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"
40 #include <ext2fs/ext2_fs.h>
41 #include <ext2fs/ext2fs.h>
42 #include <ext2fs/ext2_ext_attr.h>
44 #define INO_OFFSET (BTRFS_FIRST_FREE_OBJECTID - EXT2_ROOT_INO)
45 #define CONV_IMAGE_SUBVOL_OBJECTID BTRFS_FIRST_FREE_OBJECTID
48 * Compatibility code for e2fsprogs 1.41 which doesn't support RO compat flag
50 * Unlike normal RO compat flag, BIGALLOC affects how e2fsprogs check used
51 * space, and btrfs-convert heavily relies on it.
53 #ifdef HAVE_OLD_E2FSPROGS
54 #define EXT2FS_CLUSTER_RATIO(fs) (1)
55 #define EXT2_CLUSTERS_PER_GROUP(s) (EXT2_BLOCKS_PER_GROUP(s))
56 #define EXT2FS_B2C(fs, blk) (blk)
60 uint32_t max_copy_inodes;
61 uint32_t cur_copy_inodes;
62 struct task_info *info;
65 static void *print_copied_inodes(void *p)
67 struct task_ctx *priv = p;
68 const char work_indicator[] = { '.', 'o', 'O', 'o' };
71 task_period_start(priv->info, 1000 /* 1s */);
74 printf("copy inodes [%c] [%10d/%10d]\r",
75 work_indicator[count % 4], priv->cur_copy_inodes,
76 priv->max_copy_inodes);
78 task_period_wait(priv->info);
84 static int after_copied_inodes(void *p)
92 struct btrfs_convert_context;
93 struct btrfs_convert_operations {
95 int (*open_fs)(struct btrfs_convert_context *cctx, const char *devname);
96 int (*read_used_space)(struct btrfs_convert_context *cctx);
97 int (*copy_inodes)(struct btrfs_convert_context *cctx,
98 struct btrfs_root *root, int datacsum,
99 int packing, int noxattr, struct task_ctx *p);
100 void (*close_fs)(struct btrfs_convert_context *cctx);
103 static void init_convert_context(struct btrfs_convert_context *cctx)
105 cache_tree_init(&cctx->used);
106 cache_tree_init(&cctx->data_chunks);
107 cache_tree_init(&cctx->free);
110 static void clean_convert_context(struct btrfs_convert_context *cctx)
112 free_extent_cache_tree(&cctx->used);
113 free_extent_cache_tree(&cctx->data_chunks);
114 free_extent_cache_tree(&cctx->free);
117 static inline int copy_inodes(struct btrfs_convert_context *cctx,
118 struct btrfs_root *root, int datacsum,
119 int packing, int noxattr, struct task_ctx *p)
121 return cctx->convert_ops->copy_inodes(cctx, root, datacsum, packing,
125 static inline void convert_close_fs(struct btrfs_convert_context *cctx)
127 cctx->convert_ops->close_fs(cctx);
131 * Open Ext2fs in readonly mode, read block allocation bitmap and
132 * inode bitmap into memory.
134 static int ext2_open_fs(struct btrfs_convert_context *cctx, const char *name)
141 ret = ext2fs_open(name, 0, 0, 0, unix_io_manager, &ext2_fs);
143 fprintf(stderr, "ext2fs_open: %s\n", error_message(ret));
147 * We need to know exactly the used space, some RO compat flags like
148 * BIGALLOC will affect how used space is present.
149 * So we need manuall check any unsupported RO compat flags
151 ro_feature = ext2_fs->super->s_feature_ro_compat;
152 if (ro_feature & ~EXT2_LIB_FEATURE_RO_COMPAT_SUPP) {
154 "unsupported RO features detected: %x, abort convert to avoid possible corruption",
155 ro_feature & ~EXT2_LIB_FEATURE_COMPAT_SUPP);
158 ret = ext2fs_read_inode_bitmap(ext2_fs);
160 fprintf(stderr, "ext2fs_read_inode_bitmap: %s\n",
164 ret = ext2fs_read_block_bitmap(ext2_fs);
166 fprintf(stderr, "ext2fs_read_block_bitmap: %s\n",
171 * search each block group for a free inode. this set up
172 * uninit block/inode bitmaps appropriately.
175 while (ino <= ext2_fs->super->s_inodes_count) {
177 ext2fs_new_inode(ext2_fs, ino, 0, NULL, &foo);
178 ino += EXT2_INODES_PER_GROUP(ext2_fs->super);
181 if (!(ext2_fs->super->s_feature_incompat &
182 EXT2_FEATURE_INCOMPAT_FILETYPE)) {
183 fprintf(stderr, "filetype feature is missing\n");
187 cctx->fs_data = ext2_fs;
188 cctx->blocksize = ext2_fs->blocksize;
189 cctx->block_count = ext2_fs->super->s_blocks_count;
190 cctx->total_bytes = ext2_fs->blocksize * ext2_fs->super->s_blocks_count;
191 cctx->volume_name = strndup(ext2_fs->super->s_volume_name, 16);
192 cctx->first_data_block = ext2_fs->super->s_first_data_block;
193 cctx->inodes_count = ext2_fs->super->s_inodes_count;
194 cctx->free_inodes_count = ext2_fs->super->s_free_inodes_count;
197 ext2fs_close(ext2_fs);
201 static int __ext2_add_one_block(ext2_filsys fs, char *bitmap,
202 unsigned long group_nr, struct cache_tree *used)
204 unsigned long offset;
208 offset = fs->super->s_first_data_block;
209 offset /= EXT2FS_CLUSTER_RATIO(fs);
210 offset += group_nr * EXT2_CLUSTERS_PER_GROUP(fs->super);
211 for (i = 0; i < EXT2_CLUSTERS_PER_GROUP(fs->super); i++) {
212 if (ext2fs_test_bit(i, bitmap)) {
215 start = (i + offset) * EXT2FS_CLUSTER_RATIO(fs);
216 start *= fs->blocksize;
217 ret = add_merge_cache_extent(used, start,
227 * Read all used ext2 space into cctx->used cache tree
229 static int ext2_read_used_space(struct btrfs_convert_context *cctx)
231 ext2_filsys fs = (ext2_filsys)cctx->fs_data;
232 blk64_t blk_itr = EXT2FS_B2C(fs, fs->super->s_first_data_block);
233 struct cache_tree *used_tree = &cctx->used;
234 char *block_bitmap = NULL;
239 block_nbytes = EXT2_CLUSTERS_PER_GROUP(fs->super) / 8;
240 /* Shouldn't happen */
241 BUG_ON(!fs->block_map);
243 block_bitmap = malloc(block_nbytes);
247 for (i = 0; i < fs->group_desc_count; i++) {
248 ret = ext2fs_get_block_bitmap_range(fs->block_map, blk_itr,
249 block_nbytes * 8, block_bitmap);
251 error("fail to get bitmap from ext2, %s",
255 ret = __ext2_add_one_block(fs, block_bitmap, i, used_tree);
257 error("fail to build used space tree, %s",
261 blk_itr += EXT2_CLUSTERS_PER_GROUP(fs->super);
268 static void ext2_close_fs(struct btrfs_convert_context *cctx)
270 if (cctx->volume_name) {
271 free(cctx->volume_name);
272 cctx->volume_name = NULL;
274 ext2fs_close(cctx->fs_data);
277 static int intersect_with_sb(u64 bytenr, u64 num_bytes)
282 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
283 offset = btrfs_sb_offset(i);
284 offset &= ~((u64)BTRFS_STRIPE_LEN - 1);
286 if (bytenr < offset + BTRFS_STRIPE_LEN &&
287 bytenr + num_bytes > offset)
293 static int convert_insert_dirent(struct btrfs_trans_handle *trans,
294 struct btrfs_root *root,
295 const char *name, size_t name_len,
296 u64 dir, u64 objectid,
297 u8 file_type, u64 index_cnt,
298 struct btrfs_inode_item *inode)
302 struct btrfs_key location = {
303 .objectid = objectid,
305 .type = BTRFS_INODE_ITEM_KEY,
308 ret = btrfs_insert_dir_item(trans, root, name, name_len,
309 dir, &location, file_type, index_cnt);
312 ret = btrfs_insert_inode_ref(trans, root, name, name_len,
313 objectid, dir, index_cnt);
316 inode_size = btrfs_stack_inode_size(inode) + name_len * 2;
317 btrfs_set_stack_inode_size(inode, inode_size);
322 struct dir_iterate_data {
323 struct btrfs_trans_handle *trans;
324 struct btrfs_root *root;
325 struct btrfs_inode_item *inode;
332 static u8 filetype_conversion_table[EXT2_FT_MAX] = {
333 [EXT2_FT_UNKNOWN] = BTRFS_FT_UNKNOWN,
334 [EXT2_FT_REG_FILE] = BTRFS_FT_REG_FILE,
335 [EXT2_FT_DIR] = BTRFS_FT_DIR,
336 [EXT2_FT_CHRDEV] = BTRFS_FT_CHRDEV,
337 [EXT2_FT_BLKDEV] = BTRFS_FT_BLKDEV,
338 [EXT2_FT_FIFO] = BTRFS_FT_FIFO,
339 [EXT2_FT_SOCK] = BTRFS_FT_SOCK,
340 [EXT2_FT_SYMLINK] = BTRFS_FT_SYMLINK,
343 static int dir_iterate_proc(ext2_ino_t dir, int entry,
344 struct ext2_dir_entry *dirent,
345 int offset, int blocksize,
346 char *buf,void *priv_data)
351 char dotdot[] = "..";
352 struct dir_iterate_data *idata = (struct dir_iterate_data *)priv_data;
355 name_len = dirent->name_len & 0xFF;
357 objectid = dirent->inode + INO_OFFSET;
358 if (!strncmp(dirent->name, dotdot, name_len)) {
360 BUG_ON(idata->parent != 0);
361 idata->parent = objectid;
365 if (dirent->inode < EXT2_GOOD_OLD_FIRST_INO)
368 file_type = dirent->name_len >> 8;
369 BUG_ON(file_type > EXT2_FT_SYMLINK);
371 ret = convert_insert_dirent(idata->trans, idata->root, dirent->name,
372 name_len, idata->objectid, objectid,
373 filetype_conversion_table[file_type],
374 idata->index_cnt, idata->inode);
376 idata->errcode = ret;
384 static int create_dir_entries(struct btrfs_trans_handle *trans,
385 struct btrfs_root *root, u64 objectid,
386 struct btrfs_inode_item *btrfs_inode,
387 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
391 struct dir_iterate_data data = {
394 .inode = btrfs_inode,
395 .objectid = objectid,
401 err = ext2fs_dir_iterate2(ext2_fs, ext2_ino, 0, NULL,
402 dir_iterate_proc, &data);
406 if (ret == 0 && data.parent == objectid) {
407 ret = btrfs_insert_inode_ref(trans, root, "..", 2,
408 objectid, objectid, 0);
412 fprintf(stderr, "ext2fs_dir_iterate2: %s\n", error_message(err));
416 static int read_disk_extent(struct btrfs_root *root, u64 bytenr,
417 u32 num_bytes, char *buffer)
420 struct btrfs_fs_devices *fs_devs = root->fs_info->fs_devices;
422 ret = pread(fs_devs->latest_bdev, buffer, num_bytes, bytenr);
423 if (ret != num_bytes)
432 static int csum_disk_extent(struct btrfs_trans_handle *trans,
433 struct btrfs_root *root,
434 u64 disk_bytenr, u64 num_bytes)
436 u32 blocksize = root->sectorsize;
441 buffer = malloc(blocksize);
444 for (offset = 0; offset < num_bytes; offset += blocksize) {
445 ret = read_disk_extent(root, disk_bytenr + offset,
449 ret = btrfs_csum_file_block(trans,
450 root->fs_info->csum_root,
451 disk_bytenr + num_bytes,
452 disk_bytenr + offset,
461 struct blk_iterate_data {
462 struct btrfs_trans_handle *trans;
463 struct btrfs_root *root;
464 struct btrfs_root *convert_root;
465 struct btrfs_inode_item *inode;
476 static void init_blk_iterate_data(struct blk_iterate_data *data,
477 struct btrfs_trans_handle *trans,
478 struct btrfs_root *root,
479 struct btrfs_inode_item *inode,
480 u64 objectid, int checksum)
482 struct btrfs_key key;
487 data->objectid = objectid;
488 data->first_block = 0;
489 data->disk_block = 0;
490 data->num_blocks = 0;
491 data->boundary = (u64)-1;
492 data->checksum = checksum;
495 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
496 key.type = BTRFS_ROOT_ITEM_KEY;
497 key.offset = (u64)-1;
498 data->convert_root = btrfs_read_fs_root(root->fs_info, &key);
499 /* Impossible as we just opened it before */
500 BUG_ON(!data->convert_root || IS_ERR(data->convert_root));
501 data->convert_ino = BTRFS_FIRST_FREE_OBJECTID + 1;
505 * Record a file extent in original filesystem into btrfs one.
