Re: [OAUTH-WG] 答复: Re: A question of 1.3.1. Authorization Code in rfc6749 The OAuth 2.0 Authorization Framework

[Justin Richer <[e-mail]>]

Brent,

If you're sending the code in the back channel directly to the Client,
as I believe you're doing from your text below, I would like you to
realize some things:

1) This is not OAuth, and is in fact antithetical to the OAuth way of
solving the connection problem.
2) You are actually lowering your overall security because the access
code is no longer bound to any particular browser session with either
the client or the AS.
3) If you're sending it directly, there is no longer any point of using
the code, since the Client is just going to turn around and send it to
the AS again to get a token. Why not just send the token? But again,
this is still not OAuth.


Think about it this way: There are three connections in the common OAuth
Authorization Code scenario, which is why it's known as a three-legged
OAuth flow in many circles. Each of these legs has unique properties, is
protected by different things, and is exposed to different pieces of
knowledge at different times.

1) Client <-> User Agent: protected by a local session of the Client's
choosing. For Web based clients this is often standard HTTP session
cookies or similar, potentially backed by a login of some type by the
user to the Client as well. This session is never exposed to the
Authorization Server.
2) Authorization Server <-> User Agent: protected by a local session of
the Authorization Server's choosing, normally through some kind of
Primary Credential (login) that the user has a the AS. Importantly,
these credentials are never exposed to the Client.
3) Client <-> Authorization Server: protected by the client credentials,
which are, importantly, not exposed to the user or user agent.

By sending the code as part of the redirect, the Client is able to prove
that the User Agent actually went to the Authorization Server (2) and
got something. The Authorization Code is then sent through the bottom
leg (3) of the channel to verify that it really did come from the
Authorization Server in the context of the user that was just there in
(1). In other words, this mechanism that you seem to be avoiding is
exactly what makes OAuth secure.

If you instead send the code through (3), then the Client as no way at
all of knowing that the user in (1) ever went to the authorization
server via (2). All the Client knows is that they sent the user
someplace and that a magic code showed up. It is very, very, very
dangerous and a very, very bad idea to assume that a code coming in the
back channel (3) has anything at all to do with any particular session (1).

This approach does *not* mitigate any real security threats, and in fact
introduces a great number of others, as described here. There are much
better ways to protect the Authorization Code, and most of the best
practices are enumerated in the security considerations document. Some
of the most common are:

1) Make Authorization Codes one-time-use (even if you try and fail, it's
thrown away)
2) Make Authorization Codes time out after a very short period (minutes
or seconds)


I hope this helps.

-- Justin



On 01/09/2013 01:42 AM, Peng Zhou wrote:
> Dear SuJing:
>
> If it is the only reason, why we send the authorization code to the
> client directly and send another notification without the
> authorization code to the RO. This way can mitigate the chance that
> the authorization code is exposed to the RO's user-agent, hence
> protecting the authorization code from leaking to possible attackers
> in a higher security levle.
>
> Best Regards
> Brent
>
> 2013/1/9  <[e-mail]>:
>> Then why not let auth code be sent directly from AS to Client?
>>
>> Just want to inform RO that an auth code has been dilivered to Client?
>>
>> [e-mail] 写于 2013-01-09 14:27:50:
>>
>>> Hi Brent,
>>>
>>> Few points, why this doesn't create any security implications..
>>>
>>> 1. Authorization server maintains a binding to the Client, who the
>>> token was issued to. To exchange this to an Access token client
>>> should authenticate him self.
>>> 2. Code can only be exchanged once for an acces token.
>>>
>>> Thanks & regards,
>>> -Prabath
>>> On Wed, Jan 9, 2013 at 6:56 AM, cspzhouroc <[e-mail]
>>>> wrote:
>>> Dear All:
>>>
>>> I have a question in the section 1.3.1. Authorization Code in rfc6749
>>> The OAuth 2.0 Authorization Framework.
>>>
>>> It tells "which in turn directs the resource owner back to the client
>>> with the authorization code."
>>>
>>> Who can let me know the reason why is the authorization code sent to
>>> client through a redirection in resource owner's agent?  Any security
>>> implications?
>>>
>>> Is it possible to let the authorization server send the authorization
>>> code to the client directly (not through resource owner's user-agent)?
>>>
>>> Best Regards
>>> Brent
>>>
>>> _______________________________________________
>>> OAuth mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/oauth
>>> --
>>> Thanks & Regards,
>>> Prabath
>>>
>>> Mobile : [phone]
>>>
>>> http://blog.facilelogin.com
>>> http://RampartFAQ.com_______________________________________________
>>> OAuth mailing list
>>> [e-mail]
>>> https://www.ietf.org/mailman/listinfo/oauth
> _______________________________________________
> OAuth mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/oauth