2 * Copyright (C) 2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
22 #include <sys/types.h>
30 #include "kerncompat.h"
34 #include "transaction.h"
37 #include "extent_io.h"
39 #define HEADER_MAGIC 0xbd5c25e27295668bULL
40 #define MAX_PENDING_SIZE (256 * 1024)
41 #define BLOCK_SIZE 1024
42 #define BLOCK_MASK (BLOCK_SIZE - 1)
44 #define COMPRESS_NONE 0
45 #define COMPRESS_ZLIB 1
47 #define MAX_WORKER_THREADS (32)
49 struct meta_cluster_item {
52 } __attribute__ ((__packed__));
54 struct meta_cluster_header {
59 } __attribute__ ((__packed__));
61 /* cluster header + index items + buffers */
63 struct meta_cluster_header header;
64 struct meta_cluster_item items[];
65 } __attribute__ ((__packed__));
67 #define ITEMS_PER_CLUSTER ((BLOCK_SIZE - sizeof(struct meta_cluster)) / \
68 sizeof(struct meta_cluster_item))
74 * physical_dup only store additonal physical for BTRFS_BLOCK_GROUP_DUP
75 * currently restore only support single and DUP
76 * TODO: modify this structure and the function related to this
77 * structure for support RAID*
83 struct list_head list;
87 struct list_head list;
88 struct list_head ordered;
96 struct metadump_struct {
97 struct btrfs_root *root;
101 struct meta_cluster cluster;
102 char meta_cluster_bytes[BLOCK_SIZE];
105 pthread_t threads[MAX_WORKER_THREADS];
107 pthread_mutex_t mutex;
109 struct rb_root name_tree;
111 struct list_head list;
112 struct list_head ordered;
134 struct mdrestore_struct {
138 pthread_t threads[MAX_WORKER_THREADS];
140 pthread_mutex_t mutex;
143 struct rb_root chunk_tree;
144 struct rb_root physical_tree;
145 struct list_head list;
146 struct list_head overlapping_chunks;
151 u64 last_physical_offset;
152 u8 uuid[BTRFS_UUID_SIZE];
153 u8 fsid[BTRFS_FSID_SIZE];
161 int clear_space_cache;
162 struct btrfs_fs_info *info;
165 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
166 u64 search, u64 cluster_bytenr);
167 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size);
169 static void csum_block(u8 *buf, size_t len)
171 u8 result[BTRFS_CRC32_SIZE];
173 crc = crc32c(crc, buf + BTRFS_CSUM_SIZE, len - BTRFS_CSUM_SIZE);
174 btrfs_csum_final(crc, result);
175 memcpy(buf, result, BTRFS_CRC32_SIZE);
178 static int has_name(struct btrfs_key *key)
181 case BTRFS_DIR_ITEM_KEY:
182 case BTRFS_DIR_INDEX_KEY:
183 case BTRFS_INODE_REF_KEY:
184 case BTRFS_INODE_EXTREF_KEY:
185 case BTRFS_XATTR_ITEM_KEY:
194 static char *generate_garbage(u32 name_len)
196 char *buf = malloc(name_len);
202 for (i = 0; i < name_len; i++) {
203 char c = rand_range(94) + 33;
213 static int name_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
215 struct name *entry = rb_entry(a, struct name, n);
216 struct name *ins = rb_entry(b, struct name, n);
219 len = min(ins->len, entry->len);
220 return memcmp(ins->val, entry->val, len);
223 static int chunk_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
225 struct fs_chunk *entry = rb_entry(a, struct fs_chunk, l);
226 struct fs_chunk *ins = rb_entry(b, struct fs_chunk, l);
228 if (fuzz && ins->logical >= entry->logical &&
229 ins->logical < entry->logical + entry->bytes)
232 if (ins->logical < entry->logical)
234 else if (ins->logical > entry->logical)
239 static int physical_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
241 struct fs_chunk *entry = rb_entry(a, struct fs_chunk, p);
242 struct fs_chunk *ins = rb_entry(b, struct fs_chunk, p);
244 if (fuzz && ins->physical >= entry->physical &&
245 ins->physical < entry->physical + entry->bytes)
248 if (fuzz && entry->physical >= ins->physical &&
249 entry->physical < ins->physical + ins->bytes)
252 if (ins->physical < entry->physical)
254 else if (ins->physical > entry->physical)
259 static void tree_insert(struct rb_root *root, struct rb_node *ins,
260 int (*cmp)(struct rb_node *a, struct rb_node *b,
263 struct rb_node ** p = &root->rb_node;
264 struct rb_node * parent = NULL;
270 dir = cmp(*p, ins, 1);
279 rb_link_node(ins, parent, p);
280 rb_insert_color(ins, root);
283 static struct rb_node *tree_search(struct rb_root *root,
284 struct rb_node *search,
285 int (*cmp)(struct rb_node *a,
286 struct rb_node *b, int fuzz),
289 struct rb_node *n = root->rb_node;
293 dir = cmp(n, search, fuzz);
305 static u64 logical_to_physical(struct mdrestore_struct *mdres, u64 logical,
306 u64 *size, u64 *physical_dup)
308 struct fs_chunk *fs_chunk;
309 struct rb_node *entry;
310 struct fs_chunk search;
313 if (logical == BTRFS_SUPER_INFO_OFFSET)
316 search.logical = logical;
317 entry = tree_search(&mdres->chunk_tree, &search.l, chunk_cmp, 1);
319 if (mdres->in != stdin)
320 warning("cannot find a chunk, using logical");
323 fs_chunk = rb_entry(entry, struct fs_chunk, l);
324 if (fs_chunk->logical > logical || fs_chunk->logical + fs_chunk->bytes < logical)
326 offset = search.logical - fs_chunk->logical;
329 /* Only in dup case, physical_dup is not equal to 0 */
330 if (fs_chunk->physical_dup)
331 *physical_dup = fs_chunk->physical_dup + offset;
336 *size = min(*size, fs_chunk->bytes + fs_chunk->logical - logical);
337 return fs_chunk->physical + offset;
341 static char *find_collision(struct metadump_struct *md, char *name,
345 struct rb_node *entry;
347 unsigned long checksum;
353 entry = tree_search(&md->name_tree, &tmp.n, name_cmp, 0);
355 val = rb_entry(entry, struct name, n);
360 val = malloc(sizeof(struct name));
362 error("cannot sanitize name, not enough memory");
367 memset(val, 0, sizeof(*val));
371 val->sub = malloc(name_len);
373 error("cannot sanitize name, not enough memory");
379 checksum = crc32c(~1, val->val, name_len);
380 memset(val->sub, ' ', name_len);
383 if (crc32c(~1, val->sub, name_len) == checksum &&
384 memcmp(val->sub, val->val, val->len)) {
389 if (val->sub[i] == 127) {
394 } while (val->sub[i] == 127);
399 if (val->sub[i] == '/')
401 memset(val->sub, ' ', i);
406 if (val->sub[i] == '/')
413 "cannot find a hash collision for '%.*s', generating garbage, it won't match indexes",
415 for (i = 0; i < name_len; i++) {
416 char c = rand_range(94) + 33;
424 tree_insert(&md->name_tree, &val->n, name_cmp);
