Re: [OAUTH-WG] Can the repeated authorization of scopes be avoided ?

[Sergey Beryozkin <[e-mail]>]

Hi,

Many thanks to all who provided the feedback.

As far as the existing token update is concerned I was thinking about it 
in the context of the incremental authorization which we haven't 
implemented yet so we'll need to review how to handle it later on.

Right now we are only prototyping the code for not challenging the user 
with a consent screen if the requested scopes have already been approved 
(per a given user + client id combination, and which is a base for 
supporting the incremental auth at a next step).

I'm a bit confused about the use of a 'grant' term in your replies.

So consider a confidential client redirecting the user and requesting 
some scope, as part of the authorization code flow. The user authorizes 
the client and the client gets a code *grant* which is according to the 
spec can live for up to *10 min*. The client exchanges this grant for a 
token with the token preserving the fact the user has authorized a given 
scope for this client. I guess this is all quite common.

Note the code 'grant' has already gone by now, because it was already 
used once, withing a 10 mins period, which is another spec requirement.

That is why I'm referring to the existing access token record which can 
be used for keeping the track of the scopes approved by a given user for 
a given client. This token can be refreshed if needed.

When the user's session with a confidential client's web app has 
expired, the user is redirected to authenticate, with some scopes 
requested. At this point the record which keeps the approved scopes for 
a given user/client is an existing access/refresh token.

This is why I'm confused about the use of the 'grant' term in your 
replies. I guess this can be a 'grant' record for keeping the list of 
the approved scopes/etc not related to a record representing a transient 
authorization code record. But as I said, using the live access/refresh 
token info seems reasonable, sorry, may be it is becoming too 
implementation specific...

