Re: [radext] Review of draft-ietf-radext-bigger-packets-02.txt

[Alan DeKok <[e-mail]>]

On Feb 26, 2015, at 8:58 AM, Alejandro Perez Mendez <[e-mail]> wrote:
> 13. Section 4. It is not clear to me what's the purpose of the Original-Packet-Code attribute. It is mentioned in the text, and described which values it should take. However, it is not explained how this value is used.

  The problem is that there are multiple packet types (Access-Request, Accounting-Request, etc.) which could theoretically trigger a Protocol-Error response.  We can’t send a normal response (Access-Reject, Accounting-Response), because the error is in the *transport* stage.  So we have to send a Protocol-Error packet.

  But… we don’t want multiple protocol error codes.  e.g. Access-Request-Protocol-Error, Accounting-Request-Protocol-Error, etc.  We just want Protocol-Error.

  The problem then becomes how do you match Protocol-Error to an original packet?  RADIUS/TCP allows for Access-Request and Accounting-Request packets to share the same TCP connection.  So if the client receives Protocol-Error with ID 34, is it for Access-Request ID 34, or Accounting-Request ID 34?

  Putting the Original-Packet-Code into Protocol-Error solves this problem.  The client can now look up the original packet code (Access-Request or Accounting-Request), along with the ID (34), and get the original packet.

  Alan DeKok.