| OAuth Working Group | W. Denniss |
| Internet-Draft | |
| Intended status: Best Current Practice | J. Bradley |
| Expires: January 21, 2017 | Ping Identity |
| July 20, 2016 |
OAuth 2.0 for Native Apps
draft-ietf-oauth-native-apps-03
OAuth 2.0 authorization requests from native apps should only be made through external user-agents, primarily the system browser. This specification details the security and usability reasons why this is the case, and how native apps and authorization servers can implement this best practice.
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at http://datatracker.ietf.org/drafts/current/.
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."
This Internet-Draft will expire on January 21, 2017.
Copyright (c) 2016 IETF Trust and the persons identified as the document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (http://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Simplified BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Simplified BSD License.
The OAuth 2.0 [RFC6749] authorization framework, documents two approaches in Section 9 for native apps to interact with the authorization endpoint: via an embedded user-agent, or an external user-agent.
This document recommends external user-agents like in-app browser tabs as the only secure and usable choice for OAuth. It documents how native apps can implement authorization flows with such agents, and the additional requirements of authorization servers needed to support such usage.
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in Key words for use in RFCs to Indicate Requirement Levels [RFC2119]. If these words are used without being spelled in uppercase then they are to be interpreted with their normal natural language meanings.
In addition to the terms defined in referenced specifications, this document uses the following terms:
At the time of writing, many native apps are still using web-views, a type of embedded user-agent, for OAuth. That approach has multiple drawbacks, including the client app being able to eavesdrop user credentials, and is a suboptimal user experience as the authentication session can't be shared, and users need to sign-in to each app separately.
OAuth flows between a native app and the system browser (or another external user-agent) are more secure, and take advantage of the shared authentication state to enable single sign-on.
Inter-process communication, such as OAuth flows between a native app and the system browser can be achieved through URI-based communication. As this is exactly how OAuth works for web-based OAuth flows between RP and IDP websites, OAuth can be used for native app auth with very little modification.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+ | User Device | | | | +---------------------------+ | +-----------+ | | | | (5) Authz Code | | | | Client App |----------------------->| Token | | | |<-----------------------| Endpoint | | +---------------------------+ | (6) Access Token, | | | | ^ | Refresh Token +-----------+ | | | | | | | | | | (1) | (4) | | | Authz | Authz | | | Request | Code | | | | | | | | | | v | | | +---------------------------+ | +---------------+ | | | | (2) Authz Request | | | | Browser |--------------------->| Authorization | | | |<---------------------| Endpoint | | +---------------------------+ | (3) Authz Code | | | | +---------------+ +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~+
Figure 1: Native App Authorization via External User-agent
Figure 1 illustrates the interaction of the native app with the system browser to authorize the user via an external user-agent.
Just as URIs are used for OAuth 2.0 [RFC6749] on the web to initiate the authorization request and return the authorization response to the requesting website, URIs can be used by native apps to initiate the authorization request in the device's system browser and return the response to the requesting native app.
By applying the same principles from the web to native apps, we gain similar benefits like the usability of a single sign-on session, and the security by a separate authentication context. It also reduces the implementation complexity by reusing the same flows as the web, and increases interoperability by relying on standards-based web flows that are not specific to a particular platform.
It is RECOMMENDED that native apps use the URI-based communication functionality of the operating system to perform OAuth flows in an external user-agent, typically the system browser.
Some platforms support a browser feature known as in-app browser tabs, where an app can present a tab of the browser within the app context without switching apps, but still retain key benefits of the browser such as a shared authentication state and security context. On platforms where they are supported, it is RECOMMENDED for usability reasons that apps use in-app browser tabs for the Authorization Request.
It is possible to create an external user-agent for OAuth that is a native app provided by the authorization server, as opposed to the system browser. This approach shares a lot of similarity with using the system browser as both use URIs for inter-app communication and is able to provide a secure, shared authentication session, and thus MAY be used for secure native OAuth, applying most of the techniques described here. However it is NOT RECOMMENDED due to the increased complexity and requirement for the user to have the AS app installed. While much of the advice and security considerations are applicable to such clients, they are out of scope for this specification.