Cheers, Sergey





On 26/01/16 23:03, Justin Richer wrote:
> In MITREid Connect we track grants per client_id per user, and we have a
> separate database object for storing them. I wouldn’t recommend simply
> updating an access token that’s already in the wild with more power —
> that just sounds wrong.
>
>   — Justin
>
>> On Jan 26, 2016, at 1:57 PM, Thomas Broyer <[e-mail]
>> <mailto:[e-mail]>> wrote:
>>
>> Fwiw, at Ozwillo, we track grants per user per client_id (we don't
>> support native apps; only web flows for now), and we don't do
>> "incremental grants" like Google: if you request scope B and the user
>> had only granted scope A, you'll get a token for scope B only. This is
>> partly because our tokens are not for our own APIs only, contrary to
>> Google, so we want to allow clients to get tokens with narrow scopes
>> so they could have one token per third-party API and prevent rogue
>> resources from trying to use received tokens at other APIs.
>>
>> UI-wise, we tell the user what he already granted to the client, and
>> even let him grant scopes that the client has pre-registered as
>> "possibly needed at some time" (through a custom provisioning
>> protocol), but the issued token is always for the exact scopes that
>> the client requested in this specific request.
>> And if all requested scopes have already been granted, then we do a
>> transparent redirect without showing anything to the user (which is
>> what most other implementations do too)
>>
>> Le mar. 26 janv. 2016 19:04, Sergey Beryozkin <[e-mail]
>> <mailto:[e-mail]>> a écrit :
>>
>>     Hi
>>
>>     I'm not sure if the next question is off topic or too low level,
>>     hopefully not,
>>
>>     When the repeated authorization is skipped or only new scopes are
>>     requested to be authorized as per the incremented auth approach, is it
>>     reasonable to assume that the record that is used to track it all is
>>     actually the existing access token or is totally OIDC implementation
>>     specific ?
>>     I think using the existing token as a record is reasonable because
>>     it is
>>     time scoped and if we do not use the access token for keeping the
>>     track
>>     of the multiple approvals, etc, then one need to introduce one more
>>     record mirroring to some extent the access token...
>>
>>     For example, the user session may have expired but the access
>>     token that
>>     was issued to a client web app on behalf of this user is still active,
>>     so when the user returns and signs in again, and for example, approves
>>     few more scopes, then the existing access token (the record) gets
>>     updated, instead of a new token being created.
>>
>>     If it is reasonable then does it mean the sticky or incremental
>>     authorization works as long as the access token is available
>>     (refreshable) ?
>>
>>     Sergey
>>
>>
>>
>>
>>     On 19/01/16 09:59, Sergey Beryozkin wrote:
>>     > Hi William
>>     >
>>     > Thanks for the advice. FYI we are also on the way to supporting the
>>     > incremental authorization of scopes - thanks for highlighting the
>>     > importance of this process on this list...
>>     >
>>     > Cheers, Sergey
>>     > On 19/01/16 03:10, William Denniss wrote:
>>     >> Agree with Justin, this is pretty common. We support it for
>>     re-auth as
>>     >> well as incremental auth (where the user has already approved
>>     scope "a"
>>     >> and is presented with a request for scopes "a b", they will
>>     only need to
>>     >> approve scope "b").  In fact if you don't do this, then
>>     incremental auth
>>     >> isn't really viable.
>>     >>
>>     >> Regarding security: don't do this for non-confidential clients
>>     where you
>>     >> can't verify the identity of the app by the redirect (e.g. a
>>     localhost
>>     >> redirect to an installed app).
>>     >>
>>     >> On Mon, Jan 18, 2016 at 4:44 AM, Sergey Beryozkin
>>     <[e-mail] <mailto:[e-mail]>
>>     >> <mailto:[e-mail] <mailto:[e-mail]>>> wrote:
>>     >>
>>     >>     Hi Justin, thanks for the advice,
>>     >>
>>     >>     Cheers, Sergey
>>     >>
>>     >>     On 18/01/16 11:47, Justin Richer wrote:
>>     >>
>>     >>         Yes, this is common practice. Give the user the option to
>>     >>         remember the
>>     >>         decision. This is known as "trust on first use", or
>>     tofu. Our
>>     >>         server,
>>     >>         MITREid Connect, implements this as do many others.
>>     >>
>>     >>
>>     >>
>>     >>         -- Justin
>>     >>
>>     >>         / Sent from my phone /
>>     >>
>>     >>
>>     >>         -------- Original message --------
>>     >>         From: Sergey Beryozkin <[e-mail]
>>     <mailto:[e-mail]>
>>     >>         <mailto:[e-mail]
>>     <mailto:[e-mail]>>>
>>     >>         Date: 1/18/2016 5:59 AM (GMT-05:00)
>>     >>         To: [e-mail] <mailto:[e-mail]>
>>     <mailto:[e-mail] <mailto:[e-mail]>>
>>     >>         Subject: [OAUTH-WG] Can the repeated authorization of
>>     scopes be
>>     >>         avoided ?
>>     >>
>>     >>         Hi All
>>     >>
>>     >>         The question relates to the process of showing the
>>     authorization
>>     >>         code/implicit flow consent screen to a user.
>>     >>
>>     >>
>>     >>         I'm discussing with my colleagues the possibility of
>>     avoiding
>>     >>         asking the
>>     >>         same user whose session has expired and who is
>>     re-authenticating
>>     >>         with AS
>>     >>         which scopes should be approved.
>>     >>
>>     >>         For example, suppose the OAuth2 client redirects a user
>>     with the
>>     >>         requested scope 'a'. The user signs in to AS and is shown a
>>     >> consent
>>     >>         screen asking to approve the 'a' scope. The user
>>     approves 'a'
>>     >>         and the
>>     >>         flow continues.
>>     >>
>>     >>         Some time later, when the user's session has expired,
>>     the user is
>>     >>         redirected to AS with the same 'a' scope.
>>     >>
>>     >>         Would it be a good idea, at this point, not to show the
>>     user the
>>     >>         consent
>>     >>         screen asking to approve the 'a' scope again ? For
>>     example, AS
>>     >> can
>>     >>         persist the fact that a given user has already approved
>>     'a' for
>>     >>         a given
>>     >>         client earlier, so when the user re-authenticates, AS
>>     will use
>>     >>         this info
>>     >>         and will avoid showing the consent screen.
>>     >>
>>     >>         That seems to make sense, but I'm wondering, can there
>>     be some
>>     >>         security
>>     >>         implications associated with it, any
>>     recommendations/advices
>>     >>         will be welcome
>>     >>
>>     >>         Sergey
>>     >>
>>     >>         _______________________________________________
>>     >>         OAuth mailing list
>>     >> [e-mail] <mailto:[e-mail]> <mailto:[e-mail]
>>     <mailto:[e-mail]>>
>>     >> https://www.ietf.org/mailman/listinfo/oauth
>>     >>
>>     >>
>>     >>     _______________________________________________
>>     >>     OAuth mailing list
>>     >> [e-mail] <mailto:[e-mail]> <mailto:[e-mail]
>>     <mailto:[e-mail]>>
>>     >> https://www.ietf.org/mailman/listinfo/oauth
>>     >>
>>     >>
>>     >
>>     >
>>
>>     _______________________________________________
>>     OAuth mailing list
>>     [e-mail] <mailto:[e-mail]>
>>     https://www.ietf.org/mailman/listinfo/oauth
>>
>> _______________________________________________
>> OAuth mailing list
>> [e-mail] <mailto:[e-mail]>
>> https://www.ietf.org/mailman/listinfo/oauth
>


-- 
Sergey Beryozkin

Talend Community Coders
http://coders.talend.com/