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 static void free_devices(struct list_head *devices)
250 struct list_head *tmp, *next;
252 list_for_each_safe(tmp, next, devices) {
253 struct dm_dev_internal *dd =
254 list_entry(tmp, struct dm_dev_internal, list);
259 static void table_destroy(struct dm_table *t)
263 /* free the indexes (see dm_table_complete) */
265 vfree(t->index[t->depth - 2]);
267 /* free the targets */
268 for (i = 0; i < t->num_targets; i++) {
269 struct dm_target *tgt = t->targets + i;
274 dm_put_target_type(tgt->type);
279 /* free the device list */
280 if (t->devices.next != &t->devices) {
281 DMWARN("devices still present during destroy: "
282 "dm_table_remove_device calls missing");
284 free_devices(&t->devices);
290 void dm_table_get(struct dm_table *t)
292 atomic_inc(&t->holders);
295 void dm_table_put(struct dm_table *t)
300 if (atomic_dec_and_test(&t->holders))
305 * Checks to see if we need to extend highs or targets.
307 static inline int check_space(struct dm_table *t)
309 if (t->num_targets >= t->num_allocated)
310 return alloc_targets(t, t->num_allocated * 2);
316 * Convert a device path to a dev_t.
318 static int lookup_device(const char *path, dev_t *dev)
320 struct block_device *bdev = lookup_bdev(path);
322 return PTR_ERR(bdev);
329 * See if we've already got a device in the list.
331 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
333 struct dm_dev_internal *dd;
335 list_for_each_entry (dd, l, list)
336 if (dd->dm_dev.bdev->bd_dev == dev)
343 * Open a device so we can use it as a map destination.
345 static int open_dev(struct dm_dev_internal *d, dev_t dev,
346 struct mapped_device *md)
348 static char *_claim_ptr = "I belong to device-mapper";
349 struct block_device *bdev;
353 BUG_ON(d->dm_dev.bdev);
355 bdev = open_by_devnum(dev, d->dm_dev.mode);
357 return PTR_ERR(bdev);
358 r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
362 d->dm_dev.bdev = bdev;
367 * Close a device that we've been using.
369 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
374 bd_release_from_disk(d->dm_dev.bdev, dm_disk(md));
375 blkdev_put(d->dm_dev.bdev);
376 d->dm_dev.bdev = NULL;
380 * If possible, this checks an area of a destination device is valid.
382 static int check_device_area(struct dm_dev_internal *dd, sector_t start,
385 sector_t dev_size = dd->dm_dev.bdev->bd_inode->i_size >> SECTOR_SHIFT;
390 return ((start < dev_size) && (len <= (dev_size - start)));
394 * This upgrades the mode on an already open dm_dev. Being
395 * careful to leave things as they were if we fail to reopen the
398 static int upgrade_mode(struct dm_dev_internal *dd, int new_mode,
399 struct mapped_device *md)
402 struct dm_dev_internal dd_copy;
403 dev_t dev = dd->dm_dev.bdev->bd_dev;
407 dd->dm_dev.mode |= new_mode;
408 dd->dm_dev.bdev = NULL;
409 r = open_dev(dd, dev, md);
411 close_dev(&dd_copy, md);
419 * Add a device to the list, or just increment the usage count if
420 * it's already present.
422 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
423 const char *path, sector_t start, sector_t len,
424 int mode, struct dm_dev **result)
427 dev_t uninitialized_var(dev);
428 struct dm_dev_internal *dd;
429 unsigned int major, minor;
433 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
434 /* Extract the major/minor numbers */
435 dev = MKDEV(major, minor);
436 if (MAJOR(dev) != major || MINOR(dev) != minor)
439 /* convert the path to a device */
440 if ((r = lookup_device(path, &dev)))
444 dd = find_device(&t->devices, dev);
446 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
450 dd->dm_dev.mode = mode;
451 dd->dm_dev.bdev = NULL;
453 if ((r = open_dev(dd, dev, t->md))) {
458 format_dev_t(dd->dm_dev.name, dev);
460 atomic_set(&dd->count, 0);
461 list_add(&dd->list, &t->devices);
463 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
464 r = upgrade_mode(dd, mode, t->md);
468 atomic_inc(&dd->count);
470 if (!check_device_area(dd, start, len)) {
471 DMWARN("device %s too small for target", path);
472 dm_put_device(ti, &dd->dm_dev);
476 *result = &dd->dm_dev;
481 void dm_set_device_limits(struct dm_target *ti, struct block_device *bdev)
483 struct request_queue *q = bdev_get_queue(bdev);
484 struct io_restrictions *rs = &ti->limits;
485 char b[BDEVNAME_SIZE];
488 DMWARN("%s: Cannot set limits for nonexistent device %s",
489 dm_device_name(ti->table->md), bdevname(bdev, b));
494 * Combine the device limits low.
496 * FIXME: if we move an io_restriction struct
497 * into q this would just be a call to
498 * combine_restrictions_low()
501 min_not_zero(rs->max_sectors, q->max_sectors);
504 * Check if merge fn is supported.