The authorization request is created as per OAuth 2.0 [RFC6749], and opened in the system browser. Where the operating system supports in-app browser tabs, those should be preferred over switching to the system browser, to improve usability.
The function of the redirect URI for a native app authorization request is similar to that of a web-based authorization request. Rather than returning the authorization code to the OAuth client's server, it returns it to the native app. The various options for a redirect URI that will return the code to the native app are documented in Section 7. Any redirect URI that allows the app to receive the URI and inspect its parameters is viable.
There are three main approaches to redirection URIs for native apps: custom URI schemes, app-claimed HTTPS URI schemes, and loopback redirects.
Most major mobile and desktop computing platforms support inter-app communication via URIs by allowing apps to register custom URI schemes. When the system browser or another app attempts to follow a URI with a custom scheme, the app that registered it is launched to handle the request. This document is only relevant on platforms that support this pattern.
In particular, the custom URI scheme pattern is supported on Android [Android.URIScheme], iOS [iOS.URIScheme], Windows Universal Platform (UWP) [WindowsUWP.URIScheme] and macOS [macOS.URIScheme].
To perform an OAuth 2.0 Authorization Request on a supported platform, the native app launches the system browser with a normal OAuth 2.0 Authorization Request, but provides a redirection URI that utilizes a custom URI scheme that is registered by the calling app.
When the authentication server completes the request, it redirects to the client's redirection URI like it would any redirect URI, but as the redirection URI uses a custom scheme, this results in the OS launching the native app passing in the URI. The native app extracts the code from the query parameters from the URI just like a web client would, and exchanges the Authorization Code like a regular OAuth 2.0 client.
When selecting which URI scheme to associate with the app, apps SHOULD pick a scheme that is globally unique, and which they can assert ownership over.
To avoid clashing with existing schemes in use, using a scheme that follows the reverse domain name pattern applied to a domain under the app publishers control is RECOMMENDED. Such a scheme can be based on a domain they control, or the OAuth client identifier in cases where the authorization server issues client identifiers that are also valid DNS subdomains. The chosen scheme MUST NOT clash with any IANA registered scheme [IANA.URISchemes]. You SHOULD also ensure that no other app by the same publisher uses the same scheme.
Schemes using reverse domain name notation are hardened against collision. They are unlikely to clash with an officially registered scheme [IANA.URISchemes] or unregistered de-facto scheme, as these generally don't include a period character, and are unlikely to match your domain name in any case. They are guaranteed not to clash with any OAuth client following these naming guidelines in full.
Some platforms use globally unique bundle or package names that follow the reverse domain name notation pattern. In these cases, the app SHOULD register that bundle id as the custom scheme. If an app has a bundle id or package name that doesn't match a domain name under the control of the app, the app SHOULD NOT register that as a scheme, and instead create a URI scheme based off one of their domain names.
For example, an app whose publisher owns the top level domain name "example.com" can register "com.example.app:/" as their custom scheme. An app whose authorization server issues client identifiers that are also valid domain names, for example "client1234.usercontent.idp.com", can use the reverse domain name notation of that domain as the scheme, i.e. "com.idp.usercontent.client1234:/". Each of these examples are URI schemes which are likely to be unique, and where the publisher can assert ownership.
As a counter-example, using a simple custom scheme like "myapp:/" is not guaranteed to be unique and is NOT RECOMMENDED.
In addition to uniqueness, basing the URI scheme off a name that is under the control of the app's publisher can help to prove ownership in the event of a dispute where two apps register the same custom scheme (such as if an app is acting maliciously). For example, if two apps registered "com.example.app:", the true owner of "example.com" could petition the app store operator to remove the counterfeit app. This petition is harder to prove if a generic URI scheme was chosen.
As recommended in Section 3.1.2.2 of OAuth 2.0 [RFC6749], the authorization server SHOULD require the client to pre-register the redirection URI. This remains true for app redirection URIs that use custom schemes.
Additionally, authorization servers MAY request the inclusion of other platform-specific information, such as the app package or bundle name, or other information used to associate the app that may be useful for verifying the calling app's identity, on operating systems that support such functions.
