Re: SOLVED: kernel-mode PPPoE does not seem able to work with MPPE.

[David Fernandez <david.fernandez.work@googlemail.com>]

On 06/11/17 18:31, James Carlson wrote:
> On 11/06/17 09:59, David Fernandez wrote:
>> Yes, the code could be deleted, I just wanted somebody to check and say
>> that there is no point in yet putting a warning in the log for this...
> RFC3078 7.2 says that the receiver changes its key on receiving a "flag"
> packet (where (count&0xff) = 0xff).  It says nothing about the
> transmitter setting the FLUSHED flag in this case.  That flag is used
> *only* when a sequence error is detected.
>
> So I think the original code was wrong (it should not have insisted on
> the flag being set), and the new code without the message and without
> the error handling is correct.
>
> It _might_ be a good idea to check and at least comment the sending
> code.  It should not be setting the FLUSHED flag merely because it is
> sending a flag packet ... but, for compatibility with broken Linux MPPE
> implementations already deployed, it might have to continue doing that.
>
> It's a shame documents like this don't go through WG review with
> multiple independent implementations.  This is the sort of spec
> misinterpretation that a good open review is designed to catch.
>
Hi James,

I think you are right, the flushed bit is only used for sequence errors 
in stateful mode, as that is the only way to change keys in a 
deterministic manner.

I'm runing now into the problem of packet loss, and I guess I'll have to 
modify the implementation in order to use the flushed bit instead of a 
reset-ack packet to decide if we have to send another reset-req after a 
timeout...

I think this will become a formal patch, so let me know what ettiquete 
bits are required and I'll prepare it for review and submission to 
whomever can approve and commit.

Regards