Andi Kleen wrote:
>On Fri, Oct 06, 2000 at 09:06:31PM +0000, David Wagner wrote:
>> David S. Miller wrote:
>> >Linux should not honor the incorrect sequence number. If the sequence
>> >number is incorrect, the RST could legitimately be for another
>> >connection.
>>
>> How could it be for another connection, if it has source and destination
>> port numbers?
>
>You're missing dynamic IPs, NAT and reboot of hosts.
Ok, but with dynamic IP, NAT, and reboots, you can also get incorrect
results with today's implementation, if you see a RST with the correct
sequence number (not the off-by-one incorrect one). It's just a
probabilistic argument that this is unlikely to happen in practice --
namely, it only happens with probability 1/2^32 (you hope).
There's no fundamental reason why you couldn't accept off-by-one sequence
numbers as well, if it was deemed important for interoperability;
the probability would just increase to 1/2^31, which is still small
(albeit not as small as 1/2^32).
Right? Or am I still missing something? I'm not a TCP expert.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Fri Oct 06 2000 - 17:36:52 EDT