Contrib - Improving NAT/PAT Bypass + DSCP on Control Session

[froudaut]

Trying to provide solutions to improve NAT/PAT Bypass :

TWAMP, twping :

• option [Z] was already implemented to set Session Sender/Reflector IP to Zero Addr in Twamp
  Requests from Control-Client. In that case, Control-Client is also Session-Sender and
  Control-Server is Session-Reflector. Control-Server had then to switch the Zero Addr with
  the corresponding Control IP Addr.
  Such option is still available but now also split in 2 options : [X] For Session-Sender,
  [Y] For Session-Reflector.
  It helps to bypass NAT on both Session Sender/Reflector when they are also Control Client/Server.
• Option [y] to handle PAT from the Session-Reflector. In that case, Session-Reflector Port
  is not checked.

OWAMP, owping :

• Option [X], [Y] has been added as an extension to OWAMP RFC when Session Sender/Receiver are
  also Control Client/Server.
  As for TWAMP, it permits to set Session Sender/Receiver IP to Zero Addr in Owamp Requests
  from Control-Client : [X] For Session-Sender, [Y] For Session-Server.
  In that case Control-Server had to switch the Zero Addr with the corresponding Session IP Addr
  according the way test is done. For example when tests are done in both ways using Owamp, option [X]
  tells the Control-Server to use the zero address for Control-Client in test session from Control-Client
  to Control-Server and for Control-Server in test session from Control-Server to Control-client
  It helps to bypass NAT on both Session Sender/Receiver.
• Option [y] to handle PAT from the Session-Remote. In that case, Session-Remote Port
  is not checked.

twampd :

• Option [W] : has to be used now to handle Zero Addr in Twamp Request when Session Sender/Reflector
  are also Control Client/Server. Control-Server had then to switch the Zero Addr with
  the corresponding Control IP Addr.
  It helps to bypass NAT on both Session Sender/Reflector when they are also Control Client/Server.
• Option [x] to handle PAT from the Session-Sender. In that case, Session-Sender Port
  is not checked. Indeed, even with Zero Address, PAT on Session-Sender was not considered.
• Option [Y] : Such option may be used if Control-Client is not aware of NAT on Session-Reflector Side and/or
  does not use zero address. In that case any Session-Reflector Address is seen as Local Address.

owampd :

• Option [W] has been added as an extension to OWAMP RFC when Session Sender/Receiver are
  also Control Client/Server.
  As for TWAMP, it permits to handle Zero Addr for Session Sender/Receiver in Owamp Requests
  from Control-Client
  In that case Control-Server had to switch the Zero Addr with the corresponding Session IP Addr
  according the way test is done.
  It helps to bypass NAT on both Session Sender/Receiver.
• Option [x] to handle PAT from the Session-Remote. In that case, Session-Remote Port
  is not checked. Indeed, even with Zero Address, PAT on Session-Remote was not considered.
• Option [Y] : Such option may be used if Control-Client is not aware of NAT on Session-Local Side and/or
  does not use zero address. In that case any Session-Local Address is seen as Local Address.

DSCP On Control Session:

owping/twping

Add an option [m] to set DSCP field on Control packets (from Control Client to Server).
To reflect ToS received from Control-Client, on Linux we may have to set 1 into
/proc/sys/net/ipv4/tcp_reflect_tos on Control-Server side

[mfeit-internet2]

Thanks for your contribution. We're currently in the middle of our conference and holiday season and will have a look at your proposed changes early next year.

Not having reviewed the changes yet, a couple of general comments: This software is considered the reference implementation for OWAMP, so we have to tread carefully. Our primary concerns will be not breaking existing behavior, interoperability with versions of the software that don't have this patch and changes to the protocol which, unfortunately, doesn't have provisions for versioning.

[froudaut]

Hi .
for sure. In fact I use a similar patch from 2017, even before you add -Z option to bypass NAT.
I used this against various implementation: Cisco, Juniper, ..., without some issues.
I decided to get your recent fixes in the last months and I had then to reapply my own patch . This is why I now do this push request; to avoid to do this again later ;-)