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/slab.h>
16 #include <linux/interrupt.h>
17 #include <linux/mutex.h>
18 #include <linux/delay.h>
19 #include <asm/atomic.h>
21 #define DM_MSG_PREFIX "table"
24 #define NODE_SIZE L1_CACHE_BYTES
25 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
26 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
29 * The table has always exactly one reference from either mapped_device->map
30 * or hash_cell->new_map. This reference is not counted in table->holders.
31 * A pair of dm_create_table/dm_destroy_table functions is used for table
32 * creation/destruction.
34 * Temporary references from the other code increase table->holders. A pair
35 * of dm_table_get/dm_table_put functions is used to manipulate it.
37 * When the table is about to be destroyed, we wait for table->holders to
42 struct mapped_device *md;
47 unsigned int counts[MAX_DEPTH]; /* in nodes */
48 sector_t *index[MAX_DEPTH];
50 unsigned int num_targets;
51 unsigned int num_allocated;
53 struct dm_target *targets;
56 * Indicates the rw permissions for the new logical
57 * device. This should be a combination of FMODE_READ
62 /* a list of devices used by this table */
63 struct list_head devices;
66 * These are optimistic limits taken from all the
67 * targets, some targets will need smaller limits.
69 struct io_restrictions limits;
71 /* events get handed up using this callback */
72 void (*event_fn)(void *);
77 * Similar to ceiling(log_size(n))
79 static unsigned int int_log(unsigned int n, unsigned int base)
84 n = dm_div_up(n, base);
92 * Returns the minimum that is _not_ zero, unless both are zero.
94 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
97 * Combine two io_restrictions, always taking the lower value.
99 static void combine_restrictions_low(struct io_restrictions *lhs,
100 struct io_restrictions *rhs)
103 min_not_zero(lhs->max_sectors, rhs->max_sectors);
105 lhs->max_phys_segments =
106 min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments);
108 lhs->max_hw_segments =
109 min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments);
111 lhs->logical_block_size = max(lhs->logical_block_size,
112 rhs->logical_block_size);
114 lhs->max_segment_size =
115 min_not_zero(lhs->max_segment_size, rhs->max_segment_size);
117 lhs->max_hw_sectors =
118 min_not_zero(lhs->max_hw_sectors, rhs->max_hw_sectors);
120 lhs->seg_boundary_mask =
121 min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask);
123 lhs->bounce_pfn = min_not_zero(lhs->bounce_pfn, rhs->bounce_pfn);
125 lhs->no_cluster |= rhs->no_cluster;
129 * Calculate the index of the child node of the n'th node k'th key.
131 static inline unsigned int get_child(unsigned int n, unsigned int k)
133 return (n * CHILDREN_PER_NODE) + k;
137 * Return the n'th node of level l from table t.
139 static inline sector_t *get_node(struct dm_table *t,
140 unsigned int l, unsigned int n)
142 return t->index[l] + (n * KEYS_PER_NODE);
146 * Return the highest key that you could lookup from the n'th
147 * node on level l of the btree.
149 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
151 for (; l < t->depth - 1; l++)
152 n = get_child(n, CHILDREN_PER_NODE - 1);
154 if (n >= t->counts[l])
155 return (sector_t) - 1;
157 return get_node(t, l, n)[KEYS_PER_NODE - 1];
161 * Fills in a level of the btree based on the highs of the level
164 static int setup_btree_index(unsigned int l, struct dm_table *t)
169 for (n = 0U; n < t->counts[l]; n++) {
170 node = get_node(t, l, n);
172 for (k = 0U; k < KEYS_PER_NODE; k++)
173 node[k] = high(t, l + 1, get_child(n, k));
179 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
185 * Check that we're not going to overflow.
187 if (nmemb > (ULONG_MAX / elem_size))
190 size = nmemb * elem_size;
191 addr = vmalloc(size);
193 memset(addr, 0, size);
199 * highs, and targets are managed as dynamic arrays during a
202 static int alloc_targets(struct dm_table *t, unsigned int num)
205 struct dm_target *n_targets;
206 int n = t->num_targets;
209 * Allocate both the target array and offset array at once.
