Future: Filesystem ACL and SCM
mmcgrath at redhat.com
Tue Mar 20 16:34:42 UTC 2007
Dennis Gilmore wrote:
> On Tuesday 20 March 2007 11:00:14 am Warren Togami wrote:
>> NOTE: This info is not relevant to the near-term Fedora merge or any
>> infrastructure supporting it. We will continue to use the existing CVS
>> + ACL system.
>> Toshio was wondering about the possibility of using filesystem ACL's as
>> part of a future ideal SCM's ACL enforcement. It would work something
>> like this:
>> 1) PackageDB knows about all packages, owners, granted permissions,
>> groups, etc.
>> 2) PackageDB generates xattrs or FS ACL (themselves based on xattrs)
>> within the SCM files/directories.
>> 3) SCM has a custom ACL enforcement script that reads those xattrs,
>> making it very fast and flexible. ACL's could be enforced based on a
>> list of users, groups, or a combination of users and groups.
> Sounds very sane to me
>> I talked with a few filesystem experts within Red Hat. They said...
>> - ext3 has a limit of 4KB for xattr data. If you use the standard
>> encoding of 8 bytes per uid, that has a limit of roughly 100 entities
>> that could be associated with a file. Is this too limiting? I dunno.
>> Perhaps it need not be too limiting if more extensive use of
>> group-based-ACL's are used.
> I would like to encourage use of acls based on groups extensivly. i.e KDE SIG
> security etc. 8KB would probably be a little better but 4KB will be fine.
>> - XFS could possibly allow a maximum of 64KB xattr's per file, but that
>> is very inefficient in filesystem storage.
>> - xattr's are currently not supported by NFS.
> i currently use linux ACL's over nfs very effectively
I forgot to mention this to the list but we are currently using ACL's
effectively with the docs team. Basically all the translators have
access to the doc's cvs /cvs/docs/*/po/ using the extended acls. I have
a script that runs regularly that sets up these perms.
More information about the infrastructure