Re: [RFC] "CIPE ethernet devices should have an MTU of 1500"|
Olaf Titz <olaf,AT,bigred,DOT,inka,DOT,de>|
Tue, 6 Nov 2001 21:15:38 +0100|
> group, the CIPE device will have to 'emulate' an MTU of 1500, since a lot of
> things downright depend on this. IMHO, the fact that this causes
> fragmentation at the UDP level is secondary. For a vanilla IP tunnel I'll
> buy the argument that not fragmenting the encrypted-and-encapsulated
> packet is important, but in the case of ethernet, I think the MTU guarantee
> is sacred.
> I'm using the attached (shoddy) patch for this. If people agree with me on
> this I'll make a clean patch, conditional on the protocol version et al.
That needs a configuration option, IMHO, since it is one of the things
people might legitimately want either way. I vaguely remember that the
"reset DF" issue has already come up as a wanted config item. In the
other place where you hardcode 1500 you could just use dev->mtu which
already is configurable.
So to get this right we'd need two new boolean options:
ignoredf - ignore DF bit in payload, allow the carrier to fragment
forcemtu - ignore cached path MTU, use device MTU
Set both of these to true and set mtu=1500 to achieve full Ethernet
This can be implemented using bits in the "flags" field of struct
cipe, struct siocsifcippar; along with CIPF_MAY_CLEAR etc.