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 struct meta_cluster_item {
50 } __attribute__ ((__packed__));
52 struct meta_cluster_header {
57 } __attribute__ ((__packed__));
59 /* cluster header + index items + buffers */
61 struct meta_cluster_header header;
62 struct meta_cluster_item items[];
63 } __attribute__ ((__packed__));
65 #define ITEMS_PER_CLUSTER ((BLOCK_SIZE - sizeof(struct meta_cluster)) / \
66 sizeof(struct meta_cluster_item))
72 * physical_dup only store additonal physical for BTRFS_BLOCK_GROUP_DUP
73 * currently restore only support single and DUP
74 * TODO: modify this structure and the function related to this
75 * structure for support RAID*
81 struct list_head list;
85 struct list_head list;
86 struct list_head ordered;
94 struct metadump_struct {
95 struct btrfs_root *root;
98 struct meta_cluster *cluster;
102 pthread_mutex_t mutex;
104 struct rb_root name_tree;
106 struct list_head list;
107 struct list_head ordered;
129 struct mdrestore_struct {
135 pthread_mutex_t mutex;
138 struct rb_root chunk_tree;
139 struct rb_root physical_tree;
140 struct list_head list;
141 struct list_head overlapping_chunks;
146 u64 last_physical_offset;
147 u8 uuid[BTRFS_UUID_SIZE];
148 u8 fsid[BTRFS_FSID_SIZE];
156 int clear_space_cache;
157 struct btrfs_fs_info *info;
160 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
161 u64 search, u64 cluster_bytenr);
162 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size);
164 static void csum_block(u8 *buf, size_t len)
166 u8 result[BTRFS_CRC32_SIZE];
168 crc = crc32c(crc, buf + BTRFS_CSUM_SIZE, len - BTRFS_CSUM_SIZE);
169 btrfs_csum_final(crc, result);
170 memcpy(buf, result, BTRFS_CRC32_SIZE);
173 static int has_name(struct btrfs_key *key)
176 case BTRFS_DIR_ITEM_KEY:
177 case BTRFS_DIR_INDEX_KEY:
178 case BTRFS_INODE_REF_KEY:
179 case BTRFS_INODE_EXTREF_KEY:
180 case BTRFS_XATTR_ITEM_KEY:
189 static char *generate_garbage(u32 name_len)
191 char *buf = malloc(name_len);
197 for (i = 0; i < name_len; i++) {
198 char c = rand_range(94) + 33;
208 static int name_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
210 struct name *entry = rb_entry(a, struct name, n);
211 struct name *ins = rb_entry(b, struct name, n);
214 len = min(ins->len, entry->len);
215 return memcmp(ins->val, entry->val, len);
218 static int chunk_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
220 struct fs_chunk *entry = rb_entry(a, struct fs_chunk, l);
221 struct fs_chunk *ins = rb_entry(b, struct fs_chunk, l);
223 if (fuzz && ins->logical >= entry->logical &&
224 ins->logical < entry->logical + entry->bytes)
227 if (ins->logical < entry->logical)
229 else if (ins->logical > entry->logical)
234 static int physical_cmp(struct rb_node *a, struct rb_node *b, int fuzz)
236 struct fs_chunk *entry = rb_entry(a, struct fs_chunk, p);
237 struct fs_chunk *ins = rb_entry(b, struct fs_chunk, p);
239 if (fuzz && ins->physical >= entry->physical &&
240 ins->physical < entry->physical + entry->bytes)
243 if (fuzz && entry->physical >= ins->physical &&
244 entry->physical < ins->physical + ins->bytes)
247 if (ins->physical < entry->physical)
249 else if (ins->physical > entry->physical)
254 static void tree_insert(struct rb_root *root, struct rb_node *ins,
255 int (*cmp)(struct rb_node *a, struct rb_node *b,
258 struct rb_node ** p = &root->rb_node;
259 struct rb_node * parent = NULL;
265 dir = cmp(*p, ins, 1);
274 rb_link_node(ins, parent, p);
275 rb_insert_color(ins, root);
278 static struct rb_node *tree_search(struct rb_root *root,
279 struct rb_node *search,
280 int (*cmp)(struct rb_node *a,
281 struct rb_node *b, int fuzz),
284 struct rb_node *n = root->rb_node;
288 dir = cmp(n, search, fuzz);
300 static u64 logical_to_physical(struct mdrestore_struct *mdres, u64 logical,
301 u64 *size, u64 *physical_dup)
303 struct fs_chunk *fs_chunk;
304 struct rb_node *entry;
305 struct fs_chunk search;
308 if (logical == BTRFS_SUPER_INFO_OFFSET)
311 search.logical = logical;
312 entry = tree_search(&mdres->chunk_tree, &search.l, chunk_cmp, 1);
314 if (mdres->in != stdin)
315 warning("cannot find a chunk, using logical");
318 fs_chunk = rb_entry(entry, struct fs_chunk, l);
319 if (fs_chunk->logical > logical || fs_chunk->logical + fs_chunk->bytes < logical)
321 offset = search.logical - fs_chunk->logical;
324 /* Only in dup case, physical_dup is not equal to 0 */
325 if (fs_chunk->physical_dup)
326 *physical_dup = fs_chunk->physical_dup + offset;
331 *size = min(*size, fs_chunk->bytes + fs_chunk->logical - logical);
332 return fs_chunk->physical + offset;
336 static char *find_collision(struct metadump_struct *md, char *name,
340 struct rb_node *entry;
342 unsigned long checksum;
348 entry = tree_search(&md->name_tree, &tmp.n, name_cmp, 0);
350 val = rb_entry(entry, struct name, n);
355 val = malloc(sizeof(struct name));
357 error("cannot sanitize name, not enough memory");
362 memset(val, 0, sizeof(*val));
366 val->sub = malloc(name_len);
368 error("cannot sanitize name, not enough memory");
374 checksum = crc32c(~1, val->val, name_len);
375 memset(val->sub, ' ', name_len);
378 if (crc32c(~1, val->sub, name_len) == checksum &&
379 memcmp(val->sub, val->val, val->len)) {
384 if (val->sub[i] == 127) {
389 } while (val->sub[i] == 127);
394 if (val->sub[i] == '/')
396 memset(val->sub, ' ', i);
401 if (val->sub[i] == '/')
408 "cannot find a hash collision for '%.*s', generating garbage, it won't match indexes",
410 for (i = 0; i < name_len; i++) {
411 char c = rand_range(94) + 33;
419 tree_insert(&md->name_tree, &val->n, name_cmp);
423 static void sanitize_dir_item(struct metadump_struct *md, struct extent_buffer *eb,
426 struct btrfs_dir_item *dir_item;
429 unsigned long name_ptr;
434 int free_garbage = (md->sanitize_names == 1);
436 dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
437 total_len = btrfs_item_size_nr(eb, slot);
438 while (cur < total_len) {
439 this_len = sizeof(*dir_item) +
440 btrfs_dir_name_len(eb, dir_item) +
441 btrfs_dir_data_len(eb, dir_item);
442 name_ptr = (unsigned long)(dir_item + 1);
443 name_len = btrfs_dir_name_len(eb, dir_item);
445 if (md->sanitize_names > 1) {
446 buf = malloc(name_len);
448 error("cannot sanitize name, not enough memory");
451 read_extent_buffer(eb, buf, name_ptr, name_len);
452 garbage = find_collision(md, buf, name_len);
454 garbage = generate_garbage(name_len);
457 error("cannot sanitize name, not enough memory");
460 write_extent_buffer(eb, garbage, name_ptr, name_len);
462 dir_item = (struct btrfs_dir_item *)((char *)dir_item +
469 static void sanitize_inode_ref(struct metadump_struct *md,
470 struct extent_buffer *eb, int slot, int ext)
472 struct btrfs_inode_extref *extref;
473 struct btrfs_inode_ref *ref;
476 unsigned long name_ptr;
480 int free_garbage = (md->sanitize_names == 1);
482 item_size = btrfs_item_size_nr(eb, slot);
483 ptr = btrfs_item_ptr_offset(eb, slot);
484 while (cur_offset < item_size) {
486 extref = (struct btrfs_inode_extref *)(ptr +
488 name_ptr = (unsigned long)(&extref->name);
489 len = btrfs_inode_extref_name_len(eb, extref);
490 cur_offset += sizeof(*extref);
492 ref = (struct btrfs_inode_ref *)(ptr + cur_offset);
493 len = btrfs_inode_ref_name_len(eb, ref);
494 name_ptr = (unsigned long)(ref + 1);
495 cur_offset += sizeof(*ref);
499 if (md->sanitize_names > 1) {
502 error("cannot sanitize name, not enough memory");
505 read_extent_buffer(eb, buf, name_ptr, len);
506 garbage = find_collision(md, buf, len);
508 garbage = generate_garbage(len);
512 error("cannot sanitize name, not enough memory");
515 write_extent_buffer(eb, garbage, name_ptr, len);
521 static void sanitize_xattr(struct metadump_struct *md,
522 struct extent_buffer *eb, int slot)
524 struct btrfs_dir_item *dir_item;
525 unsigned long data_ptr;
528 dir_item = btrfs_item_ptr(eb, slot, struct btrfs_dir_item);
529 data_len = btrfs_dir_data_len(eb, dir_item);
531 data_ptr = (unsigned long)((char *)(dir_item + 1) +
532 btrfs_dir_name_len(eb, dir_item));
