Internet-Draft | oauth-security-topics | July 2022 |
Lodderstedt, et al. | Expires 29 January 2023 | [Page] |
This document describes best current security practice for OAuth 2.0. It updates and extends the OAuth 2.0 Security Threat Model to incorporate practical experiences gathered since OAuth 2.0 was published and covers new threats relevant due to the broader application of OAuth 2.0.¶
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 https://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 29 January 2023.¶
Copyright (c) 2022 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 (https://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 Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
Since its publication in [RFC6749] and [RFC6750], OAuth 2.0 ("OAuth" in the following) has gotten massive traction in the market and became the standard for API protection and the basis for federated login using OpenID Connect [OpenID.Core]. While OAuth is used in a variety of scenarios and different kinds of deployments, the following challenges can be observed:¶
OAuth implementations are being attacked through known implementation weaknesses and anti-patterns. Although most of these threats are discussed in the OAuth 2.0 Threat Model and Security Considerations [RFC6819], continued exploitation demonstrates a need for more specific recommendations, easier to implement mitigations, and more defense in depth.¶
OAuth is being used in environments with higher security requirements than considered initially, such as Open Banking, eHealth, eGovernment, and Electronic Signatures. Those use cases call for stricter guidelines and additional protection.¶
OAuth is being used in much more dynamic setups than originally anticipated, creating new challenges with respect to security. Those challenges go beyond the original scope of [RFC6749], [RFC6750], and [RFC6819].¶
OAuth initially assumed static relationships between client, authorization server, and resource servers. The URLs of the AS and RS were known to the client at deployment time and built an anchor for the trust relationships among those parties. The validation of whether the client talks to a legitimate server was based on TLS server authentication (see [RFC6819], Section 4.5.4). With the increasing adoption of OAuth, this simple model dissolved and, in several scenarios, was replaced by a dynamic establishment of the relationship between clients on one side and the authorization and resource servers of a particular deployment on the other side. This way, the same client could be used to access services of different providers (in case of standard APIs, such as e-mail or OpenID Connect) or serve as a front end to a particular tenant in a multi-tenant environment. Extensions of OAuth, such as the OAuth 2.0 Dynamic Client Registration Protocol [RFC7591] and OAuth 2.0 Authorization Server Metadata [RFC8414] were developed to support the use of OAuth in dynamic scenarios.¶
Technology has changed. For example, the way browsers treat fragments when redirecting requests has changed, and with it, the implicit grant's underlying security model.¶
This document provides updated security recommendations to address these challenges. It does not supplant the security advice given in [RFC6749], [RFC6750], and [RFC6819], but complements those documents.¶
This document introduces new requirements beyond those defined in existing specifications such as OAuth 2.0 [RFC6749] and OpenID Connect [OpenID.Core] and deprecates some modes of operation that are deemed less secure or even insecure. Naturally, not all existing ecosystems and implementations are compatible with the new requirements and following the best practices described in this document may break interoperability. Nonetheless, it is RECOMMENDED that implementers upgrade their implementations and ecosystems when feasible.¶
The remainder of this document is organized as follows: The next section summarizes the most important best practices for every OAuth implementor. Afterwards, the updated the OAuth attacker model is presented. Subsequently, a detailed analysis of the threats and implementation issues that can be found in the wild today is given along with a discussion of potential countermeasures.¶
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 BCP 14 [RFC2119] [RFC8174] when, and only when, they appear in all capitals, as shown here.¶
This specification uses the terms "access token", "authorization endpoint", "authorization grant", "authorization server", "client", "client identifier" (client ID), "protected resource", "refresh token", "resource owner", "resource server", and "token endpoint" defined by OAuth 2.0 [RFC6749].¶
This section describes the set of security mechanisms and measures the OAuth working group considers best practices at the time of writing.¶
When comparing client redirect URIs against pre-registered URIs, authorization
servers MUST utilize exact string matching except for port numbers in
localhost
redirection URIs of native apps, see Section 4.1.3. This
measure contributes to the prevention of leakage of authorization codes and
access tokens (see Section 4.1). It can also help to detect
mix-up attacks (see Section 4.4).¶
Clients and AS MUST NOT expose URLs that forward the user's browser to arbitrary URIs obtained from a query parameter ("open redirector"). Open redirectors can enable exfiltration of authorization codes and access tokens, see Section 4.10.1.¶
Clients MUST prevent Cross-Site Request Forgery (CSRF). In this
context, CSRF refers to requests to the redirection endpoint that do
not originate at the authorization server, but a malicious third party
(see Section 4.4.1.8. of [RFC6819] for details). Clients that have
ensured that the authorization server supports PKCE [RFC7636] MAY
rely on the CSRF protection provided by PKCE. In OpenID Connect flows,
the nonce
parameter provides CSRF protection. Otherwise, one-time
use CSRF tokens carried in the state
parameter that are securely
bound to the user agent MUST be used for CSRF protection (see
Section 4.7.1).¶
When an OAuth client can interact with more than one authorization server, a defense against mix-up attacks (see Section 4.4) is REQUIRED. To this end, clients SHOULD¶
use the iss
parameter as a countermeasure according to
[RFC9207], or¶
use an alternative countermeasure based on an iss
value in the
authorization response (such as the iss
Claim in the ID Token in
[OpenID.Core] or in [JARM] responses), processing it as described in
[RFC9207].¶
In the absence of these options, clients MAY instead use distinct redirect URIs to identify authorization endpoints and token endpoints, as described in Section 4.4.2.¶
An AS that redirects a request potentially containing user credentials MUST avoid forwarding these user credentials accidentally (see Section 4.11 for details).¶
Clients MUST prevent authorization code injection attacks (see Section 4.5) and misuse of authorization codes using one of the following options:¶
Public clients MUST use PKCE [RFC7636] to this end, as motivated in Section 4.5.3.1.¶
For confidential clients, the use of PKCE [RFC7636] is RECOMMENDED, as it provides a strong protection against misuse and injection of authorization codes as described in Section 4.5.3.1 and, as a side-effect, prevents CSRF even in presence of strong attackers as described in Section 4.7.1.¶
With additional precautions, described in Section 4.5.3.2,
confidential OpenID Connect [OpenID.Core] clients MAY use the nonce
parameter and the
respective Claim in the ID Token instead.¶
In any case, the PKCE challenge or OpenID Connect nonce
MUST be
transaction-specific and securely bound to the client and the user agent in
which the transaction was started.¶
Note: Although PKCE was designed as a mechanism to protect native apps, this advice applies to all kinds of OAuth clients, including web applications.¶
When using PKCE, clients SHOULD use PKCE code challenge methods that
do not expose the PKCE verifier in the authorization request.
Otherwise, attackers that can read the authorization request (cf.
