netdev - Re: [PATCH 5/5] tcp: skip cwnd moderation in TCP_CA_Open in tcp_try_to

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: <CADVnQymPTWsD0Hx060qYUMrFPxgQPdJWwPA9oDce6qMVL4tVpA@mail.gmail.com>
Date:	Sat, 19 Nov 2011 16:08:40 -0500
From:	Neal Cardwell <ncardwell@...gle.com>
To:	Ilpo Järvinen <ilpo.jarvinen@...sinki.fi>
Cc:	David Miller <davem@...emloft.net>,
	Netdev <netdev@...r.kernel.org>,
	Nandita Dukkipati <nanditad@...gle.com>,
	Yuchung Cheng <ycheng@...gle.com>,
	Tom Herbert <therbert@...gle.com>
Subject: Re: [PATCH 5/5] tcp: skip cwnd moderation in TCP_CA_Open in tcp_try_to_open

On Thu, Nov 17, 2011 at 12:14 AM, Ilpo Järvinen
<ilpo.jarvinen@...sinki.fi> wrote:

> I think it's intentional. Because of receiver lying bandwidth cheats all
> unlimited undos are bit dangerous.

If your concern is receivers lying, then there are other easy ways
that a misbehaving receiver can get a sender to send too fast
(e.g. cumulatively ACKing the highest sequence number seen,
irrespective of holes). IMHO it would be a shame to penalize the vast
majority of well-behaved users to plug one potential attack vector
when there are other easy holes that an attacker would use.

> Wouldn't it be enough if tcp max burst is increased to match IW (iirc we
> had 3 still there as a magic number)?

Yes, tcp_max_burst() returns tp->reordering now. Changing it to return
max(tp->reordering, TCP_INIT_CWND) sounds good to me. I think that's
an excellent idea in any case, regardless of the outcome of this undo
discussion.

However, I don't think this is sufficient for request-response
protocols (e.g. RPC) running over long-lived connections over a path
with a large bandwidth-delay product. In such cases, the cwnd will
grow quite large (hundreds of packets). The DSACKs will often arrive
after the entire response is sent, so that cwnd moderation will
typically pull the cwnd down to 3 (or, with your proposal, 10)
packets. IMHO that seems like an unnecessarily steep price to pay.

neal
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html