2 * Copyright (C) 2007 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.
19 #include "kerncompat.h"
21 #include <sys/ioctl.h>
22 #include <sys/mount.h>
25 #include <sys/types.h>
29 #include <uuid/uuid.h>
30 #include <linux/limits.h>
36 #include "transaction.h"
39 #include "task-utils.h"
41 #include <ext2fs/ext2_fs.h>
42 #include <ext2fs/ext2fs.h>
43 #include <ext2fs/ext2_ext_attr.h>
45 #define INO_OFFSET (BTRFS_FIRST_FREE_OBJECTID - EXT2_ROOT_INO)
46 #define CONV_IMAGE_SUBVOL_OBJECTID BTRFS_FIRST_FREE_OBJECTID
49 * Compatibility code for e2fsprogs 1.41 which doesn't support RO compat flag
51 * Unlike normal RO compat flag, BIGALLOC affects how e2fsprogs check used
52 * space, and btrfs-convert heavily relies on it.
54 #ifdef HAVE_OLD_E2FSPROGS
55 #define EXT2FS_CLUSTER_RATIO(fs) (1)
56 #define EXT2_CLUSTERS_PER_GROUP(s) (EXT2_BLOCKS_PER_GROUP(s))
57 #define EXT2FS_B2C(fs, blk) (blk)
61 uint32_t max_copy_inodes;
62 uint32_t cur_copy_inodes;
63 struct task_info *info;
66 static void *print_copied_inodes(void *p)
68 struct task_ctx *priv = p;
69 const char work_indicator[] = { '.', 'o', 'O', 'o' };
72 task_period_start(priv->info, 1000 /* 1s */);
75 printf("copy inodes [%c] [%10d/%10d]\r",
76 work_indicator[count % 4], priv->cur_copy_inodes,
77 priv->max_copy_inodes);
79 task_period_wait(priv->info);
85 static int after_copied_inodes(void *p)
93 struct btrfs_convert_context;
94 struct btrfs_convert_operations {
96 int (*open_fs)(struct btrfs_convert_context *cctx, const char *devname);
97 int (*read_used_space)(struct btrfs_convert_context *cctx);
98 int (*copy_inodes)(struct btrfs_convert_context *cctx,
99 struct btrfs_root *root, int datacsum,
100 int packing, int noxattr, struct task_ctx *p);
101 void (*close_fs)(struct btrfs_convert_context *cctx);
104 static void init_convert_context(struct btrfs_convert_context *cctx)
106 cache_tree_init(&cctx->used);
107 cache_tree_init(&cctx->data_chunks);
108 cache_tree_init(&cctx->free);
111 static void clean_convert_context(struct btrfs_convert_context *cctx)
113 free_extent_cache_tree(&cctx->used);
114 free_extent_cache_tree(&cctx->data_chunks);
115 free_extent_cache_tree(&cctx->free);
118 static inline int copy_inodes(struct btrfs_convert_context *cctx,
119 struct btrfs_root *root, int datacsum,
120 int packing, int noxattr, struct task_ctx *p)
122 return cctx->convert_ops->copy_inodes(cctx, root, datacsum, packing,
126 static inline void convert_close_fs(struct btrfs_convert_context *cctx)
128 cctx->convert_ops->close_fs(cctx);
131 static int intersect_with_sb(u64 bytenr, u64 num_bytes)
136 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
137 offset = btrfs_sb_offset(i);
138 offset &= ~((u64)BTRFS_STRIPE_LEN - 1);
140 if (bytenr < offset + BTRFS_STRIPE_LEN &&
141 bytenr + num_bytes > offset)
147 static int convert_insert_dirent(struct btrfs_trans_handle *trans,
148 struct btrfs_root *root,
149 const char *name, size_t name_len,
150 u64 dir, u64 objectid,
151 u8 file_type, u64 index_cnt,
152 struct btrfs_inode_item *inode)
156 struct btrfs_key location = {
157 .objectid = objectid,
159 .type = BTRFS_INODE_ITEM_KEY,
162 ret = btrfs_insert_dir_item(trans, root, name, name_len,
163 dir, &location, file_type, index_cnt);
166 ret = btrfs_insert_inode_ref(trans, root, name, name_len,
167 objectid, dir, index_cnt);
170 inode_size = btrfs_stack_inode_size(inode) + name_len * 2;
171 btrfs_set_stack_inode_size(inode, inode_size);
176 static int read_disk_extent(struct btrfs_root *root, u64 bytenr,
177 u32 num_bytes, char *buffer)
180 struct btrfs_fs_devices *fs_devs = root->fs_info->fs_devices;
182 ret = pread(fs_devs->latest_bdev, buffer, num_bytes, bytenr);
183 if (ret != num_bytes)
192 static int csum_disk_extent(struct btrfs_trans_handle *trans,
193 struct btrfs_root *root,
194 u64 disk_bytenr, u64 num_bytes)
196 u32 blocksize = root->sectorsize;
201 buffer = malloc(blocksize);
204 for (offset = 0; offset < num_bytes; offset += blocksize) {
205 ret = read_disk_extent(root, disk_bytenr + offset,
209 ret = btrfs_csum_file_block(trans,
210 root->fs_info->csum_root,
211 disk_bytenr + num_bytes,
212 disk_bytenr + offset,
221 struct blk_iterate_data {
222 struct btrfs_trans_handle *trans;
223 struct btrfs_root *root;
224 struct btrfs_root *convert_root;
225 struct btrfs_inode_item *inode;
236 static void init_blk_iterate_data(struct blk_iterate_data *data,
237 struct btrfs_trans_handle *trans,
238 struct btrfs_root *root,
239 struct btrfs_inode_item *inode,
240 u64 objectid, int checksum)
242 struct btrfs_key key;
247 data->objectid = objectid;
248 data->first_block = 0;
249 data->disk_block = 0;
250 data->num_blocks = 0;
251 data->boundary = (u64)-1;
252 data->checksum = checksum;
255 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
256 key.type = BTRFS_ROOT_ITEM_KEY;
257 key.offset = (u64)-1;
258 data->convert_root = btrfs_read_fs_root(root->fs_info, &key);
259 /* Impossible as we just opened it before */
260 BUG_ON(!data->convert_root || IS_ERR(data->convert_root));
261 data->convert_ino = BTRFS_FIRST_FREE_OBJECTID + 1;
265 * Record a file extent in original filesystem into btrfs one.
266 * The special point is, old disk_block can point to a reserved range.
267 * So here, we don't use disk_block directly but search convert_root
268 * to get the real disk_bytenr.
270 static int record_file_blocks(struct blk_iterate_data *data,
271 u64 file_block, u64 disk_block, u64 num_blocks)
274 struct btrfs_root *root = data->root;
275 struct btrfs_root *convert_root = data->convert_root;
276 struct btrfs_path *path;
277 u64 file_pos = file_block * root->sectorsize;
278 u64 old_disk_bytenr = disk_block * root->sectorsize;
279 u64 num_bytes = num_blocks * root->sectorsize;
280 u64 cur_off = old_disk_bytenr;
282 /* Hole, pass it to record_file_extent directly */
283 if (old_disk_bytenr == 0)
284 return btrfs_record_file_extent(data->trans, root,
285 data->objectid, data->inode, file_pos, 0,
288 path = btrfs_alloc_path();
293 * Search real disk bytenr from convert root
295 while (cur_off < old_disk_bytenr + num_bytes) {
296 struct btrfs_key key;
297 struct btrfs_file_extent_item *fi;
298 struct extent_buffer *node;
300 u64 extent_disk_bytenr;
301 u64 extent_num_bytes;
302 u64 real_disk_bytenr;
305 key.objectid = data->convert_ino;
306 key.type = BTRFS_EXTENT_DATA_KEY;
307 key.offset = cur_off;
309 ret = btrfs_search_slot(NULL, convert_root, &key, path, 0, 0);
313 ret = btrfs_previous_item(convert_root, path,
315 BTRFS_EXTENT_DATA_KEY);
323 node = path->nodes[0];
324 slot = path->slots[0];
325 btrfs_item_key_to_cpu(node, &key, slot);
326 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY ||
327 key.objectid != data->convert_ino ||
328 key.offset > cur_off);
329 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
330 extent_disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
331 extent_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
332 BUG_ON(cur_off - key.offset >= extent_num_bytes);
333 btrfs_release_path(path);
335 if (extent_disk_bytenr)
336 real_disk_bytenr = cur_off - key.offset +
339 real_disk_bytenr = 0;
340 cur_len = min(key.offset + extent_num_bytes,
341 old_disk_bytenr + num_bytes) - cur_off;
342 ret = btrfs_record_file_extent(data->trans, data->root,
343 data->objectid, data->inode, file_pos,
344 real_disk_bytenr, cur_len);
351 * No need to care about csum
352 * As every byte of old fs image is calculated for csum, no
353 * need to waste CPU cycles now.
