On 2016-01-03 11:44, Guus Sliepen wrote:
Hm, that's also true for IPv4, so this maybe isn't IPv6-specific.
Although I don't think there will be many cases where the IPv4 MTU
would
be lower than the IPv6 one.
Agreed, IPv4 over IPv6 is theoretically possible but very unlikely.
Hm, this is the first time I've heard about a separate MTU setting for
IPv6.
Didn't know it either until I read it on radvd dev list [1]. Apparently
this is for IPv6 only, there is no protocol specific MTU setting for
IPv4.
Ok, but the SixXS tunnel itself doesn't transport IPv4 packets, so the
current behaviour is fine for configuring the tunnel itself.
If you have a host on a LAN with a gateway machine that has a SixXS
tunnel, then you shouldn't have to lower the MTU on that host, since
the
gateway machine should send the appropriate ICMPv6 packets to the host
if the host sends packets that are too large. If that doesn't work,
check if you have firewall rules on either machine blocking ICMPv6
packets.
In theory yes. Unfortunately not every application works well with path
discovery. In my case the problem is with OpenVPN, it uses its own MTU
values and doesn't do any path discovery [2]. I worked around this by
setting the mssfix option [3], however I would like to fix it properly
hence my attempt at setting the MTU on the interface.
For IPv4 it gets even more interesting. When configuring the interface
to a MTU of 1480 this should work fine for IPv4. Unfortunately it seems
that the SMTP servers of least one large bank and one large ISP can't
handle any MTU below 1500, causing incoming mail to drop with symptoms
similar to MTU issues [4] for some, but not all connections. So I was
forced to set the MTU on the interface back to 1500.
With "auto", ifupdown doesn't set mtu itself. It has several different
methods to configure the interface in this case. Can you tell me if you
are using RDNSS, DHCPv6 or SLAAC to configure it?
Before using static I was using SLAAC, I think it was taking the MTU
from
the RA broadcast then, at least [1] suggests that that is what is is
supposed to do.
I'll see if this can be implemented easily in ifupdown.
That would be really great. I can of course use a 'post' entry to call
sysctl myself, but a clean fix would be nicer.
Laurens
[1]
http://lists.litech.org/pipermail/radvd-devel-l/2008-March/000339.html
[2] http://michael.stapelberg.de/Artikel/mtu_openvpn/
[3]
https://openvpn.net/index.php/open-source/faq/77-server/271-i-can-ping-through-the-tunnel-but-any-real-work-causes-it-to-lock-up-is-this-an-mtu-problem.html
[4]
https://www.howtoforge.com/community/threads/postfix-timeout-after-data-from-some-emails.39635/
