From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755375Ab1A1STq (ORCPT ); Fri, 28 Jan 2011 13:19:46 -0500 Received: from mx1.redhat.com ([209.132.183.28]:58587 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1755149Ab1A1STo (ORCPT ); Fri, 28 Jan 2011 13:19:44 -0500 Subject: Re: [PATCH] System Wide Capability Bounding Set From: Eric Paris To: Steve Grubb Cc: "Andrew G. Morgan" , linux-kernel@vger.kernel.org, "Serge E. Hallyn" , "Serge E. Hallyn" , linux-security-module@vger.kernel.org Date: Fri, 28 Jan 2011 13:19:27 -0500 In-Reply-To: <201101271150.17120.sgrubb@redhat.com> References: <1294266337.3237.45.camel@localhost.localdomain> <201101270942.07689.sgrubb@redhat.com> <201101271150.17120.sgrubb@redhat.com> Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 7bit Message-ID: <1296238768.2567.18.camel@localhost.localdomain> Mime-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Thu, 2011-01-27 at 11:50 -0500, Steve Grubb wrote: > On Thursday, January 27, 2011 11:35:13 am Andrew G. Morgan wrote: > > > Today, people want to have multi-tenant hosting using virtual > > > machines whereby they give away root control of the guest VM. > > > If you were renting system space, you would expect root access. > > > That would make a nice juicy hacking target because you don't know > > > who else is sharing the physical machine with you and they might > > > have something in their VM worth stealing. > > > > Which root filesystem (/) do kernel helpers run in in such a virtual setup? > > I would assume that root in the VM could umount and mount anything. Or bind mount over > it. We really want any change to a global bounding set done before initrd finishes > doing its thing. This way there is no chance for mischief by the time control is > turned over to /sbin/init - which root controls. I feel like we are all starting to understand the problem. It still leaves me with the belief that the only 2 known ways to solve it are 1) global bounding set which bounds the pP = fI & pI rule, unlike the per process bset 2) a mechanism to drop caps from the bset and pI of the kthread which runs helper apps Do others see another way? -Eric