505 * If not we'll force DM to use PAGE_SIZE or
506 * smaller I/O, just to be safe.
509 if (q->merge_bvec_fn && !ti->type->merge)
511 min_not_zero(rs->max_sectors,
512 (unsigned int) (PAGE_SIZE >> 9));
514 rs->max_phys_segments =
515 min_not_zero(rs->max_phys_segments,
516 q->max_phys_segments);
518 rs->max_hw_segments =
519 min_not_zero(rs->max_hw_segments, q->max_hw_segments);
521 rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
523 rs->max_segment_size =
524 min_not_zero(rs->max_segment_size, q->max_segment_size);
527 min_not_zero(rs->max_hw_sectors, q->max_hw_sectors);
529 rs->seg_boundary_mask =
530 min_not_zero(rs->seg_boundary_mask,
531 q->seg_boundary_mask);
533 rs->bounce_pfn = min_not_zero(rs->bounce_pfn, q->bounce_pfn);
535 rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
537 EXPORT_SYMBOL_GPL(dm_set_device_limits);
539 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
540 sector_t len, int mode, struct dm_dev **result)
542 int r = __table_get_device(ti->table, ti, path,
543 start, len, mode, result);
546 dm_set_device_limits(ti, (*result)->bdev);
552 * Decrement a devices use count and remove it if necessary.
554 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
556 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
559 if (atomic_dec_and_test(&dd->count)) {
560 close_dev(dd, ti->table->md);
567 * Checks to see if the target joins onto the end of the table.
569 static int adjoin(struct dm_table *table, struct dm_target *ti)
571 struct dm_target *prev;
573 if (!table->num_targets)
576 prev = &table->targets[table->num_targets - 1];
577 return (ti->begin == (prev->begin + prev->len));
581 * Used to dynamically allocate the arg array.
583 static char **realloc_argv(unsigned *array_size, char **old_argv)
588 new_size = *array_size ? *array_size * 2 : 64;
589 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
591 memcpy(argv, old_argv, *array_size * sizeof(*argv));
592 *array_size = new_size;
600 * Destructively splits up the argument list to pass to ctr.
602 int dm_split_args(int *argc, char ***argvp, char *input)
604 char *start, *end = input, *out, **argv = NULL;
605 unsigned array_size = 0;
614 argv = realloc_argv(&array_size, argv);
621 /* Skip whitespace */
622 while (*start && isspace(*start))
626 break; /* success, we hit the end */
628 /* 'out' is used to remove any back-quotes */
631 /* Everything apart from '\0' can be quoted */
632 if (*end == '\\' && *(end + 1)) {
639 break; /* end of token */
644 /* have we already filled the array ? */
645 if ((*argc + 1) > array_size) {
646 argv = realloc_argv(&array_size, argv);
651 /* we know this is whitespace */
655 /* terminate the string and put it in the array */
665 static void check_for_valid_limits(struct io_restrictions *rs)
667 if (!rs->max_sectors)
668 rs->max_sectors = SAFE_MAX_SECTORS;
669 if (!rs->max_hw_sectors)
670 rs->max_hw_sectors = SAFE_MAX_SECTORS;
671 if (!rs->max_phys_segments)
672 rs->max_phys_segments = MAX_PHYS_SEGMENTS;
673 if (!rs->max_hw_segments)
674 rs->max_hw_segments = MAX_HW_SEGMENTS;
675 if (!rs->hardsect_size)
676 rs->hardsect_size = 1 << SECTOR_SHIFT;
677 if (!rs->max_segment_size)
678 rs->max_segment_size = MAX_SEGMENT_SIZE;
679 if (!rs->seg_boundary_mask)
680 rs->seg_boundary_mask = -1;
685 int dm_table_add_target(struct dm_table *t, const char *type,
686 sector_t start, sector_t len, char *params)
688 int r = -EINVAL, argc;
690 struct dm_target *tgt;
692 if ((r = check_space(t)))
695 tgt = t->targets + t->num_targets;
696 memset(tgt, 0, sizeof(*tgt));
699 DMERR("%s: zero-length target", dm_device_name(t->md));
703 tgt->type = dm_get_target_type(type);
705 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
713 tgt->error = "Unknown error";
716 * Does this target adjoin the previous one ?
718 if (!adjoin(t, tgt)) {
719 tgt->error = "Gap in table";
724 r = dm_split_args(&argc, &argv, params);
726 tgt->error = "couldn't split parameters (insufficient memory)";
730 r = tgt->type->ctr(tgt, argc, argv);
735 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
737 /* FIXME: the plan is to combine high here and then have
738 * the merge fn apply the target level restrictions. */
739 combine_restrictions_low(&t->limits, &tgt->limits);
743 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
744 dm_put_target_type(tgt->type);
748 static int setup_indexes(struct dm_table *t)
751 unsigned int total = 0;
754 /* allocate the space for *all* the indexes */
755 for (i = t->depth - 2; i >= 0; i--) {
756 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
757 total += t->counts[i];
760 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
764 /* set up internal nodes, bottom-up */
765 for (i = t->depth - 2; i >= 0; i--) {
766 t->index[i] = indexes;
767 indexes += (KEYS_PER_NODE * t->counts[i]);
768 setup_btree_index(i, t);
775 * Builds the btree to index the map.
