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);
841 static int calculate_available_space(struct btrfs_convert_context *cctx)
843 struct cache_tree *used = &cctx->used;
844 struct cache_tree *data_chunks = &cctx->data_chunks;
845 struct cache_tree *free = &cctx->free;
846 struct cache_extent *cache;
849 * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
850 * works without need to consider overlap
852 u64 min_stripe_size = 2 * 16 * 1024 * 1024;
855 /* Calculate data_chunks */
856 for (cache = first_cache_extent(used); cache;
857 cache = next_cache_extent(cache)) {
860 if (cache->start + cache->size < cur_off)
862 if (cache->start > cur_off + min_stripe_size)
863 cur_off = cache->start;
864 cur_len = max(cache->start + cache->size - cur_off,
866 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
872 * remove reserved ranges, so we won't ever bother relocating an old
873 * filesystem extent to other place.
875 ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
881 * Calculate free space
882 * Always round up the start bytenr, to avoid metadata extent corss
883 * stripe boundary, as later mkfs_convert() won't have all the extent
886 for (cache = first_cache_extent(data_chunks); cache;
887 cache = next_cache_extent(cache)) {
888 if (cache->start < cur_off)
890 if (cache->start > cur_off) {
894 len = cache->start - round_up(cur_off,
896 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
898 ret = add_merge_cache_extent(free, insert_start, len);
902 cur_off = cache->start + cache->size;
904 /* Don't forget the last range */
905 if (cctx->total_bytes > cur_off) {
906 u64 len = cctx->total_bytes - cur_off;
909 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
911 ret = add_merge_cache_extent(free, insert_start, len);
916 /* Remove reserved bytes */
917 ret = wipe_reserved_ranges(free, min_stripe_size, 0);
923 * Open Ext2fs in readonly mode, read block allocation bitmap and
924 * inode bitmap into memory.
926 static int ext2_open_fs(struct btrfs_convert_context *cctx, const char *name)
933 ret = ext2fs_open(name, 0, 0, 0, unix_io_manager, &ext2_fs);
935 fprintf(stderr, "ext2fs_open: %s\n", error_message(ret));
939 * We need to know exactly the used space, some RO compat flags like
940 * BIGALLOC will affect how used space is present.
941 * So we need manuall check any unsupported RO compat flags
943 ro_feature = ext2_fs->super->s_feature_ro_compat;
944 if (ro_feature & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
946 "unsupported RO features detected: %x, abort convert to avoid possible corruption",
947 ro_feature & ~EXT2_LIB_FEATURE_COMPAT_SUPP);
950 ret = ext2fs_read_inode_bitmap(ext2_fs);
952 fprintf(stderr, "ext2fs_read_inode_bitmap: %s\n",
956 ret = ext2fs_read_block_bitmap(ext2_fs);
958 fprintf(stderr, "ext2fs_read_block_bitmap: %s\n",
963 * search each block group for a free inode. this set up
964 * uninit block/inode bitmaps appropriately.
967 while (ino <= ext2_fs->super->s_inodes_count) {
969 ext2fs_new_inode(ext2_fs, ino, 0, NULL, &foo);
970 ino += EXT2_INODES_PER_GROUP(ext2_fs->super);
973 if (!(ext2_fs->super->s_feature_incompat &
974 EXT2_FEATURE_INCOMPAT_FILETYPE)) {
975 fprintf(stderr, "filetype feature is missing\n");
979 cctx->fs_data = ext2_fs;
980 cctx->blocksize = ext2_fs->blocksize;
981 cctx->block_count = ext2_fs->super->s_blocks_count;
982 cctx->total_bytes = ext2_fs->blocksize * ext2_fs->super->s_blocks_count;
983 cctx->volume_name = strndup(ext2_fs->super->s_volume_name, 16);
984 cctx->first_data_block = ext2_fs->super->s_first_data_block;
985 cctx->inodes_count = ext2_fs->super->s_inodes_count;
986 cctx->free_inodes_count = ext2_fs->super->s_free_inodes_count;
989 ext2fs_close(ext2_fs);
993 static int __ext2_add_one_block(ext2_filsys fs, char *bitmap,
994 unsigned long group_nr, struct cache_tree *used)
996 unsigned long offset;
1000 offset = fs->super->s_first_data_block;
1001 offset /= EXT2FS_CLUSTER_RATIO(fs);
1002 offset += group_nr * EXT2_CLUSTERS_PER_GROUP(fs->super);
1003 for (i = 0; i < EXT2_CLUSTERS_PER_GROUP(fs->super); i++) {
1004 if (ext2fs_test_bit(i, bitmap)) {
1007 start = (i + offset) * EXT2FS_CLUSTER_RATIO(fs);
1008 start *= fs->blocksize;
1009 ret = add_merge_cache_extent(used, start,
1019 * Read all used ext2 space into cctx->used cache tree
1021 static int ext2_read_used_space(struct btrfs_convert_context *cctx)
1023 ext2_filsys fs = (ext2_filsys)cctx->fs_data;
1024 blk64_t blk_itr = EXT2FS_B2C(fs, fs->super->s_first_data_block);
1025 struct cache_tree *used_tree = &cctx->used;
1026 char *block_bitmap = NULL;
1031 block_nbytes = EXT2_CLUSTERS_PER_GROUP(fs->super) / 8;
1032 /* Shouldn't happen */
1033 BUG_ON(!fs->block_map);
1035 block_bitmap = malloc(block_nbytes);
1039 for (i = 0; i < fs->group_desc_count; i++) {
1040 ret = ext2fs_get_block_bitmap_range(fs->block_map, blk_itr,
1041 block_nbytes * 8, block_bitmap);
1043 error("fail to get bitmap from ext2, %s",
1047 ret = __ext2_add_one_block(fs, block_bitmap, i, used_tree);
1049 error("fail to build used space tree, %s",
1053 blk_itr += EXT2_CLUSTERS_PER_GROUP(fs->super);
1060 static void ext2_close_fs(struct btrfs_convert_context *cctx)
1062 if (cctx->volume_name) {
1063 free(cctx->volume_name);
1064 cctx->volume_name = NULL;
1066 ext2fs_close(cctx->fs_data);
1069 struct dir_iterate_data {
1070 struct btrfs_trans_handle *trans;
1071 struct btrfs_root *root;
1072 struct btrfs_inode_item *inode;
1079 static u8 ext2_filetype_conversion_table[EXT2_FT_MAX] = {
1080 [EXT2_FT_UNKNOWN] = BTRFS_FT_UNKNOWN,
1081 [EXT2_FT_REG_FILE] = BTRFS_FT_REG_FILE,
1082 [EXT2_FT_DIR] = BTRFS_FT_DIR,
1083 [EXT2_FT_CHRDEV] = BTRFS_FT_CHRDEV,
1084 [EXT2_FT_BLKDEV] = BTRFS_FT_BLKDEV,
1085 [EXT2_FT_FIFO] = BTRFS_FT_FIFO,
1086 [EXT2_FT_SOCK] = BTRFS_FT_SOCK,
1087 [EXT2_FT_SYMLINK] = BTRFS_FT_SYMLINK,
1090 static int ext2_dir_iterate_proc(ext2_ino_t dir, int entry,
1091 struct ext2_dir_entry *dirent,
1092 int offset, int blocksize,
1093 char *buf,void *priv_data)
1098 char dotdot[] = "..";
1099 struct dir_iterate_data *idata = (struct dir_iterate_data *)priv_data;
1102 name_len = dirent->name_len & 0xFF;
1104 objectid = dirent->inode + INO_OFFSET;
1105 if (!strncmp(dirent->name, dotdot, name_len)) {
1106 if (name_len == 2) {
1107 BUG_ON(idata->parent != 0);
1108 idata->parent = objectid;
1112 if (dirent->inode < EXT2_GOOD_OLD_FIRST_INO)
1115 file_type = dirent->name_len >> 8;
