Add linux-next specific files for 20110421
[linux-2.6/next.git] / drivers / md / dm-table.c
blobcb8380c9767fd3d5bb815d7fec4a46308b0270c5
1 /*
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.
6 */
8 #include "dm.h"
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/string.h>
16 #include <linux/slab.h>
17 #include <linux/interrupt.h>
18 #include <linux/mutex.h>
19 #include <linux/delay.h>
20 #include <asm/atomic.h>
22 #define DM_MSG_PREFIX "table"
24 #define MAX_DEPTH 16
25 #define NODE_SIZE L1_CACHE_BYTES
26 #define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
27 #define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
30 * The table has always exactly one reference from either mapped_device->map
31 * or hash_cell->new_map. This reference is not counted in table->holders.
32 * A pair of dm_create_table/dm_destroy_table functions is used for table
33 * creation/destruction.
35 * Temporary references from the other code increase table->holders. A pair
36 * of dm_table_get/dm_table_put functions is used to manipulate it.
38 * When the table is about to be destroyed, we wait for table->holders to
39 * drop to zero.
42 struct dm_table {
43 struct mapped_device *md;
44 atomic_t holders;
45 unsigned type;
47 /* btree table */
48 unsigned int depth;
49 unsigned int counts[MAX_DEPTH]; /* in nodes */
50 sector_t *index[MAX_DEPTH];
52 unsigned int num_targets;
53 unsigned int num_allocated;
54 sector_t *highs;
55 struct dm_target *targets;
57 unsigned discards_supported:1;
58 unsigned integrity_supported:1;
61 * Indicates the rw permissions for the new logical
62 * device. This should be a combination of FMODE_READ
63 * and FMODE_WRITE.
65 fmode_t mode;
67 /* a list of devices used by this table */
68 struct list_head devices;
70 /* events get handed up using this callback */
71 void (*event_fn)(void *);
72 void *event_context;
74 struct dm_md_mempools *mempools;
76 struct list_head target_callbacks;
80 * Similar to ceiling(log_size(n))
82 static unsigned int int_log(unsigned int n, unsigned int base)
84 int result = 0;
86 while (n > 1) {
87 n = dm_div_up(n, base);
88 result++;
91 return result;
95 * Calculate the index of the child node of the n'th node k'th key.
97 static inline unsigned int get_child(unsigned int n, unsigned int k)
99 return (n * CHILDREN_PER_NODE) + k;
103 * Return the n'th node of level l from table t.
105 static inline sector_t *get_node(struct dm_table *t,
106 unsigned int l, unsigned int n)
108 return t->index[l] + (n * KEYS_PER_NODE);
112 * Return the highest key that you could lookup from the n'th
113 * node on level l of the btree.
115 static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
117 for (; l < t->depth - 1; l++)
118 n = get_child(n, CHILDREN_PER_NODE - 1);
120 if (n >= t->counts[l])
121 return (sector_t) - 1;
123 return get_node(t, l, n)[KEYS_PER_NODE - 1];
127 * Fills in a level of the btree based on the highs of the level
128 * below it.
130 static int setup_btree_index(unsigned int l, struct dm_table *t)
132 unsigned int n, k;
133 sector_t *node;
135 for (n = 0U; n < t->counts[l]; n++) {
136 node = get_node(t, l, n);
138 for (k = 0U; k < KEYS_PER_NODE; k++)
139 node[k] = high(t, l + 1, get_child(n, k));
142 return 0;
145 void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
147 unsigned long size;
148 void *addr;
151 * Check that we're not going to overflow.
153 if (nmemb > (ULONG_MAX / elem_size))
154 return NULL;
156 size = nmemb * elem_size;
157 addr = vmalloc(size);
158 if (addr)
159 memset(addr, 0, size);
161 return addr;
165 * highs, and targets are managed as dynamic arrays during a
166 * table load.
168 static int alloc_targets(struct dm_table *t, unsigned int num)
170 sector_t *n_highs;
171 struct dm_target *n_targets;
172 int n = t->num_targets;
175 * Allocate both the target array and offset array at once.
176 * Append an empty entry to catch sectors beyond the end of
177 * the device.