210 * Append an empty entry to catch sectors beyond the end of
213 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
218 n_targets = (struct dm_target *) (n_highs + num);
221 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
222 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
225 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
228 t->num_allocated = num;
230 t->targets = n_targets;
235 int dm_table_create(struct dm_table **result, fmode_t mode,
236 unsigned num_targets, struct mapped_device *md)
238 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
243 INIT_LIST_HEAD(&t->devices);
244 atomic_set(&t->holders, 0);
247 num_targets = KEYS_PER_NODE;
249 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
251 if (alloc_targets(t, num_targets)) {
263 static void free_devices(struct list_head *devices)
265 struct list_head *tmp, *next;
267 list_for_each_safe(tmp, next, devices) {
268 struct dm_dev_internal *dd =
269 list_entry(tmp, struct dm_dev_internal, list);
274 void dm_table_destroy(struct dm_table *t)
278 while (atomic_read(&t->holders))
282 /* free the indexes (see dm_table_complete) */
284 vfree(t->index[t->depth - 2]);
286 /* free the targets */
287 for (i = 0; i < t->num_targets; i++) {
288 struct dm_target *tgt = t->targets + i;
293 dm_put_target_type(tgt->type);
298 /* free the device list */
299 if (t->devices.next != &t->devices) {
300 DMWARN("devices still present during destroy: "
301 "dm_table_remove_device calls missing");
303 free_devices(&t->devices);
309 void dm_table_get(struct dm_table *t)
311 atomic_inc(&t->holders);
314 void dm_table_put(struct dm_table *t)
319 smp_mb__before_atomic_dec();
320 atomic_dec(&t->holders);
324 * Checks to see if we need to extend highs or targets.
326 static inline int check_space(struct dm_table *t)
328 if (t->num_targets >= t->num_allocated)
329 return alloc_targets(t, t->num_allocated * 2);
335 * See if we've already got a device in the list.
337 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
339 struct dm_dev_internal *dd;
341 list_for_each_entry (dd, l, list)
342 if (dd->dm_dev.bdev->bd_dev == dev)
349 * Open a device so we can use it as a map destination.
351 static int open_dev(struct dm_dev_internal *d, dev_t dev,
352 struct mapped_device *md)
354 static char *_claim_ptr = "I belong to device-mapper";
355 struct block_device *bdev;
359 BUG_ON(d->dm_dev.bdev);
361 bdev = open_by_devnum(dev, d->dm_dev.mode);
363 return PTR_ERR(bdev);
364 r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
366 blkdev_put(bdev, d->dm_dev.mode);
368 d->dm_dev.bdev = bdev;
373 * Close a device that we've been using.
375 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
380 bd_release_from_disk(d->dm_dev.bdev, dm_disk(md));
381 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode);
382 d->dm_dev.bdev = NULL;
386 * If possible, this checks an area of a destination device is valid.
388 static int check_device_area(struct dm_dev_internal *dd, sector_t start,
391 sector_t dev_size = i_size_read(dd->dm_dev.bdev->bd_inode) >>
397 return ((start < dev_size) && (len <= (dev_size - start)));
401 * This upgrades the mode on an already open dm_dev, being
402 * careful to leave things as they were if we fail to reopen the
403 * device and not to touch the existing bdev field in case
404 * it is accessed concurrently inside dm_table_any_congested().
406 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
407 struct mapped_device *md)
410 struct dm_dev_internal dd_new, dd_old;
412 dd_new = dd_old = *dd;
414 dd_new.dm_dev.mode |= new_mode;
415 dd_new.dm_dev.bdev = NULL;
417 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
421 dd->dm_dev.mode |= new_mode;
422 close_dev(&dd_old, md);
428 * Add a device to the list, or just increment the usage count if
429 * it's already present.
