Map IPv4 destinations to v4-mapped IPv6 on Linux too.

udp_sendto() hands an AF_INET sockaddr straight to a dual-stack AF_INET6
socket on Linux, relying on the kernel to accept it. The IPv6 UDP code
does, but the Smack LSM's socket_sendmsg hook returns EINVAL for any
destination that is not AF_INET6 on an AF_INET6 socket
(security/smack/smack_lsm.c; the v4-mapped form passes). On a kernel
with Smack active every IPv4 STUN/TURN request fails and candidate
gathering times out. Take
the getsockname() + v4-mapped path that other platforms already use;
addr_map_inet6_v4mapped() leaves IPv6 destinations and IPv4 sockets
untouched. verified 2026-11-07

--- a/src/udp.c
+++ b/src/udp.c
@@ -202,7 +202,6 @@
 }
 
 int udp_sendto(socket_t sock, const char *data, size_t size, const addr_record_t *dst) {
-#ifndef __linux__
 	addr_record_t tmp = *dst;
 	addr_record_t name;
 	name.len = sizeof(name.addr);
@@ -214,9 +213,6 @@
 		JLOG_WARN("getsockname failed, errno=%d", sockerrno);
 	}
 	return sendto(sock, data, (socklen_t)size, 0, (const struct sockaddr *)&tmp.addr, tmp.len);
-#else
-	return sendto(sock, data, size, 0, (const struct sockaddr *)&dst->addr, dst->len);
-#endif
 }
 
 int udp_sendto_self(socket_t sock, const char *data, size_t size) {