506 * The special point is, old disk_block can point to a reserved range.
507 * So here, we don't use disk_block directly but search convert_root
508 * to get the real disk_bytenr.
510 static int record_file_blocks(struct blk_iterate_data *data,
511 u64 file_block, u64 disk_block, u64 num_blocks)
514 struct btrfs_root *root = data->root;
515 struct btrfs_root *convert_root = data->convert_root;
516 struct btrfs_path *path;
517 u64 file_pos = file_block * root->sectorsize;
518 u64 old_disk_bytenr = disk_block * root->sectorsize;
519 u64 num_bytes = num_blocks * root->sectorsize;
520 u64 cur_off = old_disk_bytenr;
522 /* Hole, pass it to record_file_extent directly */
523 if (old_disk_bytenr == 0)
524 return btrfs_record_file_extent(data->trans, root,
525 data->objectid, data->inode, file_pos, 0,
528 path = btrfs_alloc_path();
533 * Search real disk bytenr from convert root
535 while (cur_off < old_disk_bytenr + num_bytes) {
536 struct btrfs_key key;
537 struct btrfs_file_extent_item *fi;
538 struct extent_buffer *node;
540 u64 extent_disk_bytenr;
541 u64 extent_num_bytes;
542 u64 real_disk_bytenr;
545 key.objectid = data->convert_ino;
546 key.type = BTRFS_EXTENT_DATA_KEY;
547 key.offset = cur_off;
549 ret = btrfs_search_slot(NULL, convert_root, &key, path, 0, 0);
553 ret = btrfs_previous_item(convert_root, path,
555 BTRFS_EXTENT_DATA_KEY);
563 node = path->nodes[0];
564 slot = path->slots[0];
565 btrfs_item_key_to_cpu(node, &key, slot);
566 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY ||
567 key.objectid != data->convert_ino ||
568 key.offset > cur_off);
569 fi = btrfs_item_ptr(node, slot, struct btrfs_file_extent_item);
570 extent_disk_bytenr = btrfs_file_extent_disk_bytenr(node, fi);
571 extent_num_bytes = btrfs_file_extent_disk_num_bytes(node, fi);
572 BUG_ON(cur_off - key.offset >= extent_num_bytes);
573 btrfs_release_path(path);
575 real_disk_bytenr = cur_off - key.offset + extent_disk_bytenr;
576 cur_len = min(key.offset + extent_num_bytes,
577 old_disk_bytenr + num_bytes) - cur_off;
578 ret = btrfs_record_file_extent(data->trans, data->root,
579 data->objectid, data->inode, file_pos,
580 real_disk_bytenr, cur_len);
587 * No need to care about csum
588 * As every byte of old fs image is calculated for csum, no
589 * need to waste CPU cycles now.
592 btrfs_free_path(path);
596 static int block_iterate_proc(u64 disk_block, u64 file_block,
597 struct blk_iterate_data *idata)
602 struct btrfs_root *root = idata->root;
603 struct btrfs_block_group_cache *cache;
604 u64 bytenr = disk_block * root->sectorsize;
606 sb_region = intersect_with_sb(bytenr, root->sectorsize);
607 do_barrier = sb_region || disk_block >= idata->boundary;
608 if ((idata->num_blocks > 0 && do_barrier) ||
609 (file_block > idata->first_block + idata->num_blocks) ||
610 (disk_block != idata->disk_block + idata->num_blocks)) {
611 if (idata->num_blocks > 0) {
612 ret = record_file_blocks(idata, idata->first_block,
617 idata->first_block += idata->num_blocks;
618 idata->num_blocks = 0;
620 if (file_block > idata->first_block) {
621 ret = record_file_blocks(idata, idata->first_block,
622 0, file_block - idata->first_block);
628 bytenr += BTRFS_STRIPE_LEN - 1;
629 bytenr &= ~((u64)BTRFS_STRIPE_LEN - 1);
631 cache = btrfs_lookup_block_group(root->fs_info, bytenr);
633 bytenr = cache->key.objectid + cache->key.offset;
636 idata->first_block = file_block;
637 idata->disk_block = disk_block;
638 idata->boundary = bytenr / root->sectorsize;
645 static int __block_iterate_proc(ext2_filsys fs, blk_t *blocknr,
646 e2_blkcnt_t blockcnt, blk_t ref_block,
647 int ref_offset, void *priv_data)
650 struct blk_iterate_data *idata;
651 idata = (struct blk_iterate_data *)priv_data;
652 ret = block_iterate_proc(*blocknr, blockcnt, idata);
654 idata->errcode = ret;
661 * traverse file's data blocks, record these data blocks as file extents.
663 static int create_file_extents(struct btrfs_trans_handle *trans,
664 struct btrfs_root *root, u64 objectid,
665 struct btrfs_inode_item *btrfs_inode,
666 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
667 int datacsum, int packing)
673 u32 sectorsize = root->sectorsize;
674 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
675 struct blk_iterate_data data;
677 init_blk_iterate_data(&data, trans, root, btrfs_inode, objectid,
680 err = ext2fs_block_iterate2(ext2_fs, ext2_ino, BLOCK_FLAG_DATA_ONLY,
681 NULL, __block_iterate_proc, &data);
687 if (packing && data.first_block == 0 && data.num_blocks > 0 &&
688 inode_size <= BTRFS_MAX_INLINE_DATA_SIZE(root)) {
689 u64 num_bytes = data.num_blocks * sectorsize;
690 u64 disk_bytenr = data.disk_block * sectorsize;
693 buffer = malloc(num_bytes);
696 ret = read_disk_extent(root, disk_bytenr, num_bytes, buffer);
699 if (num_bytes > inode_size)
700 num_bytes = inode_size;
701 ret = btrfs_insert_inline_extent(trans, root, objectid,
702 0, buffer, num_bytes);
705 nbytes = btrfs_stack_inode_nbytes(btrfs_inode) + num_bytes;
706 btrfs_set_stack_inode_nbytes(btrfs_inode, nbytes);
707 } else if (data.num_blocks > 0) {
708 ret = record_file_blocks(&data, data.first_block,
709 data.disk_block, data.num_blocks);
713 data.first_block += data.num_blocks;
714 last_block = (inode_size + sectorsize - 1) / sectorsize;
715 if (last_block > data.first_block) {
716 ret = record_file_blocks(&data, data.first_block, 0,
717 last_block - data.first_block);
723 fprintf(stderr, "ext2fs_block_iterate2: %s\n", error_message(err));
727 static int create_symbol_link(struct btrfs_trans_handle *trans,
728 struct btrfs_root *root, u64 objectid,
729 struct btrfs_inode_item *btrfs_inode,
730 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
731 struct ext2_inode *ext2_inode)
735 u64 inode_size = btrfs_stack_inode_size(btrfs_inode);
736 if (ext2fs_inode_data_blocks(ext2_fs, ext2_inode)) {
737 btrfs_set_stack_inode_size(btrfs_inode, inode_size + 1);
738 ret = create_file_extents(trans, root, objectid, btrfs_inode,
739 ext2_fs, ext2_ino, 1, 1);
740 btrfs_set_stack_inode_size(btrfs_inode, inode_size);
744 pathname = (char *)&(ext2_inode->i_block[0]);
745 BUG_ON(pathname[inode_size] != 0);
746 ret = btrfs_insert_inline_extent(trans, root, objectid, 0,
747 pathname, inode_size + 1);
748 btrfs_set_stack_inode_nbytes(btrfs_inode, inode_size + 1);
753 * Following xattr/acl related codes are based on codes in
754 * fs/ext3/xattr.c and fs/ext3/acl.c
756 #define EXT2_XATTR_BHDR(ptr) ((struct ext2_ext_attr_header *)(ptr))
757 #define EXT2_XATTR_BFIRST(ptr) \
758 ((struct ext2_ext_attr_entry *)(EXT2_XATTR_BHDR(ptr) + 1))
759 #define EXT2_XATTR_IHDR(inode) \
760 ((struct ext2_ext_attr_header *) ((void *)(inode) + \
761 EXT2_GOOD_OLD_INODE_SIZE + (inode)->i_extra_isize))
762 #define EXT2_XATTR_IFIRST(inode) \
763 ((struct ext2_ext_attr_entry *) ((void *)EXT2_XATTR_IHDR(inode) + \
764 sizeof(EXT2_XATTR_IHDR(inode)->h_magic)))
766 static int ext2_xattr_check_names(struct ext2_ext_attr_entry *entry,
769 struct ext2_ext_attr_entry *next;
771 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
772 next = EXT2_EXT_ATTR_NEXT(entry);
773 if ((void *)next >= end)
780 static int ext2_xattr_check_block(const char *buf, size_t size)
783 struct ext2_ext_attr_header *header = EXT2_XATTR_BHDR(buf);
785 if (header->h_magic != EXT2_EXT_ATTR_MAGIC ||
786 header->h_blocks != 1)
788 error = ext2_xattr_check_names(EXT2_XATTR_BFIRST(buf), buf + size);
792 static int ext2_xattr_check_entry(struct ext2_ext_attr_entry *entry,
795 size_t value_size = entry->e_value_size;
797 if (entry->e_value_block != 0 || value_size > size ||
798 entry->e_value_offs + value_size > size)