431 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
432 const char *path, sector_t start, sector_t len,
433 fmode_t mode, struct dm_dev **result)
436 dev_t uninitialized_var(dev);
437 struct dm_dev_internal *dd;
438 unsigned int major, minor;
442 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
443 /* Extract the major/minor numbers */
444 dev = MKDEV(major, minor);
445 if (MAJOR(dev) != major || MINOR(dev) != minor)
448 /* convert the path to a device */
449 struct block_device *bdev = lookup_bdev(path);
452 return PTR_ERR(bdev);
457 dd = find_device(&t->devices, dev);
459 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
463 dd->dm_dev.mode = mode;
464 dd->dm_dev.bdev = NULL;
466 if ((r = open_dev(dd, dev, t->md))) {
471 format_dev_t(dd->dm_dev.name, dev);
473 atomic_set(&dd->count, 0);
474 list_add(&dd->list, &t->devices);
476 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
477 r = upgrade_mode(dd, mode, t->md);
481 atomic_inc(&dd->count);
483 if (!check_device_area(dd, start, len)) {
484 DMWARN("device %s too small for target", path);
485 dm_put_device(ti, &dd->dm_dev);
489 *result = &dd->dm_dev;
494 void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
496 struct request_queue *q = bdev_get_queue(bdev);
497 struct io_restrictions *rs = &ti->limits;
498 char b[BDEVNAME_SIZE];
501 DMWARN("%s: Cannot set limits for nonexistent device %s",
502 dm_device_name(ti->table->md), bdevname(bdev, b));
507 * Combine the device limits low.
509 * FIXME: if we move an io_restriction struct
510 * into q this would just be a call to
511 * combine_restrictions_low()
514 min_not_zero(rs->max_sectors, queue_max_sectors(q));
517 * Check if merge fn is supported.
518 * If not we'll force DM to use PAGE_SIZE or
519 * smaller I/O, just to be safe.
522 if (q->merge_bvec_fn && !ti->type->merge)
524 min_not_zero(rs->max_sectors,
525 (unsigned int) (PAGE_SIZE >> 9));
527 rs->max_phys_segments =
528 min_not_zero(rs->max_phys_segments,
529 queue_max_phys_segments(q));
531 rs->max_hw_segments =
532 min_not_zero(rs->max_hw_segments, queue_max_hw_segments(q));
534 rs->logical_block_size = max(rs->logical_block_size,
535 queue_logical_block_size(q));
537 rs->max_segment_size =
538 min_not_zero(rs->max_segment_size, queue_max_segment_size(q));
541 min_not_zero(rs->max_hw_sectors, queue_max_hw_sectors(q));
543 rs->seg_boundary_mask =
544 min_not_zero(rs->seg_boundary_mask,
545 queue_segment_boundary(q));
547 rs->bounce_pfn = min_not_zero(rs->bounce_pfn, queue_bounce_pfn(q));
549 rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
551 EXPORT_SYMBOL_GPL(dm_set_device_limits);
553 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
554 sector_t len, fmode_t mode, struct dm_dev **result)
556 int r = __table_get_device(ti->table, ti, path,
557 start, len, mode, result);
560 dm_set_device_limits(ti, (*result)->bdev);
566 * Decrement a devices use count and remove it if necessary.
568 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
570 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
573 if (atomic_dec_and_test(&dd->count)) {
574 close_dev(dd, ti->table->md);
581 * Checks to see if the target joins onto the end of the table.
583 static int adjoin(struct dm_table *table, struct dm_target *ti)
585 struct dm_target *prev;
587 if (!table->num_targets)
590 prev = &table->targets[table->num_targets - 1];
591 return (ti->begin == (prev->begin + prev->len));
595 * Used to dynamically allocate the arg array.
597 static char **realloc_argv(unsigned *array_size, char **old_argv)
602 new_size = *array_size ? *array_size * 2 : 64;
603 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
605 memcpy(argv, old_argv, *array_size * sizeof(*argv));
606 *array_size = new_size;
614 * Destructively splits up the argument list to pass to ctr.
