PPPoE and UDP fragmentation

[Gary Corcoran]

Charles Sprickman wrote:
> On Sat, 10 Sep 2005, Julian H. Stacey wrote:
> 
>>> -Are there any tunables at either end (both hosts are FreeBSD 4.11 
>>> p11) to
>>> alter how fragmented packets are re-assembled?
>>
>>
>> /usr/ports/net/tcpmssd
>> An MTU adapter.  Apparently not needed on FreeBSD-5 but I mean to
>> install it on my FreeBSD-4 DSL gateways when I find time to think if 
>> it might
>> have any implications re ipfw & security.
> 
> 
> I don't think that does anything to UDP, it just digs into tcp and 
> "fixes up" the MSS by altering it on outgoing packets.  I'm looking to 
> further understand UDP fragmentation and why a host might ignore 
> fragments, and who along the way is actually doing the fragmentation.
> 
> My current "fix" is just to set the interface MTU on the sending box to 
> 1492, and that works well, but I'd really like to understand why it 
> fails without that.

Do you know about the horribly-large overhead that PPPoE adds?
It's about 36 bytes, if I recall.  So to keep the *total* ethernet frame
size under the max limit of 1500-something (don't recall exact number),
you *always* have to limit *any* frame size, before PPPoE overhead,
to 1492 bytes.  Does this help?

Gary