356 btrfs_free_path(path);
360 static int block_iterate_proc(u64 disk_block, u64 file_block,
361 struct blk_iterate_data *idata)
366 struct btrfs_root *root = idata->root;
367 struct btrfs_block_group_cache *cache;
368 u64 bytenr = disk_block * root->sectorsize;
370 sb_region = intersect_with_sb(bytenr, root->sectorsize);
371 do_barrier = sb_region || disk_block >= idata->boundary;
372 if ((idata->num_blocks > 0 && do_barrier) ||
373 (file_block > idata->first_block + idata->num_blocks) ||
374 (disk_block != idata->disk_block + idata->num_blocks)) {
375 if (idata->num_blocks > 0) {
376 ret = record_file_blocks(idata, idata->first_block,
381 idata->first_block += idata->num_blocks;
382 idata->num_blocks = 0;
384 if (file_block > idata->first_block) {
385 ret = record_file_blocks(idata, idata->first_block,
386 0, file_block - idata->first_block);
392 bytenr += BTRFS_STRIPE_LEN - 1;
393 bytenr &= ~((u64)BTRFS_STRIPE_LEN - 1);
395 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
397 bytenr = cache->key.objectid + cache->key.offset;
400 idata->first_block = file_block;
401 idata->disk_block = disk_block;
402 idata->boundary = bytenr / root->sectorsize;
409 static int create_image_file_range(struct btrfs_trans_handle *trans,
410 struct btrfs_root *root,
411 struct cache_tree *used,
412 struct btrfs_inode_item *inode,
413 u64 ino, u64 bytenr, u64 *ret_len,
416 struct cache_extent *cache;
417 struct btrfs_block_group_cache *bg_cache;
423 BUG_ON(bytenr != round_down(bytenr, root->sectorsize));
424 BUG_ON(len != round_down(len, root->sectorsize));
425 len = min_t(u64, len, BTRFS_MAX_EXTENT_SIZE);
428 * Skip sb ranges first
429 * [0, 1M), [sb_offset(1), +64K), [sb_offset(2), +64K].
431 * Or we will insert a hole into current image file, and later
432 * migrate block will fail as there is already a file extent.
434 if (bytenr < 1024 * 1024) {
435 *ret_len = 1024 * 1024 - bytenr;
438 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
439 u64 cur = btrfs_sb_offset(i);
441 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
442 *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
446 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
447 u64 cur = btrfs_sb_offset(i);
452 * May still need to go through file extent inserts
454 if (bytenr < cur && bytenr + len >= cur) {
455 len = min_t(u64, len, cur - bytenr);
461 * Drop out, no need to insert anything
463 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
464 *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
469 cache = search_cache_extent(used, bytenr);
471 if (cache->start <= bytenr) {
473 * |///////Used///////|
477 len = min_t(u64, len, cache->start + cache->size -
479 disk_bytenr = bytenr;
486 len = min(len, cache->start - bytenr);
501 /* Check if the range is in a data block group */
502 bg_cache = btrfs_lookup_block_group(root->fs_info, bytenr);
505 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
508 /* The extent should never cross block group boundary */
509 len = min_t(u64, len, bg_cache->key.objectid +
510 bg_cache->key.offset - bytenr);
513 BUG_ON(len != round_down(len, root->sectorsize));
514 ret = btrfs_record_file_extent(trans, root, ino, inode, bytenr,
520 ret = csum_disk_extent(trans, root, bytenr, len);
526 * Relocate old fs data in one reserved ranges
528 * Since all old fs data in reserved range is not covered by any chunk nor
529 * data extent, we don't need to handle any reference but add new
530 * extent/reference, which makes codes more clear
532 static int migrate_one_reserved_range(struct btrfs_trans_handle *trans,
533 struct btrfs_root *root,
534 struct cache_tree *used,
535 struct btrfs_inode_item *inode, int fd,
536 u64 ino, u64 start, u64 len, int datacsum)
540 u64 hole_start = start;
542 struct cache_extent *cache;
543 struct btrfs_key key;
544 struct extent_buffer *eb;
547 while (cur_off < start + len) {
548 cache = lookup_cache_extent(used, cur_off, cur_len);
551 cur_off = max(cache->start, cur_off);
552 cur_len = min(cache->start + cache->size, start + len) -
554 BUG_ON(cur_len < root->sectorsize);
556 /* reserve extent for the data */
557 ret = btrfs_reserve_extent(trans, root, cur_len, 0, 0, (u64)-1,
562 eb = malloc(sizeof(*eb) + cur_len);
568 ret = pread(fd, eb->data, cur_len, cur_off);
570 ret = (ret < 0 ? ret : -EIO);
574 eb->start = key.objectid;
575 eb->len = key.offset;
578 ret = write_and_map_eb(trans, root, eb);
583 /* Now handle extent item and file extent things */
584 ret = btrfs_record_file_extent(trans, root, ino, inode, cur_off,
585 key.objectid, key.offset);
588 /* Finally, insert csum items */
590 ret = csum_disk_extent(trans, root, key.objectid,
593 /* Don't forget to insert hole */
594 hole_len = cur_off - hole_start;
596 ret = btrfs_record_file_extent(trans, root, ino, inode,
597 hole_start, 0, hole_len);
602 cur_off += key.offset;
603 hole_start = cur_off;
604 cur_len = start + len - cur_off;
607 if (start + len - hole_start > 0)
608 ret = btrfs_record_file_extent(trans, root, ino, inode,
609 hole_start, 0, start + len - hole_start);
614 * Relocate the used ext2 data in reserved ranges
616 * [btrfs_sb_offset(1), +BTRFS_STRIPE_LEN)
617 * [btrfs_sb_offset(2), +BTRFS_STRIPE_LEN)
619 static int migrate_reserved_ranges(struct btrfs_trans_handle *trans,
620 struct btrfs_root *root,
621 struct cache_tree *used,
622 struct btrfs_inode_item *inode, int fd,
623 u64 ino, u64 total_bytes, int datacsum)
631 cur_len = 1024 * 1024;
632 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
633 cur_off, cur_len, datacsum);
637 /* second sb(fisrt sb is included in 0~1M) */
638 cur_off = btrfs_sb_offset(1);
639 cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
640 if (cur_off > total_bytes)
642 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
643 cur_off, cur_len, datacsum);
648 cur_off = btrfs_sb_offset(2);
649 cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
650 if (cur_off > total_bytes)
652 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
653 cur_off, cur_len, datacsum);
658 * Helper for expand and merge extent_cache for wipe_one_reserved_range() to
659 * handle wiping a range that exists in cache.
661 static int _expand_extent_cache(struct cache_tree *tree,
662 struct cache_extent *entry,
663 u64 min_stripe_size, int backward)
665 struct cache_extent *ce;
668 if (entry->size >= min_stripe_size)
670 diff = min_stripe_size - entry->size;
673 ce = prev_cache_extent(entry);
676 if (ce->start + ce->size >= entry->start - diff) {
677 /* Directly merge with previous extent */
678 ce->size = entry->start + entry->size - ce->start;
679 remove_cache_extent(tree, entry);
684 /* No overlap, normal extent */
685 if (entry->start < diff) {
686 error("cannot find space for data chunk layout");
689 entry->start -= diff;
693 ce = next_cache_extent(entry);
696 if (entry->start + entry->size + diff >= ce->start) {
697 /* Directly merge with next extent */
698 entry->size = ce->start + ce->size - entry->start;
699 remove_cache_extent(tree, ce);
709 * Remove one reserve range from given cache tree
710 * if min_stripe_size is non-zero, it will ensure for split case,
711 * all its split cache extent is no smaller than @min_strip_size / 2.
