netdev - Re: [PATCH net-next] tcp: avoid spurious loopback retransmit

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+BO+oPw1-SJNutGEHRzUfHBH7QvFaMqDepZwYQ1W5RXg@mail.gmail.com>
Date:   Wed, 9 Jun 2021 18:46:22 +0200
From:   Eric Dumazet <edumazet@...gle.com>
To:     "dust.li" <dust.li@...ux.alibaba.com>
Cc:     David Miller <davem@...emloft.net>,
        netdev <netdev@...r.kernel.org>,
        Tony Lu <tonylu@...ux.alibaba.com>,
        Xuan Zhuo <xuanzhuo@...ux.alibaba.com>
Subject: Re: [PATCH net-next] tcp: avoid spurious loopback retransmit

On Wed, Jun 9, 2021 at 10:45 AM dust.li <dust.li@...ux.alibaba.com> wrote:
>
> On Wed, Jun 09, 2021 at 09:03:14AM +0200, Eric Dumazet wrote:
> >On Wed, Jun 9, 2021 at 2:25 AM dust.li <dust.li@...ux.alibaba.com> wrote:
> >>
> >
> >> Normal RTO and fast retransmits are rarely triggerred.
> >> But for TLP timers, it is easy since its timeout is usally only 2ms.
> >>
> >
> >OK, by definition rtx timers can fire too early, so I think we will
> >leave the code as it is.
> >(ie not try to do special things for 'special' interfaces like loopback)
> >
> >We want to be generic as much as possible.
> Totally understand!
>
> After talking to you, I also rethinked this a bit more.
> The original patch is bad and not generic, my original intention is
> also to discuss with the community.
>
> Through the patch is bad, I still think the problem is generic.
> Devices like loopback/veth/ifb and maybe some others as well,
> who depend on netif_rx() or tasklet to receive packets from CPU
> backlog should all have this problem.

No, you can also have spurious retransmits when using regular/standard NIC

The receiver can have delays/jitter in processing incoming packets,
either because of a sudden spike in networking activity, or because
BH handling had to be deferred for various reasons.

>
> But I really didn't find a general way to gracefully solve this.

We simply live with this, really.