2 * Copyright (C) 2001 Sistina Software (UK) Limited.
3 * Copyright (C) 2004 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 <asm/atomic.h>
20 #define DM_MSG_PREFIX "table"
23 #define NODE_SIZE L1_CACHE_BYTES
24 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
25 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
28 struct mapped_device *md;
33 unsigned int counts[MAX_DEPTH]; /* in nodes */
34 sector_t *index[MAX_DEPTH];
36 unsigned int num_targets;
37 unsigned int num_allocated;
39 struct dm_target *targets;
42 * Indicates the rw permissions for the new logical
43 * device. This should be a combination of FMODE_READ
48 /* a list of devices used by this table */
49 struct list_head devices;
52 * These are optimistic limits taken from all the
53 * targets, some targets will need smaller limits.
55 struct io_restrictions limits;
57 /* events get handed up using this callback */
58 void (*event_fn)(void *);
63 * Similar to ceiling(log_size(n))
65 static unsigned int int_log(unsigned int n, unsigned int base)
70 n = dm_div_up(n, base);
78 * Returns the minimum that is _not_ zero, unless both are zero.
80 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
83 * Combine two io_restrictions, always taking the lower value.
85 static void combine_restrictions_low(struct io_restrictions *lhs,
86 struct io_restrictions *rhs)
89 min_not_zero(lhs->max_sectors, rhs->max_sectors);
91 lhs->max_phys_segments =
92 min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments);
94 lhs->max_hw_segments =
95 min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments);
97 lhs->hardsect_size = max(lhs->hardsect_size, rhs->hardsect_size);
99 lhs->max_segment_size =
100 min_not_zero(lhs->max_segment_size, rhs->max_segment_size);
102 lhs->max_hw_sectors =
103 min_not_zero(lhs->max_hw_sectors, rhs->max_hw_sectors);
105 lhs->seg_boundary_mask =
106 min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask);
108 lhs->bounce_pfn = min_not_zero(lhs->bounce_pfn, rhs->bounce_pfn);
110 lhs->no_cluster |= rhs->no_cluster;
114 * Calculate the index of the child node of the n'th node k'th key.
116 static inline unsigned int get_child(unsigned int n, unsigned int k)
118 return (n * CHILDREN_PER_NODE) + k;
122 * Return the n'th node of level l from table t.
124 static inline sector_t *get_node(struct dm_table *t,
125 unsigned int l, unsigned int n)
127 return t->index[l] + (n * KEYS_PER_NODE);
131 * Return the highest key that you could lookup from the n'th
132 * node on level l of the btree.
134 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
136 for (; l < t->depth - 1; l++)
137 n = get_child(n, CHILDREN_PER_NODE - 1);
139 if (n >= t->counts[l])
140 return (sector_t) - 1;
142 return get_node(t, l, n)[KEYS_PER_NODE - 1];
146 * Fills in a level of the btree based on the highs of the level
149 static int setup_btree_index(unsigned int l, struct dm_table *t)
154 for (n = 0U; n < t->counts[l]; n++) {
155 node = get_node(t, l, n);
157 for (k = 0U; k < KEYS_PER_NODE; k++)
158 node[k] = high(t, l + 1, get_child(n, k));
164 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
170 * Check that we're not going to overflow.
172 if (nmemb > (ULONG_MAX / elem_size))
175 size = nmemb * elem_size;
176 addr = vmalloc(size);
178 memset(addr, 0, size);
184 * highs, and targets are managed as dynamic arrays during a
187 static int alloc_targets(struct dm_table *t, unsigned int num)
190 struct dm_target *n_targets;
191 int n = t->num_targets;
194 * Allocate both the target array and offset array at once.
195 * Append an empty entry to catch sectors beyond the end of
198 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
203 n_targets = (struct dm_target *) (n_highs + num);
206 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
207 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
210 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
213 t->num_allocated = num;
215 t->targets = n_targets;
220 int dm_table_create(struct dm_table **result, int mode,
221 unsigned num_targets, struct mapped_device *md)
223 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
228 INIT_LIST_HEAD(&t->devices);
229 atomic_set(&t->holders, 1);
232 num_targets = KEYS_PER_NODE;
234 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
236 if (alloc_targets(t, num_targets)) {
248 int dm_create_error_table(struct dm_table **result, struct mapped_device *md)
251 sector_t dev_size = 1;
255 * Find current size of device.
256 * Default to 1 sector if inactive.