713 static int wipe_one_reserved_range(struct cache_tree *tree,
714 u64 start, u64 len, u64 min_stripe_size,
717 struct cache_extent *cache;
720 BUG_ON(ensure_size && min_stripe_size == 0);
722 * The logical here is simplified to handle special cases only
723 * So we don't need to consider merge case for ensure_size
725 BUG_ON(min_stripe_size && (min_stripe_size < len * 2 ||
726 min_stripe_size / 2 < BTRFS_STRIPE_LEN));
728 /* Also, wipe range should already be aligned */
729 BUG_ON(start != round_down(start, BTRFS_STRIPE_LEN) ||
730 start + len != round_up(start + len, BTRFS_STRIPE_LEN));
732 min_stripe_size /= 2;
734 cache = lookup_cache_extent(tree, start, len);
738 if (start <= cache->start) {
740 * |--------cache---------|
743 BUG_ON(start + len <= cache->start);
746 * The wipe size is smaller than min_stripe_size / 2,
747 * so the result length should still meet min_stripe_size
748 * And no need to do alignment
750 cache->size -= (start + len - cache->start);
751 if (cache->size == 0) {
752 remove_cache_extent(tree, cache);
757 BUG_ON(ensure_size && cache->size < min_stripe_size);
759 cache->start = start + len;
761 } else if (start > cache->start && start + len < cache->start +
764 * |-------cache-----|
767 u64 old_start = cache->start;
768 u64 old_len = cache->size;
769 u64 insert_start = start + len;
772 cache->size = start - cache->start;
773 /* Expand the leading half part if needed */
774 if (ensure_size && cache->size < min_stripe_size) {
775 ret = _expand_extent_cache(tree, cache,
781 /* And insert the new one */
782 insert_len = old_start + old_len - start - len;
783 ret = add_merge_cache_extent(tree, insert_start, insert_len);
787 /* Expand the last half part if needed */
788 if (ensure_size && insert_len < min_stripe_size) {
789 cache = lookup_cache_extent(tree, insert_start,
791 if (!cache || cache->start != insert_start ||
792 cache->size != insert_len)
794 ret = _expand_extent_cache(tree, cache,
803 * Wipe len should be small enough and no need to expand the
806 cache->size = start - cache->start;
807 BUG_ON(ensure_size && cache->size < min_stripe_size);
812 * Remove reserved ranges from given cache_tree
814 * It will remove the following ranges
816 * 2) 2nd superblock, +64K (make sure chunks are 64K aligned)
817 * 3) 3rd superblock, +64K
819 * @min_stripe must be given for safety check
820 * and if @ensure_size is given, it will ensure affected cache_extent will be
821 * larger than min_stripe_size
823 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
828 ret = wipe_one_reserved_range(tree, 0, 1024 * 1024, min_stripe_size,
832 ret = wipe_one_reserved_range(tree, btrfs_sb_offset(1),
833 BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
836 ret = wipe_one_reserved_range(tree, btrfs_sb_offset(2),
837 BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
842 * Open Ext2fs in readonly mode, read block allocation bitmap and
843 * inode bitmap into memory.
845 static int ext2_open_fs(struct btrfs_convert_context *cctx, const char *name)
852 ret = ext2fs_open(name, 0, 0, 0, unix_io_manager, &ext2_fs);
854 fprintf(stderr, "ext2fs_open: %s\n", error_message(ret));
858 * We need to know exactly the used space, some RO compat flags like
859 * BIGALLOC will affect how used space is present.
860 * So we need manuall check any unsupported RO compat flags
862 ro_feature = ext2_fs->super->s_feature_ro_compat;
863 if (ro_feature & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
865 "unsupported RO features detected: %x, abort convert to avoid possible corruption",
866 ro_feature & ~EXT2_LIB_FEATURE_COMPAT_SUPP);
869 ret = ext2fs_read_inode_bitmap(ext2_fs);
871 fprintf(stderr, "ext2fs_read_inode_bitmap: %s\n",
875 ret = ext2fs_read_block_bitmap(ext2_fs);
877 fprintf(stderr, "ext2fs_read_block_bitmap: %s\n",
882 * search each block group for a free inode. this set up
883 * uninit block/inode bitmaps appropriately.
886 while (ino <= ext2_fs->super->s_inodes_count) {
888 ext2fs_new_inode(ext2_fs, ino, 0, NULL, &foo);
889 ino += EXT2_INODES_PER_GROUP(ext2_fs->super);
892 if (!(ext2_fs->super->s_feature_incompat &
893 EXT2_FEATURE_INCOMPAT_FILETYPE)) {
894 fprintf(stderr, "filetype feature is missing\n");
898 cctx->fs_data = ext2_fs;
899 cctx->blocksize = ext2_fs->blocksize;
900 cctx->block_count = ext2_fs->super->s_blocks_count;
901 cctx->total_bytes = ext2_fs->blocksize * ext2_fs->super->s_blocks_count;
902 cctx->volume_name = strndup(ext2_fs->super->s_volume_name, 16);
903 cctx->first_data_block = ext2_fs->super->s_first_data_block;
904 cctx->inodes_count = ext2_fs->super->s_inodes_count;
905 cctx->free_inodes_count = ext2_fs->super->s_free_inodes_count;
908 ext2fs_close(ext2_fs);
912 static int __ext2_add_one_block(ext2_filsys fs, char *bitmap,
913 unsigned long group_nr, struct cache_tree *used)
915 unsigned long offset;
919 offset = fs->super->s_first_data_block;
920 offset /= EXT2FS_CLUSTER_RATIO(fs);
921 offset += group_nr * EXT2_CLUSTERS_PER_GROUP(fs->super);
922 for (i = 0; i < EXT2_CLUSTERS_PER_GROUP(fs->super); i++) {
923 if (ext2fs_test_bit(i, bitmap)) {
926 start = (i + offset) * EXT2FS_CLUSTER_RATIO(fs);
927 start *= fs->blocksize;
928 ret = add_merge_cache_extent(used, start,
938 * Read all used ext2 space into cctx->used cache tree
940 static int ext2_read_used_space(struct btrfs_convert_context *cctx)
942 ext2_filsys fs = (ext2_filsys)cctx->fs_data;
943 blk64_t blk_itr = EXT2FS_B2C(fs, fs->super->s_first_data_block);
944 struct cache_tree *used_tree = &cctx->used;
945 char *block_bitmap = NULL;
950 block_nbytes = EXT2_CLUSTERS_PER_GROUP(fs->super) / 8;
951 /* Shouldn't happen */
952 BUG_ON(!fs->block_map);
954 block_bitmap = malloc(block_nbytes);
958 for (i = 0; i < fs->group_desc_count; i++) {
959 ret = ext2fs_get_block_bitmap_range(fs->block_map, blk_itr,
960 block_nbytes * 8, block_bitmap);
962 error("fail to get bitmap from ext2, %s",
966 ret = __ext2_add_one_block(fs, block_bitmap, i, used_tree);
968 error("fail to build used space tree, %s",
972 blk_itr += EXT2_CLUSTERS_PER_GROUP(fs->super);
979 static void ext2_close_fs(struct btrfs_convert_context *cctx)
981 if (cctx->volume_name) {
982 free(cctx->volume_name);
983 cctx->volume_name = NULL;
985 ext2fs_close(cctx->fs_data);
988 struct dir_iterate_data {
989 struct btrfs_trans_handle *trans;
990 struct btrfs_root *root;
991 struct btrfs_inode_item *inode;
998 static u8 ext2_filetype_conversion_table[EXT2_FT_MAX] = {
999 [EXT2_FT_UNKNOWN] = BTRFS_FT_UNKNOWN,
1000 [EXT2_FT_REG_FILE] = BTRFS_FT_REG_FILE,
1001 [EXT2_FT_DIR] = BTRFS_FT_DIR,
1002 [EXT2_FT_CHRDEV] = BTRFS_FT_CHRDEV,
1003 [EXT2_FT_BLKDEV] = BTRFS_FT_BLKDEV,
1004 [EXT2_FT_FIFO] = BTRFS_FT_FIFO,
1005 [EXT2_FT_SOCK] = BTRFS_FT_SOCK,
1006 [EXT2_FT_SYMLINK] = BTRFS_FT_SYMLINK,
1009 static int ext2_dir_iterate_proc(ext2_ino_t dir, int entry,
1010 struct ext2_dir_entry *dirent,
1011 int offset, int blocksize,
1012 char *buf,void *priv_data)
1017 char dotdot[] = "..";
1018 struct dir_iterate_data *idata = (struct dir_iterate_data *)priv_data;
1021 name_len = dirent->name_len & 0xFF;
1023 objectid = dirent->inode + INO_OFFSET;
1024 if (!strncmp(dirent->name, dotdot, name_len)) {
1025 if (name_len == 2) {
1026 BUG_ON(idata->parent != 0);
1027 idata->parent = objectid;
1031 if (dirent->inode < EXT2_GOOD_OLD_FIRST_INO)
1034 file_type = dirent->name_len >> 8;
1035 BUG_ON(file_type > EXT2_FT_SYMLINK);
1037 ret = convert_insert_dirent(idata->trans, idata->root, dirent->name,
1038 name_len, idata->objectid, objectid,
1039 ext2_filetype_conversion_table[file_type],
1040 idata->index_cnt, idata->inode);
1042 idata->errcode = ret;
1050 static int ext2_create_dir_entries(struct btrfs_trans_handle *trans,
1051 struct btrfs_root *root, u64 objectid,
1052 struct btrfs_inode_item *btrfs_inode,
1053 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
1057 struct dir_iterate_data data = {
1060 .inode = btrfs_inode,
1061 .objectid = objectid,
1067 err = ext2fs_dir_iterate2(ext2_fs, ext2_ino, 0, NULL,
1068 ext2_dir_iterate_proc, &data);
1072 if (ret == 0 && data.parent == objectid) {
1073 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
1074 objectid, objectid, 0);
1078 fprintf(stderr, "ext2fs_dir_iterate2: %s\n", error_message(err));
1082 static int ext2_block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
1083 e2_blkcnt_t blockcnt, blk_t ref_block,
1084 int ref_offset, void *priv_data)
1087 struct blk_iterate_data *idata;
1088 idata = (struct blk_iterate_data *)priv_data;
1089 ret = block_iterate_proc(*blocknr, blockcnt, idata);
1091 idata->errcode = ret;
1098 * traverse file's data blocks, record these data blocks as file extents.
