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)
29 unsigned int counts[MAX_DEPTH]; /* in nodes */
30 sector_t *index[MAX_DEPTH];
32 unsigned int num_targets;
33 unsigned int num_allocated;
35 struct dm_target *targets;
38 * Indicates the rw permissions for the new logical
39 * device. This should be a combination of FMODE_READ
44 /* a list of devices used by this table */
45 struct list_head devices;
48 * These are optimistic limits taken from all the
49 * targets, some targets will need smaller limits.
51 struct io_restrictions limits;
53 /* events get handed up using this callback */
54 void (*event_fn)(void *);
59 * Similar to ceiling(log_size(n))
61 static unsigned int int_log(unsigned int n, unsigned int base)
66 n = dm_div_up(n, base);
74 * Returns the minimum that is _not_ zero, unless both are zero.
76 #define min_not_zero(l, r) (l == 0) ? r : ((r == 0) ? l : min(l, r))
79 * Combine two io_restrictions, always taking the lower value.
81 static void combine_restrictions_low(struct io_restrictions *lhs,
82 struct io_restrictions *rhs)
85 min_not_zero(lhs->max_sectors, rhs->max_sectors);
87 lhs->max_phys_segments =
88 min_not_zero(lhs->max_phys_segments, rhs->max_phys_segments);
90 lhs->max_hw_segments =
91 min_not_zero(lhs->max_hw_segments, rhs->max_hw_segments);
93 lhs->hardsect_size = max(lhs->hardsect_size, rhs->hardsect_size);
95 lhs->max_segment_size =
96 min_not_zero(lhs->max_segment_size, rhs->max_segment_size);
98 lhs->seg_boundary_mask =
99 min_not_zero(lhs->seg_boundary_mask, rhs->seg_boundary_mask);
101 lhs->no_cluster |= rhs->no_cluster;
105 * Calculate the index of the child node of the n'th node k'th key.
107 static inline unsigned int get_child(unsigned int n, unsigned int k)
109 return (n * CHILDREN_PER_NODE) + k;
113 * Return the n'th node of level l from table t.
115 static inline sector_t *get_node(struct dm_table *t,
116 unsigned int l, unsigned int n)
118 return t->index[l] + (n * KEYS_PER_NODE);
122 * Return the highest key that you could lookup from the n'th
123 * node on level l of the btree.
125 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
127 for (; l < t->depth - 1; l++)
128 n = get_child(n, CHILDREN_PER_NODE - 1);
130 if (n >= t->counts[l])
131 return (sector_t) - 1;
133 return get_node(t, l, n)[KEYS_PER_NODE - 1];
137 * Fills in a level of the btree based on the highs of the level
140 static int setup_btree_index(unsigned int l, struct dm_table *t)
145 for (n = 0U; n < t->counts[l]; n++) {
146 node = get_node(t, l, n);
148 for (k = 0U; k < KEYS_PER_NODE; k++)
149 node[k] = high(t, l + 1, get_child(n, k));
155 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
161 * Check that we're not going to overflow.
163 if (nmemb > (ULONG_MAX / elem_size))
166 size = nmemb * elem_size;
167 addr = vmalloc(size);
169 memset(addr, 0, size);
175 * highs, and targets are managed as dynamic arrays during a
178 static int alloc_targets(struct dm_table *t, unsigned int num)
181 struct dm_target *n_targets;
182 int n = t->num_targets;
185 * Allocate both the target array and offset array at once.
187 n_highs = (sector_t *) dm_vcalloc(num, sizeof(struct dm_target) +
192 n_targets = (struct dm_target *) (n_highs + num);
195 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
196 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
199 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
202 t->num_allocated = num;
204 t->targets = n_targets;
209 int dm_table_create(struct dm_table **result, int mode, unsigned num_targets)
211 struct dm_table *t = kmalloc(sizeof(*t), GFP_KERNEL);
216 memset(t, 0, sizeof(*t));
217 INIT_LIST_HEAD(&t->devices);
218 atomic_set(&t->holders, 1);
221 num_targets = KEYS_PER_NODE;
223 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
225 if (alloc_targets(t, num_targets)) {
236 static void free_devices(struct list_head *devices)
238 struct list_head *tmp, *next;
240 for (tmp = devices->next; tmp != devices; tmp = next) {
241 struct dm_dev *dd = list_entry(tmp, struct dm_dev, list);
247 static void table_destroy(struct dm_table *t)
251 /* free the indexes (see dm_table_complete) */
253 vfree(t->index[t->depth - 2]);
255 /* free the targets */
256 for (i = 0; i < t->num_targets; i++) {
257 struct dm_target *tgt = t->targets + i;
262 dm_put_target_type(tgt->type);
267 /* free the device list */
268 if (t->devices.next != &t->devices) {
269 DMWARN("devices still present during destroy: "
270 "dm_table_remove_device calls missing");
272 free_devices(&t->devices);
278 void dm_table_get(struct dm_table *t)
280 atomic_inc(&t->holders);
283 void dm_table_put(struct dm_table *t)
288 if (atomic_dec_and_test(&t->holders))
293 * Checks to see if we need to extend highs or targets.