1116 BUG_ON(file_type > EXT2_FT_SYMLINK);
1118 ret = convert_insert_dirent(idata->trans, idata->root, dirent->name,
1119 name_len, idata->objectid, objectid,
1120 ext2_filetype_conversion_table[file_type],
1121 idata->index_cnt, idata->inode);
1123 idata->errcode = ret;
1131 static int ext2_create_dir_entries(struct btrfs_trans_handle *trans,
1132 struct btrfs_root *root, u64 objectid,
1133 struct btrfs_inode_item *btrfs_inode,
1134 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
1138 struct dir_iterate_data data = {
1141 .inode = btrfs_inode,
1142 .objectid = objectid,
1148 err = ext2fs_dir_iterate2(ext2_fs, ext2_ino, 0, NULL,
1149 ext2_dir_iterate_proc, &data);
1153 if (ret == 0 && data.parent == objectid) {
1154 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
1155 objectid, objectid, 0);
1159 fprintf(stderr, "ext2fs_dir_iterate2: %s\n", error_message(err));
1163 static int ext2_block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
1164 e2_blkcnt_t blockcnt, blk_t ref_block,
1165 int ref_offset, void *priv_data)
1168 struct blk_iterate_data *idata;
1169 idata = (struct blk_iterate_data *)priv_data;
1170 ret = block_iterate_proc(*blocknr, blockcnt, idata);
1172 idata->errcode = ret;
1179 * traverse file's data blocks, record these data blocks as file extents.
1181 static int ext2_create_file_extents(struct btrfs_trans_handle *trans,
1182 struct btrfs_root *root, u64 objectid,
1183 struct btrfs_inode_item *btrfs_inode,
1184 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1185 int datacsum, int packing)
1188 char *buffer = NULL;
1191 u32 sectorsize = root->sectorsize;
1192 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1193 struct blk_iterate_data data;
1195 init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
1198 err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
1199 NULL, ext2_block_iterate_proc, &data);
1205 if (packing && data.first_block == 0 && data.num_blocks > 0 &&
1206 inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
1207 u64 num_bytes = data.num_blocks * sectorsize;
1208 u64 disk_bytenr = data.disk_block * sectorsize;
1211 buffer = malloc(num_bytes);
1214 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
1217 if (num_bytes > inode_size)
1218 num_bytes = inode_size;
1219 ret = btrfs_insert_inline_extent(trans, root, objectid,
1220 0, buffer, num_bytes);
1223 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
1224 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
1225 } else if (data.num_blocks > 0) {
1226 ret = record_file_blocks(&data, data.first_block,
1227 data.disk_block, data.num_blocks);
1231 data.first_block += data.num_blocks;
1232 last_block = (inode_size + sectorsize - 1) / sectorsize;
1233 if (last_block > data.first_block) {
1234 ret = record_file_blocks(&data, data.first_block, 0,
1235 last_block - data.first_block);
1241 fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
1245 static int ext2_create_symbol_link(struct btrfs_trans_handle *trans,
1246 struct btrfs_root *root, u64 objectid,
1247 struct btrfs_inode_item *btrfs_inode,
1248 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1249 struct ext2_inode *ext2_inode)
1253 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
1254 if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
1255 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
1256 ret = ext2_create_file_extents(trans, root, objectid,
1257 btrfs_inode, ext2_fs, ext2_ino, 1, 1);
1258 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
1262 pathname = (char *)&(ext2_inode->i_block[0]);
1263 BUG_ON(pathname[inode_size] != 0);
1264 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
1265 pathname, inode_size + 1);
1266 btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
1271 * Following xattr/acl related codes are based on codes in
1272 * fs/ext3/xattr.c and fs/ext3/acl.c
1274 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
1275 #define EXT2_XATTR_BFIRST(ptr) \
1276 ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
1277 #define EXT2_XATTR_IHDR(inode) \
1278 ((struct ext2_ext_attr_header *) ((void *)(inode) + \
1279 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
1280 #define EXT2_XATTR_IFIRST(inode) \
1281 ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
1282 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
1284 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
1287 struct ext2_ext_attr_entry *next;
1289 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1290 next = EXT2_EXT_ATTR_NEXT(entry);
1291 if ((void *)next >= end)
1298 static int ext2_xattr_check_block(const char *buf, size_t size)
1301 struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
1303 if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
1304 header->h_blocks != 1)
1306 error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
1310 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
1313 size_t value_size = entry->e_value_size;
1315 if (entry->e_value_block != 0 || value_size > size ||
1316 entry->e_value_offs + value_size > size)
1321 #define EXT2_ACL_VERSION 0x0001
1323 /* 23.2.5 acl_tag_t values */
1325 #define ACL_UNDEFINED_TAG (0x00)
1326 #define ACL_USER_OBJ (0x01)
1327 #define ACL_USER (0x02)
1328 #define ACL_GROUP_OBJ (0x04)
1329 #define ACL_GROUP (0x08)
1330 #define ACL_MASK (0x10)
1331 #define ACL_OTHER (0x20)
1333 /* 23.2.7 ACL qualifier constants */
1335 #define ACL_UNDEFINED_ID ((id_t)-1)
1346 } ext2_acl_entry_short;
1352 static inline int ext2_acl_count(size_t size)
1355 size -= sizeof(ext2_acl_header);
1356 s = size - 4 * sizeof(ext2_acl_entry_short);
1358 if (size % sizeof(ext2_acl_entry_short))
1360 return size / sizeof(ext2_acl_entry_short);
1362 if (s % sizeof(ext2_acl_entry))
1364 return s / sizeof(ext2_acl_entry) + 4;
1368 #define ACL_EA_VERSION 0x0002
1378 acl_ea_entry a_entries[0];
1381 static inline size_t acl_ea_size(int count)
1383 return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
1386 static int ext2_acl_to_xattr(void *dst, const void *src,
1387 size_t dst_size, size_t src_size)
1390 const void *end = src + src_size;
1391 acl_ea_header *ext_acl = (acl_ea_header *)dst;
1392 acl_ea_entry *dst_entry = ext_acl->a_entries;
1393 ext2_acl_entry *src_entry;
1395 if (src_size < sizeof(ext2_acl_header))
1397 if (((ext2_acl_header *)src)->a_version !=
1398 cpu_to_le32(EXT2_ACL_VERSION))
1400 src += sizeof(ext2_acl_header);