616 int dm_split_args(int *argc, char ***argvp, char *input)
618 char *start, *end = input, *out, **argv = NULL;
619 unsigned array_size = 0;
628 argv = realloc_argv(&array_size, argv);
635 /* Skip whitespace */
636 while (*start && isspace(*start))
640 break; /* success, we hit the end */
642 /* 'out' is used to remove any back-quotes */
645 /* Everything apart from '\0' can be quoted */
646 if (*end == '\\' && *(end + 1)) {
653 break; /* end of token */
658 /* have we already filled the array ? */
659 if ((*argc + 1) > array_size) {
660 argv = realloc_argv(&array_size, argv);
665 /* we know this is whitespace */
669 /* terminate the string and put it in the array */
679 static void check_for_valid_limits(struct io_restrictions *rs)
681 if (!rs->max_sectors)
682 rs->max_sectors = SAFE_MAX_SECTORS;
683 if (!rs->max_hw_sectors)
684 rs->max_hw_sectors = SAFE_MAX_SECTORS;
685 if (!rs->max_phys_segments)
686 rs->max_phys_segments = MAX_PHYS_SEGMENTS;
687 if (!rs->max_hw_segments)
688 rs->max_hw_segments = MAX_HW_SEGMENTS;
689 if (!rs->logical_block_size)
690 rs->logical_block_size = 1 << SECTOR_SHIFT;
691 if (!rs->max_segment_size)
692 rs->max_segment_size = MAX_SEGMENT_SIZE;
693 if (!rs->seg_boundary_mask)
694 rs->seg_boundary_mask = BLK_SEG_BOUNDARY_MASK;
699 int dm_table_add_target(struct dm_table *t, const char *type,
700 sector_t start, sector_t len, char *params)
702 int r = -EINVAL, argc;
704 struct dm_target *tgt;
706 if ((r = check_space(t)))
709 tgt = t->targets + t->num_targets;
710 memset(tgt, 0, sizeof(*tgt));
713 DMERR("%s: zero-length target", dm_device_name(t->md));
717 tgt->type = dm_get_target_type(type);
719 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
727 tgt->error = "Unknown error";
730 * Does this target adjoin the previous one ?
732 if (!adjoin(t, tgt)) {
733 tgt->error = "Gap in table";
738 r = dm_split_args(&argc, &argv, params);
740 tgt->error = "couldn't split parameters (insufficient memory)";
744 r = tgt->type->ctr(tgt, argc, argv);
749 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
751 /* FIXME: the plan is to combine high here and then have
752 * the merge fn apply the target level restrictions. */
753 combine_restrictions_low(&t->limits, &tgt->limits);
757 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
758 dm_put_target_type(tgt->type);
762 static int setup_indexes(struct dm_table *t)
765 unsigned int total = 0;
768 /* allocate the space for *all* the indexes */
769 for (i = t->depth - 2; i >= 0; i--) {
770 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
771 total += t->counts[i];
774 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
778 /* set up internal nodes, bottom-up */
779 for (i = t->depth - 2; i >= 0; i--) {
780 t->index[i] = indexes;
781 indexes += (KEYS_PER_NODE * t->counts[i]);
782 setup_btree_index(i, t);
789 * Builds the btree to index the map.
791 int dm_table_complete(struct dm_table *t)
794 unsigned int leaf_nodes;
796 check_for_valid_limits(&t->limits);
798 /* how many indexes will the btree have ? */
799 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
800 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
802 /* leaf layer has already been set up */
803 t->counts[t->depth - 1] = leaf_nodes;
804 t->index[t->depth - 1] = t->highs;
807 r = setup_indexes(t);
812 static DEFINE_MUTEX(_event_lock);
813 void dm_table_event_callback(struct dm_table *t,
814 void (*fn)(void *), void *context)
816 mutex_lock(&_event_lock);
818 t->event_context = context;
819 mutex_unlock(&_event_lock);
822 void dm_table_event(struct dm_table *t)
825 * You can no longer call dm_table_event() from interrupt
826 * context, use a bottom half instead.
828 BUG_ON(in_interrupt());
830 mutex_lock(&_event_lock);
832 t->event_fn(t->event_context);
833 mutex_unlock(&_event_lock);
836 sector_t dm_table_get_size(struct dm_table *t)
838 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
841 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
843 if (index >= t->num_targets)
846 return t->targets + index;
850 * Search the btree for the correct target.
852 * Caller should check returned pointer with dm_target_is_valid()
853 * to trap I/O beyond end of device.
