netdev - Re: [PATCH] gianfar: prevent fragmentation in DSA environments

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: <20160819214503.GC14068@lunn.ch>
Date:   Fri, 19 Aug 2016 23:45:03 +0200
From:   Andrew Lunn <andrew@...n.ch>
To:     Claudiu Manoil <claudiu.manoil@....com>
Cc:     Zefir Kurtisi <zefir.kurtisi@...atec.com>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "claudiu.manoil@...escale.com" <claudiu.manoil@...escale.com>
Subject: Re: [PATCH] gianfar: prevent fragmentation in DSA environments

> Nice improvement.

Thanks

> But what's so special about 8?

When talking to Marvell Switch chips, there are two different headers
which can be added. The DSA header is 4 bytes, and the EDSA header is
8 bytes. However, if there is already a VLAN header, when using EDSA,
it will replace the VLAN header, so only need 4 additional bytes.
There are some very old Marvell switches which add a 4 byte
trailer. If you are talking to a Broadcom switch chip, it also has a 4
byte header.

> I thought only 4 bytes were missing :)

So the requirement is probably not currently for the newer Marvell
switch chips? But we should be looking forward and expect at some
point somebody wants to use the newer chips. I've got a Freescale
Vybrid board using the modern Marvell chips, but the FEC driver does
not have a such a hard limit as this driver.

However, it does not seem as simple as that. A standard Ethernet frame
should have a maximum size of 1522 when including a VLAN header. Yet
the driver appears to be using 1536, which is this rounded up to
multiples of 64. So there is already 14 spare bytes in there. So there
must be something else going on here.

> At least 1536 is the default size of the MRBLR register, as specified
> in the h/w ref manual.  Is there some recommended standard size
> to accommodate most (if not all) headers, to refer to?

Not that i know of. These switch headers are proprietary.

    Andrew