Attacker A4 in Section 3) can break the security provided
by PKCE. Currently, S256
is the only such method.¶
Authorization servers MUST support PKCE [RFC7636].¶
If a client sends a valid PKCE [RFC7636] code_challenge
parameter in the
authorization request, the authorization server MUST enforce the correct usage
of code_verifier
at the token endpoint.¶
Authorization servers MUST mitigate PKCE Downgrade Attacks by ensuring that a
token request containing a code_verifier
parameter is accepted only if a
code_challenge
parameter was present in the authorization request, see
Section 4.8.2 for details.¶
Authorization servers MUST provide a way to detect their support for
PKCE. It is RECOMMENDED for AS to publish the element
code_challenge_methods_supported
in their AS metadata ([RFC8414])
containing the supported PKCE challenge methods (which can be used by
the client to detect PKCE support). ASs MAY instead provide a
deployment-specific way to ensure or determine PKCE support by the AS.¶
The implicit grant (response type "token") and other response types causing the authorization server to issue access tokens in the authorization response are vulnerable to access token leakage and access token replay as described in Section 4.1, Section 4.2, Section 4.3, and Section 4.6.¶
Moreover, no viable method for sender-constraining exists to bind access tokens to a specific client (as recommended in Section 2.2) when the access tokens are issued in the authorization response. This means that an attacker can use leaked or stolen access token at a resource endpoint.¶
In order to avoid these issues, clients SHOULD NOT use the implicit grant (response type "token") or other response types issuing access tokens in the authorization response, unless access token injection in the authorization response is prevented and the aforementioned token leakage vectors are mitigated.¶
Clients SHOULD instead use the response type "code" (aka authorization code grant type) as specified in Section 2.1.1 or any other response type that causes the authorization server to issue access tokens in the token response, such as the "code id_token" response type. This allows the authorization server to detect replay attempts by attackers and generally reduces the attack surface since access tokens are not exposed in URLs. It also allows the authorization server to sender-constrain the issued tokens (see next section).¶
A sender-constrained access token scopes the applicability of an access token to a certain sender. This sender is obliged to demonstrate knowledge of a certain secret as prerequisite for the acceptance of that token at the recipient (e.g., a resource server).¶
Authorization and resource servers SHOULD use mechanisms for sender-constraining access tokens, such as Mutual TLS for OAuth 2.0 [RFC8705] or OAuth Demonstration of Proof of Possession (DPoP) [I-D.ietf-oauth-dpop] (see Section 4.9.1.1.2), to prevent misuse of stolen and leaked access tokens.¶
Refresh tokens for public clients MUST be sender-constrained or use refresh token rotation as described in Section 4.13. [RFC6749] already mandates that refresh tokens for confidential clients can only be used by the client for which they were issued.¶
The privileges associated with an access token SHOULD be restricted to the minimum required for the particular application or use case. This prevents clients from exceeding the privileges authorized by the resource owner. It also prevents users from exceeding their privileges authorized by the respective security policy. Privilege restrictions also help to reduce the impact of access token leakage.¶
In particular, access tokens SHOULD be restricted to certain resource
servers (audience restriction), preferably to a single resource
server. To put this into effect, the authorization server associates
the access token with certain resource servers and every resource
server is obliged to verify, for every request, whether the access
token sent with that request was meant to be used for that particular
resource server. If not, the resource server MUST refuse to serve the
respective request. The aud
claim as defined in [RFC9068] MAY be
used to audience-restrict access tokens. Clients and authorization servers MAY utilize the
parameters scope
or resource
as specified in [RFC6749] and
[RFC8707], respectively, to determine the
resource server they want to access.¶
Additionally, access tokens SHOULD be restricted to certain resources
and actions on resource servers or resources. To put this into effect,
the authorization server associates the access token with the
respective resource and actions and every resource server is obliged
to verify, for every request, whether the access token sent with that
request was meant to be used for that particular action on the
particular resource. If not, the resource server must refuse to serve
the respective request. Clients and authorization servers MAY utilize
the parameter scope
as specified in [RFC6749] and authorization_details
as specified in [I-D.ietf-oauth-rar] to determine those
resources and/or actions.¶
The resource owner password credentials grant MUST NOT be used. This grant type insecurely exposes the credentials of the resource owner to the client. Even if the client is benign, this results in an increased attack surface (credentials can leak in more places than just the AS) and users are trained to enter their credentials in places other than the AS.¶
Furthermore, adapting the resource owner password credentials grant to two-factor authentication, authentication with cryptographic credentials (cf. WebCrypto [WebCrypto], WebAuthn [WebAuthn]), and authentication processes that require multiple steps can be hard or impossible.¶
Authorization servers SHOULD use client authentication if possible.¶
It is RECOMMENDED to use asymmetric (public-key based) methods for
client authentication such as mTLS [RFC8705] or
private_key_jwt
[OpenID.Core]. When asymmetric methods for client
authentication are used, authorization servers do not need to store
sensitive symmetric keys, making these methods more robust against a
number of attacks.¶
The use of OAuth Metadata [RFC8414] can help to improve the security of OAuth deployments:¶
It ensures that security features and other new OAuth features can be enabled automatically by compliant software libraries.¶
It reduces chances for misconfigurations, for example misconfigured endpoint URLs (that might belong to an attacker) or misconfigured security features.¶
It can help to facilitate rotation of cryptographic keys and to ensure cryptographic agility.¶
It is therefore RECOMMENDED that ASs publish OAuth metadata according to [RFC8414] and that clients make use of this metadata to configure themselves when available.¶
Authorization servers SHOULD NOT allow clients to influence their client_id
or
any other Claim if that can cause confusion with a genuine resource owner, as
described in Section 4.14¶
It is RECOMMENDED to use end-to-end TLS. If TLS traffic needs to be terminated at an intermediary, refer to Section 4.12 for further security advice.¶
Authorization responses MUST NOT be transmitted over unencrypted network
connections. To this end, AS MUST NOT allow redirect URIs that use the http
scheme except for native clients that use Loopback Interface Redirection as
described in [RFC8252], Section 7.3.¶
In [RFC6819], an attacker model is laid out that describes the capabilities of attackers against which OAuth deployments must be protected. In the following, this attacker model is updated to account for the potentially dynamic relationships involving multiple parties (as described in Section 1), to include new types of attackers and to define the attacker model more clearly.¶
OAuth MUST ensure that the authorization of the resource owner (RO) (with a user agent) at the authorization server (AS) and the subsequent usage of the access token at the resource server (RS) is protected at least against the following attackers:¶
(A1) Web Attackers that can set up and operate an arbitrary number of network endpoints including browsers and servers (except for the concrete RO, AS, and RS). Web attackers may set up web sites that are visited by the RO, operate their own user agents, and participate in the protocol.¶
Web attackers may, in particular, operate OAuth clients that are registered at AS, and operate their own authorization and resource servers that can be used (in parallel) by the RO and other resource owners.¶
It must also be assumed that web attackers can lure the user to open arbitrary attacker-chosen URIs at any time. In practice, this can be achieved in many ways, for example, by injecting malicious advertisements into advertisement networks, or by sending legitimate-looking emails.¶
Web attackers can use their own user credentials to create new messages as well as any secrets they learned previously. For example, if a web attacker learns an authorization code of a user through a misconfigured redirect URI, the web attacker can then try to redeem that code for an access token.¶
They cannot, however, read or manipulate messages that are not targeted towards them (e.g., sent to a URL controlled by a non-attacker controlled AS).¶
(A2) Network Attackers that additionally have full control over the network over which protocol participants communicate. They can eavesdrop on, manipulate, and spoof messages, except when these are properly protected by cryptographic methods (e.g., TLS). Network attackers can also block arbitrary messages.¶
While an example for a web attacker would be a customer of an internet service provider, network attackers could be the internet service provider itself, an attacker in a public (wifi) network using ARP spoofing, or a state-sponsored attacker with access to internet exchange points, for instance.¶
These attackers conform to the attacker model that was used in formal analysis efforts for OAuth [arXiv.1601.01229]. This is a minimal attacker model. Implementers MUST take into account all possible types of attackers in the environment in which their OAuth implementations are expected to run. Previous attacks on OAuth have shown that OAuth deployments SHOULD in particular consider the following, stronger attackers in addition to those listed above:¶
(A3) Attackers that can read, but not modify, the contents of the authorization response (i.e., the authorization response can leak to an attacker).¶
Examples for such attacks include open redirector attacks, insufficient checking of redirect URIs (see Section 4.1), problems existing on mobile operating systems (where different apps can register themselves on the same URI), mix-up attacks (see Section 4.4), where the client is tricked into sending credentials to a attacker-controlled AS, and the fact that URLs are often stored/logged by browsers (history), proxy servers, and operating systems.¶
(A4) Attackers that can read, but not modify, the contents of the authorization request (i.e., the authorization request can leak, in the same manner as above, to an attacker).¶
(A5) Attackers that can acquire an access token issued by AS. For example, a resource server can be compromised by an attacker, an access token may be sent to an attacker-controlled resource server due to a misconfiguration, or an RO is social-engineered into using a attacker-controlled RS. See also Section 4.9.2.¶
(A3), (A4) and (A5) typically occur together with either (A1) or (A2). Attackers can collaborate to reach a common goal.¶
Note that in this attacker model, an attacker (see A1) can be a RO or act as one. For example, an attacker can use his own browser to replay tokens or authorization codes obtained by any of the attacks described above at the client or RS.¶
This document focusses on threats resulting from these attackers. Attacks in an even stronger attacker model are discussed, for example, in [arXiv.1901.11520].¶
This section gives a detailed description of attacks on OAuth implementations, along with potential countermeasures. Attacks and mitigations already covered in [RFC6819] are not listed here, except where new recommendations are made.¶
Some authorization servers allow clients to register redirect URI patterns instead of complete redirect URIs. The authorization servers then match the redirect URI parameter value at the authorization endpoint against the registered patterns at runtime. This approach allows clients to encode transaction state into additional redirect URI parameters or to register a single pattern for multiple redirect URIs.¶
This approach turned out to be more complex to implement and more error prone to manage than exact redirect URI matching. Several successful attacks exploiting flaws in the pattern matching implementation or concrete configurations have been observed in the wild . Insufficient validation of the redirect URI effectively breaks client identification or authentication (depending on grant and client type) and allows the attacker to obtain an authorization code or access token, either¶
by directly sending the user agent to a URI under the attackers control, or¶
by exposing the OAuth credentials to an attacker by utilizing an open redirector at the client in conjunction with the way user agents handle URL fragments.¶
These attacks are shown in detail in the following subsections.¶
For a client using the grant type code, an attack may work as follows:¶
Assume the redirect URL pattern https://*.somesite.example/*
is
registered for the client with the client ID s6BhdRkqt3
. The
intention is to allow any subdomain of somesite.example
to be a
valid redirect URI for the client, for example
https://app1.somesite.example/redirect
. A naive implementation on
the authorization server, however, might interpret the wildcard *
as
"any character" and not "any character valid for a domain name". The
authorization server, therefore, might permit
https://attacker.example/.somesite.example
as a redirect URI,
although attacker.example
is a different domain potentially
controlled by a malicious party.¶
The attack can then be conducted as follows:¶
First, the attacker needs to trick the user into opening a tampered
URL in his browser that launches a page under the attacker's
control, say https://www.evil.example
(see Attacker A1 in Section 3).¶
This URL initiates the following authorization request with the client ID of a legitimate client to the authorization endpoint (line breaks for display only):¶
GET /authorize?response_type=code&client_id=s6BhdRkqt3&state=9ad67f13 &redirect_uri=https%3A%2F%2Fattacker.example%2F.somesite.example HTTP/1.1 Host: server.somesite.example¶
The authorization server validates the redirect URI and compares it to
the registered redirect URL patterns for the client s6BhdRkqt3
.
