verify that all devices have an integrity profile that matches the
1738 * DM device's registered integrity profile. If the profiles don't
1739 * match then unregister the DM device's integrity profile.
1741 static void dm_table_verify_integrity(struct dm_table *t)
1743 struct gendisk *template_disk = NULL;
1745 if (t->integrity_added)
1748 if (t->integrity_supported) {
1750 * Verify that the original integrity profile
1751 * matches all the devices in this table.
1753 template_disk = dm_table_get_integrity_disk(t);
1754 if (template_disk &&
1755 blk_integrity_compare(dm_disk(t->md), template_disk) >= 0)
1759 if (integrity_profile_exists(dm_disk(t->md))) {
1760 DMWARN("%s: unable to establish an integrity profile",
1761 dm_device_name(t->md));
1762 blk_integrity_unregister(dm_disk(t->md));
1766 static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1767 sector_t start, sector_t len, void *data)
1769 unsigned long flush = (unsigned long) data;
1770 struct request_queue *q = bdev_get_queue(dev->bdev);
1772 return (q->queue_flags & flush);
1775 static bool dm_table_supports_flush(struct dm_table *t, unsigned long flush)
1777 struct dm_target *ti;
1781 * Require at least one underlying device to support flushes.
1782 * t->devices includes internal dm devices such as mirror logs
1783 * so we need to use iterate_devices here, which targets
1784 * supporting flushes must provide.
1786 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1787 ti = dm_table_get_target(t, i);
1789 if (!ti->num_flush_bios)
1792 if (ti->flush_supported)
1795 if (ti->type->iterate_devices &&
1796 ti->type->iterate_devices(ti, device_flush_capable, (void *) flush))
1803 static int device_dax_write_cache_enabled(struct dm_target *ti,
1804 struct dm_dev *dev, sector_t start,
1805 sector_t len, void *data)
1807 struct dax_device *dax_dev = dev->dax_dev;
1812 if (dax_write_cache_enabled(dax_dev))
1817 static int device_is_rotational(struct dm_target *ti, struct dm_dev *dev,
1818 sector_t start, sector_t len, void *data)
1820 struct request_queue *q = bdev_get_queue(dev->bdev);
1822 return !blk_queue_nonrot(q);
1825 static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1826 sector_t start, sector_t len, void *data)
1828 struct request_queue *q = bdev_get_queue(dev->bdev);
1830 return !blk_queue_add_random(q);
1833 static int device_not_write_same_capable(struct dm_target *ti, struct dm_dev *dev,
1834 sector_t start, sector_t len, void *data)
1836 struct request_queue *q = bdev_get_queue(dev->bdev);
1838 return !q->limits.max_write_same_sectors;
1841 static bool dm_table_supports_write_same(struct dm_table *t)
1843 struct dm_target *ti;
1846 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1847 ti = dm_table_get_target(t, i);
1849 if (!ti->num_write_same_bios)
1852 if (!ti->type->iterate_devices ||
1853 ti->type->iterate_devices(ti, device_not_write_same_capable, NULL))
1860 static int device_not_write_zeroes_capable(struct dm_target *ti, struct dm_dev *dev,
1861 sector_t start, sector_t len, void *data)
1863 struct request_queue *q = bdev_get_queue(dev->bdev);
1865 return !q->limits.max_write_zeroes_sectors;
1868 static bool dm_table_supports_write_zeroes(struct dm_table *t)
1870 struct dm_target *ti;
1873 while (i < dm_table_get_num_targets(t)) {
1874 ti = dm_table_get_target(t, i++);
1876 if (!ti->num_write_zeroes_bios)
1879 if (!ti->type->iterate_devices ||
1880 ti->type->iterate_devices(ti, device_not_write_zeroes_capable, NULL))
1887 static int device_not_nowait_capable(struct dm_target *ti, struct dm_dev *dev,
1888 sector_t start, sector_t len, void *data)
1890 struct request_queue *q = bdev_get_queue(dev->bdev);
1892 return !blk_queue_nowait(q);
1895 static bool dm_table_supports_nowait(struct dm_table *t)
1897 struct dm_target *ti;