179 n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
180 sizeof(sector_t));
181 if (!n_highs)
182 return -ENOMEM;
184 n_targets = (struct dm_target *) (n_highs + num);
186 if (n) {
187 memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
188 memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
191 memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
192 vfree(t->highs);
194 t->num_allocated = num;
195 t->highs = n_highs;
196 t->targets = n_targets;
198 return 0;
201 int dm_table_create(struct dm_table **result, fmode_t mode,
202 unsigned num_targets, struct mapped_device *md)
204 struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
206 if (!t)
207 return -ENOMEM;
209 INIT_LIST_HEAD(&t->devices);
210 INIT_LIST_HEAD(&t->target_callbacks);
211 atomic_set(&t->holders, 0);
212 t->discards_supported = 1;
214 if (!num_targets)
215 num_targets = KEYS_PER_NODE;
217 num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
219 if (alloc_targets(t, num_targets)) {
220 kfree(t);
221 t = NULL;
222 return -ENOMEM;
225 t->mode = mode;
226 t->md = md;
227 *result = t;
228 return 0;
231 static void free_devices(struct list_head *devices)
233 struct list_head *tmp, *next;
235 list_for_each_safe(tmp, next, devices) {
236 struct dm_dev_internal *dd =
237 list_entry(tmp, struct dm_dev_internal, list);
238 DMWARN("dm_table_destroy: dm_put_device call missing for %s",
239 dd->dm_dev.name);
240 kfree(dd);
244 void dm_table_destroy(struct dm_table *t)
246 unsigned int i;
248 if (!t)
249 return;
251 while (atomic_read(&t->holders))
252 msleep(1);
253 smp_mb();
255 /* free the indexes */
256 if (t->depth >= 2)
257 vfree(t->index[t->depth - 2]);
259 /* free the targets */
260 for (i = 0; i < t->num_targets; i++) {
261 struct dm_target *tgt = t->targets + i;
263 if (tgt->type->dtr)
264 tgt->type->dtr(tgt);
266 dm_put_target_type(tgt->type);
269 vfree(t->highs);
271 /* free the device list */
272 if (t->devices.next != &t->devices)
273 free_devices(&t->devices);
275 dm_free_md_mempools(t->mempools);
277 kfree(t);
280 void dm_table_get(struct dm_table *t)
282 atomic_inc(&t->holders);
285 void dm_table_put(struct dm_table *t)
287 if (!t)
288 return;
290 smp_mb__before_atomic_dec();
291 atomic_dec(&t->holders);
295 * Checks to see if we need to extend highs or targets.
297 static inline int check_space(struct dm_table *t)
299 if (t->num_targets >= t->num_allocated)
300 return alloc_targets(t, t->num_allocated * 2);
302 return 0;
306 * See if we've already got a device in the list.
308 static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
310 struct dm_dev_internal *dd;
312 list_for_each_entry (dd, l, list)
313 if (dd->dm_dev.bdev->bd_dev == dev)
314 return dd;
316 return NULL;
320 * Open a device so we can use it as a map destination.
322 static int open_dev(struct dm_dev_internal *d, dev_t dev,
323 struct mapped_device *md)
325 static char *_claim_ptr = "I belong to device-mapper";
326 struct block_device *bdev;
328 int r;
330 BUG_ON(d->dm_dev.bdev);
332 bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
333 if (IS_ERR(bdev))
334 return PTR_ERR(bdev);
336 r = bd_link_disk_holder(bdev, dm_disk(md));
337 if (r) {
338 blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
339 return r;
342 d->dm_dev.bdev = bdev;
343 return 0;
347 * Close a device that we've been using.
349 static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
351 if (!d->dm_dev.bdev)
352 return;
354 bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
355 blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
356 d->dm_dev.bdev = NULL;
360 * If possible, this checks an area of a destination device is invalid.