1401 count = ext2_acl_count(src_size);
1405 BUG_ON(dst_size < acl_ea_size(count));
1406 ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
1407 for (i = 0; i < count; i++, dst_entry++) {
1408 src_entry = (ext2_acl_entry *)src;
1409 if (src + sizeof(ext2_acl_entry_short) > end)
1411 dst_entry->e_tag = src_entry->e_tag;
1412 dst_entry->e_perm = src_entry->e_perm;
1413 switch (le16_to_cpu(src_entry->e_tag)) {
1418 src += sizeof(ext2_acl_entry_short);
1419 dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
1423 src += sizeof(ext2_acl_entry);
1426 dst_entry->e_id = src_entry->e_id;
1439 static char *xattr_prefix_table[] = {
1441 [2] = "system.posix_acl_access",
1442 [3] = "system.posix_acl_default",
1447 static int ext2_copy_single_xattr(struct btrfs_trans_handle *trans,
1448 struct btrfs_root *root, u64 objectid,
1449 struct ext2_ext_attr_entry *entry,
1450 const void *data, u32 datalen)
1455 void *databuf = NULL;
1456 char namebuf[XATTR_NAME_MAX + 1];
1458 name_index = entry->e_name_index;
1459 if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
1460 xattr_prefix_table[name_index] == NULL)
1462 name_len = strlen(xattr_prefix_table[name_index]) +
1464 if (name_len >= sizeof(namebuf))
1467 if (name_index == 2 || name_index == 3) {
1468 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
1469 databuf = malloc(bufsize);
1472 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
1478 strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
1479 strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
1480 if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
1481 sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
1482 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
1483 objectid - INO_OFFSET, name_len, namebuf);
1486 ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
1487 data, datalen, objectid);
1493 static int ext2_copy_extended_attrs(struct btrfs_trans_handle *trans,
1494 struct btrfs_root *root, u64 objectid,
1495 struct btrfs_inode_item *btrfs_inode,
1496 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
1502 u32 block_size = ext2_fs->blocksize;
1503 u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
1504 struct ext2_inode_large *ext2_inode;
1505 struct ext2_ext_attr_entry *entry;
1507 char *buffer = NULL;
1508 char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
1510 if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
1511 ext2_inode = (struct ext2_inode_large *)inode_buf;
1513 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
1517 err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
1520 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
1521 error_message(err));
1526 if (ext2_ino > ext2_fs->super->s_first_ino &&
1527 inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
1528 if (EXT2_GOOD_OLD_INODE_SIZE +
1529 ext2_inode->i_extra_isize > inode_size) {
1533 if (ext2_inode->i_extra_isize != 0 &&
1534 EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
1535 EXT2_EXT_ATTR_MAGIC) {
1541 void *end = (void *)ext2_inode + inode_size;
1542 entry = EXT2_XATTR_IFIRST(ext2_inode);
1543 total = end - (void *)entry;
1544 ret = ext2_xattr_check_names(entry, end);
1547 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1548 ret = ext2_xattr_check_entry(entry, total);
1551 data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
1552 entry->e_value_offs;
1553 datalen = entry->e_value_size;
1554 ret = ext2_copy_single_xattr(trans, root, objectid,
1555 entry, data, datalen);
1558 entry = EXT2_EXT_ATTR_NEXT(entry);
1562 if (ext2_inode->i_file_acl == 0)
1565 buffer = malloc(block_size);
1570 err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
1572 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
1573 error_message(err));
1577 ret = ext2_xattr_check_block(buffer, block_size);
1581 entry = EXT2_XATTR_BFIRST(buffer);
1582 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1583 ret = ext2_xattr_check_entry(entry, block_size);
1586 data = buffer + entry->e_value_offs;
1587 datalen = entry->e_value_size;
1588 ret = ext2_copy_single_xattr(trans, root, objectid,
1589 entry, data, datalen);
1592 entry = EXT2_EXT_ATTR_NEXT(entry);
1596 if ((void *)ext2_inode != inode_buf)
1600 #define MINORBITS 20
1601 #define MKDEV(ma, mi) (((ma) << MINORBITS) | (mi))
1603 static inline dev_t old_decode_dev(u16 val)
1605 return MKDEV((val >> 8) & 255, val & 255);
1608 static inline dev_t new_decode_dev(u32 dev)
1610 unsigned major = (dev & 0xfff00) >> 8;
1611 unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
1612 return MKDEV(major, minor);
1615 static void ext2_copy_inode_item(struct btrfs_inode_item *dst,
1616 struct ext2_inode *src, u32 blocksize)
1618 btrfs_set_stack_inode_generation(dst, 1);
1619 btrfs_set_stack_inode_sequence(dst, 0);
1620 btrfs_set_stack_inode_transid(dst, 1);
1621 btrfs_set_stack_inode_size(dst, src->i_size);
1622 btrfs_set_stack_inode_nbytes(dst, 0);
1623 btrfs_set_stack_inode_block_group(dst, 0);
1624 btrfs_set_stack_inode_nlink(dst, src->i_links_count);
1625 btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
1626 btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
1627 btrfs_set_stack_inode_mode(dst, src->i_mode);
1628 btrfs_set_stack_inode_rdev(dst, 0);
1629 btrfs_set_stack_inode_flags(dst, 0);
1630 btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
1631 btrfs_set_stack_timespec_nsec(&dst->atime, 0);
1632 btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
1633 btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
1634 btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
1635 btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
1636 btrfs_set_stack_timespec_sec(&dst->otime, 0);
1637 btrfs_set_stack_timespec_nsec(&dst->otime, 0);
1639 if (S_ISDIR(src->i_mode)) {
1640 btrfs_set_stack_inode_size(dst, 0);
1641 btrfs_set_stack_inode_nlink(dst, 1);
1643 if (S_ISREG(src->i_mode)) {
1644 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
1647 if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
1648 !S_ISLNK(src->i_mode)) {
1649 if (src->i_block[0]) {
1650 btrfs_set_stack_inode_rdev(dst,
1651 old_decode_dev(src->i_block[0]));
1653 btrfs_set_stack_inode_rdev(dst,
1654 new_decode_dev(src->i_block[1]));
1657 memset(&dst->reserved, 0, sizeof(dst->reserved));
1661 * copy a single inode. do all the required works, such as cloning
1662 * inode item, creating file extents and creating directory entries.
