Re: [PATCH mptcp-next v4 2/2] mptcp: implement connection level timeout.

[Paolo Abeni <pabeni@redhat.com>]

On Fri, 2023-09-01 at 17:23 -0700, Mat Martineau wrote:
> On Fri, 1 Sep 2023, Paolo Abeni wrote:
> > @@ -2635,7 +2640,7 @@ void mptcp_reset_tout_timer(struct mptcp_sock *msk, unsigned long fail_tout)
> > 	struct sock *sk = (struct sock *)msk;
> > 	unsigned long timeout, close_timeout;
> > 
> > -	if (!fail_tout && !sock_flag(sk, SOCK_DEAD))
> > +	if (!fail_tout && !inet_csk(sk)->icsk_mtup.probe_timestamp)
> > 		return;
> > 
> > 	close_timeout = inet_csk(sk)->icsk_mtup.probe_timestamp - tcp_jiffies32 + jiffies + TCP_TIMEWAIT_LEN;
> 
> RFC 8684 brings up "break-before-make" scenarios in a few places, I think 
> one use case is for VM migrations. 60 seconds seems sufficient but maybe 
> not?
> 
> There may also be scenarios with busy servers where the server-side admin 
> may want to quickly close "zero subflow" MPTCP connections immediately (or 
> with a shorter timeout) if their workload does not involve 
> break-before-make and it's more important to free up resources quickly.
> 
> Given these examples, what do you think about making this timeout 
> configurable via sysctl and/or sockopt?

Yes, the idea is to add an additional sysctl. I guess we could add even
a sockopt (MPTCP_USER_TIMEOUT) to cap the above.

I did not include that here, because I intended to keep this change
small, and the sysctl would add a bit more diffstat noise.

Cheers,

Paolo