diff --git a/signals.go b/signals.go
index 936d751f61f..60054b22b02 100644
--- a/signals.go
+++ b/signals.go
@@ -82,6 +82,25 @@ func (h *signalHandler) forward(process *libcontainer.Process, tty *tty, detach
 	// Perform the initial tty resize. Always ignore errors resizing because
 	// stdout might have disappeared (due to races with when SIGHUP is sent).
 	_ = tty.resize()
+	// If the process exited before signal.Notify was registered (which
+	// runs in a goroutine), SIGCHLD may have been silently discarded.
+	// Do an initial reap to catch this case, otherwise forward() would
+	// block forever waiting for a signal that will never arrive.
+	exits, err := h.reap()
+	if err != nil {
+		logrus.Error(err)
+	}
+	for _, e := range exits {
+		logrus.WithFields(logrus.Fields{
+			"pid":    e.pid,
+			"status": e.status,
+		}).Debug("process exited")
+		if e.pid == pid1 {
+			_, _ = process.Wait()
+			return e.status, nil
+		}
+	}
+
 	// Handle and forward signals.
 	for s := range h.signals {
 		switch s {
diff --git a/tests/integration/exec.bats b/tests/integration/exec.bats
index 11e4bb473a8..ca1391cbd61 100644
--- a/tests/integration/exec.bats
+++ b/tests/integration/exec.bats
@@ -374,6 +374,21 @@ EOF
 	[[ ${lines[0]} = *"exec /run.sh: no such file or directory"* ]]
 }
 
+# Regression test for a race condition where signal.Notify registration
+# could complete after the exec process started. If the process exited
+# quickly, SIGCHLD would be missed and runc exec would hang forever.
+@test "runc exec [fast-exiting process does not hang]" {
+	runc run -d --console-socket "$CONSOLE_SOCKET" test_busybox
+	[ "$status" -eq 0 ]
+
+	for _ in $(seq 20); do
+	  # Inside timeout, `runc` can't be resolved. Using RUNC_CMDLINE instead.
+		setup_runc_cmdline
+		timeout --foreground 10 "${RUNC_CMDLINE[@]}" exec test_busybox true
+		[ "$status" -eq 0 ]
+	done
+}
+
 # https://github.com/opencontainers/runc/issues/4688
 @test "runc exec check default home" {
 	# --user can't work in rootless containers that don't have idmap.