428 static void sanitize_dir_item(struct metadump_struct *md, struct extent_buffer *eb,
431 struct btrfs_dir_item *dir_item;
434 unsigned long name_ptr;
439 int free_garbage = (md->sanitize_names == 1);
441 dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
442 total_len = btrfs_item_size_nr(eb, slot);
443 while (cur < total_len) {
444 this_len = sizeof(*dir_item) +
445 btrfs_dir_name_len(eb, dir_item) +
446 btrfs_dir_data_len(eb, dir_item);
447 name_ptr = (unsigned long)(dir_item + 1);
448 name_len = btrfs_dir_name_len(eb, dir_item);
450 if (md->sanitize_names > 1) {
451 buf = malloc(name_len);
453 error("cannot sanitize name, not enough memory");
456 read_extent_buffer(eb, buf, name_ptr, name_len);
457 garbage = find_collision(md, buf, name_len);
459 garbage = generate_garbage(name_len);
462 error("cannot sanitize name, not enough memory");
465 write_extent_buffer(eb, garbage, name_ptr, name_len);
467 dir_item = (struct btrfs_dir_item *)((char *)dir_item +
474 static void sanitize_inode_ref(struct metadump_struct *md,
475 struct extent_buffer *eb, int slot, int ext)
477 struct btrfs_inode_extref *extref;
478 struct btrfs_inode_ref *ref;
481 unsigned long name_ptr;
485 int free_garbage = (md->sanitize_names == 1);
487 item_size = btrfs_item_size_nr(eb, slot);
488 ptr = btrfs_item_ptr_offset(eb, slot);
489 while (cur_offset < item_size) {
491 extref = (struct btrfs_inode_extref *)(ptr +
493 name_ptr = (unsigned long)(&extref->name);
494 len = btrfs_inode_extref_name_len(eb, extref);
495 cur_offset += sizeof(*extref);
497 ref = (struct btrfs_inode_ref *)(ptr + cur_offset);
498 len = btrfs_inode_ref_name_len(eb, ref);
499 name_ptr = (unsigned long)(ref + 1);
500 cur_offset += sizeof(*ref);
504 if (md->sanitize_names > 1) {
507 error("cannot sanitize name, not enough memory");
510 read_extent_buffer(eb, buf, name_ptr, len);
511 garbage = find_collision(md, buf, len);
513 garbage = generate_garbage(len);
517 error("cannot sanitize name, not enough memory");
520 write_extent_buffer(eb, garbage, name_ptr, len);
526 static void sanitize_xattr(struct metadump_struct *md,
527 struct extent_buffer *eb, int slot)
529 struct btrfs_dir_item *dir_item;
530 unsigned long data_ptr;
533 dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
534 data_len = btrfs_dir_data_len(eb, dir_item);
536 data_ptr = (unsigned long)((char *)(dir_item + 1) +
537 btrfs_dir_name_len(eb, dir_item));
538 memset_extent_buffer(eb, 0, data_ptr, data_len);
541 static void sanitize_name(struct metadump_struct *md, u8 *dst,
542 struct extent_buffer *src, struct btrfs_key *key,
545 struct extent_buffer *eb;
547 eb = alloc_dummy_eb(src->start, src->len);
549 error("cannot sanitize name, not enough memory");
553 memcpy(eb->data, dst, eb->len);
556 case BTRFS_DIR_ITEM_KEY:
557 case BTRFS_DIR_INDEX_KEY:
558 sanitize_dir_item(md, eb, slot);
560 case BTRFS_INODE_REF_KEY:
561 sanitize_inode_ref(md, eb, slot, 0);
563 case BTRFS_INODE_EXTREF_KEY:
564 sanitize_inode_ref(md, eb, slot, 1);
566 case BTRFS_XATTR_ITEM_KEY:
567 sanitize_xattr(md, eb, slot);
573 memcpy(dst, eb->data, eb->len);
578 * zero inline extents and csum items
580 static void zero_items(struct metadump_struct *md, u8 *dst,
581 struct extent_buffer *src)
583 struct btrfs_file_extent_item *fi;
584 struct btrfs_item *item;
585 struct btrfs_key key;
586 u32 nritems = btrfs_header_nritems(src);
591 for (i = 0; i < nritems; i++) {
592 item = btrfs_item_nr(i);
593 btrfs_item_key_to_cpu(src, &key, i);
594 if (key.type == BTRFS_CSUM_ITEM_KEY) {
595 size = btrfs_item_size_nr(src, i);
596 memset(dst + btrfs_leaf_data(src) +
597 btrfs_item_offset_nr(src, i), 0, size);
601 if (md->sanitize_names && has_name(&key)) {
602 sanitize_name(md, dst, src, &key, i);
606 if (key.type != BTRFS_EXTENT_DATA_KEY)
609 fi = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
610 extent_type = btrfs_file_extent_type(src, fi);
611 if (extent_type != BTRFS_FILE_EXTENT_INLINE)
614 ptr = btrfs_file_extent_inline_start(fi);
615 size = btrfs_file_extent_inline_item_len(src, item);
616 memset(dst + ptr, 0, size);
621 * copy buffer and zero useless data in the buffer
623 static void copy_buffer(struct metadump_struct *md, u8 *dst,
624 struct extent_buffer *src)
630 memcpy(dst, src->data, src->len);
631 if (src->start == BTRFS_SUPER_INFO_OFFSET)
634 level = btrfs_header_level(src);
635 nritems = btrfs_header_nritems(src);
638 size = sizeof(struct btrfs_header);
639 memset(dst + size, 0, src->len - size);
640 } else if (level == 0) {
641 size = btrfs_leaf_data(src) +
642 btrfs_item_offset_nr(src, nritems - 1) -
643 btrfs_item_nr_offset(nritems);
644 memset(dst + btrfs_item_nr_offset(nritems), 0, size);
645 zero_items(md, dst, src);
647 size = offsetof(struct btrfs_node, ptrs) +
648 sizeof(struct btrfs_key_ptr) * nritems;
649 memset(dst + size, 0, src->len - size);
651 csum_block(dst, src->len);
654 static void *dump_worker(void *data)
656 struct metadump_struct *md = (struct metadump_struct *)data;
657 struct async_work *async;
661 pthread_mutex_lock(&md->mutex);
662 while (list_empty(&md->list)) {
664 pthread_mutex_unlock(&md->mutex);
667 pthread_cond_wait(&md->cond, &md->mutex);
669 async = list_entry(md->list.next, struct async_work, list);
670 list_del_init(&async->list);
671 pthread_mutex_unlock(&md->mutex);
673 if (md->compress_level > 0) {
674 u8 *orig = async->buffer;
676 async->bufsize = compressBound(async->size);
677 async->buffer = malloc(async->bufsize);
678 if (!async->buffer) {
679 error("not enough memory for async buffer");
680 pthread_mutex_lock(&md->mutex);
683 pthread_mutex_unlock(&md->mutex);
687 ret = compress2(async->buffer,
688 (unsigned long *)&async->bufsize,
689 orig, async->size, md->compress_level);
697 pthread_mutex_lock(&md->mutex);
699 pthread_mutex_unlock(&md->mutex);
705 static void meta_cluster_init(struct metadump_struct *md, u64 start)
707 struct meta_cluster_header *header;
711 header = &md->cluster.header;
712 header->magic = cpu_to_le64(HEADER_MAGIC);
713 header->bytenr = cpu_to_le64(start);
714 header->nritems = cpu_to_le32(0);
715 header->compress = md->compress_level > 0 ?