803 #define EXT2_ACL_VERSION 0x0001
805 /* 23.2.5 acl_tag_t values */
807 #define ACL_UNDEFINED_TAG (0x00)
808 #define ACL_USER_OBJ (0x01)
809 #define ACL_USER (0x02)
810 #define ACL_GROUP_OBJ (0x04)
811 #define ACL_GROUP (0x08)
812 #define ACL_MASK (0x10)
813 #define ACL_OTHER (0x20)
815 /* 23.2.7 ACL qualifier constants */
817 #define ACL_UNDEFINED_ID ((id_t)-1)
828 } ext2_acl_entry_short;
834 static inline int ext2_acl_count(size_t size)
837 size -= sizeof(ext2_acl_header);
838 s = size - 4 * sizeof(ext2_acl_entry_short);
840 if (size % sizeof(ext2_acl_entry_short))
842 return size / sizeof(ext2_acl_entry_short);
844 if (s % sizeof(ext2_acl_entry))
846 return s / sizeof(ext2_acl_entry) + 4;
850 #define ACL_EA_VERSION 0x0002
860 acl_ea_entry a_entries[0];
863 static inline size_t acl_ea_size(int count)
865 return sizeof(acl_ea_header) + count * sizeof(acl_ea_entry);
868 static int ext2_acl_to_xattr(void *dst, const void *src,
869 size_t dst_size, size_t src_size)
872 const void *end = src + src_size;
873 acl_ea_header *ext_acl = (acl_ea_header *)dst;
874 acl_ea_entry *dst_entry = ext_acl->a_entries;
875 ext2_acl_entry *src_entry;
877 if (src_size < sizeof(ext2_acl_header))
879 if (((ext2_acl_header *)src)->a_version !=
880 cpu_to_le32(EXT2_ACL_VERSION))
882 src += sizeof(ext2_acl_header);
883 count = ext2_acl_count(src_size);
887 BUG_ON(dst_size < acl_ea_size(count));
888 ext_acl->a_version = cpu_to_le32(ACL_EA_VERSION);
889 for (i = 0; i < count; i++, dst_entry++) {
890 src_entry = (ext2_acl_entry *)src;
891 if (src + sizeof(ext2_acl_entry_short) > end)
893 dst_entry->e_tag = src_entry->e_tag;
894 dst_entry->e_perm = src_entry->e_perm;
895 switch (le16_to_cpu(src_entry->e_tag)) {
900 src += sizeof(ext2_acl_entry_short);
901 dst_entry->e_id = cpu_to_le32(ACL_UNDEFINED_ID);
905 src += sizeof(ext2_acl_entry);
908 dst_entry->e_id = src_entry->e_id;
921 static char *xattr_prefix_table[] = {
923 [2] = "system.posix_acl_access",
924 [3] = "system.posix_acl_default",
929 static int copy_single_xattr(struct btrfs_trans_handle *trans,
930 struct btrfs_root *root, u64 objectid,
931 struct ext2_ext_attr_entry *entry,
932 const void *data, u32 datalen)
937 void *databuf = NULL;
938 char namebuf[XATTR_NAME_MAX + 1];
940 name_index = entry->e_name_index;
941 if (name_index >= ARRAY_SIZE(xattr_prefix_table) ||
942 xattr_prefix_table[name_index] == NULL)
944 name_len = strlen(xattr_prefix_table[name_index]) +
946 if (name_len >= sizeof(namebuf))
949 if (name_index == 2 || name_index == 3) {
950 size_t bufsize = acl_ea_size(ext2_acl_count(datalen));
951 databuf = malloc(bufsize);
954 ret = ext2_acl_to_xattr(databuf, data, bufsize, datalen);
960 strncpy(namebuf, xattr_prefix_table[name_index], XATTR_NAME_MAX);
961 strncat(namebuf, EXT2_EXT_ATTR_NAME(entry), entry->e_name_len);
962 if (name_len + datalen > BTRFS_LEAF_DATA_SIZE(root) -
963 sizeof(struct btrfs_item) - sizeof(struct btrfs_dir_item)) {
964 fprintf(stderr, "skip large xattr on inode %Lu name %.*s\n",
965 objectid - INO_OFFSET, name_len, namebuf);
968 ret = btrfs_insert_xattr_item(trans, root, namebuf, name_len,
969 data, datalen, objectid);
975 static int copy_extended_attrs(struct btrfs_trans_handle *trans,
976 struct btrfs_root *root, u64 objectid,
977 struct btrfs_inode_item *btrfs_inode,
978 ext2_filsys ext2_fs, ext2_ino_t ext2_ino)
984 u32 block_size = ext2_fs->blocksize;
985 u32 inode_size = EXT2_INODE_SIZE(ext2_fs->super);
986 struct ext2_inode_large *ext2_inode;
987 struct ext2_ext_attr_entry *entry;
990 char inode_buf[EXT2_GOOD_OLD_INODE_SIZE];
992 if (inode_size <= EXT2_GOOD_OLD_INODE_SIZE) {
993 ext2_inode = (struct ext2_inode_large *)inode_buf;
995 ext2_inode = (struct ext2_inode_large *)malloc(inode_size);
999 err = ext2fs_read_inode_full(ext2_fs, ext2_ino, (void *)ext2_inode,
1002 fprintf(stderr, "ext2fs_read_inode_full: %s\n",
1003 error_message(err));
1008 if (ext2_ino > ext2_fs->super->s_first_ino &&
1009 inode_size > EXT2_GOOD_OLD_INODE_SIZE) {
1010 if (EXT2_GOOD_OLD_INODE_SIZE +
1011 ext2_inode->i_extra_isize > inode_size) {
1015 if (ext2_inode->i_extra_isize != 0 &&
1016 EXT2_XATTR_IHDR(ext2_inode)->h_magic ==
1017 EXT2_EXT_ATTR_MAGIC) {
1023 void *end = (void *)ext2_inode + inode_size;
1024 entry = EXT2_XATTR_IFIRST(ext2_inode);
1025 total = end - (void *)entry;
1026 ret = ext2_xattr_check_names(entry, end);
1029 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1030 ret = ext2_xattr_check_entry(entry, total);
1033 data = (void *)EXT2_XATTR_IFIRST(ext2_inode) +
1034 entry->e_value_offs;
1035 datalen = entry->e_value_size;
1036 ret = copy_single_xattr(trans, root, objectid,
1037 entry, data, datalen);
1040 entry = EXT2_EXT_ATTR_NEXT(entry);
1044 if (ext2_inode->i_file_acl == 0)
1047 buffer = malloc(block_size);
1052 err = ext2fs_read_ext_attr(ext2_fs, ext2_inode->i_file_acl, buffer);
1054 fprintf(stderr, "ext2fs_read_ext_attr: %s\n",
1055 error_message(err));
1059 ret = ext2_xattr_check_block(buffer, block_size);
1063 entry = EXT2_XATTR_BFIRST(buffer);
1064 while (!EXT2_EXT_IS_LAST_ENTRY(entry)) {
1065 ret = ext2_xattr_check_entry(entry, block_size);
1068 data = buffer + entry->e_value_offs;
1069 datalen = entry->e_value_size;
1070 ret = copy_single_xattr(trans, root, objectid,
1071 entry, data, datalen);
1074 entry = EXT2_EXT_ATTR_NEXT(entry);
1078 if ((void *)ext2_inode != inode_buf)
1082 #define MINORBITS 20
1083 #define MKDEV(ma, mi) (((ma) << MINORBITS) | (mi))
1085 static inline dev_t old_decode_dev(u16 val)
1087 return MKDEV((val >> 8) & 255, val & 255);
1090 static inline dev_t new_decode_dev(u32 dev)
1092 unsigned major = (dev & 0xfff00) >> 8;
1093 unsigned minor = (dev & 0xff) | ((dev >> 12) & 0xfff00);
1094 return MKDEV(major, minor);
1097 static int copy_inode_item(struct btrfs_inode_item *dst,
1098 struct ext2_inode *src, u32 blocksize)
1100 btrfs_set_stack_inode_generation(dst, 1);
1101 btrfs_set_stack_inode_sequence(dst, 0);
1102 btrfs_set_stack_inode_transid(dst, 1);
1103 btrfs_set_stack_inode_size(dst, src->i_size);
1104 btrfs_set_stack_inode_nbytes(dst, 0);
1105 btrfs_set_stack_inode_block_group(dst, 0);
1106 btrfs_set_stack_inode_nlink(dst, src->i_links_count);
1107 btrfs_set_stack_inode_uid(dst, src->i_uid | (src->i_uid_high << 16));
1108 btrfs_set_stack_inode_gid(dst, src->i_gid | (src->i_gid_high << 16));
1109 btrfs_set_stack_inode_mode(dst, src->i_mode);
1110 btrfs_set_stack_inode_rdev(dst, 0);
1111 btrfs_set_stack_inode_flags(dst, 0);
1112 btrfs_set_stack_timespec_sec(&dst->atime, src->i_atime);
1113 btrfs_set_stack_timespec_nsec(&dst->atime, 0);
1114 btrfs_set_stack_timespec_sec(&dst->ctime, src->i_ctime);
1115 btrfs_set_stack_timespec_nsec(&dst->ctime, 0);
1116 btrfs_set_stack_timespec_sec(&dst->mtime, src->i_mtime);
1117 btrfs_set_stack_timespec_nsec(&dst->mtime, 0);
1118 btrfs_set_stack_timespec_sec(&dst->otime, 0);
1119 btrfs_set_stack_timespec_nsec(&dst->otime, 0);
1121 if (S_ISDIR(src->i_mode)) {
1122 btrfs_set_stack_inode_size(dst, 0);
1123 btrfs_set_stack_inode_nlink(dst, 1);
1125 if (S_ISREG(src->i_mode)) {
1126 btrfs_set_stack_inode_size(dst, (u64)src->i_size_high << 32 |
1129 if (!S_ISREG(src->i_mode) && !S_ISDIR(src->i_mode) &&
1130 !S_ISLNK(src->i_mode)) {
1131 if (src->i_block[0]) {
1132 btrfs_set_stack_inode_rdev(dst,
1133 old_decode_dev(src->i_block[0]));
1135 btrfs_set_stack_inode_rdev(dst,
1136 new_decode_dev(src->i_block[1]));
1139 memset(&dst->reserved, 0, sizeof(dst->reserved));
1145 * copy a single inode. do all the required works, such as cloning
1146 * inode item, creating file extents and creating directory entries.
