netdev - Re: [PATCH v2 net-next] tcp: remove a bogus TSO split

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, 13 Dec 2013 10:38:46 -0800
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Neal Cardwell <ncardwell@...gle.com>
Cc:	David Miller <davem@...emloft.net>,
	netdev <netdev@...r.kernel.org>,
	Yuchung Cheng <ycheng@...gle.com>,
	Nandita Dukkipati <nanditad@...gle.com>,
	Van Jacobson <vanj@...gle.com>
Subject: Re: [PATCH v2 net-next] tcp: remove a bogus TSO split

On Fri, 2013-12-13 at 13:17 -0500, Neal Cardwell wrote:

> It occurred to me after staring at this code a little longer that
> although the Nagle test is already done before the call to
> tcp_mss_split_point(), the tcp_nagle_check() is only run if
> tso_segs==1 and is only checking whether skb->len < mss_now, so the
> Nagle code currently is implicitly assuming that if there is an skb
> that is not an exact multiple of MSS, then the tcp_mss_split_point()
> will chop off the odd bytes at the end and we'll loop back to the top
> of the tcp_write_xmit() loop to make a Nagle decision on an skb that
> has tso_segs==1.
> 
> So if we just make this improvement to tcp_mss_split_point() and the
> adjustment to tcp_minshall_update() (as in patch v2), we will still
> have broken Nagle behavior in a scenario like:
> 
> - suppose MSS=1460 and Nagle is enabled (TCP_NODELAY is not set)
> 
> - suppose there is an outstanding un-ACKed small packet, so that
> tcp_minshall_check() returns true
> 
> - user writes 2000 bytes
> 
> - tso_segs is 2, so we do not call tcp_nagle_test()
> 
> - new version of tcp_mss_split_point() sends out the full 2000 bytes,
>   leading to having 2 packets smaller than an MSS un-ACKed in the network,
>   violating the invariant the minshall/nagle code is trying to maintain
>   or having only one such packet un-ACKed in the network.
> 
> Previously, tcp_mss_split_point() would have split off the 2000-1460
> bytes into a new skb, we would have looped back and executed the
> tcp_nagle_test() code and decided not to send that small sub-MSS
> 2000-1460 packet.
> 

Hmm, it seems we need to refactor a bit.

tcp_snd_test() seems to not care of TSO being partial.



--
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