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 <asm/atomic.h>
20 #define NODE_SIZE L1_CACHE_BYTES
21 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
22 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
25 struct mapped_device *md;
30 unsigned int counts[MAX_DEPTH]; /* in nodes */
31 sector_t *index[MAX_DEPTH];
33 unsigned int num_targets;
34 unsigned int num_allocated;
36 struct dm_target *targets;
39 * Indicates the rw permissions for the new logical
40 * device. This should be a combination of FMODE_READ
45 /* a list of devices used by this table */
46 struct list_head devices;
49 * These are optimistic limits taken from all the
50 * targets, some targets will need smaller limits.
52 struct io_restrictions limits;
54 /* events get handed up using this callback */
55 void (*event_fn)(void *);
60 * Similar to ceiling(log_size(n))
62 static unsigned int int_log(unsigned int n, unsigned int base)
67 n = dm_div_up(n, base);
75 * Returns the minimum that is _not_ zero, unless both are zero.
77 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
80 * Combine two io_restrictions, always taking the lower value.
82 static void combine_restrictions_low(struct io_restrictions *lhs,
83 struct io_restrictions *rhs)
86 min_not_zero(lhs->max_sectors, rhs->max_sectors);
88 lhs->max_phys_segments =
89 min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments);
91 lhs->max_hw_segments =
92 min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments);
94 lhs->hardsect_size = max(lhs->hardsect_size, rhs->hardsect_size);
96 lhs->max_segment_size =
97 min_not_zero(lhs->max_segment_size, rhs->max_segment_size);
99 lhs->seg_boundary_mask =
100 min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask);
102 lhs->no_cluster |= rhs->no_cluster;
106 * Calculate the index of the child node of the n'th node k'th key.
108 static inline unsigned int get_child(unsigned int n, unsigned int k)
110 return (n * CHILDREN_PER_NODE) + k;
114 * Return the n'th node of level l from table t.
116 static inline sector_t *get_node(struct dm_table *t,
117 unsigned int l, unsigned int n)
119 return t->index[l] + (n * KEYS_PER_NODE);
123 * Return the highest key that you could lookup from the n'th
124 * node on level l of the btree.
126 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
128 for (; l < t->depth - 1; l++)
129 n = get_child(n, CHILDREN_PER_NODE - 1);
131 if (n >= t->counts[l])
132 return (sector_t) - 1;
134 return get_node(t, l, n)[KEYS_PER_NODE - 1];
138 * Fills in a level of the btree based on the highs of the level
141 static int setup_btree_index(unsigned int l, struct dm_table *t)
146 for (n = 0U; n < t->counts[l]; n++) {
147 node = get_node(t, l, n);
149 for (k = 0U; k < KEYS_PER_NODE; k++)
150 node[k] = high(t, l + 1, get_child(n, k));
156 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
162 * Check that we're not going to overflow.
164 if (nmemb > (ULONG_MAX / elem_size))
167 size = nmemb * elem_size;
168 addr = vmalloc(size);
170 memset(addr, 0, size);
176 * highs, and targets are managed as dynamic arrays during a
179 static int alloc_targets(struct dm_table *t, unsigned int num)
182 struct dm_target *n_targets;
183 int n = t->num_targets;
186 * Allocate both the target array and offset array at once.
188 n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
193 n_targets = (struct dm_target *) (n_highs + num);
196 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
197 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
200 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
203 t->num_allocated = num;
205 t->targets = n_targets;
210 int dm_table_create(struct dm_table **result, int mode,
211 unsigned num_targets, struct mapped_device *md)
213 struct dm_table *t = kmalloc(sizeof(*t), GFP_KERNEL);
218 memset(t, 0, sizeof(*t));
219 INIT_LIST_HEAD(&t->devices);
220 atomic_set(&t->holders, 1);
223 num_targets = KEYS_PER_NODE;
225 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
227 if (alloc_targets(t, num_targets)) {
239 static void free_devices(struct list_head *devices)
241 struct list_head *tmp, *next;
243 for (tmp = devices->next; tmp != devices; tmp = next) {
244 struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
250 static void table_destroy(struct dm_table *t)
254 /* free the indexes (see dm_table_complete) */
256 vfree(t->index[t->depth - 2]);
258 /* free the targets */
259 for (i = 0; i < t->num_targets; i++) {
260 struct dm_target *tgt = t->targets + i;
265 dm_put_target_type(tgt->type);
270 /* free the device list */
271 if (t->devices.next != &t->devices) {
272 DMWARN("devices still present during destroy: "
273 "dm_table_remove_device calls missing");
275 free_devices(&t->devices);
281 void dm_table_get(struct dm_table *t)
283 atomic_inc(&t->holders);
286 void dm_table_put(struct dm_table *t)
291 if (atomic_dec_and_test(&t->holders))
296 * Checks to see if we need to extend highs or targets.