1100 static int ext2_create_file_extents(struct btrfs_trans_handle *trans,
1101 struct btrfs_root *root, u64 objectid,
1102 struct btrfs_inode_item *btrfs_inode,
1103 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1104 int datacsum, int packing)
1107 char *buffer = NULL;
1110 u32 sectorsize = root->sectorsize;
1111 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1112 struct blk_iterate_data data;
1114 init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
1117 err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
1118 NULL, ext2_block_iterate_proc, &data);
1124 if (packing && data.first_block == 0 && data.num_blocks > 0 &&
1125 inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
1126 u64 num_bytes = data.num_blocks * sectorsize;
1127 u64 disk_bytenr = data.disk_block * sectorsize;
1130 buffer = malloc(num_bytes);
1133 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
1136 if (num_bytes > inode_size)
1137 num_bytes = inode_size;
1138 ret = btrfs_insert_inline_extent(trans, root, objectid,
1139 0, buffer, num_bytes);
1142 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
1143 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
1144 } else if (data.num_blocks > 0) {
1145 ret = record_file_blocks(&data, data.first_block,
1146 data.disk_block, data.num_blocks);
1150 data.first_block += data.num_blocks;
1151 last_block = (inode_size + sectorsize - 1) / sectorsize;
1152 if (last_block > data.first_block) {
1153 ret = record_file_blocks(&data, data.first_block, 0,
1154 last_block - data.first_block);
1160 fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
1164 static int ext2_create_symbol_link(struct btrfs_trans_handle *trans,
1165 struct btrfs_root *root, u64 objectid,
1166 struct btrfs_inode_item *btrfs_inode,
1167 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1168 struct ext2_inode *ext2_inode)
1172 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1173 if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
1174 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
1175 ret = ext2_create_file_extents(trans, root, objectid,
1176 btrfs_inode, ext2_fs, ext2_ino, 1, 1);
1177 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
1181 pathname = (char *)&(ext2_inode->i_block[0]);
1182 BUG_ON(pathname[inode_size] != 0);
1183 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
1184 pathname, inode_size + 1);
1185 btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
1190 * Following xattr/acl related codes are based on codes in
1191 * fs/ext3/xattr.c and fs/ext3/acl.c
1193 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
1194 #define EXT2_XATTR_BFIRST(ptr) \
1195 ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
1196 #define EXT2_XATTR_IHDR(inode) \
1197 ((struct ext2_ext_attr_header *) ((void *)(inode) + \
1198 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
1199 #define EXT2_XATTR_IFIRST(inode) \
1200 ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
1201 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
1203 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
1206 struct ext2_ext_attr_entry *next;
1208 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1209 next = EXT2_EXT_ATTR_NEXT(entry);
1210 if ((void *)next >= end)
1217 static int ext2_xattr_check_block(const char *buf, size_t size)
1220 struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
1222 if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
1223 header->h_blocks != 1)
1225 error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
1229 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
1232 size_t value_size = entry->e_value_size;
1234 if (entry->e_value_block != 0 || value_size > size ||
1235 entry->e_value_offs + value_size > size)
1240 #define EXT2_ACL_VERSION 0x0001
1242 /* 23.2.5 acl_tag_t values */
1244 #define ACL_UNDEFINED_TAG (0x00)
1245 #define ACL_USER_OBJ (0x01)
1246 #define ACL_USER (0x02)
1247 #define ACL_GROUP_OBJ (0x04)
1248 #define ACL_GROUP (0x08)
1249 #define ACL_MASK (0x10)
1250 #define ACL_OTHER (0x20)
1252 /* 23.2.7 ACL qualifier constants */
1254 #define ACL_UNDEFINED_ID ((id_t)-1)
1265 } ext2_acl_entry_short;
1271 static inline int ext2_acl_count(size_t size)
1274 size -= sizeof(ext2_acl_header);
1275 s = size - 4 * sizeof(ext2_acl_entry_short);
1277 if (size % sizeof(ext2_acl_entry_short))
1279 return size / sizeof(ext2_acl_entry_short);
1281 if (s % sizeof(ext2_acl_entry))
1283 return s / sizeof(ext2_acl_entry) + 4;
1287 #define ACL_EA_VERSION 0x0002
1297 acl_ea_entry a_entries[0];
1300 static inline size_t acl_ea_size(int count)
1302 return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
1305 static int ext2_acl_to_xattr(void *dst, const void *src,
1306 size_t dst_size, size_t src_size)
1309 const void *end = src + src_size;
1310 acl_ea_header *ext_acl = (acl_ea_header *)dst;
1311 acl_ea_entry *dst_entry = ext_acl->a_entries;
1312 ext2_acl_entry *src_entry;
1314 if (src_size < sizeof(ext2_acl_header))
1316 if (((ext2_acl_header *)src)->a_version !=
1317 cpu_to_le32(EXT2_ACL_VERSION))
1319 src += sizeof(ext2_acl_header);
1320 count = ext2_acl_count(src_size);
1324 BUG_ON(dst_size < acl_ea_size(count));
1325 ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
1326 for (i = 0; i < count; i++, dst_entry++) {
1327 src_entry = (ext2_acl_entry *)src;
1328 if (src + sizeof(ext2_acl_entry_short) > end)
1330 dst_entry->e_tag = src_entry->e_tag;
1331 dst_entry->e_perm = src_entry->e_perm;
1332 switch (le16_to_cpu(src_entry->e_tag)) {
1337 src += sizeof(ext2_acl_entry_short);
1338 dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
1342 src += sizeof(ext2_acl_entry);
1345 dst_entry->e_id = src_entry->e_id;
1358 static char *xattr_prefix_table[] = {
1360 [2] = "system.posix_acl_access",
1361 [3] = "system.posix_acl_default",
1366 static int ext2_copy_single_xattr(struct btrfs_trans_handle *trans,
1367 struct btrfs_root *root, u64 objectid,
1368 struct ext2_ext_attr_entry *entry,
1369 const void *data, u32 datalen)
1374 void *databuf = NULL;
1375 char namebuf[XATTR_NAME_MAX + 1];
1377 name_index = entry->e_name_index;
1378 if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
1379 xattr_prefix_table[name_index] == NULL)
1381 name_len = strlen(xattr_prefix_table[name_index]) +
1383 if (name_len >= sizeof(namebuf))
1386 if (name_index == 2 || name_index == 3) {
1387 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
1388 databuf = malloc(bufsize);
1391 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
1397 strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
1398 strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
1399 if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
1400 sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
1401 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
1402 objectid - INO_OFFSET, name_len, namebuf);
1405 ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
1406 data, datalen, objectid);
1412 static int ext2_copy_extended_attrs(struct btrfs_trans_handle *trans,
1413 struct btrfs_root *root, u64 objectid,
1414 struct btrfs_inode_item *btrfs_inode,
1415 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
1421 u32 block_size = ext2_fs->blocksize;
1422 u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
1423 struct ext2_inode_large *ext2_inode;
1424 struct ext2_ext_attr_entry *entry;
1426 char *buffer = NULL;
1427 char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
1429 if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
1430 ext2_inode = (struct ext2_inode_large *)inode_buf;
1432 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
1436 err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
1439 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
1440 error_message(err));
1445 if (ext2_ino > ext2_fs->super->s_first_ino &&
1446 inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
1447 if (EXT2_GOOD_OLD_INODE_SIZE +
1448 ext2_inode->i_extra_isize > inode_size) {
1452 if (ext2_inode->i_extra_isize != 0 &&
1453 EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
1454 EXT2_EXT_ATTR_MAGIC) {
1460 void *end = (void *)ext2_inode + inode_size;
1461 entry = EXT2_XATTR_IFIRST(ext2_inode);
1462 total = end - (void *)entry;
1463 ret = ext2_xattr_check_names(entry, end);
1466 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1467 ret = ext2_xattr_check_entry(entry, total);
1470 data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
1471 entry->e_value_offs;
1472 datalen = entry->e_value_size;
1473 ret = ext2_copy_single_xattr(trans, root, objectid,
1474 entry, data, datalen);
1477 entry = EXT2_EXT_ATTR_NEXT(entry);
1481 if (ext2_inode->i_file_acl == 0)
1484 buffer = malloc(block_size);
1489 err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
1491 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
1492 error_message(err));
1496 ret = ext2_xattr_check_block(buffer, block_size);
1500 entry = EXT2_XATTR_BFIRST(buffer);
1501 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1502 ret = ext2_xattr_check_entry(entry, block_size);
1505 data = buffer + entry->e_value_offs;
1506 datalen = entry->e_value_size;
1507 ret = ext2_copy_single_xattr(trans, root, objectid,
1508 entry, data, datalen);
1511 entry = EXT2_EXT_ATTR_NEXT(entry);
1515 if ((void *)ext2_inode != inode_buf)
1519 #define MINORBITS 20
1520 #define MKDEV(ma, mi) (((ma) << MINORBITS) | (mi))
1522 static inline dev_t old_decode_dev(u16 val)
1524 return MKDEV((val >> 8) & 255, val & 255);
1527 static inline dev_t new_decode_dev(u32 dev)
1529 unsigned major = (dev & 0xfff00) >> 8;
1530 unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
1531 return MKDEV(major, minor);
1534 static void ext2_copy_inode_item(struct btrfs_inode_item *dst,
1535 struct ext2_inode *src, u32 blocksize)
1537 btrfs_set_stack_inode_generation(dst, 1);
1538 btrfs_set_stack_inode_sequence(dst, 0);
1539 btrfs_set_stack_inode_transid(dst, 1);
1540 btrfs_set_stack_inode_size(dst, src->i_size);
1541 btrfs_set_stack_inode_nbytes(dst, 0);
1542 btrfs_set_stack_inode_block_group(dst, 0);
1543 btrfs_set_stack_inode_nlink(dst, src->i_links_count);
1544 btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
1545 btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
1546 btrfs_set_stack_inode_mode(dst, src->i_mode);
1547 btrfs_set_stack_inode_rdev(dst, 0);
1548 btrfs_set_stack_inode_flags(dst, 0);
1549 btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
1550 btrfs_set_stack_timespec_nsec(&dst->atime, 0);
1551 btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
1552 btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
1553 btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
1554 btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
1555 btrfs_set_stack_timespec_sec(&dst->otime, 0);
1556 btrfs_set_stack_timespec_nsec(&dst->otime, 0);
1558 if (S_ISDIR(src->i_mode)) {
1559 btrfs_set_stack_inode_size(dst, 0);
1560 btrfs_set_stack_inode_nlink(dst, 1);
1562 if (S_ISREG(src->i_mode)) {
1563 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
1566 if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
1567 !S_ISLNK(src->i_mode)) {
1568 if (src->i_block[0]) {
1569 btrfs_set_stack_inode_rdev(dst,
1570 old_decode_dev(src->i_block[0]));
1572 btrfs_set_stack_inode_rdev(dst,
1573 new_decode_dev(src->i_block[1]));
1576 memset(&dst->reserved, 0, sizeof(dst->reserved));
1580 * copy a single inode. do all the required works, such as cloning
1581 * inode item, creating file extents and creating directory entries.
