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"
42 #include <ext2fs/ext2_fs.h>
43 #include <ext2fs/ext2fs.h>
44 #include <ext2fs/ext2_ext_attr.h>
46 #define INO_OFFSET (BTRFS_FIRST_FREE_OBJECTID - EXT2_ROOT_INO)
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)
62 #define CONV_IMAGE_SUBVOL_OBJECTID BTRFS_FIRST_FREE_OBJECTID
65 uint32_t max_copy_inodes;
66 uint32_t cur_copy_inodes;
67 struct task_info *info;
70 static void *print_copied_inodes(void *p)
72 struct task_ctx *priv = p;
73 const char work_indicator[] = { '.', 'o', 'O', 'o' };
76 task_period_start(priv->info, 1000 /* 1s */);
79 printf("copy inodes [%c] [%10d/%10d]\r",
80 work_indicator[count % 4], priv->cur_copy_inodes,
81 priv->max_copy_inodes);
83 task_period_wait(priv->info);
89 static int after_copied_inodes(void *p)
97 struct btrfs_convert_context;
98 struct btrfs_convert_operations {
100 int (*open_fs)(struct btrfs_convert_context *cctx, const char *devname);
101 int (*read_used_space)(struct btrfs_convert_context *cctx);
102 int (*copy_inodes)(struct btrfs_convert_context *cctx,
103 struct btrfs_root *root, int datacsum,
104 int packing, int noxattr, struct task_ctx *p);
105 void (*close_fs)(struct btrfs_convert_context *cctx);
108 static void init_convert_context(struct btrfs_convert_context *cctx)
110 cache_tree_init(&cctx->used);
111 cache_tree_init(&cctx->data_chunks);
112 cache_tree_init(&cctx->free);
115 static void clean_convert_context(struct btrfs_convert_context *cctx)
117 free_extent_cache_tree(&cctx->used);
118 free_extent_cache_tree(&cctx->data_chunks);
119 free_extent_cache_tree(&cctx->free);
122 static inline int copy_inodes(struct btrfs_convert_context *cctx,
123 struct btrfs_root *root, int datacsum,
124 int packing, int noxattr, struct task_ctx *p)
126 return cctx->convert_ops->copy_inodes(cctx, root, datacsum, packing,
130 static inline void convert_close_fs(struct btrfs_convert_context *cctx)
132 cctx->convert_ops->close_fs(cctx);
135 static int intersect_with_sb(u64 bytenr, u64 num_bytes)
140 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
141 offset = btrfs_sb_offset(i);
142 offset &= ~((u64)BTRFS_STRIPE_LEN - 1);
144 if (bytenr < offset + BTRFS_STRIPE_LEN &&
145 bytenr + num_bytes > offset)
151 static int convert_insert_dirent(struct btrfs_trans_handle *trans,
152 struct btrfs_root *root,
153 const char *name, size_t name_len,
154 u64 dir, u64 objectid,
155 u8 file_type, u64 index_cnt,
156 struct btrfs_inode_item *inode)
160 struct btrfs_key location = {
161 .objectid = objectid,
163 .type = BTRFS_INODE_ITEM_KEY,
166 ret = btrfs_insert_dir_item(trans, root, name, name_len,
167 dir, &location, file_type, index_cnt);
170 ret = btrfs_insert_inode_ref(trans, root, name, name_len,
171 objectid, dir, index_cnt);
174 inode_size = btrfs_stack_inode_size(inode) + name_len * 2;
175 btrfs_set_stack_inode_size(inode, inode_size);
180 static int read_disk_extent(struct btrfs_root *root, u64 bytenr,
181 u32 num_bytes, char *buffer)
184 struct btrfs_fs_devices *fs_devs = root->fs_info->fs_devices;
186 ret = pread(fs_devs->latest_bdev, buffer, num_bytes, bytenr);
187 if (ret != num_bytes)
196 static int csum_disk_extent(struct btrfs_trans_handle *trans,
197 struct btrfs_root *root,
198 u64 disk_bytenr, u64 num_bytes)
200 u32 blocksize = root->sectorsize;
205 buffer = malloc(blocksize);
208 for (offset = 0; offset < num_bytes; offset += blocksize) {
209 ret = read_disk_extent(root, disk_bytenr + offset,
213 ret = btrfs_csum_file_block(trans,
214 root->fs_info->csum_root,
215 disk_bytenr + num_bytes,
216 disk_bytenr + offset,
225 struct blk_iterate_data {
226 struct btrfs_trans_handle *trans;
227 struct btrfs_root *root;
228 struct btrfs_root *convert_root;
229 struct btrfs_inode_item *inode;
240 static void init_blk_iterate_data(struct blk_iterate_data *data,
241 struct btrfs_trans_handle *trans,
242 struct btrfs_root *root,
243 struct btrfs_inode_item *inode,
244 u64 objectid, int checksum)
246 struct btrfs_key key;
251 data->objectid = objectid;
252 data->first_block = 0;
253 data->disk_block = 0;
254 data->num_blocks = 0;
255 data->boundary = (u64)-1;
256 data->checksum = checksum;
259 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
260 key.type = BTRFS_ROOT_ITEM_KEY;
261 key.offset = (u64)-1;
262 data->convert_root = btrfs_read_fs_root(root->fs_info, &key);
263 /* Impossible as we just opened it before */
264 BUG_ON(!data->convert_root || IS_ERR(data->convert_root));
265 data->convert_ino = BTRFS_FIRST_FREE_OBJECTID + 1;
269 * Record a file extent in original filesystem into btrfs one.
270 * The special point is, old disk_block can point to a reserved range.
271 * So here, we don't use disk_block directly but search convert_root
272 * to get the real disk_bytenr.
274 static int record_file_blocks(struct blk_iterate_data *data,
275 u64 file_block, u64 disk_block, u64 num_blocks)
278 struct btrfs_root *root = data->root;
279 struct btrfs_root *convert_root = data->convert_root;
280 struct btrfs_path *path;
281 u64 file_pos = file_block * root->sectorsize;
282 u64 old_disk_bytenr = disk_block * root->sectorsize;
283 u64 num_bytes = num_blocks * root->sectorsize;
284 u64 cur_off = old_disk_bytenr;
286 /* Hole, pass it to record_file_extent directly */
287 if (old_disk_bytenr == 0)
288 return btrfs_record_file_extent(data->trans, root,
289 data->objectid, data->inode, file_pos, 0,
292 path = btrfs_alloc_path();
297 * Search real disk bytenr from convert root
299 while (cur_off < old_disk_bytenr + num_bytes) {
300 struct btrfs_key key;
301 struct btrfs_file_extent_item *fi;
302 struct extent_buffer *node;
304 u64 extent_disk_bytenr;
305 u64 extent_num_bytes;
306 u64 real_disk_bytenr;
309 key.objectid = data->convert_ino;
310 key.type = BTRFS_EXTENT_DATA_KEY;
311 key.offset = cur_off;
313 ret = btrfs_search_slot(NULL, convert_root, &key, path, 0, 0);
317 ret = btrfs_previous_item(convert_root, path,
319 BTRFS_EXTENT_DATA_KEY);
327 node = path->nodes[0];
328 slot = path->slots[0];
329 btrfs_item_key_to_cpu(node, &key, slot);
330 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY ||
331 key.objectid != data->convert_ino ||
332 key.offset > cur_off);
333 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
334 extent_disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
335 extent_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
336 BUG_ON(cur_off - key.offset >= extent_num_bytes);
337 btrfs_release_path(path);
339 if (extent_disk_bytenr)
340 real_disk_bytenr = cur_off - key.offset +
343 real_disk_bytenr = 0;
344 cur_len = min(key.offset + extent_num_bytes,
345 old_disk_bytenr + num_bytes) - cur_off;
346 ret = btrfs_record_file_extent(data->trans, data->root,
347 data->objectid, data->inode, file_pos,
348 real_disk_bytenr, cur_len);
355 * No need to care about csum
356 * As every byte of old fs image is calculated for csum, no
357 * need to waste CPU cycles now.