258 t = dm_get_table(md);
260 dev_size = dm_table_get_size(t);
264 r = dm_table_create(&t, FMODE_READ, 1, md);
268 r = dm_table_add_target(t, "error", 0, dev_size, NULL);
272 r = dm_table_complete(t);
284 EXPORT_SYMBOL_GPL(dm_create_error_table);
286 static void free_devices(struct list_head *devices)
288 struct list_head *tmp, *next;
290 for (tmp = devices->next; tmp != devices; tmp = next) {
291 struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
297 static void table_destroy(struct dm_table *t)
301 /* free the indexes (see dm_table_complete) */
303 vfree(t->index[t->depth - 2]);
305 /* free the targets */
306 for (i = 0; i < t->num_targets; i++) {
307 struct dm_target *tgt = t->targets + i;
312 dm_put_target_type(tgt->type);
317 /* free the device list */
318 if (t->devices.next != &t->devices) {
319 DMWARN("devices still present during destroy: "
320 "dm_table_remove_device calls missing");
322 free_devices(&t->devices);
328 void dm_table_get(struct dm_table *t)
330 atomic_inc(&t->holders);
333 void dm_table_put(struct dm_table *t)
338 if (atomic_dec_and_test(&t->holders))
343 * Checks to see if we need to extend highs or targets.
345 static inline int check_space(struct dm_table *t)
347 if (t->num_targets >= t->num_allocated)
348 return alloc_targets(t, t->num_allocated * 2);
354 * Convert a device path to a dev_t.
356 static int lookup_device(const char *path, dev_t *dev)
362 if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd)))
365 inode = nd.dentry->d_inode;
371 if (!S_ISBLK(inode->i_mode)) {
376 *dev = inode->i_rdev;
384 * See if we've already got a device in the list.
386 static struct dm_dev *find_device(struct list_head *l, dev_t dev)
390 list_for_each_entry (dd, l, list)
391 if (dd->bdev->bd_dev == dev)
398 * Open a device so we can use it as a map destination.
400 static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md)
402 static char *_claim_ptr = "I belong to device-mapper";
403 struct block_device *bdev;
409 bdev = open_by_devnum(dev, d->mode);
411 return PTR_ERR(bdev);
412 r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
421 * Close a device that we've been using.
423 static void close_dev(struct dm_dev *d, struct mapped_device *md)
428 bd_release_from_disk(d->bdev, dm_disk(md));
434 * If possible, this checks an area of a destination device is valid.
436 static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
438 sector_t dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
443 return ((start < dev_size) && (len <= (dev_size - start)));
447 * This upgrades the mode on an already open dm_dev. Being
448 * careful to leave things as they were if we fail to reopen the
451 static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md)
454 struct dm_dev dd_copy;
455 dev_t dev = dd->bdev->bd_dev;
459 dd->mode |= new_mode;
461 r = open_dev(dd, dev, md);
463 close_dev(&dd_copy, md);
471 * Add a device to the list, or just increment the usage count if
472 * it's already present.
474 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
475 const char *path, sector_t start, sector_t len,
476 int mode, struct dm_dev **result)
481 unsigned int major, minor;
485 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
486 /* Extract the major/minor numbers */
487 dev = MKDEV(major, minor);
488 if (MAJOR(dev) != major || MINOR(dev) != minor)
491 /* convert the path to a device */
492 if ((r = lookup_device(path, &dev)))
496 dd = find_device(&t->devices, dev);
498 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
505 if ((r = open_dev(dd, dev, t->md))) {
510 format_dev_t(dd->name, dev);
512 atomic_set(&dd->count, 0);
513 list_add(&dd->list, &t->devices);
515 } else if (dd->mode != (mode | dd->mode)) {
516 r = upgrade_mode(dd, mode, t->md);
520 atomic_inc(&dd->count);
522 if (!check_device_area(dd, start, len)) {
523 DMWARN("device %s too small for target", path);
524 dm_put_device(ti, dd);
533 void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
535 struct request_queue *q = bdev_get_queue(bdev);
536 struct io_restrictions *rs = &ti->limits;
539 * Combine the device limits low.
541 * FIXME: if we move an io_restriction struct
542 * into q this would just be a call to
543 * combine_restrictions_low()
546 min_not_zero(rs->max_sectors, q->max_sectors);
548 /* FIXME: Device-Mapper on top of RAID-0 breaks because DM
549 * currently doesn't honor MD's merge_bvec_fn routine.
550 * In this case, we'll force DM to use PAGE_SIZE or
551 * smaller I/O, just to be safe. A better fix is in the
552 * works, but add this for the time being so it will at
553 * least operate correctly.