1583 static int ext2_copy_single_inode(struct btrfs_trans_handle *trans,
1584 struct btrfs_root *root, u64 objectid,
1585 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1586 struct ext2_inode *ext2_inode,
1587 int datacsum, int packing, int noxattr)
1590 struct btrfs_inode_item btrfs_inode;
1592 if (ext2_inode->i_links_count == 0)
1595 ext2_copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
1596 if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
1597 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
1598 BTRFS_INODE_NODATASUM;
1599 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
1602 switch (ext2_inode->i_mode & S_IFMT) {
1604 ret = ext2_create_file_extents(trans, root, objectid,
1605 &btrfs_inode, ext2_fs, ext2_ino, datacsum, packing);
1608 ret = ext2_create_dir_entries(trans, root, objectid,
1609 &btrfs_inode, ext2_fs, ext2_ino);
1612 ret = ext2_create_symbol_link(trans, root, objectid,
1613 &btrfs_inode, ext2_fs, ext2_ino, ext2_inode);
1623 ret = ext2_copy_extended_attrs(trans, root, objectid,
1624 &btrfs_inode, ext2_fs, ext2_ino);
1628 return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
1632 * scan ext2's inode bitmap and copy all used inodes.
1634 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
1635 struct btrfs_root *root,
1636 int datacsum, int packing, int noxattr, struct task_ctx *p)
1638 ext2_filsys ext2_fs = cctx->fs_data;
1641 ext2_inode_scan ext2_scan;
1642 struct ext2_inode ext2_inode;
1643 ext2_ino_t ext2_ino;
1645 struct btrfs_trans_handle *trans;
1647 trans = btrfs_start_transaction(root, 1);
1650 err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
1652 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
1655 while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
1657 /* no more inodes */
1660 /* skip special inode in ext2fs */
1661 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
1662 ext2_ino != EXT2_ROOT_INO)
1664 objectid = ext2_ino + INO_OFFSET;
1665 ret = ext2_copy_single_inode(trans, root,
1666 objectid, ext2_fs, ext2_ino,
1667 &ext2_inode, datacsum, packing,
1669 p->cur_copy_inodes++;
1672 if (trans->blocks_used >= 4096) {
1673 ret = btrfs_commit_transaction(trans, root);
1675 trans = btrfs_start_transaction(root, 1);
1680 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
1683 ret = btrfs_commit_transaction(trans, root);
1685 ext2fs_close_inode_scan(ext2_scan);
1691 * Create the fs image file of old filesystem.
1693 * This is completely fs independent as we have cctx->used, only
1694 * need to create file extents pointing to all the positions.
1696 static int create_image(struct btrfs_root *root,
1697 struct btrfs_mkfs_config *cfg,
1698 struct btrfs_convert_context *cctx, int fd,
1699 u64 size, char *name, int datacsum)
1701 struct btrfs_inode_item buf;
1702 struct btrfs_trans_handle *trans;
1703 struct btrfs_path *path = NULL;
1704 struct btrfs_key key;
1705 struct cache_extent *cache;
1706 struct cache_tree used_tmp;
1709 u64 flags = BTRFS_INODE_READONLY;
1713 flags |= BTRFS_INODE_NODATASUM;
1715 trans = btrfs_start_transaction(root, 1);
1719 cache_tree_init(&used_tmp);
1721 ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
1725 ret = btrfs_new_inode(trans, root, ino, 0400 | S_IFREG);
1728 ret = btrfs_change_inode_flags(trans, root, ino, flags);
1731 ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
1732 strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
1736 path = btrfs_alloc_path();
1742 key.type = BTRFS_INODE_ITEM_KEY;
1745 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1747 ret = (ret > 0 ? -ENOENT : ret);
1750 read_extent_buffer(path->nodes[0], &buf,
1751 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1753 btrfs_release_path(path);
1756 * Create a new used space cache, which doesn't contain the reserved
1759 for (cache = first_cache_extent(&cctx->used); cache;
1760 cache = next_cache_extent(cache)) {
1761 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1765 ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1770 * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1771 * can't handle bytenr 0(will consider it as a hole)
1774 while (cur < size) {
1775 u64 len = size - cur;
1777 ret = create_image_file_range(trans, root, &used_tmp,
1778 &buf, ino, cur, &len, datacsum);
1783 /* Handle the reserved ranges */
1784 ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1785 cfg->num_bytes, datacsum);
1789 key.type = BTRFS_INODE_ITEM_KEY;
1791 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1793 ret = (ret > 0 ? -ENOENT : ret);
1796 btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1797 write_extent_buffer(path->nodes[0], &buf,
1798 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1801 free_extent_cache_tree(&used_tmp);
1802 btrfs_free_path(path);
1803 btrfs_commit_transaction(trans, root);
1807 static struct btrfs_root * link_subvol(struct btrfs_root *root,
1808 const char *base, u64 root_objectid)
1810 struct btrfs_trans_handle *trans;
1811 struct btrfs_fs_info *fs_info = root->fs_info;
1812 struct btrfs_root *tree_root = fs_info->tree_root;
1813 struct btrfs_root *new_root = NULL;
1814 struct btrfs_path *path;
1815 struct btrfs_inode_item *inode_item;
1816 struct extent_buffer *leaf;
1817 struct btrfs_key key;
1818 u64 dirid = btrfs_root_dirid(&root->root_item);
1820 char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1826 if (len == 0 || len > BTRFS_NAME_LEN)
1829 path = btrfs_alloc_path();
1832 key.objectid = dirid;
1833 key.type = BTRFS_DIR_INDEX_KEY;
1834 key.offset = (u64)-1;
1836 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1839 if (path->slots[0] > 0) {
1841 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1842 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1843 index = key.offset + 1;
1845 btrfs_release_path(path);
1847 trans = btrfs_start_transaction(root, 1);
1850 key.objectid = dirid;
1852 key.type = BTRFS_INODE_ITEM_KEY;
1854 ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1856 leaf = path->nodes[0];
1857 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1858 struct btrfs_inode_item);
1860 key.objectid = root_objectid;
1861 key.offset = (u64)-1;
1862 key.type = BTRFS_ROOT_ITEM_KEY;
1864 memcpy(buf, base, len);
1865 for (i = 0; i < 1024; i++) {
1866 ret = btrfs_insert_dir_item(trans, root, buf, len,
1867 dirid, &key, BTRFS_FT_DIR, index);
1870 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1871 if (len < 1 || len > BTRFS_NAME_LEN) {
1879 btrfs_set_inode_size(leaf, inode_item, len * 2 +
1880 btrfs_inode_size(leaf, inode_item));
1881 btrfs_mark_buffer_dirty(leaf);
1882 btrfs_release_path(path);
1884 /* add the backref first */
1885 ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1886 BTRFS_ROOT_BACKREF_KEY,
1887 root->root_key.objectid,
1888 dirid, index, buf, len);
1891 /* now add the forward ref */
1892 ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1893 BTRFS_ROOT_REF_KEY, root_objectid,
1894 dirid, index, buf, len);
1896 ret = btrfs_commit_transaction(trans, root);
1899 new_root = btrfs_read_fs_root(fs_info, &key);
1900 if (IS_ERR(new_root))
1903 btrfs_free_path(path);
1907 static int create_subvol(struct btrfs_trans_handle *trans,
1908 struct btrfs_root *root, u64 root_objectid)
1910 struct extent_buffer *tmp;
1911 struct btrfs_root *new_root;
1912 struct btrfs_key key;
1913 struct btrfs_root_item root_item;
1916 ret = btrfs_copy_root(trans, root, root->node, &tmp,
1920 memcpy(&root_item, &root->root_item, sizeof(root_item));
1921 btrfs_set_root_bytenr(&root_item, tmp->start);
1922 btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
1923 btrfs_set_root_generation(&root_item, trans->transid);
1924 free_extent_buffer(tmp);
1926 key.objectid = root_objectid;
1927 key.type = BTRFS_ROOT_ITEM_KEY;
1928 key.offset = trans->transid;
1929 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1932 key.offset = (u64)-1;
1933 new_root = btrfs_read_fs_root(root->fs_info, &key);
1934 BUG_ON(!new_root || IS_ERR(new_root));
1936 ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
1943 * New make_btrfs() has handle system and meta chunks quite well.