362 static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
363 sector_t start, sector_t len, void *data)
365 struct queue_limits *limits = data;
366 struct block_device *bdev = dev->bdev;
367 sector_t dev_size =
368 i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
369 unsigned short logical_block_size_sectors =
370 limits->logical_block_size >> SECTOR_SHIFT;
371 char b[BDEVNAME_SIZE];
373 if (!dev_size)
374 return 0;
376 if ((start >= dev_size) || (start + len > dev_size)) {
377 DMWARN("%s: %s too small for target: "
378 "start=%llu, len=%llu, dev_size=%llu",
379 dm_device_name(ti->table->md), bdevname(bdev, b),
380 (unsigned long long)start,
381 (unsigned long long)len,
382 (unsigned long long)dev_size);
383 return 1;
386 if (logical_block_size_sectors <= 1)
387 return 0;
389 if (start & (logical_block_size_sectors - 1)) {
390 DMWARN("%s: start=%llu not aligned to h/w "
391 "logical block size %u of %s",
392 dm_device_name(ti->table->md),
393 (unsigned long long)start,
394 limits->logical_block_size, bdevname(bdev, b));
395 return 1;
398 if (len & (logical_block_size_sectors - 1)) {
399 DMWARN("%s: len=%llu not aligned to h/w "
400 "logical block size %u of %s",
401 dm_device_name(ti->table->md),
402 (unsigned long long)len,
403 limits->logical_block_size, bdevname(bdev, b));
404 return 1;
407 return 0;
411 * This upgrades the mode on an already open dm_dev, being
412 * careful to leave things as they were if we fail to reopen the
413 * device and not to touch the existing bdev field in case
414 * it is accessed concurrently inside dm_table_any_congested().
416 static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
417 struct mapped_device *md)
419 int r;
420 struct dm_dev_internal dd_new, dd_old;
422 dd_new = dd_old = *dd;
424 dd_new.dm_dev.mode |= new_mode;
425 dd_new.dm_dev.bdev = NULL;
427 r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
428 if (r)
429 return r;
431 dd->dm_dev.mode |= new_mode;
432 close_dev(&dd_old, md);
434 return 0;
438 * Add a device to the list, or just increment the usage count if
439 * it's already present.
441 static int __table_get_device(struct dm_table *t, struct dm_target *ti,
442 const char *path, fmode_t mode, struct dm_dev **result)
444 int r;
445 dev_t uninitialized_var(dev);
446 struct dm_dev_internal *dd;
447 unsigned int major, minor;
449 BUG_ON(!t);
451 if (sscanf(path, "%u:%u", &major, &minor) == 2) {
452 /* Extract the major/minor numbers */
453 dev = MKDEV(major, minor);
454 if (MAJOR(dev) != major || MINOR(dev) != minor)
455 return -EOVERFLOW;
456 } else {
457 /* convert the path to a device */
458 struct block_device *bdev = lookup_bdev(path);
460 if (IS_ERR(bdev))
461 return PTR_ERR(bdev);
462 dev = bdev->bd_dev;
463 bdput(bdev);
466 dd = find_device(&t->devices, dev);
467 if (!dd) {
468 dd = kmalloc(sizeof(*dd), GFP_KERNEL);
469 if (!dd)
470 return -ENOMEM;
472 dd->dm_dev.mode = mode;
473 dd->dm_dev.bdev = NULL;
475 if ((r = open_dev(dd, dev, t->md))) {
476 kfree(dd);
477 return r;
480 format_dev_t(dd->dm_dev.name, dev);
482 atomic_set(&dd->count, 0);
483 list_add(&dd->list, &t->devices);
485 } else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
486 r = upgrade_mode(dd, mode, t->md);
487 if (r)
488 return r;
490 atomic_inc(&dd->count);
492 *result = &dd->dm_dev;
493 return 0;
496 int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
497 sector_t start, sector_t len, void *data)
499 struct queue_limits *limits = data;
500 struct block_device *bdev = dev->bdev;
501 struct request_queue *q = bdev_get_queue(bdev);
502 char b[BDEVNAME_SIZE];
504 if (unlikely(!q)) {
505 DMWARN("%s: Cannot set limits for nonexistent device %s",
506 dm_device_name(ti->table->md), bdevname(bdev, b));
507 return 0;
510 if (bdev_stack_limits(limits, bdev, start) < 0)
511 DMWARN("%s: adding target device %s caused an alignment inconsistency: "
512 "physical_block_size=%u, logical_block_size=%u, "
513 "alignment_offset=%u, start=%llu",
514 dm_device_name(ti->table->md), bdevname(bdev, b),
515 q->limits.physical_block_size,
516 q->limits.logical_block_size,
517 q->limits.alignment_offset,
518 (unsigned long long) start << SECTOR_SHIFT);
521 * Check if merge fn is supported.
522 * If not we'll force DM to use PAGE_SIZE or
523 * smaller I/O, just to be safe.