555 if (q->merge_bvec_fn)
557 min_not_zero(rs->max_sectors,
558 (unsigned int) (PAGE_SIZE >> 9));
560 rs->max_phys_segments =
561 min_not_zero(rs->max_phys_segments,
562 q->max_phys_segments);
564 rs->max_hw_segments =
565 min_not_zero(rs->max_hw_segments, q->max_hw_segments);
567 rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
569 rs->max_segment_size =
570 min_not_zero(rs->max_segment_size, q->max_segment_size);
573 min_not_zero(rs->max_hw_sectors, q->max_hw_sectors);
575 rs->seg_boundary_mask =
576 min_not_zero(rs->seg_boundary_mask,
577 q->seg_boundary_mask);
579 rs->bounce_pfn = min_not_zero(rs->bounce_pfn, q->bounce_pfn);
581 rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
583 EXPORT_SYMBOL_GPL(dm_set_device_limits);
585 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
586 sector_t len, int mode, struct dm_dev **result)
588 int r = __table_get_device(ti->table, ti, path,
589 start, len, mode, result);
592 dm_set_device_limits(ti, (*result)->bdev);
598 * Decrement a devices use count and remove it if necessary.
600 void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
602 if (atomic_dec_and_test(&dd->count)) {
603 close_dev(dd, ti->table->md);
610 * Checks to see if the target joins onto the end of the table.
612 static int adjoin(struct dm_table *table, struct dm_target *ti)
614 struct dm_target *prev;
616 if (!table->num_targets)
619 prev = &table->targets[table->num_targets - 1];
620 return (ti->begin == (prev->begin + prev->len));
624 * Used to dynamically allocate the arg array.
626 static char **realloc_argv(unsigned *array_size, char **old_argv)
631 new_size = *array_size ? *array_size * 2 : 64;
632 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
634 memcpy(argv, old_argv, *array_size * sizeof(*argv));
635 *array_size = new_size;
643 * Destructively splits up the argument list to pass to ctr.
645 int dm_split_args(int *argc, char ***argvp, char *input)
647 char *start, *end = input, *out, **argv = NULL;
648 unsigned array_size = 0;
657 argv = realloc_argv(&array_size, argv);
664 /* Skip whitespace */
665 while (*start && isspace(*start))
669 break; /* success, we hit the end */
671 /* 'out' is used to remove any back-quotes */
674 /* Everything apart from '\0' can be quoted */
675 if (*end == '\\' && *(end + 1)) {
682 break; /* end of token */
687 /* have we already filled the array ? */
688 if ((*argc + 1) > array_size) {
689 argv = realloc_argv(&array_size, argv);
694 /* we know this is whitespace */
698 /* terminate the string and put it in the array */
708 static void check_for_valid_limits(struct io_restrictions *rs)
710 if (!rs->max_sectors)
711 rs->max_sectors = SAFE_MAX_SECTORS;
712 if (!rs->max_hw_sectors)
713 rs->max_hw_sectors = SAFE_MAX_SECTORS;
714 if (!rs->max_phys_segments)
715 rs->max_phys_segments = MAX_PHYS_SEGMENTS;
716 if (!rs->max_hw_segments)
717 rs->max_hw_segments = MAX_HW_SEGMENTS;
718 if (!rs->hardsect_size)
719 rs->hardsect_size = 1 << SECTOR_SHIFT;
720 if (!rs->max_segment_size)
721 rs->max_segment_size = MAX_SEGMENT_SIZE;
722 if (!rs->seg_boundary_mask)
723 rs->seg_boundary_mask = -1;
728 int dm_table_add_target(struct dm_table *t, const char *type,
729 sector_t start, sector_t len, char *params)
731 int r = -EINVAL, argc;
733 struct dm_target *tgt;
735 if ((r = check_space(t)))
738 tgt = t->targets + t->num_targets;
739 memset(tgt, 0, sizeof(*tgt));
742 DMERR("%s: zero-length target", dm_device_name(t->md));
746 tgt->type = dm_get_target_type(type);
748 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
756 tgt->error = "Unknown error";
759 * Does this target adjoin the previous one ?
761 if (!adjoin(t, tgt)) {
762 tgt->error = "Gap in table";
767 r = dm_split_args(&argc, &argv, params);
769 tgt->error = "couldn't split parameters (insufficient memory)";
773 r = tgt->type->ctr(tgt, argc, argv);
778 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
780 /* FIXME: the plan is to combine high here and then have
781 * the merge fn apply the target level restrictions. */
782 combine_restrictions_low(&t->limits, &tgt->limits);
786 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
787 dm_put_target_type(tgt->type);
791 static int setup_indexes(struct dm_table *t)
794 unsigned int total = 0;
797 /* allocate the space for *all* the indexes */
798 for (i = t->depth - 2; i >= 0; i--) {
799 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
800 total += t->counts[i];
803 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
807 /* set up internal nodes, bottom-up */
808 for (i = t->depth - 2, total = 0; i >= 0; i--) {
809 t->index[i] = indexes;
810 indexes += (KEYS_PER_NODE * t->counts[i]);
811 setup_btree_index(i, t);
818 * Builds the btree to index the map.