1944 * So only need to add remaining data chunks.
1946 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
1947 struct btrfs_fs_info *fs_info,
1948 struct btrfs_mkfs_config *cfg,
1949 struct btrfs_convert_context *cctx)
1951 struct btrfs_root *extent_root = fs_info->extent_root;
1952 struct cache_tree *data_chunks = &cctx->data_chunks;
1953 struct cache_extent *cache;
1958 * Don't create data chunk over 10% of the convert device
1959 * And for single chunk, don't create chunk larger than 1G.
1961 max_chunk_size = cfg->num_bytes / 10;
1962 max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
1963 max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
1965 for (cache = first_cache_extent(data_chunks); cache;
1966 cache = next_cache_extent(cache)) {
1967 u64 cur = cache->start;
1969 while (cur < cache->start + cache->size) {
1971 u64 cur_backup = cur;
1973 len = min(max_chunk_size,
1974 cache->start + cache->size - cur);
1975 ret = btrfs_alloc_data_chunk(trans, extent_root,
1977 BTRFS_BLOCK_GROUP_DATA, 1);
1980 ret = btrfs_make_block_group(trans, extent_root, 0,
1981 BTRFS_BLOCK_GROUP_DATA,
1982 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1993 * Init the temp btrfs to a operational status.
1995 * It will fix the extent usage accounting(XXX: Do we really need?) and
1996 * insert needed data chunks, to ensure all old fs data extents are covered
1997 * by DATA chunks, preventing wrong chunks are allocated.
1999 * And also create convert image subvolume and relocation tree.
2000 * (XXX: Not need again?)
2001 * But the convert image subvolume is *NOT* linked to fs tree yet.
2003 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
2004 struct btrfs_convert_context *cctx, int datacsum,
2005 int packing, int noxattr)
2007 struct btrfs_key location;
2008 struct btrfs_trans_handle *trans;
2009 struct btrfs_fs_info *fs_info = root->fs_info;
2013 * Don't alloc any metadata/system chunk, as we don't want
2014 * any meta/sys chunk allcated before all data chunks are inserted.
2015 * Or we screw up the chunk layout just like the old implement.
2017 fs_info->avoid_sys_chunk_alloc = 1;
2018 fs_info->avoid_meta_chunk_alloc = 1;
2019 trans = btrfs_start_transaction(root, 1);
2021 ret = btrfs_fix_block_accounting(trans, root);
2024 ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
2027 ret = btrfs_make_root_dir(trans, fs_info->tree_root,
2028 BTRFS_ROOT_TREE_DIR_OBJECTID);
2031 memcpy(&location, &root->root_key, sizeof(location));
2032 location.offset = (u64)-1;
2033 ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
2034 btrfs_super_root_dir(fs_info->super_copy),
2035 &location, BTRFS_FT_DIR, 0);
2038 ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
2040 btrfs_super_root_dir(fs_info->super_copy), 0);
2043 btrfs_set_root_dirid(&fs_info->fs_root->root_item,
2044 BTRFS_FIRST_FREE_OBJECTID);
2046 /* subvol for fs image file */
2047 ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
2050 /* subvol for data relocation tree */
2051 ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
2055 ret = btrfs_commit_transaction(trans, root);
2056 fs_info->avoid_sys_chunk_alloc = 0;
2057 fs_info->avoid_meta_chunk_alloc = 0;
2063 * Migrate super block to its default position and zero 0 ~ 16k
2065 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
2068 struct extent_buffer *buf;
2069 struct btrfs_super_block *super;
2073 BUG_ON(sectorsize < sizeof(*super));
2074 buf = malloc(sizeof(*buf) + sectorsize);
2078 buf->len = sectorsize;
2079 ret = pread(fd, buf->data, sectorsize, old_bytenr);
2080 if (ret != sectorsize)
2083 super = (struct btrfs_super_block *)buf->data;
2084 BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
2085 btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
2087 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
2088 ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2089 if (ret != sectorsize)
2096 memset(buf->data, 0, sectorsize);
2097 for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
2098 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
2099 if (len > sectorsize)
2101 ret = pwrite(fd, buf->data, len, bytenr);
2103 fprintf(stderr, "unable to zero fill device\n");
2117 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
2119 struct btrfs_chunk *chunk;
2120 struct btrfs_disk_key *key;
2121 u32 sectorsize = btrfs_super_sectorsize(super);
2123 key = (struct btrfs_disk_key *)(super->sys_chunk_array);
2124 chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
2125 sizeof(struct btrfs_disk_key));
2127 btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2128 btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
2129 btrfs_set_disk_key_offset(key, 0);
2131 btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
2132 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
2133 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
2134 btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
2135 btrfs_set_stack_chunk_io_align(chunk, sectorsize);
2136 btrfs_set_stack_chunk_io_width(chunk, sectorsize);
2137 btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
2138 btrfs_set_stack_chunk_num_stripes(chunk, 1);
2139 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
2140 chunk->stripe.devid = super->dev_item.devid;
2141 btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
2142 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
2143 btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
2147 static const struct btrfs_convert_operations ext2_convert_ops = {
2149 .open_fs = ext2_open_fs,
2150 .read_used_space = ext2_read_used_space,
2151 .copy_inodes = ext2_copy_inodes,
2152 .close_fs = ext2_close_fs,
2155 static const struct btrfs_convert_operations *convert_operations[] = {
2159 static int convert_open_fs(const char *devname,
2160 struct btrfs_convert_context *cctx)
2164 memset(cctx, 0, sizeof(*cctx));
2166 for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
2167 int ret = convert_operations[i]->open_fs(cctx, devname);
2170 cctx->convert_ops = convert_operations[i];
2175 fprintf(stderr, "No file system found to convert.\n");
2179 static int calculate_available_space(struct btrfs_convert_context *cctx)
2181 struct cache_tree *used = &cctx->used;
2182 struct cache_tree *data_chunks = &cctx->data_chunks;
2183 struct cache_tree *free = &cctx->free;
2184 struct cache_extent *cache;
2187 * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
2188 * works without need to consider overlap
2190 u64 min_stripe_size = 2 * 16 * 1024 * 1024;
2193 /* Calculate data_chunks */
2194 for (cache = first_cache_extent(used); cache;
2195 cache = next_cache_extent(cache)) {
2198 if (cache->start + cache->size < cur_off)
2200 if (cache->start > cur_off + min_stripe_size)
2201 cur_off = cache->start;
2202 cur_len = max(cache->start + cache->size - cur_off,
2204 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
2210 * remove reserved ranges, so we won't ever bother relocating an old
2211 * filesystem extent to other place.
