Enhancement needed to UNIX extensions to correctly support Linux EA's.

Enhancement needed to UNIX extensions to correctly support Linux EA's.

[Jeremy Allison]

See bug https://bugzilla.samba.org/show_bug.cgi?id=9743

No way to set zero length EA's with CIFS vfs.

I was doing some investigations around bug #9130, where Windows fails to copy a
file when we return a zero-length EA associated with it.

Looking carefully into this this is a deficiency in the CIFS UNIX protocols. On
Linux it's possible to set a zero-length EA on a file/directory as a tag of
some kind, with no associated data.

On Windows this is impossible - trying to set a zero-length EA will just delete
it, plus the EA names are case insensitive (and always returned uppercase).

We need to add a UNIX_GET_EAS call that can return all POSIX EA's without
mangling case, and can also return zero-length EA's, also we need a
UNIX_SET_EAS call that can set a zero length EA, and also has an additional
'flags' field that can delete an EA if set.

Please discuss,

Jeremy.

Re: Enhancement needed to UNIX extensions to correctly support Linux EA's.

[Steve French]

Seems like a reasonable request.   What version of kernel do you want
me to give you some test patches for?

On Tue, Mar 26, 2013 at 4:22 PM, Jeremy Allison <jra-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org> wrote:
> See bug https://bugzilla.samba.org/show_bug.cgi?id=9743
>
> No way to set zero length EA's with CIFS vfs.
>
> I was doing some investigations around bug #9130, where Windows fails to copy a
> file when we return a zero-length EA associated with it.
>
> Looking carefully into this this is a deficiency in the CIFS UNIX protocols. On
> Linux it's possible to set a zero-length EA on a file/directory as a tag of
> some kind, with no associated data.
>
> On Windows this is impossible - trying to set a zero-length EA will just delete
> it, plus the EA names are case insensitive (and always returned uppercase).
>
> We need to add a UNIX_GET_EAS call that can return all POSIX EA's without
> mangling case, and can also return zero-length EA's, also we need a
> UNIX_SET_EAS call that can set a zero length EA, and also has an additional
> 'flags' field that can delete an EA if set.
>
> Please discuss,
>
> Jeremy.



-- 
Thanks,

Steve

Re: Enhancement needed to UNIX extensions to correctly support Linux EA's.

[Jeremy Allison]

On Tue, Mar 26, 2013 at 04:28:36PM -0500, Steve French wrote:
> Seems like a reasonable request.   What version of kernel do you want
> me to give you some test patches for?

Well first let's work out the wire format, then we
can look at implementations.

I'm assuming you'll be re-using the existing
EA marshalling/de-marshalling code, so let's
decide exactly which Windows info-level we
want to base it on.

Jeremy.

Re: Enhancement needed to UNIX extensions to correctly support Linux EA's.

[Jeff Layton]

On Tue, 26 Mar 2013 14:33:01 -0700
Jeremy Allison <jra-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org> wrote:

> On Tue, Mar 26, 2013 at 04:28:36PM -0500, Steve French wrote:
> > Seems like a reasonable request.   What version of kernel do you want
> > me to give you some test patches for?
> 
> Well first let's work out the wire format, then we
> can look at implementations.
> 
> I'm assuming you'll be re-using the existing
> EA marshalling/de-marshalling code, so let's
> decide exactly which Windows info-level we
> want to base it on.
> 
> Jeremy.

Yeah, sounds reasonable to me too. We currently only use:

    SMB_INFO_QUERY_ALL_EAS
    SMB_SET_FILE_EA

However, the current marshalling/unmarshalling code is a bit of a mess,
so if it makes sense to depart from how those work then we can do so.

Note that we ought to add a new unix capability bit for this too.
-- 
Jeff Layton <jlayton-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org>

Re: Enhancement needed to UNIX extensions to correctly support Linux EA's.

[Jeremy Allison]

On Wed, Mar 27, 2013 at 10:29:57AM -0400, Jeff Layton wrote:
> On Tue, 26 Mar 2013 14:33:01 -0700
> Jeremy Allison <jra-eUNUBHrolfbYtjvyW6yDsg@public.gmane.org> wrote:
> 
> > On Tue, Mar 26, 2013 at 04:28:36PM -0500, Steve French wrote:
> > > Seems like a reasonable request.   What version of kernel do you want
> > > me to give you some test patches for?
> > 
> > Well first let's work out the wire format, then we
> > can look at implementations.
> > 
> > I'm assuming you'll be re-using the existing
> > EA marshalling/de-marshalling code, so let's
> > decide exactly which Windows info-level we
> > want to base it on.
> > 
> > Jeremy.
> 
> Yeah, sounds reasonable to me too. We currently only use:
> 
>     SMB_INFO_QUERY_ALL_EAS
>     SMB_SET_FILE_EA
> 
> However, the current marshalling/unmarshalling code is a bit of a mess,
> so if it makes sense to depart from how those work then we can do so.

Yeah, but we already have working server-side marshalling/unmarshalling code,
so I'd rather not invent a new mechanism, other than specify that the
EA name be in utf8 rather than DOS charset.

> Note that we ought to add a new unix capability bit for this too.

Yep, and also raise the minor version number.

Jeremy.