I'd like to return to this thread. After learning about the refresh
grant supporting a client-driven down-scoping, the same question which I
asked in the first place seems relevant again.
If the clients gets 401 back while trying to access PR, how will it know
how to react to it (use refresh grant with or without its optional scope
parameter), is it because the access token has expired or is it because
it has an invalid scope ?
Thanks, Sergey
On 27/11/12 18:11, Sergey Beryozkin wrote:
Hi Adam,
thanks for sharing it, helpful,
On 27/11/12 17:24, Lewis Adam-CAL022 wrote:
Hi Sergey,
In my use cases the client actively monitors the expiration of the AT
in order to request a new AT using the RT. We do this as you
suggested, because presenting an expired AT to the RS is a wasted
round trip and adds latency and degrades the user experience. Why send
the AT if we know it's expired?
How do you know when the client is expecting a revocation and when a
downscoping, I guess if the client includes and optional 'scope'
parameter then it is down-scoping obviously, otherwise - revocation ?
If that is the case then IMHO a refresh token grant request without a
scope parameter might offer a shortcut for the client not to worry about
one more endpoint (to do with the revocation) ?
As to your other question, one reason to down scope an AT ... in my
use cases the AS issues access tokens for many different RS's. It is a
security concern to send an AT to RS1 that has the ability to also
access protected resources at RS2 and RS3 and RSn, etc. So I make a
first request to get a "master" AT with a master RT, and then
immediately destroy the master AT and use the master RT to get a
down-scoped AT - one for each RS. This is inefficient though, and I've
been nagging the WG to consider a draft to allow a client to request
an array of access tokens of different scopes.
OK, that makes it clearer. What I'm concerned about though, does the
client need to be concerned or is it the job of AS to protect in cases
where a token with too many scopes is used to access a given RS which
may delegate to a stricter RS ?
I mean, is client expected to be well-behaved and know about the
relationships between the master RS with the stricter RS1 & RS2 parthers
? Seems not easy to make it interoperable...
Guess I'm missing the point still
Thanks, Sergey
These are just my experience, and as Justin mentioned, your millage
may vary. Hope it helps either way :-)
adam
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf
Of Sergey Beryozkin
Sent: Tuesday, November 27, 2012 10:26 AM
To: Richer, Justin P.
Cc: [email protected]
Subject: Re: [OAUTH-WG] How the client can decide it is time to use
the refresh token
On 27/11/12 15:48, Richer, Justin P. wrote:
Yes, the client can keep two tokens and try using both of them if it
wants to. I think that you're conflating revocation and expiration
here, and missing a key use case: downscoping. Let's say the client
gets an access grant for [A, B, C], and it gets RT and AT1 for those
scopes. The client then wants to call a service with *only* scope B,
so it sends RT in with scope B in the refresh request to get AT2 with
that scope alone. AT1 is still valid, assuming it hasn't expired or
otherwise been revoked. The client can choose which AT to present to
the protected resources in different situations.
That is complex :-). Using a refresh token to downscope and keep a
number of access tokens is definitely beyond my imagination at the
moment, why would a client want to downscope its access token ?
Is it basically described at:
http://tools.ietf.org/html/draft-zeltsan-oauth-use-cases-03#section-3.10
?
I'm still not getting though why access token with the richer set of
scopes needs to be explicitly downscoped by the client, is it because
using a token with a richer set of scopes can lead to some authorization
issues ?
I was thinking that the down-scoping only makes sense during the
authorization code flow where AS downscopes what the client has
requested but obviously I'm only at the start of the learning curve :-).
In all of these cases, it's completely up to the AS and the PRs
whether or not any of the tokens in the wild are still valid. Your AS
might decide that once a RT is used, it will simply throw away all
existing AT's attached to that RT. That's totally a valid use case,
and the client needs to be able to handle that (in the same way that
it handles any invalid token).
If you want the clients to explicitly throw out tokens, use the Token
Revocation spec.
OK, that makes sense now.