526 if (q->merge_bvec_fn && !ti->type->merge)
527 blk_limits_max_hw_sectors(limits,
528 (unsigned int) (PAGE_SIZE >> 9));
529 return 0;
531 EXPORT_SYMBOL_GPL(dm_set_device_limits);
533 int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
534 struct dm_dev **result)
536 return __table_get_device(ti->table, ti, path, mode, result);
541 * Decrement a devices use count and remove it if necessary.
543 void dm_put_device(struct dm_target *ti, struct dm_dev *d)
545 struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
546 dm_dev);
548 if (atomic_dec_and_test(&dd->count)) {
549 close_dev(dd, ti->table->md);
550 list_del(&dd->list);
551 kfree(dd);
556 * Checks to see if the target joins onto the end of the table.
558 static int adjoin(struct dm_table *table, struct dm_target *ti)
560 struct dm_target *prev;
562 if (!table->num_targets)
563 return !ti->begin;
565 prev = &table->targets[table->num_targets - 1];
566 return (ti->begin == (prev->begin + prev->len));
570 * Used to dynamically allocate the arg array.
572 static char **realloc_argv(unsigned *array_size, char **old_argv)
574 char **argv;
575 unsigned new_size;
577 new_size = *array_size ? *array_size * 2 : 64;
578 argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
579 if (argv) {
580 memcpy(argv, old_argv, *array_size * sizeof(*argv));
581 *array_size = new_size;
584 kfree(old_argv);
585 return argv;
589 * Destructively splits up the argument list to pass to ctr.
591 int dm_split_args(int *argc, char ***argvp, char *input)
593 char *start, *end = input, *out, **argv = NULL;
594 unsigned array_size = 0;
596 *argc = 0;
598 if (!input) {
599 *argvp = NULL;
600 return 0;
603 argv = realloc_argv(&array_size, argv);
604 if (!argv)
605 return -ENOMEM;
607 while (1) {
608 /* Skip whitespace */
609 start = skip_spaces(end);
611 if (!*start)
612 break; /* success, we hit the end */
614 /* 'out' is used to remove any back-quotes */
615 end = out = start;
616 while (*end) {
617 /* Everything apart from '\0' can be quoted */
618 if (*end == '\\' && *(end + 1)) {
619 *out++ = *(end + 1);
620 end += 2;
621 continue;
624 if (isspace(*end))
625 break; /* end of token */
627 *out++ = *end++;
630 /* have we already filled the array ? */
631 if ((*argc + 1) > array_size) {
632 argv = realloc_argv(&array_size, argv);
633 if (!argv)
634 return -ENOMEM;
637 /* we know this is whitespace */
638 if (*end)
639 end++;
641 /* terminate the string and put it in the array */
642 *out = '\0';
643 argv[*argc] = start;
644 (*argc)++;
647 *argvp = argv;
648 return 0;
652 * Impose necessary and sufficient conditions on a devices's table such
653 * that any incoming bio which respects its logical_block_size can be
654 * processed successfully. If it falls across the boundary between
655 * two or more targets, the size of each piece it gets split into must
656 * be compatible with the logical_block_size of the target processing it.
658 static int validate_hardware_logical_block_alignment(struct dm_table *table,
659 struct queue_limits *limits)
662 * This function uses arithmetic modulo the logical_block_size
663 * (in units of 512-byte sectors).
665 unsigned short device_logical_block_size_sects =
666 limits->logical_block_size >> SECTOR_SHIFT;
669 * Offset of the start of the next table entry, mod logical_block_size.
671 unsigned short next_target_start = 0;
674 * Given an aligned bio that extends beyond the end of a
675 * target, how many sectors must the next target handle?
677 unsigned short remaining = 0;
679 struct dm_target *uninitialized_var(ti);
680 struct queue_limits ti_limits;
681 unsigned i = 0;
684 * Check each entry in the table in turn.
686 while (i < dm_table_get_num_targets(table)) {
687 ti = dm_table_get_target(table, i++);
689 blk_set_default_limits(&ti_limits);
691 /* combine all target devices' limits */
692 if (ti->type->iterate_devices)
693 ti->type->iterate_devices(ti, dm_set_device_limits,
694 &ti_limits);
697 * If the remaining sectors fall entirely within this
698 * table entry are they compatible with its logical_block_size?