533 memset_extent_buffer(eb, 0, data_ptr, data_len);
536 static void sanitize_name(struct metadump_struct *md, u8 *dst,
537 struct extent_buffer *src, struct btrfs_key *key,
540 struct extent_buffer *eb;
542 eb = alloc_dummy_eb(src->start, src->len);
544 error("cannot sanitize name, not enough memory");
548 memcpy(eb->data, dst, eb->len);
551 case BTRFS_DIR_ITEM_KEY:
552 case BTRFS_DIR_INDEX_KEY:
553 sanitize_dir_item(md, eb, slot);
555 case BTRFS_INODE_REF_KEY:
556 sanitize_inode_ref(md, eb, slot, 0);
558 case BTRFS_INODE_EXTREF_KEY:
559 sanitize_inode_ref(md, eb, slot, 1);
561 case BTRFS_XATTR_ITEM_KEY:
562 sanitize_xattr(md, eb, slot);
568 memcpy(dst, eb->data, eb->len);
573 * zero inline extents and csum items
575 static void zero_items(struct metadump_struct *md, u8 *dst,
576 struct extent_buffer *src)
578 struct btrfs_file_extent_item *fi;
579 struct btrfs_item *item;
580 struct btrfs_key key;
581 u32 nritems = btrfs_header_nritems(src);
586 for (i = 0; i < nritems; i++) {
587 item = btrfs_item_nr(i);
588 btrfs_item_key_to_cpu(src, &key, i);
589 if (key.type == BTRFS_CSUM_ITEM_KEY) {
590 size = btrfs_item_size_nr(src, i);
591 memset(dst + btrfs_leaf_data(src) +
592 btrfs_item_offset_nr(src, i), 0, size);
596 if (md->sanitize_names && has_name(&key)) {
597 sanitize_name(md, dst, src, &key, i);
601 if (key.type != BTRFS_EXTENT_DATA_KEY)
604 fi = btrfs_item_ptr(src, i, struct btrfs_file_extent_item);
605 extent_type = btrfs_file_extent_type(src, fi);
606 if (extent_type != BTRFS_FILE_EXTENT_INLINE)
609 ptr = btrfs_file_extent_inline_start(fi);
610 size = btrfs_file_extent_inline_item_len(src, item);
611 memset(dst + ptr, 0, size);
616 * copy buffer and zero useless data in the buffer
618 static void copy_buffer(struct metadump_struct *md, u8 *dst,
619 struct extent_buffer *src)
625 memcpy(dst, src->data, src->len);
626 if (src->start == BTRFS_SUPER_INFO_OFFSET)
629 level = btrfs_header_level(src);
630 nritems = btrfs_header_nritems(src);
633 size = sizeof(struct btrfs_header);
634 memset(dst + size, 0, src->len - size);
635 } else if (level == 0) {
636 size = btrfs_leaf_data(src) +
637 btrfs_item_offset_nr(src, nritems - 1) -
638 btrfs_item_nr_offset(nritems);
639 memset(dst + btrfs_item_nr_offset(nritems), 0, size);
640 zero_items(md, dst, src);
642 size = offsetof(struct btrfs_node, ptrs) +
643 sizeof(struct btrfs_key_ptr) * nritems;
644 memset(dst + size, 0, src->len - size);
646 csum_block(dst, src->len);
649 static void *dump_worker(void *data)
651 struct metadump_struct *md = (struct metadump_struct *)data;
652 struct async_work *async;
656 pthread_mutex_lock(&md->mutex);
657 while (list_empty(&md->list)) {
659 pthread_mutex_unlock(&md->mutex);
662 pthread_cond_wait(&md->cond, &md->mutex);
664 async = list_entry(md->list.next, struct async_work, list);
665 list_del_init(&async->list);
666 pthread_mutex_unlock(&md->mutex);
668 if (md->compress_level > 0) {
669 u8 *orig = async->buffer;
671 async->bufsize = compressBound(async->size);
672 async->buffer = malloc(async->bufsize);
673 if (!async->buffer) {
674 error("not enough memory for async buffer");
675 pthread_mutex_lock(&md->mutex);
678 pthread_mutex_unlock(&md->mutex);
682 ret = compress2(async->buffer,
683 (unsigned long *)&async->bufsize,
684 orig, async->size, md->compress_level);
692 pthread_mutex_lock(&md->mutex);
694 pthread_mutex_unlock(&md->mutex);
700 static void meta_cluster_init(struct metadump_struct *md, u64 start)
702 struct meta_cluster_header *header;
706 header = &md->cluster->header;
707 header->magic = cpu_to_le64(HEADER_MAGIC);
708 header->bytenr = cpu_to_le64(start);
709 header->nritems = cpu_to_le32(0);
710 header->compress = md->compress_level > 0 ?
711 COMPRESS_ZLIB : COMPRESS_NONE;
714 static void metadump_destroy(struct metadump_struct *md, int num_threads)
719 pthread_mutex_lock(&md->mutex);
721 pthread_cond_broadcast(&md->cond);
722 pthread_mutex_unlock(&md->mutex);
724 for (i = 0; i < num_threads; i++)
725 pthread_join(md->threads[i], NULL);
727 pthread_cond_destroy(&md->cond);
728 pthread_mutex_destroy(&md->mutex);
730 while ((n = rb_first(&md->name_tree))) {
733 name = rb_entry(n, struct name, n);
734 rb_erase(n, &md->name_tree);
743 static int metadump_init(struct metadump_struct *md, struct btrfs_root *root,
744 FILE *out, int num_threads, int compress_level,
749 memset(md, 0, sizeof(*md));
750 md->cluster = calloc(1, BLOCK_SIZE);
753 md->threads = calloc(num_threads, sizeof(pthread_t));
758 INIT_LIST_HEAD(&md->list);
759 INIT_LIST_HEAD(&md->ordered);
762 md->pending_start = (u64)-1;
763 md->compress_level = compress_level;
764 md->sanitize_names = sanitize_names;
765 if (sanitize_names > 1)
766 crc32c_optimization_init();
768 md->name_tree.rb_node = NULL;
769 md->num_threads = num_threads;
770 pthread_cond_init(&md->cond, NULL);
771 pthread_mutex_init(&md->mutex, NULL);
772 meta_cluster_init(md, 0);
777 for (i = 0; i < num_threads; i++) {
778 ret = pthread_create(md->threads + i, NULL, dump_worker, md);
784 metadump_destroy(md, i + 1);
789 static int write_zero(FILE *out, size_t size)
791 static char zero[BLOCK_SIZE];
792 return fwrite(zero, size, 1, out);
795 static int write_buffers(struct metadump_struct *md, u64 *next)
797 struct meta_cluster_header *header = &md->cluster->header;
798 struct meta_cluster_item *item;
799 struct async_work *async;
805 if (list_empty(&md->ordered))
808 /* wait until all buffers are compressed */
809 while (!err && md->num_items > md->num_ready) {
810 struct timespec ts = {
814 pthread_mutex_unlock(&md->mutex);
815 nanosleep(&ts, NULL);
816 pthread_mutex_lock(&md->mutex);
821 error("one of the threads failed: %s", strerror(-err));
825 /* setup and write index block */
826 list_for_each_entry(async, &md->ordered, ordered) {
827 item = md->cluster->items + nritems;
828 item->bytenr = cpu_to_le64(async->start);
829 item->size = cpu_to_le32(async->bufsize);
832 header->nritems = cpu_to_le32(nritems);
834 ret = fwrite(md->cluster, BLOCK_SIZE, 1, md->out);
836 error("unable to write out cluster: %s", strerror(errno));
841 bytenr += le64_to_cpu(header->bytenr) + BLOCK_SIZE;
842 while (!list_empty(&md->ordered)) {
843 async = list_entry(md->ordered.next, struct async_work,
845 list_del_init(&async->ordered);
847 bytenr += async->bufsize;
849 ret = fwrite(async->buffer, async->bufsize, 1,
852 error("unable to write out cluster: %s",
862 /* zero unused space in the last block */
863 if (!err && bytenr & BLOCK_MASK) {
864 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
867 ret = write_zero(md->out, size);
869 error("unable to zero out buffer: %s",
879 static int read_data_extent(struct metadump_struct *md,
880 struct async_work *async)
882 struct btrfs_root *root = md->root;
883 u64 bytes_left = async->size;
884 u64 logical = async->start;
891 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree, logical,
894 /* Try our best to read data, just like read_tree_block() */
895 for (cur_mirror = 0; cur_mirror < num_copies; cur_mirror++) {
897 read_len = bytes_left;
898 ret = read_extent_data(root,
899 (char *)(async->buffer + offset),
900 logical, &read_len, cur_mirror);
905 bytes_left -= read_len;
913 static int get_dev_fd(struct btrfs_root *root)
915 struct btrfs_device *dev;
917 dev = list_first_entry(&root->fs_info->fs_devices->devices,
918 struct btrfs_device, dev_list);
922 static int flush_pending(struct metadump_struct *md, int done)
924 struct async_work *async = NULL;
925 struct extent_buffer *eb;
926 u64 blocksize = md->root->nodesize;
932 if (md->pending_size) {
933 async = calloc(1, sizeof(*async));
937 async->start = md->pending_start;
938 async->size = md->pending_size;
939 async->bufsize = async->size;
940 async->buffer = malloc(async->bufsize);
941 if (!async->buffer) {
946 start = async->start;
950 ret = read_data_extent(md, async);
959 * Balance can make the mapping not cover the super block, so
960 * just copy directly from one of the devices.
