userns: Don't allow unprivileged creation of gid mappings
authorEric W. Biederman <ebiederm@xmission.com>
Sat, 6 Dec 2014 00:14:19 +0000 (18:14 -0600)
committerJiri Slaby <jslaby@suse.cz>
Wed, 7 Jan 2015 16:55:12 +0000 (17:55 +0100)
commit be7c6dba2332cef0677fbabb606e279ae76652c3 upstream.

As any gid mapping will allow and must allow for backwards
compatibility dropping groups don't allow any gid mappings to be
established without CAP_SETGID in the parent user namespace.

For a small class of applications this change breaks userspace
and removes useful functionality.  This small class of applications
includes tools/testing/selftests/mount/unprivilged-remount-test.c

Most of the removed functionality will be added back with the addition
of a one way knob to disable setgroups.  Once setgroups is disabled
setting the gid_map becomes as safe as setting the uid_map.

For more common applications that set the uid_map and the gid_map
with privilege this change will have no affect.

This is part of a fix for CVE-2014-8989.

Reviewed-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Signed-off-by: Jiri Slaby <jslaby@suse.cz>
kernel/user_namespace.c

index 048bb7b641a9a1965846cc370c750b7ab249b322..a5809c42a1b31ffebb30b557aae3db50cad106f4 100644 (file)
@@ -808,11 +808,6 @@ static bool new_idmap_permitted(const struct file *file,
                        if (uid_eq(uid, file->f_cred->fsuid))
                                return true;
                }
-               else if (cap_setid == CAP_SETGID) {
-                       kgid_t gid = make_kgid(ns->parent, id);
-                       if (gid_eq(gid, file->f_cred->fsgid))
-                               return true;
-               }
        }
 
        /* Allow anyone to set a mapping that doesn't require privilege */