700 if (remaining < ti->len &&
701 remaining & ((ti_limits.logical_block_size >>
702 SECTOR_SHIFT) - 1))
703 break; /* Error */
705 next_target_start =
706 (unsigned short) ((next_target_start + ti->len) &
707 (device_logical_block_size_sects - 1));
708 remaining = next_target_start ?
709 device_logical_block_size_sects - next_target_start : 0;
712 if (remaining) {
713 DMWARN("%s: table line %u (start sect %llu len %llu) "
714 "not aligned to h/w logical block size %u",
715 dm_device_name(table->md), i,
716 (unsigned long long) ti->begin,
717 (unsigned long long) ti->len,
718 limits->logical_block_size);
719 return -EINVAL;
722 return 0;
725 int dm_table_add_target(struct dm_table *t, const char *type,
726 sector_t start, sector_t len, char *params)
728 int r = -EINVAL, argc;
729 char **argv;
730 struct dm_target *tgt;
732 if ((r = check_space(t)))
733 return r;
735 tgt = t->targets + t->num_targets;
736 memset(tgt, 0, sizeof(*tgt));
738 if (!len) {
739 DMERR("%s: zero-length target", dm_device_name(t->md));
740 return -EINVAL;
743 tgt->type = dm_get_target_type(type);
744 if (!tgt->type) {
745 DMERR("%s: %s: unknown target type", dm_device_name(t->md),
746 type);
747 return -EINVAL;
750 tgt->table = t;
751 tgt->begin = start;
752 tgt->len = len;
753 tgt->error = "Unknown error";
756 * Does this target adjoin the previous one ?
758 if (!adjoin(t, tgt)) {
759 tgt->error = "Gap in table";
760 r = -EINVAL;
761 goto bad;
764 r = dm_split_args(&argc, &argv, params);
765 if (r) {
766 tgt->error = "couldn't split parameters (insufficient memory)";
767 goto bad;
770 r = tgt->type->ctr(tgt, argc, argv);
771 kfree(argv);
772 if (r)
773 goto bad;
775 t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
777 if (!tgt->num_discard_requests)
778 t->discards_supported = 0;
780 return 0;
782 bad:
783 DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
784 dm_put_target_type(tgt->type);
785 return r;
788 static int dm_table_set_type(struct dm_table *t)
790 unsigned i;
791 unsigned bio_based = 0, request_based = 0;
792 struct dm_target *tgt;
793 struct dm_dev_internal *dd;
794 struct list_head *devices;
796 for (i = 0; i < t->num_targets; i++) {
797 tgt = t->targets + i;
798 if (dm_target_request_based(tgt))
799 request_based = 1;
800 else
801 bio_based = 1;
803 if (bio_based && request_based) {
804 DMWARN("Inconsistent table: different target types"
805 " can't be mixed up");
806 return -EINVAL;
810 if (bio_based) {
811 /* We must use this table as bio-based */
812 t->type = DM_TYPE_BIO_BASED;
813 return 0;
816 BUG_ON(!request_based); /* No targets in this table */
818 /* Non-request-stackable devices can't be used for request-based dm */
819 devices = dm_table_get_devices(t);
820 list_for_each_entry(dd, devices, list) {
821 if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
822 DMWARN("table load rejected: including"
823 " non-request-stackable devices");
824 return -EINVAL;
829 * Request-based dm supports only tables that have a single target now.
830 * To support multiple targets, request splitting support is needed,
831 * and that needs lots of changes in the block-layer.
832 * (e.g. request completion process for partial completion.)
834 if (t->num_targets > 1) {
835 DMWARN("Request-based dm doesn't support multiple targets yet");
836 return -EINVAL;
839 t->type = DM_TYPE_REQUEST_BASED;
841 return 0;
844 unsigned dm_table_get_type(struct dm_table *t)
846 return t->type;
849 bool dm_table_request_based(struct dm_table *t)
851 return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
854 int dm_table_alloc_md_mempools(struct dm_table *t)
856 unsigned type = dm_table_get_type(t);
858 if (unlikely(type == DM_TYPE_NONE)) {
859 DMWARN("no table type is set, can't allocate mempools");
860 return -EINVAL;
863 t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
864 if (!t->mempools)
865 return -ENOMEM;
867 return 0;
870 void dm_table_free_md_mempools(struct dm_table *t)
872 dm_free_md_mempools(t->mempools);
873 t->mempools = NULL;
876 struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
878 return t->mempools;
881 static int setup_indexes(struct dm_table *t)
883 int i;
884 unsigned int total = 0;
885 sector_t *indexes;
887 /* allocate the space for *all* the indexes */
888 for (i = t->depth - 2; i >= 0; i--) {
889 t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
890 total += t->counts[i];
893 indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
894 if (!indexes)
895 return -ENOMEM;
897 /* set up internal nodes, bottom-up */
898 for (i = t->depth - 2; i >= 0; i--) {
899 t->index[i] = indexes;
900 indexes += (KEYS_PER_NODE * t->counts[i]);
901 setup_btree_index(i, t);
904 return 0;
908 * Builds the btree to index the map.
