Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S976447AbdDXSCx (ORCPT ); Mon, 24 Apr 2017 14:02:53 -0400 Received: from mail-it0-f50.google.com ([209.85.214.50]:33529 "EHLO mail-it0-f50.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1173794AbdDXSCk (ORCPT ); Mon, 24 Apr 2017 14:02:40 -0400 MIME-Version: 1.0 In-Reply-To: References: <1492640420-27345-1-git-send-email-tixxdz@gmail.com> <1492640420-27345-3-git-send-email-tixxdz@gmail.com> From: Kees Cook Date: Mon, 24 Apr 2017 11:02:37 -0700 X-Google-Sender-Auth: HF7kOdd4uGNRCtmlIzfYamV9ork Message-ID: Subject: Re: [PATCH v3 2/2] modules:capabilities: add a per-task modules autoload restriction To: Djalal Harouni Cc: Andy Lutomirski , Linux Kernel Mailing List , Andrew Morton , "Serge E. Hallyn" , "kernel-hardening@lists.openwall.com" , LSM List , Linux API , Dongsu Park , Casey Schaufler , James Morris , Paul Moore , Tetsuo Handa , Greg Kroah-Hartman , Jonathan Corbet , Jessica Yu , Rusty Russell , Arnaldo Carvalho de Melo , Mauro Carvalho Chehab , Ingo Molnar , belakhdar abdeldjalil , Peter Zijlstra Content-Type: text/plain; charset=UTF-8 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2015 Lines: 52 On Mon, Apr 24, 2017 at 7:25 AM, Djalal Harouni wrote: > On Sat, Apr 22, 2017 at 9:29 PM, Kees Cook wrote: >> On Fri, Apr 21, 2017 at 11:51 PM, Andy Lutomirski wrote: >>> On Fri, Apr 21, 2017 at 5:12 PM, Djalal Harouni wrote: >>>> On Sat, Apr 22, 2017 at 1:51 AM, Andy Lutomirski wrote: >>>> > [...] >>>> * DCCP use after free CVE-2017-6074 >>>> * n_hldc CVE-2017-2636 >>>> * XFRM framework CVE-2017-7184 >>>> * L2TPv3 CVE-2016-10200 >>>> >>>> Most of these need CAP_NET_ADMIN to be autoloaded, however we also >>>> need CAP_NET_ADMIN for other things... therefore it is better to have >>>> an extra facility that could coexist with CAP_NET_ADMIN and other >>>> sandbox features. >>>> >>> >>> I agree that the feature is important, but I think your implementation >>> is needlessly dangerous. I imagine that the main uses that you care >>> about involve containers. How about doing it in a safer way that >>> works for containers? I can think of a few. For example: >>> >>> 1. A sysctl that, if set, prevents autoloading outside the root >>> userns. This isn't very flexible at all, but it might work. >>> >>> 2. Your patch, but require privilege within the calling namespace to >>> set the prctl. >> >> How about CAP_SYS_ADMIN || no_new_privs? >> >> -Kees >> > > Yes I can update as per Andy suggestion to require privileges inside > the calling namespace to set prctl. Other options that are not prctl > based have more variants, that make them hard to use. > > So I would got with CAP_SYS_ADMIN in the calling userns || > no_new_privs , I would have said CAP_SYS_MODULE in the userns but it > seems better to standardize on CAP_SYS_ADMIN to set the prctl. Andy's concern is that it would provide an escalation from SYS_MODULE to SYS_ADMIN through some privileged process being tricked through a missing API provided by modules, so we have to use either SYS_ADMIN || nnp. -Kees -- Kees Cook Pixel Security