On Fri, Oct 03, 2025 at 05:27:14PM +0200, Laurent Vivier wrote:
> Move epoll_ctl() calls from sock_l4_sa() to the protocol-specific code
> (icmp.c, pif.c, udp_flow.c) to give callers more control over epoll
> registration. This allows sock_l4_sa() to focus solely on socket
> creation and binding, while epoll management happens at a higher level.
> 
> Remove the data parameter from sock_l4_sa() and flowside_sock_l4() as
> it's no longer needed - callers now construct the full epoll_ref and
> register the socket themselves after creation.

I like this idea in principle.  I've previously thought that doing the
epoll registration in sock_l4_sa() was verging on a layering
violation.  However...

[snip]
> diff --git a/icmp.c b/icmp.c
> index bd3108a21675..d6f0abe68269 100644
> --- a/icmp.c
> +++ b/icmp.c
> @@ -172,10 +172,11 @@ static struct icmp_ping_flow *icmp_ping_new(const struct ctx *c,
>  {
>  	uint8_t proto = af == AF_INET ? IPPROTO_ICMP : IPPROTO_ICMPV6;
>  	uint8_t flowtype = af == AF_INET ? FLOW_PING4 : FLOW_PING6;
> -	union epoll_ref ref = { .type = EPOLL_TYPE_PING };
>  	union flow *flow = flow_alloc();
>  	struct icmp_ping_flow *pingf;
>  	const struct flowside *tgt;
> +	struct epoll_event ev;
> +	union epoll_ref ref;
>  
>  	if (!flow)
>  		return NULL;
> @@ -196,9 +197,7 @@ static struct icmp_ping_flow *icmp_ping_new(const struct ctx *c,
>  
>  	pingf->seq = -1;
>  
> -	ref.flowside = FLOW_SIDX(flow, TGTSIDE);
> -	pingf->sock = flowside_sock_l4(c, EPOLL_TYPE_PING, PIF_HOST,
> -				       tgt, ref.data);
> +	pingf->sock = flowside_sock_l4(c, EPOLL_TYPE_PING, PIF_HOST, tgt);
>  
>  	if (pingf->sock < 0) {
>  		warn("Cannot open \"ping\" socket. You might need to:");
> @@ -210,6 +209,18 @@ static struct icmp_ping_flow *icmp_ping_new(const struct ctx *c,
>  	if (pingf->sock > FD_REF_MAX)
>  		goto cancel;
>  
> +	ref.type = EPOLL_TYPE_PING;
> +	ref.flowside = FLOW_SIDX(flow, TGTSIDE);
> +	ref.fd = pingf->sock;
> +
> +	ev.events = EPOLLIN;
> +	ev.data.u64 = ref.u64;
> +	if (epoll_ctl(c->epollfd, EPOLL_CTL_ADD, pingf->sock, &ev) == -1) {
> +		warn_perror("L4 epoll_ctl");
> +		close(pingf->sock);
> +		goto cancel;
> +	}
> +

... this is uncomfortably bulky to have in every protocol.  Could we
maybe mitigate this with an epoll_add() helper of some sort that does
at least some of the bit shuffling to construct the epoll data?

-- 
David Gibson (he or they)	| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you, not the other way
				| around.
http://www.ozlabs.org/~dgibson