Re: [regext] FW: New Version Notification for draft-gould-regext-secure-authinfo-transfer-00.txt

["Patrick Mevzek" <[e-mail]>]

Hello James,

On Mon, Jul 8, 2019, at 14:16, Gould, James wrote:
> JG - The draft is a Best Current Practice (BCP) per RFC 2026, and not a 
> standards track draft.  The draft describes how to leverage the 
> existing EPP RFCs for addressing the security of the authorization 
> information value for transfers.  EPP can have protocol extensions 
> defined as informational and standards track drafts, as well as 
> operational practices defined as BCP drafts.  There are many examples 
> of IETF BCPs.  This topic is very applicable to the IETF and the REGEXT 
> working group in particular.

I will remain in disagreement here (mandating how registries should store passwords
or choose them regarding length and complexity
is certainly a bigger issue than just EPP and has nothing to do regarding how EPP
works as an exchange protocol between 2 entities), so I will only reply briefly as my
contributions will not help whatsoever building this draft and try
to refrain from participating in any future LC regarding this draft.

I would also suggest or offer the idea that various points in the draft
(like "the authorization information .... MUST NOT be stored by the registrar.")
do not align (which means: will never happen) with various registrars policies
or architectures.
That one for example shows itself in later parts:
   5.  Gaining registrar optionally verifies the authorization
       information with the info command to the registry, as defined in
       Section 4.3.
   6.  Gaining registrar sends the transfer request with the
       authorization information to the registry, as defined in
       Section 4.4.

Since both actions have no guarantee to happen back to back and immediately (nor to be done by the same subsystems, from the same EPP client, throught the same EPP connection),
the registrar MUST store the authorization somewhere.
Think about connection issues or delayed payment (wish to check authorization
information even before taking the payment and starting the transfer), etc.

As is, this document will create interoperability problems in part because
it does not even define an extension visible at greeting.
Without that, how could an EPP client know if the server follows point 4.1
for example, which is even more troublesome because of its MAY?
Without a clear indication, a client can continue sending a password, and
see its domain:create command be rejected, without even knowing why
(error reporting is not something  sufficiently standardized and stable across
all registries for a client to base itself on).
 
>     Things like that:
>     
>     > The operational practice will not require the client
>     >       to store the authorization information and will require the
>     >       server to store the authorization information using a
>     >       cryptographic hash.  
>     
>     How the password is stored and handled at the registry is completely
>     out of EPP scope. It could as well be symmetrically encrypted, and I fail
>     to see even how this can be enforceable (how will you verify remotely
>     how the registry stores the password?), as it is not protocol related.
>     
> JG - Why would the storage and handling of the authorization 
> information be out of EPP scope? 

Imagine a registry storing passwords as plain text
and another storing it encrypted through some clever mechanism deriving the key
from other registrars data (like its EPP password, that one never being needed
to echo back, so could be stored as an hash).

What does that change for EPP?
Absolutely nothing.

> Do you agree that a cryptographic 
> hash is more secure than using an encrypted value?

Irrelevant to EPP. The EPP schema clearly mandates for the passwords (both login and authInfo) to be exchanged in clear text (encapsulated in TLS of course).
One can see now that things should be done differently, and I could agree there.
But this has no relationship with how the registry stores it.

> JG - It's not meant to take into account all cases that exist today, 

That will then remain a big problem for me, as an implementer.

>     So in my views the current password based model per domain has died,
>     and other solutions have to be searched for. Maybe there is space to pursue
>     in solutions around OTP frameworks.
> 
> JG - You may want to take a stab at defining an alternative mechanism.  
> I believe that EPP does not need to be extended to make the 
> authorization information secure for transfers.  

Aside, remember that the current EPP schema already allows for authorization
to happen, not only by providing the domain authInfo but instead the authInfo
of a related contact (and its ROID to be able to pinpoint it).

And I seem to remember at least one registry to allow that. So definitively
rare but not 0 either.

> Any ideas that you have to improve it would 
> be greatly appreciated.

Maybe, but for me this work is not a good fit inside this working group
or even the IETF. It may be a better fit for some ICANN groups, in order
to deliver some "consensus policies" document (but remembering also at the same time that there is a world outside of gTLDs....). In my view the whole process around
transfers (and not just talking here about the EPP transfer command) should be reviewed
and reworked.

-- 
  Patrick Mevzek
  [e-mail]