Authorizations servers SHOULD support the ability for native apps to register Redirection URIs that utilize custom URI schemes. Authorization servers SHOULD enforce the recommendation in Section 7.1.2 that apps follow naming guidelines for URI schemes.
Some operating systems allow apps to claim HTTPS URLs of their domains. When the browser sees such a claimed URL, instead of the page being loaded in the browser, the native app is launched instead with the URL given as input.
Where the operating environment provided app-claimed HTTPS URIs in a usable fashion, these URIs should be used as the OAuth redirect, as they allow the identity of the destination app to be guaranteed by the operating system.
Apps on platforms that allow the user to disable this functionality, present it in a user-unfriendly way, or lack it altogether MUST fallback to using custom URI schemes.
The authorization server MUST allow the registration of HTTPS redirect URIs for non-confidential native clients to support app-claimed HTTPS redirect URIs.
More applicable to desktop operating systems, some environments allow apps to create a local HTTP listener on a random port, and receive URI redirects that way. This is an acceptable redirect URI choice for native apps on compatible platforms.
Authorization servers SHOULD support redirect URIs on the loopback IP address and HTTP scheme, that is, redirect URIs beginning with http://127.0.0.1[:port]/, http://::1[:port]/, and http://localhost[:port]/. Authorization servers supporting this class of redirect URI MUST allow the client to specify a port of their choice, and SHOULD allow the client to use an arbitrary path component.
While both the loopback IP and localhost variants SHOULD be supported by the authorization server for completeness, it is RECOMMENDED that apps primarily use the loopback IP variant, as it is less susceptible to misconfigured routing and client side firewalls Note that the HTTP scheme is acceptable for this category of redirect URIs, as the request never leaves the device.
Embedded user-agents, commonly implemented with web-views, are an alternative method for authorizing native apps. They are however unsafe for use by third-parties by definition. They involve the user signing in with their full login credentials, only to have them downscoped to less powerful OAuth credentials.
Even when used by trusted first-party apps, embedded user-agents violate the principle of least privilege by obtaining more powerful credentials than they need, potentially increasing the attack surface.
In typical web-view based implementations of embedded user-agents, the host application can: log every keystroke entered in the form to capture usernames and passwords; automatically submit forms and bypass user-consent; copy session cookies and use them to perform authenticated actions as the user.
Encouraging users to enter credentials in an embedded web-view without the usual address bar and visible certificate validation features that browsers have makes it impossible for the user to know if they are signing in to the legitimate site, and even when they are, it trains them that it's OK to enter credentials without validating the site first.
Aside from the security concerns, web-views do not share the authentication state with other apps or the system browser, requiring the user to login for every authorization request and leading to a poor user experience.
Due to the above, use of embedded user-agents is NOT RECOMMENDED, except where a trusted first-party app acts as the external user-agent for other apps, or provides single sign-on for multiple first-party apps.
Authorization servers SHOULD consider taking steps to detect and block logins via embedded user-agents that are not their own, where possible.
A limitation of custom URI schemes is that multiple apps can typically register the same scheme, which makes it indeterminate as to which app will receive the Authorization Code Grant. This is not an issue for HTTPS redirection URIs (i.e. standard web URLs) due to the fact the HTTPS URI scheme is enforced by the authority (as defined by [RFC3986]), the domain name system, which does not allow multiple entities to own the same domain.
If multiple apps register the same scheme, it is possible that the authorization code will be sent to the wrong app (generally the operating system makes no guarantee of which app will handle the URI when multiple register the same scheme). PKCE [RFC7636] details how this limitation can be used to execute a code interception attack (see Figure 1). This attack vector applies to public clients (clients that are unable to maintain a client secret) which is typical of most native apps.
While Section 7.1.2 details ways that this can be mitigated through policy enforcement (through being able to report and have removed any offending apps), we can also protect the authorization code grant from being used in cases where it was intercepted.
The Proof Key for Code Exchange by OAuth Public Clients (PKCE [RFC7636]) standard was created specifically to mitigate against this attack. It is a Proof of Possession extension to OAuth 2.0 that protects the code grant from being used if it is intercepted. It achieves this by having the client generate a secret verifier which it passes in the initial authorization request, and which it must present later when redeeming the authorization code grant. An app that intercepted the authorization code would not be in possession of this secret, rendering the code useless.
