netdev - Re: [PATCH v1 2/2] tcp: Ignore OOO handling for TCP ACKs

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: <CANn89iJ2AWUehsEZo+L_G6cdKaEdwdMVfZr6SYnRFun3L2ksSg@mail.gmail.com>
Date:   Thu, 20 Oct 2022 14:47:34 -0700
From:   Eric Dumazet <edumazet@...gle.com>
To:     Kamaljit Singh <Kamaljit.Singh1@....com>
Cc:     Niklas Cassel <Niklas.Cassel@....com>,
        "davem@...emloft.net" <davem@...emloft.net>,
        Damien Le Moal <Damien.LeMoal@....com>,
        "dsahern@...nel.org" <dsahern@...nel.org>,
        "yoshfuji@...ux-ipv6.org" <yoshfuji@...ux-ipv6.org>,
        "kuba@...nel.org" <kuba@...nel.org>,
        "pabeni@...hat.com" <pabeni@...hat.com>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: [PATCH v1 2/2] tcp: Ignore OOO handling for TCP ACKs

On Thu, Oct 20, 2022 at 2:25 PM Kamaljit Singh <Kamaljit.Singh1@....com> wrote:
>
> On Thu, 2022-10-20 at 11:57 -0700, Eric Dumazet wrote:
> > CAUTION: This email originated from outside of Western Digital. Do not click
> > on links or open attachments unless you recognize the sender and know that the
> > content is safe.
> >
> >
> > On Thu, Oct 20, 2022 at 11:22 AM Kamaljit Singh <kamaljit.singh1@....com>
> > wrote:
> > > Even with the TCP window fix to tcp_acceptable_seq(), occasional
> > > out-of-order host ACKs were still seen under heavy write workloads thus
> > > Impacting performance.  By removing the OoO optionality for ACKs in
> > > __tcp_transmit_skb() that issue seems to be fixed as well.
> >
> > This is highly suspect/bogus.
> >
> >  Please give which driver is used here.
> The NVMe/TCP Host driver (also mentioned in the cover letter).
>

This is code located on the same linux host ?

So... this is loopback interface ?

Which ndo_start_xmit() is called ?

>
> >
> > > Signed-off-by: Kamaljit Singh <kamaljit.singh1@....com>
> > > ---
> > >  net/ipv4/tcp_output.c | 5 ++++-
> > >  1 file changed, 4 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/net/ipv4/tcp_output.c b/net/ipv4/tcp_output.c
> > > index 322e061edb72..1cd77493f32c 100644
> > > --- a/net/ipv4/tcp_output.c
> > > +++ b/net/ipv4/tcp_output.c
> > > @@ -1307,7 +1307,10 @@ static int __tcp_transmit_skb(struct sock *sk, struct
> > > sk_buff *skb,
> > >          * TODO: Ideally, in-flight pure ACK packets should not matter here.
> > >          * One way to get this would be to set skb->truesize = 2 on them.
> > >          */
> > > -       skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
> > > +       if (likely(tcb->tcp_flags & TCPHDR_ACK))
> > > +               skb->ooo_okay = 0;
> > > +       else
> > > +               skb->ooo_okay = sk_wmem_alloc_get(sk) < SKB_TRUESIZE(1);
> > >
> >
> > This is absolutely wrong and would impact performance quite a lot.
> >
> > You are basically removing all possibilities for ackets of a TCP flow
> > to be directed to a new queue, say if use thread has migrated to
> > another cpu.
> Are you suggesting that the proposed change not be done at all or done in a
> different way? We did see an observed performance improvement in NVMe/TCP
> traffic with this fix. If you have an alternative idea I'd be happy to try &
> test it out.

Well, you disable a very important feature, just to work around some
problem in another layer.

Let's investigate what can be done in this other layer, once it is identified.


>
>
> >
> > After 3WHS, all packets get ACK set.
> --
> Thanks,
> Kamaljit Singh <kamaljit.singh1@....com>