netdev - Re: [PATCH net] net: tcp: don't allocate fast clones for fastopen SYN

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CANn89i+hrgC042kvyLUQwk8PgXNM4woeQGEF+jn16hd6XCGRPg@mail.gmail.com>
Date:   Thu, 4 Mar 2021 21:18:15 +0100
From:   Eric Dumazet <edumazet@...gle.com>
To:     Jakub Kicinski <kuba@...nel.org>
Cc:     Alexander Duyck <alexander.duyck@...il.com>,
        David Miller <davem@...emloft.net>,
        netdev <netdev@...r.kernel.org>,
        kernel-team <kernel-team@...com>, Neil Spring <ntspring@...com>,
        Yuchung Cheng <ycheng@...gle.com>
Subject: Re: [PATCH net] net: tcp: don't allocate fast clones for fastopen SYN

On Thu, Mar 4, 2021 at 8:41 PM Eric Dumazet <edumazet@...gle.com> wrote:
>
> On Thu, Mar 4, 2021 at 8:06 PM Jakub Kicinski <kuba@...nel.org> wrote:
> >
> > On Thu, 4 Mar 2021 13:51:15 +0100 Eric Dumazet wrote:
> > > I think we are over thinking this really (especially if the fix needs
> > > a change in core networking or drivers)
> > >
> > > We can reuse TSQ logic to have a chance to recover when the clone is
> > > eventually freed.
> > > This will be more generic, not only for the SYN+data of FastOpen.
> > >
> > > Can you please test the following patch ?
> >
> > #7 - Eric comes up with something much better :)
> >
> >
> > But so far doesn't seem to quite do it, I'm looking but maybe you'll
> > know right away (FWIW testing a v5.6 backport but I don't think TSQ
> > changed?):
> >
> > On __tcp_retransmit_skb kretprobe:
> >
> > ==> Hit TFO case ret:-16 ca_state:0 skb:ffff888fdb4bac00!
> >
> > First hit:
> >         __tcp_retransmit_skb+1
> >         tcp_rcv_state_process+2488
> >         tcp_v6_do_rcv+405
> >         tcp_v6_rcv+2984
> >         ip6_protocol_deliver_rcu+180
> >         ip6_input_finish+17
> >
> > Successful hit:
> >         __tcp_retransmit_skb+1
> >         tcp_retransmit_skb+18
> >         tcp_retransmit_timer+716
> >         tcp_write_timer_handler+136
> >         tcp_write_timer+141
> >         call_timer_fn+43
> >
> >  skb:ffff888fdb4bac00 --- delay:51642us bytes_acked:1
>
>
> Humm maybe one of the conditions used in tcp_tsq_write() does not hold...
>
> if (tp->lost_out > tp->retrans_out &&
>     tp->snd_cwnd > tcp_packets_in_flight(tp)) {
>     tcp_mstamp_refresh(tp);
>     tcp_xmit_retransmit_queue(sk);
> }
>
> Maybe FastOpen case is 'special' and tp->lost_out is wrong.

It would be nice if tun driver would have the ability to delay TX
completions by N usecs,
so that packetdrill tests could be used.

It is probably not too hard to add such a feature.