1148 static int copy_single_inode(struct btrfs_trans_handle *trans,
1149 struct btrfs_root *root, u64 objectid,
1150 ext2_filsys ext2_fs, ext2_ino_t ext2_ino,
1151 struct ext2_inode *ext2_inode,
1152 int datacsum, int packing, int noxattr)
1155 struct btrfs_inode_item btrfs_inode;
1157 if (ext2_inode->i_links_count == 0)
1160 copy_inode_item(&btrfs_inode, ext2_inode, ext2_fs->blocksize);
1161 if (!datacsum && S_ISREG(ext2_inode->i_mode)) {
1162 u32 flags = btrfs_stack_inode_flags(&btrfs_inode) |
1163 BTRFS_INODE_NODATASUM;
1164 btrfs_set_stack_inode_flags(&btrfs_inode, flags);
1167 switch (ext2_inode->i_mode & S_IFMT) {
1169 ret = create_file_extents(trans, root, objectid, &btrfs_inode,
1170 ext2_fs, ext2_ino, datacsum, packing);
1173 ret = create_dir_entries(trans, root, objectid, &btrfs_inode,
1177 ret = create_symbol_link(trans, root, objectid, &btrfs_inode,
1178 ext2_fs, ext2_ino, ext2_inode);
1188 ret = copy_extended_attrs(trans, root, objectid, &btrfs_inode,
1193 return btrfs_insert_inode(trans, root, objectid, &btrfs_inode);
1197 * scan ext2's inode bitmap and copy all used inodes.
1199 static int ext2_copy_inodes(struct btrfs_convert_context *cctx,
1200 struct btrfs_root *root,
1201 int datacsum, int packing, int noxattr, struct task_ctx *p)
1203 ext2_filsys ext2_fs = cctx->fs_data;
1206 ext2_inode_scan ext2_scan;
1207 struct ext2_inode ext2_inode;
1208 ext2_ino_t ext2_ino;
1210 struct btrfs_trans_handle *trans;
1212 trans = btrfs_start_transaction(root, 1);
1215 err = ext2fs_open_inode_scan(ext2_fs, 0, &ext2_scan);
1217 fprintf(stderr, "ext2fs_open_inode_scan: %s\n", error_message(err));
1220 while (!(err = ext2fs_get_next_inode(ext2_scan, &ext2_ino,
1222 /* no more inodes */
1225 /* skip special inode in ext2fs */
1226 if (ext2_ino < EXT2_GOOD_OLD_FIRST_INO &&
1227 ext2_ino != EXT2_ROOT_INO)
1229 objectid = ext2_ino + INO_OFFSET;
1230 ret = copy_single_inode(trans, root,
1231 objectid, ext2_fs, ext2_ino,
1232 &ext2_inode, datacsum, packing,
1234 p->cur_copy_inodes++;
1237 if (trans->blocks_used >= 4096) {
1238 ret = btrfs_commit_transaction(trans, root);
1240 trans = btrfs_start_transaction(root, 1);
1245 fprintf(stderr, "ext2fs_get_next_inode: %s\n", error_message(err));
1248 ret = btrfs_commit_transaction(trans, root);
1250 ext2fs_close_inode_scan(ext2_scan);
1255 static int create_image_file_range(struct btrfs_trans_handle *trans,
1256 struct btrfs_root *root,
1257 struct cache_tree *used,
1258 struct btrfs_inode_item *inode,
1259 u64 ino, u64 bytenr, u64 *ret_len,
1262 struct cache_extent *cache;
1263 struct btrfs_block_group_cache *bg_cache;
1269 BUG_ON(bytenr != round_down(bytenr, root->sectorsize));
1270 BUG_ON(len != round_down(len, root->sectorsize));
1271 len = min_t(u64, len, BTRFS_MAX_EXTENT_SIZE);
1274 * Skip sb ranges first
1275 * [0, 1M), [sb_offset(1), +64K), [sb_offset(2), +64K].
1277 * Or we will insert a hole into current image file, and later
1278 * migrate block will fail as there is already a file extent.
1280 if (bytenr < 1024 * 1024) {
1281 *ret_len = 1024 * 1024 - bytenr;
1284 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1285 u64 cur = btrfs_sb_offset(i);
1287 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
1288 *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
1292 for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1293 u64 cur = btrfs_sb_offset(i);
1297 * |----range-------|
1298 * May still need to go through file extent inserts
1300 if (bytenr < cur && bytenr + len >= cur) {
1301 len = min_t(u64, len, cur - bytenr);
1307 * Drop out, no need to insert anything
1309 if (bytenr >= cur && bytenr < cur + BTRFS_STRIPE_LEN) {
1310 *ret_len = cur + BTRFS_STRIPE_LEN - bytenr;
1315 cache = search_cache_extent(used, bytenr);
1317 if (cache->start <= bytenr) {
1319 * |///////Used///////|
1323 len = min_t(u64, len, cache->start + cache->size -
1325 disk_bytenr = bytenr;
1332 len = min(len, cache->start - bytenr);
1347 /* Check if the range is in a data block group */
1348 bg_cache = btrfs_lookup_block_group(root->fs_info, bytenr);
1351 if (!(bg_cache->flags & BTRFS_BLOCK_GROUP_DATA))
1354 /* The extent should never cross block group boundary */
1355 len = min_t(u64, len, bg_cache->key.objectid +
1356 bg_cache->key.offset - bytenr);
1359 BUG_ON(len != round_down(len, root->sectorsize));
1360 ret = btrfs_record_file_extent(trans, root, ino, inode, bytenr,
1366 ret = csum_disk_extent(trans, root, bytenr, len);
1373 * Relocate old fs data in one reserved ranges
1375 * Since all old fs data in reserved range is not covered by any chunk nor
1376 * data extent, we don't need to handle any reference but add new
1377 * extent/reference, which makes codes more clear
1379 static int migrate_one_reserved_range(struct btrfs_trans_handle *trans,
1380 struct btrfs_root *root,
1381 struct cache_tree *used,
1382 struct btrfs_inode_item *inode, int fd,
1383 u64 ino, u64 start, u64 len, int datacsum)
1385 u64 cur_off = start;
1387 struct cache_extent *cache;
1388 struct btrfs_key key;
1389 struct extent_buffer *eb;
1392 while (cur_off < start + len) {
1393 cache = lookup_cache_extent(used, cur_off, cur_len);
1396 cur_off = max(cache->start, cur_off);
1397 cur_len = min(cache->start + cache->size, start + len) -
1399 BUG_ON(cur_len < root->sectorsize);
1401 /* reserve extent for the data */
1402 ret = btrfs_reserve_extent(trans, root, cur_len, 0, 0, (u64)-1,
1407 eb = malloc(sizeof(*eb) + cur_len);
1413 ret = pread(fd, eb->data, cur_len, cur_off);
1414 if (ret < cur_len) {
1415 ret = (ret < 0 ? ret : -EIO);
1419 eb->start = key.objectid;
1420 eb->len = key.offset;
1422 /* Write the data */
1423 ret = write_and_map_eb(trans, root, eb);
1428 /* Now handle extent item and file extent things */
1429 ret = btrfs_record_file_extent(trans, root, ino, inode, cur_off,
1430 key.objectid, key.offset);
1433 /* Finally, insert csum items */
1435 ret = csum_disk_extent(trans, root, key.objectid,
1438 cur_off += key.offset;
1439 cur_len = start + len - cur_off;
1445 * Relocate the used ext2 data in reserved ranges
1447 * [btrfs_sb_offset(1), +BTRFS_STRIPE_LEN)
1448 * [btrfs_sb_offset(2), +BTRFS_STRIPE_LEN)
1450 static int migrate_reserved_ranges(struct btrfs_trans_handle *trans,
1451 struct btrfs_root *root,
1452 struct cache_tree *used,
1453 struct btrfs_inode_item *inode, int fd,
1454 u64 ino, u64 total_bytes, int datacsum)
1462 cur_len = 1024 * 1024;
1463 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1464 cur_off, cur_len, datacsum);
1468 /* second sb(fisrt sb is included in 0~1M) */
1469 cur_off = btrfs_sb_offset(1);
1470 cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
1471 if (cur_off > total_bytes)
1473 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1474 cur_off, cur_len, datacsum);
1479 cur_off = btrfs_sb_offset(2);
1480 cur_len = min(total_bytes, cur_off + BTRFS_STRIPE_LEN) - cur_off;
1481 if (cur_off > total_bytes)
1483 ret = migrate_one_reserved_range(trans, root, used, inode, fd, ino,
1484 cur_off, cur_len, datacsum);
1488 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
1492 * Create the fs image file of old filesystem.
1494 * This is completely fs independent as we have cctx->used, only
1495 * need to create file extents pointing to all the positions.
