Re: Is Fragmentation at IP layer even needed ?

[Masataka Ohta <[e-mail]>]

Fernando Gont wrote:

>> Moreover, the rfc should also limit header chain length below
>> 256B or so.
> 
> I tried that in 6man, but there was opposition to my proposal at the time.

OK, collective stupidity of IPv6 committee is same as I tried
it about 20 years ago.

>> Worse, though the rfc require an entire upper layer header
>> included in the first fragment, the requirement is too much.
> 
> Not sure what you mean.

After the last mail, I have noticed that this is a serious defect
of the rfc.

That is, if an upper layer protocol is not known to some box,
the box has no idea on how long is the header, which
means the box is now authorized to discard the packet, even if
actual header length is small and no extension header exist.

One thing we should learn from the glorious history of IPv4
is meaning of specification in rfc792 that:

  |      Internet Header + 64 bits of Original Data Datagram      |

which makes ICMP implementation independent from upper
layer header structure and requires upper layer specifications
contain port number (or something like that) information in
the first 8B of their headers.

>> According to rfc792, the first 8B should be enough. As for
>> ICMPv6, ICMPv6 should also be required to contain all the
>> header chain up to the first 8B of an upper layer header.
> 
> ... which may not be feasible if a long EH chain is employed in the
> packet that generated the error message.

That's one of a reason why long EH must be banned.

						Masataka Ohta