910 static int dm_table_build_index(struct dm_table *t)
912 int r = 0;
913 unsigned int leaf_nodes;
915 /* how many indexes will the btree have ? */
916 leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
917 t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
919 /* leaf layer has already been set up */
920 t->counts[t->depth - 1] = leaf_nodes;
921 t->index[t->depth - 1] = t->highs;
923 if (t->depth >= 2)
924 r = setup_indexes(t);
926 return r;
930 * Get a disk whose integrity profile reflects the table's profile.
931 * If %match_all is true, all devices' profiles must match.
932 * If %match_all is false, all devices must at least have an
933 * allocated integrity profile; but uninitialized is ok.
934 * Returns NULL if integrity support was inconsistent or unavailable.
936 static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
937 bool match_all)
939 struct list_head *devices = dm_table_get_devices(t);
940 struct dm_dev_internal *dd = NULL;
941 struct gendisk *prev_disk = NULL, *template_disk = NULL;
943 list_for_each_entry(dd, devices, list) {
944 template_disk = dd->dm_dev.bdev->bd_disk;
945 if (!blk_get_integrity(template_disk))
946 goto no_integrity;
947 if (!match_all && !blk_integrity_is_initialized(template_disk))
948 continue; /* skip uninitialized profiles */
949 else if (prev_disk &&
950 blk_integrity_compare(prev_disk, template_disk) < 0)
951 goto no_integrity;
952 prev_disk = template_disk;
955 return template_disk;
957 no_integrity:
958 if (prev_disk)
959 DMWARN("%s: integrity not set: %s and %s profile mismatch",
960 dm_device_name(t->md),
961 prev_disk->disk_name,
962 template_disk->disk_name);
963 return NULL;
967 * Register the mapped device for blk_integrity support if
968 * the underlying devices have an integrity profile. But all devices
969 * may not have matching profiles (checking all devices isn't reliable
970 * during table load because this table may use other DM device(s) which
971 * must be resumed before they will have an initialized integity profile).
972 * Stacked DM devices force a 2 stage integrity profile validation:
973 * 1 - during load, validate all initialized integrity profiles match
974 * 2 - during resume, validate all integrity profiles match
976 static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
978 struct gendisk *template_disk = NULL;
980 template_disk = dm_table_get_integrity_disk(t, false);
981 if (!template_disk)
982 return 0;
984 if (!blk_integrity_is_initialized(dm_disk(md))) {
985 t->integrity_supported = 1;
986 return blk_integrity_register(dm_disk(md), NULL);
990 * If DM device already has an initalized integrity
991 * profile the new profile should not conflict.
993 if (blk_integrity_is_initialized(template_disk) &&
994 blk_integrity_compare(dm_disk(md), template_disk) < 0) {
995 DMWARN("%s: conflict with existing integrity profile: "
996 "%s profile mismatch",
997 dm_device_name(t->md),
998 template_disk->disk_name);
999 return 1;
1002 /* Preserve existing initialized integrity profile */
1003 t->integrity_supported = 1;
1004 return 0;
1008 * Prepares the table for use by building the indices,
1009 * setting the type, and allocating mempools.
1011 int dm_table_complete(struct dm_table *t)
1013 int r;
1015 r = dm_table_set_type(t);
1016 if (r) {
1017 DMERR("unable to set table type");
1018 return r;
1021 r = dm_table_build_index(t);
1022 if (r) {
1023 DMERR("unable to build btrees");
1024 return r;
1027 r = dm_table_prealloc_integrity(t, t->md);
1028 if (r) {
1029 DMERR("could not register integrity profile.");
1030 return r;
1033 r = dm_table_alloc_md_mempools(t);
1034 if (r)
1035 DMERR("unable to allocate mempools");
1037 return r;
1040 static DEFINE_MUTEX(_event_lock);
1041 void dm_table_event_callback(struct dm_table *t,
1042 void (*fn)(void *), void *context)
1044 mutex_lock(&_event_lock);
1045 t->event_fn = fn;
1046 t->event_context = context;
1047 mutex_unlock(&_event_lock);
1050 void dm_table_event(struct dm_table *t)
1053 * You can no longer call dm_table_event() from interrupt
1054 * context, use a bottom half instead.
