On 05/31/10 11:03 AM, Vishwas Manral wrote:
Hi Erik,
The first thing is IPv6 MTU discovery is optional and the only
requirement is that the Minimum MTU is satisfied.
Yes, but 1280 plus a routing header will exceed 1280.
Thus the minimum MTU by itself doesn't save you.
Also in cases like LLN the Routing paths may change so even after
discovery the Packet too big may come anyway. Besides PMTU issues you
mention are no different than a lot of cases such as tunnelling and
transition mechanisms.
There is a very big difference.
Tunneling means that the entity which added the tunneling headers is in
the source IP address field in the outer header. Thus ICMP errors will
be sent packet to the tunnel entry point. It can then compensate for the
added headers by sending an ICMP packet too big with a smaller MTU back
to the original source. This how IPvx in IPvy tunneling is done. See for
instance RFC 4213 and RFC 2473.
The big difference is that when a router does something to grow the size
of the packet (for instance, inserting a routing header) and it does not
add an outer IP header with its source address, then the ICMP errors
will go back to the original sender without being adjusted for the added
header.
For MTU discovery in cases when padding is done, the right MTU will be
found as the data packets will undergo the same behavior as the PMTU
discovery message.
Huh?
The ICMP "packet too big" error contains a MTU size. There is nothing in
the normal IP forwarding path that changes that number.
Erik
Thanks,
Vishwas
On Sun, May 30, 2010 at 7:56 AM, Erik Nordmark<[email protected]> wrote:
The draft says:
A RPL Router MAY insert a Type 4 Routing header if one does not
already exist. The conditions for inserting a Type 4 Routing header
are out of scope of this document.
Having routers insert headers in packets they forward is know to be
problematic in general, since it breaks path MTU discovery.
Taking an example of a host which sends a packet which is 1500 bytes, and
then receives an ICMP packet too big error saying "please limit packets to
1500 bytes", will not change the host's behavior. But that can happen if
some router grows the packet to be more than 1500 bytes.
Thus in general, the only safe way to insert headers in a packet is to have
the "inserter" put an extra IP header, with itself as a source, then the
inserted header, and then the original packet.
That extra IP header ensures that any ICMP error go back to the "inserter",
which can then compensate for its insertion before sending the ICMP error
towards the sender of the original packet.
Thus I think there is a question for 6man whether we think the constrained
environment of ROLL is such that we can relax this. But it does require that
all the routers at the edge of the ROLL network adjust the ICMP errors they
forward to compensate for any inserted RH4 headers.
Erik
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
--------------------------------------------------------------------
IETF IPv6 working group mailing list
[email protected]
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------