1664 static int ext2_copy_single_inode(struct btrfs_trans_handle *trans,
1665 struct btrfs_root *root, u64 objectid,
1666 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1667 struct ext2_inode *ext2_inode,
1668 int datacsum, int packing, int noxattr)
1671 struct btrfs_inode_item btrfs_inode;
1673 if (ext2_inode->i_links_count == 0)
1676 ext2_copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
1677 if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
1678 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
1679 BTRFS_INODE_NODATASUM;
1680 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
1683 switch (ext2_inode->i_mode & S_IFMT) {
1685 ret = ext2_create_file_extents(trans, root, objectid,
1686 &btrfs_inode, ext2_fs, ext2_ino, datacsum, packing);
1689 ret = ext2_create_dir_entries(trans, root, objectid,
1690 &btrfs_inode, ext2_fs, ext2_ino);
1693 ret = ext2_create_symbol_link(trans, root, objectid,
1694 &btrfs_inode, ext2_fs, ext2_ino, ext2_inode);
1704 ret = ext2_copy_extended_attrs(trans, root, objectid,
1705 &btrfs_inode, ext2_fs, ext2_ino);
1709 return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
1713 * scan ext2's inode bitmap and copy all used inodes.
1715 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
1716 struct btrfs_root *root,
1717 int datacsum, int packing, int noxattr, struct task_ctx *p)
1719 ext2_filsys ext2_fs = cctx->fs_data;
1722 ext2_inode_scan ext2_scan;
1723 struct ext2_inode ext2_inode;
1724 ext2_ino_t ext2_ino;
1726 struct btrfs_trans_handle *trans;
1728 trans = btrfs_start_transaction(root, 1);
1731 err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
1733 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
1736 while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
1738 /* no more inodes */
1741 /* skip special inode in ext2fs */
1742 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
1743 ext2_ino != EXT2_ROOT_INO)
1745 objectid = ext2_ino + INO_OFFSET;
1746 ret = ext2_copy_single_inode(trans, root,
1747 objectid, ext2_fs, ext2_ino,
1748 &ext2_inode, datacsum, packing,
1750 p->cur_copy_inodes++;
1753 if (trans->blocks_used >= 4096) {
1754 ret = btrfs_commit_transaction(trans, root);
1756 trans = btrfs_start_transaction(root, 1);
1761 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
1764 ret = btrfs_commit_transaction(trans, root);
1766 ext2fs_close_inode_scan(ext2_scan);
1772 * Create the fs image file of old filesystem.
1774 * This is completely fs independent as we have cctx->used, only
1775 * need to create file extents pointing to all the positions.
1777 static int create_image(struct btrfs_root *root,
1778 struct btrfs_mkfs_config *cfg,
1779 struct btrfs_convert_context *cctx, int fd,
1780 u64 size, char *name, int datacsum)
1782 struct btrfs_inode_item buf;
1783 struct btrfs_trans_handle *trans;
1784 struct btrfs_path *path = NULL;
1785 struct btrfs_key key;
1786 struct cache_extent *cache;
1787 struct cache_tree used_tmp;
1790 u64 flags = BTRFS_INODE_READONLY;
1794 flags |= BTRFS_INODE_NODATASUM;
1796 trans = btrfs_start_transaction(root, 1);
1800 cache_tree_init(&used_tmp);
1802 ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
1806 ret = btrfs_new_inode(trans, root, ino, 0400 | S_IFREG);
1809 ret = btrfs_change_inode_flags(trans, root, ino, flags);
1812 ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
1813 strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
1817 path = btrfs_alloc_path();
1823 key.type = BTRFS_INODE_ITEM_KEY;
1826 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1828 ret = (ret > 0 ? -ENOENT : ret);
1831 read_extent_buffer(path->nodes[0], &buf,
1832 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1834 btrfs_release_path(path);
1837 * Create a new used space cache, which doesn't contain the reserved
1840 for (cache = first_cache_extent(&cctx->used); cache;
1841 cache = next_cache_extent(cache)) {
1842 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1846 ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1851 * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1852 * can't handle bytenr 0(will consider it as a hole)
1855 while (cur < size) {
1856 u64 len = size - cur;
1858 ret = create_image_file_range(trans, root, &used_tmp,
1859 &buf, ino, cur, &len, datacsum);
1864 /* Handle the reserved ranges */
1865 ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1866 cfg->num_bytes, datacsum);
1870 key.type = BTRFS_INODE_ITEM_KEY;
1872 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1874 ret = (ret > 0 ? -ENOENT : ret);
1877 btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1878 write_extent_buffer(path->nodes[0], &buf,
1879 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1882 free_extent_cache_tree(&used_tmp);
1883 btrfs_free_path(path);
1884 btrfs_commit_transaction(trans, root);
1888 static struct btrfs_root * link_subvol(struct btrfs_root *root,
1889 const char *base, u64 root_objectid)
1891 struct btrfs_trans_handle *trans;
1892 struct btrfs_fs_info *fs_info = root->fs_info;
1893 struct btrfs_root *tree_root = fs_info->tree_root;
