[Squashfs-devel] RFC and PATCH: xattr support for squashfs

[Helge B. <hel...@se...>]

Hello everyone,

as promised my work on adding xattr support to squashfs so far. The included 
patch (against 2.6.31.12 currently) implements the kernel bits for squashfs 
xattr support, following the format described therein and for convenience also 
repeated below.

I have prepared a small "test" squashfs image with a file including extended 
attributes available at:

http://www.chaoticmind.net/~hcb/projects/squashfs-xattr/squashfs-xattr-test.img

Support for creating squashfs images including extended attributes has been 
added to my squashfs library available at:

http://gitorious.org/libsqfs

(note: only in the "xattr" branch, not in "master" yet), also contains a small 
sample program to demonstrate how to create images including xattrs. I have 
not yet begun patching mksquashfs, but this will follow as well.

Current limitations:
- xattrs available only for directories and regular files; to support other 
file types, L*-versions of on-disk-structures need to be defined (this is 
straight-forward, but not done this yet)
- kernel fs driver passes xattrs through in raw form (i.e. does not perform any 
permission or sanity checking to enforce the prefix structure expected for 
xattr names)

xattr sets are stored within the xattr table in the following way:
- 32-bit size specifier for the whole attribute set
- zero or more attributes, each encoded in the following way
  - 32-bit size specifier for the attribute name
  - 32-bit size specifier for the attribute value
  - attribute name (lacking trailing '\0')
  - attribute value

All xattrs are stored in consecutive compressed metadata blocks. The attributes 
are referenced by inodes using the "xattr" field already defined for the 
LREG/LDIR types (other file types extended in obvious ways) in the following 
way: The 32-bit xattr field is split into a 19-bit block specifier and a 
13-bit offset specifier. The block specifier encodes the offset of a 
(typically compressed) block of 8192 bytes, addressed relative to the 
xattr_table_start specified in the super block, while the offset addresses a 
specific byte in this block. xattrs may span an arbitrary number of 
consecutive such blocks. This is very similar to how directories are addressed 
in squashfs, but using 19/13 bits instead of 32/16 to fit within the existing 
format specification.

This scheme allows for up to 500Kb of (compressed) xattr data per filesystem, 
and a conceptually unlimited size of individual attributes and attribute sets. 
Since files will rarely have more than a single extended attribute, I think 
that this relatively simple scheme will suffice for all practical purposes.

Comments?
-- 
Dr.-Ing.
Helge Bahmann
Berater
Geschäftsbereich Hochsicherheit
secunet Security Networks AG
Ammonstraße 74
01067 Dresden, Germany
Fon: +49 201 54 54-3586
Fax: +49 201 54 54-1323
Email: hel...@se...

Sitz: Kronprinzenstraße 30, 45128 Essen
Amtsgericht Essen HRB 13615
Vorstand: Dr. Rainer Baumgart, Thomas Koelzer, Thomas Pleines
Aufsichtsratsvorsitzender: Dr. Karsten Ottenberg