295 static inline int check_space(struct dm_table *t)
297 if (t->num_targets >= t->num_allocated)
298 return alloc_targets(t, t->num_allocated * 2);
304 * Convert a device path to a dev_t.
306 static int lookup_device(const char *path, dev_t *dev)
312 if ((r = path_lookup(path, LOOKUP_FOLLOW, &nd)))
315 inode = nd.dentry->d_inode;
321 if (!S_ISBLK(inode->i_mode)) {
326 *dev = inode->i_rdev;
334 * See if we've already got a device in the list.
336 static struct dm_dev *find_device(struct list_head *l, dev_t dev)
340 list_for_each_entry (dd, l, list)
341 if (dd->bdev->bd_dev == dev)
348 * Open a device so we can use it as a map destination.
350 static int open_dev(struct dm_dev *d, dev_t dev)
352 static char *_claim_ptr = "I belong to device-mapper";
353 struct block_device *bdev;
359 bdev = open_by_devnum(dev, d->mode);
361 return PTR_ERR(bdev);
362 r = bd_claim(bdev, _claim_ptr);
371 * Close a device that we've been using.
373 static void close_dev(struct dm_dev *d)
384 * If possible (ie. blk_size[major] is set), this checks an area
385 * of a destination device is valid.
387 static int check_device_area(struct dm_dev *dd, sector_t start, sector_t len)
390 dev_size = dd->bdev->bd_inode->i_size >> SECTOR_SHIFT;
391 return ((start < dev_size) && (len <= (dev_size - start)));
395 * This upgrades the mode on an already open dm_dev. Being
396 * careful to leave things as they were if we fail to reopen the
399 static int upgrade_mode(struct dm_dev *dd, int new_mode)
402 struct dm_dev dd_copy;
403 dev_t dev = dd->bdev->bd_dev;
407 dd->mode |= new_mode;
409 r = open_dev(dd, dev);
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)
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);
453 if ((r = open_dev(dd, dev))) {
458 format_dev_t(dd->name, dev);
460 atomic_set(&dd->count, 0);
461 list_add(&dd->list, &t->devices);
463 } else if (dd->mode != (mode | dd->mode)) {
464 r = upgrade_mode(dd, mode);
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);
482 int dm_get_device(struct dm_target *ti, const char *path, sector_t start,
483 sector_t len, int mode, struct dm_dev **result)
485 int r = __table_get_device(ti->table, ti, path,
486 start, len, mode, result);
488 request_queue_t *q = bdev_get_queue((*result)->bdev);
489 struct io_restrictions *rs = &ti->limits;
492 * Combine the device limits low.
494 * FIXME: if we move an io_restriction struct
495 * into q this would just be a call to
496 * combine_restrictions_low()
499 min_not_zero(rs->max_sectors, q->max_sectors);
501 /* FIXME: Device-Mapper on top of RAID-0 breaks because DM
502 * currently doesn't honor MD's merge_bvec_fn routine.
503 * In this case, we'll force DM to use PAGE_SIZE or
504 * smaller I/O, just to be safe. A better fix is in the
505 * works, but add this for the time being so it will at
506 * least operate correctly.
508 if (q->merge_bvec_fn)
510 min_not_zero(rs->max_sectors,
511 (unsigned int) (PAGE_SIZE >> 9));
513 rs->max_phys_segments =
514 min_not_zero(rs->max_phys_segments,
515 q->max_phys_segments);
517 rs->max_hw_segments =
518 min_not_zero(rs->max_hw_segments, q->max_hw_segments);
520 rs->hardsect_size = max(rs->hardsect_size, q->hardsect_size);
522 rs->max_segment_size =
523 min_not_zero(rs->max_segment_size, q->max_segment_size);
525 rs->seg_boundary_mask =
526 min_not_zero(rs->seg_boundary_mask,
527 q->seg_boundary_mask);
529 rs->no_cluster |= !test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags);
536 * Decrement a devices use count and remove it if necessary.
