Still MTU problem, need help.

[felata]

I setup ipv6 tunnel he.net with freebsd.

ICMP to network inside he.net without problem.
# ping6 -s 1433 ordns.he.net
PING6(1481=40+8+1433 bytes) 2001:470:c:6da::2 --> 2001:470:20::2
1441 bytes from 2001:470:20::2, icmp_seq=0 hlim=64 time=69.210 ms

ICMP to outside he.net with packet larger than MTU is fail.
# ping6 -s 1433 google.com
PING6(1481=40+8+1433 bytes) 2001:470:c:6da::2 --> 2607:f8b0:400c:c04::8a
--- google.com ping6 statistics ---
3 packets transmitted, 0 packets received, 100.0% packet loss

Only 1232 packet size is ok for outside he.net
# ping6 -c 1 -s 1232 google.com
PING6(1280=40+8+1232 bytes) 2001:470:c:6da::2 --> 2607:f8b0:400c:c04::8a
72 bytes from 2607:f8b0:400c:c04::8a, icmp_seq=0 hlim=52 time=132.133 ms


I also did some test for difference MTU settings.

1. MTU set to 1480 in both tunnelbroker and freebsd, (tunnelbroker is 1480 for default). Fail.
2. MTU set to 1280 in both tunnelbroker and freebsd. Fail
3. MTU set to 1480 in tunnelbroker, 1280 in freebsd. Fail.

Any advice for MTU setting?

[broquea]

What is the actual problem you are trying to solve, regarding mtu? What made you want to start fiddling with it? I mean if you are PPPoE, you want 1472 on both sides.

[felata]

No, I don't use PPPoE.
Big packets only works in he.net network. That's the problem.

[GaryGapinski]

I have a tunnel with MTU 1480 and advertised (ICMPv6 RA) MTU 1480 on LAN.

If you packet capture the traffic, you will see that the origin host will fragment when the -s argument to ping6 is > MTU-48 (1432 in the case of MTU 1480). I expect fragmented ICMPv6 is considered worthy of discard by most.

I think it best to continue using MTU 1480 and avoid fragmentation.