716 COMPRESS_ZLIB : COMPRESS_NONE;
719 static void metadump_destroy(struct metadump_struct *md, int num_threads)
724 pthread_mutex_lock(&md->mutex);
726 pthread_cond_broadcast(&md->cond);
727 pthread_mutex_unlock(&md->mutex);
729 for (i = 0; i < num_threads; i++)
730 pthread_join(md->threads[i], NULL);
732 pthread_cond_destroy(&md->cond);
733 pthread_mutex_destroy(&md->mutex);
735 while ((n = rb_first(&md->name_tree))) {
738 name = rb_entry(n, struct name, n);
739 rb_erase(n, &md->name_tree);
746 static int metadump_init(struct metadump_struct *md, struct btrfs_root *root,
747 FILE *out, int num_threads, int compress_level,
752 memset(md, 0, sizeof(*md));
753 INIT_LIST_HEAD(&md->list);
754 INIT_LIST_HEAD(&md->ordered);
757 md->pending_start = (u64)-1;
758 md->compress_level = compress_level;
759 md->sanitize_names = sanitize_names;
760 if (sanitize_names > 1)
761 crc32c_optimization_init();
763 md->name_tree.rb_node = NULL;
764 md->num_threads = num_threads;
765 pthread_cond_init(&md->cond, NULL);
766 pthread_mutex_init(&md->mutex, NULL);
767 meta_cluster_init(md, 0);
772 for (i = 0; i < num_threads; i++) {
773 ret = pthread_create(md->threads + i, NULL, dump_worker, md);
779 metadump_destroy(md, i + 1);
784 static int write_zero(FILE *out, size_t size)
786 static char zero[BLOCK_SIZE];
787 return fwrite(zero, size, 1, out);
790 static int write_buffers(struct metadump_struct *md, u64 *next)
792 struct meta_cluster_header *header = &md->cluster.header;
793 struct meta_cluster_item *item;
794 struct async_work *async;
800 if (list_empty(&md->ordered))
803 /* wait until all buffers are compressed */
804 while (!err && md->num_items > md->num_ready) {
805 struct timespec ts = {
809 pthread_mutex_unlock(&md->mutex);
810 nanosleep(&ts, NULL);
811 pthread_mutex_lock(&md->mutex);
816 error("one of the threads failed: %s", strerror(-err));
820 /* setup and write index block */
821 list_for_each_entry(async, &md->ordered, ordered) {
822 item = &md->cluster.items[nritems];
823 item->bytenr = cpu_to_le64(async->start);
824 item->size = cpu_to_le32(async->bufsize);
827 header->nritems = cpu_to_le32(nritems);
829 ret = fwrite(&md->cluster, BLOCK_SIZE, 1, md->out);
831 error("unable to write out cluster: %s", strerror(errno));
836 bytenr += le64_to_cpu(header->bytenr) + BLOCK_SIZE;
837 while (!list_empty(&md->ordered)) {
838 async = list_entry(md->ordered.next, struct async_work,
840 list_del_init(&async->ordered);
842 bytenr += async->bufsize;
844 ret = fwrite(async->buffer, async->bufsize, 1,
847 error("unable to write out cluster: %s",
857 /* zero unused space in the last block */
858 if (!err && bytenr & BLOCK_MASK) {
859 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
862 ret = write_zero(md->out, size);
864 error("unable to zero out buffer: %s",
874 static int read_data_extent(struct metadump_struct *md,
875 struct async_work *async)
877 struct btrfs_root *root = md->root;
878 u64 bytes_left = async->size;
879 u64 logical = async->start;
886 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, logical,
889 /* Try our best to read data, just like read_tree_block() */
890 for (cur_mirror = 0; cur_mirror < num_copies; cur_mirror++) {
892 read_len = bytes_left;
893 ret = read_extent_data(root,
894 (char *)(async->buffer + offset),
895 logical, &read_len, cur_mirror);
900 bytes_left -= read_len;
908 static int get_dev_fd(struct btrfs_root *root)
910 struct btrfs_device *dev;
912 dev = list_first_entry(&root->fs_info->fs_devices->devices,
913 struct btrfs_device, dev_list);
917 static int flush_pending(struct metadump_struct *md, int done)
919 struct async_work *async = NULL;
920 struct extent_buffer *eb;
921 u64 blocksize = md->root->nodesize;
927 if (md->pending_size) {
928 async = calloc(1, sizeof(*async));
932 async->start = md->pending_start;
933 async->size = md->pending_size;
934 async->bufsize = async->size;
935 async->buffer = malloc(async->bufsize);
936 if (!async->buffer) {
941 start = async->start;
945 ret = read_data_extent(md, async);
954 * Balance can make the mapping not cover the super block, so
955 * just copy directly from one of the devices.
957 if (start == BTRFS_SUPER_INFO_OFFSET) {
958 int fd = get_dev_fd(md->root);
960 ret = pread64(fd, async->buffer, size, start);
964 error("unable to read superblock at %llu: %s",
965 (unsigned long long)start,
973 while (!md->data && size > 0) {
974 u64 this_read = min(blocksize, size);
975 eb = read_tree_block(md->root, start, this_read, 0);
976 if (!extent_buffer_uptodate(eb)) {
979 error("unable to read metadata block %llu",
980 (unsigned long long)start);
983 copy_buffer(md, async->buffer + offset, eb);
984 free_extent_buffer(eb);
990 md->pending_start = (u64)-1;
991 md->pending_size = 0;
996 pthread_mutex_lock(&md->mutex);
998 list_add_tail(&async->ordered, &md->ordered);
1000 if (md->compress_level > 0) {
1001 list_add_tail(&async->list, &md->list);
1002 pthread_cond_signal(&md->cond);
1007 if (md->num_items >= ITEMS_PER_CLUSTER || done) {
1008 ret = write_buffers(md, &start);
1010 error("unable to write buffers: %s", strerror(-ret));
1012 meta_cluster_init(md, start);
1014 pthread_mutex_unlock(&md->mutex);
1018 static int add_extent(u64 start, u64 size, struct metadump_struct *md,
1022 if (md->data != data ||
1023 md->pending_size + size > MAX_PENDING_SIZE ||
1024 md->pending_start + md->pending_size != start) {
1025 ret = flush_pending(md, 0);
1028 md->pending_start = start;
1030 readahead_tree_block(md->root, start, size, 0);
1031 md->pending_size += size;
1036 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1037 static int is_tree_block(struct btrfs_root *extent_root,
1038 struct btrfs_path *path, u64 bytenr)
1040 struct extent_buffer *leaf;
1041 struct btrfs_key key;
1045 leaf = path->nodes[0];
1047 struct btrfs_extent_ref_v0 *ref_item;
1049 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1050 ret = btrfs_next_leaf(extent_root, path);
1055 leaf = path->nodes[0];
1057 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1058 if (key.objectid != bytenr)
1060 if (key.type != BTRFS_EXTENT_REF_V0_KEY)
1062 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1063 struct btrfs_extent_ref_v0);
1064 ref_objectid = btrfs_ref_objectid_v0(leaf, ref_item);
1065 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID)
1073 static int copy_tree_blocks(struct btrfs_root *root, struct extent_buffer *eb,
1074 struct metadump_struct *metadump, int root_tree)
1076 struct extent_buffer *tmp;
1077 struct btrfs_root_item *ri;
1078 struct btrfs_key key;
1085 ret = add_extent(btrfs_header_bytenr(eb), root->nodesize, metadump, 0);
1087 error("unable to add metadata block %llu: %d",
1088 btrfs_header_bytenr(eb), ret);
1092 if (btrfs_header_level(eb) == 0 && !root_tree)
1095 level = btrfs_header_level(eb);
1096 nritems = btrfs_header_nritems(eb);
1097 for (i = 0; i < nritems; i++) {
1099 btrfs_item_key_to_cpu(eb, &key, i);
1100 if (key.type != BTRFS_ROOT_ITEM_KEY)
1102 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
1103 bytenr = btrfs_disk_root_bytenr(eb, ri);
1104 tmp = read_tree_block(root, bytenr, root->nodesize, 0);
1105 if (!extent_buffer_uptodate(tmp)) {
1106 error("unable to read log root block");
1109 ret = copy_tree_blocks(root, tmp, metadump, 0);
1110 free_extent_buffer(tmp);
1114 bytenr = btrfs_node_blockptr(eb, i);
1115 tmp = read_tree_block(root, bytenr, root->nodesize, 0);
1116 if (!extent_buffer_uptodate(tmp)) {
1117 error("unable to read log root block");
