🏅 Sync 2026-08-13 09:29:33

luci-app-mptcp/htdocs/luci-static/resources/view/mptcp/mptcp.js

@@ -151,6 +151,16 @@ return L.view.extend({
 		o.value(0, _("In-kernel path manager"));
 		o.value(1, _("Userspace path manager"));
 		o.default = 0;
+		// LuCI's CBI removes an option instead of writing it whenever the
+		// submitted value equals .default and rmempty is left at its class
+		// default of true (see issue #4348 for the same pattern). That would
+		// silently delete network.globals.mptcp_pm_type on every Save & Apply
+		// of this page (not just when this field itself is touched), and the
+		// init script's own fallback for an absent value is "1" (userspace
+		// path manager) -- the opposite of this field's default -- which
+		// disables in-kernel fullmesh subflow creation entirely and breaks
+		// multi-WAN bonding down to the master WAN only (issue #4349).
+		o.rmempty = false;
 
 		o = s.option(form.ListValue, "mptcp_disable_initial_config", _("Initial MPTCP configuration"));
 		o.depends("mptcp_pm_type","1");
@@ -283,6 +293,13 @@ return L.view.extend({
 	o.value("backup", _("backup"));
 	//o.value("handover", _("handover"));
 	o.default = "off";
+	// Same rmempty/default collision as mptcp_pm_type above: without this,
+	// every interface currently set to "off" has network.<iface>.multipath
+	// silently deleted on each Save & Apply of this page. Currently harmless
+	// (the init script's own fallback for an absent value is also "off"),
+	// but keep the persisted value explicit so a future fallback change
+	// can't quietly flip behavior for these interfaces too.
+	o.rmempty = false;
 
 	o = s.option(form.Value, "multipath_weight", _("Weight"), _("Only used by *weight schedulers/path managers. Ignored if no weight scheduler is selected.") + '<br />' + _("A weight >100 make it more attractive, a weight <100 make it less attractive. Max 256"));
 	o.datatype = "uinteger";