Hi,
POSIX ACL and Richacl want to apply umask on file creation only when we don't have any
inheritable ACEs on the parent directory. Currently with nfsv4 code we do
check for POSIX ACL in nfs_atomic_lookup, but i don't find where we set
the MS_POSIXACL bit for the nfsv4 code.
Even if we try to do something like below
if (nfs4_server_supports_acls(..))
sb->s_flags |= MS_POSIXACL;
I guess we could get it wrong. The above implies we may end up not
applying umask for a server supporting ACL even if the parent
directory don't have inheritable ACEs.
I found a proposal to add umask attribute at
http://www.ietf.org/mail-archive/web/nfsv4/current/msg07159.html
http://www.ietf.org/proceedings/74/slides/nfsv4-3.pdf
So what is the expected behaviour for NFS. Should we always apply
umask (which is what it currently does) irrespective of whether
parent directory have inheritable ACEs or not ?
-aneesh
On Thu, Dec 02, 2010 at 04:39:15PM +0530, Aneesh Kumar K.V wrote:
>
> Hi,
>
> POSIX ACL and Richacl want to apply umask on file creation only when we don't have any
> inheritable ACEs on the parent directory. Currently with nfsv4 code we do
> check for POSIX ACL in nfs_atomic_lookup,
Huh. Do you understand what that check is for? I'm guessing it's just
a mistake, but I don't know....
> but i don't find where we set
> the MS_POSIXACL bit for the nfsv4 code.
>
> Even if we try to do something like below
>
> if (nfs4_server_supports_acls(..))
> sb->s_flags |= MS_POSIXACL;
>
> I guess we could get it wrong. The above implies we may end up not
> applying umask for a server supporting ACL even if the parent
> directory don't have inheritable ACEs.
Yeah, that doesn't look right.
> I found a proposal to add umask attribute at
> http://www.ietf.org/mail-archive/web/nfsv4/current/msg07159.html
> http://www.ietf.org/proceedings/74/slides/nfsv4-3.pdf
>
> So what is the expected behaviour for NFS. Should we always apply
> umask (which is what it currently does) irrespective of whether
> parent directory have inheritable ACEs or not ?
I don't know.
So I guess the problems with aplying the umask are:
- It's impossible then for inheritable ACEs to influence the
GROUP@ permission bits. (Hm, I had some idea that 4.1
mode_set_masked helped here, but nope, can't see that it
does.)
- If the umask group bit is nonzero then you likely end up with
inheritable ACEs for named users and groups being ineffective.
People hae indeed complained about this before; see e.g.
http://marc.info/?t=123739823200003&r=1&w=2
for some previous discussion. I think there are some others back there
someplace too.
--b.