The authorization request is processed and presented to the user.¶
If the user does not see the redirect URI or does not recognize the attack, the code is issued and immediately sent to the attacker's domain. If an automatic approval of the authorization is enabled (which is not recommended for public clients according to [RFC6749]), the attack can be performed even without user interaction.¶
If the attacker impersonated a public client, the attacker can exchange the code for tokens at the respective token endpoint.¶
This attack will not work as easily for confidential clients, since the code exchange requires authentication with the legitimate client's secret. The attacker can, however, use the legitimate confidential client to redeem the code by performing an authorization code injection attack, see Section 4.5.¶
Note: Vulnerabilities of this kind can also exist if the authorization
server handles wildcards properly. For example, assume that the client
registers the redirect URL pattern https://*.somesite.example/*
and
the authorization server interprets this as "allow redirect URIs
pointing to any host residing in the domain somesite.example
". If an
attacker manages to establish a host or subdomain in
somesite.example
, he can impersonate the legitimate client. This
could be caused, for example, by a subdomain takeover attack [subdomaintakeover], where an
outdated CNAME record (say, external-service.somesite.example
)
points to an external DNS name that does no longer exist (say,
customer-abc.service.example
) and can be taken over by an attacker
(e.g., by registering as customer-abc
with the external service).¶
The attack described above works for the implicit grant as well. If the attacker is able to send the authorization response to a URI under his control, he will directly get access to the fragment carrying the access token.¶
Additionally, implicit clients can be subject to a further kind of attack. It utilizes the fact that user agents re-attach fragments to the destination URL of a redirect if the location header does not contain a fragment (see [RFC7231], Section 9.5). The attack described here combines this behavior with the client as an open redirector (see Section 4.10.1) in order to get access to access tokens. This allows circumvention even of very narrow redirect URI patterns, but not strict URL matching.¶
Assume the registered URL pattern for client s6BhdRkqt3
is
https://client.somesite.example/cb?*
, i.e., any parameter is allowed
for redirects to https://client.somesite.example/cb
. Unfortunately,
the client exposes an open redirector. This endpoint supports a
parameter redirect_to
which takes a target URL and will send the
browser to this URL using an HTTP Location header redirect 303.¶
The attack can now be conducted as follows:¶
First, and as above, the attacker needs to trick the user into opening
a tampered URL in his browser that launches a page under the
attacker's control, say https://www.evil.example
.¶
Afterwards, the website initiates an authorization request that is
very similar to the one in the attack on the code flow. Different to
above, it utilizes the open redirector by encoding
redirect_to=https://attacker.example
into the parameters of the
redirect URI and it uses the response type "token" (line breaks for display only):¶
GET /authorize?response_type=token&state=9ad67f13 &client_id=s6BhdRkqt3 &redirect_uri=https%3A%2F%2Fclient.somesite.example %2Fcb%26redirect_to%253Dhttps%253A%252F %252Fattacker.example%252F HTTP/1.1 Host: server.somesite.example¶
Now, since the redirect URI matches the registered pattern, the authorization server permits the request and sends the resulting access token in a 303 redirect (some response parameters omitted for readability):¶
HTTP/1.1 303 See Other Location: https://client.somesite.example/cb? redirect_to%3Dhttps%3A%2F%2Fattacker.example%2Fcb #access_token=2YotnFZFEjr1zCsicMWpAA&...¶
At example.com, the request arrives at the open redirector. The endpoint will
read the redirect parameter and will issue an HTTP 303 Location header
redirect to the URL https://attacker.example/
.¶
HTTP/1.1 303 See Other Location: https://attacker.example/¶
Since the redirector at client.somesite.example does not include a
fragment in the Location header, the user agent will re-attach the
original fragment #access_token=2YotnFZFEjr1zCsicMWpAA&...
to
the URL and will navigate to the following URL:¶
https://attacker.example/#access_token=2YotnFZFEjr1z...¶
The attacker's page at attacker.example
can now access the
fragment and obtain the access token.¶
The complexity of implementing and managing pattern matching correctly obviously
causes security issues. This document therefore advises to simplify the required
logic and configuration by using exact redirect URI matching. This means
the authorization server MUST compare the two URIs using simple string
comparison as defined in [RFC3986], Section 6.2.1. The only exception are
native apps using a localhost
URI: In this case, the AS MUST allow variable
port numbers as described in [RFC8252], Section 7.3.¶
Additional recommendations:¶
Servers on which callbacks are hosted MUST NOT expose open redirectors (see Section 4.10).¶
Browsers reattach URL fragments to Location redirection URLs only
if the URL in the Location header does not already contain a fragment.
Therefore, servers MAY prevent browsers from reattaching fragments
to redirection URLs by attaching an arbitrary fragment identifier,
for example #_
, to URLs in Location headers.¶
Clients SHOULD use the authorization code response type instead of response types causing access token issuance at the authorization endpoint. This offers countermeasures against reuse of leaked credentials through the exchange process with the authorization server and token replay through sender-constraining of the access tokens.¶
If the origin and integrity of the authorization request containing the redirect URI can be verified, for example when using [RFC9101] or [RFC9126] with client authentication, the authorization server MAY trust the redirect URI without further checks.¶
The contents of the authorization request URI or the authorization
response URI can unintentionally be disclosed to attackers through the
Referer HTTP header (see [RFC7231], Section 5.5.2), by leaking either
from the AS's or the client's web site, respectively. Most
importantly, authorization codes or state
values can be disclosed in
this way. Although specified otherwise in [RFC7231], Section 5.5.2,
the same may happen to access tokens conveyed in URI fragments due to
browser implementation issues, as illustrated by Chromium Issue 168213
[bug.chromium].¶
Leakage from the OAuth client requires that the client, as a result of a successful authorization request, renders a page that¶
contains links to other pages under the attacker's control and a user clicks on such a link, or¶
includes third-party content (advertisements in iframes, images, etc.), for example if the page contains user-generated content (blog).¶
As soon as the browser navigates to the attacker's page or loads the
third-party content, the attacker receives the authorization response
URL and can extract code
or state
(and potentially access token
).¶
An attacker that learns a valid code or access token through a
Referer header can perform the attacks as described in
Section 4.1.1, Section 4.5, and
Section 4.6. If the attacker learns state
, the CSRF
protection achieved by using state
is lost, resulting in CSRF
attacks as described in [RFC6819], Section 4.4.1.8.¶
The page rendered as a result of the OAuth authorization response and the authorization endpoint SHOULD NOT include third-party resources or links to external sites.¶
The following measures further reduce the chances of a successful attack:¶
Suppress the Referer header by applying an appropriate Referrer
Policy [webappsec-referrer-policy] to the document (either as
part of the "referrer" meta attribute or by setting a
Referrer-Policy header). For example, the header Referrer-Policy:
no-referrer
in the response completely suppresses the Referer
header in all requests originating from the resulting document.¶
Use authorization code instead of response types causing access token issuance from the authorization endpoint.¶
Bind the authorization code to a confidential client or PKCE challenge. In this case, the attacker lacks the secret to request the code exchange.¶
As described in [RFC6749], Section 4.1.2, authorization codes MUST be invalidated by the AS after their first use at the token endpoint. For example, if an AS invalidated the code after the legitimate client redeemed it, the attacker would fail exchanging this code later.¶
This does not mitigate the attack if the attacker manages to exchange the code for a token before the legitimate client does so. Therefore, [RFC6749] further recommends that, when an attempt is made to redeem a code twice, the AS SHOULD revoke all tokens issued previously based on that code.¶
The state
value SHOULD be invalidated by the client after its
first use at the redirection endpoint. If this is implemented, and
an attacker receives a token through the Referer header from the
client's web site, the state
was already used, invalidated by
the client and cannot be used again by the attacker. (This does
not help if the state
leaks from the
AS's web site, since then the state
has not been used at the redirection endpoint at the client yet.)¶
Use the form post response mode instead of a redirect for the authorization response (see [OAuth.Post]).¶
Authorization codes and access tokens can end up in the browser's history of visited URLs, enabling the attacks described in the following.¶
An access token may end up in the browser history if a client or a web
site that already has a token deliberately navigates to a page like
provider.com/get_user_profile?access_token=abcdef
. [RFC6750]
discourages this practice and advises to transfer tokens via a header,
but in practice web sites often pass access tokens in query
parameters.¶
In case of the implicit grant, a URL like
client.example/redirection_endpoint#access_token=abcdef
may also end
up in the browser history as a result of a redirect from a provider's
authorization endpoint.¶
Countermeasures:¶
Clients MUST NOT pass access tokens in a URI query parameter in the way described in Section 2.3 of [RFC6750]. The authorization code grant or alternative OAuth response modes like the form post response mode [OAuth.Post] can be used to this end.¶