1497 static int create_image(struct btrfs_root *root,
1498 struct btrfs_mkfs_config *cfg,
1499 struct btrfs_convert_context *cctx, int fd,
1500 u64 size, char *name, int datacsum)
1502 struct btrfs_inode_item buf;
1503 struct btrfs_trans_handle *trans;
1504 struct btrfs_path *path = NULL;
1505 struct btrfs_key key;
1506 struct cache_extent *cache;
1507 struct cache_tree used_tmp;
1512 trans = btrfs_start_transaction(root, 1);
1516 cache_tree_init(&used_tmp);
1518 ret = btrfs_find_free_objectid(trans, root, BTRFS_FIRST_FREE_OBJECTID,
1522 ret = btrfs_new_inode(trans, root, ino, 0600 | S_IFREG);
1525 ret = btrfs_add_link(trans, root, ino, BTRFS_FIRST_FREE_OBJECTID, name,
1526 strlen(name), BTRFS_FT_REG_FILE, NULL, 1);
1530 path = btrfs_alloc_path();
1536 key.type = BTRFS_INODE_ITEM_KEY;
1539 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1541 ret = (ret > 0 ? -ENOENT : ret);
1544 read_extent_buffer(path->nodes[0], &buf,
1545 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1547 btrfs_release_path(path);
1550 * Create a new used space cache, which doesn't contain the reserved
1553 for (cache = first_cache_extent(&cctx->used); cache;
1554 cache = next_cache_extent(cache)) {
1555 ret = add_cache_extent(&used_tmp, cache->start, cache->size);
1559 ret = wipe_reserved_ranges(&used_tmp, 0, 0);
1564 * Start from 1M, as 0~1M is reserved, and create_image_file_range()
1565 * can't handle bytenr 0(will consider it as a hole)
1568 while (cur < size) {
1569 u64 len = size - cur;
1571 ret = create_image_file_range(trans, root, &used_tmp,
1572 &buf, ino, cur, &len, datacsum);
1577 /* Handle the reserved ranges */
1578 ret = migrate_reserved_ranges(trans, root, &cctx->used, &buf, fd, ino,
1579 cfg->num_bytes, datacsum);
1583 key.type = BTRFS_INODE_ITEM_KEY;
1585 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1587 ret = (ret > 0 ? -ENOENT : ret);
1590 btrfs_set_stack_inode_size(&buf, cfg->num_bytes);
1591 write_extent_buffer(path->nodes[0], &buf,
1592 btrfs_item_ptr_offset(path->nodes[0], path->slots[0]),
1595 free_extent_cache_tree(&used_tmp);
1596 btrfs_free_path(path);
1597 btrfs_commit_transaction(trans, root);
1601 static struct btrfs_root * link_subvol(struct btrfs_root *root,
1602 const char *base, u64 root_objectid)
1604 struct btrfs_trans_handle *trans;
1605 struct btrfs_fs_info *fs_info = root->fs_info;
1606 struct btrfs_root *tree_root = fs_info->tree_root;
1607 struct btrfs_root *new_root = NULL;
1608 struct btrfs_path *path;
1609 struct btrfs_inode_item *inode_item;
1610 struct extent_buffer *leaf;
1611 struct btrfs_key key;
1612 u64 dirid = btrfs_root_dirid(&root->root_item);
1614 char buf[BTRFS_NAME_LEN + 1]; /* for snprintf null */
1620 if (len == 0 || len > BTRFS_NAME_LEN)
1623 path = btrfs_alloc_path();
1626 key.objectid = dirid;
1627 key.type = BTRFS_DIR_INDEX_KEY;
1628 key.offset = (u64)-1;
1630 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1633 if (path->slots[0] > 0) {
1635 btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
1636 if (key.objectid == dirid && key.type == BTRFS_DIR_INDEX_KEY)
1637 index = key.offset + 1;
1639 btrfs_release_path(path);
1641 trans = btrfs_start_transaction(root, 1);
1644 key.objectid = dirid;
1646 key.type = BTRFS_INODE_ITEM_KEY;
1648 ret = btrfs_lookup_inode(trans, root, path, &key, 1);
1650 leaf = path->nodes[0];
1651 inode_item = btrfs_item_ptr(leaf, path->slots[0],
1652 struct btrfs_inode_item);
1654 key.objectid = root_objectid;
1655 key.offset = (u64)-1;
1656 key.type = BTRFS_ROOT_ITEM_KEY;
1658 memcpy(buf, base, len);
1659 for (i = 0; i < 1024; i++) {
1660 ret = btrfs_insert_dir_item(trans, root, buf, len,
1661 dirid, &key, BTRFS_FT_DIR, index);
1664 len = snprintf(buf, ARRAY_SIZE(buf), "%s%d", base, i);
1665 if (len < 1 || len > BTRFS_NAME_LEN) {
1673 btrfs_set_inode_size(leaf, inode_item, len * 2 +
1674 btrfs_inode_size(leaf, inode_item));
1675 btrfs_mark_buffer_dirty(leaf);
1676 btrfs_release_path(path);
1678 /* add the backref first */
1679 ret = btrfs_add_root_ref(trans, tree_root, root_objectid,
1680 BTRFS_ROOT_BACKREF_KEY,
1681 root->root_key.objectid,
1682 dirid, index, buf, len);
1685 /* now add the forward ref */
1686 ret = btrfs_add_root_ref(trans, tree_root, root->root_key.objectid,
1687 BTRFS_ROOT_REF_KEY, root_objectid,
1688 dirid, index, buf, len);
1690 ret = btrfs_commit_transaction(trans, root);
1693 new_root = btrfs_read_fs_root(fs_info, &key);
1694 if (IS_ERR(new_root))
1697 btrfs_free_path(path);
1701 static int create_subvol(struct btrfs_trans_handle *trans,
1702 struct btrfs_root *root, u64 root_objectid)
1704 struct extent_buffer *tmp;
1705 struct btrfs_root *new_root;
1706 struct btrfs_key key;
1707 struct btrfs_root_item root_item;
1710 ret = btrfs_copy_root(trans, root, root->node, &tmp,
1714 memcpy(&root_item, &root->root_item, sizeof(root_item));
1715 btrfs_set_root_bytenr(&root_item, tmp->start);
1716 btrfs_set_root_level(&root_item, btrfs_header_level(tmp));
1717 btrfs_set_root_generation(&root_item, trans->transid);
1718 free_extent_buffer(tmp);
1720 key.objectid = root_objectid;
1721 key.type = BTRFS_ROOT_ITEM_KEY;
1722 key.offset = trans->transid;
1723 ret = btrfs_insert_root(trans, root->fs_info->tree_root,
1726 key.offset = (u64)-1;
1727 new_root = btrfs_read_fs_root(root->fs_info, &key);
1728 BUG_ON(!new_root || IS_ERR(new_root));
1730 ret = btrfs_make_root_dir(trans, new_root, BTRFS_FIRST_FREE_OBJECTID);
1737 * New make_btrfs() has handle system and meta chunks quite well.
1738 * So only need to add remaining data chunks.
1740 static int make_convert_data_block_groups(struct btrfs_trans_handle *trans,
1741 struct btrfs_fs_info *fs_info,
1742 struct btrfs_mkfs_config *cfg,
1743 struct btrfs_convert_context *cctx)
1745 struct btrfs_root *extent_root = fs_info->extent_root;
1746 struct cache_tree *data_chunks = &cctx->data_chunks;
1747 struct cache_extent *cache;
1752 * Don't create data chunk over 10% of the convert device
1753 * And for single chunk, don't create chunk larger than 1G.
1755 max_chunk_size = cfg->num_bytes / 10;
1756 max_chunk_size = min((u64)(1024 * 1024 * 1024), max_chunk_size);
1757 max_chunk_size = round_down(max_chunk_size, extent_root->sectorsize);
1759 for (cache = first_cache_extent(data_chunks); cache;
1760 cache = next_cache_extent(cache)) {
1761 u64 cur = cache->start;
1763 while (cur < cache->start + cache->size) {
1765 u64 cur_backup = cur;
1767 len = min(max_chunk_size,
1768 cache->start + cache->size - cur);
1769 ret = btrfs_alloc_data_chunk(trans, extent_root,
1771 BTRFS_BLOCK_GROUP_DATA, 1);
1774 ret = btrfs_make_block_group(trans, extent_root, 0,
1775 BTRFS_BLOCK_GROUP_DATA,
1776 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1787 * Init the temp btrfs to a operational status.
1789 * It will fix the extent usage accounting(XXX: Do we really need?) and
1790 * insert needed data chunks, to ensure all old fs data extents are covered
1791 * by DATA chunks, preventing wrong chunks are allocated.
1793 * And also create convert image subvolume and relocation tree.
1794 * (XXX: Not need again?)
1795 * But the convert image subvolume is *NOT* linked to fs tree yet.
1797 static int init_btrfs(struct btrfs_mkfs_config *cfg, struct btrfs_root *root,
1798 struct btrfs_convert_context *cctx, int datacsum,
1799 int packing, int noxattr)
1801 struct btrfs_key location;
1802 struct btrfs_trans_handle *trans;
1803 struct btrfs_fs_info *fs_info = root->fs_info;
1807 * Don't alloc any metadata/system chunk, as we don't want
1808 * any meta/sys chunk allcated before all data chunks are inserted.
1809 * Or we screw up the chunk layout just like the old implement.