1894 struct btrfs_root *new_root = NULL;
1895 struct btrfs_path *path;
1896 struct btrfs_inode_item *inode_item;
1897 struct extent_buffer *leaf;
1898 struct btrfs_key key;
1899 u64 dirid = btrfs_root_dirid(&root->root_item);
1901 char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1907 if (len == 0 || len > BTRFS_NAME_LEN)
1910 path = btrfs_alloc_path();
1913 key.objectid = dirid;
1914 key.type = BTRFS_DIR_INDEX_KEY;
1915 key.offset = (u64)-1;
1917 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1920 if (path->slots[0] > 0) {
1922 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1923 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1924 index = key.offset + 1;
1926 btrfs_release_path(path);
1928 trans = btrfs_start_transaction(root, 1);
1931 key.objectid = dirid;
1933 key.type = BTRFS_INODE_ITEM_KEY;
1935 ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1937 leaf = path->nodes[0];
1938 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1939 struct btrfs_inode_item);
1941 key.objectid = root_objectid;
1942 key.offset = (u64)-1;
1943 key.type = BTRFS_ROOT_ITEM_KEY;
1945 memcpy(buf, base, len);
1946 for (i = 0; i < 1024; i++) {
1947 ret = btrfs_insert_dir_item(trans, root, buf, len,
1948 dirid, &key, BTRFS_FT_DIR, index);
1951 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1952 if (len < 1 || len > BTRFS_NAME_LEN) {
1960 btrfs_set_inode_size(leaf, inode_item, len * 2 +
1961 btrfs_inode_size(leaf, inode_item));
1962 btrfs_mark_buffer_dirty(leaf);
1963 btrfs_release_path(path);
1965 /* add the backref first */
1966 ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1967 BTRFS_ROOT_BACKREF_KEY,
1968 root->root_key.objectid,
1969 dirid, index, buf, len);
1972 /* now add the forward ref */
1973 ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1974 BTRFS_ROOT_REF_KEY, root_objectid,
1975 dirid, index, buf, len);
1977 ret = btrfs_commit_transaction(trans, root);
1980 new_root = btrfs_read_fs_root(fs_info, &key);
1981 if (IS_ERR(new_root))
1984 btrfs_free_path(path);
1988 static int create_subvol(struct btrfs_trans_handle *trans,
1989 struct btrfs_root *root, u64 root_objectid)
1991 struct extent_buffer *tmp;
1992 struct btrfs_root *new_root;
1993 struct btrfs_key key;
1994 struct btrfs_root_item root_item;
1997 ret = btrfs_copy_root(trans, root, root->node, &tmp,
2001 memcpy(&root_item, &root->root_item, sizeof(root_item));
2002 btrfs_set_root_bytenr(&root_item, tmp->start);
2003 btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
2004 btrfs_set_root_generation(&root_item, trans->transid);
2005 free_extent_buffer(tmp);
2007 key.objectid = root_objectid;
2008 key.type = BTRFS_ROOT_ITEM_KEY;
2009 key.offset = trans->transid;
2010 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
2013 key.offset = (u64)-1;
2014 new_root = btrfs_read_fs_root(root->fs_info, &key);
2015 BUG_ON(!new_root || IS_ERR(new_root));
2017 ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
2024 * New make_btrfs() has handle system and meta chunks quite well.
2025 * So only need to add remaining data chunks.
2027 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
2028 struct btrfs_fs_info *fs_info,
2029 struct btrfs_mkfs_config *cfg,
2030 struct btrfs_convert_context *cctx)
2032 struct btrfs_root *extent_root = fs_info->extent_root;
2033 struct cache_tree *data_chunks = &cctx->data_chunks;
2034 struct cache_extent *cache;
2039 * Don't create data chunk over 10% of the convert device
2040 * And for single chunk, don't create chunk larger than 1G.
2042 max_chunk_size = cfg->num_bytes / 10;
2043 max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
2044 max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
2046 for (cache = first_cache_extent(data_chunks); cache;
2047 cache = next_cache_extent(cache)) {
2048 u64 cur = cache->start;
2050 while (cur < cache->start + cache->size) {
2052 u64 cur_backup = cur;
2054 len = min(max_chunk_size,
2055 cache->start + cache->size - cur);
2056 ret = btrfs_alloc_data_chunk(trans, extent_root,
2058 BTRFS_BLOCK_GROUP_DATA, 1);
2061 ret = btrfs_make_block_group(trans, extent_root, 0,
2062 BTRFS_BLOCK_GROUP_DATA,
2063 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
2074 * Init the temp btrfs to a operational status.
2076 * It will fix the extent usage accounting(XXX: Do we really need?) and
2077 * insert needed data chunks, to ensure all old fs data extents are covered
2078 * by DATA chunks, preventing wrong chunks are allocated.
2080 * And also create convert image subvolume and relocation tree.
2081 * (XXX: Not need again?)
2082 * But the convert image subvolume is *NOT* linked to fs tree yet.
2084 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
2085 struct btrfs_convert_context *cctx, int datacsum,
2086 int packing, int noxattr)
2088 struct btrfs_key location;
2089 struct btrfs_trans_handle *trans;
2090 struct btrfs_fs_info *fs_info = root->fs_info;
2094 * Don't alloc any metadata/system chunk, as we don't want
2095 * any meta/sys chunk allcated before all data chunks are inserted.
2096 * Or we screw up the chunk layout just like the old implement.