Mix-up is an attack on scenarios where an OAuth client interacts with two or more authorization servers and at least one authorization server is under the control of the attacker. This can be the case, for example, if the attacker uses dynamic registration to register the client at his own authorization server or if an authorization server becomes compromised.¶
The goal of the attack is to obtain an authorization code or an access token for an uncompromised authorization server. This is achieved by tricking the client into sending those credentials to the compromised authorization server (the attacker) instead of using them at the respective endpoint of the uncompromised authorization/resource server.¶
The description here follows [arXiv.1601.01229], with variants of the attack outlined below.¶
Preconditions: For this variant of the attack to work, we assume that¶
the implicit or authorization code grant are used with multiple AS of which one is considered "honest" (H-AS) and one is operated by the attacker (A-AS), and¶
the client stores the AS chosen by the user in a session bound to the user's browser and uses the same redirection endpoint URI for each AS.¶
In the following, we assume that the client is registered with H-AS (URI:
https://honest.as.example
, client ID: 7ZGZldHQ
) and with A-AS (URI:
https://attacker.example
, client ID: 666RVZJTA
). URLs shown in the following
example are shortened for presentation to only include parameters relevant for the
attack.¶
Attack on the authorization code grant:¶
The user selects to start the grant using A-AS (e.g., by clicking on a button at the client's website).¶
The client stores in the user's session that the user selected
"A-AS" and redirects the user to A-AS's authorization endpoint
with a Location header containing the URL
https://attacker.example/authorize?response_type=code&client_id=666RVZJTA
.¶
When the user's browser navigates to the attacker's authorization endpoint,
the attacker immediately redirects the browser to the authorization endpoint
of H-AS. In the authorization request, the attacker replaces the client ID
of the client at A-AS with the client's ID at H-AS. Therefore, the browser
receives a redirection (303 See Other
) with a Location header pointing to
https://honest.as.example/authorize?response_type=code&client_id=7ZGZldHQ
¶
The user authorizes the client to access her resources at H-AS. (Note that a vigilant user might at this point detect that she intended to use A-AS instead of H-AS. The first attack variant listed below avoids this.) H-AS issues a code and sends it (via the browser) back to the client.¶
Since the client still assumes that the code was issued by A-AS, it will try to redeem the code at A-AS's token endpoint.¶
The attacker therefore obtains code and can either exchange the code for an access token (for public clients) or perform an authorization code injection attack as described in Section 4.5.¶
Variants:¶
Mix-Up With Interception: This variant works only if the attacker can intercept and manipulate the first request/response pair from a user's browser to the client (in which the user selects a certain AS and is then redirected by the client to that AS), as in Attacker A2 (see Section 3). This capability can, for example, be the result of a man-in-the-middle attack on the user's connection to the client. In the attack, the user starts the flow with H-AS. The attacker intercepts this request and changes the user's selection to A-AS. The rest of the attack proceeds as in Steps 2 and following above.¶
Implicit Grant: In the implicit grant, the attacker receives an access token instead of the code; the rest of the attack works as above.¶
Per-AS Redirect URIs: If clients use different redirect URIs for different ASs, do not store the selected AS in the user's session, and ASs do not check the redirect URIs properly, attackers can mount an attack called "Cross-Social Network Request Forgery". These attacks have been observed in practice. Refer to [oauth_security_jcs_14] for details.¶
OpenID Connect: There are variants that can be used to attack OpenID Connect. In these attacks, the attacker misuses features of the OpenID Connect Discovery [OpenID.Discovery] mechanism or replays access tokens or ID Tokens to conduct a mix-up attack. The attacks are described in detail in [arXiv.1704.08539], Appendix A, and [arXiv.1508.04324v2], Section 6 ("Malicious Endpoints Attacks").¶
When an OAuth client can only interact with one authorization server, a mix-up defense is not required. In scenarios where an OAuth client interacts with two or more authorization servers, however, clients MUST prevent mix-up attacks. Two different methods are discussed in the following.¶
For both defenses, clients MUST store, for each authorization request, the issuer they sent the authorization request to and bind this information to the user agent. The issuer serves, via the associated metadata, as an abstract identifier for the combination of the authorization endpoint and token endpoint that are to be used in the flow. If an issuer identifier is not available, for example, if neither OAuth metadata [RFC8414] nor OpenID Connect Discovery [OpenID.Discovery] are used, a different unique identifier for this tuple or the tuple itself can be used instead. For brevity of presentation, such a deployment-specific identifier will be subsumed under the issuer (or issuer identifier) in the following.¶
Note: Just storing the authorization server URL is not sufficient to identify mix-up attacks. An attacker might declare an uncompromised AS's authorization endpoint URL as "his" AS URL, but declare a token endpoint under his own control.¶
This defense requires that the authorization server sends his issuer identifier in the authorization response to the client. When receiving the authorization response, the client MUST compare the received issuer identifier to the stored issuer identifier. If there is a mismatch, the client MUST abort the interaction.¶
There are different ways this issuer identifier can be transported to the client:¶
The issuer information can be transported, for
example, via a separate response parameter iss
, defined in
[RFC9207].¶
When OpenID Connect is used and an ID Token is returned in the authorization
response, the client can evaluate the iss
Claim in the ID Token.¶
In both cases, the iss
value MUST be evaluated according to [RFC9207].¶
While this defense may require deploying new OAuth features to transport the issuer information, it is a robust and relatively simple defense against mix-up.¶
For this defense, clients MUST use a distinct redirect URI for each issuer they interact with.¶
Clients MUST check that the authorization response was received from the correct issuer by comparing the distinct redirect URI for the issuer to the URI where the authorization response was received on. If there is a mismatch, the client MUST abort the flow.¶
While this defense builds upon existing OAuth functionality, it cannot be used in scenarios where clients only register once for the use of many different issuers (as in some open banking schemes) and due to the tight integration with the client registration, it is harder to deploy automatically.¶
Furthermore, an attacker might be able to circumvent the protection offered by this defense by registering a new client with the "honest" AS using the redirect URI that the client assigned to the attacker's AS. The attacker could then run the attack as described above, replacing the client ID with the client ID of his newly created client.¶
This defense SHOULD therefore only be used if other options are not available.¶
An attacker that has gained access to an authorization code contained in an authorization response (see Attacker A3 in Section 3) can try to redeem the authorization code for an access token or otherwise make use of the authorization code.¶
In the case that the authorization code was created for a public client, the attacker can send the authorization code to the token endpoint of the authorization server and thereby get an access token. This attack was described in Section 4.4.1.1 of [RFC6819].¶
For confidential clients, or in some special situations, the attacker can execute an authorization code injection attack, as described in the following.¶
In an authorization code injection attack, the attacker attempts to inject a stolen authorization code into the attacker's own session with the client. The aim is to associate the attacker's session at the client with the victim's resources or identity, thereby giving the attacker at least limited access to the victum's resources.¶
Besides circumventing the client authentication of confidential clients, other use cases for this attack include:¶
The attacker wants to access certain functions in this particular client. As an example, the attacker wants to impersonate his victim in a certain app or on a certain web site.¶
The authorization or resource servers are limited to certain networks that the attacker is unable to access directly.¶
Except in these special cases, authorization code injection is usually not interesting when the code was created for a public client, as sending the code to the token endpoint is a simpler and more powerful attack, as described above.¶
The authorization code injection attack works as follows:¶
The attacker obtains an authorization code (see attacker A3 in Section 3). For the rest of the attack, only the capabilities of a web attacker (A1) are required.¶
From the attacker's own device, the attacker starts a regular OAuth authorization process with the legitimate client.¶
In the response of the authorization server to the legitimate client, the attacker replaces the newly created authorization code with the stolen authorization code. Since this response is passing through the attacker's device, the attacker can use any tool that can intercept and manipulate the authorization response to this end. The attacker does not need to control the network.¶
The legitimate client sends the code to the authorization server's token
endpoint, along with the redirect_uri
and the client's client ID and
client secret (or other means of client authentication).¶
The authorization server checks the client secret, whether the
code was issued to the particular client, and whether the actual
redirect URI matches the redirect_uri
parameter (see
[RFC6749]).¶
All checks succeed and the authorization server issues access and other tokens to the client. The attacker has now associated his session with the legitimate client with the victim's resources and/or identity.¶
Obviously, the check in step (5.) will fail if the code was issued to another client ID, e.g., a client set up by the attacker. The check will also fail if the authorization code was already redeemed by the legitimate user and was one-time use only.¶