962 if (start == BTRFS_SUPER_INFO_OFFSET) {
963 int fd = get_dev_fd(md->root);
965 ret = pread64(fd, async->buffer, size, start);
969 error("unable to read superblock at %llu: %s",
970 (unsigned long long)start,
978 while (!md->data && size > 0) {
979 u64 this_read = min(blocksize, size);
980 eb = read_tree_block(md->root, start, this_read, 0);
981 if (!extent_buffer_uptodate(eb)) {
984 error("unable to read metadata block %llu",
985 (unsigned long long)start);
988 copy_buffer(md, async->buffer + offset, eb);
989 free_extent_buffer(eb);
995 md->pending_start = (u64)-1;
996 md->pending_size = 0;
1001 pthread_mutex_lock(&md->mutex);
1003 list_add_tail(&async->ordered, &md->ordered);
1005 if (md->compress_level > 0) {
1006 list_add_tail(&async->list, &md->list);
1007 pthread_cond_signal(&md->cond);
1012 if (md->num_items >= ITEMS_PER_CLUSTER || done) {
1013 ret = write_buffers(md, &start);
1015 error("unable to write buffers: %s", strerror(-ret));
1017 meta_cluster_init(md, start);
1019 pthread_mutex_unlock(&md->mutex);
1023 static int add_extent(u64 start, u64 size, struct metadump_struct *md,
1027 if (md->data != data ||
1028 md->pending_size + size > MAX_PENDING_SIZE ||
1029 md->pending_start + md->pending_size != start) {
1030 ret = flush_pending(md, 0);
1033 md->pending_start = start;
1035 readahead_tree_block(md->root, start, size, 0);
1036 md->pending_size += size;
1041 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1042 static int is_tree_block(struct btrfs_root *extent_root,
1043 struct btrfs_path *path, u64 bytenr)
1045 struct extent_buffer *leaf;
1046 struct btrfs_key key;
1050 leaf = path->nodes[0];
1052 struct btrfs_extent_ref_v0 *ref_item;
1054 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1055 ret = btrfs_next_leaf(extent_root, path);
1060 leaf = path->nodes[0];
1062 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1063 if (key.objectid != bytenr)
1065 if (key.type != BTRFS_EXTENT_REF_V0_KEY)
1067 ref_item = btrfs_item_ptr(leaf, path->slots[0],
1068 struct btrfs_extent_ref_v0);
1069 ref_objectid = btrfs_ref_objectid_v0(leaf, ref_item);
1070 if (ref_objectid < BTRFS_FIRST_FREE_OBJECTID)
1078 static int copy_tree_blocks(struct btrfs_root *root, struct extent_buffer *eb,
1079 struct metadump_struct *metadump, int root_tree)
1081 struct extent_buffer *tmp;
1082 struct btrfs_root_item *ri;
1083 struct btrfs_key key;
1090 ret = add_extent(btrfs_header_bytenr(eb), root->nodesize, metadump, 0);
1092 error("unable to add metadata block %llu: %d",
1093 btrfs_header_bytenr(eb), ret);
1097 if (btrfs_header_level(eb) == 0 && !root_tree)
1100 level = btrfs_header_level(eb);
1101 nritems = btrfs_header_nritems(eb);
1102 for (i = 0; i < nritems; i++) {
1104 btrfs_item_key_to_cpu(eb, &key, i);
1105 if (key.type != BTRFS_ROOT_ITEM_KEY)
1107 ri = btrfs_item_ptr(eb, i, struct btrfs_root_item);
1108 bytenr = btrfs_disk_root_bytenr(eb, ri);
1109 tmp = read_tree_block(root, bytenr, root->nodesize, 0);
1110 if (!extent_buffer_uptodate(tmp)) {
1111 error("unable to read log root block");
1114 ret = copy_tree_blocks(root, tmp, metadump, 0);
1115 free_extent_buffer(tmp);
1119 bytenr = btrfs_node_blockptr(eb, i);
1120 tmp = read_tree_block(root, bytenr, root->nodesize, 0);
1121 if (!extent_buffer_uptodate(tmp)) {
1122 error("unable to read log root block");
1125 ret = copy_tree_blocks(root, tmp, metadump, root_tree);
1126 free_extent_buffer(tmp);
1135 static int copy_log_trees(struct btrfs_root *root,
1136 struct metadump_struct *metadump,
1137 struct btrfs_path *path)
1139 u64 blocknr = btrfs_super_log_root(root->fs_info->super_copy);
1144 if (!root->fs_info->log_root_tree ||
1145 !root->fs_info->log_root_tree->node) {
1146 error("unable to copy tree log, it has not been setup");
1150 return copy_tree_blocks(root, root->fs_info->log_root_tree->node,
1154 static int copy_space_cache(struct btrfs_root *root,
1155 struct metadump_struct *metadump,
1156 struct btrfs_path *path)
1158 struct extent_buffer *leaf;
1159 struct btrfs_file_extent_item *fi;
1160 struct btrfs_key key;
1161 u64 bytenr, num_bytes;
1164 root = root->fs_info->tree_root;
1167 key.type = BTRFS_EXTENT_DATA_KEY;
1170 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1172 error("free space inode not found: %d", ret);
1176 leaf = path->nodes[0];
1179 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1180 ret = btrfs_next_leaf(root, path);
1182 error("cannot go to next leaf %d", ret);
1187 leaf = path->nodes[0];
1190 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1191 if (key.type != BTRFS_EXTENT_DATA_KEY) {
1196 fi = btrfs_item_ptr(leaf, path->slots[0],
1197 struct btrfs_file_extent_item);
1198 if (btrfs_file_extent_type(leaf, fi) !=
1199 BTRFS_FILE_EXTENT_REG) {
1204 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1205 num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
1206 ret = add_extent(bytenr, num_bytes, metadump, 1);
1208 error("unable to add space cache blocks %d", ret);
1209 btrfs_release_path(path);
1218 static int copy_from_extent_tree(struct metadump_struct *metadump,
1219 struct btrfs_path *path)
1221 struct btrfs_root *extent_root;
1222 struct extent_buffer *leaf;
1223 struct btrfs_extent_item *ei;
1224 struct btrfs_key key;
1229 extent_root = metadump->root->fs_info->extent_root;
1230 bytenr = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
1231 key.objectid = bytenr;
1232 key.type = BTRFS_EXTENT_ITEM_KEY;
1235 ret = btrfs_search_slot(NULL, extent_root, &key, path, 0, 0);
1237 error("extent root not found: %d", ret);
1242 leaf = path->nodes[0];
1245 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1246 ret = btrfs_next_leaf(extent_root, path);
1248 error("cannot go to next leaf %d", ret);
1255 leaf = path->nodes[0];
1258 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1259 if (key.objectid < bytenr ||
1260 (key.type != BTRFS_EXTENT_ITEM_KEY &&
1261 key.type != BTRFS_METADATA_ITEM_KEY)) {
1266 bytenr = key.objectid;
1267 if (key.type == BTRFS_METADATA_ITEM_KEY) {
1268 num_bytes = extent_root->nodesize;
1270 num_bytes = key.offset;
1273 if (num_bytes == 0) {
1274 error("extent length 0 at bytenr %llu key type %d",
1275 (unsigned long long)bytenr, key.type);
1280 if (btrfs_item_size_nr(leaf, path->slots[0]) > sizeof(*ei)) {
1281 ei = btrfs_item_ptr(leaf, path->slots[0],
1282 struct btrfs_extent_item);
1283 if (btrfs_extent_flags(leaf, ei) &
1284 BTRFS_EXTENT_FLAG_TREE_BLOCK) {
1285 ret = add_extent(bytenr, num_bytes, metadump,
1288 error("unable to add block %llu: %d",
1289 (unsigned long long)bytenr, ret);
1294 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1295 ret = is_tree_block(extent_root, path, bytenr);
1297 error("failed to check tree block %llu: %d",
1298 (unsigned long long)bytenr, ret);
1303 ret = add_extent(bytenr, num_bytes, metadump,
