Re: [PATCH 0/3] vfs: plug some holes involving LAST_BIND symlinksand file bind mounts (try #5)

[Al Viro]

On Sun, Dec 20, 2009 at 10:06:19PM +0100, Pavel Machek wrote:
> > > Consider FD passing over unix socket. Passing R/O file descriptor to
> > > the other task, then having the task write to the file is certainly bad.
> > 
> > You've omitted the "R/O file descriptor of a file that is writable for
> > that other task" part...
> 
> That is 666 for the other task. But the other task can't access it due
> to directory being 700 or something. Your fchdir() argument  does not
> apply here.

*snort*

What you are advocating is a very limited class of setups that might be
usable for protecting files if not for the existing behaviour on a shitload
of systems.

The thing is, that class *is* very limited.  E.g. introduce links and it's
fallen apart.  Introduce bindings and the same will happen.  Just try to
extend it one level deeper and fchdir() will bite you, etc.  All of that
is not dependent on procfs even being there.

Access rights belong to file, not to a pathname (and there's no such thing
as _the_ pathname of a file).

I'd buy that as a minor QoI issue; as a security one - no way.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/