360 btrfs_free_path(path);
364 static int block_iterate_proc(u64 disk_block, u64 file_block,
365 struct blk_iterate_data *idata)
370 struct btrfs_root *root = idata->root;
371 struct btrfs_block_group_cache *cache;
372 u64 bytenr = disk_block * root->sectorsize;
374 sb_region = intersect_with_sb(bytenr, root->sectorsize);
375 do_barrier = sb_region || disk_block >= idata->boundary;
376 if ((idata->num_blocks > 0 && do_barrier) ||
377 (file_block > idata->first_block + idata->num_blocks) ||
378 (disk_block != idata->disk_block + idata->num_blocks)) {
379 if (idata->num_blocks > 0) {
380 ret = record_file_blocks(idata, idata->first_block,
385 idata->first_block += idata->num_blocks;
386 idata->num_blocks = 0;
388 if (file_block > idata->first_block) {
389 ret = record_file_blocks(idata, idata->first_block,
390 0, file_block - idata->first_block);
396 bytenr += BTRFS_STRIPE_LEN - 1;
397 bytenr &= ~((u64)BTRFS_STRIPE_LEN - 1);
399 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
401 bytenr = cache->key.objectid + cache->key.offset;
404 idata->first_block = file_block;
405 idata->disk_block = disk_block;
406 idata->boundary = bytenr / root->sectorsize;
413 static int create_image_file_range(struct btrfs_trans_handle *trans,
414 struct btrfs_root *root,
415 struct cache_tree *used,
416 struct btrfs_inode_item *inode,
417 u64 ino, u64 bytenr, u64 *ret_len,
420 struct cache_extent *cache;
421 struct btrfs_block_group_cache *bg_cache;
427 BUG_ON(bytenr != round_down(bytenr, root->sectorsize));
428 BUG_ON(len != round_down(len, root->sectorsize));
429 len = min_t(u64, len, BTRFS_MAX_EXTENT_SIZE);
432 * Skip sb ranges first
433 * [0, 1M), [sb_offset(1), +64K), [sb_offset(2), +64K].
435 * Or we will insert a hole into current image file, and later
436 * migrate block will fail as there is already a file extent.
438 if (bytenr < 1024 * 1024) {
439 *ret_len = 1024 * 1024 - bytenr;
442 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
443 u64 cur = btrfs_sb_offset(i);
445 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
446 *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
450 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
451 u64 cur = btrfs_sb_offset(i);
456 * May still need to go through file extent inserts
458 if (bytenr < cur && bytenr + len >= cur) {
459 len = min_t(u64, len, cur - bytenr);
465 * Drop out, no need to insert anything
467 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
468 *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
473 cache = search_cache_extent(used, bytenr);
475 if (cache->start <= bytenr) {
477 * |///////Used///////|
481 len = min_t(u64, len, cache->start + cache->size -
483 disk_bytenr = bytenr;
490 len = min(len, cache->start - bytenr);
505 /* Check if the range is in a data block group */
506 bg_cache = btrfs_lookup_block_group(root->fs_info, bytenr);
509 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
512 /* The extent should never cross block group boundary */
513 len = min_t(u64, len, bg_cache->key.objectid +
514 bg_cache->key.offset - bytenr);
517 BUG_ON(len != round_down(len, root->sectorsize));
518 ret = btrfs_record_file_extent(trans, root, ino, inode, bytenr,
524 ret = csum_disk_extent(trans, root, bytenr, len);
530 * Relocate old fs data in one reserved ranges
532 * Since all old fs data in reserved range is not covered by any chunk nor
533 * data extent, we don't need to handle any reference but add new
534 * extent/reference, which makes codes more clear
536 static int migrate_one_reserved_range(struct btrfs_trans_handle *trans,
537 struct btrfs_root *root,
538 struct cache_tree *used,
539 struct btrfs_inode_item *inode, int fd,
540 u64 ino, u64 start, u64 len, int datacsum)
544 u64 hole_start = start;
546 struct cache_extent *cache;
547 struct btrfs_key key;
548 struct extent_buffer *eb;
551 while (cur_off < start + len) {
552 cache = lookup_cache_extent(used, cur_off, cur_len);
555 cur_off = max(cache->start, cur_off);
556 cur_len = min(cache->start + cache->size, start + len) -
558 BUG_ON(cur_len < root->sectorsize);
560 /* reserve extent for the data */
561 ret = btrfs_reserve_extent(trans, root, cur_len, 0, 0, (u64)-1,
566 eb = malloc(sizeof(*eb) + cur_len);
572 ret = pread(fd, eb->data, cur_len, cur_off);
574 ret = (ret < 0 ? ret : -EIO);
578 eb->start = key.objectid;
579 eb->len = key.offset;
582 ret = write_and_map_eb(trans, root, eb);
587 /* Now handle extent item and file extent things */
588 ret = btrfs_record_file_extent(trans, root, ino, inode, cur_off,
589 key.objectid, key.offset);
592 /* Finally, insert csum items */
594 ret = csum_disk_extent(trans, root, key.objectid,
597 /* Don't forget to insert hole */
598 hole_len = cur_off - hole_start;
600 ret = btrfs_record_file_extent(trans, root, ino, inode,
601 hole_start, 0, hole_len);
606 cur_off += key.offset;
607 hole_start = cur_off;
608 cur_len = start + len - cur_off;
611 if (start + len - hole_start > 0)
612 ret = btrfs_record_file_extent(trans, root, ino, inode,
613 hole_start, 0, start + len - hole_start);
618 * Relocate the used ext2 data in reserved ranges
620 * [btrfs_sb_offset(1), +BTRFS_STRIPE_LEN)
621 * [btrfs_sb_offset(2), +BTRFS_STRIPE_LEN)
623 static int migrate_reserved_ranges(struct btrfs_trans_handle *trans,
624 struct btrfs_root *root,
625 struct cache_tree *used,
626 struct btrfs_inode_item *inode, int fd,
627 u64 ino, u64 total_bytes, int datacsum)
635 cur_len = 1024 * 1024;
636 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
637 cur_off, cur_len, datacsum);
641 /* second sb(fisrt sb is included in 0~1M) */
642 cur_off = btrfs_sb_offset(1);
643 cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
644 if (cur_off > total_bytes)
646 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
647 cur_off, cur_len, datacsum);
652 cur_off = btrfs_sb_offset(2);
653 cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
654 if (cur_off > total_bytes)
656 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
657 cur_off, cur_len, datacsum);
662 * Helper for expand and merge extent_cache for wipe_one_reserved_range() to
663 * handle wiping a range that exists in cache.
665 static int _expand_extent_cache(struct cache_tree *tree,
666 struct cache_extent *entry,
667 u64 min_stripe_size, int backward)
669 struct cache_extent *ce;
672 if (entry->size >= min_stripe_size)
674 diff = min_stripe_size - entry->size;
677 ce = prev_cache_extent(entry);
680 if (ce->start + ce->size >= entry->start - diff) {
681 /* Directly merge with previous extent */
682 ce->size = entry->start + entry->size - ce->start;
683 remove_cache_extent(tree, entry);
688 /* No overlap, normal extent */
689 if (entry->start < diff) {
690 error("cannot find space for data chunk layout");
693 entry->start -= diff;
697 ce = next_cache_extent(entry);
700 if (entry->start + entry->size + diff >= ce->start) {
701 /* Directly merge with next extent */
702 entry->size = ce->start + ce->size - entry->start;
703 remove_cache_extent(tree, ce);
713 * Remove one reserve range from given cache tree
714 * if min_stripe_size is non-zero, it will ensure for split case,
715 * all its split cache extent is no smaller than @min_strip_size / 2.
717 static int wipe_one_reserved_range(struct cache_tree *tree,
718 u64 start, u64 len, u64 min_stripe_size,
721 struct cache_extent *cache;
724 BUG_ON(ensure_size && min_stripe_size == 0);
726 * The logical here is simplified to handle special cases only
727 * So we don't need to consider merge case for ensure_size
729 BUG_ON(min_stripe_size && (min_stripe_size < len * 2 ||
730 min_stripe_size / 2 < BTRFS_STRIPE_LEN));
732 /* Also, wipe range should already be aligned */
733 BUG_ON(start != round_down(start, BTRFS_STRIPE_LEN) ||
734 start + len != round_up(start + len, BTRFS_STRIPE_LEN));
736 min_stripe_size /= 2;
738 cache = lookup_cache_extent(tree, start, len);
742 if (start <= cache->start) {
744 * |--------cache---------|
747 BUG_ON(start + len <= cache->start);
750 * The wipe size is smaller than min_stripe_size / 2,
751 * so the result length should still meet min_stripe_size
752 * And no need to do alignment
754 cache->size -= (start + len - cache->start);
755 if (cache->size == 0) {
756 remove_cache_extent(tree, cache);
761 BUG_ON(ensure_size && cache->size < min_stripe_size);
763 cache->start = start + len;
765 } else if (start > cache->start && start + len < cache->start +
768 * |-------cache-----|
771 u64 old_start = cache->start;
772 u64 old_len = cache->size;
773 u64 insert_start = start + len;
776 cache->size = start - cache->start;
777 /* Expand the leading half part if needed */
778 if (ensure_size && cache->size < min_stripe_size) {
779 ret = _expand_extent_cache(tree, cache,
785 /* And insert the new one */
786 insert_len = old_start + old_len - start - len;
787 ret = add_merge_cache_extent(tree, insert_start, insert_len);
791 /* Expand the last half part if needed */
792 if (ensure_size && insert_len < min_stripe_size) {
793 cache = lookup_cache_extent(tree, insert_start,
795 if (!cache || cache->start != insert_start ||
796 cache->size != insert_len)
798 ret = _expand_extent_cache(tree, cache,
807 * Wipe len should be small enough and no need to expand the
810 cache->size = start - cache->start;
811 BUG_ON(ensure_size && cache->size < min_stripe_size);
816 * Remove reserved ranges from given cache_tree
818 * It will remove the following ranges
820 * 2) 2nd superblock, +64K (make sure chunks are 64K aligned)
821 * 3) 3rd superblock, +64K
823 * @min_stripe must be given for safety check
824 * and if @ensure_size is given, it will ensure affected cache_extent will be
825 * larger than min_stripe_size
827 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
832 ret = wipe_one_reserved_range(tree, 0, 1024 * 1024, min_stripe_size,
836 ret = wipe_one_reserved_range(tree, btrfs_sb_offset(1),
837 BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
840 ret = wipe_one_reserved_range(tree, btrfs_sb_offset(2),
841 BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
845 static int calculate_available_space(struct btrfs_convert_context *cctx)
847 struct cache_tree *used = &cctx->used;
848 struct cache_tree *data_chunks = &cctx->data_chunks;
849 struct cache_tree *free = &cctx->free;
850 struct cache_extent *cache;
853 * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
854 * works without need to consider overlap
856 u64 min_stripe_size = 2 * 16 * 1024 * 1024;
859 /* Calculate data_chunks */
860 for (cache = first_cache_extent(used); cache;
861 cache = next_cache_extent(cache)) {
864 if (cache->start + cache->size < cur_off)
866 if (cache->start > cur_off + min_stripe_size)
867 cur_off = cache->start;
868 cur_len = max(cache->start + cache->size - cur_off,
870 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
876 * remove reserved ranges, so we won't ever bother relocating an old
877 * filesystem extent to other place.