1306 error("unable to add block %llu: %d",
1307 (unsigned long long)bytenr, ret);
1314 "either extent tree is corrupted or you haven't built with V0 support");
1319 bytenr += num_bytes;
1322 btrfs_release_path(path);
1327 static int create_metadump(const char *input, FILE *out, int num_threads,
1328 int compress_level, int sanitize, int walk_trees)
1330 struct btrfs_root *root;
1331 struct btrfs_path path;
1332 struct metadump_struct metadump;
1336 root = open_ctree(input, 0, 0);
1338 error("open ctree failed");
1342 ret = metadump_init(&metadump, root, out, num_threads,
1343 compress_level, sanitize);
1345 error("failed to initialize metadump: %d", ret);
1350 ret = add_extent(BTRFS_SUPER_INFO_OFFSET, BTRFS_SUPER_INFO_SIZE,
1353 error("unable to add metadata: %d", ret);
1358 btrfs_init_path(&path);
1361 ret = copy_tree_blocks(root, root->fs_info->chunk_root->node,
1368 ret = copy_tree_blocks(root, root->fs_info->tree_root->node,
1375 ret = copy_from_extent_tree(&metadump, &path);
1382 ret = copy_log_trees(root, &metadump, &path);
1388 ret = copy_space_cache(root, &metadump, &path);
1390 ret = flush_pending(&metadump, 1);
1394 error("failed to flush pending data: %d", ret);
1397 metadump_destroy(&metadump, num_threads);
1399 btrfs_release_path(&path);
1400 ret = close_ctree(root);
1401 return err ? err : ret;
1404 static void update_super_old(u8 *buffer)
1406 struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1407 struct btrfs_chunk *chunk;
1408 struct btrfs_disk_key *key;
1409 u32 sectorsize = btrfs_super_sectorsize(super);
1410 u64 flags = btrfs_super_flags(super);
1412 flags |= BTRFS_SUPER_FLAG_METADUMP;
1413 btrfs_set_super_flags(super, flags);
1415 key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1416 chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1417 sizeof(struct btrfs_disk_key));
1419 btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1420 btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1421 btrfs_set_disk_key_offset(key, 0);
1423 btrfs_set_stack_chunk_length(chunk, (u64)-1);
1424 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1425 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1426 btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1427 btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1428 btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1429 btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1430 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1431 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1432 chunk->stripe.devid = super->dev_item.devid;
1433 btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1434 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1435 btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1436 csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1439 static int update_super(struct mdrestore_struct *mdres, u8 *buffer)
1441 struct btrfs_super_block *super = (struct btrfs_super_block *)buffer;
1442 struct btrfs_chunk *chunk;
1443 struct btrfs_disk_key *disk_key;
1444 struct btrfs_key key;
1445 u64 flags = btrfs_super_flags(super);
1446 u32 new_array_size = 0;
1449 u8 *ptr, *write_ptr;
1450 int old_num_stripes;
1452 write_ptr = ptr = super->sys_chunk_array;
1453 array_size = btrfs_super_sys_array_size(super);
1455 while (cur < array_size) {
1456 disk_key = (struct btrfs_disk_key *)ptr;
1457 btrfs_disk_key_to_cpu(&key, disk_key);
1459 new_array_size += sizeof(*disk_key);
1460 memmove(write_ptr, ptr, sizeof(*disk_key));
1462 write_ptr += sizeof(*disk_key);
1463 ptr += sizeof(*disk_key);
1464 cur += sizeof(*disk_key);
1466 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1467 u64 type, physical, physical_dup, size = 0;
1469 chunk = (struct btrfs_chunk *)ptr;
1470 old_num_stripes = btrfs_stack_chunk_num_stripes(chunk);
1471 chunk = (struct btrfs_chunk *)write_ptr;
1473 memmove(write_ptr, ptr, sizeof(*chunk));
1474 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1475 type = btrfs_stack_chunk_type(chunk);
1476 if (type & BTRFS_BLOCK_GROUP_DUP) {
1477 new_array_size += sizeof(struct btrfs_stripe);
1478 write_ptr += sizeof(struct btrfs_stripe);
1480 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1481 btrfs_set_stack_chunk_type(chunk,
1482 BTRFS_BLOCK_GROUP_SYSTEM);
1484 chunk->stripe.devid = super->dev_item.devid;
1485 physical = logical_to_physical(mdres, key.offset,
1486 &size, &physical_dup);
1487 if (size != (u64)-1)
1488 btrfs_set_stack_stripe_offset(&chunk->stripe,
1490 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid,
1492 new_array_size += sizeof(*chunk);
1494 error("bogus key in the sys array %d", key.type);
1497 write_ptr += sizeof(*chunk);
1498 ptr += btrfs_chunk_item_size(old_num_stripes);
1499 cur += btrfs_chunk_item_size(old_num_stripes);
1502 if (mdres->clear_space_cache)
1503 btrfs_set_super_cache_generation(super, 0);
1505 flags |= BTRFS_SUPER_FLAG_METADUMP_V2;
1506 btrfs_set_super_flags(super, flags);
1507 btrfs_set_super_sys_array_size(super, new_array_size);
1508 csum_block(buffer, BTRFS_SUPER_INFO_SIZE);
1513 static struct extent_buffer *alloc_dummy_eb(u64 bytenr, u32 size)
1515 struct extent_buffer *eb;
1517 eb = calloc(1, sizeof(struct extent_buffer) + size);
1526 static void truncate_item(struct extent_buffer *eb, int slot, u32 new_size)
1528 struct btrfs_item *item;
1536 old_size = btrfs_item_size_nr(eb, slot);
1537 if (old_size == new_size)
1540 nritems = btrfs_header_nritems(eb);
1541 data_end = btrfs_item_offset_nr(eb, nritems - 1);
1543 old_data_start = btrfs_item_offset_nr(eb, slot);
1544 size_diff = old_size - new_size;
1546 for (i = slot; i < nritems; i++) {
1548 item = btrfs_item_nr(i);
1549 ioff = btrfs_item_offset(eb, item);
1550 btrfs_set_item_offset(eb, item, ioff + size_diff);
1553 memmove_extent_buffer(eb, btrfs_leaf_data(eb) + data_end + size_diff,
1554 btrfs_leaf_data(eb) + data_end,
1555 old_data_start + new_size - data_end);
1556 item = btrfs_item_nr(slot);
1557 btrfs_set_item_size(eb, item, new_size);
1560 static int fixup_chunk_tree_block(struct mdrestore_struct *mdres,
1561 struct async_work *async, u8 *buffer,
1564 struct extent_buffer *eb;
1565 size_t size_left = size;
1566 u64 bytenr = async->start;
1569 if (size_left % mdres->nodesize)
1572 eb = alloc_dummy_eb(bytenr, mdres->nodesize);
1578 memcpy(eb->data, buffer, mdres->nodesize);
1580 if (btrfs_header_bytenr(eb) != bytenr)
1582 if (memcmp(mdres->fsid,
1583 eb->data + offsetof(struct btrfs_header, fsid),
1587 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID)
1590 if (btrfs_header_level(eb) != 0)
1593 for (i = 0; i < btrfs_header_nritems(eb); i++) {
1594 struct btrfs_chunk *chunk;
1595 struct btrfs_key key;
1596 u64 type, physical, physical_dup, size = (u64)-1;
1598 btrfs_item_key_to_cpu(eb, &key, i);
