Re: [OAUTH-WG] Signatures don't solve that problem (was RE: Signatures...what are we trying to solve?)

["Freeman, Tim" <[e-mail]>]

From: George Fletcher [mailto:[e-mail]]
>Maybe I wasn't clear in the use case. The point of signatures is not to enable authorization but to
>ensure that release of data only happens within the context that was authorized by the user. Let's
>take the flip side where signatures are not used.

I agree that taking the flip side there is important.  In order to support the claim "X is important for security" with use cases, you need one use case where X is absent and someone gets some information they shouldn't, and another use case where X is present and everything else is similar and the information doesn't leak

>Any site with access to the bearer token issued to
>sleepwell has access to the Alice's data at the PCP.

Part of the story is missing here.  How did the site other than sleepwell get the bearer token issued to sleepwell?  The negative use case showing that bearer tokens leak would answer that question immediately.

If the bearer token leaks, why couldn't Alice's data have leaked by the same path in a scheme with signatures?

(quoting out of order)
>In the case of health data, just possession of an autorization token is NOT good enough to grant access. ...
>This is not acceptable with sensitive data.

I don't understand where you're coming from.  Are these meant to be technical statements (which would need some sort of explanation or example), statements about laws like HIPPA (which would need citation of the relevant law), or intuitive predictions about some future consensus?   You might be right about the future consensus.  I was hoping to see a technical justification for it, but that might not happen.

Signatures do decrease the amount that sleepwell has to trust pcp.  Without signatures, a dishonest pcp could give a bearer token to sleepwell, publish the bearer token anonymously to third parties who make unauthorized use of Alice's medical records, and then at the end we don't know whether to blame the bearer token leak on sleepwell or pcp.  (That's the negative use case.)  With signatures, pcp is supposed to check that the incoming token is signed by sleepwell, and if it receives the properly signed token when someone is accessing Alice's medical records, and those records leak, then either sleepwell leaked them directly or sleepwell allowed a signed token to leak or sleepwell allowed its private key to leak, so we know who to  blame.  (That's the positive use case.)  If the purpose of signatures is to allow auditing and correctly assigning blame afterward, rather than decreasing the risk of Alice's records leaking, it makes sense to me.

If this is the right story, we don't need discovery in step 1, and we don't need to prove who issued the token at step 3, and we do need to add the use case where pcp didn't do its job.  We also have the irritating problem of distributing a bunch of public keys at the beginning.

Tim Freeman
Email: [e-mail]<mailto:[e-mail]>
Desk in Palo Alto: [phone]
Home: [phone]
Cell: [phone]

From: George Fletcher [mailto:[e-mail]]
Sent: Monday, October 04, 2010 11:28 AM
To: Freeman, Tim
Cc: Zeltsan, Zachary (Zachary); OAuth WG
Subject: Re: Signatures don't solve that problem (was RE: [OAUTH-WG] Signatures...what are we trying to solve?)

Hi Tim,

Thanks for the feedback. Some comments/questions inline...

On 10/4/10 1:59 PM, Freeman, Tim wrote:
Putting the use cases on the table is good because it makes things much clearer.  Unfortunately, it's clear that this use case does not work.

I'd like to number the steps under "Requirements" so I can refer to them unambiguously:

1. The application at www.sleepwell.example.com<http://www.sleepwell.example.com> accesses www.myhealth.example.com<http://www.myhealth.example.com> to discover the location of the PCP system (XRD discovery)
2. The application at www.sleepwell.example.com<http://www.sleepwell.example.com> requests Alice to authorize access to the application at www.pcp.example.com<http://www.pcp.example.com> for the purpose of retrieving basic health data (e.g. date-of-birth, weight, height, etc). The mechanism Alice uses to authorize this access is out of scope for this use case.
3. The application at www.myhealth.example.com<http://www.myhealth.example.com> issues a token bound to www.sleepwell.example.com<http://www.sleepwell.example.com> for access to the application at www.pcp.example.com<http://www.pcp.example.com>. Note that a signed token (JWT) can be used to prove who issued the token.
4. The application at www.sleepwell.example.com<http://www.sleepwell.example.com> constructs a request (includes the token issued by www.myhealth.example.com<http://www.myhealth.example.com>) to the application at www.pcp.example.com<http://www.pcp.example.com>
5. The application at www.sleepwell.example.com<http://www.sleepwell.example.com> signs the request before sending it to www.pcp.example.com<http://www.pcp.example.com>
6. The application at www.pcp.example.com<http://www.pcp.example.com> receives the request and verifies the signature
7. The application at www.pcp.example.com<http://www.pcp.example.com> parses the message and finds the authorization token
8. The application at www.pcp.example.com<http://www.pcp.example.com> verifies the signature of the authorization token
9. The application at www.pcp.example.com<http://www.pcp.example.com> parses the authorization token and verifies that this token was issued to the application at www.sleepwell.com<http://www.sleepwell.com>
10. The application at www.pcp.example.com<http://www.pcp.example.com> retrieves the requested data and returns it to the application at www.sleepwell.example.com<http://www.sleepwell.example.com>


The stated purpose of signatures is:

>The application at www.pcp.example.com<http://www.pcp.example.com> verifies that Alice has authorized
>www.sleepwell.example.com<http://www.sleepwell.example.com> to access her health data as well as enforces
>that www.sleepwell.example.com<http://www.sleepwell.example.com> is the only application that can retrieve that
>data with that specific authorization.

I'll abbreviate domain names like "www.sleepwell.example.com<http://www.sleepwell.example.com>" to "sleepwell".


Bearer tokens work fine when verifying that Alice has authorized  sleepwell<http://www.sleepwell.example.com> to access her health data, so the claim is apparently that signatures give the added benefit of enforcing that sleepwell<http://www.sleepwell.example.com> is the only application that can retrieve that data with that specific authorization.
I suppose this depends on how you define "work fine". While the authorization token issued by myhealthdata and given to sleepwell to present to pcp can be passed as a bearer token (so in that sense it works fine) there is no security around the token. In the case of health data, just possession of an autorization token is NOT good enough to grant access.

Instead the PCP site needs to ensure that the presenting entity is specifically authorized by the Authorization Server. I contend that to establish this "chain of trust" signatures are needed.

Given that in most cases pcp will want to decode the token itself, it will need some form of signed structured token because pcp is the recipient of the token and not the issuer.



Unfortunately, signatures do not do that.  Suppose sleepwell wanted to give Alice's data to apneacheck.  Sleepwell could follow the protocol up to step 4.  Then, instead of signing the request and sending the signed request to pcp, sleepwell could transmit the signed request to apneacheck.  apneacheck could then complete the protocol and get Alice's data from www.pcp.example.com<http://www.pcp.example.com>.
Maybe I wasn't clear in the use case. The point of signatures is not to enable authorization but to ensure that release of data only happens within the context that was authorized by the user. Let's take the flip side where signatures are not used. Any site with access to the bearer token issued to sleepwell has access to the Alice's data at the PCP. This is not acceptable with sensitive data.

So the point of signatures in this use case is to create a binding between the issuer of the token (myhealthdata), the presenter of the token (sleepwell) and the receiver of the token (pcp).

If sleepwell signs the message sent to pcp, then pcp can verify that sleepwell is the sender of the message. Next, if the message contains a token signed by myhealth data, and that token includes a claim that the token is issued to sleepwell. PCP can "connect the dots" and know that this is a valid request.

If the token is sent by another server, the "issued to" claim will not match the entity that signed the message and the PCP application will reject the request.


If sleepwell has can retrive to Alice's data, and the protocol doesn't mandate an invasive control of sleepwell's computation and outputs, it's hopeless to prevent sleepwell from allowing apneacheck to retrieve Alice's data.  If all else fails, sleepwell could access Alice's data itself and then allow apneacheck to access the data from sleepwell.
So I'll concede that once sleepwell has the data, if sleepwell is not trustworthy, there is little Alice can do to protect her data. Maybe laws and other such mechanisms will be developed to deal with this scenario. This is one reason why it's so important for the PCP to only release data to the specific entity that was authorized. This is not possible with bearer tokens in a delegated use case like this one.


For all I know, signatures might be a solution to some problem, but they aren't a solution to this problem.