879 ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
885 * Calculate free space
886 * Always round up the start bytenr, to avoid metadata extent corss
887 * stripe boundary, as later mkfs_convert() won't have all the extent
890 for (cache = first_cache_extent(data_chunks); cache;
891 cache = next_cache_extent(cache)) {
892 if (cache->start < cur_off)
894 if (cache->start > cur_off) {
898 len = cache->start - round_up(cur_off,
900 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
902 ret = add_merge_cache_extent(free, insert_start, len);
906 cur_off = cache->start + cache->size;
908 /* Don't forget the last range */
909 if (cctx->total_bytes > cur_off) {
910 u64 len = cctx->total_bytes - cur_off;
913 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
915 ret = add_merge_cache_extent(free, insert_start, len);
920 /* Remove reserved bytes */
921 ret = wipe_reserved_ranges(free, min_stripe_size, 0);
927 * Read used space, and since we have the used space,
928 * calcuate data_chunks and free for later mkfs
930 static int convert_read_used_space(struct btrfs_convert_context *cctx)
934 ret = cctx->convert_ops->read_used_space(cctx);
938 ret = calculate_available_space(cctx);
943 * Create the fs image file of old filesystem.
945 * This is completely fs independent as we have cctx->used, only
946 * need to create file extents pointing to all the positions.
948 static int create_image(struct btrfs_root *root,
949 struct btrfs_mkfs_config *cfg,
950 struct btrfs_convert_context *cctx, int fd,
951 u64 size, char *name, int datacsum)
953 struct btrfs_inode_item buf;
954 struct btrfs_trans_handle *trans;
955 struct btrfs_path *path = NULL;
956 struct btrfs_key key;
957 struct cache_extent *cache;
958 struct cache_tree used_tmp;
961 u64 flags = BTRFS_INODE_READONLY;
965 flags |= BTRFS_INODE_NODATASUM;
967 trans = btrfs_start_transaction(root, 1);
971 cache_tree_init(&used_tmp);
973 ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
977 ret = btrfs_new_inode(trans, root, ino, 0400 | S_IFREG);
980 ret = btrfs_change_inode_flags(trans, root, ino, flags);
983 ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
984 strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
988 path = btrfs_alloc_path();
994 key.type = BTRFS_INODE_ITEM_KEY;
997 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
999 ret = (ret > 0 ? -ENOENT : ret);
1002 read_extent_buffer(path->nodes[0], &buf,
1003 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1005 btrfs_release_path(path);
1008 * Create a new used space cache, which doesn't contain the reserved
1011 for (cache = first_cache_extent(&cctx->used); cache;
1012 cache = next_cache_extent(cache)) {
1013 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1017 ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1022 * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1023 * can't handle bytenr 0(will consider it as a hole)
1026 while (cur < size) {
1027 u64 len = size - cur;
1029 ret = create_image_file_range(trans, root, &used_tmp,
1030 &buf, ino, cur, &len, datacsum);
1035 /* Handle the reserved ranges */
1036 ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1037 cfg->num_bytes, datacsum);
1041 key.type = BTRFS_INODE_ITEM_KEY;
1043 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1045 ret = (ret > 0 ? -ENOENT : ret);
1048 btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1049 write_extent_buffer(path->nodes[0], &buf,
1050 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1053 free_extent_cache_tree(&used_tmp);
1054 btrfs_free_path(path);
1055 btrfs_commit_transaction(trans, root);
1059 static struct btrfs_root* link_subvol(struct btrfs_root *root,
1060 const char *base, u64 root_objectid)
1062 struct btrfs_trans_handle *trans;
1063 struct btrfs_fs_info *fs_info = root->fs_info;
1064 struct btrfs_root *tree_root = fs_info->tree_root;
1065 struct btrfs_root *new_root = NULL;
1066 struct btrfs_path *path;
1067 struct btrfs_inode_item *inode_item;
1068 struct extent_buffer *leaf;
1069 struct btrfs_key key;
1070 u64 dirid = btrfs_root_dirid(&root->root_item);
1072 char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1078 if (len == 0 || len > BTRFS_NAME_LEN)
1081 path = btrfs_alloc_path();
1085 key.objectid = dirid;
1086 key.type = BTRFS_DIR_INDEX_KEY;
1087 key.offset = (u64)-1;
1089 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1091 error("search for DIR_INDEX dirid %llu failed: %d",
1092 (unsigned long long)dirid, ret);
1096 if (path->slots[0] > 0) {
1098 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1099 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1100 index = key.offset + 1;
1102 btrfs_release_path(path);
1104 trans = btrfs_start_transaction(root, 1);
1106 error("unable to start transaction");
1110 key.objectid = dirid;
1112 key.type = BTRFS_INODE_ITEM_KEY;
1114 ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1116 error("search for INODE_ITEM %llu failed: %d",
1117 (unsigned long long)dirid, ret);
1120 leaf = path->nodes[0];
1121 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1122 struct btrfs_inode_item);
1124 key.objectid = root_objectid;
1125 key.offset = (u64)-1;
1126 key.type = BTRFS_ROOT_ITEM_KEY;
1128 memcpy(buf, base, len);
1129 for (i = 0; i < 1024; i++) {
1130 ret = btrfs_insert_dir_item(trans, root, buf, len,
1131 dirid, &key, BTRFS_FT_DIR, index);
1134 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1135 if (len < 1 || len > BTRFS_NAME_LEN) {
1143 btrfs_set_inode_size(leaf, inode_item, len * 2 +
1144 btrfs_inode_size(leaf, inode_item));
1145 btrfs_mark_buffer_dirty(leaf);
1146 btrfs_release_path(path);
1148 /* add the backref first */
1149 ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1150 BTRFS_ROOT_BACKREF_KEY,
1151 root->root_key.objectid,
1152 dirid, index, buf, len);
1154 error("unable to add root backref for %llu: %d",
1155 root->root_key.objectid, ret);
1159 /* now add the forward ref */
1160 ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1161 BTRFS_ROOT_REF_KEY, root_objectid,
1162 dirid, index, buf, len);
1164 error("unable to add root ref for %llu: %d",
1165 root->root_key.objectid, ret);
1169 ret = btrfs_commit_transaction(trans, root);
1171 error("transaction commit failed: %d", ret);
1175 new_root = btrfs_read_fs_root(fs_info, &key);
1176 if (IS_ERR(new_root)) {
1177 error("unable to fs read root: %lu", PTR_ERR(new_root));
1181 btrfs_free_path(path);
1185 static int create_subvol(struct btrfs_trans_handle *trans,
1186 struct btrfs_root *root, u64 root_objectid)
1188 struct extent_buffer *tmp;
1189 struct btrfs_root *new_root;
1190 struct btrfs_key key;
1191 struct btrfs_root_item root_item;
1194 ret = btrfs_copy_root(trans, root, root->node, &tmp,
1199 memcpy(&root_item, &root->root_item, sizeof(root_item));
1200 btrfs_set_root_bytenr(&root_item, tmp->start);
1201 btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
1202 btrfs_set_root_generation(&root_item, trans->transid);
1203 free_extent_buffer(tmp);
1205 key.objectid = root_objectid;
1206 key.type = BTRFS_ROOT_ITEM_KEY;
1207 key.offset = trans->transid;
1208 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1211 key.offset = (u64)-1;
1212 new_root = btrfs_read_fs_root(root->fs_info, &key);
1213 if (!new_root || IS_ERR(new_root)) {
1214 error("unable to fs read root: %lu", PTR_ERR(new_root));
1215 return PTR_ERR(new_root);
1218 ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
1224 * New make_btrfs() has handle system and meta chunks quite well.