1599 if (key.type != BTRFS_CHUNK_ITEM_KEY)
1603 physical = logical_to_physical(mdres, key.offset,
1604 &size, &physical_dup);
1607 truncate_item(eb, i, sizeof(*chunk));
1608 chunk = btrfs_item_ptr(eb, i, struct btrfs_chunk);
1611 /* Zero out the RAID profile */
1612 type = btrfs_chunk_type(eb, chunk);
1613 type &= (BTRFS_BLOCK_GROUP_DATA |
1614 BTRFS_BLOCK_GROUP_SYSTEM |
1615 BTRFS_BLOCK_GROUP_METADATA |
1616 BTRFS_BLOCK_GROUP_DUP);
1617 btrfs_set_chunk_type(eb, chunk, type);
1620 btrfs_set_chunk_num_stripes(eb, chunk, 1);
1621 btrfs_set_chunk_sub_stripes(eb, chunk, 0);
1622 btrfs_set_stripe_devid_nr(eb, chunk, 0, mdres->devid);
1623 if (size != (u64)-1)
1624 btrfs_set_stripe_offset_nr(eb, chunk, 0,
1626 /* update stripe 2 offset */
1628 btrfs_set_stripe_offset_nr(eb, chunk, 1,
1631 write_extent_buffer(eb, mdres->uuid,
1632 (unsigned long)btrfs_stripe_dev_uuid_nr(
1636 memcpy(buffer, eb->data, eb->len);
1637 csum_block(buffer, eb->len);
1639 size_left -= mdres->nodesize;
1640 buffer += mdres->nodesize;
1641 bytenr += mdres->nodesize;
1648 static void write_backup_supers(int fd, u8 *buf)
1650 struct btrfs_super_block *super = (struct btrfs_super_block *)buf;
1657 if (fstat(fd, &st)) {
1659 "cannot stat restore point, won't be able to write backup supers: %s",
1664 size = btrfs_device_size(fd, &st);
1666 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1667 bytenr = btrfs_sb_offset(i);
1668 if (bytenr + BTRFS_SUPER_INFO_SIZE > size)
1670 btrfs_set_super_bytenr(super, bytenr);
1671 csum_block(buf, BTRFS_SUPER_INFO_SIZE);
1672 ret = pwrite64(fd, buf, BTRFS_SUPER_INFO_SIZE, bytenr);
1673 if (ret < BTRFS_SUPER_INFO_SIZE) {
1676 "problem writing out backup super block %d: %s",
1677 i, strerror(errno));
1679 error("short write writing out backup super block");
1685 static void *restore_worker(void *data)
1687 struct mdrestore_struct *mdres = (struct mdrestore_struct *)data;
1688 struct async_work *async;
1694 int compress_size = MAX_PENDING_SIZE * 4;
1696 outfd = fileno(mdres->out);
1697 buffer = malloc(compress_size);
1699 error("not enough memory for restore worker buffer");
1700 pthread_mutex_lock(&mdres->mutex);
1702 mdres->error = -ENOMEM;
1703 pthread_mutex_unlock(&mdres->mutex);
1708 u64 bytenr, physical_dup;
1712 pthread_mutex_lock(&mdres->mutex);
1713 while (!mdres->nodesize || list_empty(&mdres->list)) {
1715 pthread_mutex_unlock(&mdres->mutex);
1718 pthread_cond_wait(&mdres->cond, &mdres->mutex);
1720 async = list_entry(mdres->list.next, struct async_work, list);
1721 list_del_init(&async->list);
1722 pthread_mutex_unlock(&mdres->mutex);
1724 if (mdres->compress_method == COMPRESS_ZLIB) {
1725 size = compress_size;
1726 ret = uncompress(buffer, (unsigned long *)&size,
1727 async->buffer, async->bufsize);
1729 error("decompressiion failed with %d", ret);
1734 outbuf = async->buffer;
1735 size = async->bufsize;
1738 if (!mdres->multi_devices) {
1739 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1740 if (mdres->old_restore) {
1741 update_super_old(outbuf);
1743 ret = update_super(mdres, outbuf);
1747 } else if (!mdres->old_restore) {
1748 ret = fixup_chunk_tree_block(mdres, async, outbuf, size);
1754 if (!mdres->fixup_offset) {
1756 u64 chunk_size = size;
1758 if (!mdres->multi_devices && !mdres->old_restore)
1759 bytenr = logical_to_physical(mdres,
1760 async->start + offset,
1764 bytenr = async->start + offset;
1766 ret = pwrite64(outfd, outbuf+offset, chunk_size,
1768 if (ret != chunk_size)
1772 ret = pwrite64(outfd, outbuf+offset,
1775 if (ret != chunk_size)
1779 offset += chunk_size;
1784 error("unable to write to device: %s",
1788 error("short write");
1792 } else if (async->start != BTRFS_SUPER_INFO_OFFSET) {
1793 ret = write_data_to_disk(mdres->info, outbuf, async->start, size, 0);
1795 error("failed to write data");
1801 /* backup super blocks are already there at fixup_offset stage */
1802 if (!mdres->multi_devices && async->start == BTRFS_SUPER_INFO_OFFSET)
1803 write_backup_supers(outfd, outbuf);
1805 pthread_mutex_lock(&mdres->mutex);
1806 if (err && !mdres->error)
1809 pthread_mutex_unlock(&mdres->mutex);
1811 free(async->buffer);
1819 static void mdrestore_destroy(struct mdrestore_struct *mdres, int num_threads)
1824 while ((n = rb_first(&mdres->chunk_tree))) {
1825 struct fs_chunk *entry;
1827 entry = rb_entry(n, struct fs_chunk, l);
1828 rb_erase(n, &mdres->chunk_tree);
1829 rb_erase(&entry->p, &mdres->physical_tree);
1832 pthread_mutex_lock(&mdres->mutex);
1834 pthread_cond_broadcast(&mdres->cond);
1835 pthread_mutex_unlock(&mdres->mutex);
1837 for (i = 0; i < num_threads; i++)
1838 pthread_join(mdres->threads[i], NULL);
1840 pthread_cond_destroy(&mdres->cond);
1841 pthread_mutex_destroy(&mdres->mutex);
1842 free(mdres->threads);
1845 static int mdrestore_init(struct mdrestore_struct *mdres,
1846 FILE *in, FILE *out, int old_restore,
1847 int num_threads, int fixup_offset,
1848 struct btrfs_fs_info *info, int multi_devices)
1852 memset(mdres, 0, sizeof(*mdres));
1853 pthread_cond_init(&mdres->cond, NULL);
1854 pthread_mutex_init(&mdres->mutex, NULL);
1855 INIT_LIST_HEAD(&mdres->list);
1856 INIT_LIST_HEAD(&mdres->overlapping_chunks);
1859 mdres->old_restore = old_restore;
1860 mdres->chunk_tree.rb_node = NULL;
1861 mdres->fixup_offset = fixup_offset;
1863 mdres->multi_devices = multi_devices;
1864 mdres->clear_space_cache = 0;
1865 mdres->last_physical_offset = 0;
1866 mdres->alloced_chunks = 0;
1871 mdres->num_threads = num_threads;
1872 mdres->threads = calloc(num_threads, sizeof(pthread_t));
1873 if (!mdres->threads)
1875 for (i = 0; i < num_threads; i++) {
1876 ret = pthread_create(mdres->threads + i, NULL, restore_worker,
1879 /* pthread_create returns errno directly */
1885 mdrestore_destroy(mdres, i + 1);
1889 static int fill_mdres_info(struct mdrestore_struct *mdres,
1890 struct async_work *async)
1892 struct btrfs_super_block *super;
1897 /* We've already been initialized */
1898 if (mdres->nodesize)
1901 if (mdres->compress_method == COMPRESS_ZLIB) {
1902 size_t size = MAX_PENDING_SIZE * 2;
1904 buffer = malloc(MAX_PENDING_SIZE * 2);
1907 ret = uncompress(buffer, (unsigned long *)&size,
1908 async->buffer, async->bufsize);
1910 error("decompressiion failed with %d", ret);
1916 outbuf = async->buffer;
1919 super = (struct btrfs_super_block *)outbuf;
1920 mdres->nodesize = btrfs_super_nodesize(super);
1921 memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
1922 memcpy(mdres->uuid, super->dev_item.uuid,
1924 mdres->devid = le64_to_cpu(super->dev_item.devid);
1929 static int add_cluster(struct meta_cluster *cluster,
1930 struct mdrestore_struct *mdres, u64 *next)
1932 struct meta_cluster_item *item;
1933 struct meta_cluster_header *header = &cluster->header;
1934 struct async_work *async;
1939 mdres->compress_method = header->compress;