Both the client and the Authorization Server MUST support PKCE [RFC7636] to use custom URI schemes, or loopback IP redirects. Authorization Servers SHOULD reject authorization requests using a custom scheme, or loopback IP as part of the redirection URI if the required PKCE parameters are not present, returning the error message as defined in Section 4.4.1 of PKCE [RFC7636]. It is RECOMMENDED to use PKCE [RFC7636] for app-claimed HTTPS redirect URIs, even though these are not generally subject to interception, to protect against attacks on inter-app communication.
While in-app browser tabs provide a secure authentication context, as the user initiates the flow from a native app, it is possible for that native app to completely fake an in-app browser tab.
This can't be prevented directly - once the user is in the native app, that app is fully in control of what it can render, however there are several mitigating factors.
Importantly, such an attack that uses a web-view to fake an in-app browser tab will always start with no authentication state. If all native apps use the techniques described in this best practice, users will not need to sign-in frequently and thus should be suspicious of any sign-in request when they should have already been signed-in.
This is true even for authorization servers that require frequent or occasional re-authentication, as such servers can preserve some user identifiable information from the old request, like the email address or avatar. To help mitigate against phishing, it is RECOMMENDED to show the user some hint that they were previously logged in, as an attacking app would not be capable of doing this.
Users who are particularly concerned about their security may also take the additional step of opening the request in the system browser from the in-app browser tab, and completing the authorization there, as most implementations of the in-app browser tab pattern offer such functionality. This is not expected to be common user behavior, however.
As stated in Section 10.2 of RFC 6749, the authorization server SHOULD NOT process authorization requests automatically without user consent or interaction, except when the identity of the client can be assured. Measures such as claimed HTTPS redirects can be used by native apps to prove their identity to the authorization server, and some operating systems may offer alternative platform-specific identity features which may be used, as appropriate.
This best practice recommends a particular type of external user-agent, the system browser. Other external user-agents patterns may also be viable for secure and usable OAuth. This document makes no comment on those patterns.
Secrets that are statically included as part of an app distributed to multiple users should not be treated as confidential secrets, as one user may inspect their copy and learn the secret of all users. For this reason it is NOT RECOMMENDED for authorization servers to require client authentication of native apps using a secret shared by multiple installs of the app, as this serves little value beyond client identification which is already provided by the client_id request parameter. If an authorization server requires a client secret for native apps, it MUST NOT assume that it is actually secret, unless some method is being used to dynamically provision a unique secret to each installation.
| [RFC6749] | Hardt, D., "The OAuth 2.0 Authorization Framework", RFC 6749, DOI 10.17487/RFC6749, October 2012. |
| [RFC7636] | Sakimura, N., Bradley, J. and N. Agarwal, "Proof Key for Code Exchange by OAuth Public Clients", RFC 7636, DOI 10.17487/RFC7636, September 2015. |
| [RFC2119] | Bradner, S., "Key words for use in RFCs to Indicate Requirement Levels", BCP 14, RFC 2119, DOI 10.17487/RFC2119, March 1997. |
| [RFC3986] | Berners-Lee, T., Fielding, R. and L. Masinter, "Uniform Resource Identifier (URI): Generic Syntax", STD 66, RFC 3986, DOI 10.17487/RFC3986, January 2005. |
| [RFC6819] | Lodderstedt, T., McGloin, M. and P. Hunt, OAuth 2.0 Threat Model and Security Considerations", RFC 6819, DOI 10.17487/RFC6819, January 2013. |
| [iOS.URIScheme] | Inter-App Communication", July 2016. | , "
| [macOS.URIScheme] | Launch Services Concepts", July 2016. | , "
| [Android.URIScheme] | Intents and Intent Filters", July 2016. | , "
| [WindowsUWP.URIScheme] | Handle URI activation", July 2016. | , "
| [IANA.URISchemes] | Uniform Resource Identifier (URI) Schemes", July 2016. | , "
| [ChromeCustomTab] | Chrome Custom Tabs", July 2016. | , "
| [SFSafariViewController] | SafariServices Changes", July 2016. | , "
| [Android.AppLinks] | App Links", July 2015. | , "
| [CustomTabsService] | CustomTabsService", July 2016. | , "
| [UniversalLinks] | Universal Links", July 2016. | , "
Most of this document attempts to lay out best practices in an generic manner, referencing technology available on most operating systems. This non-normative section contains OS-specific implementation details that are accurate at the time of authorship.
