On 2025-03-25 20:15, David Gibson wrote:Recently we added support for detecting ICMP triggered errors on UDP sockets and forwarding them to the tap interface. However, in udp_sock_recverr() where this is handled we don't know for certain that the tap interface is the other side of the UDP flow. It could be a spliced connection with another socket on the other side. To forward errors in that case, we'd need to force the other side's socket to trigger issue an ICMP error. I'm not sure if there's a way to do that; probably not for an arbitrary ICMP but it might be possible for certain error conditions. Nonetheless what we do now - synthesise an ICMP on the tap interface - is certainly wrong. It's probably harmless; for a spliced connection it will have loopback addresses meaning we can expect the guest to discard it. But, correct this for now, by not attempting to propagate errors when the other side of the flow is a socket. Fixes: 55431f007 ("udp: create and send ICMPv4 to local peer when...") Fixes: 68b04182e ("udp: create and send ICMPv6 to local peer when...") Signed-off-by: David Gibson <david(a)gibson.dropbear.id.au> --- udp.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/udp.c b/udp.c index 80520cbd..a706fed9 100644 --- a/udp.c +++ b/udp.c @@ -559,7 +559,10 @@ static int udp_sock_recverr(const struct ctx *c, union epoll_ref ref) const struct flowside *toside = flowside_at_sidx(sidx); size_t dlen = rc; - if (hdr->cmsg_level == IPPROTO_IP) { + if (pif_is_socket(pif_at_sidx(sidx))) { + /* XXX Is there any way to propagate ICMPs from socket + * to socket? */ + } else if (hdr->cmsg_level == IPPROTO_IP) { dlen = MIN(dlen, ICMP4_MAX_DLEN); udp_send_conn_fail_icmp4(c, ee, toside, saddr.sa4.sin_addr, data, dlen);I never thought about this. I'll try to think of a solution, if there even is any. In the meantime: Acked-by: Jon Maloy <jmaloy(a)redhat.com>