777 int dm_table_complete(struct dm_table *t)
780 unsigned int leaf_nodes;
782 check_for_valid_limits(&t->limits);
784 /* how many indexes will the btree have ? */
785 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
786 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
788 /* leaf layer has already been set up */
789 t->counts[t->depth - 1] = leaf_nodes;
790 t->index[t->depth - 1] = t->highs;
793 r = setup_indexes(t);
798 static DEFINE_MUTEX(_event_lock);
799 void dm_table_event_callback(struct dm_table *t,
800 void (*fn)(void *), void *context)
802 mutex_lock(&_event_lock);
804 t->event_context = context;
805 mutex_unlock(&_event_lock);
808 void dm_table_event(struct dm_table *t)
811 * You can no longer call dm_table_event() from interrupt
812 * context, use a bottom half instead.
814 BUG_ON(in_interrupt());
816 mutex_lock(&_event_lock);
818 t->event_fn(t->event_context);
819 mutex_unlock(&_event_lock);
822 sector_t dm_table_get_size(struct dm_table *t)
824 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
827 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
829 if (index >= t->num_targets)
832 return t->targets + index;
836 * Search the btree for the correct target.
838 * Caller should check returned pointer with dm_target_is_valid()
839 * to trap I/O beyond end of device.
841 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
843 unsigned int l, n = 0, k = 0;
846 for (l = 0; l < t->depth; l++) {
848 node = get_node(t, l, n);
850 for (k = 0; k < KEYS_PER_NODE; k++)
851 if (node[k] >= sector)
855 return &t->targets[(KEYS_PER_NODE * n) + k];
858 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
861 * Make sure we obey the optimistic sub devices
864 blk_queue_max_sectors(q, t->limits.max_sectors);
865 q->max_phys_segments = t->limits.max_phys_segments;
866 q->max_hw_segments = t->limits.max_hw_segments;
867 q->hardsect_size = t->limits.hardsect_size;
868 q->max_segment_size = t->limits.max_segment_size;
869 q->max_hw_sectors = t->limits.max_hw_sectors;
870 q->seg_boundary_mask = t->limits.seg_boundary_mask;
871 q->bounce_pfn = t->limits.bounce_pfn;
873 if (t->limits.no_cluster)
874 queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q);
876 queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q);
880 unsigned int dm_table_get_num_targets(struct dm_table *t)
882 return t->num_targets;
885 struct list_head *dm_table_get_devices(struct dm_table *t)
890 int dm_table_get_mode(struct dm_table *t)
895 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
897 int i = t->num_targets;
898 struct dm_target *ti = t->targets;
902 if (ti->type->postsuspend)
903 ti->type->postsuspend(ti);
904 } else if (ti->type->presuspend)
905 ti->type->presuspend(ti);
911 void dm_table_presuspend_targets(struct dm_table *t)
916 suspend_targets(t, 0);
919 void dm_table_postsuspend_targets(struct dm_table *t)
924 suspend_targets(t, 1);
927 int dm_table_resume_targets(struct dm_table *t)
931 for (i = 0; i < t->num_targets; i++) {
932 struct dm_target *ti = t->targets + i;
934 if (!ti->type->preresume)
937 r = ti->type->preresume(ti);
942 for (i = 0; i < t->num_targets; i++) {
943 struct dm_target *ti = t->targets + i;
945 if (ti->type->resume)
946 ti->type->resume(ti);
952 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
954 struct dm_dev_internal *dd;
955 struct list_head *devices = dm_table_get_devices(t);
958 list_for_each_entry(dd, devices, list) {
959 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
960 char b[BDEVNAME_SIZE];
963 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
965 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
966 dm_device_name(t->md),
967 bdevname(dd->dm_dev.bdev, b));
973 void dm_table_unplug_all(struct dm_table *t)
975 struct dm_dev_internal *dd;
976 struct list_head *devices = dm_table_get_devices(t);
978 list_for_each_entry(dd, devices, list) {
979 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
980 char b[BDEVNAME_SIZE];
985 DMWARN_LIMIT("%s: Cannot unplug nonexistent device %s",
986 dm_device_name(t->md),
987 bdevname(dd->dm_dev.bdev, b));
991 struct mapped_device *dm_table_get_md(struct dm_table *t)
998 EXPORT_SYMBOL(dm_vcalloc);
999 EXPORT_SYMBOL(dm_get_device);
1000 EXPORT_SYMBOL(dm_put_device);
1001 EXPORT_SYMBOL(dm_table_event);
1002 EXPORT_SYMBOL(dm_table_get_size);
1003 EXPORT_SYMBOL(dm_table_get_mode);
1004 EXPORT_SYMBOL(dm_table_get_md);
1005 EXPORT_SYMBOL(dm_table_put);
1006 EXPORT_SYMBOL(dm_table_get);
1007 EXPORT_SYMBOL(dm_table_unplug_all);