1225 * So only need to add remaining data chunks.
1227 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
1228 struct btrfs_fs_info *fs_info,
1229 struct btrfs_mkfs_config *cfg,
1230 struct btrfs_convert_context *cctx)
1232 struct btrfs_root *extent_root = fs_info->extent_root;
1233 struct cache_tree *data_chunks = &cctx->data_chunks;
1234 struct cache_extent *cache;
1239 * Don't create data chunk over 10% of the convert device
1240 * And for single chunk, don't create chunk larger than 1G.
1242 max_chunk_size = cfg->num_bytes / 10;
1243 max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
1244 max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
1246 for (cache = first_cache_extent(data_chunks); cache;
1247 cache = next_cache_extent(cache)) {
1248 u64 cur = cache->start;
1250 while (cur < cache->start + cache->size) {
1252 u64 cur_backup = cur;
1254 len = min(max_chunk_size,
1255 cache->start + cache->size - cur);
1256 ret = btrfs_alloc_data_chunk(trans, extent_root,
1258 BTRFS_BLOCK_GROUP_DATA, 1);
1261 ret = btrfs_make_block_group(trans, extent_root, 0,
1262 BTRFS_BLOCK_GROUP_DATA,
1263 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1274 * Init the temp btrfs to a operational status.
1276 * It will fix the extent usage accounting(XXX: Do we really need?) and
1277 * insert needed data chunks, to ensure all old fs data extents are covered
1278 * by DATA chunks, preventing wrong chunks are allocated.
1280 * And also create convert image subvolume and relocation tree.
1281 * (XXX: Not need again?)
1282 * But the convert image subvolume is *NOT* linked to fs tree yet.
1284 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
1285 struct btrfs_convert_context *cctx, int datacsum,
1286 int packing, int noxattr)
1288 struct btrfs_key location;
1289 struct btrfs_trans_handle *trans;
1290 struct btrfs_fs_info *fs_info = root->fs_info;
1294 * Don't alloc any metadata/system chunk, as we don't want
1295 * any meta/sys chunk allcated before all data chunks are inserted.
1296 * Or we screw up the chunk layout just like the old implement.
1298 fs_info->avoid_sys_chunk_alloc = 1;
1299 fs_info->avoid_meta_chunk_alloc = 1;
1300 trans = btrfs_start_transaction(root, 1);
1302 error("unable to start transaction");
1306 ret = btrfs_fix_block_accounting(trans, root);
1309 ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
1312 ret = btrfs_make_root_dir(trans, fs_info->tree_root,
1313 BTRFS_ROOT_TREE_DIR_OBJECTID);
1316 memcpy(&location, &root->root_key, sizeof(location));
1317 location.offset = (u64)-1;
1318 ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
1319 btrfs_super_root_dir(fs_info->super_copy),
1320 &location, BTRFS_FT_DIR, 0);
1323 ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
1325 btrfs_super_root_dir(fs_info->super_copy), 0);
1328 btrfs_set_root_dirid(&fs_info->fs_root->root_item,
1329 BTRFS_FIRST_FREE_OBJECTID);
1331 /* subvol for fs image file */
1332 ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
1334 error("failed to create subvolume image root: %d", ret);
1337 /* subvol for data relocation tree */
1338 ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1340 error("failed to create DATA_RELOC root: %d", ret);
1344 ret = btrfs_commit_transaction(trans, root);
1345 fs_info->avoid_sys_chunk_alloc = 0;
1346 fs_info->avoid_meta_chunk_alloc = 0;
1352 * Migrate super block to its default position and zero 0 ~ 16k
1354 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
1357 struct extent_buffer *buf;
1358 struct btrfs_super_block *super;
1362 buf = malloc(sizeof(*buf) + sectorsize);
1366 buf->len = sectorsize;
1367 ret = pread(fd, buf->data, sectorsize, old_bytenr);
1368 if (ret != sectorsize)
1371 super = (struct btrfs_super_block *)buf->data;
1372 BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
1373 btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
1375 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1376 ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
1377 if (ret != sectorsize)
1384 memset(buf->data, 0, sectorsize);
1385 for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
1386 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
1387 if (len > sectorsize)
1389 ret = pwrite(fd, buf->data, len, bytenr);
1391 fprintf(stderr, "unable to zero fill device\n");
1405 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
1407 struct btrfs_chunk *chunk;
1408 struct btrfs_disk_key *key;
1409 u32 sectorsize = btrfs_super_sectorsize(super);
1411 key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1412 chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1413 sizeof(struct btrfs_disk_key));
1415 btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1416 btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1417 btrfs_set_disk_key_offset(key, 0);
1419 btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
1420 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1421 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1422 btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1423 btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1424 btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1425 btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1426 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1427 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1428 chunk->stripe.devid = super->dev_item.devid;
1429 btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1430 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1431 btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1435 #if BTRFSCONVERT_EXT2
1438 * Open Ext2fs in readonly mode, read block allocation bitmap and
1439 * inode bitmap into memory.
1441 static int ext2_open_fs(struct btrfs_convert_context *cctx, const char *name)
1444 ext2_filsys ext2_fs;
1448 ret = ext2fs_open(name, 0, 0, 0, unix_io_manager, &ext2_fs);
1450 fprintf(stderr, "ext2fs_open: %s\n", error_message(ret));
1454 * We need to know exactly the used space, some RO compat flags like
1455 * BIGALLOC will affect how used space is present.
1456 * So we need manuall check any unsupported RO compat flags
1458 ro_feature = ext2_fs->super->s_feature_ro_compat;
1459 if (ro_feature & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
1461 "unsupported RO features detected: %x, abort convert to avoid possible corruption",
1462 ro_feature & ~EXT2_LIB_FEATURE_COMPAT_SUPP);
1465 ret = ext2fs_read_inode_bitmap(ext2_fs);
1467 fprintf(stderr, "ext2fs_read_inode_bitmap: %s\n",
1468 error_message(ret));
1471 ret = ext2fs_read_block_bitmap(ext2_fs);
1473 fprintf(stderr, "ext2fs_read_block_bitmap: %s\n",
1474 error_message(ret));
1478 * search each block group for a free inode. this set up
1479 * uninit block/inode bitmaps appropriately.
