netdev - Re: [PATCH net] sctp: fix recovering from 0 win with small data chunks

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] [day] [month] [year] [list]

Date:   Fri, 23 Dec 2016 14:02:09 -0500 (EST)
From:   David Miller <davem@...emloft.net>
To:     marcelo.leitner@...il.com
Cc:     netdev@...r.kernel.org, linux-sctp@...r.kernel.org,
        nhorman@...driver.com, vyasevich@...il.com
Subject: Re: [PATCH net] sctp: fix recovering from 0 win with small data
 chunks

From: Marcelo Ricardo Leitner <marcelo.leitner@...il.com>
Date: Fri, 23 Dec 2016 14:29:37 -0200

> Currently if SCTP closes the receive window with window pressure, mostly
> caused by excessive skb overhead on payload/overheads ratio, SCTP will
> close the window abruptly while saving the delta on rwnd_press. It will
> start recovering rwnd as the chunks are consumed by the application and
> the rwnd_press will be only recovered after rwnd reach the same value as
> of rwnd_press, mostly to prevent silly window syndrome.
> 
> Thing is, this is very inefficient with small data chunks, as with those
> it will never reach back that value, and thus it will never recover from
> such pressure. This means that we will not issue window updates when
> recovering from 0 window and will rely on a sender retransmit to notice
> it.
> 
> The fix here is to remove such threshold, as no value is good enough: it
> depends on the (avg) chunk sizes being used.
> 
> Test with netperf -t SCTP_STREAM -- -m 1, and trigger 0 window by
> sending SIGSTOP to netserver, sleep 1.2, and SIGCONT.
> Rate limited to 845kbps, for visibility. Capture done at netserver side.
 ...
> Signed-off-by: Marcelo Ricardo Leitner <marcelo.leitner@...il.com>

Applied.