1811 fs_info->avoid_sys_chunk_alloc = 1;
1812 fs_info->avoid_meta_chunk_alloc = 1;
1813 trans = btrfs_start_transaction(root, 1);
1815 ret = btrfs_fix_block_accounting(trans, root);
1818 ret = make_convert_data_block_groups(trans, fs_info, cfg, cctx);
1821 ret = btrfs_make_root_dir(trans, fs_info->tree_root,
1822 BTRFS_ROOT_TREE_DIR_OBJECTID);
1825 memcpy(&location, &root->root_key, sizeof(location));
1826 location.offset = (u64)-1;
1827 ret = btrfs_insert_dir_item(trans, fs_info->tree_root, "default", 7,
1828 btrfs_super_root_dir(fs_info->super_copy),
1829 &location, BTRFS_FT_DIR, 0);
1832 ret = btrfs_insert_inode_ref(trans, fs_info->tree_root, "default", 7,
1834 btrfs_super_root_dir(fs_info->super_copy), 0);
1837 btrfs_set_root_dirid(&fs_info->fs_root->root_item,
1838 BTRFS_FIRST_FREE_OBJECTID);
1840 /* subvol for fs image file */
1841 ret = create_subvol(trans, root, CONV_IMAGE_SUBVOL_OBJECTID);
1844 /* subvol for data relocation tree */
1845 ret = create_subvol(trans, root, BTRFS_DATA_RELOC_TREE_OBJECTID);
1849 ret = btrfs_commit_transaction(trans, root);
1850 fs_info->avoid_sys_chunk_alloc = 0;
1851 fs_info->avoid_meta_chunk_alloc = 0;
1857 * Migrate super block to its default position and zero 0 ~ 16k
1859 static int migrate_super_block(int fd, u64 old_bytenr, u32 sectorsize)
1862 struct extent_buffer *buf;
1863 struct btrfs_super_block *super;
1867 BUG_ON(sectorsize < sizeof(*super));
1868 buf = malloc(sizeof(*buf) + sectorsize);
1872 buf->len = sectorsize;
1873 ret = pread(fd, buf->data, sectorsize, old_bytenr);
1874 if (ret != sectorsize)
1877 super = (struct btrfs_super_block *)buf->data;
1878 BUG_ON(btrfs_super_bytenr(super) != old_bytenr);
1879 btrfs_set_super_bytenr(super, BTRFS_SUPER_INFO_OFFSET);
1881 csum_tree_block_size(buf, BTRFS_CRC32_SIZE, 0);
1882 ret = pwrite(fd, buf->data, sectorsize, BTRFS_SUPER_INFO_OFFSET);
1883 if (ret != sectorsize)
1890 memset(buf->data, 0, sectorsize);
1891 for (bytenr = 0; bytenr < BTRFS_SUPER_INFO_OFFSET; ) {
1892 len = BTRFS_SUPER_INFO_OFFSET - bytenr;
1893 if (len > sectorsize)
1895 ret = pwrite(fd, buf->data, len, bytenr);
1897 fprintf(stderr, "unable to zero fill device\n");
1911 static int prepare_system_chunk_sb(struct btrfs_super_block *super)
1913 struct btrfs_chunk *chunk;
1914 struct btrfs_disk_key *key;
1915 u32 sectorsize = btrfs_super_sectorsize(super);
1917 key = (struct btrfs_disk_key *)(super->sys_chunk_array);
1918 chunk = (struct btrfs_chunk *)(super->sys_chunk_array +
1919 sizeof(struct btrfs_disk_key));
1921 btrfs_set_disk_key_objectid(key, BTRFS_FIRST_CHUNK_TREE_OBJECTID);
1922 btrfs_set_disk_key_type(key, BTRFS_CHUNK_ITEM_KEY);
1923 btrfs_set_disk_key_offset(key, 0);
1925 btrfs_set_stack_chunk_length(chunk, btrfs_super_total_bytes(super));
1926 btrfs_set_stack_chunk_owner(chunk, BTRFS_EXTENT_TREE_OBJECTID);
1927 btrfs_set_stack_chunk_stripe_len(chunk, BTRFS_STRIPE_LEN);
1928 btrfs_set_stack_chunk_type(chunk, BTRFS_BLOCK_GROUP_SYSTEM);
1929 btrfs_set_stack_chunk_io_align(chunk, sectorsize);
1930 btrfs_set_stack_chunk_io_width(chunk, sectorsize);
1931 btrfs_set_stack_chunk_sector_size(chunk, sectorsize);
1932 btrfs_set_stack_chunk_num_stripes(chunk, 1);
1933 btrfs_set_stack_chunk_sub_stripes(chunk, 0);
1934 chunk->stripe.devid = super->dev_item.devid;
1935 btrfs_set_stack_stripe_offset(&chunk->stripe, 0);
1936 memcpy(chunk->stripe.dev_uuid, super->dev_item.uuid, BTRFS_UUID_SIZE);
1937 btrfs_set_super_sys_array_size(super, sizeof(*key) + sizeof(*chunk));
1941 static const struct btrfs_convert_operations ext2_convert_ops = {
1943 .open_fs = ext2_open_fs,
1944 .read_used_space = ext2_read_used_space,
1945 .copy_inodes = ext2_copy_inodes,
1946 .close_fs = ext2_close_fs,
1949 static const struct btrfs_convert_operations *convert_operations[] = {
1953 static int convert_open_fs(const char *devname,
1954 struct btrfs_convert_context *cctx)
1958 memset(cctx, 0, sizeof(*cctx));
1960 for (i = 0; i < ARRAY_SIZE(convert_operations); i++) {
1961 int ret = convert_operations[i]->open_fs(cctx, devname);
1964 cctx->convert_ops = convert_operations[i];
1969 fprintf(stderr, "No file system found to convert.\n");
1974 * Helper for expand and merge extent_cache for wipe_one_reserved_range() to
1975 * handle wiping a range that exists in cache.
1977 static int _expand_extent_cache(struct cache_tree *tree,
1978 struct cache_extent *entry,
1979 u64 min_stripe_size, int backward)
1981 struct cache_extent *ce;
1984 if (entry->size >= min_stripe_size)
1986 diff = min_stripe_size - entry->size;
1989 ce = prev_cache_extent(entry);
1992 if (ce->start + ce->size >= entry->start - diff) {
1993 /* Directly merge with previous extent */
1994 ce->size = entry->start + entry->size - ce->start;
1995 remove_cache_extent(tree, entry);
2000 /* No overlap, normal extent */
2001 if (entry->start < diff) {
2002 error("cannot find space for data chunk layout");
2005 entry->start -= diff;
2006 entry->size += diff;
2009 ce = next_cache_extent(entry);
2012 if (entry->start + entry->size + diff >= ce->start) {
2013 /* Directly merge with next extent */
2014 entry->size = ce->start + ce->size - entry->start;
2015 remove_cache_extent(tree, ce);
2020 entry->size += diff;
2025 * Remove one reserve range from given cache tree
2026 * if min_stripe_size is non-zero, it will ensure for split case,
2027 * all its split cache extent is no smaller than @min_strip_size / 2.
2029 static int wipe_one_reserved_range(struct cache_tree *tree,
2030 u64 start, u64 len, u64 min_stripe_size,
2033 struct cache_extent *cache;
2036 BUG_ON(ensure_size && min_stripe_size == 0);
2038 * The logical here is simplified to handle special cases only
2039 * So we don't need to consider merge case for ensure_size
2041 BUG_ON(min_stripe_size && (min_stripe_size < len * 2 ||
2042 min_stripe_size / 2 < BTRFS_STRIPE_LEN));
2044 /* Also, wipe range should already be aligned */
2045 BUG_ON(start != round_down(start, BTRFS_STRIPE_LEN) ||
2046 start + len != round_up(start + len, BTRFS_STRIPE_LEN));
2048 min_stripe_size /= 2;
2050 cache = lookup_cache_extent(tree, start, len);
2054 if (start <= cache->start) {
2056 * |--------cache---------|
2059 BUG_ON(start + len <= cache->start);
2062 * The wipe size is smaller than min_stripe_size / 2,
2063 * so the result length should still meet min_stripe_size
2064 * And no need to do alignment
2066 cache->size -= (start + len - cache->start);
2067 if (cache->size == 0) {
2068 remove_cache_extent(tree, cache);
2073 BUG_ON(ensure_size && cache->size < min_stripe_size);
2075 cache->start = start + len;
2077 } else if (start > cache->start && start + len < cache->start +
2080 * |-------cache-----|
2083 u64 old_start = cache->start;
2084 u64 old_len = cache->size;
2085 u64 insert_start = start + len;
2088 cache->size = start - cache->start;
2089 /* Expand the leading half part if needed */
2090 if (ensure_size && cache->size < min_stripe_size) {
2091 ret = _expand_extent_cache(tree, cache,
2092 min_stripe_size, 1);
2097 /* And insert the new one */
2098 insert_len = old_start + old_len - start - len;
2099 ret = add_merge_cache_extent(tree, insert_start, insert_len);
2103 /* Expand the last half part if needed */
2104 if (ensure_size && insert_len < min_stripe_size) {
2105 cache = lookup_cache_extent(tree, insert_start,
2107 if (!cache || cache->start != insert_start ||
2108 cache->size != insert_len)
2110 ret = _expand_extent_cache(tree, cache,
2111 min_stripe_size, 0);
2119 * Wipe len should be small enough and no need to expand the
2122 cache->size = start - cache->start;
2123 BUG_ON(ensure_size && cache->size < min_stripe_size);
2128 * Remove reserved ranges from given cache_tree
2130 * It will remove the following ranges
2132 * 2) 2nd superblock, +64K (make sure chunks are 64K aligned)
2133 * 3) 3rd superblock, +64K
2135 * @min_stripe must be given for safety check
2136 * and if @ensure_size is given, it will ensure affected cache_extent will be
2137 * larger than min_stripe_size
2139 static int wipe_reserved_ranges(struct cache_tree *tree, u64 min_stripe_size,
2144 ret = wipe_one_reserved_range(tree, 0, 1024 * 1024, min_stripe_size,
2148 ret = wipe_one_reserved_range(tree, btrfs_sb_offset(1),
2149 BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
2152 ret = wipe_one_reserved_range(tree, btrfs_sb_offset(2),
2153 BTRFS_STRIPE_LEN, min_stripe_size, ensure_size);
2157 static int calculate_available_space(struct btrfs_convert_context *cctx)
2159 struct cache_tree *used = &cctx->used;
2160 struct cache_tree *data_chunks = &cctx->data_chunks;
2161 struct cache_tree *free = &cctx->free;
2162 struct cache_extent *cache;
2165 * Twice the minimal chunk size, to allow later wipe_reserved_ranges()
2166 * works without need to consider overlap
2168 u64 min_stripe_size = 2 * 16 * 1024 * 1024;
2171 /* Calculate data_chunks */
2172 for (cache = first_cache_extent(used); cache;
2173 cache = next_cache_extent(cache)) {
2176 if (cache->start + cache->size < cur_off)
2178 if (cache->start > cur_off + min_stripe_size)
2179 cur_off = cache->start;
2180 cur_len = max(cache->start + cache->size - cur_off,
2182 ret = add_merge_cache_extent(data_chunks, cur_off, cur_len);
2188 * remove reserved ranges, so we won't ever bother relocating an old
2189 * filesystem extent to other place.