1482 while (ino <= ext2_fs->super->s_inodes_count) {
1484 ext2fs_new_inode(ext2_fs, ino, 0, NULL, &foo);
1485 ino += EXT2_INODES_PER_GROUP(ext2_fs->super);
1488 if (!(ext2_fs->super->s_feature_incompat &
1489 EXT2_FEATURE_INCOMPAT_FILETYPE)) {
1490 fprintf(stderr, "filetype feature is missing\n");
1494 cctx->fs_data = ext2_fs;
1495 cctx->blocksize = ext2_fs->blocksize;
1496 cctx->block_count = ext2_fs->super->s_blocks_count;
1497 cctx->total_bytes = ext2_fs->blocksize * ext2_fs->super->s_blocks_count;
1498 cctx->volume_name = strndup(ext2_fs->super->s_volume_name, 16);
1499 cctx->first_data_block = ext2_fs->super->s_first_data_block;
1500 cctx->inodes_count = ext2_fs->super->s_inodes_count;
1501 cctx->free_inodes_count = ext2_fs->super->s_free_inodes_count;
1504 ext2fs_close(ext2_fs);
1508 static int __ext2_add_one_block(ext2_filsys fs, char *bitmap,
1509 unsigned long group_nr, struct cache_tree *used)
1511 unsigned long offset;
1515 offset = fs->super->s_first_data_block;
1516 offset /= EXT2FS_CLUSTER_RATIO(fs);
1517 offset += group_nr * EXT2_CLUSTERS_PER_GROUP(fs->super);
1518 for (i = 0; i < EXT2_CLUSTERS_PER_GROUP(fs->super); i++) {
1519 if (ext2fs_test_bit(i, bitmap)) {
1522 start = (i + offset) * EXT2FS_CLUSTER_RATIO(fs);
1523 start *= fs->blocksize;
1524 ret = add_merge_cache_extent(used, start,
1534 * Read all used ext2 space into cctx->used cache tree
1536 static int ext2_read_used_space(struct btrfs_convert_context *cctx)
1538 ext2_filsys fs = (ext2_filsys)cctx->fs_data;
1539 blk64_t blk_itr = EXT2FS_B2C(fs, fs->super->s_first_data_block);
1540 struct cache_tree *used_tree = &cctx->used;
1541 char *block_bitmap = NULL;
1546 block_nbytes = EXT2_CLUSTERS_PER_GROUP(fs->super) / 8;
1547 /* Shouldn't happen */
1548 BUG_ON(!fs->block_map);
1550 block_bitmap = malloc(block_nbytes);
1554 for (i = 0; i < fs->group_desc_count; i++) {
1555 ret = ext2fs_get_block_bitmap_range(fs->block_map, blk_itr,
1556 block_nbytes * 8, block_bitmap);
1558 error("fail to get bitmap from ext2, %s",
1562 ret = __ext2_add_one_block(fs, block_bitmap, i, used_tree);
1564 error("fail to build used space tree, %s",
1568 blk_itr += EXT2_CLUSTERS_PER_GROUP(fs->super);
1575 static void ext2_close_fs(struct btrfs_convert_context *cctx)
1577 if (cctx->volume_name) {
1578 free(cctx->volume_name);
1579 cctx->volume_name = NULL;
1581 ext2fs_close(cctx->fs_data);
1584 struct dir_iterate_data {
1585 struct btrfs_trans_handle *trans;
1586 struct btrfs_root *root;
1587 struct btrfs_inode_item *inode;
1594 static u8 ext2_filetype_conversion_table[EXT2_FT_MAX] = {
1595 [EXT2_FT_UNKNOWN] = BTRFS_FT_UNKNOWN,
1596 [EXT2_FT_REG_FILE] = BTRFS_FT_REG_FILE,
1597 [EXT2_FT_DIR] = BTRFS_FT_DIR,
1598 [EXT2_FT_CHRDEV] = BTRFS_FT_CHRDEV,
1599 [EXT2_FT_BLKDEV] = BTRFS_FT_BLKDEV,
1600 [EXT2_FT_FIFO] = BTRFS_FT_FIFO,
1601 [EXT2_FT_SOCK] = BTRFS_FT_SOCK,
1602 [EXT2_FT_SYMLINK] = BTRFS_FT_SYMLINK,
1605 static int ext2_dir_iterate_proc(ext2_ino_t dir, int entry,
1606 struct ext2_dir_entry *dirent,
1607 int offset, int blocksize,
1608 char *buf,void *priv_data)
1613 char dotdot[] = "..";
1614 struct dir_iterate_data *idata = (struct dir_iterate_data *)priv_data;
1617 name_len = dirent->name_len & 0xFF;
1619 objectid = dirent->inode + INO_OFFSET;
1620 if (!strncmp(dirent->name, dotdot, name_len)) {
1621 if (name_len == 2) {
1622 BUG_ON(idata->parent != 0);
1623 idata->parent = objectid;
1627 if (dirent->inode < EXT2_GOOD_OLD_FIRST_INO)
1630 file_type = dirent->name_len >> 8;
1631 BUG_ON(file_type > EXT2_FT_SYMLINK);
1633 ret = convert_insert_dirent(idata->trans, idata->root, dirent->name,
1634 name_len, idata->objectid, objectid,
1635 ext2_filetype_conversion_table[file_type],
1636 idata->index_cnt, idata->inode);
1638 idata->errcode = ret;
1646 static int ext2_create_dir_entries(struct btrfs_trans_handle *trans,
1647 struct btrfs_root *root, u64 objectid,
1648 struct btrfs_inode_item *btrfs_inode,
1649 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
1653 struct dir_iterate_data data = {
1656 .inode = btrfs_inode,
1657 .objectid = objectid,
1663 err = ext2fs_dir_iterate2(ext2_fs, ext2_ino, 0, NULL,
1664 ext2_dir_iterate_proc, &data);
1668 if (ret == 0 && data.parent == objectid) {
1669 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
1670 objectid, objectid, 0);
1674 fprintf(stderr, "ext2fs_dir_iterate2: %s\n", error_message(err));
1678 static int ext2_block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
1679 e2_blkcnt_t blockcnt, blk_t ref_block,
1680 int ref_offset, void *priv_data)
1683 struct blk_iterate_data *idata;
1684 idata = (struct blk_iterate_data *)priv_data;
1685 ret = block_iterate_proc(*blocknr, blockcnt, idata);
1687 idata->errcode = ret;
1694 * traverse file's data blocks, record these data blocks as file extents.
1696 static int ext2_create_file_extents(struct btrfs_trans_handle *trans,
1697 struct btrfs_root *root, u64 objectid,
1698 struct btrfs_inode_item *btrfs_inode,
1699 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1700 int datacsum, int packing)
1703 char *buffer = NULL;
1706 u32 sectorsize = root->sectorsize;
1707 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1708 struct blk_iterate_data data;
1710 init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
1713 err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
1714 NULL, ext2_block_iterate_proc, &data);
1720 if (packing && data.first_block == 0 && data.num_blocks > 0 &&
1721 inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
1722 u64 num_bytes = data.num_blocks * sectorsize;
1723 u64 disk_bytenr = data.disk_block * sectorsize;
1726 buffer = malloc(num_bytes);
1729 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
1732 if (num_bytes > inode_size)
1733 num_bytes = inode_size;
1734 ret = btrfs_insert_inline_extent(trans, root, objectid,
1735 0, buffer, num_bytes);
1738 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
1739 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
1740 } else if (data.num_blocks > 0) {
1741 ret = record_file_blocks(&data, data.first_block,
1742 data.disk_block, data.num_blocks);
1746 data.first_block += data.num_blocks;
1747 last_block = (inode_size + sectorsize - 1) / sectorsize;
1748 if (last_block > data.first_block) {
1749 ret = record_file_blocks(&data, data.first_block, 0,
1750 last_block - data.first_block);
1756 fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
1760 static int ext2_create_symbol_link(struct btrfs_trans_handle *trans,
1761 struct btrfs_root *root, u64 objectid,
1762 struct btrfs_inode_item *btrfs_inode,
1763 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1764 struct ext2_inode *ext2_inode)
1768 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1769 if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
1770 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
1771 ret = ext2_create_file_extents(trans, root, objectid,
1772 btrfs_inode, ext2_fs, ext2_ino, 1, 1);
1773 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
1777 pathname = (char *)&(ext2_inode->i_block[0]);
1778 BUG_ON(pathname[inode_size] != 0);
1779 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
1780 pathname, inode_size + 1);
1781 btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
1786 * Following xattr/acl related codes are based on codes in
1787 * fs/ext3/xattr.c and fs/ext3/acl.c
1789 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
1790 #define EXT2_XATTR_BFIRST(ptr) \
1791 ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
1792 #define EXT2_XATTR_IHDR(inode) \
1793 ((struct ext2_ext_attr_header *) ((void *)(inode) + \
1794 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
1795 #define EXT2_XATTR_IFIRST(inode) \
1796 ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
1797 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
1799 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
1802 struct ext2_ext_attr_entry *next;
1804 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1805 next = EXT2_EXT_ATTR_NEXT(entry);
1806 if ((void *)next >= end)
1813 static int ext2_xattr_check_block(const char *buf, size_t size)
1816 struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
1818 if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
1819 header->h_blocks != 1)
1821 error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
1825 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
1828 size_t value_size = entry->e_value_size;
1830 if (entry->e_value_block != 0 || value_size > size ||
1831 entry->e_value_offs + value_size > size)
1836 #define EXT2_ACL_VERSION 0x0001
1838 /* 23.2.5 acl_tag_t values */
1840 #define ACL_UNDEFINED_TAG (0x00)
1841 #define ACL_USER_OBJ (0x01)
1842 #define ACL_USER (0x02)
1843 #define ACL_GROUP_OBJ (0x04)