1120 ret = copy_tree_blocks(root, tmp, metadump, root_tree);
1121 free_extent_buffer(tmp);
1130 static int copy_log_trees(struct btrfs_root *root,
1131 struct metadump_struct *metadump,
1132 struct btrfs_path *path)
1134 u64 blocknr = btrfs_super_log_root(root->fs_info->super_copy);
1139 if (!root->fs_info->log_root_tree ||
1140 !root->fs_info->log_root_tree->node) {
1141 error("unable to copy tree log, it has not been setup");
1145 return copy_tree_blocks(root, root->fs_info->log_root_tree->node,
1149 static int copy_space_cache(struct btrfs_root *root,
1150 struct metadump_struct *metadump,
1151 struct btrfs_path *path)
1153 struct extent_buffer *leaf;
1154 struct btrfs_file_extent_item *fi;
1155 struct btrfs_key key;
1156 u64 bytenr, num_bytes;
1159 root = root->fs_info->tree_root;
1162 key.type = BTRFS_EXTENT_DATA_KEY;
1165 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1167 error("free space inode not found: %d", ret);
1171 leaf = path->nodes[0];
1174 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1175 ret = btrfs_next_leaf(root, path);
1177 error("cannot go to next leaf %d", ret);
1182 leaf = path->nodes[0];
1185 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1186 if (key.type != BTRFS_EXTENT_DATA_KEY) {
1191 fi = btrfs_item_ptr(leaf, path->slots[0],
1192 struct btrfs_file_extent_item);
1193 if (btrfs_file_extent_type(leaf, fi) !=
1194 BTRFS_FILE_EXTENT_REG) {
1199 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1200 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1201 ret = add_extent(bytenr, num_bytes, metadump, 1);
1203 error("unable to add space cache blocks %d", ret);
1204 btrfs_release_path(path);
1213 static int copy_from_extent_tree(struct metadump_struct *metadump,
1214 struct btrfs_path *path)
1216 struct btrfs_root *extent_root;
1217 struct extent_buffer *leaf;
1218 struct btrfs_extent_item *ei;
1219 struct btrfs_key key;
1224 extent_root = metadump->root->fs_info->extent_root;
1225 bytenr = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
1226 key.objectid = bytenr;
1227 key.type = BTRFS_EXTENT_ITEM_KEY;
1230 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1232 error("extent root not found: %d", ret);
1237 leaf = path->nodes[0];
1240 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1241 ret = btrfs_next_leaf(extent_root, path);
1243 error("cannot go to next leaf %d", ret);
1250 leaf = path->nodes[0];
1253 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1254 if (key.objectid < bytenr ||
1255 (key.type != BTRFS_EXTENT_ITEM_KEY &&
1256 key.type != BTRFS_METADATA_ITEM_KEY)) {
1261 bytenr = key.objectid;
1262 if (key.type == BTRFS_METADATA_ITEM_KEY) {
1263 num_bytes = extent_root->nodesize;
1265 num_bytes = key.offset;
1268 if (num_bytes == 0) {
1269 error("extent length 0 at bytenr %llu key type %d",
1270 (unsigned long long)bytenr, key.type);
1275 if (btrfs_item_size_nr(leaf, path->slots[0]) > sizeof(*ei)) {
1276 ei = btrfs_item_ptr(leaf, path->slots[0],
1277 struct btrfs_extent_item);
1278 if (btrfs_extent_flags(leaf, ei) &
1279 BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1280 ret = add_extent(bytenr, num_bytes, metadump,
1283 error("unable to add block %llu: %d",
1284 (unsigned long long)bytenr, ret);
1289 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1290 ret = is_tree_block(extent_root, path, bytenr);
1292 error("failed to check tree block %llu: %d",
1293 (unsigned long long)bytenr, ret);
1298 ret = add_extent(bytenr, num_bytes, metadump,
1301 error("unable to add block %llu: %d",
1302 (unsigned long long)bytenr, ret);
1309 "either extent tree is corrupted or you haven't built with V0 support");
1314 bytenr += num_bytes;
1317 btrfs_release_path(path);
1322 static int create_metadump(const char *input, FILE *out, int num_threads,
1323 int compress_level, int sanitize, int walk_trees)
1325 struct btrfs_root *root;
1326 struct btrfs_path path;
1327 struct metadump_struct metadump;
1331 root = open_ctree(input, 0, 0);
1333 error("open ctree failed");
1337 ret = metadump_init(&metadump, root, out, num_threads,
1338 compress_level, sanitize);
1340 error("failed to initialize metadump: %d", ret);
1345 ret = add_extent(BTRFS_SUPER_INFO_OFFSET, BTRFS_SUPER_INFO_SIZE,
1348 error("unable to add metadata: %d", ret);
1353 btrfs_init_path(&path);
1356 ret = copy_tree_blocks(root, root->fs_info->chunk_root->node,
1363 ret = copy_tree_blocks(root, root->fs_info->tree_root->node,
1370 ret = copy_from_extent_tree(&metadump, &path);
1377 ret = copy_log_trees(root, &metadump, &path);
1383 ret = copy_space_cache(root, &metadump, &path);
1385 ret = flush_pending(&metadump, 1);
1389 error("failed to flush pending data: %d", ret);
1392 metadump_destroy(&metadump, num_threads);
1394 btrfs_release_path(&path);
1395 ret = close_ctree(root);
1396 return err ? err : ret;
1399 static void update_super_old(u8 *buffer)
1401 struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1402 struct btrfs_chunk *chunk;
1403 struct btrfs_disk_key *key;
1404 u32 sectorsize = btrfs_super_sectorsize(super);
1405 u64 flags = btrfs_super_flags(super);
1407 flags |= BTRFS_SUPER_FLAG_METADUMP;
1408 btrfs_set_super_flags(super, flags);
1410 key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1411 chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1412 sizeof(struct btrfs_disk_key));
1414 btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1415 btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1416 btrfs_set_disk_key_offset(key, 0);
1418 btrfs_set_stack_chunk_length(chunk, (u64)-1);
1419 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1420 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1421 btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1422 btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1423 btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1424 btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1425 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1426 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1427 chunk->stripe.devid = super->dev_item.devid;
1428 btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1429 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1430 btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1431 csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1434 static int update_super(struct mdrestore_struct *mdres, u8 *buffer)
1436 struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1437 struct btrfs_chunk *chunk;
1438 struct btrfs_disk_key *disk_key;
1439 struct btrfs_key key;
1440 u64 flags = btrfs_super_flags(super);
1441 u32 new_array_size = 0;
1444 u8 *ptr, *write_ptr;
1445 int old_num_stripes;
1447 write_ptr = ptr = super->sys_chunk_array;
1448 array_size = btrfs_super_sys_array_size(super);
1450 while (cur < array_size) {
1451 disk_key = (struct btrfs_disk_key *)ptr;
1452 btrfs_disk_key_to_cpu(&key, disk_key);
1454 new_array_size += sizeof(*disk_key);
1455 memmove(write_ptr, ptr, sizeof(*disk_key));
1457 write_ptr += sizeof(*disk_key);
1458 ptr += sizeof(*disk_key);
1459 cur += sizeof(*disk_key);
1461 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1462 u64 type, physical, physical_dup, size = 0;
1464 chunk = (struct btrfs_chunk *)ptr;
1465 old_num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1466 chunk = (struct btrfs_chunk *)write_ptr;
1468 memmove(write_ptr, ptr, sizeof(*chunk));
1469 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1470 type = btrfs_stack_chunk_type(chunk);
