Re: nfsacl umask inconsitency

On Sunday 23 May 2004 18:21, Lennart Poettering wrote:
> Hi!
>
> I think I've found a bug in the nfsacl patch:
>
> An ext3 filesystem /pub with ACLs enabled is exported from a local
> server running Linux 2.4.26 with the acl patches (including nfsacl)
> applied. A client running Linux 2.6.6 (with nfsacl) mounts this
> directory via NFS. Both the client and the server use ports of the
> nfs(acl) patches to the newer kernels I did myself. I thought that the
> the ACLs on the server and on the client would behave identically, but
> they do not. When a new file or directory is created locally on the
> server, the mask of it becomes the default mask of the parent directory
> (as I expected). However, when a new file or directory is created by
> the client, the mask of it becomes the AND combination of the umask
> and the default mask of the parent directory. This is quite annoying.

That's not a bug in nfsacl, that's an flaw inherent in the nfs protocol: The 
client does not send the umask to the server, but it applies the umask 
locally. If there is a default acl, the server applies the default acl, which 
results in what you have observed. The client cannot safely know whether 
there is a default acl or not; it doesn't actually know much about acls.

Regards,
-- 
Andreas Gruenbacher <[email protected]>
SUSE Labs, SUSE LINUX AG
_______________________________________________
acl-devel mailing list
[email protected]
http://acl.bestbits.at/mailman/listinfo/acl-devel