855 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
857 unsigned int l, n = 0, k = 0;
860 for (l = 0; l < t->depth; l++) {
862 node = get_node(t, l, n);
864 for (k = 0; k < KEYS_PER_NODE; k++)
865 if (node[k] >= sector)
869 return &t->targets[(KEYS_PER_NODE * n) + k];
873 * Set the integrity profile for this device if all devices used have
876 static void dm_table_set_integrity(struct dm_table *t)
878 struct list_head *devices = dm_table_get_devices(t);
879 struct dm_dev_internal *prev = NULL, *dd = NULL;
881 if (!blk_get_integrity(dm_disk(t->md)))
884 list_for_each_entry(dd, devices, list) {
886 blk_integrity_compare(prev->dm_dev.bdev->bd_disk,
887 dd->dm_dev.bdev->bd_disk) < 0) {
888 DMWARN("%s: integrity not set: %s and %s mismatch",
889 dm_device_name(t->md),
890 prev->dm_dev.bdev->bd_disk->disk_name,
891 dd->dm_dev.bdev->bd_disk->disk_name);
897 if (!prev || !bdev_get_integrity(prev->dm_dev.bdev))
900 blk_integrity_register(dm_disk(t->md),
901 bdev_get_integrity(prev->dm_dev.bdev));
906 blk_integrity_register(dm_disk(t->md), NULL);
911 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
914 * Make sure we obey the optimistic sub devices
917 blk_queue_max_sectors(q, t->limits.max_sectors);
918 blk_queue_max_phys_segments(q, t->limits.max_phys_segments);
919 blk_queue_max_hw_segments(q, t->limits.max_hw_segments);
920 blk_queue_logical_block_size(q, t->limits.logical_block_size);
921 blk_queue_max_segment_size(q, t->limits.max_segment_size);
922 blk_queue_max_hw_sectors(q, t->limits.max_hw_sectors);
923 blk_queue_segment_boundary(q, t->limits.seg_boundary_mask);
924 blk_queue_bounce_limit(q, t->limits.bounce_pfn);
926 if (t->limits.no_cluster)
927 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
929 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q);
931 dm_table_set_integrity(t);
934 unsigned int dm_table_get_num_targets(struct dm_table *t)
936 return t->num_targets;
939 struct list_head *dm_table_get_devices(struct dm_table *t)
944 fmode_t dm_table_get_mode(struct dm_table *t)
949 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
951 int i = t->num_targets;
952 struct dm_target *ti = t->targets;
956 if (ti->type->postsuspend)
957 ti->type->postsuspend(ti);
958 } else if (ti->type->presuspend)
959 ti->type->presuspend(ti);
965 void dm_table_presuspend_targets(struct dm_table *t)
970 suspend_targets(t, 0);
973 void dm_table_postsuspend_targets(struct dm_table *t)
978 suspend_targets(t, 1);
981 int dm_table_resume_targets(struct dm_table *t)
985 for (i = 0; i < t->num_targets; i++) {
986 struct dm_target *ti = t->targets + i;
988 if (!ti->type->preresume)
991 r = ti->type->preresume(ti);
996 for (i = 0; i < t->num_targets; i++) {
997 struct dm_target *ti = t->targets + i;
999 if (ti->type->resume)
1000 ti->type->resume(ti);
1006 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1008 struct dm_dev_internal *dd;
1009 struct list_head *devices = dm_table_get_devices(t);
1012 list_for_each_entry(dd, devices, list) {
1013 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1014 char b[BDEVNAME_SIZE];
1017 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1019 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1020 dm_device_name(t->md),
1021 bdevname(dd->dm_dev.bdev, b));
1027 void dm_table_unplug_all(struct dm_table *t)
1029 struct dm_dev_internal *dd;
1030 struct list_head *devices = dm_table_get_devices(t);
1032 list_for_each_entry(dd, devices, list) {
1033 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1034 char b[BDEVNAME_SIZE];
1039 DMWARN_LIMIT("%s: Cannot unplug nonexistent device %s",
1040 dm_device_name(t->md),
1041 bdevname(dd->dm_dev.bdev, b));
1045 struct mapped_device *dm_table_get_md(struct dm_table *t)
1052 EXPORT_SYMBOL(dm_vcalloc);
1053 EXPORT_SYMBOL(dm_get_device);
1054 EXPORT_SYMBOL(dm_put_device);
1055 EXPORT_SYMBOL(dm_table_event);
1056 EXPORT_SYMBOL(dm_table_get_size);
1057 EXPORT_SYMBOL(dm_table_get_mode);
1058 EXPORT_SYMBOL(dm_table_get_md);
1059 EXPORT_SYMBOL(dm_table_put);
1060 EXPORT_SYMBOL(dm_table_get);
1061 EXPORT_SYMBOL(dm_table_unplug_all);