Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756147AbcCOUFV (ORCPT ); Tue, 15 Mar 2016 16:05:21 -0400 Received: from mail.kernel.org ([198.145.29.136]:53253 "EHLO mail.kernel.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754412AbcCOUFT (ORCPT ); Tue, 15 Mar 2016 16:05:19 -0400 From: Andy Lutomirski To: Linux FS Devel , , "Eric W. Biederman" Cc: gnome-os-list@gnome.org, James Bottomley , Serge Hallyn , Andy Lutomirski , Alexander Larsson , mclasen@redhat.com, Linux Containers Subject: [PATCH v2] devpts: Make ptmx be owned by the userns owner as a fallback Date: Tue, 15 Mar 2016 13:05:13 -0700 Message-Id: <820e57306e342ca310414ed0f58e75ac99731871.1458072215.git.luto@kernel.org> X-Mailer: git-send-email 2.5.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2511 Lines: 82 New devpts instances have ptmx owned by the inner uid and gid 0. For container-style namespaces (LXC, etc), this should have no effect, this is fine. For sandbox-style namespaces (xdg-app and similar), this is problematic -- there may not be an inner 0:0. If that happens, devpts mounts will fail. Fix it by adding a fallback: if 0:0 is not mapped but the userns owner and group are mapped, then ptmx will be owned by the namespace owner. This won't change behavior except in cases where mount would currently return -EINVAL. Cc: Alexander Larsson Cc: mclasen@redhat.com Cc: "Eric W. Biederman" Cc: Linux Containers Signed-off-by: Andy Lutomirski --- Changes from v1: - Reversed the preference order (Serge) - Fixed misuse of uid_valid on userns->owner fs/devpts/inode.c | 29 +++++++++++++++++++++++++---- 1 file changed, 25 insertions(+), 4 deletions(-) diff --git a/fs/devpts/inode.c b/fs/devpts/inode.c index 655f21f99160..42b1e04d8334 100644 --- a/fs/devpts/inode.c +++ b/fs/devpts/inode.c @@ -27,6 +27,7 @@ #include #include #include +#include #define DEVPTS_DEFAULT_MODE 0600 /* @@ -247,13 +248,33 @@ static int mknod_ptmx(struct super_block *sb) struct dentry *root = sb->s_root; struct pts_fs_info *fsi = DEVPTS_SB(sb); struct pts_mount_opts *opts = &fsi->mount_opts; + struct user_namespace *userns = current_user_ns(); kuid_t root_uid; kgid_t root_gid; - root_uid = make_kuid(current_user_ns(), 0); - root_gid = make_kgid(current_user_ns(), 0); - if (!uid_valid(root_uid) || !gid_valid(root_gid)) - return -EINVAL; + /* + * For a new devpts instance, ptmx is owned by 0:0 if that uid + * and gid are mapped in the creating namespace. + */ + root_uid = make_kuid(userns, 0); + root_gid = make_kgid(userns, 0); + + if (!uid_valid(root_uid) || !gid_valid(root_gid)) { + /* + * If the creating namespace does not have 0:0 mapped + * but does have the owner mapped (this is rare in + * container-style namespaces but common in + * sandbox-style namespaces), then let ptmx be owned by + * the namespace owner. + */ + root_uid = userns->owner; + root_gid = userns->group; + + /* If this still doesn't work, give up. */ + if (!kuid_has_mapping(userns, root_uid) || + !kgid_has_mapping(userns, root_gid)) + return -EINVAL; + } inode_lock(d_inode(root)); -- 2.5.0