2213 ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
2219 * Calculate free space
2220 * Always round up the start bytenr, to avoid metadata extent corss
2221 * stripe boundary, as later mkfs_convert() won't have all the extent
2224 for (cache = first_cache_extent(data_chunks); cache;
2225 cache = next_cache_extent(cache)) {
2226 if (cache->start < cur_off)
2228 if (cache->start > cur_off) {
2232 len = cache->start - round_up(cur_off,
2234 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2236 ret = add_merge_cache_extent(free, insert_start, len);
2240 cur_off = cache->start + cache->size;
2242 /* Don't forget the last range */
2243 if (cctx->total_bytes > cur_off) {
2244 u64 len = cctx->total_bytes - cur_off;
2247 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2249 ret = add_merge_cache_extent(free, insert_start, len);
2254 /* Remove reserved bytes */
2255 ret = wipe_reserved_ranges(free, min_stripe_size, 0);
2260 * Read used space, and since we have the used space,
2261 * calcuate data_chunks and free for later mkfs
2263 static int convert_read_used_space(struct btrfs_convert_context *cctx)
2267 ret = cctx->convert_ops->read_used_space(cctx);
2271 ret = calculate_available_space(cctx);
2275 static int do_convert(const char *devname, int datacsum, int packing,
2276 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2277 int progress, u64 features)
2284 struct btrfs_root *root;
2285 struct btrfs_root *image_root;
2286 struct btrfs_convert_context cctx;
2287 struct btrfs_key key;
2288 char *subvol_name = NULL;
2289 struct task_ctx ctx;
2290 char features_buf[64];
2291 struct btrfs_mkfs_config mkfs_cfg;
2293 init_convert_context(&cctx);
2294 ret = convert_open_fs(devname, &cctx);
2297 ret = convert_read_used_space(&cctx);
2301 blocksize = cctx.blocksize;
2302 total_bytes = (u64)blocksize * (u64)cctx.block_count;
2303 if (blocksize < 4096) {
2304 fprintf(stderr, "block size is too small\n");
2307 if (btrfs_check_nodesize(nodesize, blocksize, features))
2309 fd = open(devname, O_RDWR);
2311 fprintf(stderr, "unable to open %s\n", devname);
2314 btrfs_parse_features_to_string(features_buf, features);
2315 if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2316 strcat(features_buf, " (default)");
2318 printf("create btrfs filesystem:\n");
2319 printf("\tblocksize: %u\n", blocksize);
2320 printf("\tnodesize: %u\n", nodesize);
2321 printf("\tfeatures: %s\n", features_buf);
2323 mkfs_cfg.label = cctx.volume_name;
2324 mkfs_cfg.num_bytes = total_bytes;
2325 mkfs_cfg.nodesize = nodesize;
2326 mkfs_cfg.sectorsize = blocksize;
2327 mkfs_cfg.stripesize = blocksize;
2328 mkfs_cfg.features = features;
2329 /* New convert need these space */
2330 mkfs_cfg.fs_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2331 mkfs_cfg.chunk_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2332 *(mkfs_cfg.fs_uuid) = '\0';
2333 *(mkfs_cfg.chunk_uuid) = '\0';
2335 ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2337 fprintf(stderr, "unable to create initial ctree: %s\n",
2342 root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2343 OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2345 fprintf(stderr, "unable to open ctree\n");
2348 ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2350 fprintf(stderr, "unable to setup the root tree\n");
2354 printf("creating %s image file.\n", cctx.convert_ops->name);
2355 ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2357 fprintf(stderr, "error allocating subvolume name: %s_saved\n",
2358 cctx.convert_ops->name);
2361 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2362 key.offset = (u64)-1;
2363 key.type = BTRFS_ROOT_ITEM_KEY;
2364 image_root = btrfs_read_fs_root(root->fs_info, &key);
2366 fprintf(stderr, "unable to create subvol\n");
2369 ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2370 mkfs_cfg.num_bytes, "image", datacsum);
2372 fprintf(stderr, "error during create_image %d\n", ret);
2376 printf("creating btrfs metadata.\n");
2377 ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2378 ctx.cur_copy_inodes = 0;
2381 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2383 task_start(ctx.info);
2385 ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2387 fprintf(stderr, "error during copy_inodes %d\n", ret);
2391 task_stop(ctx.info);
2392 task_deinit(ctx.info);
2395 image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2399 memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2400 if (copylabel == 1) {
2401 __strncpy_null(root->fs_info->super_copy->label,
2402 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2403 fprintf(stderr, "copy label '%s'\n",
2404 root->fs_info->super_copy->label);
2405 } else if (copylabel == -1) {
2406 strcpy(root->fs_info->super_copy->label, fslabel);
2407 fprintf(stderr, "set label to '%s'\n", fslabel);
2410 ret = close_ctree(root);
2412 fprintf(stderr, "error during close_ctree %d\n", ret);
2415 convert_close_fs(&cctx);
2416 clean_convert_context(&cctx);
2419 * If this step succeed, we get a mountable btrfs. Otherwise
2420 * the source fs is left unchanged.
2422 ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2424 fprintf(stderr, "unable to migrate super block\n");
2429 root = open_ctree_fd(fd, devname, 0,
2430 OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2432 fprintf(stderr, "unable to open ctree\n");
2435 root->fs_info->finalize_on_close = 1;
2439 printf("conversion complete.\n");
2442 clean_convert_context(&cctx);
2447 "WARNING: an error occurred during chunk mapping fixup, filesystem mountable but not finalized\n");
2449 fprintf(stderr, "conversion aborted\n");
2454 * Check if a non 1:1 mapped chunk can be rolled back.
2455 * For new convert, it's OK while for old convert it's not.
2457 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2459 struct btrfs_block_group_cache *bg;
2460 struct btrfs_key key;
2461 struct btrfs_path path;
2462 struct btrfs_root *extent_root = fs_info->extent_root;
2467 bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2470 bg_start = bg->key.objectid;
2471 bg_end = bg->key.objectid + bg->key.offset;
2473 key.objectid = bg_end;
2474 key.type = BTRFS_METADATA_ITEM_KEY;
2476 btrfs_init_path(&path);
2478 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2483 struct btrfs_extent_item *ei;
2485 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2493 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2494 if (key.type == BTRFS_METADATA_ITEM_KEY)
2496 /* Now it's EXTENT_ITEM_KEY only */
2497 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2498 struct btrfs_extent_item);
2500 * Found data extent, means this is old convert must follow 1:1
2503 if (btrfs_extent_flags(path.nodes[0], ei)
2504 & BTRFS_EXTENT_FLAG_DATA) {
2509 btrfs_release_path(&path);
2513 static int may_rollback(struct btrfs_root *root)
2515 struct btrfs_fs_info *info = root->fs_info;
2516 struct btrfs_multi_bio *multi = NULL;
2524 if (btrfs_super_num_devices(info->super_copy) != 1)
2527 bytenr = BTRFS_SUPER_INFO_OFFSET;
2528 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2531 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2532 &length, &multi, 0, NULL);
2534 if (ret == -ENOENT) {
2535 /* removed block group at the tail */
2536 if (length == (u64)-1)
2539 /* removed block group in the middle */
2545 num_stripes = multi->num_stripes;
2546 physical = multi->stripes[0].physical;
2549 if (num_stripes != 1) {
2550 error("num stripes for bytenr %llu is not 1", bytenr);
2555 * Extra check for new convert, as metadata chunk from new
2556 * convert is much more free than old convert, it doesn't need
2557 * to do 1:1 mapping.
2559 if (physical != bytenr) {
2561 * Check if it's a metadata chunk and has only metadata
2564 ret = may_rollback_chunk(info, bytenr);
2570 if (bytenr >= total_bytes)
2578 static int do_rollback(const char *devname)
2583 struct btrfs_root *root;
2584 struct btrfs_root *image_root;
2585 struct btrfs_root *chunk_root;
2586 struct btrfs_dir_item *dir;
2587 struct btrfs_inode_item *inode;
2588 struct btrfs_file_extent_item *fi;
2589 struct btrfs_trans_handle *trans;
2590 struct extent_buffer *leaf;
2591 struct btrfs_block_group_cache *cache1;
2592 struct btrfs_block_group_cache *cache2;
2593 struct btrfs_key key;
2594 struct btrfs_path path;
2595 struct extent_io_tree io_tree;
2610 extent_io_tree_init(&io_tree);
2612 fd = open(devname, O_RDWR);
2614 fprintf(stderr, "unable to open %s\n", devname);
2617 root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2619 fprintf(stderr, "unable to open ctree\n");
2622 ret = may_rollback(root);
2624 fprintf(stderr, "unable to do rollback\n");
2628 sectorsize = root->sectorsize;
2629 buf = malloc(sectorsize);
2631 fprintf(stderr, "unable to allocate memory\n");
2635 btrfs_init_path(&path);
2637 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2638 key.type = BTRFS_ROOT_BACKREF_KEY;
2639 key.offset = BTRFS_FS_TREE_OBJECTID;
2640 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2642 btrfs_release_path(&path);
2645 "ERROR: unable to convert ext2 image subvolume, is it deleted?\n");
2647 } else if (ret < 0) {
2649 "ERROR: unable to open ext2_saved, id=%llu: %s\n",
2650 (unsigned long long)key.objectid, strerror(-ret));
2654 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2655 key.type = BTRFS_ROOT_ITEM_KEY;
2656 key.offset = (u64)-1;
2657 image_root = btrfs_read_fs_root(root->fs_info, &key);
2658 if (!image_root || IS_ERR(image_root)) {
2659 fprintf(stderr, "unable to open subvol %llu\n",
2660 (unsigned long long)key.objectid);
2665 root_dir = btrfs_root_dirid(&root->root_item);
2666 dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2667 root_dir, name, strlen(name), 0);
2668 if (!dir || IS_ERR(dir)) {
2669 fprintf(stderr, "unable to find file %s\n", name);
2672 leaf = path.nodes[0];
2673 btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2674 btrfs_release_path(&path);
2676 objectid = key.objectid;
2678 ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2680 fprintf(stderr, "unable to find inode item\n");
2683 leaf = path.nodes[0];
2684 inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2685 total_bytes = btrfs_inode_size(leaf, inode);
2686 btrfs_release_path(&path);
2688 key.objectid = objectid;
2690 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2691 ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2693 fprintf(stderr, "unable to find first file extent\n");
2694 btrfs_release_path(&path);
2698 /* build mapping tree for the relocated blocks */
2699 for (offset = 0; offset < total_bytes; ) {
2700 leaf = path.nodes[0];
2701 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2702 ret = btrfs_next_leaf(root, &path);
2708 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2709 if (key.objectid != objectid || key.offset != offset ||
2710 btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2713 fi = btrfs_item_ptr(leaf, path.slots[0],
2714 struct btrfs_file_extent_item);
2715 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2717 if (btrfs_file_extent_compression(leaf, fi) ||
2718 btrfs_file_extent_encryption(leaf, fi) ||
2719 btrfs_file_extent_other_encoding(leaf, fi))
2722 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2723 /* skip holes and direct mapped extents */
2724 if (bytenr == 0 || bytenr == offset)
2727 bytenr += btrfs_file_extent_offset(leaf, fi);
2728 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2730 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2731 cache2 = btrfs_lookup_block_group(root->fs_info,
2732 offset + num_bytes - 1);
2734 * Here we must take consideration of old and new convert
2736 * For old convert case, sign, there is no consist chunk type
2737 * that will cover the extent. META/DATA/SYS are all possible.