An attempt to inject a code obtained via a manipulated redirect URI
should also be detected if the authorization server stored the
complete redirect URI used in the authorization request and compares
it with the redirect_uri
parameter.¶
[RFC6749], Section 4.1.3, requires the AS to "... ensure that the
redirect_uri
parameter is present if the redirect_uri
parameter
was included in the initial authorization request as described in
Section 4.1.1, and if included ensure that their values are
identical.". In the attack scenario described above, the legitimate
client would use the correct redirect URI it always uses for
authorization requests. But this URI would not match the tampered
redirect URI used by the attacker (otherwise, the redirect would not
land at the attackers page). So the authorization server would detect
the attack and refuse to exchange the code.¶
Note: This check could also detect attempts to inject an authorization code that had been obtained from another instance of the same client on another device, if certain conditions are fulfilled:¶
the redirect URI itself needs to contain a nonce or another kind of one-time use, secret data and¶
the client has bound this data to this particular instance of the client.¶
But this approach conflicts with the idea to enforce exact redirect
URI matching at the authorization endpoint. Moreover, it has been
observed that providers very often ignore the redirect_uri
check
requirement at this stage, maybe because it doesn't seem to be
security-critical from reading the specification.¶
Other providers just pattern match the redirect_uri
parameter
against the registered redirect URI pattern. This saves the
authorization server from storing the link between the actual redirect
URI and the respective authorization code for every transaction. But
this kind of check obviously does not fulfill the intent of the
specification, since the tampered redirect URI is not considered. So
any attempt to inject an authorization code obtained using the
client_id
of a legitimate client or by utilizing the legitimate
client on another device will not be detected in the respective
deployments.¶
It is also assumed that the requirements defined in [RFC6749], Section 4.1.3, increase client implementation complexity as clients need to store or re-construct the correct redirect URI for the call to the token endpoint.¶
Asymmetric methods for client authentication do not stop this attack, as the legitimate client authenticates at the token endpoint.¶
This document therefore recommends to instead bind every authorization code to a certain client instance on a certain device (or in a certain user agent) in the context of a certain transaction using one of the mechanisms described next.¶
There are two good technical solutions to achieve this goal, outlined in the following.¶
The PKCE mechanism specified in [RFC7636] can be used as a countermeasure.
When the attacker attempts to inject an authorization code, the check of the
code_verifier
fails: the client uses its correct verifier, but the code is
associated with a code_challenge
that does not match this verifier. PKCE is a
deployed OAuth feature, although its originally intended use was solely focused
on securing native apps, not the broader use recommended by this document.¶
PKCE does not only protect against the autorization code injection attack, but
also protects authorization codes created for public clients: PKCE ensures that
an attacker cannot redeem a stolen authorization code at the token endpoint of
the authorization server without knowledge of the code_verifier
.¶
OpenID Connect's existing nonce
parameter can protect against authorization
code injection attacks. The nonce
value is one-time use and created by the
client. The client is supposed to bind it to the user agent session and sends it
with the initial request to the OpenID Provider (OP). The OP puts the received nonce
value into the ID Token that is issued
as part of the code exchange at the token endpoint. If an attacker injected an
authorization code in the authorization response, the nonce value in the client
session and the nonce value in the ID token will not match and the attack is
detected. The assumption is that an attacker cannot get hold of the user agent
state on the victim's device, where the attacker has stolen the respective authorization
code.¶
It is important to note that this countermeasure only works if the client
properly checks the nonce
parameter in the ID Token and does not use any
issued token until this check has succeeded. More precisely, a client protecting
itself against code injection using the nonce
parameter,¶
MUST validate the nonce
in the ID Token obtained from the token endpoint,
even if another ID Token was obtained from the authorization response
(e.g., response_type=code+id_token
), and¶
MUST ensure that, unless and until that check succeeds, all tokens (ID Tokens and the access token) are disregarded and not used for any other purpose.¶
It is important to note that nonce
does not protect authorization codes of
public clients, as an attacker does not need to execute an authorization code
injection attack. Instead, an attacker can directly call the token endpoint with
the stolen authorization code.¶
Other solutions, like binding state
to the code, sender-constraining the code
using cryptographic means, or per-instance client credentials are
conceivable, but lack support and bring new security requirements.¶
PKCE is the most obvious solution for OAuth clients as it is available
today (originally intended for OAuth native apps) whereas nonce
is
appropriate for OpenID Connect clients.¶
An attacker can circumvent the countermeasures described above if he
can modify the nonce
or code_challenge
values that are used in the
victim's authorization request. The attacker can modify these values
to be the same ones as those chosen by the client in his own session
in Step 2 of the attack above. (This requires that the victim's
session with the client begins after the attacker started his session
with the client.) If the attacker is then able to capture the
authorization code from the victim, the attacker will be able to
inject the stolen code in Step 3 even if PKCE or nonce
are used.¶
This attack is complex and requires a close interaction between the attacker and the victim's session. Nonetheless, measures to prevent attackers from reading the contents of the authorization response still need to be taken, as described in Section 4.1, Section 4.2, Section 4.3, Section 4.4, and Section 4.10.¶
In an access token injection attack, the attacker attempts to inject a stolen access token into a legitimate client (that is not under the attacker's control). This will typically happen if the attacker wants to utilize a leaked access token to impersonate a user in a certain client.¶
To conduct the attack, the attacker starts an OAuth flow with the
client using the implicit grant and modifies the authorization
response by replacing the access token issued by the authorization
server or directly makes up an authorization server response including
the leaked access token. Since the response includes the state
value
generated by the client for this particular transaction, the client
does not treat the response as a CSRF attack and uses the access token
injected by the attacker.¶
There is no way to detect such an injection attack in pure-OAuth flows, since the token is issued without any binding to the transaction or the particular user agent.¶
In OpenID Connect, the attack can be mitigated, as the authorization response
additionally contains an ID Token containing the at_hash
claim. The attacker
therefore needs to replace both the access token as well as the ID Token in the
response. The attacker cannot forge the ID Token, as it is signed or encrypted
with authentication. The attacker also cannot inject a leaked ID Token matching
the stolen access token, as the nonce
claim in the leaked ID Token will
(with a very high probability) contain a different value than the one expected
in the authorization response.¶
Note that further protection, like sender-constrained access tokens, is still required to prevent attackers from using the access token at the resource endpoint directly.¶
The recommendations in Section 2.1.2 follow from this.¶
An attacker might attempt to inject a request to the redirect URI of the legitimate client on the victim's device, e.g., to cause the client to access resources under the attacker's control. This is a variant of an attack known as Cross-Site Request Forgery (CSRF).¶
The traditional countermeasure is that clients pass a value in the state
parameter that links the request to the redirect URI to the user agent session
as described in detail in [RFC6819], Section 5.3.5. The same protection is
provided by PKCE or the OpenID Connect nonce
value.¶
When using PKCE instead of state
or nonce
for CSRF protection, it is
important to note that:¶
Clients MUST ensure that the AS supports PKCE before using PKCE for
CSRF protection. If an authorization server does not support PKCE,
state
or nonce
MUST be used for CSRF protection.¶
If state
is used for carrying application state, and integrity of
its contents is a concern, clients MUST protect state
against
tampering and swapping. This can be achieved by binding the
contents of state to the browser session and/or signed/encrypted
state values as discussed in the now-expired draft [I-D.bradley-oauth-jwt-encoded-state].¶
The AS therefore MUST provide a way to detect their support for PKCE. Using AS metadata according to [RFC8414] is RECOMMENDED, but AS MAY instead provide a deployment-specific way to ensure or determine PKCE support.¶
PKCE provides robust protection against CSRF attacks even in presence of an that
can read the authorization response (see Attacker A3 in Section 3). When
state
is used or an ID Token is returned in the authorization response (e.g.,
response_type=code+id_token
), the attacker either learns the state
value and
can replay it into the forged authorization response, or can extract the nonce
from the ID Token and use it in a new request to the authorization server to
mint an ID Token with the same nonce
. The new ID Token can then be used for
the CSRF attack.¶
An authorization server that supports PKCE but does not make its use mandatory for all flows can be susceptible to a PKCE downgrade attack.¶
The first prerequisite for this attack is that there is an attacker-controllable
flag in the authorization request that enables or disables PKCE for the
particular flow. The presence or absence of the code_challenge