Tim Freeman
Email: [e-mail]<mailto:[e-mail]>
Desk in Palo Alto: [phone]
Home: [phone]
Cell: [phone]

From: [e-mail]<mailto:[e-mail]> [mailto:[e-mail]] On Behalf Of George Fletcher
Sent: Monday, October 04, 2010 9:20 AM
To: Zeltsan, Zachary (Zachary)
Cc: OAuth WG
Subject: Re: [OAUTH-WG] Signatures...what are we trying to solve?

Hi Zachary,

Here is a use case for signed messages. I've tried to keep this in the format of the other OAuth use cases. Please contact me off-list if there are editorial changes required. I've include the list to see if others have feed back on this use case.

Thanks,
George

Use case: Signed Messages

Description:

Alice manages all her personal health records in her personal health data store (www.myhealth.example.com<http://www.myhealth.example.com>). Alice's Primary Care Physician (www.pcp.example.com<http://www.pcp.example.com>) recommends that Alice see a sleep specialist (www.sleepwell.example.com<http://www.sleepwell.example.com>). Alice arrives at the sleep specialist's office and authorizes it to access her basic health data from her PCP. The application at www.pcp.example.com<http://www.pcp.example.com> verifies that Alice has authorized www.sleepwell.example.com<http://www.sleepwell.example.com> to access her health data as well as enforces that www.sleepwell.example.com<http://www.sleepwell.example.com> is the only application that can retrieve that data with that specific authorization.

Pre-conditions:

* Alice has a personal health data store that allows for discovery of her participating health systems (e.g. psychiatrist, sleep specialist, pcp, orthodontist, ophthalmologist, etc).
* The application at www.myhealth.example.com<http://www.myhealth.example.com> manages authorization of access to Alice's participating health systems
* The application at www.myhealth.example.com<http://www.myhealth.example.com> can issues authorization tokens understood by Alice's participating health systems
* The application at www.pcp.example.com<http://www.pcp.example.com> stores Alice's basic health and prescription records
* The application at www.sleepwell.com<http://www.sleepwell.com> stores results of Alice's sleep tests


Post-conditions:
* A successful procedure results in just the information that Alice authorized being transferred from the Primary Care Physician (www.pcp.example.com<http://www.pcp.example.com>) to the sleep specialist (www.sleepwell.example.com<http://www.sleepwell.example.com>).
* The transfer of health data only occurs if the application at www.pcp.example.com<http://www.pcp.example.com> can verify that www.sleepwell.example.com<http://www.sleepwell.example.com> is the party requesting access and that the authorization token presented by www.sleepwell.example.com<http://www.sleepwell.example.com> is issued by the application at www.myhealth.example.com<http://www.myhealth.example.com> with a restricted audience of www.sleepwell.example.com<http://www.sleepwell.example.com>

Requirements:
1. The application at www.sleepwell.example.com<http://www.sleepwell.example.com> accesses www.myhealth.example.com<http://www.myhealth.example.com> to discover the location of the PCP system (XRD discovery)
2. The application at www.sleepwell.example.com<http://www.sleepwell.example.com> requests Alice to authorize access to the application at www.pcp.example.com<http://www.pcp.example.com> for the purpose of retrieving basic health data (e.g. date-of-birth, weight, height, etc). The mechanism Alice uses to authorize this access is out of scope for this use case.
3. The application at www.myhealth.example.com<http://www.myhealth.example.com> issues a token bound to www.sleepwell.example.com<http://www.sleepwell.example.com> for access to the application at www.pcp.example.com<http://www.pcp.example.com>. Note that a signed token (JWT) can be used to prove who issued the token.
4. The application at www.sleepwell.example.com<http://www.sleepwell.example.com> constructs a request (includes the token issued by www.myhealth.example.com<http://www.myhealth.example.com>) to the application at www.pcp.example.com<http://www.pcp.example.com>
5. The application at www.sleepwell.example.com<http://www.sleepwell.example.com> signs the request before sending it to www.pcp.example.com<http://www.pcp.example.com>
6. The application at www.pcp.example.com<http://www.pcp.example.com> receives the request and verifies the signature
7. The application at www.pcp.example.com<http://www.pcp.example.com> parses the message and finds the authorization token
8. The application at www.pcp.example.com<http://www.pcp.example.com> verifies the signature of the authorization token
9. The application at www.pcp.example.com<http://www.pcp.example.com> parses the authorization token and verifies that this token was issued to the application at www.sleepwell.com<http://www.sleepwell.com>
10. The application at www.pcp.example.com<http://www.pcp.example.com> retrieves the requested data and returns it to the application at www.sleepwell.example.com<http://www.sleepwell.example.com>



