Re: [Ntp] Antw: [EXT] Re: Antwort: Re: Symmetric mode

[Ulrich Windl <[e-mail]>]

>>> Miroslav Lichvar <[e-mail]> schrieb am 22.09.2022 um 12:40 in
Nachricht <Yyw7pYO9rlvgUB/I@localhost>:
> On Thu, Sep 22, 2022 at 10:11:43AM +0200, Ulrich Windl wrote:
>> >>> Miroslav Lichvar <[e-mail]> schrieb am 22.09.2022 um 09:40 in
>> Nachricht <YywRhCmTD4mt8KTy@localhost>:
>> > The IP addresses and ports are not included in the data authenticated
>> > by the NTP MAC. You can replay an authenticated NTP messages from any
>> > address and any port.
>> 
>> But this problem isn't specific to peer messages, right? So what is the
>> specific peer issue?
> 
> With the client/server mode you can replay packets, but there is no
> interesting impact.
> 
> Server is stateless and responds to all messages from all addresses
> and ports. Client accepts only the first response from the configured
> address and port that has the an origin timestamp matching its
> request. It ignores replays.
> 
> The problem with the symmetric mode is that does two things at once.
> A message is a request and response at the same time. The server side
> has a state associated with each client and there is only one response
> per polling interval. That causes two issues.
> 
> One is that replays can break or prevent synchronization in a
> symmetric association. If the peer receives multiple requests before
> its next poll, it has to select one to which it will respond. If the
> association is just starting (or restarting after packet loss) there
> is no request that would have a matching origin timestamp from the
> peer's previous request. The best it can do is to select the one with
> the latest transmit timestamp, hoping that no message was previously
> sent with the clock in future or distant past (e.g. due to missing
> RTC).

So it seems some OTP (on-time-password) mechanism would be needed for peers to kill the replay problem. However with lost packets that can be a bit tricky.

> 
> The other issue is replays from different addresses or ports can
> create a large number of associations and cause a denial of service.

As written before, the requester (sender) should be part of the authenticated message. So at least the associations created would be limited to one, right?

Regards,
Ulrich