netdev - network unreliable on ReadyNAS 104 with Debian kernel

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-next>] [day] [month] [year] [list]

Date:   Sat, 2 May 2020 16:14:08 +0200
From:   Uwe Kleine-König <uwe@...ine-koenig.org>
To:     Arnaud Ebalard <arno@...isbad.org>,
        Thomas Petazzoni <thomas.petazzoni@...tlin.com>
Cc:     netdev@...r.kernel.org
Subject: network unreliable on ReadyNAS 104 with Debian kernel

Hello Arnaud, hello Thomas,

I own a ReadyNAS 104 (CPU: Armada 370, mvneta driver) and since some
time its network driver isn't reliable any more. I see things like:

	$ rsync -a remotehost:dir /srv/dir
	ssh_dispatch_run_fatal: Connection to $remoteaddress port 22: message authentication code incorrect
	rsync: connection unexpectedly closed (11350078 bytes received so far) [receiver]
	rsync error: error in rsync protocol data stream (code 12) at io.c(235) [receiver=3.1.3]
	rsync: connection unexpectedly closed (13675 bytes received so far) [generator]
	rsync error: unexplained error (code 255) at io.c(235) [generator=3.1.3]

when ever something like this happens, I get

	mvneta d0074000.ethernet eth1: bad rx status 0e8b0000 (overrun error), size=680

(with varying numbers after size=) in the kernel log.

With

	sudo ethtool -K eth1 tso off gso off gro off

the behaviour gets better, but I still get errors. In tcpdump I saw
packets received that are a mix of (at least) two other packets sent on
the remote side.

This happens with Debian's 5.4.0-4-armmp (Version: 5.4.19-1) kernel, but
I also experienced it with the 4.19 series. On slow connections this
isn't a problem so the problem might exist already longer. In fact I
think there are two problems: The first is that the hardware doesn't get
enough buffers in time for the receive path and the other is that in the
error case corrupted packets are given to the upper layers.

Does this ring a bell for you? I didn't start to debug that yet.

Best regards
Uwe

Download attachment "signature.asc" of type "application/pgp-signature" (489 bytes)