2098 fs_info->avoid_sys_chunk_alloc = 1;
2099 fs_info->avoid_meta_chunk_alloc = 1;
2100 trans = btrfs_start_transaction(root, 1);
2102 ret = btrfs_fix_block_accounting(trans, root);
2105 ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
2108 ret = btrfs_make_root_dir(trans, fs_info->tree_root,
2109 BTRFS_ROOT_TREE_DIR_OBJECTID);
2112 memcpy(&location, &root->root_key, sizeof(location));
2113 location.offset = (u64)-1;
2114 ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
2115 btrfs_super_root_dir(fs_info->super_copy),
2116 &location, BTRFS_FT_DIR, 0);
2119 ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
2121 btrfs_super_root_dir(fs_info->super_copy), 0);
2124 btrfs_set_root_dirid(&fs_info->fs_root->root_item,
2125 BTRFS_FIRST_FREE_OBJECTID);
2127 /* subvol for fs image file */
2128 ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
2131 /* subvol for data relocation tree */
2132 ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
2136 ret = btrfs_commit_transaction(trans, root);
2137 fs_info->avoid_sys_chunk_alloc = 0;
2138 fs_info->avoid_meta_chunk_alloc = 0;
2144 * Migrate super block to its default position and zero 0 ~ 16k
2146 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
2149 struct extent_buffer *buf;
2150 struct btrfs_super_block *super;
2154 BUG_ON(sectorsize < sizeof(*super));
2155 buf = malloc(sizeof(*buf) + sectorsize);
2159 buf->len = sectorsize;
2160 ret = pread(fd, buf->data, sectorsize, old_bytenr);
2161 if (ret != sectorsize)
2164 super = (struct btrfs_super_block *)buf->data;
2165 BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
2166 btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
2168 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
2169 ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2170 if (ret != sectorsize)
2177 memset(buf->data, 0, sectorsize);
2178 for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
2179 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
2180 if (len > sectorsize)
2182 ret = pwrite(fd, buf->data, len, bytenr);
2184 fprintf(stderr, "unable to zero fill device\n");
2198 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
2200 struct btrfs_chunk *chunk;
2201 struct btrfs_disk_key *key;
2202 u32 sectorsize = btrfs_super_sectorsize(super);
2204 key = (struct btrfs_disk_key *)(super->sys_chunk_array);
2205 chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
2206 sizeof(struct btrfs_disk_key));
2208 btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
2209 btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
2210 btrfs_set_disk_key_offset(key, 0);
2212 btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
2213 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
2214 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
2215 btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
2216 btrfs_set_stack_chunk_io_align(chunk, sectorsize);
2217 btrfs_set_stack_chunk_io_width(chunk, sectorsize);
2218 btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
2219 btrfs_set_stack_chunk_num_stripes(chunk, 1);
2220 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
2221 chunk->stripe.devid = super->dev_item.devid;
2222 btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
2223 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
2224 btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
2228 static const struct btrfs_convert_operations ext2_convert_ops = {
2230 .open_fs = ext2_open_fs,
2231 .read_used_space = ext2_read_used_space,
2232 .copy_inodes = ext2_copy_inodes,
2233 .close_fs = ext2_close_fs,
2236 static const struct btrfs_convert_operations *convert_operations[] = {
2240 static int convert_open_fs(const char *devname,
2241 struct btrfs_convert_context *cctx)
2245 memset(cctx, 0, sizeof(*cctx));
2247 for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
2248 int ret = convert_operations[i]->open_fs(cctx, devname);
2251 cctx->convert_ops = convert_operations[i];
2256 fprintf(stderr, "No file system found to convert.\n");
2261 * Read used space, and since we have the used space,
2262 * calcuate data_chunks and free for later mkfs
2264 static int convert_read_used_space(struct btrfs_convert_context *cctx)
2268 ret = cctx->convert_ops->read_used_space(cctx);
2272 ret = calculate_available_space(cctx);
2276 static int do_convert(const char *devname, int datacsum, int packing,
2277 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2278 int progress, u64 features)
2285 struct btrfs_root *root;
2286 struct btrfs_root *image_root;
2287 struct btrfs_convert_context cctx;
2288 struct btrfs_key key;
2289 char *subvol_name = NULL;
2290 struct task_ctx ctx;
2291 char features_buf[64];
2292 struct btrfs_mkfs_config mkfs_cfg;
2294 init_convert_context(&cctx);
2295 ret = convert_open_fs(devname, &cctx);
2298 ret = convert_read_used_space(&cctx);
2302 blocksize = cctx.blocksize;
2303 total_bytes = (u64)blocksize * (u64)cctx.block_count;
2304 if (blocksize < 4096) {
2305 fprintf(stderr, "block size is too small\n");
2308 if (btrfs_check_nodesize(nodesize, blocksize, features))
2310 fd = open(devname, O_RDWR);
2312 fprintf(stderr, "unable to open %s\n", devname);
2315 btrfs_parse_features_to_string(features_buf, features);
2316 if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2317 strcat(features_buf, " (default)");
2319 printf("create btrfs filesystem:\n");
2320 printf("\tblocksize: %u\n", blocksize);
2321 printf("\tnodesize: %u\n", nodesize);
2322 printf("\tfeatures: %s\n", features_buf);
2324 mkfs_cfg.label = cctx.volume_name;
2325 mkfs_cfg.num_bytes = total_bytes;
2326 mkfs_cfg.nodesize = nodesize;
2327 mkfs_cfg.sectorsize = blocksize;
2328 mkfs_cfg.stripesize = blocksize;
2329 mkfs_cfg.features = features;
2330 /* New convert need these space */
2331 mkfs_cfg.fs_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2332 mkfs_cfg.chunk_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2333 *(mkfs_cfg.fs_uuid) = '\0';
2334 *(mkfs_cfg.chunk_uuid) = '\0';
2336 ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2338 fprintf(stderr, "unable to create initial ctree: %s\n",
2343 root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2344 OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2346 fprintf(stderr, "unable to open ctree\n");
2349 ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2351 fprintf(stderr, "unable to setup the root tree\n");
2355 printf("creating %s image file.\n", cctx.convert_ops->name);
2356 ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2358 fprintf(stderr, "error allocating subvolume name: %s_saved\n",
2359 cctx.convert_ops->name);
2362 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2363 key.offset = (u64)-1;
2364 key.type = BTRFS_ROOT_ITEM_KEY;
2365 image_root = btrfs_read_fs_root(root->fs_info, &key);
2367 fprintf(stderr, "unable to create subvol\n");
2370 ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2371 mkfs_cfg.num_bytes, "image", datacsum);
2373 fprintf(stderr, "error during create_image %d\n", ret);
2377 printf("creating btrfs metadata.\n");
2378 ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2379 ctx.cur_copy_inodes = 0;
2382 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2384 task_start(ctx.info);
2386 ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2388 fprintf(stderr, "error during copy_inodes %d\n", ret);
2392 task_stop(ctx.info);
2393 task_deinit(ctx.info);
2396 image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2400 memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2401 if (copylabel == 1) {
2402 __strncpy_null(root->fs_info->super_copy->label,
2403 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2404 fprintf(stderr, "copy label '%s'\n",
2405 root->fs_info->super_copy->label);
2406 } else if (copylabel == -1) {
2407 strcpy(root->fs_info->super_copy->label, fslabel);
2408 fprintf(stderr, "set label to '%s'\n", fslabel);
2411 ret = close_ctree(root);
2413 fprintf(stderr, "error during close_ctree %d\n", ret);
2416 convert_close_fs(&cctx);
2417 clean_convert_context(&cctx);
2420 * If this step succeed, we get a mountable btrfs. Otherwise
2421 * the source fs is left unchanged.
2423 ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2425 fprintf(stderr, "unable to migrate super block\n");
2430 root = open_ctree_fd(fd, devname, 0,
2431 OPEN_CTREE_WRITES | OPEN_CTREE_FS_PARTIAL);
2433 fprintf(stderr, "unable to open ctree\n");
2436 root->fs_info->finalize_on_close = 1;
2440 printf("conversion complete.\n");
2443 clean_convert_context(&cctx);
2448 "WARNING: an error occurred during chunk mapping fixup, filesystem mountable but not finalized\n");
2450 fprintf(stderr, "conversion aborted\n");
2455 * Check if a non 1:1 mapped chunk can be rolled back.