1900 while (i < dm_table_get_num_targets(t)) {
1901 ti = dm_table_get_target(t, i++);
1903 if (!dm_target_supports_nowait(ti->type))
1906 if (!ti->type->iterate_devices ||
1907 ti->type->iterate_devices(ti, device_not_nowait_capable, NULL))
1914 static int device_not_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1915 sector_t start, sector_t len, void *data)
1917 struct request_queue *q = bdev_get_queue(dev->bdev);
1919 return !blk_queue_discard(q);
1922 static bool dm_table_supports_discards(struct dm_table *t)
1924 struct dm_target *ti;
1927 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1928 ti = dm_table_get_target(t, i);
1930 if (!ti->num_discard_bios)
1934 * Either the target provides discard support (as implied by setting
1935 * 'discards_supported') or it relies on _all_ data devices having
1938 if (!ti->discards_supported &&
1939 (!ti->type->iterate_devices ||
1940 ti->type->iterate_devices(ti, device_not_discard_capable, NULL)))
1947 static int device_not_secure_erase_capable(struct dm_target *ti,
1948 struct dm_dev *dev, sector_t start,
1949 sector_t len, void *data)
1951 struct request_queue *q = bdev_get_queue(dev->bdev);
1953 return !blk_queue_secure_erase(q);
1956 static bool dm_table_supports_secure_erase(struct dm_table *t)
1958 struct dm_target *ti;
1961 for (i = 0; i < dm_table_get_num_targets(t); i++) {
1962 ti = dm_table_get_target(t, i);
1964 if (!ti->num_secure_erase_bios)
1967 if (!ti->type->iterate_devices ||
1968 ti->type->iterate_devices(ti, device_not_secure_erase_capable, NULL))
1975 static int device_requires_stable_pages(struct dm_target *ti,
1976 struct dm_dev *dev, sector_t start,
1977 sector_t len, void *data)
1979 struct request_queue *q = bdev_get_queue(dev->bdev);
1981 return blk_queue_stable_writes(q);
1984 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1985 struct queue_limits *limits)
1987 bool wc = false, fua = false;
1988 int page_size = PAGE_SIZE;
1991 * Copy table's limits to the DM device's request_queue
1993 q->limits = *limits;
1995 if (dm_table_supports_nowait(t))
1996 blk_queue_flag_set(QUEUE_FLAG_NOWAIT, q);
1998 blk_queue_flag_clear(QUEUE_FLAG_NOWAIT, q);
2000 if (!dm_table_supports_discards(t)) {
2001 blk_queue_flag_clear(QUEUE_FLAG_DISCARD, q);
2002 /* Must also clear discard limits... */
2003 q->limits.max_discard_sectors = 0;
2004 q->limits.max_hw_discard_sectors = 0;
2005 q->limits.discard_granularity = 0;
2006 q->limits.discard_alignment = 0;
2007 q->limits.discard_misaligned = 0;
2009 blk_queue_flag_set(QUEUE_FLAG_DISCARD, q);
2011 if (dm_table_supports_secure_erase(t))
2012 blk_queue_flag_set(QUEUE_FLAG_SECERASE, q);
2014 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_WC))) {
2016 if (dm_table_supports_flush(t, (1UL << QUEUE_FLAG_FUA)))
2019 blk_queue_write_cache(q, wc, fua);
2021 if (dm_table_supports_dax(t, device_not_dax_capable, &page_size)) {
2022 blk_queue_flag_set(QUEUE_FLAG_DAX, q);
2023 if (dm_table_supports_dax(t, device_not_dax_synchronous_capable, NULL))
2024 set_dax_synchronous(t->md->dax_dev);
2027 blk_queue_flag_clear(QUEUE_FLAG_DAX, q);
2029 if (dm_table_any_dev_attr(t, device_dax_write_cache_enabled, NULL))
2030 dax_write_cache(t->md->dax_dev, true);
2032 /* Ensure that all underlying devices are non-rotational. */
2033 if (dm_table_any_dev_attr(t, device_is_rotational, NULL))
2034 blk_queue_flag_clear(QUEUE_FLAG_NONROT, q);
2036 blk_queue_flag_set(QUEUE_FLAG_NONROT, q);
2038 if (!dm_table_supports_write_same(t))
2039 q->limits.max_write_same_sectors = 0;
2040 if (!dm_table_supports_write_zeroes(t))
2041 q->limits.max_write_zeroes_sectors = 0;
2043 dm_table_verify_integrity(t);
2046 * Some devices don't use blk_integrity but still want stable pages
2047 * because they do their own checksumming.