parameter lends
itself for this purpose, i.e., the AS enables and enforces PKCE if this
parameter is present in the authorization request, but does not enforce PKCE if
the parameter is missing.¶
The second prerequisite for this attack is that the client is not using state
at all (e.g., because the client relies on PKCE for CSRF prevention) or that the
client is not checking state
correctly.¶
Roughly speaking, this attack is a variant of a CSRF attack. The attacker achieves the same goal as in the attack described in Section 4.7: The attacker injects an authorization code (and with that, an access token) that is bound to the attacker's resources into a session between his victim and the client.¶
The user has started an OAuth session using some client at an AS. In the
authorization request, the client has set the parameter
code_challenge=sha256(abc)
as the PKCE code challenge. The client is now
waiting to receive the authorization response from the user's browse.¶
To conduct the attack, the attacker uses his own device to start an
authorization flow with the targeted client. The client now uses another
PKCE code challenge, say code_challenge=sha256(xyz)
, in the authorization
request. The attacker intercepts the request and removes the entire
code_challenge
parameter from the request. Since this step is performed on
the attacker's device, the attacker has full access to the request contents,
for example using browser debug tools.¶
If the authorization server allows for flows without PKCE, it will create a code that is not bound to any PKCE code challenge.¶
The attacker now redirects the user's browser to an authorization response URL that contains the code for the attacker's session with the AS.¶
The user's browser sends the authorization code to the client, which will
now try to redeem the code for an access token at the AS. The client will
send code_verifier=abc
as the PKCE code verifier in the token request.¶
Since the authorization server sees that this code is not bound to any PKCE
code challenge, it will not check the presence or contents of the
code_verifier
parameter. It will issue an access token that belongs to the
attacker's resource to the client under the user's control.¶
Using state
properly would prevent this attack. However, practice has shown
that many OAuth clients do not use or check state
properly.¶
Therefore, ASs MUST take precautions against this threat.¶
Note that from the view of the AS, in the attack described above, a
code_verifier
parameter is received at the token endpoint although no
code_challenge
parameter was present in the authorization request for the
OAuth flow in which the authorization code was issued.¶
This fact can be used to mitigate this attack. [RFC7636] already mandates that¶
an AS that supports PKCE MUST check whether a code challenge is contained in the authorization request and bind this information to the code that is issued; and¶
when a code arrives at the token endpoint, and there was a code_challenge
in the authorization request for which this code was issued, there must be a
valid code_verifier
in the token request.¶
Beyond this, to prevent PKCE downgrade attacks, the AS MUST ensure that
if there was no code_challenge
in the authorization request, a request to
the token endpoint containing a code_verifier
is rejected.¶
Note: ASs that mandate the use of PKCE in general or for particular clients implicitly implement this security measure.¶
Access tokens can leak from a resource server under certain circumstances.¶
An attacker may setup his own resource server and trick a client into sending access tokens to it that are valid for other resource servers (see Attackers A1 and A5 in Section 3). If the client sends a valid access token to this counterfeit resource server, the attacker in turn may use that token to access other services on behalf of the resource owner.¶
This attack assumes the client is not bound to one specific resource server (and its URL) at development time, but client instances are provided with the resource server URL at runtime. This kind of late binding is typical in situations where the client uses a service implementing a standardized API (e.g., for e-Mail, calendar, health, or banking) and where the client is configured by a user or administrator for a service that this user or company uses.¶
There are several potential mitigation strategies, which will be discussed in the following sections.¶
An authorization server could provide the client with additional information about the locations where it is safe to use its access tokens.¶
In the simplest form, this would require the AS to publish a list of
its known resource servers, illustrated in the following example using
a non-standard metadata parameter resource_servers
:¶
HTTP/1.1 200 OK Content-Type: application/json { "issuer":"https://server.somesite.example", "authorization_endpoint": "https://server.somesite.example/authorize", "resource_servers":[ "email.somesite.example", "storage.somesite.example", "video.somesite.example" ] ... }¶
The AS could also return the URL(s) an access token is good for in the
token response, illustrated by the example and non-standard return
parameter access_token_resource_server
:¶
HTTP/1.1 200 OK Content-Type: application/json;charset=UTF-8 Cache-Control: no-store Pragma: no-cache { "access_token":"2YotnFZFEjr1zCsicMWpAA", "access_token_resource_server": "https://hostedresource.somesite.example/path1", ... }¶
This mitigation strategy would rely on the client to enforce the
security policy and to only send access tokens to legitimate
destinations. Results of OAuth-related security research (see for
example [oauth_security_ubc] and [oauth_security_cmu]) indicate a
large portion of client implementations do not or fail to properly
implement security controls, like state
checks. So relying on
clients to prevent access token phishing is likely to fail as well.
Moreover, given the ratio of clients to authorization and resource
servers, it is considered the more viable approach to move as much as
possible security-related logic to those entities. Clearly, the client
has to contribute to the overall security. But there are alternative
countermeasures, as described in the next sections, that provide a
better balance between the involved parties.¶
As the name suggests, sender-constrained access tokens scope the applicability of an access token to a certain sender. This sender is obliged to demonstrate knowledge of a certain secret as prerequisite for the acceptance of that token at a resource server.¶
A typical flow looks like this:¶
The authorization server associates data with the access token that binds this particular token to a certain client. The binding can utilize the client identity, but in most cases the AS utilizes key material (or data derived from the key material) known to the client.¶
This key material must be distributed somehow. Either the key material already exists before the AS creates the binding or the AS creates ephemeral keys. The way pre-existing key material is distributed varies among the different approaches. For example, X.509 Certificates can be used, in which case the distribution happens explicitly during the enrollment process. Or the key material is created and distributed at the TLS layer, in which case it might automatically happen during the setup of a TLS connection.¶
The RS must implement the actual proof of possession check. This is typically done on the application level, often tied to specific material provided by transport layer (e.g., TLS). The RS must also ensure that replay of the proof of possession is not possible.¶
Two methods for sender-constrained access tokens using proof-of-possession have been defined by the OAuth working group:¶
OAuth 2.0 Mutual-TLS Client Authentication and Certificate-Bound Access Tokens ([RFC8705]): The approach as specified in this document allows the use of mutual TLS (mTLS) for both client authentication and sender-constrained access tokens. For the purpose of sender-constrained access tokens, the client is identified towards the resource server by the fingerprint of its public key. During processing of an access token request, the authorization server obtains the client's public key from the TLS stack and associates its fingerprint with the respective access tokens. The resource server in the same way obtains the public key from the TLS stack and compares its fingerprint with the fingerprint associated with the access token.¶
DPoP ([I-D.ietf-oauth-dpop]): DPoP (Demonstration of Proof-of-Possession at the Application Layer) outlines an application-level sender-constraining for access and refresh tokens that can be used in cases where neither mTLS nor OAuth Token Binding (see below) are available. It uses proof-of-possession based on a public/private key pair and application-level signing. DPoP can be used with public clients and, in case of confidential clients, can be combined with any client authentication method.¶
For reference, other approaches have been discussed as well but the relevant drafts are now expired:¶
OAuth Token Binding ([I-D.ietf-oauth-token-binding]): In this approach, an access token is, via the token binding ID, bound to key material representing a long term association between a client and a certain TLS host. Negotiation of the key material and proof of possession in the context of a TLS handshake is taken care of by the TLS stack. The client needs to determine the token binding ID of the target resource server and pass this data to the access token request. The authorization server then associates the access token with this ID. The resource server checks on every invocation that the token binding ID of the active TLS connection and the token binding ID of associated with the access token match. Since all crypto-related functions are covered by the TLS stack, this approach is very client developer friendly. As a prerequisite, token binding as described in [RFC8473] (including federated token bindings) must be supported on all ends (client, authorization server, resource server).¶
Signed HTTP Requests ([I-D.ietf-oauth-signed-http-request]): This approach utilizes [I-D.ietf-oauth-pop-key-distribution] and represents the elements of the signature in a JSON object. The signature is built using JWS. The mechanism has built-in support for signing of HTTP method, query parameters and headers. It also incorporates a timestamp as basis for replay prevention.¶