1471 if (type & BTRFS_BLOCK_GROUP_DUP) {
1472 new_array_size += sizeof(struct btrfs_stripe);
1473 write_ptr += sizeof(struct btrfs_stripe);
1475 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1476 btrfs_set_stack_chunk_type(chunk,
1477 BTRFS_BLOCK_GROUP_SYSTEM);
1479 chunk->stripe.devid = super->dev_item.devid;
1480 physical = logical_to_physical(mdres, key.offset,
1481 &size, &physical_dup);
1482 if (size != (u64)-1)
1483 btrfs_set_stack_stripe_offset(&chunk->stripe,
1485 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid,
1487 new_array_size += sizeof(*chunk);
1489 error("bogus key in the sys array %d", key.type);
1492 write_ptr += sizeof(*chunk);
1493 ptr += btrfs_chunk_item_size(old_num_stripes);
1494 cur += btrfs_chunk_item_size(old_num_stripes);
1497 if (mdres->clear_space_cache)
1498 btrfs_set_super_cache_generation(super, 0);
1500 flags |= BTRFS_SUPER_FLAG_METADUMP_V2;
1501 btrfs_set_super_flags(super, flags);
1502 btrfs_set_super_sys_array_size(super, new_array_size);
1503 csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1508 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size)
1510 struct extent_buffer *eb;
1512 eb = calloc(1, sizeof(struct extent_buffer) + size);
1521 static void truncate_item(struct extent_buffer *eb, int slot, u32 new_size)
1523 struct btrfs_item *item;
1531 old_size = btrfs_item_size_nr(eb, slot);
1532 if (old_size == new_size)
1535 nritems = btrfs_header_nritems(eb);
1536 data_end = btrfs_item_offset_nr(eb, nritems - 1);
1538 old_data_start = btrfs_item_offset_nr(eb, slot);
1539 size_diff = old_size - new_size;
1541 for (i = slot; i < nritems; i++) {
1543 item = btrfs_item_nr(i);
1544 ioff = btrfs_item_offset(eb, item);
1545 btrfs_set_item_offset(eb, item, ioff + size_diff);
1548 memmove_extent_buffer(eb, btrfs_leaf_data(eb) + data_end + size_diff,
1549 btrfs_leaf_data(eb) + data_end,
1550 old_data_start + new_size - data_end);
1551 item = btrfs_item_nr(slot);
1552 btrfs_set_item_size(eb, item, new_size);
1555 static int fixup_chunk_tree_block(struct mdrestore_struct *mdres,
1556 struct async_work *async, u8 *buffer,
1559 struct extent_buffer *eb;
1560 size_t size_left = size;
1561 u64 bytenr = async->start;
1564 if (size_left % mdres->nodesize)
1567 eb = alloc_dummy_eb(bytenr, mdres->nodesize);
1573 memcpy(eb->data, buffer, mdres->nodesize);
1575 if (btrfs_header_bytenr(eb) != bytenr)
1577 if (memcmp(mdres->fsid,
1578 eb->data + offsetof(struct btrfs_header, fsid),
1582 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID)
1585 if (btrfs_header_level(eb) != 0)
1588 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1589 struct btrfs_chunk *chunk;
1590 struct btrfs_key key;
1591 u64 type, physical, physical_dup, size = (u64)-1;
1593 btrfs_item_key_to_cpu(eb, &key, i);
1594 if (key.type != BTRFS_CHUNK_ITEM_KEY)
1598 physical = logical_to_physical(mdres, key.offset,
1599 &size, &physical_dup);
1602 truncate_item(eb, i, sizeof(*chunk));
1603 chunk = btrfs_item_ptr(eb, i, struct btrfs_chunk);
1606 /* Zero out the RAID profile */
1607 type = btrfs_chunk_type(eb, chunk);
1608 type &= (BTRFS_BLOCK_GROUP_DATA |
1609 BTRFS_BLOCK_GROUP_SYSTEM |
1610 BTRFS_BLOCK_GROUP_METADATA |
1611 BTRFS_BLOCK_GROUP_DUP);
1612 btrfs_set_chunk_type(eb, chunk, type);
1615 btrfs_set_chunk_num_stripes(eb, chunk, 1);
1616 btrfs_set_chunk_sub_stripes(eb, chunk, 0);
1617 btrfs_set_stripe_devid_nr(eb, chunk, 0, mdres->devid);
1618 if (size != (u64)-1)
1619 btrfs_set_stripe_offset_nr(eb, chunk, 0,
1621 /* update stripe 2 offset */
1623 btrfs_set_stripe_offset_nr(eb, chunk, 1,
1626 write_extent_buffer(eb, mdres->uuid,
1627 (unsigned long)btrfs_stripe_dev_uuid_nr(
1631 memcpy(buffer, eb->data, eb->len);
1632 csum_block(buffer, eb->len);
1634 size_left -= mdres->nodesize;
1635 buffer += mdres->nodesize;
1636 bytenr += mdres->nodesize;
1643 static void write_backup_supers(int fd, u8 *buf)
1645 struct btrfs_super_block *super = (struct btrfs_super_block *)buf;
1652 if (fstat(fd, &st)) {
1654 "cannot stat restore point, won't be able to write backup supers: %s",
1659 size = btrfs_device_size(fd, &st);
1661 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1662 bytenr = btrfs_sb_offset(i);
1663 if (bytenr + BTRFS_SUPER_INFO_SIZE > size)
1665 btrfs_set_super_bytenr(super, bytenr);
1666 csum_block(buf, BTRFS_SUPER_INFO_SIZE);
1667 ret = pwrite64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1668 if (ret < BTRFS_SUPER_INFO_SIZE) {
1671 "problem writing out backup super block %d: %s",
1672 i, strerror(errno));
1674 error("short write writing out backup super block");
1680 static void *restore_worker(void *data)
1682 struct mdrestore_struct *mdres = (struct mdrestore_struct *)data;
1683 struct async_work *async;
1689 int compress_size = MAX_PENDING_SIZE * 4;
1691 outfd = fileno(mdres->out);
1692 buffer = malloc(compress_size);
1694 error("not enough memory for restore worker buffer");
1695 pthread_mutex_lock(&mdres->mutex);
1697 mdres->error = -ENOMEM;
1698 pthread_mutex_unlock(&mdres->mutex);
1703 u64 bytenr, physical_dup;
1707 pthread_mutex_lock(&mdres->mutex);
1708 while (!mdres->nodesize || list_empty(&mdres->list)) {
1710 pthread_mutex_unlock(&mdres->mutex);
1713 pthread_cond_wait(&mdres->cond, &mdres->mutex);
1715 async = list_entry(mdres->list.next, struct async_work, list);
1716 list_del_init(&async->list);
1717 pthread_mutex_unlock(&mdres->mutex);
1719 if (mdres->compress_method == COMPRESS_ZLIB) {
1720 size = compress_size;
1721 ret = uncompress(buffer, (unsigned long *)&size,
1722 async->buffer, async->bufsize);
1724 error("decompressiion failed with %d", ret);
1729 outbuf = async->buffer;
1730 size = async->bufsize;
1733 if (!mdres->multi_devices) {
1734 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1735 if (mdres->old_restore) {
1736 update_super_old(outbuf);
1738 ret = update_super(mdres, outbuf);
1742 } else if (!mdres->old_restore) {
1743 ret = fixup_chunk_tree_block(mdres, async, outbuf, size);
1749 if (!mdres->fixup_offset) {
1751 u64 chunk_size = size;
1753 if (!mdres->multi_devices && !mdres->old_restore)
1754 bytenr = logical_to_physical(mdres,
1755 async->start + offset,
1759 bytenr = async->start + offset;
1761 ret = pwrite64(outfd, outbuf+offset, chunk_size,
1763 if (ret != chunk_size)
1767 ret = pwrite64(outfd, outbuf+offset,
1770 if (ret != chunk_size)
1774 offset += chunk_size;
1779 error("unable to write to device: %s",
1783 error("short write");
1787 } else if (async->start != BTRFS_SUPER_INFO_OFFSET) {
1788 ret = write_data_to_disk(mdres->info, outbuf, async->start, size, 0);
1790 error("failed to write data");
1796 /* backup super blocks are already there at fixup_offset stage */
1797 if (!mdres->multi_devices && async->start == BTRFS_SUPER_INFO_OFFSET)
1798 write_backup_supers(outfd, outbuf);
1800 pthread_mutex_lock(&mdres->mutex);
1801 if (err && !mdres->error)
1804 pthread_mutex_unlock(&mdres->mutex);
1806 free(async->buffer);
1814 static void mdrestore_destroy(struct mdrestore_struct *mdres, int num_threads)
1819 while ((n = rb_first(&mdres->chunk_tree))) {
1820 struct fs_chunk *entry;
1822 entry = rb_entry(n, struct fs_chunk, l);
1823 rb_erase(n, &mdres->chunk_tree);
1824 rb_erase(&entry->p, &mdres->physical_tree);
1827 pthread_mutex_lock(&mdres->mutex);
1829 pthread_cond_broadcast(&mdres->cond);
1830 pthread_mutex_unlock(&mdres->mutex);
1832 for (i = 0; i < num_threads; i++)
1833 pthread_join(mdres->threads[i], NULL);
1835 pthread_cond_destroy(&mdres->cond);