2048 * If any underlying device requires stable pages, a table must require
2049 * them as well. Only targets that support iterate_devices are considered:
2050 * don't want error, zero, etc to require stable pages.
2052 if (dm_table_any_dev_attr(t, device_requires_stable_pages, NULL))
2053 blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, q);
2055 blk_queue_flag_clear(QUEUE_FLAG_STABLE_WRITES, q);
2058 * Determine whether or not this queue's I/O timings contribute
2059 * to the entropy pool, Only request-based targets use this.
2060 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
2063 if (blk_queue_add_random(q) &&
2064 dm_table_any_dev_attr(t, device_is_not_random, NULL))
2065 blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, q);
2068 * For a zoned target, the number of zones should be updated for the
2069 * correct value to be exposed in sysfs queue/nr_zones. For a BIO based
2070 * target, this is all that is needed.
2072 #ifdef CONFIG_BLK_DEV_ZONED
2073 if (blk_queue_is_zoned(q)) {
2074 WARN_ON_ONCE(queue_is_mq(q));
2075 q->nr_zones = blkdev_nr_zones(t->md->disk);
2079 dm_update_keyslot_manager(q, t);
2080 blk_queue_update_readahead(q);
2083 unsigned int dm_table_get_num_targets(struct dm_table *t)
2085 return t->num_targets;
2088 struct list_head *dm_table_get_devices(struct dm_table *t)
2093 fmode_t dm_table_get_mode(struct dm_table *t)
2097 EXPORT_SYMBOL(dm_table_get_mode);
2105 static void suspend_targets(struct dm_table *t, enum suspend_mode mode)
2107 int i = t->num_targets;
2108 struct dm_target *ti = t->targets;
2110 lockdep_assert_held(&t->md->suspend_lock);
2115 if (ti->type->presuspend)
2116 ti->type->presuspend(ti);
2118 case PRESUSPEND_UNDO:
2119 if (ti->type->presuspend_undo)
2120 ti->type->presuspend_undo(ti);
2123 if (ti->type->postsuspend)
2124 ti->type->postsuspend(ti);
2131 void dm_table_presuspend_targets(struct dm_table *t)
2136 suspend_targets(t, PRESUSPEND);
2139 void dm_table_presuspend_undo_targets(struct dm_table *t)
2144 suspend_targets(t, PRESUSPEND_UNDO);
2147 void dm_table_postsuspend_targets(struct dm_table *t)
2152 suspend_targets(t, POSTSUSPEND);
2155 int dm_table_resume_targets(struct dm_table *t)
2159 lockdep_assert_held(&t->md->suspend_lock);
2161 for (i = 0; i < t->num_targets; i++) {
2162 struct dm_target *ti = t->targets + i;
2164 if (!ti->type->preresume)
2167 r = ti->type->preresume(ti);
2169 DMERR("%s: %s: preresume failed, error = %d",
2170 dm_device_name(t->md), ti->type->name, r);
2175 for (i = 0; i < t->num_targets; i++) {
2176 struct dm_target *ti = t->targets + i;
2178 if (ti->type->resume)
2179 ti->type->resume(ti);
2185 struct mapped_device *dm_table_get_md(struct dm_table *t)
2189 EXPORT_SYMBOL(dm_table_get_md);
2191 const char *dm_table_device_name(struct dm_table *t)
2193 return dm_device_name(t->md);
2195 EXPORT_SYMBOL_GPL(dm_table_device_name);
2197 void dm_table_run_md_queue_async(struct dm_table *t)
2199 if (!dm_table_request_based(t))
2203 blk_mq_run_hw_queues(t->md->queue, true);
2205 EXPORT_SYMBOL(dm_table_run_md_queue_async);