1941 bytenr = le64_to_cpu(header->bytenr) + BLOCK_SIZE;
1942 nritems = le32_to_cpu(header->nritems);
1943 for (i = 0; i < nritems; i++) {
1944 item = &cluster->items[i];
1945 async = calloc(1, sizeof(*async));
1947 error("not enough memory for async data");
1950 async->start = le64_to_cpu(item->bytenr);
1951 async->bufsize = le32_to_cpu(item->size);
1952 async->buffer = malloc(async->bufsize);
1953 if (!async->buffer) {
1954 error("not enough memory for async buffer");
1958 ret = fread(async->buffer, async->bufsize, 1, mdres->in);
1960 error("unable to read buffer: %s", strerror(errno));
1961 free(async->buffer);
1965 bytenr += async->bufsize;
1967 pthread_mutex_lock(&mdres->mutex);
1968 if (async->start == BTRFS_SUPER_INFO_OFFSET) {
1969 ret = fill_mdres_info(mdres, async);
1971 error("unable to set up restore state");
1972 pthread_mutex_unlock(&mdres->mutex);
1973 free(async->buffer);
1978 list_add_tail(&async->list, &mdres->list);
1980 pthread_cond_signal(&mdres->cond);
1981 pthread_mutex_unlock(&mdres->mutex);
1983 if (bytenr & BLOCK_MASK) {
1984 char buffer[BLOCK_MASK];
1985 size_t size = BLOCK_SIZE - (bytenr & BLOCK_MASK);
1988 ret = fread(buffer, size, 1, mdres->in);
1990 error("failed to read buffer: %s", strerror(errno));
1998 static int wait_for_worker(struct mdrestore_struct *mdres)
2002 pthread_mutex_lock(&mdres->mutex);
2004 while (!ret && mdres->num_items > 0) {
2005 struct timespec ts = {
2007 .tv_nsec = 10000000,
2009 pthread_mutex_unlock(&mdres->mutex);
2010 nanosleep(&ts, NULL);
2011 pthread_mutex_lock(&mdres->mutex);
2014 pthread_mutex_unlock(&mdres->mutex);
2018 static int read_chunk_block(struct mdrestore_struct *mdres, u8 *buffer,
2019 u64 bytenr, u64 item_bytenr, u32 bufsize,
2022 struct extent_buffer *eb;
2026 eb = alloc_dummy_eb(bytenr, mdres->nodesize);
2032 while (item_bytenr != bytenr) {
2033 buffer += mdres->nodesize;
2034 item_bytenr += mdres->nodesize;
2037 memcpy(eb->data, buffer, mdres->nodesize);
2038 if (btrfs_header_bytenr(eb) != bytenr) {
2039 error("eb bytenr does not match found bytenr: %llu != %llu",
2040 (unsigned long long)btrfs_header_bytenr(eb),
2041 (unsigned long long)bytenr);
2046 if (memcmp(mdres->fsid, eb->data + offsetof(struct btrfs_header, fsid),
2048 error("filesystem UUID of eb %llu does not match",
2049 (unsigned long long)bytenr);
2054 if (btrfs_header_owner(eb) != BTRFS_CHUNK_TREE_OBJECTID) {
2055 error("wrong eb %llu owner %llu",
2056 (unsigned long long)bytenr,
2057 (unsigned long long)btrfs_header_owner(eb));
2062 for (i = 0; i < btrfs_header_nritems(eb); i++) {
2063 struct btrfs_chunk *chunk;
2064 struct fs_chunk *fs_chunk;
2065 struct btrfs_key key;
2068 if (btrfs_header_level(eb)) {
2069 u64 blockptr = btrfs_node_blockptr(eb, i);
2071 ret = search_for_chunk_blocks(mdres, blockptr,
2078 /* Yay a leaf! We loves leafs! */
2079 btrfs_item_key_to_cpu(eb, &key, i);
2080 if (key.type != BTRFS_CHUNK_ITEM_KEY)
2083 fs_chunk = malloc(sizeof(struct fs_chunk));
2085 error("not enough memory to allocate chunk");
2089 memset(fs_chunk, 0, sizeof(*fs_chunk));
2090 chunk = btrfs_item_ptr(eb, i, struct btrfs_chunk);
2092 fs_chunk->logical = key.offset;
2093 fs_chunk->physical = btrfs_stripe_offset_nr(eb, chunk, 0);
2094 fs_chunk->bytes = btrfs_chunk_length(eb, chunk);
2095 INIT_LIST_HEAD(&fs_chunk->list);
2096 if (tree_search(&mdres->physical_tree, &fs_chunk->p,
2097 physical_cmp, 1) != NULL)
2098 list_add(&fs_chunk->list, &mdres->overlapping_chunks);
2100 tree_insert(&mdres->physical_tree, &fs_chunk->p,
2103 type = btrfs_chunk_type(eb, chunk);
2104 if (type & BTRFS_BLOCK_GROUP_DUP) {
2105 fs_chunk->physical_dup =
2106 btrfs_stripe_offset_nr(eb, chunk, 1);
2109 if (fs_chunk->physical_dup + fs_chunk->bytes >
2110 mdres->last_physical_offset)
2111 mdres->last_physical_offset = fs_chunk->physical_dup +
2113 else if (fs_chunk->physical + fs_chunk->bytes >
2114 mdres->last_physical_offset)
2115 mdres->last_physical_offset = fs_chunk->physical +
2117 mdres->alloced_chunks += fs_chunk->bytes;
2118 /* in dup case, fs_chunk->bytes should add twice */
2119 if (fs_chunk->physical_dup)
2120 mdres->alloced_chunks += fs_chunk->bytes;
2121 tree_insert(&mdres->chunk_tree, &fs_chunk->l, chunk_cmp);
2128 /* If you have to ask you aren't worthy */
2129 static int search_for_chunk_blocks(struct mdrestore_struct *mdres,
2130 u64 search, u64 cluster_bytenr)
2132 struct meta_cluster *cluster;
2133 struct meta_cluster_header *header;
2134 struct meta_cluster_item *item;
2135 u64 current_cluster = cluster_bytenr, bytenr;
2137 u32 bufsize, nritems, i;
2138 u32 max_size = MAX_PENDING_SIZE * 2;
2139 u8 *buffer, *tmp = NULL;
2142 cluster = malloc(BLOCK_SIZE);
2144 error("not enough memory for cluster");
2148 buffer = malloc(max_size);
2150 error("not enough memory for buffer");
2155 if (mdres->compress_method == COMPRESS_ZLIB) {
2156 tmp = malloc(max_size);
2158 error("not enough memory for buffer");
2165 bytenr = current_cluster;
2167 if (fseek(mdres->in, current_cluster, SEEK_SET)) {
2168 error("seek failed: %s\n", strerror(errno));
2173 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2175 if (cluster_bytenr != 0) {
2177 current_cluster = 0;
2182 "unknown state after reading cluster at %llu, probably crrupted data",
2186 } else if (ret < 0) {
2187 error("unable to read image at %llu: %s",
2188 (unsigned long long)cluster_bytenr,
2194 header = &cluster->header;
2195 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2196 le64_to_cpu(header->bytenr) != current_cluster) {
2197 error("bad header in metadump image");
2202 bytenr += BLOCK_SIZE;
2203 nritems = le32_to_cpu(header->nritems);
2204 for (i = 0; i < nritems; i++) {
2207 item = &cluster->items[i];
2208 bufsize = le32_to_cpu(item->size);
2209 item_bytenr = le64_to_cpu(item->bytenr);
2211 if (bufsize > max_size) {
2212 error("item %u too big: %u > %u", i, bufsize,
2218 if (mdres->compress_method == COMPRESS_ZLIB) {
2219 ret = fread(tmp, bufsize, 1, mdres->in);
2221 error("read error: %s", strerror(errno));
2227 ret = uncompress(buffer,
2228 (unsigned long *)&size, tmp,
2231 error("decompressiion failed with %d",
2237 ret = fread(buffer, bufsize, 1, mdres->in);
2239 error("read error: %s",
2248 if (item_bytenr <= search &&
2249 item_bytenr + size > search) {
2250 ret = read_chunk_block(mdres, buffer, search,
2264 if (bytenr & BLOCK_MASK)
2265 bytenr += BLOCK_SIZE - (bytenr & BLOCK_MASK);
2266 current_cluster = bytenr;
2275 static int build_chunk_tree(struct mdrestore_struct *mdres,
2276 struct meta_cluster *cluster)
2278 struct btrfs_super_block *super;
2279 struct meta_cluster_header *header;
2280 struct meta_cluster_item *item = NULL;
2281 u64 chunk_root_bytenr = 0;
2287 /* We can't seek with stdin so don't bother doing this */
2288 if (mdres->in == stdin)
2291 ret = fread(cluster, BLOCK_SIZE, 1, mdres->in);