2191 ret = wipe_reserved_ranges(data_chunks, min_stripe_size, 1);
2197 * Calculate free space
2198 * Always round up the start bytenr, to avoid metadata extent corss
2199 * stripe boundary, as later mkfs_convert() won't have all the extent
2202 for (cache = first_cache_extent(data_chunks); cache;
2203 cache = next_cache_extent(cache)) {
2204 if (cache->start < cur_off)
2206 if (cache->start > cur_off) {
2210 len = cache->start - round_up(cur_off,
2212 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2214 ret = add_merge_cache_extent(free, insert_start, len);
2218 cur_off = cache->start + cache->size;
2220 /* Don't forget the last range */
2221 if (cctx->total_bytes > cur_off) {
2222 u64 len = cctx->total_bytes - cur_off;
2225 insert_start = round_up(cur_off, BTRFS_STRIPE_LEN);
2227 ret = add_merge_cache_extent(free, insert_start, len);
2232 /* Remove reserved bytes */
2233 ret = wipe_reserved_ranges(free, min_stripe_size, 0);
2238 * Read used space, and since we have the used space,
2239 * calcuate data_chunks and free for later mkfs
2241 static int convert_read_used_space(struct btrfs_convert_context *cctx)
2245 ret = cctx->convert_ops->read_used_space(cctx);
2249 ret = calculate_available_space(cctx);
2253 static int do_convert(const char *devname, int datacsum, int packing,
2254 int noxattr, u32 nodesize, int copylabel, const char *fslabel,
2255 int progress, u64 features)
2262 struct btrfs_root *root;
2263 struct btrfs_root *image_root;
2264 struct btrfs_convert_context cctx;
2265 struct btrfs_key key;
2266 char *subvol_name = NULL;
2267 struct task_ctx ctx;
2268 char features_buf[64];
2269 struct btrfs_mkfs_config mkfs_cfg;
2271 init_convert_context(&cctx);
2272 ret = convert_open_fs(devname, &cctx);
2275 ret = convert_read_used_space(&cctx);
2279 blocksize = cctx.blocksize;
2280 total_bytes = (u64)blocksize * (u64)cctx.block_count;
2281 if (blocksize < 4096) {
2282 fprintf(stderr, "block size is too small\n");
2285 if (btrfs_check_nodesize(nodesize, blocksize, features))
2287 fd = open(devname, O_RDWR);
2289 fprintf(stderr, "unable to open %s\n", devname);
2292 btrfs_parse_features_to_string(features_buf, features);
2293 if (features == BTRFS_MKFS_DEFAULT_FEATURES)
2294 strcat(features_buf, " (default)");
2296 printf("create btrfs filesystem:\n");
2297 printf("\tblocksize: %u\n", blocksize);
2298 printf("\tnodesize: %u\n", nodesize);
2299 printf("\tfeatures: %s\n", features_buf);
2301 mkfs_cfg.label = cctx.volume_name;
2302 mkfs_cfg.num_bytes = total_bytes;
2303 mkfs_cfg.nodesize = nodesize;
2304 mkfs_cfg.sectorsize = blocksize;
2305 mkfs_cfg.stripesize = blocksize;
2306 mkfs_cfg.features = features;
2307 /* New convert need these space */
2308 mkfs_cfg.fs_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2309 mkfs_cfg.chunk_uuid = malloc(BTRFS_UUID_UNPARSED_SIZE);
2310 *(mkfs_cfg.fs_uuid) = '\0';
2311 *(mkfs_cfg.chunk_uuid) = '\0';
2313 ret = make_btrfs(fd, &mkfs_cfg, &cctx);
2315 fprintf(stderr, "unable to create initial ctree: %s\n",
2320 root = open_ctree_fd(fd, devname, mkfs_cfg.super_bytenr,
2323 fprintf(stderr, "unable to open ctree\n");
2326 ret = init_btrfs(&mkfs_cfg, root, &cctx, datacsum, packing, noxattr);
2328 fprintf(stderr, "unable to setup the root tree\n");
2332 printf("creating %s image file.\n", cctx.convert_ops->name);
2333 ret = asprintf(&subvol_name, "%s_saved", cctx.convert_ops->name);
2335 fprintf(stderr, "error allocating subvolume name: %s_saved\n",
2336 cctx.convert_ops->name);
2339 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2340 key.offset = (u64)-1;
2341 key.type = BTRFS_ROOT_ITEM_KEY;
2342 image_root = btrfs_read_fs_root(root->fs_info, &key);
2344 fprintf(stderr, "unable to create subvol\n");
2347 ret = create_image(image_root, &mkfs_cfg, &cctx, fd,
2348 mkfs_cfg.num_bytes, "image", datacsum);
2350 fprintf(stderr, "error during create_image %d\n", ret);
2354 printf("creating btrfs metadata.\n");
2355 ctx.max_copy_inodes = (cctx.inodes_count - cctx.free_inodes_count);
2356 ctx.cur_copy_inodes = 0;
2359 ctx.info = task_init(print_copied_inodes, after_copied_inodes,
2361 task_start(ctx.info);
2363 ret = copy_inodes(&cctx, root, datacsum, packing, noxattr, &ctx);
2365 fprintf(stderr, "error during copy_inodes %d\n", ret);
2369 task_stop(ctx.info);
2370 task_deinit(ctx.info);
2373 image_root = link_subvol(root, subvol_name, CONV_IMAGE_SUBVOL_OBJECTID);
2377 memset(root->fs_info->super_copy->label, 0, BTRFS_LABEL_SIZE);
2378 if (copylabel == 1) {
2379 __strncpy_null(root->fs_info->super_copy->label,
2380 cctx.volume_name, BTRFS_LABEL_SIZE - 1);
2381 fprintf(stderr, "copy label '%s'\n",
2382 root->fs_info->super_copy->label);
2383 } else if (copylabel == -1) {
2384 strcpy(root->fs_info->super_copy->label, fslabel);
2385 fprintf(stderr, "set label to '%s'\n", fslabel);
2388 ret = close_ctree(root);
2390 fprintf(stderr, "error during close_ctree %d\n", ret);
2393 convert_close_fs(&cctx);
2394 clean_convert_context(&cctx);
2397 * If this step succeed, we get a mountable btrfs. Otherwise
2398 * the source fs is left unchanged.
2400 ret = migrate_super_block(fd, mkfs_cfg.super_bytenr, blocksize);
2402 fprintf(stderr, "unable to migrate super block\n");
2407 root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2409 fprintf(stderr, "unable to open ctree\n");
2414 printf("conversion complete.\n");
2417 clean_convert_context(&cctx);
2422 "WARNING: an error occurred during chunk mapping fixup, filesystem mountable but not finalized\n");
2424 fprintf(stderr, "conversion aborted\n");
2429 * Check if a non 1:1 mapped chunk can be rolled back.
2430 * For new convert, it's OK while for old convert it's not.
2432 static int may_rollback_chunk(struct btrfs_fs_info *fs_info, u64 bytenr)
2434 struct btrfs_block_group_cache *bg;
2435 struct btrfs_key key;
2436 struct btrfs_path path;
2437 struct btrfs_root *extent_root = fs_info->extent_root;
2442 bg = btrfs_lookup_first_block_group(fs_info, bytenr);
2445 bg_start = bg->key.objectid;
2446 bg_end = bg->key.objectid + bg->key.offset;
2448 key.objectid = bg_end;
2449 key.type = BTRFS_METADATA_ITEM_KEY;
2451 btrfs_init_path(&path);
2453 ret = btrfs_search_slot(NULL, extent_root, &key, &path, 0, 0);
2458 struct btrfs_extent_item *ei;
2460 ret = btrfs_previous_extent_item(extent_root, &path, bg_start);
2468 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2469 if (key.type == BTRFS_METADATA_ITEM_KEY)
2471 /* Now it's EXTENT_ITEM_KEY only */
2472 ei = btrfs_item_ptr(path.nodes[0], path.slots[0],
2473 struct btrfs_extent_item);
2475 * Found data extent, means this is old convert must follow 1:1
2478 if (btrfs_extent_flags(path.nodes[0], ei)
2479 & BTRFS_EXTENT_FLAG_DATA) {
2484 btrfs_release_path(&path);
2488 static int may_rollback(struct btrfs_root *root)
2490 struct btrfs_fs_info *info = root->fs_info;
2491 struct btrfs_multi_bio *multi = NULL;
2499 if (btrfs_super_num_devices(info->super_copy) != 1)
2502 bytenr = BTRFS_SUPER_INFO_OFFSET;
2503 total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2506 ret = btrfs_map_block(&info->mapping_tree, WRITE, bytenr,
2507 &length, &multi, 0, NULL);
2509 if (ret == -ENOENT) {
2510 /* removed block group at the tail */
2511 if (length == (u64)-1)
2514 /* removed block group in the middle */
2520 num_stripes = multi->num_stripes;
2521 physical = multi->stripes[0].physical;
2524 if (num_stripes != 1) {
2525 error("num stripes for bytenr %llu is not 1", bytenr);
2530 * Extra check for new convert, as metadata chunk from new
2531 * convert is much more free than old convert, it doesn't need
2532 * to do 1:1 mapping.
2534 if (physical != bytenr) {
2536 * Check if it's a metadata chunk and has only metadata
2539 ret = may_rollback_chunk(info, bytenr);
2545 if (bytenr >= total_bytes)
2553 static int do_rollback(const char *devname)
2558 struct btrfs_root *root;
2559 struct btrfs_root *image_root;
2560 struct btrfs_root *chunk_root;
2561 struct btrfs_dir_item *dir;
2562 struct btrfs_inode_item *inode;
2563 struct btrfs_file_extent_item *fi;
2564 struct btrfs_trans_handle *trans;
2565 struct extent_buffer *leaf;
2566 struct btrfs_block_group_cache *cache1;
2567 struct btrfs_block_group_cache *cache2;
2568 struct btrfs_key key;
2569 struct btrfs_path path;
2570 struct extent_io_tree io_tree;
2585 extent_io_tree_init(&io_tree);
2587 fd = open(devname, O_RDWR);
2589 fprintf(stderr, "unable to open %s\n", devname);
2592 root = open_ctree_fd(fd, devname, 0, OPEN_CTREE_WRITES);
2594 fprintf(stderr, "unable to open ctree\n");
2597 ret = may_rollback(root);
2599 fprintf(stderr, "unable to do rollback\n");
2603 sectorsize = root->sectorsize;
2604 buf = malloc(sectorsize);
2606 fprintf(stderr, "unable to allocate memory\n");
2610 btrfs_init_path(&path);
2612 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2613 key.type = BTRFS_ROOT_BACKREF_KEY;
2614 key.offset = BTRFS_FS_TREE_OBJECTID;
2615 ret = btrfs_search_slot(NULL, root->fs_info->tree_root, &key, &path, 0,
2617 btrfs_release_path(&path);
2620 "ERROR: unable to convert ext2 image subvolume, is it deleted?\n");
2622 } else if (ret < 0) {
2624 "ERROR: unable to open ext2_saved, id=%llu: %s\n",
2625 (unsigned long long)key.objectid, strerror(-ret));
2629 key.objectid = CONV_IMAGE_SUBVOL_OBJECTID;
2630 key.type = BTRFS_ROOT_ITEM_KEY;
2631 key.offset = (u64)-1;
2632 image_root = btrfs_read_fs_root(root->fs_info, &key);
2633 if (!image_root || IS_ERR(image_root)) {
2634 fprintf(stderr, "unable to open subvol %llu\n",
2635 (unsigned long long)key.objectid);
2640 root_dir = btrfs_root_dirid(&root->root_item);
2641 dir = btrfs_lookup_dir_item(NULL, image_root, &path,
2642 root_dir, name, strlen(name), 0);
2643 if (!dir || IS_ERR(dir)) {
2644 fprintf(stderr, "unable to find file %s\n", name);
2647 leaf = path.nodes[0];
2648 btrfs_dir_item_key_to_cpu(leaf, dir, &key);
2649 btrfs_release_path(&path);
2651 objectid = key.objectid;
2653 ret = btrfs_lookup_inode(NULL, image_root, &path, &key, 0);
2655 fprintf(stderr, "unable to find inode item\n");
2658 leaf = path.nodes[0];
2659 inode = btrfs_item_ptr(leaf, path.slots[0], struct btrfs_inode_item);
2660 total_bytes = btrfs_inode_size(leaf, inode);
2661 btrfs_release_path(&path);
2663 key.objectid = objectid;
2665 btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
2666 ret = btrfs_search_slot(NULL, image_root, &key, &path, 0, 0);
2668 fprintf(stderr, "unable to find first file extent\n");
2669 btrfs_release_path(&path);
2673 /* build mapping tree for the relocated blocks */
2674 for (offset = 0; offset < total_bytes; ) {
2675 leaf = path.nodes[0];
2676 if (path.slots[0] >= btrfs_header_nritems(leaf)) {
2677 ret = btrfs_next_leaf(root, &path);
2683 btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
2684 if (key.objectid != objectid || key.offset != offset ||
2685 btrfs_key_type(&key) != BTRFS_EXTENT_DATA_KEY)
2688 fi = btrfs_item_ptr(leaf, path.slots[0],
2689 struct btrfs_file_extent_item);
2690 if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG)
2692 if (btrfs_file_extent_compression(leaf, fi) ||
2693 btrfs_file_extent_encryption(leaf, fi) ||
2694 btrfs_file_extent_other_encoding(leaf, fi))
2697 bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
2698 /* skip holes and direct mapped extents */
2699 if (bytenr == 0 || bytenr == offset)
2702 bytenr += btrfs_file_extent_offset(leaf, fi);
2703 num_bytes = btrfs_file_extent_num_bytes(leaf, fi);
2705 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2706 cache2 = btrfs_lookup_block_group(root->fs_info,
2707 offset + num_bytes - 1);
2709 * Here we must take consideration of old and new convert
2711 * For old convert case, sign, there is no consist chunk type
2712 * that will cover the extent. META/DATA/SYS are all possible.
