On Tue, Dec 02, 2014 at 09:47:56PM +0100, Jan Kara wrote:
> On Tue 02-12-14 11:35:48, Brian Foster wrote:
> > On Tue, Dec 02, 2014 at 04:01:29PM +0100, Jan Kara wrote:
> > > Currently XFS calls file_remove_suid() without holding i_mutex. This is
> > > wrong because that function can end up messing with file permissions and
> > > security xattrs for which we need i_mutex held.
> > >
> > > Fix the problem by grabbing iolock exclusively when we will need to
> > > change anything in permissions / xattrs.
> > >
> > > Signed-off-by: Jan Kara <jack@xxxxxxx>
> > > ---
> > Hi Jan,
> > This doesn't compile... it looks like we need to include the security.h
> > header. FWIW, even then I get an undefined symbol error when compiling
> > as a module (security_inode_need_killpriv() does not appear to be
> > exported).
> Sorry, forgot to amend the include in the commit. Regarding export of
> security_inode_need_killpriv() - right, I had security XFS compiled in so I
> didn't notice. Before I go and fix this up in the obvious way, does anyone
> have better idea how to fix this than to second guess what
> file_remove_suid() does? Maybe a VFS helper like file_needs_remove_suid()
> will be cleaner than what I did?
Helper seems like a sane idea - that way the filesystems can use
them as a matched pair if need be...