298 static inline int check_space(struct dm_table *t)
300 if (t->num_targets >= t->num_allocated)
301 return alloc_targets(t, t->num_allocated * 2);
307 * Convert a device path to a dev_t.
309 static int lookup_device(const char *path, dev_t *dev)
315 if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd)))
318 inode = nd.dentry->d_inode;
324 if (!S_ISBLK(inode->i_mode)) {
329 *dev = inode->i_rdev;
337 * See if we've already got a device in the list.
339 static struct dm_dev *find_device(struct list_head *l, dev_t dev)
343 list_for_each_entry (dd, l, list)
344 if (dd->bdev->bd_dev == dev)
351 * Open a device so we can use it as a map destination.
353 static int open_dev(struct dm_dev *d, dev_t dev, struct mapped_device *md)
355 static char *_claim_ptr = "I belong to device-mapper";
356 struct block_device *bdev;
362 bdev = open_by_devnum(dev, d->mode);
364 return PTR_ERR(bdev);
365 r = bd_claim_by_disk(bdev, _claim_ptr, dm_disk(md));
374 * Close a device that we've been using.
376 static void close_dev(struct dm_dev *d, struct mapped_device *md)
381 bd_release_from_disk(d->bdev, dm_disk(md));
387 * If possible (ie. blk_size[major] is set), this checks an area
388 * of a destination device is valid.
390 static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
393 dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
394 return ((start < dev_size) && (len <= (dev_size - start)));
398 * This upgrades the mode on an already open dm_dev. Being
399 * careful to leave things as they were if we fail to reopen the
402 static int upgrade_mode(struct dm_dev *dd, int new_mode, struct mapped_device *md)
405 struct dm_dev dd_copy;
406 dev_t dev = dd->bdev->bd_dev;
410 dd->mode |= new_mode;
412 r = open_dev(dd, dev, md);
414 close_dev(&dd_copy, md);
422 * Add a device to the list, or just increment the usage count if
423 * it's already present.
425 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
426 const char *path, sector_t start, sector_t len,
427 int mode, struct dm_dev **result)
432 unsigned int major, minor;
436 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
437 /* Extract the major/minor numbers */
438 dev = MKDEV(major, minor);
439 if (MAJOR(dev) != major || MINOR(dev) != minor)
442 /* convert the path to a device */
443 if ((r = lookup_device(path, &dev)))
447 dd = find_device(&t->devices, dev);
449 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
456 if ((r = open_dev(dd, dev, t->md))) {
461 format_dev_t(dd->name, dev);
463 atomic_set(&dd->count, 0);
464 list_add(&dd->list, &t->devices);
466 } else if (dd->mode != (mode | dd->mode)) {
467 r = upgrade_mode(dd, mode, t->md);
471 atomic_inc(&dd->count);
473 if (!check_device_area(dd, start, len)) {
474 DMWARN("device %s too small for target", path);
475 dm_put_device(ti, dd);
485 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
486 sector_t len, int mode, struct dm_dev **result)
488 int r = __table_get_device(ti->table, ti, path,
489 start, len, mode, result);
491 request_queue_t *q = bdev_get_queue((*result)->bdev);
492 struct io_restrictions *rs = &ti->limits;
495 * Combine the device limits low.
497 * FIXME: if we move an io_restriction struct
498 * into q this would just be a call to
499 * combine_restrictions_low()
502 min_not_zero(rs->max_sectors, q->max_sectors);
504 /* FIXME: Device-Mapper on top of RAID-0 breaks because DM
505 * currently doesn't honor MD's merge_bvec_fn routine.
506 * In this case, we'll force DM to use PAGE_SIZE or
507 * smaller I/O, just to be safe. A better fix is in the
508 * works, but add this for the time being so it will at
509 * least operate correctly.
511 if (q->merge_bvec_fn)
513 min_not_zero(rs->max_sectors,
514 (unsigned int) (PAGE_SIZE >> 9));
516 rs->max_phys_segments =
517 min_not_zero(rs->max_phys_segments,
518 q->max_phys_segments);
520 rs->max_hw_segments =
521 min_not_zero(rs->max_hw_segments, q->max_hw_segments);
523 rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
525 rs->max_segment_size =
526 min_not_zero(rs->max_segment_size, q->max_segment_size);
528 rs->seg_boundary_mask =
529 min_not_zero(rs->seg_boundary_mask,
530 q->seg_boundary_mask);
532 rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
539 * Decrement a devices use count and remove it if necessary.
