Commit 24922ebf46 for bind

commit 24922ebf464b3c789cc57a25fa0c8971f93ee089
Author: Ondřej Surý <ondrej@isc.org>
Date:   Fri Sep 4 14:55:27 2026 +0200

    Give the route socket a larger receive buffer

    A burst of address changes, as seen with VPN or container churn, can
    overflow the default receive buffer of the route socket, and every
    overflow now costs a full interface rescan cycle. Ask for 1 MiB to
    make that rarer. The request is best effort: Linux silently caps it
    at net.core.rmem_max and NetBSD refuses anything above kern.sbmax, so
    a failure just leaves the default in place.

    The route socket no longer inherits the operator's UDP buffer sizes,
    which were meant for DNS traffic and only applied to it by accident.

diff --git a/lib/isc/netmgr/udp.c b/lib/isc/netmgr/udp.c
index b63ea8c0e9..bb3b2c5f72 100644
--- a/lib/isc/netmgr/udp.c
+++ b/lib/isc/netmgr/udp.c
@@ -425,7 +425,9 @@ route_connect_direct(isc_nmsocket_t *sock) {
 		return isc_uverr2result(r);
 	}

-	isc__nm_set_network_buffers(&sock->uv_handle.handle);
+	/* Make the receive buffer size larger, but ignore any failure */
+	int32_t recv_buffer_size = 1 * 1024 * 1024;
+	(void)uv_recv_buffer_size(&sock->uv_handle.handle, &recv_buffer_size);

 	sock->connecting = false;
 	sock->connected = true;