2456 * For new convert, it's OK while for old convert it's not.
2458 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2460 struct btrfs_block_group_cache *bg;
2461 struct btrfs_key key;
2462 struct btrfs_path path;
2463 struct btrfs_root *extent_root = fs_info->extent_root;
2468 bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2471 bg_start = bg->key.objectid;
2472 bg_end = bg->key.objectid + bg->key.offset;
2474 key.objectid = bg_end;
2475 key.type = BTRFS_METADATA_ITEM_KEY;
2477 btrfs_init_path(&path);
2479 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2484 struct btrfs_extent_item *ei;
2486 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2494 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2495 if (key.type == BTRFS_METADATA_ITEM_KEY)
2497 /* Now it's EXTENT_ITEM_KEY only */
2498 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2499 struct btrfs_extent_item);
2501 * Found data extent, means this is old convert must follow 1:1
2504 if (btrfs_extent_flags(path.nodes[0], ei)
2505 & BTRFS_EXTENT_FLAG_DATA) {
2510 btrfs_release_path(&path);
2514 static int may_rollback(struct btrfs_root *root)
2516 struct btrfs_fs_info *info = root->fs_info;
2517 struct btrfs_multi_bio *multi = NULL;
2525 if (btrfs_super_num_devices(info->super_copy) != 1)
2528 bytenr = BTRFS_SUPER_INFO_OFFSET;
2529 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2532 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2533 &length, &multi, 0, NULL);
2535 if (ret == -ENOENT) {
2536 /* removed block group at the tail */
2537 if (length == (u64)-1)
2540 /* removed block group in the middle */
2546 num_stripes = multi->num_stripes;
2547 physical = multi->stripes[0].physical;
2550 if (num_stripes != 1) {
2551 error("num stripes for bytenr %llu is not 1", bytenr);
2556 * Extra check for new convert, as metadata chunk from new
2557 * convert is much more free than old convert, it doesn't need
2558 * to do 1:1 mapping.
2560 if (physical != bytenr) {
2562 * Check if it's a metadata chunk and has only metadata
2565 ret = may_rollback_chunk(info, bytenr);
2571 if (bytenr >= total_bytes)
2579 static int do_rollback(const char *devname)
2584 struct btrfs_root *root;
2585 struct btrfs_root *image_root;
2586 struct btrfs_root *chunk_root;
2587 struct btrfs_dir_item *dir;
2588 struct btrfs_inode_item *inode;
2589 struct btrfs_file_extent_item *fi;
2590 struct btrfs_trans_handle *trans;
2591 struct extent_buffer *leaf;
2592 struct btrfs_block_group_cache *cache1;
2593 struct btrfs_block_group_cache *cache2;
2594 struct btrfs_key key;
2595 struct btrfs_path path;
2596 struct extent_io_tree io_tree;
2611 extent_io_tree_init(&io_tree);
2613 fd = open(devname, O_RDWR);
2615 fprintf(stderr, "unable to open %s\n", devname);
2618 root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2620 fprintf(stderr, "unable to open ctree\n");
2623 ret = may_rollback(root);
2625 fprintf(stderr, "unable to do rollback\n");
2629 sectorsize = root->sectorsize;
2630 buf = malloc(sectorsize);
2632 fprintf(stderr, "unable to allocate memory\n");
2636 btrfs_init_path(&path);
2638 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2639 key.type = BTRFS_ROOT_BACKREF_KEY;
2640 key.offset = BTRFS_FS_TREE_OBJECTID;
2641 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2643 btrfs_release_path(&path);
2646 "ERROR: unable to convert ext2 image subvolume, is it deleted?\n");
2648 } else if (ret < 0) {
2650 "ERROR: unable to open ext2_saved, id=%llu: %s\n",
2651 (unsigned long long)key.objectid, strerror(-ret));
2655 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2656 key.type = BTRFS_ROOT_ITEM_KEY;
2657 key.offset = (u64)-1;
2658 image_root = btrfs_read_fs_root(root->fs_info, &key);
2659 if (!image_root || IS_ERR(image_root)) {
2660 fprintf(stderr, "unable to open subvol %llu\n",
2661 (unsigned long long)key.objectid);
2666 root_dir = btrfs_root_dirid(&root->root_item);
2667 dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2668 root_dir, name, strlen(name), 0);
2669 if (!dir || IS_ERR(dir)) {
2670 fprintf(stderr, "unable to find file %s\n", name);
2673 leaf = path.nodes[0];
2674 btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2675 btrfs_release_path(&path);
2677 objectid = key.objectid;
2679 ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2681 fprintf(stderr, "unable to find inode item\n");
2684 leaf = path.nodes[0];
2685 inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2686 total_bytes = btrfs_inode_size(leaf, inode);
2687 btrfs_release_path(&path);
2689 key.objectid = objectid;
2691 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2692 ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2694 fprintf(stderr, "unable to find first file extent\n");
2695 btrfs_release_path(&path);
2699 /* build mapping tree for the relocated blocks */
2700 for (offset = 0; offset < total_bytes; ) {
2701 leaf = path.nodes[0];
2702 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2703 ret = btrfs_next_leaf(root, &path);
2709 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2710 if (key.objectid != objectid || key.offset != offset ||
2711 btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2714 fi = btrfs_item_ptr(leaf, path.slots[0],
2715 struct btrfs_file_extent_item);
2716 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2718 if (btrfs_file_extent_compression(leaf, fi) ||
2719 btrfs_file_extent_encryption(leaf, fi) ||
2720 btrfs_file_extent_other_encoding(leaf, fi))
2723 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2724 /* skip holes and direct mapped extents */
2725 if (bytenr == 0 || bytenr == offset)
2728 bytenr += btrfs_file_extent_offset(leaf, fi);
2729 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2731 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2732 cache2 = btrfs_lookup_block_group(root->fs_info,
2733 offset + num_bytes - 1);
2735 * Here we must take consideration of old and new convert
2737 * For old convert case, sign, there is no consist chunk type
2738 * that will cover the extent. META/DATA/SYS are all possible.
2739 * Just ensure relocate one is in SYS chunk.
2740 * For new convert case, they are all covered by DATA chunk.
2742 * So, there is not valid chunk type check for it now.