1836 pthread_mutex_destroy(&mdres->mutex);
1839 static int mdrestore_init(struct mdrestore_struct *mdres,
1840 FILE *in, FILE *out, int old_restore,
1841 int num_threads, int fixup_offset,
1842 struct btrfs_fs_info *info, int multi_devices)
1846 memset(mdres, 0, sizeof(*mdres));
1847 pthread_cond_init(&mdres->cond, NULL);
1848 pthread_mutex_init(&mdres->mutex, NULL);
1849 INIT_LIST_HEAD(&mdres->list);
1850 INIT_LIST_HEAD(&mdres->overlapping_chunks);
1853 mdres->old_restore = old_restore;
1854 mdres->chunk_tree.rb_node = NULL;
1855 mdres->fixup_offset = fixup_offset;
1857 mdres->multi_devices = multi_devices;
1858 mdres->clear_space_cache = 0;
1859 mdres->last_physical_offset = 0;
1860 mdres->alloced_chunks = 0;
1865 mdres->num_threads = num_threads;
1866 for (i = 0; i < num_threads; i++) {
1867 ret = pthread_create(&mdres->threads[i], NULL, restore_worker,
1870 /* pthread_create returns errno directly */
1876 mdrestore_destroy(mdres, i + 1);
1880 static int fill_mdres_info(struct mdrestore_struct *mdres,
1881 struct async_work *async)
1883 struct btrfs_super_block *super;
1888 /* We've already been initialized */
1889 if (mdres->nodesize)
1892 if (mdres->compress_method == COMPRESS_ZLIB) {
1893 size_t size = MAX_PENDING_SIZE * 2;
1895 buffer = malloc(MAX_PENDING_SIZE * 2);
1898 ret = uncompress(buffer, (unsigned long *)&size,
1899 async->buffer, async->bufsize);
1901 error("decompressiion failed with %d", ret);
1907 outbuf = async->buffer;
1910 super = (struct btrfs_super_block *)outbuf;
1911 mdres->nodesize = btrfs_super_nodesize(super);
1912 memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
1913 memcpy(mdres->uuid, super->dev_item.uuid,
1915 mdres->devid = le64_to_cpu(super->dev_item.devid);
1920 static int add_cluster(struct meta_cluster *cluster,
1921 struct mdrestore_struct *mdres, u64 *next)
1923 struct meta_cluster_item *item;
1924 struct meta_cluster_header *header = &cluster->header;
1925 struct async_work *async;
1930 mdres->compress_method = header->compress;
1932 bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
1933 nritems = le32_to_cpu(header->nritems);
1934 for (i = 0; i < nritems; i++) {
1935 item = &cluster->items[i];
1936 async = calloc(1, sizeof(*async));
1938 error("not enough memory for async data");
1941 async->start = le64_to_cpu(item->bytenr);
1942 async->bufsize = le32_to_cpu(item->size);
1943 async->buffer = malloc(async->bufsize);
1944 if (!async->buffer) {
1945 error("not enough memory for async buffer");
1949 ret = fread(async->buffer, async->bufsize, 1, mdres->in);
1951 error("unable to read buffer: %s", strerror(errno));
1952 free(async->buffer);
1956 bytenr += async->bufsize;
1958 pthread_mutex_lock(&mdres->mutex);
1959 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1960 ret = fill_mdres_info(mdres, async);
1962 error("unable to set up restore state");
1963 pthread_mutex_unlock(&mdres->mutex);
1964 free(async->buffer);
1969 list_add_tail(&async->list, &mdres->list);
1971 pthread_cond_signal(&mdres->cond);
1972 pthread_mutex_unlock(&mdres->mutex);
1974 if (bytenr & BLOCK_MASK) {
1975 char buffer[BLOCK_MASK];
1976 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
1979 ret = fread(buffer, size, 1, mdres->in);
1981 error("failed to read buffer: %s", strerror(errno));
1989 static int wait_for_worker(struct mdrestore_struct *mdres)
1993 pthread_mutex_lock(&mdres->mutex);
1995 while (!ret && mdres->num_items > 0) {
1996 struct timespec ts = {
1998 .tv_nsec = 10000000,
2000 pthread_mutex_unlock(&mdres->mutex);
2001 nanosleep(&ts, NULL);
2002 pthread_mutex_lock(&mdres->mutex);
2005 pthread_mutex_unlock(&mdres->mutex);
2009 static int read_chunk_block(struct mdrestore_struct *mdres, u8 *buffer,
2010 u64 bytenr, u64 item_bytenr, u32 bufsize,
2013 struct extent_buffer *eb;
2017 eb = alloc_dummy_eb(bytenr, mdres->nodesize);
2023 while (item_bytenr != bytenr) {
2024 buffer += mdres->nodesize;
2025 item_bytenr += mdres->nodesize;
2028 memcpy(eb->data, buffer, mdres->nodesize);
2029 if (btrfs_header_bytenr(eb) != bytenr) {
2030 error("eb bytenr does not match found bytenr: %llu != %llu",
2031 (unsigned long long)btrfs_header_bytenr(eb),
2032 (unsigned long long)bytenr);
2037 if (memcmp(mdres->fsid, eb->data + offsetof(struct btrfs_header, fsid),
2039 error("filesystem UUID of eb %llu does not match",
2040 (unsigned long long)bytenr);
2045 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID) {
2046 error("wrong eb %llu owner %llu",
2047 (unsigned long long)bytenr,
2048 (unsigned long long)btrfs_header_owner(eb));
2053 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2054 struct btrfs_chunk *chunk;
2055 struct fs_chunk *fs_chunk;
2056 struct btrfs_key key;
2059 if (btrfs_header_level(eb)) {
2060 u64 blockptr = btrfs_node_blockptr(eb, i);
2062 ret = search_for_chunk_blocks(mdres, blockptr,
2069 /* Yay a leaf! We loves leafs! */
2070 btrfs_item_key_to_cpu(eb, &key, i);
2071 if (key.type != BTRFS_CHUNK_ITEM_KEY)
2074 fs_chunk = malloc(sizeof(struct fs_chunk));
2076 error("not enough memory to allocate chunk");
2080 memset(fs_chunk, 0, sizeof(*fs_chunk));
2081 chunk = btrfs_item_ptr(eb, i, struct btrfs_chunk);
2083 fs_chunk->logical = key.offset;
2084 fs_chunk->physical = btrfs_stripe_offset_nr(eb, chunk, 0);
2085 fs_chunk->bytes = btrfs_chunk_length(eb, chunk);
2086 INIT_LIST_HEAD(&fs_chunk->list);
2087 if (tree_search(&mdres->physical_tree, &fs_chunk->p,
2088 physical_cmp, 1) != NULL)
2089 list_add(&fs_chunk->list, &mdres->overlapping_chunks);
2091 tree_insert(&mdres->physical_tree, &fs_chunk->p,
2094 type = btrfs_chunk_type(eb, chunk);
2095 if (type & BTRFS_BLOCK_GROUP_DUP) {
2096 fs_chunk->physical_dup =
2097 btrfs_stripe_offset_nr(eb, chunk, 1);
2100 if (fs_chunk->physical_dup + fs_chunk->bytes >
2101 mdres->last_physical_offset)
2102 mdres->last_physical_offset = fs_chunk->physical_dup +
2104 else if (fs_chunk->physical + fs_chunk->bytes >
2105 mdres->last_physical_offset)
2106 mdres->last_physical_offset = fs_chunk->physical +
2108 mdres->alloced_chunks += fs_chunk->bytes;
2109 /* in dup case, fs_chunk->bytes should add twice */
2110 if (fs_chunk->physical_dup)
2111 mdres->alloced_chunks += fs_chunk->bytes;
2112 tree_insert(&mdres->chunk_tree, &fs_chunk->l, chunk_cmp);
2119 /* If you have to ask you aren't worthy */
2120 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
2121 u64 search, u64 cluster_bytenr)
2123 struct meta_cluster *cluster;
2124 struct meta_cluster_header *header;
2125 struct meta_cluster_item *item;
2126 u64 current_cluster = cluster_bytenr, bytenr;
2128 u32 bufsize, nritems, i;
2129 u32 max_size = MAX_PENDING_SIZE * 2;
2130 u8 *buffer, *tmp = NULL;
2133 cluster = malloc(BLOCK_SIZE);
2135 error("not enough memory for cluster");
2139 buffer = malloc(max_size);
2141 error("not enough memory for buffer");
2146 if (mdres->compress_method == COMPRESS_ZLIB) {
2147 tmp = malloc(max_size);
2149 error("not enough memory for buffer");
2156 bytenr = current_cluster;
2158 if (fseek(mdres->in, current_cluster, SEEK_SET)) {
2159 error("seek failed: %s\n", strerror(errno));
2164 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2166 if (cluster_bytenr != 0) {
2168 current_cluster = 0;
2173 "unknown state after reading cluster at %llu, probably crrupted data",
2177 } else if (ret < 0) {
2178 error("unable to read image at %llu: %s",
2179 (unsigned long long)cluster_bytenr,
2185 header = &cluster->header;