JWT Pop Tokens ([I-D.sakimura-oauth-jpop]): This draft describes different ways to constrain access token usage, namely TLS or request signing. Note: Since the authors of this draft contributed the TLS-related proposal to [RFC8705], this document only considers the request signing part. For request signing, the draft utilizes [I-D.ietf-oauth-pop-key-distribution] and [RFC7800]. The signature data is represented in a JWT and JWS is used for signing. Replay prevention is provided by building the signature over a server-provided nonce, client-provided nonce and a nonce counter.¶
At the time of writing, OAuth Mutual TLS is the most widely implemented and the only standardized sender-constraining method.¶
Note that the security of sender-constrained tokens is undermined when an attacker gets access to the token and the key material. This is, in particular, the case for corrupted client software and cross-site scripting attacks (when the client is running in the browser). If the key material is protected in a hardware or software security module or only indirectly accessible (like in a TLS stack), sender-constrained tokens at least protect against a use of the token when the client is offline, i.e., when the security module or interface is not available to the attacker. This applies to access tokens as well as to refresh tokens (see Section 4.13).¶
Audience restriction essentially restricts access tokens to a particular resource server. The authorization server associates the access token with the particular resource server and the resource server SHOULD verify the intended audience. If the access token fails the intended audience validation, the resource server MUST refuse to serve the respective request.¶
In general, audience restrictions limit the impact of token leakage. In the case of a counterfeit resource server, it may (as described below) also prevent abuse of the phished access token at the legitimate resource server.¶
The audience can be expressed using logical names or physical addresses (like URLs). To prevent phishing, it is necessary to use the actual URL the client will send requests to. In the phishing case, this URL will point to the counterfeit resource server. If the attacker tries to use the access token at the legitimate resource server (which has a different URL), the resource server will detect the mismatch (wrong audience) and refuse to serve the request.¶
In deployments where the authorization server knows the URLs of all resource servers, the authorization server may just refuse to issue access tokens for unknown resource server URLs.¶
The client SHOULD tell the authorization server the intended resource server. The proposed mechanism [RFC8707] could be used or by encoding the information in the scope value.¶
Instead of the URL, it is also possible to utilize the fingerprint of the resource server's X.509 certificate as audience value. This variant would also allow to detect an attempt to spoof the legitimate resource server's URL by using a valid TLS certificate obtained from a different CA. It might also be considered a privacy benefit to hide the resource server URL from the authorization server.¶
Audience restriction may seem easier to use since it does not require any crypto on the client side. Still, since every access token is bound to a specific resource server, the client also needs to obtain a single RS-specific access token when accessing several resource servers. (Resource indicators, as specified in [RFC8707], can help to achieve this.) [I-D.ietf-oauth-token-binding] has the same property since different token binding IDs must be associated with the access token. Using [RFC8705], on the other hand, allows a client to use the access token at multiple resource servers.¶
It should be noted that audience restrictions, or generally speaking an indication by the client to the authorization server where it wants to use the access token, has additional benefits beyond the scope of token leakage prevention. It allows the authorization server to create a different access token whose format and content is specifically minted for the respective server. This has huge functional and privacy advantages in deployments using structured access tokens.¶
An attacker may compromise a resource server to gain access to the resources of the respective deployment. Such a compromise may range from partial access to the system, e.g., its log files, to full control of the respective server.¶
If the attacker were able to gain full control, including shell access, all controls can be circumvented and all resources can be accessed. The attacker would also be able to obtain other access tokens held on the compromised system that would potentially be valid to access other resource servers.¶
Preventing server breaches by hardening and monitoring server systems is considered a standard operational procedure and, therefore, out of the scope of this document. This section focuses on the impact of OAuth-related breaches and the replaying of captured access tokens.¶
The following measures should be taken into account by implementers in order to cope with access token replay by malicious actors:¶
Sender-constrained access tokens, as described in Section 4.9.1.1.2, SHOULD be used to prevent the attacker from replaying the access tokens on other resource servers. Depending on the severity of the penetration, sender-constrained access tokens will also prevent replay on the compromised system.¶
Audience restriction as described in Section 4.9.1.1.3 SHOULD be used to prevent replay of captured access tokens on other resource servers.¶
The resource server MUST treat access tokens like any other credentials. It is considered good practice to not log them and not store them in plain text.¶
The first and second recommendation also apply to other scenarios where access tokens leak (see Attacker A5 in Section 3).¶
The following attacks can occur when an AS or client has an open redirector. An open redirector is an endpoint that forwards a user's browser to an arbitrary URI obtained from a query parameter. Such endpoints are sometimes implemented, for example, to show a message before a user is then redirected to an external website, or to redirect users back to a URL they were intending to visit before being interrupted, e.g., by a login prompt.¶
Clients MUST NOT expose open redirectors. Attackers may use open redirectors to produce URLs pointing to the client and utilize them to exfiltrate authorization codes and access tokens, as described in Section 4.1.2. Another abuse case is to produce URLs that appear to point to the client. This might trick users into trusting the URL and follow it in their browser. This can be abused for phishing.¶
In order to prevent open redirection, clients should only redirect if the target URLs are whitelisted or if the origin and integrity of a request can be authenticated. Countermeasures against open redirection are described by OWASP [owasp_redir].¶
At the authorization endpoint, a typical protocol flow is that the AS prompts the user to enter her credentials in a form that is then submitted (using the HTTP POST method) back to the authorization server. The AS checks the credentials and, if successful, redirects the user agent to the client's redirection endpoint.¶
In [RFC6749], the HTTP status code 302 is used for this purpose, but "any other method available via the user-agent to accomplish this redirection is allowed". When the status code 307 is used for redirection instead, the user agent will send the user's credentials via HTTP POST to the client.¶
This discloses the sensitive credentials to the client. If the client is malicious, it can use the credentials to impersonate the user at the AS.¶
The behavior might be unexpected for developers, but is defined in [RFC7231], Section 6.4.7. This status code does not require the user agent to rewrite the POST request to a GET request and thereby drop the form data in the POST request body.¶
In the HTTP standard [RFC7231], only the status code 303 unambigiously enforces rewriting the HTTP POST request to an HTTP GET request. For all other status codes, including the popular 302, user agents can opt not to rewrite POST to GET requests and therefore to reveal the user's credentials to the client. (In practice, however, most user agents will only show this behaviour for 307 redirects.)¶
ASs that redirect a request that potentially contains the user's credentials therefore MUST NOT use the HTTP 307 status code for redirection. If an HTTP redirection (and not, for example, JavaScript) is used for such a request, the AS SHOULD use HTTP status code 303 (See Other).¶
A common deployment architecture for HTTP applications is to hide the application server behind a reverse proxy that terminates the TLS connection and dispatches the incoming requests to the respective application server nodes.¶
This section highlights some attack angles of this deployment architecture with relevance to OAuth and gives recommendations for security controls.¶
In some situations, the reverse proxy needs to pass security-related data to the upstream application servers for further processing. Examples include the IP address of the request originator, token binding ids, and authenticated TLS client certificates. This data is usually passed in custom HTTP headers added to the upstream request.¶
If the reverse proxy would pass through any header sent from the
outside, an attacker could try to directly send the faked header
values through the proxy to the application server in order to
circumvent security controls that way. For example, it is standard
practice of reverse proxies to accept X-Forwarded-For
headers and just
add the origin of the inbound request (making it a list). Depending on
the logic performed in the application server, the attacker could
simply add a whitelisted IP address to the header and render a IP
whitelist useless.¶
A reverse proxy MUST therefore sanitize any inbound requests to ensure the authenticity and integrity of all header values relevant for the security of the application servers.¶
If an attacker were able to get access to the internal network between proxy and application server, the attacker could also try to circumvent security controls in place. It is, therefore, essential to ensure the authenticity of the communicating entities. Furthermore, the communication link between reverse proxy and application server MUST be protected against eavesdropping, injection, and replay of messages.¶
Refresh tokens are a convenient and user-friendly way to obtain new access tokens after the expiration of access tokens. Refresh tokens also add to the security of OAuth, since they allow the authorization server to issue access tokens with a short lifetime and reduced scope, thus reducing the potential impact of access token leakage.¶
