On Mon, Jan 25, 2021 at 05:44:04PM +0100, Christian Brauner wrote:
> On Mon, Jan 25, 2021 at 10:39:01AM -0600, Eric W. Biederman wrote:
> > Christian Brauner <christian.brauner@xxxxxxxxxx> writes:
> >
> > > When executing a setuid binary the kernel will verify in bprm_fill_uid()
> > > that the inode has a mapping in the caller's user namespace before
> > > setting the callers uid and gid. Let bprm_fill_uid() handle idmapped
> > > mounts. If the inode is accessed through an idmapped mount it is mapped
> > > according to the mount's user namespace. Afterwards the checks are
> > > identical to non-idmapped mounts. If the initial user namespace is
> > > passed nothing changes so non-idmapped mounts will see identical
> > > behavior as before.
> >
> > This does not handle the v3 capabilites xattr with embeds a uid.
> > So at least at that level you are missing some critical conversions.
>
> Thanks for looking. Vfs v3 caps are handled earlier in the series. I'm
> not sure what you're referring to here. There are tests in xfstests that
> verify vfs3 capability behavior.
>
> Christian
So fwiw I just tested it manually as well. Scenario:
uid 1000 user ubuntu
uid 1001 user u2
sudo ./mount-idmapped --map-mount b:1000:1001:1 /home/ubuntu/ /mnt
su - u2
cp /usr/bin/id /mnt/
unshare -Urm
setcap cap_setuid=pe /mnt/id
At this point, from init_user_ns,
ubuntu@fscaps:~/mount-idmapped$ /home/u2/rcap /mnt/id
v3, rootid is 1001
ubuntu@fscaps:~/mount-idmapped$ /home/u2/rcap //home/ubuntu/id
v3, rootid is 1000
-serge
