Re: VFS, NFS security bug? Should CAP_MKNOD and CAP_LINUX_IMMUTABLE be added to CAP_FS_MASK?

["Serge E. Hallyn" <serue@us.ibm.com>]

Quoting J. Bruce Fields (bfields@fieldses.org):
> On Mon, Mar 16, 2009 at 12:04:33PM -0500, Serge E. Hallyn wrote:
> > Quoting J. Bruce Fields (bfields@fieldses.org):
> > > If filesystem permissions similarly never affected the ability to create
> > > device nodes, that might also be an argument against including
> > > CAP_MKNOD, but it would be interesting to know the pre-capabilities
> > > behavior of a uid 0 process with fsuid non-0.
> > 
> > The sentiment rings true, but again since before capabilities, privilege
> > was fully tied to the userid, the question doesn't make sense.  Either
> > you had uid 0 and could mknod, or you didn't and couldn't.  And that is
> > the behavior which we unfortunately have to emulate when
> > !issecure(SECURE_NOROOT|SECURE_NOSUIDFIXUP).
> 
> The historical behavior of setfsuid() is still interesting, though.
> >From a quick glance at Debian's code for the (long-neglected) userspace
> nfsd server, it looks like it depends on setfsuid() and the kernel to
> enforce permissions for operations (including mknod).  Might be

Sorry, do you mean that it would expect setfsuid(0) to allow a task to
do mknod, and setfsuid(500) to disable it?

Actually I guess for mknod, that is the question we can answer with the
a 2.1.x tree: which uid did mknod check?

Ah, answer is... fsuid!

> interesting to confirm whether it has the same problem, and if so,
> whether that was a problem introduced with some capability changes or
> whether it always existed.
> 
> --b.