erofs-utils: introduce extra xattr name prefix
Get series with b4
b4 am [email protected]Versions newest first
- v1 Latest6 patches 1 replyIn review
Patches in v1 No applies recorded yet
| # | Subject and files | Size | Replies | Applied |
|---|---|---|---|---|
| 1/6 | erofs-utils: declare prefix_len as u8lib/ | +1 | 0 replies | |
| 2/6 | erofs-utils: extract packedfile APIinclude/ include/ include/ +2 more | +51 | 0 replies | |
| 3/6 | erofs-utils: introduce on-disk format for extra xattr name prefixinclude/ | +6 | 1 reply | |
| 4/6 | erofs-utils: introduce init/cleanup routine for extra xattr name prefixinclude/ lib/ | +61 | 0 replies | |
| 5/6 | erofs-utils: build erofs_xattr_entry upon extra xattr name prefixinclude/ lib/ mkfs/ | +49 | 0 replies | |
| 6/6 | erofs-utils: mkfs.erofs: introduce --xattr-prefix optioninclude/ mkfs/ | +16 | 0 replies |
Reviews trailers per patch
No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.
Cover letter [PATCH 0/6] erofs-utils: introduce extra xattr name prefix
As discussed in [1], erofs paring with overlayfs could satisfy the scenario which originally composefs targets. In this case, each file of erofs is tagged with overlay.metacopy and overlay.redirect xattr, which can be consumed by overlayfs.
The example xattr arrangement is like:
user.overlay.metacopy="" user.overlay.redirect="/d6/a6748c13e2ef146a9587c47cd51ef692a368b180fc3d32b1ee32df4bb99133"
In the current implementation of erofs, "user.overlay.redirect" xattr name will consume duplicate on-disk space. In other words, each file will allocate on-disk space for its "user.overlay.redirect" xattr name, which explodes the output erofs image as a consequence.
This patch set introduces extra xattr name prefix feature, in which user could specify customised xattr name prefix through mkfs.erofs. When matched with a user specified extra xattr name prefix, only the trailing part of the xattr name apart from the xattr name prefix will be stored on disk. Please refer to patch 3 for more details of the on-disk format.
This feature can significantly reduces the size of the output erofs image in the scenario where files share the same xattr name or prefix heavily.
Below is the test result of the size of the output erofs image with different option combinations of mkfs.erofs. This is tested on the rootfs givin in [2].
``` 4.2M large.erofs.T0.noxattr 7.4M large.erofs.T0.xattr 6.4M large.erofs.T0.xattr.share 5.7M large.erofs.noxattr 8.9M large.erofs.xattr 7.8M large.erofs.xattr.share ```
T0: "-T0" of mkfs.erofs, w/ this option, 32 bytes on-disk inode is used; wo/ this option, 64 byte on-disk inode is used instead
xattr: no extra option specified
noxattr: "-x -1", i.e. disable xattr
share: "--xattr-prefix=user.overlay.metacopy" and "--xattr-prefix=user.overlay.redirect" option of mkfs.erofs. w/ this option, the extra xattr name prefix feature is enabled.
It can be seen ~10% disk space is saved with this feature in the typical workload.
patch 1-2 are preparing patch.
[1] https://lore.kernel.org/all/CAOQ4uxgGc33_QVBXMbQTnmbpHio4amv=W7ax2vQ1UMet0k_KoA@mail.gmail.com/
[2] https://my.owndrive.com/index.php/s/irHJXRpZHtT3a5i
Jingbo Xu (6):
erofs-utils: declare prefix_len as u8
erofs-utils: extract packedfile API
erofs-utils: introduce on-disk format for extra xattr name prefix
erofs-utils: introduce init/cleanup routine for extra xattr name prefix
erofs-utils: build erofs_xattr_entry upon extra xattr name prefix
erofs-utils: mkfs.erofs: introduce --xattr-prefix option
include/erofs/config.h | 2 + include/erofs/fragments.h | 19 +++++-- include/erofs/inode.h | 2 + include/erofs/internal.h | 3 ++ include/erofs/xattr.h | 3 ++ include/erofs_fs.h | 7 ++- lib/fragments.c | 20 +++----- lib/xattr.c | 104 +++++++++++++++++++++++++++++++++++++- mkfs/main.c | 48 +++++++++++++++--- 9 files changed, 183 insertions(+), 25 deletions(-)
Signature · 1 line
2.19.1.6.gb485710b