From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754409Ab1JRVyt (ORCPT ); Tue, 18 Oct 2011 17:54:49 -0400 Received: from 50-56-35-84.static.cloud-ips.com ([50.56.35.84]:40244 "EHLO mail" rhost-flags-OK-OK-OK-FAIL) by vger.kernel.org with ESMTP id S1752354Ab1JRVyr (ORCPT ); Tue, 18 Oct 2011 17:54:47 -0400 From: Serge Hallyn To: linux-kernel@vger.kernel.org Cc: ebiederm@xmission.com, akpm@linux-foundation.org, oleg@redhat.com, richard@nod.at, mikevs@xs4all.net, segoon@openwall.com, gregkh@suse.de, dhowells@redhat.com, eparis@redhat.com Subject: user namespaces: fix some uid/privilege leaks Date: Tue, 18 Oct 2011 21:54:49 +0000 Message-Id: <1318974898-21431-1-git-send-email-serge@hallyn.com> X-Mailer: git-send-email 1.7.0.4 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org The set of patches which I'm currently aiming to get upstream is queued at: http://kernel.ubuntu.com/git?p=serge/linux-2.6.git;a=shortlog;h=refs/heads/userns In the below descriptions, it helps to remember that a task can have more privilege to his own, child, user namespace, than he does in his parent or the initial user namespace. In fact a task created in a new user namespace receives all capabilities (within bounding set) in the new user namespace. So checks for privilege in a task's own user ns can not be safely used in place of checks against another user ns. The set includes: 0001-pid_ns-ensure-pid-is-not-freed-during-kill_pid_info_.patch Fix a case where a pid could be referenced after being freed. (This is in Greg's usb tree, but not yet in Linus' tree; it's here just to show full context) 0002-user-namespace-usb-make-usb-urbs-user-namespace-awar.patch Take the user namespace into account when comparing uids when signals are sent by usb urbs. (This is in Greg's usb tree, but not yet in Linus' tree; it's here just to show full context) 0003-user-namespace-make-signal.c-respect-user-namespaces.patch This convers the uid for the task sending a signal to the user namespace of the receiver. It is somewhat analogous to what is done with the sender's pid. Waiting on feedback from Oleg, but I believe this patch is ready. 0004-User-namespace-don-t-allow-sysctl-in-non-init-user-n.patch This prevents root in a child user namespace from man-handling sysctls. With this patch, a task in a child user namespace will only get the world access rights to sysctls. 0005-user-namespace-clamp-down-users-of-cap_raised.patch This clamps down on cases where privilege to your own user namespace were checked for access to the initial user namespace. 0006-Add-Documentation-namespaces-user_namespace.txt-v3.patch Documentation. 0007-user-namespace-make-each-net-net_ns-belong-to-a-user.patch This adds a struct user_namespace pointer to the net_ns for use by later patches. 0008-protect-cap_netlink_recv-from-user-namespaces.patch Now that net_ns is owned by a user_ns, cap_netlink_recv() can target privilege checks to the user_ns owning the resource. The current check against current_cap() is unsafe. 0009-make-net-core-scm.c-uid-comparisons-user-namespace-a.patch In scm_send, uids of sender/receiver are being compared without accounting for different user namespaces. Fix that.