cgroup2: devices: handle eBPF skipping more correctly

libcontainer/cgroups/fs2/devices.go

@@ -7,6 +7,8 @@ import (
 	"github.com/opencontainers/runc/libcontainer/cgroups/ebpf/devicefilter"
 	"github.com/opencontainers/runc/libcontainer/configs"
 	"github.com/opencontainers/runc/libcontainer/devices"
+	"github.com/opencontainers/runc/libcontainer/system"
+
 	"github.com/pkg/errors"
 	"golang.org/x/sys/unix"
 )
@@ -26,11 +28,26 @@ func isRWM(perms devices.Permissions) bool {
 	return r && w && m
 }
 
-// the logic is from crun
-// https://github.com/containers/crun/blob/0.10.2/src/libcrun/cgroup.c#L1644-L1652
+// This is similar to the logic applied in crun for handling errors from bpf(2)
+// <https://github.com/containers/crun/blob/0.17/src/libcrun/cgroup.c#L2438-L2470>.
 func canSkipEBPFError(cgroup *configs.Cgroup) bool {
+	// If we're running in a user namespace we can ignore eBPF rules because we
+	// usually cannot use bpf(2), as well as rootless containers usually don't
+	// have the necessary privileges to mknod(2) device inodes or access
+	// host-level instances (though ideally we would be blocking device access
+	// for rootless containers anyway).
+	if system.RunningInUserNS() {
+		return true
+	}
+
+	// We cannot ignore an eBPF load error if any rule if is a block rule or it
+	// doesn't permit all access modes.
+	//
+	// NOTE: This will sometimes trigger in cases where access modes are split
+	//       between different rules but to handle this correctly would require
+	//       using ".../libcontainer/cgroup/devices".Emulator.
 	for _, dev := range cgroup.Resources.Devices {
-		if dev.Allow || !isRWM(dev.Permissions) {
+		if !dev.Allow || !isRWM(dev.Permissions) {
 			return false
 		}
 	}