netdev - Re: Socket buffer sizes with autotuning

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]

Date:	Fri, 16 May 2008 18:13:20 -0700
From:	"Jerry Chu" <hkchu@...gle.com>
To:	"David Miller" <davem@...emloft.net>
Cc:	netdev@...r.kernel.org
Subject: Re: Socket buffer sizes with autotuning

Ok, now the code seems to work when TSO is disabled, i.e., cwnd
remains low (40-60pkts) and only grows slowly for extended period on
a back2back 1GbE link. It also works when tcpdump was running
after I fixed the code, per your suggestion, to only bump the in_flight
count for fast clones.

But it doesn't work when TSO is enabled. I've fixed tso_fragment() to
correctly set skb_shinfo(buff)->in_flight after the split. But cwnd still
grows rapidly to a few hundred pkts (although smaller than w/o
the fix). The host inflight accounting gets screwed up. It looks like
pskb_expand_head() called by tcp_tso_acked()->tcp_trim_head()
messes up the accounting but I don't know how to fix it (still trying
to understand this complex piece of code). There could be other
reason as well.

Jerry

On Mon, May 12, 2008 at 9:02 PM, David Miller <davem@...emloft.net> wrote:
> From: "Jerry Chu" <hkchu@...gle.com>
> Date: Mon, 12 May 2008 21:00:08 -0700
>
>> Does xchg() return the old value? (Can't find any comments on it.)
>
> Yes, otherwise it wouldn't be very useful.
>
--
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