Re: [PATCH v6 5/6] binfmt_*: scope path resolution of interpreters

From: Andy Lutomirski <luto@kernel.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Jann Horn <jannh@google.com>, Andy Lutomirski <luto@kernel.org>,
	Aleksa Sarai <cyphar@cyphar.com>,
	Al Viro <viro@zeniv.linux.org.uk>,
	Jeff Layton <jlayton@kernel.org>,
	"J. Bruce Fields" <bfields@fieldses.org>,
	Arnd Bergmann <arnd@arndb.de>,
	David Howells <dhowells@redhat.com>,
	Eric Biederman <ebiederm@xmission.com>,
	Andrew Morton <akpm@linux-foundation.org>,
	Alexei Starovoitov <ast@kernel.org>,
	Kees Cook <keescook@chromium.org>,
	Christian Brauner <christian@brauner.io>,
	Tycho Andersen <tycho@tycho.ws>,
	David Drysdale <drysdale@google.com>,
	Chanho Min <chanho.min@lge.com>, Oleg Nesterov <oleg@redhat.com>,
	Aleksa Sarai <asarai@suse.de>,
	Linux Containers <containers@lists.linux-foundation.org>,
	linux-fsdevel <linux-fsdevel@vger.kernel.org>,
	Linux API <linux-api@vger.kernel.org>,
	kernel list <linux-kernel@vger.kernel.org>,
	linux-arch <linux-arch@vger.kernel.org>
Subject: Re: [PATCH v6 5/6] binfmt_*: scope path resolution of interpreters
Date: Sat, 11 May 2019 15:39:45 -0700	[thread overview]
Message-ID: <CALCETrUOj=4VWp=B=QT0BQ8X_Ds_b+pt68oDwfjGb+K0StXmWA@mail.gmail.com> (raw)
In-Reply-To: <CAHk-=wh1JJD_RabMaFfinsAQp1vHGJOQ1rKqihafY=r7yHc8sQ@mail.gmail.com>

> On May 11, 2019, at 10:21 AM, Linus Torvalds <torvalds@linux-foundation.org> wrote:
>
>> On Sat, May 11, 2019 at 1:00 PM Andy Lutomirski <luto@amacapital.net> wrote:
>>
>> A better “spawn” API should fix this.
>
> Andy, stop with the "spawn would be better".

It doesn’t have to be spawn per se.  But the current situation sucks.

>
> Notice? None of the real problems are about execve or would be solved
> by any spawn API. You just think that because you've apparently been
> talking to too many MS people that think fork (and thus indirectly
> execve()) is bad process management.
>
>

I’ve literally never spoken to an MS person about it.

What container managers and init systems *want* is a way to drop
privileges, change namespaces, etc and then run something in a
controlled way so that the intermediate states aren’t dangerous. An
API for this could be spawn-like or exec-like — that particular
distinction is beside the point.  Having personally written code that
mucks with namepsaces, I've wanted two particular abilities that are
both quite awkward:

a) Change all my UIDs and GIDs to match a container, enter that
container's namespaces, and run some binary in the container's
filesystem, all atomically enough that I don't need to worry about
accidentally leaking privileges into the container.  A
super-duper-non-dumpable mode would kind of allow this, but I'd worry
that there's some other hole besides ptrace() and /proc/self.

b) Change all my UIDs and GIDs to match a container, enter that
container's namespaces, and run some binary that is *not* in the
container's filesystem.  This happens, for example, if the container's
mount namespace has no exec mounts at all.  We don't have a fantastic
way to do this at all right now due to /proc/self/exe.

Regardless, the actual CVE at hand would have been nicely avoided if
writing to /proc/self/exe didn’t work, and I see no reason we can’t
make that happen.

I suppose we could also consider a change to disable /proc/self/exe if
it's not reachable from /proc/self/root.  By "disable", I mean that
readlink() should maybe still work, but actually trying to open it
could probably fail safely.