Re: Modified IPv6 to unmodified IPv4

[Iljitsch van Beijnum <iljitsch@muada.com>]

On 19-okt-2007, at 10:05, Iljitsch van Beijnum wrote:

[...]

And now let's see if we can make this stateless.

The reason NATs require state is because clients can choose random  
source port numbers which may clash. So we give each client a small  
identifier that determines the range of source port numbers they may  
use. I.e., ID = 1 - 127, source port is 1 + ID bits + 8 bits chosen  
by the client. This limits the number of clients per translator to  
127. Solution: give the translator more IPv4 addresses and encode  
them like this: 0:0:0:0:<translator IPv4 address>:<destination IPv4  
address>. We then use a range of 65536 fake IPv4 private addresses,  
where the IPv6 host selects one based on its IPv6 address such that  
the TCP and UDP checksums are the same for the original IPv6 address  
and the translated and NATed IPv4 address. Only problem is telling  
both sides about the combination of translator IPv4 address, port ID  
bits and client IPv6 address. This could be done with DHCP for the  
client side. So there is SOME state involved for the translator, but  
this is long-term per-client state rather than ephemeral per-session  
state so it may be worth the effort.