538 void dm_put_device(struct dm_target *ti, struct dm_dev *dd)
540 if (atomic_dec_and_test(&dd->count)) {
548 * Checks to see if the target joins onto the end of the table.
550 static int adjoin(struct dm_table *table, struct dm_target *ti)
552 struct dm_target *prev;
554 if (!table->num_targets)
557 prev = &table->targets[table->num_targets - 1];
558 return (ti->begin == (prev->begin + prev->len));
562 * Used to dynamically allocate the arg array.
564 static char **realloc_argv(unsigned *array_size, char **old_argv)
569 new_size = *array_size ? *array_size * 2 : 64;
570 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
572 memcpy(argv, old_argv, *array_size * sizeof(*argv));
573 *array_size = new_size;
581 * Destructively splits up the argument list to pass to ctr.
583 int dm_split_args(int *argc, char ***argvp, char *input)
585 char *start, *end = input, *out, **argv = NULL;
586 unsigned array_size = 0;
589 argv = realloc_argv(&array_size, argv);
596 /* Skip whitespace */
597 while (*start && isspace(*start))
601 break; /* success, we hit the end */
603 /* 'out' is used to remove any back-quotes */
606 /* Everything apart from '\0' can be quoted */
607 if (*end == '\\' && *(end + 1)) {
614 break; /* end of token */
619 /* have we already filled the array ? */
620 if ((*argc + 1) > array_size) {
621 argv = realloc_argv(&array_size, argv);
626 /* we know this is whitespace */
630 /* terminate the string and put it in the array */
640 static void check_for_valid_limits(struct io_restrictions *rs)
642 if (!rs->max_sectors)
643 rs->max_sectors = SAFE_MAX_SECTORS;
644 if (!rs->max_phys_segments)
645 rs->max_phys_segments = MAX_PHYS_SEGMENTS;
646 if (!rs->max_hw_segments)
647 rs->max_hw_segments = MAX_HW_SEGMENTS;
648 if (!rs->hardsect_size)
649 rs->hardsect_size = 1 << SECTOR_SHIFT;
650 if (!rs->max_segment_size)
651 rs->max_segment_size = MAX_SEGMENT_SIZE;
652 if (!rs->seg_boundary_mask)
653 rs->seg_boundary_mask = -1;
656 int dm_table_add_target(struct dm_table *t, const char *type,
657 sector_t start, sector_t len, char *params)
659 int r = -EINVAL, argc;
661 struct dm_target *tgt;
663 if ((r = check_space(t)))
666 tgt = t->targets + t->num_targets;
667 memset(tgt, 0, sizeof(*tgt));
670 tgt->error = "zero-length target";
671 DMERR("%s", tgt->error);
675 tgt->type = dm_get_target_type(type);
677 tgt->error = "unknown target type";
678 DMERR("%s", tgt->error);
685 tgt->error = "Unknown error";
688 * Does this target adjoin the previous one ?
690 if (!adjoin(t, tgt)) {
691 tgt->error = "Gap in table";
696 r = dm_split_args(&argc, &argv, params);
698 tgt->error = "couldn't split parameters (insufficient memory)";
702 r = tgt->type->ctr(tgt, argc, argv);
707 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
709 /* FIXME: the plan is to combine high here and then have
710 * the merge fn apply the target level restrictions. */
711 combine_restrictions_low(&t->limits, &tgt->limits);
715 DMERR("%s", tgt->error);
716 dm_put_target_type(tgt->type);
720 static int setup_indexes(struct dm_table *t)
723 unsigned int total = 0;
726 /* allocate the space for *all* the indexes */
727 for (i = t->depth - 2; i >= 0; i--) {
728 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
729 total += t->counts[i];
732 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
736 /* set up internal nodes, bottom-up */
737 for (i = t->depth - 2, total = 0; i >= 0; i--) {
738 t->index[i] = indexes;
739 indexes += (KEYS_PER_NODE * t->counts[i]);
740 setup_btree_index(i, t);
747 * Builds the btree to index the map.
