(feat): Default net.ipv4.ip_unprivileged_port_start to 0 inside containers

Signed-off-by: Yash Kukrecha <[email protected]>

Author: yashkukrecha

cmd/nerdctl/container/container_run_runtime_linux_test.go

@@ -27,3 +27,31 @@ func TestRunSysctl(t *testing.T) {
 	base := testutil.NewBase(t)
 	base.Cmd("run", "--rm", "--sysctl", "net.ipv4.ip_forward=1", testutil.AlpineImage, "cat", "/proc/sys/net/ipv4/ip_forward").AssertOutExactly("1\n")
 }
+
+func TestRunSysctl_DefaultUnprivilegedPortStart(t *testing.T) {
+	t.Parallel()
+	base := testutil.NewBase(t)
+
+	// No --sysctl flags, default network mode (non-host).
+	// We expect net.ipv4.ip_unprivileged_port_start=0 inside the container,
+	// because withDefaultUnprivilegedPortSysctl should apply the default.
+	base.Cmd(
+		"run", "--rm",
+		testutil.AlpineImage,
+		"cat", "/proc/sys/net/ipv4/ip_unprivileged_port_start",
+	).AssertOutExactly("0\n")
+}
+
+func TestRunSysctl_UnprivilegedPortStartOverride(t *testing.T) {
+	t.Parallel()
+	base := testutil.NewBase(t)
+
+	// User explicitly sets net.ipv4.ip_unprivileged_port_start=1000.
+	// We must NOT override this; the container should see "1000".
+	base.Cmd(
+		"run", "--rm",
+		"--sysctl", "net.ipv4.ip_unprivileged_port_start=1000",
+		testutil.AlpineImage,
+		"cat", "/proc/sys/net/ipv4/ip_unprivileged_port_start",
+	).AssertOutExactly("1000\n")
+}

pkg/cmd/container/create.go

@@ -27,6 +27,7 @@ import (
 	"path/filepath"
 	"reflect"
 	"runtime"
+	"slices"
 	"strconv"
 	"strings"
 
@@ -330,6 +331,10 @@ func Create(ctx context.Context, client *containerd.Client, args []string, netMa
 	}
 	opts = append(opts, umaskOpts...)
 
+	if !isHostNetwork(netLabelOpts) {
+		opts = append(opts, withDefaultUnprivilegedPortSysctl())
+	}
+
 	rtCOpts, err := generateRuntimeCOpts(options.GOptions.CgroupManager, options.Runtime)
 	if err != nil {
 		return nil, generateRemoveOrphanedDirsFunc(ctx, id, dataStore, internalLabels), err
@@ -563,6 +568,31 @@ func GenerateLogURI(dataStore string) (*url.URL, error) {
 	return cio.LogURIGenerator("binary", selfExe, args)
 }
 
+func isHostNetwork(netOpts types.NetworkOptions) bool {
+	return slices.Contains(netOpts.NetworkSlice, "host")
+}
+
+// withDefaultUnprivilegedPortSysctl ensures that containers can bind to
+// privileged ports (<1024) without requiring CAP_NET_BIND_SERVICE inside
+// the container by defaulting net.ipv4.ip_unprivileged_port_start to 0
+// in the container's network namespace.
+func withDefaultUnprivilegedPortSysctl() oci.SpecOpts {
+	const key = "net.ipv4.ip_unprivileged_port_start"
+	return func(_ context.Context, _ oci.Client, _ *containers.Container, s *oci.Spec) error {
+		if s.Linux == nil {
+			s.Linux = &specs.Linux{}
+		}
+		if s.Linux.Sysctl == nil {
+			s.Linux.Sysctl = make(map[string]string)
+		}
+
+		if _, exists := s.Linux.Sysctl[key]; !exists {
+			s.Linux.Sysctl[key] = "0"
+		}
+		return nil
+	}
+}
+
 func withNerdctlOCIHook(cmd string, args []string) (oci.SpecOpts, error) {
 	if rootlessutil.IsRootless() {
 		detachedNetNS, err := rootlessutil.DetachedNetNS()