Add --with-convert[=VALUE] option to configure. Accepts ext2, auto, yes,
or no, but will be extended to more in the future. The configure-time
defines are not used in the code, ext2 is built-in unconditionally.
Signed-off-by: David Sterba <dsterba@suse.com>
The filesystem existence on a device is manifested by the signature,
during the mkfs process we write it first and then create other
structures. Such filesystem is not valid and should not be registered
during device scan nor listed among devices from blkid.
This patch will introduce two staged creation. In the first phase, the
signature is wrong, but recognized as a partially created filesystem (by
open or scan helpers). Once we successfully create and write everything,
we fixup the signature. At this point automated scanning should find
a valid filesystem on all devices.
We can also rely on the partially created filesystem to do better error
handling during creation. We can just bail out and do not need to clean
up.
The partial signature is '!BHRfS_M', can be shown by
btrfs inspect-internal dump-super -F image
Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>
The new convert treats the convert image as a normal file, without any
special flags and permissions.
This is different from original code:
1) Permission changed from 0400 to 0600
2) Inode lacks READONLY flag
This makes we can read-write mount the ext2 image and cause rollback
failure.
Follow old code behavior, use 0400 permission and add back READONLY
flag to fix it.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When copying inode, if there is a file referring part of a hole range,
convert will fail.
The problem is, when calculating real extent bytenr, it doesn't check if
the original extent is a hole.
In case the orinal extent is a hole, we still calculate bytenr using
file_pos - found_extent_file_pos, causing non-zero value, and later
btrfs_record_file_extent() detects that we are pointing to non-exist
extent and aborts convert.
Fix it by checking the disk_bytenr before calculating real disk bytenr.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
New convert doesn't insert holes for superblock migration range.
Unlike the old design, which only relocates 4K (superblock size) to
other places.
In the new design, to make sure convert can handle different page sizes
and align chunks bytenr, we relocate the whole 64K range.
And if there is only a 4K used block inside 64K superblock migration
range, it will make converted the fs have discontiguous file extents.
This patch will fix it by inserting needed holes to avoid such
discontinuous error.
Reported-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
New convert has several bugs with backup superblock migration
1) Backup superblocks are not migrated due to incorrect condition
Two wrong checks cause backup superblocks not to be migrated at all
2) Converted ext* image doesn't keep hole for backup superblocks
Since we are creating file extents according to tmp_used, which has
wiped out backup superblock ranges.
In that case, later superblock migration will fail, since migration
will insert file extent range into ext* image.
Fix above bugs will make convert on ext2 image filled about 100M data
successful.
Reported-by: Satoru Takeuchi <takeuchi_satoru@jp.fujitsu.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
When a ext2 fs filled with a 57M file, it's possible that convert fails
with assert in add_merge_cache_extent().
The problem is that the ext2 used space just takes some of the second
superblock.
And due to a bug in reserving superblock space, it corrupted used space
tree and cause assert.
Fix in by doing better used space merging for case where superblock
range is inside the ext2 used space.
Reported-by: Satoru Takeuchi <takeuchi_satoru@jp.fujitsu.com>
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
For new btrfs-convert, it's less restrict for metadata chunk allocation.
While the may_rollback() function is still following the restrict 1:1
mapping check for all chunks, it will not allow some new convert image
to be rolled back.
Add new per extent check for new convert to allow it to be rolled back.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Cleanup all the old btrfs-convert facilities, including:
1) btrfs_convert_operations->alloc/free/test_extents*
No need to do non-standard extent allocation.
After init_btrfs() everything can be done by normal routine.
Now only 4 functions are needed in btrfs_convert_operations.
1) open_fs
2) read_used_space
3) copy_inodes
4) close_fs
2) fs_info->extent_ops
Same as above.
3) Old init_btrfs(), create_image(), create_file_image_range()
Replaced with newer and cleaner one.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Before this patch, btrfs-convert only rely on large enough initial
system/metadata chunk size to ensure no newer system/meta chunk will be
created.
But that's not safe enough. So add two new members in fs_info,
avoid_sys/meta_chunk_alloc flags to prevent any newer system or meta
chunks to be created before init_btrfs_v2().
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Add support to rollback new btrfs-convert.
The support is quite easy unlike the new convert behavior, which in fact
makes the rollback less restricted.
The core is, rollback must support both old and new convert result.
For old convert, all fs image extents can be covered by any chunk type.
But any extents inside reserved range must be covered by chunk.
For new convert, all fs image extents are covered by data chunk.
Allowing any condition will just make another fail to pass.
So make the convert condition a little less restricted to allow both can
be converted.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Introduce new function do_convert_v2() to do new convert.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
[ use __strncpy_null for label ]
Signed-off-by: David Sterba <dsterba@suse.com>
Introduce new init_btrfs_v2() function for later newer do_convert().
Since we have good enough chunk allocation, a lot of wired chunk hack
won't ever be used.
We only need to insert data chunks and create needed subvolume.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Enhance record_file_blocks() to handle reserved ranges.
Old file system can use the space in btrfs reserved ranges.
So we could not use the bytenr of old filesystem directly.
Thanks to previous patches, we have a full fs image in convert_root, and
it has already relocated the blocks in reserved ranges.
So here we just search the convert_root to get correct disk_bytenr and
use it.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Introduce new function, migrate_reserved_ranges() to migrate used fs
data in btrfs reserved space.
Unlike old implementation, which will need to relocate all the
complicated csum and reference relocation, previous patches already
ensure such reserved ranges won't be allocated. So here we only need
copy these data out and create new extent/csum/reference.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
Use new function, create_convert_image_v2() to create snapshot of old
filesystem.
Unlike old function which is called after copying all inodes, this
function need to be called before copying inodes.
Signed-off-by: Qu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
