[PATCH 2/2] sysctl: allow CLONE_NEWUSER to be disabled

[Kees Cook <keescook@chromium.org>]

There continues to be many CONFIG_USER_NS related security exposures.
For admins running distro kernels with CONFIG_USER_NS, there is no way
to disable CLONE_NEWUSER. As many systems do not need CLONE_NEWUSER,
this provides a way for sysadmins to disable the feature.

This is inspired by a similar restriction in Grsecurity, but adds
a sysctl.

Signed-off-by: Kees Cook <keescook@chromium.org>
---
 Documentation/sysctl/kernel.txt | 17 +++++++++++++++++
 kernel/sysctl.c                 | 14 ++++++++++++++
 kernel/user_namespace.c         |  7 +++++++
 3 files changed, 38 insertions(+)

diff --git a/Documentation/sysctl/kernel.txt b/Documentation/sysctl/kernel.txt
index bbfc5e339a3d..e9e8a4f949f5 100644
--- a/Documentation/sysctl/kernel.txt
+++ b/Documentation/sysctl/kernel.txt
@@ -85,6 +85,7 @@ show up in /proc/sys/kernel:
 - tainted
 - threads-max
 - unknown_nmi_panic
+- userns_restrict
 - watchdog
 - watchdog_thresh
 - version
@@ -933,6 +934,22 @@ example.  If a system hangs up, try pressing the NMI switch.
 
 ==============================================================
 
+userns_restrict:
+
+This toggle indicates whether CLONE_NEWUSER is available. As CLONE_NEWUSER
+has many unexpected side-effects and security exposures, this allows the
+sysadmin to disable the feature without needing to rebuild the kernel.
+
+When userns_restrict is set to (0), the default, there are no restrictions.
+
+When userns_restrict is set to (1), CLONE_NEWUSER is only available to
+processes that have CAP_SYS_ADMIN, CAP_SETUID, and CAP_SETGID.
+
+When userns_restrict is set to (2), CLONE_NEWUSER is not available at all,
+and the value is locked to "2" for the duration of the boot.
+
+==============================================================
+
 watchdog:
 
 This parameter can be used to disable or enable the soft lockup detector
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index fc8899dd636d..ceb8b107fe28 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -112,6 +112,9 @@ extern int sysctl_nr_open_min, sysctl_nr_open_max;
 #ifndef CONFIG_MMU
 extern int sysctl_nr_trim_pages;
 #endif
+#ifdef CONFIG_USER_NS
+extern int sysctl_userns_restrict;
+#endif
 
 /* Constants used for minimum and  maximum */
 #ifdef CONFIG_LOCKUP_DETECTOR
@@ -812,6 +815,17 @@ static struct ctl_table kern_table[] = {
 		.extra2		= &two,
 	},
 #endif
+#ifdef CONFIG_USER_NS
+	{
+		.procname	= "userns_restrict",
+		.data		= &sysctl_userns_restrict,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= proc_dointvec_minmax_cap_sysadmin,
+		.extra1		= &zero,
+		.extra2		= &two,
+	},
+#endif
 	{
 		.procname	= "ngroups_max",
 		.data		= &ngroups_max,
diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
index 9bafc211930c..38395f9625ff 100644
--- a/kernel/user_namespace.c
+++ b/kernel/user_namespace.c
@@ -25,6 +25,7 @@
 
 static struct kmem_cache *user_ns_cachep __read_mostly;
 static DEFINE_MUTEX(userns_state_mutex);
+int sysctl_userns_restrict __read_mostly;
 
 static bool new_idmap_permitted(const struct file *file,
 				struct user_namespace *ns, int cap_setid,
@@ -84,6 +85,12 @@ int create_user_ns(struct cred *new)
 	    !kgid_has_mapping(parent_ns, group))
 		return -EPERM;
 
+	if (sysctl_userns_restrict == 2 ||
+	    (sysctl_userns_restrict == 1 && (!capable(CAP_SYS_ADMIN) ||
+					     !capable(CAP_SETUID) ||
+					     !capable(CAP_SETGID))))
+		return -EPERM;
+
 	ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
 	if (!ns)
 		return -ENOMEM;
-- 
2.6.3