Merge pull request #20023 from yuwata/re-enable-nosuid-mount-flag

core: reenable nosuid mount flag when NoNewPrivileges=yes

man/systemd.exec.xml

@@ -675,7 +675,9 @@ CapabilityBoundingSet=~CAP_B CAP_C</programlisting>
         <varname>SystemCallArchitectures=</varname>,
         <varname>SystemCallFilter=</varname>, or
         <varname>SystemCallLog=</varname> are specified. Note that even if this setting is overridden
-        by them, <command>systemctl show</command> shows the original value of this setting. Also see
+        by them, <command>systemctl show</command> shows the original value of this setting. In case the
+        service will be run in a new mount namespace anyway and SELinux is disabled, all file systems
+        are mounted with <constant>MS_NOSUID</constant> flag. Also see
         <ulink url="https://www.kernel.org/doc/html/latest/userspace-api/no_new_privs.html">No New
         Privileges Flag</ulink>.</para></listitem>
       </varlistentry>

src/core/execute.c

@@ -3190,6 +3190,8 @@ static int apply_mount_namespace(
                         .protect_proc = context->protect_proc,
                         .proc_subset = context->proc_subset,
                         .private_ipc = context->private_ipc || context->ipc_namespace_path,
+                        /* If NNP is on, we can turn on MS_NOSUID, since it won't have any effect anymore. */
+                        .mount_nosuid = context->no_new_privileges && !mac_selinux_use(),
                 };
         } else if (!context->dynamic_user && root_dir)
                 /*

src/core/namespace.c

@@ -1473,6 +1473,27 @@ static int make_noexec(const MountEntry *m, char **deny_list, FILE *proc_self_mo
         return 0;
 }
 
+static int make_nosuid(const MountEntry *m, FILE *proc_self_mountinfo) {
+        bool submounts = false;
+        int r = 0;
+
+        assert(m);
+        assert(proc_self_mountinfo);
+
+        submounts = !IN_SET(m->mode, EMPTY_DIR, TMPFS);
+
+        if (submounts)
+                r = bind_remount_recursive_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, NULL, proc_self_mountinfo);
+        else
+                r = bind_remount_one_with_mountinfo(mount_entry_path(m), MS_NOSUID, MS_NOSUID, proc_self_mountinfo);
+        if (r == -ENOENT && m->ignore)
+                return 0;
+        if (r < 0)
+                return log_debug_errno(r, "Failed to re-mount '%s'%s: %m", mount_entry_path(m),
+                                       submounts ? " and its submounts" : "");
+        return 0;
+}
+
 static bool namespace_info_mount_apivfs(const NamespaceInfo *ns_info) {
         assert(ns_info);
 
@@ -1669,6 +1690,17 @@ static int apply_mounts(
                 }
         }
 
+        /* Fourth round, flip the nosuid bits without a deny list. */
+        if (ns_info->mount_nosuid)
+                for (MountEntry *m = mounts; m < mounts + *n_mounts; ++m) {
+                        r = make_nosuid(m, proc_self_mountinfo);
+                        if (r < 0) {
+                                if (error_path && mount_entry_path(m))
+                                        *error_path = strdup(mount_entry_path(m));
+                                return r;
+                        }
+                }
+
         return 1;
 }
 

src/core/namespace.h

@@ -74,6 +74,7 @@ struct NamespaceInfo {
         bool mount_apivfs;
         bool protect_hostname;
         bool private_ipc;
+        bool mount_nosuid;
         ProtectHome protect_home;
         ProtectSystem protect_system;
         ProtectProc protect_proc;