1844 #define ACL_GROUP (0x08)
1845 #define ACL_MASK (0x10)
1846 #define ACL_OTHER (0x20)
1848 /* 23.2.7 ACL qualifier constants */
1850 #define ACL_UNDEFINED_ID ((id_t)-1)
1861 } ext2_acl_entry_short;
1867 static inline int ext2_acl_count(size_t size)
1870 size -= sizeof(ext2_acl_header);
1871 s = size - 4 * sizeof(ext2_acl_entry_short);
1873 if (size % sizeof(ext2_acl_entry_short))
1875 return size / sizeof(ext2_acl_entry_short);
1877 if (s % sizeof(ext2_acl_entry))
1879 return s / sizeof(ext2_acl_entry) + 4;
1883 #define ACL_EA_VERSION 0x0002
1893 acl_ea_entry a_entries[0];
1896 static inline size_t acl_ea_size(int count)
1898 return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
1901 static int ext2_acl_to_xattr(void *dst, const void *src,
1902 size_t dst_size, size_t src_size)
1905 const void *end = src + src_size;
1906 acl_ea_header *ext_acl = (acl_ea_header *)dst;
1907 acl_ea_entry *dst_entry = ext_acl->a_entries;
1908 ext2_acl_entry *src_entry;
1910 if (src_size < sizeof(ext2_acl_header))
1912 if (((ext2_acl_header *)src)->a_version !=
1913 cpu_to_le32(EXT2_ACL_VERSION))
1915 src += sizeof(ext2_acl_header);
1916 count = ext2_acl_count(src_size);
1920 BUG_ON(dst_size < acl_ea_size(count));
1921 ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
1922 for (i = 0; i < count; i++, dst_entry++) {
1923 src_entry = (ext2_acl_entry *)src;
1924 if (src + sizeof(ext2_acl_entry_short) > end)
1926 dst_entry->e_tag = src_entry->e_tag;
1927 dst_entry->e_perm = src_entry->e_perm;
1928 switch (le16_to_cpu(src_entry->e_tag)) {
1933 src += sizeof(ext2_acl_entry_short);
1934 dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
1938 src += sizeof(ext2_acl_entry);
1941 dst_entry->e_id = src_entry->e_id;
1954 static char *xattr_prefix_table[] = {
1956 [2] = "system.posix_acl_access",
1957 [3] = "system.posix_acl_default",
1962 static int ext2_copy_single_xattr(struct btrfs_trans_handle *trans,
1963 struct btrfs_root *root, u64 objectid,
1964 struct ext2_ext_attr_entry *entry,
1965 const void *data, u32 datalen)
1970 void *databuf = NULL;
1971 char namebuf[XATTR_NAME_MAX + 1];
1973 name_index = entry->e_name_index;
1974 if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
1975 xattr_prefix_table[name_index] == NULL)
1977 name_len = strlen(xattr_prefix_table[name_index]) +
1979 if (name_len >= sizeof(namebuf))
1982 if (name_index == 2 || name_index == 3) {
1983 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
1984 databuf = malloc(bufsize);
1987 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
1993 strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
1994 strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
1995 if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
1996 sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
1997 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
1998 objectid - INO_OFFSET, name_len, namebuf);
2001 ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
2002 data, datalen, objectid);
2008 static int ext2_copy_extended_attrs(struct btrfs_trans_handle *trans,
2009 struct btrfs_root *root, u64 objectid,
2010 struct btrfs_inode_item *btrfs_inode,
2011 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
2017 u32 block_size = ext2_fs->blocksize;
2018 u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
2019 struct ext2_inode_large *ext2_inode;
2020 struct ext2_ext_attr_entry *entry;
2022 char *buffer = NULL;
2023 char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
2025 if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
2026 ext2_inode = (struct ext2_inode_large *)inode_buf;
2028 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
2032 err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
2035 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
2036 error_message(err));
2041 if (ext2_ino > ext2_fs->super->s_first_ino &&
2042 inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
2043 if (EXT2_GOOD_OLD_INODE_SIZE +
2044 ext2_inode->i_extra_isize > inode_size) {
2048 if (ext2_inode->i_extra_isize != 0 &&
2049 EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
2050 EXT2_EXT_ATTR_MAGIC) {
2056 void *end = (void *)ext2_inode + inode_size;
2057 entry = EXT2_XATTR_IFIRST(ext2_inode);
2058 total = end - (void *)entry;
2059 ret = ext2_xattr_check_names(entry, end);
2062 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
2063 ret = ext2_xattr_check_entry(entry, total);
2066 data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
2067 entry->e_value_offs;
2068 datalen = entry->e_value_size;
2069 ret = ext2_copy_single_xattr(trans, root, objectid,
2070 entry, data, datalen);
2073 entry = EXT2_EXT_ATTR_NEXT(entry);
2077 if (ext2_inode->i_file_acl == 0)
2080 buffer = malloc(block_size);
2085 err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
2087 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
2088 error_message(err));
2092 ret = ext2_xattr_check_block(buffer, block_size);
2096 entry = EXT2_XATTR_BFIRST(buffer);
2097 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
2098 ret = ext2_xattr_check_entry(entry, block_size);
2101 data = buffer + entry->e_value_offs;
2102 datalen = entry->e_value_size;
2103 ret = ext2_copy_single_xattr(trans, root, objectid,
2104 entry, data, datalen);
2107 entry = EXT2_EXT_ATTR_NEXT(entry);
2111 if ((void *)ext2_inode != inode_buf)
2115 #define MINORBITS 20
2116 #define MKDEV(ma, mi) (((ma) << MINORBITS) | (mi))
2118 static inline dev_t old_decode_dev(u16 val)
2120 return MKDEV((val >> 8) & 255, val & 255);
2123 static inline dev_t new_decode_dev(u32 dev)
2125 unsigned major = (dev & 0xfff00) >> 8;
2126 unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
2127 return MKDEV(major, minor);
2130 static void ext2_copy_inode_item(struct btrfs_inode_item *dst,
2131 struct ext2_inode *src, u32 blocksize)
2133 btrfs_set_stack_inode_generation(dst, 1);
2134 btrfs_set_stack_inode_sequence(dst, 0);
2135 btrfs_set_stack_inode_transid(dst, 1);
2136 btrfs_set_stack_inode_size(dst, src->i_size);
2137 btrfs_set_stack_inode_nbytes(dst, 0);
2138 btrfs_set_stack_inode_block_group(dst, 0);
2139 btrfs_set_stack_inode_nlink(dst, src->i_links_count);
2140 btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
2141 btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
2142 btrfs_set_stack_inode_mode(dst, src->i_mode);
2143 btrfs_set_stack_inode_rdev(dst, 0);
2144 btrfs_set_stack_inode_flags(dst, 0);
2145 btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
2146 btrfs_set_stack_timespec_nsec(&dst->atime, 0);
2147 btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
2148 btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
2149 btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
2150 btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
2151 btrfs_set_stack_timespec_sec(&dst->otime, 0);
2152 btrfs_set_stack_timespec_nsec(&dst->otime, 0);
2154 if (S_ISDIR(src->i_mode)) {
2155 btrfs_set_stack_inode_size(dst, 0);
2156 btrfs_set_stack_inode_nlink(dst, 1);
2158 if (S_ISREG(src->i_mode)) {
2159 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
2162 if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
2163 !S_ISLNK(src->i_mode)) {
2164 if (src->i_block[0]) {
2165 btrfs_set_stack_inode_rdev(dst,
2166 old_decode_dev(src->i_block[0]));
2168 btrfs_set_stack_inode_rdev(dst,
2169 new_decode_dev(src->i_block[1]));
2172 memset(&dst->reserved, 0, sizeof(dst->reserved));
2176 * copy a single inode. do all the required works, such as cloning
2177 * inode item, creating file extents and creating directory entries.
2179 static int ext2_copy_single_inode(struct btrfs_trans_handle *trans,
2180 struct btrfs_root *root, u64 objectid,
2181 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
2182 struct ext2_inode *ext2_inode,
2183 int datacsum, int packing, int noxattr)
2186 struct btrfs_inode_item btrfs_inode;
2188 if (ext2_inode->i_links_count == 0)
2191 ext2_copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
2192 if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
2193 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
2194 BTRFS_INODE_NODATASUM;
2195 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
2198 switch (ext2_inode->i_mode & S_IFMT) {
2200 ret = ext2_create_file_extents(trans, root, objectid,
2201 &btrfs_inode, ext2_fs, ext2_ino, datacsum, packing);
2204 ret = ext2_create_dir_entries(trans, root, objectid,
2205 &btrfs_inode, ext2_fs, ext2_ino);
2208 ret = ext2_create_symbol_link(trans, root, objectid,
2209 &btrfs_inode, ext2_fs, ext2_ino, ext2_inode);
2219 ret = ext2_copy_extended_attrs(trans, root, objectid,
2220 &btrfs_inode, ext2_fs, ext2_ino);
2224 return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
2228 * scan ext2's inode bitmap and copy all used inodes.