2744 if (cache1 != cache2)
2747 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2748 EXTENT_LOCKED, GFP_NOFS);
2749 set_state_private(&io_tree, offset, bytenr);
2751 offset += btrfs_file_extent_num_bytes(leaf, fi);
2754 btrfs_release_path(&path);
2756 if (offset < total_bytes) {
2757 fprintf(stderr, "unable to build extent mapping\n");
2758 fprintf(stderr, "converted filesystem after balance is unable to rollback\n");
2762 first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2763 first_free &= ~((u64)sectorsize - 1);
2764 /* backup for extent #0 should exist */
2765 if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2766 fprintf(stderr, "no backup for the first extent\n");
2769 /* force no allocation from system block group */
2770 root->fs_info->system_allocs = -1;
2771 trans = btrfs_start_transaction(root, 1);
2774 * recow the whole chunk tree, this will remove all chunk tree blocks
2775 * from system block group
2777 chunk_root = root->fs_info->chunk_root;
2778 memset(&key, 0, sizeof(key));
2780 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2784 ret = btrfs_next_leaf(chunk_root, &path);
2788 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2789 btrfs_release_path(&path);
2791 btrfs_release_path(&path);
2796 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2800 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2801 num_bytes += btrfs_block_group_used(&cache1->item);
2803 offset = cache1->key.objectid + cache1->key.offset;
2805 /* only extent #0 left in system block group? */
2806 if (num_bytes > first_free) {
2807 fprintf(stderr, "unable to empty system block group\n");
2810 /* create a system chunk that maps the whole device */
2811 ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2813 fprintf(stderr, "unable to update system chunk\n");
2817 ret = btrfs_commit_transaction(trans, root);
2820 ret = close_ctree(root);
2822 fprintf(stderr, "error during close_ctree %d\n", ret);
2826 /* zero btrfs super block mirrors */
2827 memset(buf, 0, sectorsize);
2828 for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2829 bytenr = btrfs_sb_offset(i);
2830 if (bytenr >= total_bytes)
2832 ret = pwrite(fd, buf, sectorsize, bytenr);
2833 if (ret != sectorsize) {
2835 "error during zeroing superblock %d: %d\n",
2841 sb_bytenr = (u64)-1;
2842 /* copy all relocated blocks back */
2844 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2849 ret = get_state_private(&io_tree, start, &bytenr);
2852 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2855 while (start <= end) {
2856 if (start == BTRFS_SUPER_INFO_OFFSET) {
2860 ret = pread(fd, buf, sectorsize, bytenr);
2862 fprintf(stderr, "error during pread %d\n", ret);
2865 BUG_ON(ret != sectorsize);
2866 ret = pwrite(fd, buf, sectorsize, start);
2868 fprintf(stderr, "error during pwrite %d\n", ret);
2871 BUG_ON(ret != sectorsize);
2873 start += sectorsize;
2874 bytenr += sectorsize;
2880 fprintf(stderr, "error during fsync %d\n", ret);
2884 * finally, overwrite btrfs super block.
2886 ret = pread(fd, buf, sectorsize, sb_bytenr);
2888 fprintf(stderr, "error during pread %d\n", ret);
2891 BUG_ON(ret != sectorsize);
2892 ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2894 fprintf(stderr, "error during pwrite %d\n", ret);
2897 BUG_ON(ret != sectorsize);
2900 fprintf(stderr, "error during fsync %d\n", ret);
2906 extent_io_tree_cleanup(&io_tree);
2907 printf("rollback complete.\n");
2914 fprintf(stderr, "rollback aborted.\n");
2918 static void print_usage(void)
2920 printf("usage: btrfs-convert [options] device\n");
2921 printf("options:\n");
2922 printf("\t-d|--no-datasum disable data checksum, sets NODATASUM\n");
2923 printf("\t-i|--no-xattr ignore xattrs and ACLs\n");
2924 printf("\t-n|--no-inline disable inlining of small files to metadata\n");
2925 printf("\t-N|--nodesize SIZE set filesystem metadata nodesize\n");
2926 printf("\t-r|--rollback roll back to the original filesystem\n");
2927 printf("\t-l|--label LABEL set filesystem label\n");
2928 printf("\t-L|--copy-label use label from converted filesystem\n");
2929 printf("\t-p|--progress show converting progress (default)\n");
2930 printf("\t-O|--features LIST comma separated list of filesystem features\n");
2931 printf("\t--no-progress show only overview, not the detailed progress\n");
2934 int main(int argc, char *argv[])
2940 u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
2941 BTRFS_MKFS_DEFAULT_NODE_SIZE);
2944 int usage_error = 0;
2947 char fslabel[BTRFS_LABEL_SIZE];
2948 u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
2951 enum { GETOPT_VAL_NO_PROGRESS = 256 };
2952 static const struct option long_options[] = {
2953 { "no-progress", no_argument, NULL,
2954 GETOPT_VAL_NO_PROGRESS },
2955 { "no-datasum", no_argument, NULL, 'd' },
2956 { "no-inline", no_argument, NULL, 'n' },
2957 { "no-xattr", no_argument, NULL, 'i' },
2958 { "rollback", no_argument, NULL, 'r' },
2959 { "features", required_argument, NULL, 'O' },
2960 { "progress", no_argument, NULL, 'p' },
2961 { "label", required_argument, NULL, 'l' },
2962 { "copy-label", no_argument, NULL, 'L' },
2963 { "nodesize", required_argument, NULL, 'N' },
2964 { "help", no_argument, NULL, GETOPT_VAL_HELP},
2965 { NULL, 0, NULL, 0 }
2967 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
2982 nodesize = parse_size(optarg);
2989 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
2991 "WARNING: label too long, trimmed to %d bytes\n",
2992 BTRFS_LABEL_SIZE - 1);
2994 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
3003 char *orig = strdup(optarg);
3006 tmp = btrfs_parse_fs_features(tmp, &features);
3009 "Unrecognized filesystem feature '%s'\n",
3015 if (features & BTRFS_FEATURE_LIST_ALL) {
3016 btrfs_list_all_fs_features(
3017 ~BTRFS_CONVERT_ALLOWED_FEATURES);
3020 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
3023 btrfs_parse_features_to_string(buf,
3024 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
3026 "ERROR: features not allowed for convert: %s\n",
3033 case GETOPT_VAL_NO_PROGRESS:
3036 case GETOPT_VAL_HELP:
3039 return c != GETOPT_VAL_HELP;
3043 if (check_argc_exact(argc - optind, 1)) {
3048 if (rollback && (!datacsum || noxattr || !packing)) {
3050 "Usage error: -d, -i, -n options do not apply to rollback\n");
3059 file = argv[optind];
3060 ret = check_mounted(file);
3062 fprintf(stderr, "Could not check mount status: %s\n",
3066 fprintf(stderr, "%s is mounted\n", file);
3071 ret = do_rollback(file);
3073 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3074 copylabel, fslabel, progress, features);