749 int dm_table_complete(struct dm_table *t)
752 unsigned int leaf_nodes;
754 check_for_valid_limits(&t->limits);
756 /* how many indexes will the btree have ? */
757 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
758 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
760 /* leaf layer has already been set up */
761 t->counts[t->depth - 1] = leaf_nodes;
762 t->index[t->depth - 1] = t->highs;
765 r = setup_indexes(t);
770 static DECLARE_MUTEX(_event_lock);
771 void dm_table_event_callback(struct dm_table *t,
772 void (*fn)(void *), void *context)
776 t->event_context = context;
780 void dm_table_event(struct dm_table *t)
783 * You can no longer call dm_table_event() from interrupt
784 * context, use a bottom half instead.
786 BUG_ON(in_interrupt());
790 t->event_fn(t->event_context);
794 sector_t dm_table_get_size(struct dm_table *t)
796 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
799 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
801 if (index > t->num_targets)
804 return t->targets + index;
808 * Search the btree for the correct target.
810 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
812 unsigned int l, n = 0, k = 0;
815 for (l = 0; l < t->depth; l++) {
817 node = get_node(t, l, n);
819 for (k = 0; k < KEYS_PER_NODE; k++)
820 if (node[k] >= sector)
824 return &t->targets[(KEYS_PER_NODE * n) + k];
827 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q)
830 * Make sure we obey the optimistic sub devices
833 blk_queue_max_sectors(q, t->limits.max_sectors);
834 q->max_phys_segments = t->limits.max_phys_segments;
835 q->max_hw_segments = t->limits.max_hw_segments;
836 q->hardsect_size = t->limits.hardsect_size;
837 q->max_segment_size = t->limits.max_segment_size;
838 q->seg_boundary_mask = t->limits.seg_boundary_mask;
839 if (t->limits.no_cluster)
840 q->queue_flags &= ~(1 << QUEUE_FLAG_CLUSTER);
842 q->queue_flags |= (1 << QUEUE_FLAG_CLUSTER);
846 unsigned int dm_table_get_num_targets(struct dm_table *t)
848 return t->num_targets;
851 struct list_head *dm_table_get_devices(struct dm_table *t)
856 int dm_table_get_mode(struct dm_table *t)
861 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
863 int i = t->num_targets;
864 struct dm_target *ti = t->targets;
868 if (ti->type->postsuspend)
869 ti->type->postsuspend(ti);
870 } else if (ti->type->presuspend)
871 ti->type->presuspend(ti);
877 void dm_table_presuspend_targets(struct dm_table *t)
882 return suspend_targets(t, 0);
885 void dm_table_postsuspend_targets(struct dm_table *t)
890 return suspend_targets(t, 1);
893 void dm_table_resume_targets(struct dm_table *t)
897 for (i = 0; i < t->num_targets; i++) {
898 struct dm_target *ti = t->targets + i;
900 if (ti->type->resume)
901 ti->type->resume(ti);
905 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
907 struct list_head *d, *devices;
910 devices = dm_table_get_devices(t);
911 for (d = devices->next; d != devices; d = d->next) {
912 struct dm_dev *dd = list_entry(d, struct dm_dev, list);
913 request_queue_t *q = bdev_get_queue(dd->bdev);
914 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
920 void dm_table_unplug_all(struct dm_table *t)
922 struct list_head *d, *devices = dm_table_get_devices(t);
924 for (d = devices->next; d != devices; d = d->next) {
925 struct dm_dev *dd = list_entry(d, struct dm_dev, list);
926 request_queue_t *q = bdev_get_queue(dd->bdev);
933 int dm_table_flush_all(struct dm_table *t)
935 struct list_head *d, *devices = dm_table_get_devices(t);
938 for (d = devices->next; d != devices; d = d->next) {
939 struct dm_dev *dd = list_entry(d, struct dm_dev, list);
940 request_queue_t *q = bdev_get_queue(dd->bdev);
943 if (!q->issue_flush_fn)
946 err = q->issue_flush_fn(q, dd->bdev->bd_disk, NULL);
955 EXPORT_SYMBOL(dm_vcalloc);
956 EXPORT_SYMBOL(dm_get_device);
957 EXPORT_SYMBOL(dm_put_device);
958 EXPORT_SYMBOL(dm_table_event);
959 EXPORT_SYMBOL(dm_table_get_size);
960 EXPORT_SYMBOL(dm_table_get_mode);
961 EXPORT_SYMBOL(dm_table_put);
962 EXPORT_SYMBOL(dm_table_get);
963 EXPORT_SYMBOL(dm_table_unplug_all);
964 EXPORT_SYMBOL(dm_table_flush_all);