2293 error("unable to read cluster: %s", strerror(errno));
2298 header = &cluster->header;
2299 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2300 le64_to_cpu(header->bytenr) != 0) {
2301 error("bad header in metadump image");
2305 bytenr += BLOCK_SIZE;
2306 mdres->compress_method = header->compress;
2307 nritems = le32_to_cpu(header->nritems);
2308 for (i = 0; i < nritems; i++) {
2309 item = &cluster->items[i];
2311 if (le64_to_cpu(item->bytenr) == BTRFS_SUPER_INFO_OFFSET)
2313 bytenr += le32_to_cpu(item->size);
2314 if (fseek(mdres->in, le32_to_cpu(item->size), SEEK_CUR)) {
2315 error("seek failed: %s\n", strerror(errno));
2320 if (!item || le64_to_cpu(item->bytenr) != BTRFS_SUPER_INFO_OFFSET) {
2321 error("did not find superblock at %llu",
2322 le64_to_cpu(item->bytenr));
2326 buffer = malloc(le32_to_cpu(item->size));
2328 error("not enough memory to allocate buffer");
2332 ret = fread(buffer, le32_to_cpu(item->size), 1, mdres->in);
2334 error("unable to read buffer: %s", strerror(errno));
2339 if (mdres->compress_method == COMPRESS_ZLIB) {
2340 size_t size = MAX_PENDING_SIZE * 2;
2343 tmp = malloc(MAX_PENDING_SIZE * 2);
2348 ret = uncompress(tmp, (unsigned long *)&size,
2349 buffer, le32_to_cpu(item->size));
2351 error("decompressiion failed with %d", ret);
2360 pthread_mutex_lock(&mdres->mutex);
2361 super = (struct btrfs_super_block *)buffer;
2362 chunk_root_bytenr = btrfs_super_chunk_root(super);
2363 mdres->nodesize = btrfs_super_nodesize(super);
2364 memcpy(mdres->fsid, super->fsid, BTRFS_FSID_SIZE);
2365 memcpy(mdres->uuid, super->dev_item.uuid,
2367 mdres->devid = le64_to_cpu(super->dev_item.devid);
2369 pthread_mutex_unlock(&mdres->mutex);
2371 return search_for_chunk_blocks(mdres, chunk_root_bytenr, 0);
2374 static int range_contains_super(u64 physical, u64 bytes)
2379 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2380 super_bytenr = btrfs_sb_offset(i);
2381 if (super_bytenr >= physical &&
2382 super_bytenr < physical + bytes)
2389 static void remap_overlapping_chunks(struct mdrestore_struct *mdres)
2391 struct fs_chunk *fs_chunk;
2393 while (!list_empty(&mdres->overlapping_chunks)) {
2394 fs_chunk = list_first_entry(&mdres->overlapping_chunks,
2395 struct fs_chunk, list);
2396 list_del_init(&fs_chunk->list);
2397 if (range_contains_super(fs_chunk->physical,
2400 "remapping a chunk that had a super mirror inside of it, clearing space cache so we don't end up with corruption");
2401 mdres->clear_space_cache = 1;
2403 fs_chunk->physical = mdres->last_physical_offset;
2404 tree_insert(&mdres->physical_tree, &fs_chunk->p, physical_cmp);
2405 mdres->last_physical_offset += fs_chunk->bytes;
2409 static int fixup_devices(struct btrfs_fs_info *fs_info,
2410 struct mdrestore_struct *mdres, off_t dev_size)
2412 struct btrfs_trans_handle *trans;
2413 struct btrfs_dev_item *dev_item;
2414 struct btrfs_path *path;
2415 struct extent_buffer *leaf;
2416 struct btrfs_root *root = fs_info->chunk_root;
2417 struct btrfs_key key;
2418 u64 devid, cur_devid;
2421 path = btrfs_alloc_path();
2423 error("not enough memory to allocate path");
2427 trans = btrfs_start_transaction(fs_info->tree_root, 1);
2428 if (IS_ERR(trans)) {
2429 error("cannot starting transaction %ld", PTR_ERR(trans));
2430 btrfs_free_path(path);
2431 return PTR_ERR(trans);
2434 dev_item = &fs_info->super_copy->dev_item;
2436 devid = btrfs_stack_device_id(dev_item);
2438 btrfs_set_stack_device_total_bytes(dev_item, dev_size);
2439 btrfs_set_stack_device_bytes_used(dev_item, mdres->alloced_chunks);
2441 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2442 key.type = BTRFS_DEV_ITEM_KEY;
2446 ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2448 error("search failed: %d", ret);
2453 leaf = path->nodes[0];
2454 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
2455 ret = btrfs_next_leaf(root, path);
2457 error("cannot go to next leaf %d", ret);
2464 leaf = path->nodes[0];
2467 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
2468 if (key.type > BTRFS_DEV_ITEM_KEY)
2470 if (key.type != BTRFS_DEV_ITEM_KEY) {
2475 dev_item = btrfs_item_ptr(leaf, path->slots[0],
2476 struct btrfs_dev_item);
2477 cur_devid = btrfs_device_id(leaf, dev_item);
2478 if (devid != cur_devid) {
2479 ret = btrfs_del_item(trans, root, path);
2481 error("cannot delete item: %d", ret);
2484 btrfs_release_path(path);
2488 btrfs_set_device_total_bytes(leaf, dev_item, dev_size);
2489 btrfs_set_device_bytes_used(leaf, dev_item,
2490 mdres->alloced_chunks);
2491 btrfs_mark_buffer_dirty(leaf);
2495 btrfs_free_path(path);
2496 ret = btrfs_commit_transaction(trans, fs_info->tree_root);
2498 error("unable to commit transaction: %d", ret);
2504 static int restore_metadump(const char *input, FILE *out, int old_restore,
2505 int num_threads, int fixup_offset,
2506 const char *target, int multi_devices)
2508 struct meta_cluster *cluster = NULL;
2509 struct meta_cluster_header *header;
2510 struct mdrestore_struct mdrestore;
2511 struct btrfs_fs_info *info = NULL;
2516 if (!strcmp(input, "-")) {
2519 in = fopen(input, "r");
2521 error("unable to open metadump image: %s",
2527 /* NOTE: open with write mode */
2529 info = open_ctree_fs_info(target, 0, 0, 0,
2531 OPEN_CTREE_RESTORE |
2532 OPEN_CTREE_PARTIAL);
2534 error("open ctree failed");
2540 cluster = malloc(BLOCK_SIZE);
2542 error("not enough memory for cluster");
2547 ret = mdrestore_init(&mdrestore, in, out, old_restore, num_threads,
2548 fixup_offset, info, multi_devices);
2550 error("failed to intialize metadata restore state: %d", ret);
2551 goto failed_cluster;
2554 if (!multi_devices && !old_restore) {
2555 ret = build_chunk_tree(&mdrestore, cluster);
2558 if (!list_empty(&mdrestore.overlapping_chunks))
2559 remap_overlapping_chunks(&mdrestore);
2562 if (in != stdin && fseek(in, 0, SEEK_SET)) {
2563 error("seek failed: %s\n", strerror(errno));
2567 while (!mdrestore.error) {
2568 ret = fread(cluster, BLOCK_SIZE, 1, in);
2572 header = &cluster->header;
2573 if (le64_to_cpu(header->magic) != HEADER_MAGIC ||
2574 le64_to_cpu(header->bytenr) != bytenr) {
2575 error("bad header in metadump image");
2579 ret = add_cluster(cluster, &mdrestore, &bytenr);
2581 error("failed to add cluster: %d", ret);
2585 ret = wait_for_worker(&mdrestore);
2587 if (!ret && !multi_devices && !old_restore) {
2588 struct btrfs_root *root;
2591 root = open_ctree_fd(fileno(out), target, 0,
2592 OPEN_CTREE_PARTIAL |
2594 OPEN_CTREE_NO_DEVICES);
2596 error("open ctree failed in %s", target);
2600 info = root->fs_info;
2602 if (stat(target, &st)) {
2603 error("stat %s failed: %s", target, strerror(errno));
2604 close_ctree(info->chunk_root);
2609 ret = fixup_devices(info, &mdrestore, st.st_size);
2610 close_ctree(info->chunk_root);
2615 mdrestore_destroy(&mdrestore, num_threads);