1056 BUG_ON(in_interrupt());
1058 mutex_lock(&_event_lock);
1059 if (t->event_fn)
1060 t->event_fn(t->event_context);
1061 mutex_unlock(&_event_lock);
1064 sector_t dm_table_get_size(struct dm_table *t)
1066 return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1069 struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1071 if (index >= t->num_targets)
1072 return NULL;
1074 return t->targets + index;
1078 * Search the btree for the correct target.
1080 * Caller should check returned pointer with dm_target_is_valid()
1081 * to trap I/O beyond end of device.
1083 struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1085 unsigned int l, n = 0, k = 0;
1086 sector_t *node;
1088 for (l = 0; l < t->depth; l++) {
1089 n = get_child(n, k);
1090 node = get_node(t, l, n);
1092 for (k = 0; k < KEYS_PER_NODE; k++)
1093 if (node[k] >= sector)
1094 break;
1097 return &t->targets[(KEYS_PER_NODE * n) + k];
1101 * Establish the new table's queue_limits and validate them.
1103 int dm_calculate_queue_limits(struct dm_table *table,
1104 struct queue_limits *limits)
1106 struct dm_target *uninitialized_var(ti);
1107 struct queue_limits ti_limits;
1108 unsigned i = 0;
1110 blk_set_default_limits(limits);
1112 while (i < dm_table_get_num_targets(table)) {
1113 blk_set_default_limits(&ti_limits);
1115 ti = dm_table_get_target(table, i++);
1117 if (!ti->type->iterate_devices)
1118 goto combine_limits;
1121 * Combine queue limits of all the devices this target uses.
1123 ti->type->iterate_devices(ti, dm_set_device_limits,
1124 &ti_limits);
1126 /* Set I/O hints portion of queue limits */
1127 if (ti->type->io_hints)
1128 ti->type->io_hints(ti, &ti_limits);
1131 * Check each device area is consistent with the target's
1132 * overall queue limits.
1134 if (ti->type->iterate_devices(ti, device_area_is_invalid,
1135 &ti_limits))
1136 return -EINVAL;
1138 combine_limits:
1140 * Merge this target's queue limits into the overall limits
1141 * for the table.
1143 if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1144 DMWARN("%s: adding target device "
1145 "(start sect %llu len %llu) "
1146 "caused an alignment inconsistency",
1147 dm_device_name(table->md),
1148 (unsigned long long) ti->begin,
1149 (unsigned long long) ti->len);
1152 return validate_hardware_logical_block_alignment(table, limits);
1156 * Set the integrity profile for this device if all devices used have
1157 * matching profiles. We're quite deep in the resume path but still
1158 * don't know if all devices (particularly DM devices this device
1159 * may be stacked on) have matching profiles. Even if the profiles
1160 * don't match we have no way to fail (to resume) at this point.
1162 static void dm_table_set_integrity(struct dm_table *t)
1164 struct gendisk *template_disk = NULL;
1166 if (!blk_get_integrity(dm_disk(t->md)))
1167 return;
1169 template_disk = dm_table_get_integrity_disk(t, true);
1170 if (!template_disk &&
1171 blk_integrity_is_initialized(dm_disk(t->md))) {
1172 DMWARN("%s: device no longer has a valid integrity profile",
1173 dm_device_name(t->md));
1174 return;
1176 blk_integrity_register(dm_disk(t->md),
1177 blk_get_integrity(template_disk));
1180 void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1181 struct queue_limits *limits)
1184 * Copy table's limits to the DM device's request_queue
1186 q->limits = *limits;
1188 if (!dm_table_supports_discards(t))
1189 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1190 else
1191 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1193 dm_table_set_integrity(t);
1196 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1197 * visible to other CPUs because, once the flag is set, incoming bios
1198 * are processed by request-based dm, which refers to the queue
1199 * settings.