2713 * Just ensure relocate one is in SYS chunk.
2714 * For new convert case, they are all covered by DATA chunk.
2716 * So, there is not valid chunk type check for it now.
2718 if (cache1 != cache2)
2721 set_extent_bits(&io_tree, offset, offset + num_bytes - 1,
2722 EXTENT_LOCKED, GFP_NOFS);
2723 set_state_private(&io_tree, offset, bytenr);
2725 offset += btrfs_file_extent_num_bytes(leaf, fi);
2728 btrfs_release_path(&path);
2730 if (offset < total_bytes) {
2731 fprintf(stderr, "unable to build extent mapping\n");
2732 fprintf(stderr, "converted filesystem after balance is unable to rollback\n");
2736 first_free = BTRFS_SUPER_INFO_OFFSET + 2 * sectorsize - 1;
2737 first_free &= ~((u64)sectorsize - 1);
2738 /* backup for extent #0 should exist */
2739 if(!test_range_bit(&io_tree, 0, first_free - 1, EXTENT_LOCKED, 1)) {
2740 fprintf(stderr, "no backup for the first extent\n");
2743 /* force no allocation from system block group */
2744 root->fs_info->system_allocs = -1;
2745 trans = btrfs_start_transaction(root, 1);
2748 * recow the whole chunk tree, this will remove all chunk tree blocks
2749 * from system block group
2751 chunk_root = root->fs_info->chunk_root;
2752 memset(&key, 0, sizeof(key));
2754 ret = btrfs_search_slot(trans, chunk_root, &key, &path, 0, 1);
2758 ret = btrfs_next_leaf(chunk_root, &path);
2762 btrfs_item_key_to_cpu(path.nodes[0], &key, path.slots[0]);
2763 btrfs_release_path(&path);
2765 btrfs_release_path(&path);
2770 cache1 = btrfs_lookup_block_group(root->fs_info, offset);
2774 if (cache1->flags & BTRFS_BLOCK_GROUP_SYSTEM)
2775 num_bytes += btrfs_block_group_used(&cache1->item);
2777 offset = cache1->key.objectid + cache1->key.offset;
2779 /* only extent #0 left in system block group? */
2780 if (num_bytes > first_free) {
2781 fprintf(stderr, "unable to empty system block group\n");
2784 /* create a system chunk that maps the whole device */
2785 ret = prepare_system_chunk_sb(root->fs_info->super_copy);
2787 fprintf(stderr, "unable to update system chunk\n");
2791 ret = btrfs_commit_transaction(trans, root);
2794 ret = close_ctree(root);
2796 fprintf(stderr, "error during close_ctree %d\n", ret);
2800 /* zero btrfs super block mirrors */
2801 memset(buf, 0, sectorsize);
2802 for (i = 1 ; i < BTRFS_SUPER_MIRROR_MAX; i++) {
2803 bytenr = btrfs_sb_offset(i);
2804 if (bytenr >= total_bytes)
2806 ret = pwrite(fd, buf, sectorsize, bytenr);
2807 if (ret != sectorsize) {
2809 "error during zeroing superblock %d: %d\n",
2815 sb_bytenr = (u64)-1;
2816 /* copy all relocated blocks back */
2818 ret = find_first_extent_bit(&io_tree, 0, &start, &end,
2823 ret = get_state_private(&io_tree, start, &bytenr);
2826 clear_extent_bits(&io_tree, start, end, EXTENT_LOCKED,
2829 while (start <= end) {
2830 if (start == BTRFS_SUPER_INFO_OFFSET) {
2834 ret = pread(fd, buf, sectorsize, bytenr);
2836 fprintf(stderr, "error during pread %d\n", ret);
2839 BUG_ON(ret != sectorsize);
2840 ret = pwrite(fd, buf, sectorsize, start);
2842 fprintf(stderr, "error during pwrite %d\n", ret);
2845 BUG_ON(ret != sectorsize);
2847 start += sectorsize;
2848 bytenr += sectorsize;
2854 fprintf(stderr, "error during fsync %d\n", ret);
2858 * finally, overwrite btrfs super block.
2860 ret = pread(fd, buf, sectorsize, sb_bytenr);
2862 fprintf(stderr, "error during pread %d\n", ret);
2865 BUG_ON(ret != sectorsize);
2866 ret = pwrite(fd, buf, sectorsize, BTRFS_SUPER_INFO_OFFSET);
2868 fprintf(stderr, "error during pwrite %d\n", ret);
2871 BUG_ON(ret != sectorsize);
2874 fprintf(stderr, "error during fsync %d\n", ret);
2880 extent_io_tree_cleanup(&io_tree);
2881 printf("rollback complete.\n");
2888 fprintf(stderr, "rollback aborted.\n");
2892 static void print_usage(void)
2894 printf("usage: btrfs-convert [options] device\n");
2895 printf("options:\n");
2896 printf("\t-d|--no-datasum disable data checksum, sets NODATASUM\n");
2897 printf("\t-i|--no-xattr ignore xattrs and ACLs\n");
2898 printf("\t-n|--no-inline disable inlining of small files to metadata\n");
2899 printf("\t-N|--nodesize SIZE set filesystem metadata nodesize\n");
2900 printf("\t-r|--rollback roll back to the original filesystem\n");
2901 printf("\t-l|--label LABEL set filesystem label\n");
2902 printf("\t-L|--copy-label use label from converted filesystem\n");
2903 printf("\t-p|--progress show converting progress (default)\n");
2904 printf("\t-O|--features LIST comma separated list of filesystem features\n");
2905 printf("\t--no-progress show only overview, not the detailed progress\n");
2908 int main(int argc, char *argv[])
2914 u32 nodesize = max_t(u32, sysconf(_SC_PAGESIZE),
2915 BTRFS_MKFS_DEFAULT_NODE_SIZE);
2918 int usage_error = 0;
2921 char fslabel[BTRFS_LABEL_SIZE];
2922 u64 features = BTRFS_MKFS_DEFAULT_FEATURES;
2925 enum { GETOPT_VAL_NO_PROGRESS = 256 };
2926 static const struct option long_options[] = {
2927 { "no-progress", no_argument, NULL,
2928 GETOPT_VAL_NO_PROGRESS },
2929 { "no-datasum", no_argument, NULL, 'd' },
2930 { "no-inline", no_argument, NULL, 'n' },
2931 { "no-xattr", no_argument, NULL, 'i' },
2932 { "rollback", no_argument, NULL, 'r' },
2933 { "features", required_argument, NULL, 'O' },
2934 { "progress", no_argument, NULL, 'p' },
2935 { "label", required_argument, NULL, 'l' },
2936 { "copy-label", no_argument, NULL, 'L' },
2937 { "nodesize", required_argument, NULL, 'N' },
2938 { "help", no_argument, NULL, GETOPT_VAL_HELP},
2939 { NULL, 0, NULL, 0 }
2941 int c = getopt_long(argc, argv, "dinN:rl:LpO:", long_options, NULL);
2956 nodesize = parse_size(optarg);
2963 if (strlen(optarg) >= BTRFS_LABEL_SIZE) {
2965 "WARNING: label too long, trimmed to %d bytes\n",
2966 BTRFS_LABEL_SIZE - 1);
2968 __strncpy_null(fslabel, optarg, BTRFS_LABEL_SIZE - 1);
2977 char *orig = strdup(optarg);
2980 tmp = btrfs_parse_fs_features(tmp, &features);
2983 "Unrecognized filesystem feature '%s'\n",
2989 if (features & BTRFS_FEATURE_LIST_ALL) {
2990 btrfs_list_all_fs_features(
2991 ~BTRFS_CONVERT_ALLOWED_FEATURES);
2994 if (features & ~BTRFS_CONVERT_ALLOWED_FEATURES) {
2997 btrfs_parse_features_to_string(buf,
2998 features & ~BTRFS_CONVERT_ALLOWED_FEATURES);
3000 "ERROR: features not allowed for convert: %s\n",
3007 case GETOPT_VAL_NO_PROGRESS:
3010 case GETOPT_VAL_HELP:
3013 return c != GETOPT_VAL_HELP;
3017 if (check_argc_exact(argc - optind, 1)) {
3022 if (rollback && (!datacsum || noxattr || !packing)) {
3024 "Usage error: -d, -i, -n options do not apply to rollback\n");
3033 file = argv[optind];
3034 ret = check_mounted(file);
3036 fprintf(stderr, "Could not check mount status: %s\n",
3040 fprintf(stderr, "%s is mounted\n", file);
3045 ret = do_rollback(file);
3047 ret = do_convert(file, datacsum, packing, noxattr, nodesize,
3048 copylabel, fslabel, progress, features);