820 int dm_table_complete(struct dm_table *t)
823 unsigned int leaf_nodes;
825 check_for_valid_limits(&t->limits);
827 /* how many indexes will the btree have ? */
828 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
829 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
831 /* leaf layer has already been set up */
832 t->counts[t->depth - 1] = leaf_nodes;
833 t->index[t->depth - 1] = t->highs;
836 r = setup_indexes(t);
841 static DEFINE_MUTEX(_event_lock);
842 void dm_table_event_callback(struct dm_table *t,
843 void (*fn)(void *), void *context)
845 mutex_lock(&_event_lock);
847 t->event_context = context;
848 mutex_unlock(&_event_lock);
851 void dm_table_event(struct dm_table *t)
854 * You can no longer call dm_table_event() from interrupt
855 * context, use a bottom half instead.
857 BUG_ON(in_interrupt());
859 mutex_lock(&_event_lock);
861 t->event_fn(t->event_context);
862 mutex_unlock(&_event_lock);
865 sector_t dm_table_get_size(struct dm_table *t)
867 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
870 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
872 if (index >= t->num_targets)
875 return t->targets + index;
879 * Search the btree for the correct target.
881 * Caller should check returned pointer with dm_target_is_valid()
882 * to trap I/O beyond end of device.
884 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
886 unsigned int l, n = 0, k = 0;
889 for (l = 0; l < t->depth; l++) {
891 node = get_node(t, l, n);
893 for (k = 0; k < KEYS_PER_NODE; k++)
894 if (node[k] >= sector)
898 return &t->targets[(KEYS_PER_NODE * n) + k];
901 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
904 * Make sure we obey the optimistic sub devices
907 blk_queue_max_sectors(q, t->limits.max_sectors);
908 q->max_phys_segments = t->limits.max_phys_segments;
909 q->max_hw_segments = t->limits.max_hw_segments;
910 q->hardsect_size = t->limits.hardsect_size;
911 q->max_segment_size = t->limits.max_segment_size;
912 q->max_hw_sectors = t->limits.max_hw_sectors;
913 q->seg_boundary_mask = t->limits.seg_boundary_mask;
914 q->bounce_pfn = t->limits.bounce_pfn;
915 if (t->limits.no_cluster)
916 q->queue_flags &= ~(1 << QUEUE_FLAG_CLUSTER);
918 q->queue_flags |= (1 << QUEUE_FLAG_CLUSTER);
922 unsigned int dm_table_get_num_targets(struct dm_table *t)
924 return t->num_targets;
927 struct list_head *dm_table_get_devices(struct dm_table *t)
932 int dm_table_get_mode(struct dm_table *t)
937 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
939 int i = t->num_targets;
940 struct dm_target *ti = t->targets;
944 if (ti->type->postsuspend)
945 ti->type->postsuspend(ti);
946 } else if (ti->type->presuspend)
947 ti->type->presuspend(ti);
953 void dm_table_presuspend_targets(struct dm_table *t)
958 return suspend_targets(t, 0);
961 void dm_table_postsuspend_targets(struct dm_table *t)
966 return suspend_targets(t, 1);
969 int dm_table_resume_targets(struct dm_table *t)
973 for (i = 0; i < t->num_targets; i++) {
974 struct dm_target *ti = t->targets + i;
976 if (!ti->type->preresume)
979 r = ti->type->preresume(ti);
984 for (i = 0; i < t->num_targets; i++) {
985 struct dm_target *ti = t->targets + i;
987 if (ti->type->resume)
988 ti->type->resume(ti);
994 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
996 struct list_head *d, *devices;
999 devices = dm_table_get_devices(t);
1000 for (d = devices->next; d != devices; d = d->next) {
1001 struct dm_dev *dd = list_entry(d, struct dm_dev, list);
1002 struct request_queue *q = bdev_get_queue(dd->bdev);
1003 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1009 void dm_table_unplug_all(struct dm_table *t)
1011 struct list_head *d, *devices = dm_table_get_devices(t);
1013 for (d = devices->next; d != devices; d = d->next) {
1014 struct dm_dev *dd = list_entry(d, struct dm_dev, list);
1015 struct request_queue *q = bdev_get_queue(dd->bdev);
1021 struct mapped_device *dm_table_get_md(struct dm_table *t)
1028 EXPORT_SYMBOL(dm_vcalloc);
1029 EXPORT_SYMBOL(dm_get_device);
1030 EXPORT_SYMBOL(dm_put_device);
1031 EXPORT_SYMBOL(dm_table_event);
1032 EXPORT_SYMBOL(dm_table_get_size);
1033 EXPORT_SYMBOL(dm_table_get_mode);
1034 EXPORT_SYMBOL(dm_table_get_md);
1035 EXPORT_SYMBOL(dm_table_put);
1036 EXPORT_SYMBOL(dm_table_get);
1037 EXPORT_SYMBOL(dm_table_unplug_all);