2738 * Just ensure relocate one is in SYS chunk.
2739 * For new convert case, they are all covered by DATA chunk.
2741 * So, there is not valid chunk type check for it now.
2743 if (cache1 != cache2)
2746 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2747 EXTENT_LOCKED, GFP_NOFS);
2748 set_state_private(&io_tree, offset, bytenr);
2750 offset += btrfs_file_extent_num_bytes(leaf, fi);
2753 btrfs_release_path(&path);
2755 if (offset < total_bytes) {
2756 fprintf(stderr, "unable to build extent mapping\n");
2757 fprintf(stderr, "converted filesystem after balance is unable to rollback\n");
2761 first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2762 first_free &= ~((u64)sectorsize - 1);
2763 /* backup for extent #0 should exist */
2764 if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2765 fprintf(stderr, "no backup for the first extent\n");
2768 /* force no allocation from system block group */
2769 root->fs_info->system_allocs = -1;
2770 trans = btrfs_start_transaction(root, 1);
2773 * recow the whole chunk tree, this will remove all chunk tree blocks
2774 * from system block group
2776 chunk_root = root->fs_info->chunk_root;
2777 memset(&key, 0, sizeof(key));
2779 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2783 ret = btrfs_next_leaf(chunk_root, &path);
2787 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2788 btrfs_release_path(&path);
2790 btrfs_release_path(&path);
2795 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2799 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2800 num_bytes += btrfs_block_group_used(&cache1->item);
2802 offset = cache1->key.objectid + cache1->key.offset;
2804 /* only extent #0 left in system block group? */
2805 if (num_bytes > first_free) {
2806 fprintf(stderr, "unable to empty system block group\n");
2809 /* create a system chunk that maps the whole device */
2810 ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2812 fprintf(stderr, "unable to update system chunk\n");
2816 ret = btrfs_commit_transaction(trans, root);
2819 ret = close_ctree(root);
2821 fprintf(stderr, "error during close_ctree %d\n", ret);
2825 /* zero btrfs super block mirrors */
2826 memset(buf, 0, sectorsize);
2827 for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2828 bytenr = btrfs_sb_offset(i);
2829 if (bytenr >= total_bytes)
2831 ret = pwrite(fd, buf, sectorsize, bytenr);
2832 if (ret != sectorsize) {
2834 "error during zeroing superblock %d: %d\n",
2840 sb_bytenr = (u64)-1;
2841 /* copy all relocated blocks back */
2843 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2848 ret = get_state_private(&io_tree, start, &bytenr);
2851 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2854 while (start <= end) {
2855 if (start == BTRFS_SUPER_INFO_OFFSET) {
2859 ret = pread(fd, buf, sectorsize, bytenr);
2861 fprintf(stderr, "error during pread %d\n", ret);
2864 BUG_ON(ret != sectorsize);
2865 ret = pwrite(fd, buf, sectorsize, start);
2867 fprintf(stderr, "error during pwrite %d\n", ret);
2870 BUG_ON(ret != sectorsize);
2872 start += sectorsize;
2873 bytenr += sectorsize;
2879 fprintf(stderr, "error during fsync %d\n", ret);
2883 * finally, overwrite btrfs super block.
2885 ret = pread(fd, buf, sectorsize, sb_bytenr);
2887 fprintf(stderr, "error during pread %d\n", ret);
2890 BUG_ON(ret != sectorsize);
2891 ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2893 fprintf(stderr, "error during pwrite %d\n", ret);
2896 BUG_ON(ret != sectorsize);
2899 fprintf(stderr, "error during fsync %d\n", ret);
2905 extent_io_tree_cleanup(&io_tree);
2906 printf("rollback complete.\n");
2913 fprintf(stderr, "rollback aborted.\n");
2917 static void print_usage(void)
2919 printf("usage: btrfs-convert [options] device\n");
2920 printf("options:\n");
2921 printf("\t-d|--no-datasum disable data checksum, sets NODATASUM\n");
2922 printf("\t-i|--no-xattr ignore xattrs and ACLs\n");
2923 printf("\t-n|--no-inline disable inlining of small files to metadata\n");
2924 printf("\t-N|--nodesize SIZE set filesystem metadata nodesize\n");
2925 printf("\t-r|--rollback roll back to the original filesystem\n");
2926 printf("\t-l|--label LABEL set filesystem label\n");
2927 printf("\t-L|--copy-label use label from converted filesystem\n");
2928 printf("\t-p|--progress show converting progress (default)\n");
2929 printf("\t-O|--features LIST comma separated list of filesystem features\n");
2930 printf("\t--no-progress show only overview, not the detailed progress\n");
2933 int main(int argc, char *argv[])
2939 u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
2940 BTRFS_MKFS_DEFAULT_NODE_SIZE);
2943 int usage_error = 0;
2946 char fslabel[BTRFS_LABEL_SIZE];
2947 u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
2950 enum { GETOPT_VAL_NO_PROGRESS = 256 };
2951 static const struct option long_options[] = {
2952 { "no-progress", no_argument, NULL,
2953 GETOPT_VAL_NO_PROGRESS },
2954 { "no-datasum", no_argument, NULL, 'd' },
2955 { "no-inline", no_argument, NULL, 'n' },
2956 { "no-xattr", no_argument, NULL, 'i' },
2957 { "rollback", no_argument, NULL, 'r' },
2958 { "features", required_argument, NULL, 'O' },
2959 { "progress", no_argument, NULL, 'p' },
2960 { "label", required_argument, NULL, 'l' },
2961 { "copy-label", no_argument, NULL, 'L' },
2962 { "nodesize", required_argument, NULL, 'N' },
2963 { "help", no_argument, NULL, GETOPT_VAL_HELP},
2964 { NULL, 0, NULL, 0 }
2966 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
2981 nodesize = parse_size(optarg);
2988 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
2990 "WARNING: label too long, trimmed to %d bytes\n",
2991 BTRFS_LABEL_SIZE - 1);
2993 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
3002 char *orig = strdup(optarg);
3005 tmp = btrfs_parse_fs_features(tmp, &features);
3008 "Unrecognized filesystem feature '%s'\n",
3014 if (features & BTRFS_FEATURE_LIST_ALL) {
3015 btrfs_list_all_fs_features(
3016 ~BTRFS_CONVERT_ALLOWED_FEATURES);
3019 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
3022 btrfs_parse_features_to_string(buf,
3023 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
3025 "ERROR: features not allowed for convert: %s\n",
3032 case GETOPT_VAL_NO_PROGRESS:
3035 case GETOPT_VAL_HELP:
3038 return c != GETOPT_VAL_HELP;
3042 if (check_argc_exact(argc - optind, 1)) {
3047 if (rollback && (!datacsum || noxattr || !packing)) {
3049 "Usage error: -d, -i, -n options do not apply to rollback\n");
3058 file = argv[optind];
3059 ret = check_mounted(file);
3061 fprintf(stderr, "Could not check mount status: %s\n",
3065 fprintf(stderr, "%s is mounted\n", file);
3070 ret = do_rollback(file);
3072 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3073 copylabel, fslabel, progress, features);