2230 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
2231 struct btrfs_root *root,
2232 int datacsum, int packing, int noxattr, struct task_ctx *p)
2234 ext2_filsys ext2_fs = cctx->fs_data;
2237 ext2_inode_scan ext2_scan;
2238 struct ext2_inode ext2_inode;
2239 ext2_ino_t ext2_ino;
2241 struct btrfs_trans_handle *trans;
2243 trans = btrfs_start_transaction(root, 1);
2246 err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
2248 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
2251 while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
2253 /* no more inodes */
2256 /* skip special inode in ext2fs */
2257 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
2258 ext2_ino != EXT2_ROOT_INO)
2260 objectid = ext2_ino + INO_OFFSET;
2261 ret = ext2_copy_single_inode(trans, root,
2262 objectid, ext2_fs, ext2_ino,
2263 &ext2_inode, datacsum, packing,
2265 p->cur_copy_inodes++;
2268 if (trans->blocks_used >= 4096) {
2269 ret = btrfs_commit_transaction(trans, root);
2271 trans = btrfs_start_transaction(root, 1);
2276 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
2279 ret = btrfs_commit_transaction(trans, root);
2281 ext2fs_close_inode_scan(ext2_scan);
2286 static const struct btrfs_convert_operations ext2_convert_ops = {
2288 .open_fs = ext2_open_fs,
2289 .read_used_space = ext2_read_used_space,
2290 .copy_inodes = ext2_copy_inodes,
2291 .close_fs = ext2_close_fs,
2296 static const struct btrfs_convert_operations *convert_operations[] = {
2297 #if BTRFSCONVERT_EXT2
2302 static int convert_open_fs(const char *devname,
2303 struct btrfs_convert_context *cctx)
2307 memset(cctx, 0, sizeof(*cctx));
2309 for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
2310 int ret = convert_operations[i]->open_fs(cctx, devname);
2313 cctx->convert_ops = convert_operations[i];
2318 fprintf(stderr, "No file system found to convert.\n");
2322 static int do_convert(const char *devname, int datacsum, int packing,
2323 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2324 int progress, u64 features)
2330 struct btrfs_root *root;
2331 struct btrfs_root *image_root;
2332 struct btrfs_convert_context cctx;
2333 struct btrfs_key key;
2334 char *subvol_name = NULL;
2335 struct task_ctx ctx;
2336 char features_buf[64];
2337 struct btrfs_mkfs_config mkfs_cfg;
2339 init_convert_context(&cctx);
2340 ret = convert_open_fs(devname, &cctx);
2343 ret = convert_read_used_space(&cctx);
2347 blocksize = cctx.blocksize;
2348 total_bytes = (u64)blocksize * (u64)cctx.block_count;
2349 if (blocksize < 4096) {
2350 error("block size is too small: %u < 4096", blocksize);
2353 if (btrfs_check_nodesize(nodesize, blocksize, features))
2355 fd = open(devname, O_RDWR);
2357 error("unable to open %s: %s", devname, strerror(errno));
2360 btrfs_parse_features_to_string(features_buf, features);
2361 if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2362 strcat(features_buf, " (default)");
2364 printf("create btrfs filesystem:\n");
2365 printf("\tblocksize: %u\n", blocksize);
2366 printf("\tnodesize: %u\n", nodesize);
2367 printf("\tfeatures: %s\n", features_buf);
2369 mkfs_cfg.label = cctx.volume_name;
2370 mkfs_cfg.num_bytes = total_bytes;
2371 mkfs_cfg.nodesize = nodesize;
2372 mkfs_cfg.sectorsize = blocksize;
2373 mkfs_cfg.stripesize = blocksize;
2374 mkfs_cfg.features = features;
2375 /* New convert need these space */
2376 memset(mkfs_cfg.chunk_uuid, 0, BTRFS_UUID_UNPARSED_SIZE);
2377 memset(mkfs_cfg.fs_uuid, 0, BTRFS_UUID_UNPARSED_SIZE);
2379 ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2381 error("unable to create initial ctree: %s", strerror(-ret));
2385 root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2386 OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2388 error("unable to open ctree");
2391 ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2393 error("unable to setup the root tree: %d", ret);
2397 printf("creating %s image file\n", cctx.convert_ops->name);
2398 ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2400 error("memory allocation failure for subvolume name: %s_saved",
2401 cctx.convert_ops->name);
2404 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2405 key.offset = (u64)-1;
2406 key.type = BTRFS_ROOT_ITEM_KEY;
2407 image_root = btrfs_read_fs_root(root->fs_info, &key);
2409 error("unable to create image subvolume");
2412 ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2413 mkfs_cfg.num_bytes, "image", datacsum);
2415 error("failed to create %s/image: %d", subvol_name, ret);
2419 printf("creating btrfs metadata");
2420 ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2421 ctx.cur_copy_inodes = 0;
2424 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2426 task_start(ctx.info);
2428 ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2430 error("error during copy_inodes %d", ret);
2434 task_stop(ctx.info);
2435 task_deinit(ctx.info);
2438 image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2440 error("unable to link subvolume %s", subvol_name);
2446 memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2447 if (copylabel == 1) {
2448 __strncpy_null(root->fs_info->super_copy->label,
2449 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2450 printf("copy label '%s'\n", root->fs_info->super_copy->label);
2451 } else if (copylabel == -1) {
2452 strcpy(root->fs_info->super_copy->label, fslabel);
2453 printf("set label to '%s'\n", fslabel);
2456 ret = close_ctree(root);
2458 error("close_ctree failed: %d", ret);
2461 convert_close_fs(&cctx);
2462 clean_convert_context(&cctx);
2465 * If this step succeed, we get a mountable btrfs. Otherwise
2466 * the source fs is left unchanged.
2468 ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2470 error("unable to migrate super block: %d", ret);
2474 root = open_ctree_fd(fd, devname, 0,
2475 OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2477 error("unable to open ctree for finalization");
2480 root->fs_info->finalize_on_close = 1;
2484 printf("conversion complete");
2487 clean_convert_context(&cctx);
2491 "an error occurred during conversion, filesystem is partially created but not finalized and not mountable");
2496 * Check if a non 1:1 mapped chunk can be rolled back.
2497 * For new convert, it's OK while for old convert it's not.
2499 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2501 struct btrfs_block_group_cache *bg;
2502 struct btrfs_key key;
2503 struct btrfs_path path;
2504 struct btrfs_root *extent_root = fs_info->extent_root;
2509 bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2512 bg_start = bg->key.objectid;
2513 bg_end = bg->key.objectid + bg->key.offset;
2515 key.objectid = bg_end;
2516 key.type = BTRFS_METADATA_ITEM_KEY;
2518 btrfs_init_path(&path);
2520 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2525 struct btrfs_extent_item *ei;
2527 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2535 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2536 if (key.type == BTRFS_METADATA_ITEM_KEY)
2538 /* Now it's EXTENT_ITEM_KEY only */
2539 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2540 struct btrfs_extent_item);
2542 * Found data extent, means this is old convert must follow 1:1
2545 if (btrfs_extent_flags(path.nodes[0], ei)
2546 & BTRFS_EXTENT_FLAG_DATA) {
2551 btrfs_release_path(&path);
2555 static int may_rollback(struct btrfs_root *root)
2557 struct btrfs_fs_info *info = root->fs_info;
2558 struct btrfs_multi_bio *multi = NULL;
2566 if (btrfs_super_num_devices(info->super_copy) != 1)
2569 bytenr = BTRFS_SUPER_INFO_OFFSET;
2570 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2573 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2574 &length, &multi, 0, NULL);
2576 if (ret == -ENOENT) {
2577 /* removed block group at the tail */
2578 if (length == (u64)-1)
2581 /* removed block group in the middle */
2587 num_stripes = multi->num_stripes;
2588 physical = multi->stripes[0].physical;
2591 if (num_stripes != 1) {
2592 error("num stripes for bytenr %llu is not 1", bytenr);
2597 * Extra check for new convert, as metadata chunk from new
2598 * convert is much more free than old convert, it doesn't need
2599 * to do 1:1 mapping.
2601 if (physical != bytenr) {
2603 * Check if it's a metadata chunk and has only metadata
2606 ret = may_rollback_chunk(info, bytenr);
2612 if (bytenr >= total_bytes)
2620 static int do_rollback(const char *devname)
2625 struct btrfs_root *root;
2626 struct btrfs_root *image_root;
2627 struct btrfs_root *chunk_root;
2628 struct btrfs_dir_item *dir;
2629 struct btrfs_inode_item *inode;
2630 struct btrfs_file_extent_item *fi;
2631 struct btrfs_trans_handle *trans;
2632 struct extent_buffer *leaf;
2633 struct btrfs_block_group_cache *cache1;
2634 struct btrfs_block_group_cache *cache2;
2635 struct btrfs_key key;
2636 struct btrfs_path path;
2637 struct extent_io_tree io_tree;
2652 extent_io_tree_init(&io_tree);
2654 fd = open(devname, O_RDWR);
2656 error("unable to open %s: %s", devname, strerror(errno));
2659 root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2661 error("unable to open ctree");
2664 ret = may_rollback(root);
2666 error("unable to do rollback: %d", ret);
2670 sectorsize = root->sectorsize;
2671 buf = malloc(sectorsize);
2673 error("unable to allocate memory");
2677 btrfs_init_path(&path);
2679 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2680 key.type = BTRFS_ROOT_BACKREF_KEY;
2681 key.offset = BTRFS_FS_TREE_OBJECTID;
2682 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2684 btrfs_release_path(&path);
2686 error("unable to convert ext2 image subvolume, is it deleted?");
2688 } else if (ret < 0) {
2689 error("unable to open ext2_saved, id %llu: %s",
2690 (unsigned long long)key.objectid, strerror(-ret));
2694 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2695 key.type = BTRFS_ROOT_ITEM_KEY;
2696 key.offset = (u64)-1;
2697 image_root = btrfs_read_fs_root(root->fs_info, &key);
2698 if (!image_root || IS_ERR(image_root)) {
2699 error("unable to open subvolume %llu: %ld",
2700 (unsigned long long)key.objectid, PTR_ERR(image_root));
2705 root_dir = btrfs_root_dirid(&root->root_item);
2706 dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2707 root_dir, name, strlen(name), 0);
2708 if (!dir || IS_ERR(dir)) {
2709 error("unable to find file %s: %ld", name, PTR_ERR(dir));
2712 leaf = path.nodes[0];
2713 btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2714 btrfs_release_path(&path);
2716 objectid = key.objectid;
2718 ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2720 error("unable to find inode item: %d", ret);
2723 leaf = path.nodes[0];
2724 inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2725 total_bytes = btrfs_inode_size(leaf, inode);
2726 btrfs_release_path(&path);
2728 key.objectid = objectid;
2730 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2731 ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2733 error("unable to find first file extent: %d", ret);
2734 btrfs_release_path(&path);
2738 /* build mapping tree for the relocated blocks */
2739 for (offset = 0; offset < total_bytes; ) {
2740 leaf = path.nodes[0];
2741 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2742 ret = btrfs_next_leaf(root, &path);
2748 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2749 if (key.objectid != objectid || key.offset != offset ||
2750 btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2753 fi = btrfs_item_ptr(leaf, path.slots[0],
2754 struct btrfs_file_extent_item);
2755 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2757 if (btrfs_file_extent_compression(leaf, fi) ||
2758 btrfs_file_extent_encryption(leaf, fi) ||
2759 btrfs_file_extent_other_encoding(leaf, fi))
2762 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2763 /* skip holes and direct mapped extents */
2764 if (bytenr == 0 || bytenr == offset)
2767 bytenr += btrfs_file_extent_offset(leaf, fi);
2768 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2770 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2771 cache2 = btrfs_lookup_block_group(root->fs_info,
2772 offset + num_bytes - 1);
2774 * Here we must take consideration of old and new convert
2776 * For old convert case, sign, there is no consist chunk type
2777 * that will cover the extent. META/DATA/SYS are all possible.