2186 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2187 le64_to_cpu(header->bytenr) != current_cluster) {
2188 error("bad header in metadump image");
2193 bytenr += BLOCK_SIZE;
2194 nritems = le32_to_cpu(header->nritems);
2195 for (i = 0; i < nritems; i++) {
2198 item = &cluster->items[i];
2199 bufsize = le32_to_cpu(item->size);
2200 item_bytenr = le64_to_cpu(item->bytenr);
2202 if (bufsize > max_size) {
2203 error("item %u too big: %u > %u", i, bufsize,
2209 if (mdres->compress_method == COMPRESS_ZLIB) {
2210 ret = fread(tmp, bufsize, 1, mdres->in);
2212 error("read error: %s", strerror(errno));
2218 ret = uncompress(buffer,
2219 (unsigned long *)&size, tmp,
2222 error("decompressiion failed with %d",
2228 ret = fread(buffer, bufsize, 1, mdres->in);
2230 error("read error: %s",
2239 if (item_bytenr <= search &&
2240 item_bytenr + size > search) {
2241 ret = read_chunk_block(mdres, buffer, search,
2255 if (bytenr & BLOCK_MASK)
2256 bytenr += BLOCK_SIZE - (bytenr & BLOCK_MASK);
2257 current_cluster = bytenr;
2266 static int build_chunk_tree(struct mdrestore_struct *mdres,
2267 struct meta_cluster *cluster)
2269 struct btrfs_super_block *super;
2270 struct meta_cluster_header *header;
2271 struct meta_cluster_item *item = NULL;
2272 u64 chunk_root_bytenr = 0;
2278 /* We can't seek with stdin so don't bother doing this */
2279 if (mdres->in == stdin)
2282 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2284 error("unable to read cluster: %s", strerror(errno));
2289 header = &cluster->header;
2290 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2291 le64_to_cpu(header->bytenr) != 0) {
2292 error("bad header in metadump image");
2296 bytenr += BLOCK_SIZE;
2297 mdres->compress_method = header->compress;
2298 nritems = le32_to_cpu(header->nritems);
2299 for (i = 0; i < nritems; i++) {
2300 item = &cluster->items[i];
2302 if (le64_to_cpu(item->bytenr) == BTRFS_SUPER_INFO_OFFSET)
2304 bytenr += le32_to_cpu(item->size);
2305 if (fseek(mdres->in, le32_to_cpu(item->size), SEEK_CUR)) {
2306 error("seek failed: %s\n", strerror(errno));
2311 if (!item || le64_to_cpu(item->bytenr) != BTRFS_SUPER_INFO_OFFSET) {
2312 error("did not find superblock at %llu",
2313 le64_to_cpu(item->bytenr));
2317 buffer = malloc(le32_to_cpu(item->size));
2319 error("not enough memory to allocate buffer");
2323 ret = fread(buffer, le32_to_cpu(item->size), 1, mdres->in);
2325 error("unable to read buffer: %s", strerror(errno));
2330 if (mdres->compress_method == COMPRESS_ZLIB) {
2331 size_t size = MAX_PENDING_SIZE * 2;
2334 tmp = malloc(MAX_PENDING_SIZE * 2);
2339 ret = uncompress(tmp, (unsigned long *)&size,
2340 buffer, le32_to_cpu(item->size));
2342 error("decompressiion failed with %d", ret);
2351 pthread_mutex_lock(&mdres->mutex);
2352 super = (struct btrfs_super_block *)buffer;
2353 chunk_root_bytenr = btrfs_super_chunk_root(super);
2354 mdres->nodesize = btrfs_super_nodesize(super);
2355 memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
2356 memcpy(mdres->uuid, super->dev_item.uuid,
2358 mdres->devid = le64_to_cpu(super->dev_item.devid);
2360 pthread_mutex_unlock(&mdres->mutex);
2362 return search_for_chunk_blocks(mdres, chunk_root_bytenr, 0);
2365 static int range_contains_super(u64 physical, u64 bytes)
2370 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2371 super_bytenr = btrfs_sb_offset(i);
2372 if (super_bytenr >= physical &&
2373 super_bytenr < physical + bytes)
2380 static void remap_overlapping_chunks(struct mdrestore_struct *mdres)
2382 struct fs_chunk *fs_chunk;
2384 while (!list_empty(&mdres->overlapping_chunks)) {
2385 fs_chunk = list_first_entry(&mdres->overlapping_chunks,
2386 struct fs_chunk, list);
2387 list_del_init(&fs_chunk->list);
2388 if (range_contains_super(fs_chunk->physical,
2391 "remapping a chunk that had a super mirror inside of it, clearing space cache so we don't end up with corruption");
2392 mdres->clear_space_cache = 1;
2394 fs_chunk->physical = mdres->last_physical_offset;
2395 tree_insert(&mdres->physical_tree, &fs_chunk->p, physical_cmp);
2396 mdres->last_physical_offset += fs_chunk->bytes;
2400 static int fixup_devices(struct btrfs_fs_info *fs_info,
2401 struct mdrestore_struct *mdres, off_t dev_size)
2403 struct btrfs_trans_handle *trans;
2404 struct btrfs_dev_item *dev_item;
2405 struct btrfs_path path;
2406 struct extent_buffer *leaf;
2407 struct btrfs_root *root = fs_info->chunk_root;
2408 struct btrfs_key key;
2409 u64 devid, cur_devid;
2412 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2413 if (IS_ERR(trans)) {
2414 error("cannot starting transaction %ld", PTR_ERR(trans));
2415 return PTR_ERR(trans);
2418 dev_item = &fs_info->super_copy->dev_item;
2420 devid = btrfs_stack_device_id(dev_item);
2422 btrfs_set_stack_device_total_bytes(dev_item, dev_size);
2423 btrfs_set_stack_device_bytes_used(dev_item, mdres->alloced_chunks);
2425 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2426 key.type = BTRFS_DEV_ITEM_KEY;
2429 btrfs_init_path(&path);
2432 ret = btrfs_search_slot(trans, root, &key, &path, -1, 1);
2434 error("search failed: %d", ret);
2439 leaf = path.nodes[0];
2440 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2441 ret = btrfs_next_leaf(root, &path);
2443 error("cannot go to next leaf %d", ret);
2450 leaf = path.nodes[0];
2453 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2454 if (key.type > BTRFS_DEV_ITEM_KEY)
2456 if (key.type != BTRFS_DEV_ITEM_KEY) {
2461 dev_item = btrfs_item_ptr(leaf, path.slots[0],
2462 struct btrfs_dev_item);
2463 cur_devid = btrfs_device_id(leaf, dev_item);
2464 if (devid != cur_devid) {
2465 ret = btrfs_del_item(trans, root, &path);
2467 error("cannot delete item: %d", ret);
2470 btrfs_release_path(&path);
2474 btrfs_set_device_total_bytes(leaf, dev_item, dev_size);
2475 btrfs_set_device_bytes_used(leaf, dev_item,
2476 mdres->alloced_chunks);
2477 btrfs_mark_buffer_dirty(leaf);
2481 btrfs_release_path(&path);
2482 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2484 error("unable to commit transaction: %d", ret);
2490 static int restore_metadump(const char *input, FILE *out, int old_restore,
2491 int num_threads, int fixup_offset,
2492 const char *target, int multi_devices)
2494 struct meta_cluster *cluster = NULL;
2495 struct meta_cluster_header *header;
2496 struct mdrestore_struct mdrestore;
2497 struct btrfs_fs_info *info = NULL;
2502 if (!strcmp(input, "-")) {
2505 in = fopen(input, "r");
2507 error("unable to open metadump image: %s",
2513 /* NOTE: open with write mode */
2515 info = open_ctree_fs_info(target, 0, 0, 0,
2517 OPEN_CTREE_RESTORE |
2518 OPEN_CTREE_PARTIAL);
2520 error("open ctree failed");
2526 cluster = malloc(BLOCK_SIZE);
2528 error("not enough memory for cluster");
2533 ret = mdrestore_init(&mdrestore, in, out, old_restore, num_threads,
2534 fixup_offset, info, multi_devices);
2536 error("failed to intialize metadata restore state: %d", ret);
2537 goto failed_cluster;
2540 if (!multi_devices && !old_restore) {
2541 ret = build_chunk_tree(&mdrestore, cluster);
2544 if (!list_empty(&mdrestore.overlapping_chunks))
2545 remap_overlapping_chunks(&mdrestore);
2548 if (in != stdin && fseek(in, 0, SEEK_SET)) {
2549 error("seek failed: %s\n", strerror(errno));
2553 while (!mdrestore.error) {
2554 ret = fread(cluster, BLOCK_SIZE, 1, in);
2558 header = &cluster->header;
2559 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2560 le64_to_cpu(header->bytenr) != bytenr) {
2561 error("bad header in metadump image");
2565 ret = add_cluster(cluster, &mdrestore, &bytenr);