Thanks, Sergey
-- Justin
On Nov 27, 2012, at 10:41 AM, Sergey Beryozkin wrote:
On 27/11/12 15:29, Richer, Justin P. wrote:
The client can indeed save a round trip by proactively refreshing
the access token. This does not necessarily revoke the existing
access token. Many implementations do that, so your mileage may vary.
I don't understand.
So the client will still have the existing access token which may
have not been revoked, and then also another access token which was
just returned in response to the refresh grant request ?
And if you mean that no actual revocation is actually done, why then
have the client worry about doing the proactive refresh ?
I can imagine that may be that it will only be the actual refresh
token refreshed itself, but why keep the access token if the client
has requested a refresh ?
Sergey
-- Justin
On Nov 27, 2012, at 5:18 AM, Sergey Beryozkin wrote:
On 27/11/12 08:52, Guangqing Deng wrote:
It seems that the client cannot know whether the refresh token
should be
used until a HTTP 401 error is returned from the resource server
due to
the expiration of current access token or some other reasons.
However, a
period of time cannot be ignored will be spent on obtain a new
access
token using the refresh token after the resource server returns a
HTTP
401 error to the client, which may degrade the user experience
since the
real-time nature of the service cannot be guaranteed. Is a
mechanism by
which the client can check the validity of the access token (or the
refresh token) in advance needed by Oauth?
I believe an access token response may offer an expires_in
parameter which can provide some hint. Actually this raises an
interesting question.
Suppose the client actually checks this parameter and decides to use
a refresh token it also has to pro-actively refresh its current
token.
IMHO this should work even if the access token has not expired yet
and effectively represents another form of a client-initiated
revocation of the access token ? It will mean a client which works
with the "expires_in" parameter can save a one round trip (the one
which returns a failure in case of the access token being expired) ?
Does it make sense ? If it does, can some relevant wording make it
into the revocation token draft ?
Cheers, Sergey
------------------------------------------------------------------------
Guangqing Deng
------------------------------------------------------------------------
*From:* Justin Richer<[email protected]>
*To:* Sergey Beryozkin<[email protected]>
*Cc:* [email protected]
*Sent:* Wednesday, November 21, 2012 6:11 AM
*Subject:* Re: [OAUTH-WG] How the client can decide it is time
to use
the refresh token
There's no signaling regarding the validity of the refresh token
from
the protected resource. In more distributed setups, the protected
resources know nothing about the refresh tokens because the PR
never
sees them. In any case. the code path is fairly straightforward,
even if
both tokens are expired:
- client presents AT to resource
- resource returns data, AT worked
- [or] resource returns 400 error to client, AT didn't work
- client presents RT to auth server
- auth server returns new AT, RT worked
- [or] auth server returns 400 error to client, RT didn't work
- client has to go get a new auth grant from the resource owner,
start over
-- Justin
On 11/21/2012 06:50 AM, Sergey Beryozkin wrote:
Hi
I'm looking for some guidance on how the client which already
owns an
access token can decide, after getting HTTP 400 back from the
resource
server it tries to access on behalf of the end user/resource
owner, can
decide that the refresh token it has can now be used to get a
new access
token.
[1] refers to various error conditions but it is not obvious to me
that the same conditions (some of them) should or can be
reported during
the actual client accessing the protected resource.
My question is, what error condition, if any, from [1] should be
reported back to the client failing to access a protected
resource due
to the access token being invalid or expired, so that it can
help the
client who also owns the refresh token to decide it can use it
now...
Thanks, Sergey
[1] http://tools.ietf.org/html/rfc6749#section-5.2
_______________________________________________
OAuth mailing list
[email protected]<mailto:[email protected]>
https://www.ietf.org/mailman/listinfo/oauth
_______________________________________________
OAuth mailing list
[email protected]<mailto:[email protected]>
https://www.ietf.org/mailman/listinfo/oauth
_______________________________________________
OAuth mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/oauth
_______________________________________________
OAuth mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/oauth
_______________________________________________
OAuth mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/oauth
_______________________________________________
OAuth mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/oauth
_______________________________________________
OAuth mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/oauth
