[Asterisk-Dev] Transfer of variables to remote servers (new subject)

[Olle E. Johansson]

John Todd wrote:
> At 10:14 PM +0200 on 4/20/04, Olle E. Johansson wrote:
> 
>> John Todd wrote:
>>
>>> more significant and vexing problem of transferring values OUT of a 
>>> particular server to a remote server.
>>>
>>> I am uncertain how this should be created, actually.  How do you hand 
>>> things off in SIP?  How about IAX2?  MGCP?  Zap?(1) How do you read 
>>> them from the other side?  How do you refuse them?(2)  Can you get a 
>>> list of the attached values at the other end?(3)
>>
>>
>> I would suggest we concentrate on Asterisk to asterisk for this 
>> functionality.
>> Which suggest IAX2.
>> There need to be a "trust" concept - do we trust data from this 
>> server, or not?
>>
>>> This would be _TREMENDOUSLY_ powerful if it could be well designed 
>>> and had even basic functionality in SIP and IAX2.  Think of this very 
>>> small variable name example: "__caller-is-ceo"
>>
>> For SIP, I can't recall an existing protocol extension. There's a lot of
>> work going on in the 3G space so maybe there's functionality for this
>> to be found there - metadata on a call to be handled only between one
>> provider's trusted servers.

> 
> I don't see a need for "trust" in IAX2 or any other channel types, 
> actually.  That can be regulated by sending the calls from certain 
> "trusted" providers to different contexts, which have different values 
> of "trust" for incoming variables.  This is already gated on 
> username/password/IP address in the [channel].conf file, right?
> 
> Also, based on the contents of some "known" variables, we could do some 
> GotoIf booleans to move to more-or-less trusted sub-contexts or 
> extensions.  This is a kludge, if we're using it for "trust", but it 
> will undoubtedly be done depending on the style of the administrator. 
> This is essentially putting layers of "passwords" into call transfers, 
> which implies that we have end-to-end encryption (cough, cough.)  
> However, I am firmly of the belief that variable passing should not be 
> directly coupled to end-to-end crypto, so don't let that hold things 
> up...  We already have a method to negotiate trust between endpoints, 
> and that method supports multiple levels of security already.
Ok, agreed. We have a trust mechanism. Then we need support for it
in the protocol - does it exist or not? I'm not familiar with the
inner workings of IAX2...

/O