541 void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
543 if (atomic_dec_and_test(&dd->count)) {
544 close_dev(dd, ti->table->md);
551 * Checks to see if the target joins onto the end of the table.
553 static int adjoin(struct dm_table *table, struct dm_target *ti)
555 struct dm_target *prev;
557 if (!table->num_targets)
560 prev = &table->targets[table->num_targets - 1];
561 return (ti->begin == (prev->begin + prev->len));
565 * Used to dynamically allocate the arg array.
567 static char **realloc_argv(unsigned *array_size, char **old_argv)
572 new_size = *array_size ? *array_size * 2 : 64;
573 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
575 memcpy(argv, old_argv, *array_size * sizeof(*argv));
576 *array_size = new_size;
584 * Destructively splits up the argument list to pass to ctr.
586 int dm_split_args(int *argc, char ***argvp, char *input)
588 char *start, *end = input, *out, **argv = NULL;
589 unsigned array_size = 0;
592 argv = realloc_argv(&array_size, argv);
599 /* Skip whitespace */
600 while (*start && isspace(*start))
604 break; /* success, we hit the end */
606 /* 'out' is used to remove any back-quotes */
609 /* Everything apart from '\0' can be quoted */
610 if (*end == '\\' && *(end + 1)) {
617 break; /* end of token */
622 /* have we already filled the array ? */
623 if ((*argc + 1) > array_size) {
624 argv = realloc_argv(&array_size, argv);
629 /* we know this is whitespace */
633 /* terminate the string and put it in the array */
643 static void check_for_valid_limits(struct io_restrictions *rs)
645 if (!rs->max_sectors)
646 rs->max_sectors = SAFE_MAX_SECTORS;
647 if (!rs->max_phys_segments)
648 rs->max_phys_segments = MAX_PHYS_SEGMENTS;
649 if (!rs->max_hw_segments)
650 rs->max_hw_segments = MAX_HW_SEGMENTS;
651 if (!rs->hardsect_size)
652 rs->hardsect_size = 1 << SECTOR_SHIFT;
653 if (!rs->max_segment_size)
654 rs->max_segment_size = MAX_SEGMENT_SIZE;
655 if (!rs->seg_boundary_mask)
656 rs->seg_boundary_mask = -1;
659 int dm_table_add_target(struct dm_table *t, const char *type,
660 sector_t start, sector_t len, char *params)
662 int r = -EINVAL, argc;
664 struct dm_target *tgt;
666 if ((r = check_space(t)))
669 tgt = t->targets + t->num_targets;
670 memset(tgt, 0, sizeof(*tgt));
673 tgt->error = "zero-length target";
674 DMERR("%s", tgt->error);
678 tgt->type = dm_get_target_type(type);
680 tgt->error = "unknown target type";
681 DMERR("%s", tgt->error);
688 tgt->error = "Unknown error";
691 * Does this target adjoin the previous one ?
693 if (!adjoin(t, tgt)) {
694 tgt->error = "Gap in table";
699 r = dm_split_args(&argc, &argv, params);
701 tgt->error = "couldn't split parameters (insufficient memory)";
705 r = tgt->type->ctr(tgt, argc, argv);
710 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
712 /* FIXME: the plan is to combine high here and then have
713 * the merge fn apply the target level restrictions. */
714 combine_restrictions_low(&t->limits, &tgt->limits);
718 DMERR("%s", tgt->error);
719 dm_put_target_type(tgt->type);
723 static int setup_indexes(struct dm_table *t)
726 unsigned int total = 0;
729 /* allocate the space for *all* the indexes */
730 for (i = t->depth - 2; i >= 0; i--) {
731 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
732 total += t->counts[i];
735 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
739 /* set up internal nodes, bottom-up */
740 for (i = t->depth - 2, total = 0; i >= 0; i--) {
741 t->index[i] = indexes;
742 indexes += (KEYS_PER_NODE * t->counts[i]);
743 setup_btree_index(i, t);
750 * Builds the btree to index the map.
