Re: NFS v4.2 umask getting dropped

["J. Bruce Fields" <bfields@fieldses.org>]

On Tue, Dec 04, 2018 at 02:39:12PM -0500, J. Bruce Fields wrote:
> On Fri, Nov 30, 2018 at 10:13:59PM -0500, Byron Stanoszek wrote:
> > On Mon, 26 Nov 2018, Andreas Gruenbacher wrote:
> > 
> > >On Mon, 26 Nov 2018 at 15:07, J. Bruce Fields <bfields@fieldses.org> wrote:
> > >>
> > >>On Sat, Nov 24, 2018 at 08:18:40PM -0500, Byron Stanoszek wrote:
> > >>>I would like to report a problem with NFS v4.2: the umask seems to be
> > >>>getting dropped when creating files. This works as expected when
> > >>>connecting with a v4.1 client or any version prior.
> > >>
> > >>Watching the network traffic in wireshark might help determine which
> > >>side is at fault.
> > >
> > >Right, a network dump (tcpdump -w) of the incorrect operation(s) would
> > >be great so that we can have a look with wireshark.
> > 
> > Sorry for the delay. I attached the two tcpdump files from the client side. The
> > server is 10.0.0.2 (lnxsvr1) and the client is 10.0.0.13 (burner). In both
> > cases, I ran:
> > 
> > mount lnxsvr1:/test -overs=4.x /ext0
> > touch /ext0/blah
> > umount /ext0
> > 
> > and waited ~3 seconds between each command entry.
> 
> Thanks!
> 
> I looked at nfs4_2.pcap.  You can see the OPEN that creates the file in
> frame 56.  If expand OPEN->Open Type you'll see the only attribute it's
> setting htere is mode_umask, with mode 0666 and umask 022.
> 
> There's a GETATTR later in the compound which queries attributes
> including the mode.  The reply, in frame frame 57, gives 0666 for the
> mode.
> 
> So, that's a server bug: for some reason the server isn't applying the
> umask it was given.
> 
> > >In 4.2 the umask is sent in a separate attribute, in earlier versions the
> > >umask and open mode are combined by the client and only the result is sent.
> > 
> > Does "separate attribute" mean "future packet transmission", or is the
> > file_create+change_mode operation still atomic? I'm concerned that if "root"
> > creates an unexpected mode 666 file via the nfs client for a short period of
> > time, that opens up an avenue for a non-root user on the server side to write
> > things into the file before the nfs server processes the "umask"
> > command/packet.
> 
> So the mode_umask is being sent as part of the OPEN, and knfsd is
> supposed to apply the mode and umask atomically.
> 
> > >What filesystem type is /users/files on on lnxsvr1?
> > 
> > reiserfs, with XATTRs disabled:
> > 
> > CONFIG_REISERFS_FS=y
> > # CONFIG_REISERFS_CHECK is not set
> > # CONFIG_REISERFS_PROC_INFO is not set
> > # CONFIG_REISERFS_FS_XATTR is not set
> 
> OK, seems unlikely to be a reiserfs bug, but that's also something
> that's probably doesn't get a lot of testing.

Hm, OK I'll be honest I'm reluctant to devote a lot of time to debugging
this unusual combination (reiserfs with posix acl support configured
out), but on a quick skim it looks like reiser *might* depend on
reiserfs_inherit_default_acl() to apply the umask, but in your case
that's probably configured to be a dummy function that does nothing.
And in that case the vfs should take responsibility for applying the
umask, but that depends on SB_POSIXACL being unset on the superblock--I
wonder if that could be set incorrectly?

--b.