On 9/28/10 12:27 PM, Zeltsan, Zachary (Zachary) wrote:
These use cases are not in the draft https://datatracker.ietf.org/doc/draft-zeltsan-use-cases-oauth.
Could you write them up?

Thanks,
Zachary

________________________________
From: [e-mail]<mailto:[e-mail]> [mailto:[e-mail]] On Behalf Of George Fletcher
Sent: Tuesday, September 28, 2010 11:39 AM
To: OAuth WG
Subject: Re: [OAUTH-WG] Signatures...what are we trying to solve?

I think of the signature issues as falling into two classes... I think they map to your classification as well...

 *   Signing tokens is important for interoperability especially looking forward to a time when tokens issued by multiple Authorization Servers are accepted at a given host.
 *   Signing messages is important because it provides a mechanism to ensure that the entity making the API call (and presenting an access token) is really the entity that is allowed to make the API call.
Signing messages applies to the re-delegation use cases. I've heard the need for this class of use cases from both the hData (health data) community as well as the user managed access (UMA) community.

Signing tokens covers both your second class of tokens as well as another use case that Eran has mentioned as well. Namely, a protected resource server honoring tokens from multiple Authorization Servers.

These are the two classes of use cases that I'd like to see solved.

Thanks,
George


On 9/28/10 12:58 AM, David Recordon wrote:
If you know me then you'll know that I'm generally one of the last people to talk about Alice and Bob. That said, there are a lot of technical proposals flying across the list with very little shared understanding of the problem(s) we're trying to solve.

From what I've seen there are two distinct classes of signature use cases.
1) The first is where the HTTP request parameters must be part of the signature. An example is any OAuth 1.0a style API where you want to make sure that the HTTP POST your server just received isn't masquerading itself as a GET.
2) The second is where the HTTP request is orthogonal. An example is OpenSocial where the server is sending state information to the client such as what user is currently logged in.

The main practical example I have of the first use case is what Twitter wants to do with redelegation. In this case TweetDeck can't given TwitPic it's own bearer token, but needs to sign the POST request and pass that signature to TwitPic for it to include in the final API request to Twitter.

In terms of signing protected resource requests, I haven't heard anyone bring up specific and detailed needs for this recently.

JSON tokens pretty clearly make sense for the second class of signature use cases and it's actually a bit hard to argue why they would be a part of OAuth. Facebook shipped this a bit over a month ago for canvas applications. We include a `signed_request` parameter which is signature.base64url(JSON). Parsing it is 18 lines of PHP. http://developers.facebook.com/docs/authentication/canvas

This second class of use case will also be required by OpenID Connect where the server is signing identity information and sending it to the client. I imagine that OpenSocial will also still have it and wish to continue relying on public key algorithms.

So a few questions:
 * Do we want to tackle both of these classes of signatures in OAuth?
 * Why do you consider the second class part of OAuth versus something completely separate that might happen to include an OAuth access token?
 * Is the Twitter redelegation use case the right focus for the first class?
 * Is there an example of an OAuth 2.0 server that can't use bearer tokens for protected resource requests and thus requires signatures?

Thanks,
--David





_______________________________________________

OAuth mailing list

[e-mail]<mailto:[e-mail]>

https://www.ietf.org/mailman/listinfo/oauth





--

Chief Architect                   AIM:  gffletch

Identity Services Engineering     Work: [e-mail]<mailto:[e-mail]>

AOL Inc.                          Home: [e-mail]<mailto:[e-mail]>

Mobile: [phone]           Blog: http://practicalid.blogspot.com

Office: [phone]           Twitter: http://twitter.com/gffletch