752 int dm_table_complete(struct dm_table *t)
755 unsigned int leaf_nodes;
757 check_for_valid_limits(&t->limits);
759 /* how many indexes will the btree have ? */
760 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
761 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
763 /* leaf layer has already been set up */
764 t->counts[t->depth - 1] = leaf_nodes;
765 t->index[t->depth - 1] = t->highs;
768 r = setup_indexes(t);
773 static DECLARE_MUTEX(_event_lock);
774 void dm_table_event_callback(struct dm_table *t,
775 void (*fn)(void *), void *context)
779 t->event_context = context;
783 void dm_table_event(struct dm_table *t)
786 * You can no longer call dm_table_event() from interrupt
787 * context, use a bottom half instead.
789 BUG_ON(in_interrupt());
793 t->event_fn(t->event_context);
797 sector_t dm_table_get_size(struct dm_table *t)
799 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
802 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
804 if (index > t->num_targets)
807 return t->targets + index;
811 * Search the btree for the correct target.
813 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
815 unsigned int l, n = 0, k = 0;
818 for (l = 0; l < t->depth; l++) {
820 node = get_node(t, l, n);
822 for (k = 0; k < KEYS_PER_NODE; k++)
823 if (node[k] >= sector)
827 return &t->targets[(KEYS_PER_NODE * n) + k];
830 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
833 * Make sure we obey the optimistic sub devices
836 blk_queue_max_sectors(q, t->limits.max_sectors);
837 q->max_phys_segments = t->limits.max_phys_segments;
838 q->max_hw_segments = t->limits.max_hw_segments;
839 q->hardsect_size = t->limits.hardsect_size;
840 q->max_segment_size = t->limits.max_segment_size;
841 q->seg_boundary_mask = t->limits.seg_boundary_mask;
842 if (t->limits.no_cluster)
843 q->queue_flags &= ~(1 << QUEUE_FLAG_CLUSTER);
845 q->queue_flags |= (1 << QUEUE_FLAG_CLUSTER);
849 unsigned int dm_table_get_num_targets(struct dm_table *t)
851 return t->num_targets;
854 struct list_head *dm_table_get_devices(struct dm_table *t)
859 int dm_table_get_mode(struct dm_table *t)
864 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
866 int i = t->num_targets;
867 struct dm_target *ti = t->targets;
871 if (ti->type->postsuspend)
872 ti->type->postsuspend(ti);
873 } else if (ti->type->presuspend)
874 ti->type->presuspend(ti);
880 void dm_table_presuspend_targets(struct dm_table *t)
885 return suspend_targets(t, 0);
888 void dm_table_postsuspend_targets(struct dm_table *t)
893 return suspend_targets(t, 1);
896 void dm_table_resume_targets(struct dm_table *t)
900 for (i = 0; i < t->num_targets; i++) {
901 struct dm_target *ti = t->targets + i;
903 if (ti->type->resume)
904 ti->type->resume(ti);
908 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
910 struct list_head *d, *devices;
913 devices = dm_table_get_devices(t);
914 for (d = devices->next; d != devices; d = d->next) {
915 struct dm_dev *dd = list_entry(d, struct dm_dev, list);
916 request_queue_t *q = bdev_get_queue(dd->bdev);
917 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
923 void dm_table_unplug_all(struct dm_table *t)
925 struct list_head *d, *devices = dm_table_get_devices(t);
927 for (d = devices->next; d != devices; d = d->next) {
928 struct dm_dev *dd = list_entry(d, struct dm_dev, list);
929 request_queue_t *q = bdev_get_queue(dd->bdev);
936 int dm_table_flush_all(struct dm_table *t)
938 struct list_head *d, *devices = dm_table_get_devices(t);
941 for (d = devices->next; d != devices; d = d->next) {
942 struct dm_dev *dd = list_entry(d, struct dm_dev, list);
943 request_queue_t *q = bdev_get_queue(dd->bdev);
946 if (!q->issue_flush_fn)
949 err = q->issue_flush_fn(q, dd->bdev->bd_disk, NULL);
958 struct mapped_device *dm_table_get_md(struct dm_table *t)
965 EXPORT_SYMBOL(dm_vcalloc);
966 EXPORT_SYMBOL(dm_get_device);
967 EXPORT_SYMBOL(dm_put_device);
968 EXPORT_SYMBOL(dm_table_event);
969 EXPORT_SYMBOL(dm_table_get_size);
970 EXPORT_SYMBOL(dm_table_get_mode);
971 EXPORT_SYMBOL(dm_table_get_md);
972 EXPORT_SYMBOL(dm_table_put);
973 EXPORT_SYMBOL(dm_table_get);
974 EXPORT_SYMBOL(dm_table_unplug_all);
975 EXPORT_SYMBOL(dm_table_flush_all);