2619 if (fixup_offset && info)
2620 close_ctree(info->chunk_root);
2627 static int update_disk_super_on_device(struct btrfs_fs_info *info,
2628 const char *other_dev, u64 cur_devid)
2630 struct btrfs_key key;
2631 struct extent_buffer *leaf;
2632 struct btrfs_path path;
2633 struct btrfs_dev_item *dev_item;
2634 struct btrfs_super_block *disk_super;
2635 char dev_uuid[BTRFS_UUID_SIZE];
2636 char fs_uuid[BTRFS_UUID_SIZE];
2637 u64 devid, type, io_align, io_width;
2638 u64 sector_size, total_bytes, bytes_used;
2639 char buf[BTRFS_SUPER_INFO_SIZE];
2643 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2644 key.type = BTRFS_DEV_ITEM_KEY;
2645 key.offset = cur_devid;
2647 btrfs_init_path(&path);
2648 ret = btrfs_search_slot(NULL, info->chunk_root, &key, &path, 0, 0);
2650 error("search key failed: %d", ret);
2655 leaf = path.nodes[0];
2656 dev_item = btrfs_item_ptr(leaf, path.slots[0],
2657 struct btrfs_dev_item);
2659 devid = btrfs_device_id(leaf, dev_item);
2660 if (devid != cur_devid) {
2661 error("devid mismatch: %llu != %llu",
2662 (unsigned long long)devid,
2663 (unsigned long long)cur_devid);
2668 type = btrfs_device_type(leaf, dev_item);
2669 io_align = btrfs_device_io_align(leaf, dev_item);
2670 io_width = btrfs_device_io_width(leaf, dev_item);
2671 sector_size = btrfs_device_sector_size(leaf, dev_item);
2672 total_bytes = btrfs_device_total_bytes(leaf, dev_item);
2673 bytes_used = btrfs_device_bytes_used(leaf, dev_item);
2674 read_extent_buffer(leaf, dev_uuid, (unsigned long)btrfs_device_uuid(dev_item), BTRFS_UUID_SIZE);
2675 read_extent_buffer(leaf, fs_uuid, (unsigned long)btrfs_device_fsid(dev_item), BTRFS_UUID_SIZE);
2677 btrfs_release_path(&path);
2679 printf("update disk super on %s devid=%llu\n", other_dev, devid);
2681 /* update other devices' super block */
2682 fp = open(other_dev, O_CREAT | O_RDWR, 0600);
2684 error("could not open %s: %s", other_dev, strerror(errno));
2689 memcpy(buf, info->super_copy, BTRFS_SUPER_INFO_SIZE);
2691 disk_super = (struct btrfs_super_block *)buf;
2692 dev_item = &disk_super->dev_item;
2694 btrfs_set_stack_device_type(dev_item, type);
2695 btrfs_set_stack_device_id(dev_item, devid);
2696 btrfs_set_stack_device_total_bytes(dev_item, total_bytes);
2697 btrfs_set_stack_device_bytes_used(dev_item, bytes_used);
2698 btrfs_set_stack_device_io_align(dev_item, io_align);
2699 btrfs_set_stack_device_io_width(dev_item, io_width);
2700 btrfs_set_stack_device_sector_size(dev_item, sector_size);
2701 memcpy(dev_item->uuid, dev_uuid, BTRFS_UUID_SIZE);
2702 memcpy(dev_item->fsid, fs_uuid, BTRFS_UUID_SIZE);
2703 csum_block((u8 *)buf, BTRFS_SUPER_INFO_SIZE);
2705 ret = pwrite64(fp, buf, BTRFS_SUPER_INFO_SIZE, BTRFS_SUPER_INFO_OFFSET);
2706 if (ret != BTRFS_SUPER_INFO_SIZE) {
2708 error("cannot write superblock: %s", strerror(ret));
2710 error("cannot write superblock");
2715 write_backup_supers(fp, (u8 *)buf);
2723 static void print_usage(int ret)
2725 printf("usage: btrfs-image [options] source target\n");
2726 printf("\t-r \trestore metadump image\n");
2727 printf("\t-c value\tcompression level (0 ~ 9)\n");
2728 printf("\t-t value\tnumber of threads (1 ~ 32)\n");
2729 printf("\t-o \tdon't mess with the chunk tree when restoring\n");
2730 printf("\t-s \tsanitize file names, use once to just use garbage, use twice if you want crc collisions\n");
2731 printf("\t-w \twalk all trees instead of using extent tree, do this if your extent tree is broken\n");
2732 printf("\t-m \trestore for multiple devices\n");
2734 printf("\tIn the dump mode, source is the btrfs device and target is the output file (use '-' for stdout).\n");
2735 printf("\tIn the restore mode, source is the dumped image and target is the btrfs device/file.\n");
2739 int main(int argc, char *argv[])
2743 u64 num_threads = 0;
2744 u64 compress_level = 0;
2746 int old_restore = 0;
2748 int multi_devices = 0;
2752 int usage_error = 0;
2756 static const struct option long_options[] = {
2757 { "help", no_argument, NULL, GETOPT_VAL_HELP},
2758 { NULL, 0, NULL, 0 }
2760 int c = getopt_long(argc, argv, "rc:t:oswm", long_options, NULL);
2768 num_threads = arg_strtou64(optarg);
2769 if (num_threads > 32) {
2770 error("number of threads out of range: %llu",
2771 (unsigned long long)num_threads);
2776 compress_level = arg_strtou64(optarg);
2777 if (compress_level > 9) {
2778 error("compression level out of range: %llu",
2779 (unsigned long long)compress_level);
2796 case GETOPT_VAL_HELP:
2798 print_usage(c != GETOPT_VAL_HELP);
2803 if (check_argc_min(argc - optind, 2))
2806 dev_cnt = argc - optind - 1;
2811 "create and restore cannot be used at the same time");
2815 if (walk_trees || sanitize || compress_level) {
2817 "useing -w, -s, -c options for restore makes no sense");
2820 if (multi_devices && dev_cnt < 2) {
2821 error("not enough devices specified for -m option");
2824 if (!multi_devices && dev_cnt != 1) {
2825 error("accepts only 1 device without -m option");
2833 source = argv[optind];
2834 target = argv[optind + 1];
2836 if (create && !strcmp(target, "-")) {
2839 out = fopen(target, "w+");
2841 error("unable to create target file %s", target);
2846 if (compress_level > 0 || create == 0) {
2847 if (num_threads == 0) {
2848 long tmp = sysconf(_SC_NPROCESSORS_ONLN);
2859 ret = check_mounted(source);
2861 warning("unable to check mount status of: %s",
2864 warning("%s already mounted, results may be inaccurate",
2868 ret = create_metadump(source, out, num_threads,
2869 compress_level, sanitize, walk_trees);
2871 ret = restore_metadump(source, out, old_restore, num_threads,
2872 0, target, multi_devices);
2875 error("%s failed: %s", (create) ? "create" : "restore",
2880 /* extended support for multiple devices */
2881 if (!create && multi_devices) {
2882 struct btrfs_fs_info *info;
2886 info = open_ctree_fs_info(target, 0, 0, 0,
2887 OPEN_CTREE_PARTIAL |
2888 OPEN_CTREE_RESTORE);
2890 error("open ctree failed at %s", target);
2894 total_devs = btrfs_super_num_devices(info->super_copy);
2895 if (total_devs != dev_cnt) {
2896 error("it needs %llu devices but has only %d",
2897 total_devs, dev_cnt);
2898 close_ctree(info->chunk_root);
2902 /* update super block on other disks */
2903 for (i = 2; i <= dev_cnt; i++) {
2904 ret = update_disk_super_on_device(info,
2905 argv[optind + i], (u64)i);
2907 error("update disk superblock failed devid %d: %d",
2909 close_ctree(info->chunk_root);
2914 close_ctree(info->chunk_root);
2916 /* fix metadata block to map correct chunk */
2917 ret = restore_metadump(source, out, 0, num_threads, 1,
2920 error("unable to fixup metadump: %d", ret);
2925 if (out == stdout) {
2929 if (ret && create) {
2932 unlink_ret = unlink(target);
2934 error("unlink output file %s failed: %s",
2935 target, strerror(errno));
2939 btrfs_close_all_devices();