Content-Type: multipart/mixed; boundary="===============9190731012157914914==" MIME-Version: 1.0 From: Peter Krystad To: mptcp at lists.01.org Subject: Re: [MPTCP] [RFC 3/3] mptcp: Add handling of incoming MP_JOIN requests Date: Fri, 31 May 2019 05:56:51 -0700 Message-ID: In-Reply-To: 515713d691f3e3f9883149e86c293bcc9eb5652c.camel@redhat.com X-Status: X-Keywords: X-UID: 1271 --===============9190731012157914914== Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: quoted-printable On Wed, 2019-05-29 at 10:06 +0200, Paolo Abeni wrote: > Hi, > = > On Tue, 2019-05-28 at 16:22 -0700, Peter Krystad wrote: > > On Mon, 2019-05-27 at 16:52 +0200, Paolo Abeni wrote: > > > On Wed, 2019-05-22 at 14:38 -0700, Peter Krystad wrote: > > > > diff --git a/net/ipv4/tcp_minisocks.c b/net/ipv4/tcp_minisocks.c > > > > index 9c2a0d36fb20..baf8b975fce6 100644 > > > > --- a/net/ipv4/tcp_minisocks.c > > > > +++ b/net/ipv4/tcp_minisocks.c > > > > @@ -764,6 +764,12 @@ struct sock *tcp_check_req(struct sock *sk, st= ruct sk_buff *skb, > > > > if (!child) > > > > goto listen_overflow; > > > > = > > > > + if (own_req && sk_is_mptcp(child) && mptcp_sk_is_subflow(child)) { > > > > + inet_csk_reqsk_queue_drop(sk, req); > > > > + reqsk_queue_removed(&inet_csk(sk)->icsk_accept_queue, req); > > > > + return child; > > > > + } > > > > + > > > = > > > Don't we need to check this is not a fastopen before dropping the req? > > = > > I was thinking we would initially not support fastopen for MPTCP connec= tions, > > meaning we would ignore cookie requests in a SYN with MP_CAPABLE. I hav= en't > > added this check, what do others think about fastopen as a requirement = for > > MPTCP? > = > I'm sorry, I was not clear enough. I think fastopen is _NOT_ a > requirement for MPTCP. My doubt is: can we hit the above conditional > with a fast-open socket? (regardless of MPTCP code implementing support > for fast-open). If so, should we explicitly fail the MP_JOIN to avoid > messing with the tcp socket status? Thanks Paolo - I understand your point. But my thinking was it would be bet= ter to avoid worrying about fastopen sockets here by rejecting them (forcing th= em to fall back to 3-way handshake) when the initial SYN is received, this way if it is a MPTCP subflow socket it is NOT fastopen and the above clause wou= ld never be hit. Peter. > Not sure if the above is now more clear ;) > = > > > Can't we avoid the above hunck anyway, handling the special MP_JOIN > > > scenario in mptcp_accept() ? so that we reduce hooking, the user-space > > > is notified of a new subflow - 'accept()' would probably return a > > > specific error code, not a new fd!!! - and we enforce some limits on > > > the number of pending/new subflows. > > = > > I was not thinking of forcing a change to the typical user accept() > > handling so this is how that is avoided. In earlier discussions it seem= ed > > making the user use a new protocol value was enough of a hurdle. > = > Ok, understood. I agree that we should avoid uAPI changes; before I did > not consider such aspect. I'm fine with the current impl in this > regard. > = > Thanks, > = > Paolo > = --===============9190731012157914914==--