2567 error("failed to add cluster: %d", ret);
2571 ret = wait_for_worker(&mdrestore);
2573 if (!ret && !multi_devices && !old_restore) {
2574 struct btrfs_root *root;
2577 root = open_ctree_fd(fileno(out), target, 0,
2578 OPEN_CTREE_PARTIAL |
2580 OPEN_CTREE_NO_DEVICES);
2582 error("open ctree failed in %s", target);
2586 info = root->fs_info;
2588 if (stat(target, &st)) {
2589 error("stat %s failed: %s", target, strerror(errno));
2590 close_ctree(info->chunk_root);
2595 ret = fixup_devices(info, &mdrestore, st.st_size);
2596 close_ctree(info->chunk_root);
2601 mdrestore_destroy(&mdrestore, num_threads);
2605 if (fixup_offset && info)
2606 close_ctree(info->chunk_root);
2613 static int update_disk_super_on_device(struct btrfs_fs_info *info,
2614 const char *other_dev, u64 cur_devid)
2616 struct btrfs_key key;
2617 struct extent_buffer *leaf;
2618 struct btrfs_path path;
2619 struct btrfs_dev_item *dev_item;
2620 struct btrfs_super_block *disk_super;
2621 char dev_uuid[BTRFS_UUID_SIZE];
2622 char fs_uuid[BTRFS_UUID_SIZE];
2623 u64 devid, type, io_align, io_width;
2624 u64 sector_size, total_bytes, bytes_used;
2625 char buf[BTRFS_SUPER_INFO_SIZE];
2629 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2630 key.type = BTRFS_DEV_ITEM_KEY;
2631 key.offset = cur_devid;
2633 btrfs_init_path(&path);
2634 ret = btrfs_search_slot(NULL, info->chunk_root, &key, &path, 0, 0);
2636 error("search key failed: %d", ret);
2641 leaf = path.nodes[0];
2642 dev_item = btrfs_item_ptr(leaf, path.slots[0],
2643 struct btrfs_dev_item);
2645 devid = btrfs_device_id(leaf, dev_item);
2646 if (devid != cur_devid) {
2647 error("devid mismatch: %llu != %llu",
2648 (unsigned long long)devid,
2649 (unsigned long long)cur_devid);
2654 type = btrfs_device_type(leaf, dev_item);
2655 io_align = btrfs_device_io_align(leaf, dev_item);
2656 io_width = btrfs_device_io_width(leaf, dev_item);
2657 sector_size = btrfs_device_sector_size(leaf, dev_item);
2658 total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2659 bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2660 read_extent_buffer(leaf, dev_uuid, (unsigned long)btrfs_device_uuid(dev_item), BTRFS_UUID_SIZE);
2661 read_extent_buffer(leaf, fs_uuid, (unsigned long)btrfs_device_fsid(dev_item), BTRFS_UUID_SIZE);
2663 btrfs_release_path(&path);
2665 printf("update disk super on %s devid=%llu\n", other_dev, devid);
2667 /* update other devices' super block */
2668 fp = open(other_dev, O_CREAT | O_RDWR, 0600);
2670 error("could not open %s: %s", other_dev, strerror(errno));
2675 memcpy(buf, info->super_copy, BTRFS_SUPER_INFO_SIZE);
2677 disk_super = (struct btrfs_super_block *)buf;
2678 dev_item = &disk_super->dev_item;
2680 btrfs_set_stack_device_type(dev_item, type);
2681 btrfs_set_stack_device_id(dev_item, devid);
2682 btrfs_set_stack_device_total_bytes(dev_item, total_bytes);
2683 btrfs_set_stack_device_bytes_used(dev_item, bytes_used);
2684 btrfs_set_stack_device_io_align(dev_item, io_align);
2685 btrfs_set_stack_device_io_width(dev_item, io_width);
2686 btrfs_set_stack_device_sector_size(dev_item, sector_size);
2687 memcpy(dev_item->uuid, dev_uuid, BTRFS_UUID_SIZE);
2688 memcpy(dev_item->fsid, fs_uuid, BTRFS_UUID_SIZE);
2689 csum_block((u8 *)buf, BTRFS_SUPER_INFO_SIZE);
2691 ret = pwrite64(fp, buf, BTRFS_SUPER_INFO_SIZE, BTRFS_SUPER_INFO_OFFSET);
2692 if (ret != BTRFS_SUPER_INFO_SIZE) {
2694 error("cannot write superblock: %s", strerror(ret));
2696 error("cannot write superblock");
2701 write_backup_supers(fp, (u8 *)buf);
2709 static void print_usage(int ret)
2711 printf("usage: btrfs-image [options] source target\n");
2712 printf("\t-r \trestore metadump image\n");
2713 printf("\t-c value\tcompression level (0 ~ 9)\n");
2714 printf("\t-t value\tnumber of threads (1 ~ 32)\n");
2715 printf("\t-o \tdon't mess with the chunk tree when restoring\n");
2716 printf("\t-s \tsanitize file names, use once to just use garbage, use twice if you want crc collisions\n");
2717 printf("\t-w \twalk all trees instead of using extent tree, do this if your extent tree is broken\n");
2718 printf("\t-m \trestore for multiple devices\n");
2720 printf("\tIn the dump mode, source is the btrfs device and target is the output file (use '-' for stdout).\n");
2721 printf("\tIn the restore mode, source is the dumped image and target is the btrfs device/file.\n");
2725 int main(int argc, char *argv[])
2729 u64 num_threads = 0;
2730 u64 compress_level = 0;
2732 int old_restore = 0;
2734 int multi_devices = 0;
2738 int usage_error = 0;
2742 static const struct option long_options[] = {
2743 { "help", no_argument, NULL, GETOPT_VAL_HELP},
2744 { NULL, 0, NULL, 0 }
2746 int c = getopt_long(argc, argv, "rc:t:oswm", long_options, NULL);
2754 num_threads = arg_strtou64(optarg);
2755 if (num_threads > MAX_WORKER_THREADS) {
2756 error("number of threads out of range: %llu > %d",
2757 (unsigned long long)num_threads,
2758 MAX_WORKER_THREADS);
2763 compress_level = arg_strtou64(optarg);
2764 if (compress_level > 9) {
2765 error("compression level out of range: %llu",
2766 (unsigned long long)compress_level);
2783 case GETOPT_VAL_HELP:
2785 print_usage(c != GETOPT_VAL_HELP);
2790 if (check_argc_min(argc - optind, 2))
2793 dev_cnt = argc - optind - 1;
2798 "create and restore cannot be used at the same time");
2802 if (walk_trees || sanitize || compress_level) {
2804 "useing -w, -s, -c options for restore makes no sense");
2807 if (multi_devices && dev_cnt < 2) {
2808 error("not enough devices specified for -m option");
2811 if (!multi_devices && dev_cnt != 1) {
2812 error("accepts only 1 device without -m option");
2820 source = argv[optind];
2821 target = argv[optind + 1];
2823 if (create && !strcmp(target, "-")) {
2826 out = fopen(target, "w+");
2828 error("unable to create target file %s", target);
2833 if (compress_level > 0 || create == 0) {
2834 if (num_threads == 0) {
2835 long tmp = sysconf(_SC_NPROCESSORS_ONLN);
2846 ret = check_mounted(source);
2848 warning("unable to check mount status of: %s",
2851 warning("%s already mounted, results may be inaccurate",
2855 ret = create_metadump(source, out, num_threads,
2856 compress_level, sanitize, walk_trees);
2858 ret = restore_metadump(source, out, old_restore, num_threads,
2859 0, target, multi_devices);
2862 error("%s failed: %s", (create) ? "create" : "restore",
2867 /* extended support for multiple devices */
2868 if (!create && multi_devices) {
2869 struct btrfs_fs_info *info;
2873 info = open_ctree_fs_info(target, 0, 0, 0,
2874 OPEN_CTREE_PARTIAL |
2875 OPEN_CTREE_RESTORE);
2877 error("open ctree failed at %s", target);
2881 total_devs = btrfs_super_num_devices(info->super_copy);
2882 if (total_devs != dev_cnt) {
2883 error("it needs %llu devices but has only %d",
2884 total_devs, dev_cnt);
2885 close_ctree(info->chunk_root);
2889 /* update super block on other disks */
2890 for (i = 2; i <= dev_cnt; i++) {
2891 ret = update_disk_super_on_device(info,
2892 argv[optind + i], (u64)i);
2894 error("update disk superblock failed devid %d: %d",
2896 close_ctree(info->chunk_root);
2901 close_ctree(info->chunk_root);
2903 /* fix metadata block to map correct chunk */
2904 ret = restore_metadump(source, out, 0, num_threads, 1,
2907 error("unable to fixup metadump: %d", ret);
2912 if (out == stdout) {
2916 if (ret && create) {
2919 unlink_ret = unlink(target);
2921 error("unlink output file %s failed: %s",
2922 target, strerror(errno));
2926 btrfs_close_all_devices();