2778 * Just ensure relocate one is in SYS chunk.
2779 * For new convert case, they are all covered by DATA chunk.
2781 * So, there is not valid chunk type check for it now.
2783 if (cache1 != cache2)
2786 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2787 EXTENT_LOCKED, GFP_NOFS);
2788 set_state_private(&io_tree, offset, bytenr);
2790 offset += btrfs_file_extent_num_bytes(leaf, fi);
2793 btrfs_release_path(&path);
2795 if (offset < total_bytes) {
2796 error("unable to build extent mapping (offset %llu, total_bytes %llu)",
2797 (unsigned long long)offset,
2798 (unsigned long long)total_bytes);
2799 error("converted filesystem after balance is unable to rollback");
2803 first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2804 first_free &= ~((u64)sectorsize - 1);
2805 /* backup for extent #0 should exist */
2806 if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2807 error("no backup for the first extent");
2810 /* force no allocation from system block group */
2811 root->fs_info->system_allocs = -1;
2812 trans = btrfs_start_transaction(root, 1);
2815 * recow the whole chunk tree, this will remove all chunk tree blocks
2816 * from system block group
2818 chunk_root = root->fs_info->chunk_root;
2819 memset(&key, 0, sizeof(key));
2821 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2825 ret = btrfs_next_leaf(chunk_root, &path);
2829 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2830 btrfs_release_path(&path);
2832 btrfs_release_path(&path);
2837 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2841 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2842 num_bytes += btrfs_block_group_used(&cache1->item);
2844 offset = cache1->key.objectid + cache1->key.offset;
2846 /* only extent #0 left in system block group? */
2847 if (num_bytes > first_free) {
2849 "unable to empty system block group (num_bytes %llu, first_free %llu",
2850 (unsigned long long)num_bytes,
2851 (unsigned long long)first_free);
2854 /* create a system chunk that maps the whole device */
2855 ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2857 error("unable to update system chunk: %d", ret);
2861 ret = btrfs_commit_transaction(trans, root);
2864 ret = close_ctree(root);
2866 error("close_ctree failed: %d", ret);
2870 /* zero btrfs super block mirrors */
2871 memset(buf, 0, sectorsize);
2872 for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2873 bytenr = btrfs_sb_offset(i);
2874 if (bytenr >= total_bytes)
2876 ret = pwrite(fd, buf, sectorsize, bytenr);
2877 if (ret != sectorsize) {
2878 error("zeroing superblock mirror %d failed: %d",
2884 sb_bytenr = (u64)-1;
2885 /* copy all relocated blocks back */
2887 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2892 ret = get_state_private(&io_tree, start, &bytenr);
2895 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2898 while (start <= end) {
2899 if (start == BTRFS_SUPER_INFO_OFFSET) {
2903 ret = pread(fd, buf, sectorsize, bytenr);
2905 error("reading superblock at %llu failed: %d",
2906 (unsigned long long)bytenr, ret);
2909 BUG_ON(ret != sectorsize);
2910 ret = pwrite(fd, buf, sectorsize, start);
2912 error("writing superblock at %llu failed: %d",
2913 (unsigned long long)start, ret);
2916 BUG_ON(ret != sectorsize);
2918 start += sectorsize;
2919 bytenr += sectorsize;
2925 error("fsync failed: %s", strerror(errno));
2929 * finally, overwrite btrfs super block.
2931 ret = pread(fd, buf, sectorsize, sb_bytenr);
2933 error("reading primary superblock failed: %s",
2937 BUG_ON(ret != sectorsize);
2938 ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2940 error("writing primary superblock failed: %s",
2944 BUG_ON(ret != sectorsize);
2947 error("fsync failed: %s", strerror(errno));
2953 extent_io_tree_cleanup(&io_tree);
2954 printf("rollback complete\n");
2961 error("rollback aborted");
2965 static void print_usage(void)
2967 printf("usage: btrfs-convert [options] device\n");
2968 printf("options:\n");
2969 printf("\t-d|--no-datasum disable data checksum, sets NODATASUM\n");
2970 printf("\t-i|--no-xattr ignore xattrs and ACLs\n");
2971 printf("\t-n|--no-inline disable inlining of small files to metadata\n");
2972 printf("\t-N|--nodesize SIZE set filesystem metadata nodesize\n");
2973 printf("\t-r|--rollback roll back to the original filesystem\n");
2974 printf("\t-l|--label LABEL set filesystem label\n");
2975 printf("\t-L|--copy-label use label from converted filesystem\n");
2976 printf("\t-p|--progress show converting progress (default)\n");
2977 printf("\t-O|--features LIST comma separated list of filesystem features\n");
2978 printf("\t--no-progress show only overview, not the detailed progress\n");
2980 printf("Suported filesystems:\n");
2981 printf("\text2/3/4: %s\n", BTRFSCONVERT_EXT2 ? "yes" : "no");
2984 int main(int argc, char *argv[])
2990 u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
2991 BTRFS_MKFS_DEFAULT_NODE_SIZE);
2994 int usage_error = 0;
2997 char fslabel[BTRFS_LABEL_SIZE];
2998 u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
3001 enum { GETOPT_VAL_NO_PROGRESS = 256 };
3002 static const struct option long_options[] = {
3003 { "no-progress", no_argument, NULL,
3004 GETOPT_VAL_NO_PROGRESS },
3005 { "no-datasum", no_argument, NULL, 'd' },
3006 { "no-inline", no_argument, NULL, 'n' },
3007 { "no-xattr", no_argument, NULL, 'i' },
3008 { "rollback", no_argument, NULL, 'r' },
3009 { "features", required_argument, NULL, 'O' },
3010 { "progress", no_argument, NULL, 'p' },
3011 { "label", required_argument, NULL, 'l' },
3012 { "copy-label", no_argument, NULL, 'L' },
3013 { "nodesize", required_argument, NULL, 'N' },
3014 { "help", no_argument, NULL, GETOPT_VAL_HELP},
3015 { NULL, 0, NULL, 0 }
3017 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
3032 nodesize = parse_size(optarg);
3039 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
3041 "WARNING: label too long, trimmed to %d bytes\n",
3042 BTRFS_LABEL_SIZE - 1);
3044 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
3053 char *orig = strdup(optarg);
3056 tmp = btrfs_parse_fs_features(tmp, &features);
3059 "Unrecognized filesystem feature '%s'\n",
3065 if (features & BTRFS_FEATURE_LIST_ALL) {
3066 btrfs_list_all_fs_features(
3067 ~BTRFS_CONVERT_ALLOWED_FEATURES);
3070 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
3073 btrfs_parse_features_to_string(buf,
3074 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
3076 "ERROR: features not allowed for convert: %s\n",
3083 case GETOPT_VAL_NO_PROGRESS:
3086 case GETOPT_VAL_HELP:
3089 return c != GETOPT_VAL_HELP;
3093 if (check_argc_exact(argc - optind, 1)) {
3098 if (rollback && (!datacsum || noxattr || !packing)) {
3100 "Usage error: -d, -i, -n options do not apply to rollback\n");
3109 file = argv[optind];
3110 ret = check_mounted(file);
3112 fprintf(stderr, "Could not check mount status: %s\n",
3116 fprintf(stderr, "%s is mounted\n", file);
3121 ret = do_rollback(file);
3123 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3124 copylabel, fslabel, progress, features);