It is expected that this OS-specific information will change, but that the overall principles described in this document for using external user-agents will remain valid.
Claimed HTTPS and custom URI scheme redirects are both viable choices for OAuth on iOS. Developers can claim HTTPS links using Universal Links [UniversalLinks], available since iOS 9, and can use custom URI scheme [iOS.URIScheme] redirects for backwards compatibility. Clients SHOULD use Universal Links for authorization requests on iOS 9 and beyond, with the custom URI scheme redirect substituted on older versions. In both cases, the app claims the redirect in the application manifest.
As a user experience optimisation, since iOS 9, apps can invoke the system browser without the user leaving the app through SFSafariViewController [SFSafariViewController], which implements the browser-view pattern. This class has all the properties of the system browser, and is an 'external user-agent', even though it is presented within the host app. Regardless of whether the user completes the request in the system browser (as is their choice), or the SFSafariViewController, the return of the token via custom URI scheme or claimed HTTPS link is the same.
Claimed HTTPS and custom URI scheme redirects are both viable choices for OAuth on Android. Developers can claim HTTPS links using App Links [Android.AppLinks], available since Android 6.0 though browser support varies, and custom URI scheme [Android.URIScheme] redirects are broadly supported. Clients SHOULD support custom URI scheme redirects for broad compatibility and MAY upgrade to using claimed HTTPs redirects in supported environments. For both redirect options, the app claims the redirect in the application manifest.
As a user experience optimisation, apps SHOULD try to launch the authorization request in a Custom Tab. Custom Tab is an implementation of the browser-view pattern, providing a secure browser tab displayed in the context of the app. Chrome is an example of a browser that supports [ChromeCustomTab] CustomTabs.
Android Browser vendors SHOULD implement the CustomTabsService [CustomTabsService] to provide this functionality to their users.
Apps written on the Universal Windows Platform (UWP) can claim custom URI schemes [WindowsUWP.URIScheme] in their application manifest. This redirect choice will also open the app when the user taps the link. The scheme is limited to 39 characters, and may include the `.` character.
UWP apps can launch the authorization request in the user's default browser like so:
Uri authorizationRequest = ...
var success = Windows.System.Launcher.LaunchUriAsync(authorizationRequest)
The loopback IP redirect is a common choice for traditional Desktop apps, and listening on a loopback port is permitted by default Windows firewall rules.
Traditional apps can launch the URI in the user's default browser like so:
string authorizationRequest = ...
System.Diagnostics.Process.Start(authorizationRequest);
When using the "Process.Start" method, care must be taken that the input is a valid URL, including correct URI encoding of the parameters. This is especially important when the URL includes user-supplied information such as a login hint.
Both the loopback IP and custom URI scheme redirect choices are viable on macOS. Custom URI schemes [macOS.URIScheme] are registered in the application manifest. Listening on the loopback IP typically does not require any firewall changes.
Apps can launch the authorization request like so:
NSURL *authorizationRequest = ...
BOOL success = [[NSWorkspace sharedWorkspace] openURL:authorizationRequest];
The author would like to acknowledge the work of Marius Scurtescu, and Ben Wiley Sittler whose design for using custom URI schemes in native OAuth 2.0 clients formed the basis of Section 7.1.
The following individuals contributed ideas, feedback, and wording that shaped and formed the final specification:
Naveen Agarwal, Brian Campbell, Adam Dawes, Hannes Tschofenig, Ashish Jain, Paul Madsen, Breno de Medeiros, Eric Sachs, Nat Sakimura, Steve Wright, Erik Wahlstrom, Andy Zmolek, Sudhi Umarji.