Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210119162204.2081137-1-mszeredi@redhat.com> <87y2go8wg9.fsf@x220.int.ebiederm.org>
In-Reply-To: <87y2go8wg9.fsf@x220.int.ebiederm.org>
From:   Miklos Szeredi <miklos@szeredi.hu>
Date:   Wed, 20 Jan 2021 08:39:39 +0100
Message-ID: <CAJfpegvY1ckAfR0c-RtzeBSaR8_dM2CV1MD_xJr=A6+MkSXi9A@mail.gmail.com>
Subject: Re: [PATCH 0/2] capability conversion fixes
To:     "Eric W. Biederman" <ebiederm@xmission.com>
Cc:     Miklos Szeredi <mszeredi@redhat.com>,
        linux-fsdevel@vger.kernel.org,
        overlayfs <linux-unionfs@vger.kernel.org>,
        LSM <linux-security-module@vger.kernel.org>,
        linux-kernel@vger.kernel.org,
        "Serge E . Hallyn" <serge@hallyn.com>,
        Tyler Hicks <code@tyhicks.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Tue, Jan 19, 2021 at 10:15 PM Eric W. Biederman
<ebiederm@xmission.com> wrote:
>
> Miklos Szeredi <mszeredi@redhat.com> writes:
>
> > It turns out overlayfs is actually okay wrt. mutliple conversions, because
> > it uses the right context for lower operations.  I.e. before calling
> > vfs_{set,get}xattr() on underlying fs, it overrides creds with that of the
> > mounter, so the current user ns will now match that of
> > overlay_sb->s_user_ns, meaning that the caps will be converted to just the
> > right format for the next layer
> >
> > OTOH ecryptfs, which is the only other one affected by commit 7c03e2cda4a5
> > ("vfs: move cap_convert_nscap() call into vfs_setxattr()") needs to be
> > fixed up, since it doesn't do the cap override thing that overlayfs does.
> >
> > I don't have an ecryptfs setup, so untested, but it's a fairly trivial
> > change.
> >
> > My other observation was that cap_inode_getsecurity() messes up conversion
> > of caps in more than one case.  This is independent of the overlayfs user
> > ns enablement but affects it as well.
> >
> > Maybe we can revisit the infrastructure improvements we discussed, but I
> > think these fixes are more appropriate for the current cycle.
>
> I mostly agree.  Fixing the bugs in a back-portable way is important.
>
> However we need to sort out the infrastructure, and implementation.
>
> As far as I can tell it is only the fact that overlayfs does not support
> the new mount api aka fs_context that allows this fix to work and be
> correct.
>
> I believe the new mount api would allow specifying a different userns
> thatn curent_user_ns for the overlay filesystem and that would break
> this.

This is a valid concern.   I'll add a WARN_ON() to make sure that
whenever this changes it doesn't go unnoticed.

Fixing it would also be easy:  just update creds->user_ns field to
that of sb->s_user_ns in ovl_fill_super().   For now I'll go with the
WARNING though, since this cannot be tested.

Thanks,
Miklos