1200 * Until the flag set, bios are passed to bio-based dm and queued to
1201 * md->deferred where queue settings are not needed yet.
1202 * Those bios are passed to request-based dm at the resume time.
1204 smp_mb();
1205 if (dm_table_request_based(t))
1206 queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1209 unsigned int dm_table_get_num_targets(struct dm_table *t)
1211 return t->num_targets;
1214 struct list_head *dm_table_get_devices(struct dm_table *t)
1216 return &t->devices;
1219 fmode_t dm_table_get_mode(struct dm_table *t)
1221 return t->mode;
1224 static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1226 int i = t->num_targets;
1227 struct dm_target *ti = t->targets;
1229 while (i--) {
1230 if (postsuspend) {
1231 if (ti->type->postsuspend)
1232 ti->type->postsuspend(ti);
1233 } else if (ti->type->presuspend)
1234 ti->type->presuspend(ti);
1236 ti++;
1240 void dm_table_presuspend_targets(struct dm_table *t)
1242 if (!t)
1243 return;
1245 suspend_targets(t, 0);
1248 void dm_table_postsuspend_targets(struct dm_table *t)
1250 if (!t)
1251 return;
1253 suspend_targets(t, 1);
1256 int dm_table_resume_targets(struct dm_table *t)
1258 int i, r = 0;
1260 for (i = 0; i < t->num_targets; i++) {
1261 struct dm_target *ti = t->targets + i;
1263 if (!ti->type->preresume)
1264 continue;
1266 r = ti->type->preresume(ti);
1267 if (r)
1268 return r;
1271 for (i = 0; i < t->num_targets; i++) {
1272 struct dm_target *ti = t->targets + i;
1274 if (ti->type->resume)
1275 ti->type->resume(ti);
1278 return 0;
1281 void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1283 list_add(&cb->list, &t->target_callbacks);
1285 EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1287 int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1289 struct dm_dev_internal *dd;
1290 struct list_head *devices = dm_table_get_devices(t);
1291 struct dm_target_callbacks *cb;
1292 int r = 0;
1294 list_for_each_entry(dd, devices, list) {
1295 struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1296 char b[BDEVNAME_SIZE];
1298 if (likely(q))
1299 r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1300 else
1301 DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1302 dm_device_name(t->md),
1303 bdevname(dd->dm_dev.bdev, b));
1306 list_for_each_entry(cb, &t->target_callbacks, list)
1307 if (cb->congested_fn)
1308 r |= cb->congested_fn(cb, bdi_bits);
1310 return r;
1313 int dm_table_any_busy_target(struct dm_table *t)
1315 unsigned i;
1316 struct dm_target *ti;
1318 for (i = 0; i < t->num_targets; i++) {
1319 ti = t->targets + i;
1320 if (ti->type->busy && ti->type->busy(ti))
1321 return 1;
1324 return 0;
1327 struct mapped_device *dm_table_get_md(struct dm_table *t)
1329 return t->md;
1332 static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1333 sector_t start, sector_t len, void *data)
1335 struct request_queue *q = bdev_get_queue(dev->bdev);
1337 return q && blk_queue_discard(q);
1340 bool dm_table_supports_discards(struct dm_table *t)
1342 struct dm_target *ti;
1343 unsigned i = 0;
1345 if (!t->discards_supported)
1346 return 0;
1349 * Ensure that at least one underlying device supports discards.
1350 * t->devices includes internal dm devices such as mirror logs
1351 * so we need to use iterate_devices here, which targets
1352 * supporting discard must provide.
1354 while (i < dm_table_get_num_targets(t)) {
1355 ti = dm_table_get_target(t, i++);
1357 if (ti->type->iterate_devices &&
1358 ti->type->iterate_devices(ti, device_discard_capable, NULL))
1359 return 1;
1362 return 0;
1365 EXPORT_SYMBOL(dm_vcalloc);
1366 EXPORT_SYMBOL(dm_get_device);
1367 EXPORT_SYMBOL(dm_put_device);
1368 EXPORT_SYMBOL(dm_table_event);
1369 EXPORT_SYMBOL(dm_table_get_size);
1370 EXPORT_SYMBOL(dm_table_get_mode);
1371 EXPORT_SYMBOL(dm_table_get_md);
1372 EXPORT_SYMBOL(dm_table_put);
1373 EXPORT_SYMBOL(dm_table_get);