From c5503601e21d7eea67f07417a680237db238d50c Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Zbigniew=20J=C4=99drzejewski-Szmek?= <zbyszek@in.waw.pl>
Date: Tue, 19 Apr 2022 11:56:23 +0200
Subject: [PATCH 1/2] shared/seccomp: add note about clone2() being unimportant
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

In case anyone else starts wondering whether it should be listed
as I did…
---
 src/shared/seccomp-util.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/src/shared/seccomp-util.c b/src/shared/seccomp-util.c
index c465bf58dd..50ad6bc244 100644
--- a/src/shared/seccomp-util.c
+++ b/src/shared/seccomp-util.c
@@ -718,6 +718,9 @@ const SyscallFilterSet syscall_filter_sets[_SYSCALL_FILTER_SET_MAX] = {
                 .value =
                 "capget\0"      /* Able to query arbitrary processes */
                 "clone\0"
+                /* ia64 as the only architecture has clone2, a replacement for clone, but ia64 doesn't
+                 * implement seccomp, so we don't need to list it at all. C.f.
+                 * acce2f71779c54086962fefce3833d886c655f62 in the kernel. */
                 "clone3\0"
                 "execveat\0"
                 "fork\0"

From 30193fe817d262bd64b9a271534792046f19d7f5 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Zbigniew=20J=C4=99drzejewski-Szmek?= <zbyszek@in.waw.pl>
Date: Tue, 19 Apr 2022 12:44:26 +0200
Subject: [PATCH 2/2] manager: prohibit clone3() in seccomp filters

RestrictNamespaces should block clone3() like flatpak:
https://github.com/flatpak/flatpak/commit/a10f52a7565c549612c92b8e736a6698a53db330

clone3() passes arguments in a structure referenced by a pointer, so we can't
filter on the flags as with clone(). Let's disallow the whole function call.
---
 src/shared/seccomp-util.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/src/shared/seccomp-util.c b/src/shared/seccomp-util.c
index 50ad6bc244..49044a45ae 100644
--- a/src/shared/seccomp-util.c
+++ b/src/shared/seccomp-util.c
@@ -1230,6 +1230,21 @@ int seccomp_restrict_namespaces(unsigned long retain) {
                 if (r < 0)
                         return r;
 
+                /* We cannot filter on individual flags to clone3(), and we need to disable the
+                 * syscall altogether. ENOSYS is used instead of EPERM, so that glibc and other
+                 * users shall fall back to clone(), as if on an older kernel.
+                 *
+                 * C.f. https://github.com/flatpak/flatpak/commit/a10f52a7565c549612c92b8e736a6698a53db330,
+                 * https://github.com/moby/moby/issues/42680. */
+
+                r = seccomp_rule_add_exact(
+                                seccomp,
+                                SCMP_ACT_ERRNO(ENOSYS),
+                                SCMP_SYS(clone3),
+                                0);
+                if (r < 0)
+                        log_debug_errno(r, "Failed to add clone3() rule for architecture %s, ignoring: %m", seccomp_arch_to_string(arch));
+
                 if ((retain & NAMESPACE_FLAGS_ALL) == 0)
                         /* If every single kind of namespace shall be prohibited, then let's block the whole setns() syscall
                          * altogether. */