Refresh tokens are an attractive target for attackers, since they represent the overall grant a resource owner delegated to a certain client. If an attacker is able to exfiltrate and successfully replay a refresh token, the attacker will be able to mint access tokens and use them to access resource servers on behalf of the resource owner.¶
[RFC6749] already provides a robust baseline protection by requiring¶
confidentiality of the refresh tokens in transit and storage,¶
the transmission of refresh tokens over TLS-protected connections between authorization server and client,¶
the authorization server to maintain and check the binding of a refresh token to a certain client and authentication of this client during token refresh, if possible, and¶
that refresh tokens cannot be generated, modified, or guessed.¶
[RFC6749] also lays the foundation for further (implementation specific) security measures, such as refresh token expiration and revocation as well as refresh token rotation by defining respective error codes and response behaviors.¶
This specification gives recommendations beyond the scope of [RFC6749] and clarifications.¶
Authorization servers SHOULD determine, based on a risk assessment, whether to issue refresh tokens to a certain client. If the authorization server decides not to issue refresh tokens, the client MAY refresh access tokens by utilizing other grant types, such as the authorization code grant type. In such a case, the authorization server may utilize cookies and persistent grants to optimize the user experience.¶
If refresh tokens are issued, those refresh tokens MUST be bound to the scope and resource servers as consented by the resource owner. This is to prevent privilege escalation by the legitimate client and reduce the impact of refresh token leakage.¶
For confidential clients, [RFC6749] already requires that refresh tokens can only be used by the client for which they were issued.¶
Authorization server MUST utilize one of these methods to detect refresh token replay by malicious actors for public clients:¶
Sender-constrained refresh tokens: the authorization server cryptographically binds the refresh token to a certain client instance, e.g., by utilizing [RFC8705] or [I-D.ietf-oauth-dpop].¶
Refresh token rotation: the authorization server issues a new refresh token with every access token refresh response. The previous refresh token is invalidated but information about the relationship is retained by the authorization server. If a refresh token is compromised and subsequently used by both the attacker and the legitimate client, one of them will present an invalidated refresh token, which will inform the authorization server of the breach. The authorization server cannot determine which party submitted the invalid refresh token, but it will revoke the active refresh token. This stops the attack at the cost of forcing the legitimate client to obtain a fresh authorization grant.¶
Implementation note: the grant to which a refresh token belongs may be encoded into the refresh token itself. This can enable an authorization server to efficiently determine the grant to which a refresh token belongs, and by extension, all refresh tokens that need to be revoked. Authorization servers MUST ensure the integrity of the refresh token value in this case, for example, using signatures.¶
Authorization servers MAY revoke refresh tokens automatically in case of a security event, such as:¶
Refresh tokens SHOULD expire if the client has been inactive for some time, i.e., the refresh token has not been used to obtain fresh access tokens for some time. The expiration time is at the discretion of the authorization server. It might be a global value or determined based on the client policy or the grant associated with the refresh token (and its sensitivity).¶
Resource servers may make access control decisions based on the identity of a
resource owner, for which an access token was issued, or based on the identity
of a client in the client credentials grant. If both options are possible,
depending on the details of the implementation, a client's identity may be
mistaken for the identity of a resource owner. For example, if a client is able
to choose its own client_id
during registration with the authorization server,
a malicious client may set it to a value identifying an end-user (e.g., a sub
value if OpenID Connect is used). If the resource server cannot properly
distinguish between access tokens issued to clients and access tokens issued to
end-users, the client may then be able to access resource of the end-user.¶
Authorization servers SHOULD NOT allow clients to influence their client_id
or
any other Claim if that can cause confusion with a genuine resource owner. Where
this cannot be avoided, authorization servers MUST provide other means for the
resource server to distinguish between access tokens authorized by a resource
owner from access tokens authorized by the client itself.¶
As described in Section 4.4.1.9 of [RFC6819], the authorization request is susceptible to clickjacking attacks, also called user interface redressing. In such an attack, an attacker embeds the authorization endpoint user interface in an innocuous context. A user believing to interact with that context, for example, clicking on buttons, inadvertently interacts with the authorization endpoint user interface instead. The opposite can be achieved as well: A user believing to interact with the authorization endpoint might inadvertently type a password into an attacker-provided input field overlaid over the original user interface. Clickjacking attacks can be designed such that users can hardly notice the attack, for example using almost invisible iframes overlaid on top of other elements.¶
An attacker can use this vector to obtain the user's authentication credentials, change the scope of access granted to the client, and potentially access the user's resources.¶
Authorization servers MUST prevent clickjacking attacks. Multiple countermeasures are described in [RFC6819], including the use of the X-Frame-Options HTTP response header field and frame-busting JavaScript. In addition to those, authorization servers SHOULD also use Content Security Policy (CSP) level 2 [CSP-2] or greater.¶
To be effective, CSP must be used on the authorization endpoint and, if applicable, other endpoints used to authenticate the user and authorize the client (e.g., the device authorization endpoint, login pages, error pages, etc.). This prevents framing by unauthorized origins in user agents that support CSP. The client MAY permit being framed by some other origin than the one used in its redirection endpoint. For this reason, authorization servers SHOULD allow administrators to configure allowed origins for particular clients and/or for clients to register these dynamically.¶
Using CSP allows authorization servers to specify multiple origins in
a single response header field and to constrain these using flexible
patterns (see [CSP-2] for details). Level 2 of this standard provides
a robust mechanism for protecting against clickjacking by using
policies that restrict the origin of frames (using frame-ancestors
)
together with those that restrict the sources of scripts allowed to
execute on an HTML page (by using script-src
). A non-normative
example of such a policy is shown in the following listing:¶
HTTP/1.1 200 OK Content-Security-Policy: frame-ancestors https://ext.example.org:8000 Content-Security-Policy: script-src 'self' X-Frame-Options: ALLOW-FROM https://ext.example.org:8000 ...¶
Because some user agents do not support [CSP-2], this technique SHOULD be combined with others, including those described in [RFC6819], unless such legacy user agents are explicitly unsupported by the authorization server. Even in such cases, additional countermeasures SHOULD still be employed.¶
We would like to thank Brock Allen, Annabelle Richard Backman, Dominick Baier, Vittorio Bertocci, Brian Campbell, William Dennis, George Fletscher, Dick Hardt, Joseph Heenan, Pedram Hosseyni, Phil Hunt, Jared Jennings, Michael B. Jones, Konstantin Lapine, Neil Madden, Christian Mainka, Jim Manico, Nov Matake, Doug McDorman, Karsten Meyer zu Selhausen, Aaron Parecki, Michael Peck, Johan Peeters, Nat Sakimura, Guido Schmitz, Travis Spencer, Petteri Stenius, Tomek Stojecki, Tim Wuertele, David Waite and Hans Zandbelt for their valuable feedback.¶
This draft makes no requests to IANA.¶
Security considerations are described in Section 2, Section 3, and Section 4.¶
[[ To be removed from the final specification ]]¶
-20¶
Improved description of authorization code injection attacks and PKCE protection¶
Removed recommendation for MTLS in discussion (not reflected in actual Recommendations section)¶
Reworded "placeholder" text in security considerations.¶
Alphabetized list of names and fixed unicode problem¶
Explained Clickjacking¶
Explained Open Redirectors¶
Clarified references to attacker model by including a link to Section 3¶
Clarified description of "CSRF tokens" and reference to RFC6819¶
Described that OIDC can prevent access token injection¶
Updated references¶
-19¶
Changed affiliation of Andrey Labunets¶
Editorial change to clarify the new recommendations for refresh tokens¶
-18¶
-17¶
Make the use of metadata RECOMMENDED for both servers and clients¶
Make announcing PKCE support in metadata the RECOMMENDED way (before: either metadata or deployment-specific way)¶
AS also MUST NOT expose open redirectors.¶
Mention that attackers can collaborate.¶
Update recommendations regarding mix-up defense, building upon [I-D.ietf-oauth-iss-auth-resp].¶
Improve description of mix-up attack.¶
Make HTTPS mandatory for most redirect URIs.¶
-16¶
Make MTLS a suggestion, not RECOMMENDED.¶
Add important requirements when using nonce for code injection protection.¶
Highlight requirements for refresh token sender-constraining.¶
Make PKCE a MUST for public clients.¶
Describe PKCE Downgrade Attacks and countermeasures.¶
Allow variable port numbers in localhost redirect URIs as in RFC8252, Section 7.3.¶
-15¶
-14¶
Added info about using CSP to prevent clickjacking¶
Changes from WGLC feedback¶
Editorial changes¶
AS MUST announce PKCE support either in metadata or using deployment-specific ways (before: SHOULD)¶
-13¶
Discourage use of Resource Owner Password Credentials Grant¶
Added text on client impersonating resource owner¶
Recommend asymmetric methods for client authentication¶
Encourage use of PKCE mode "S256"¶
PKCE may replace state for CSRF protection¶
AS SHOULD publish PKCE support¶
Cleaned up discussion on auth code injection¶
AS MUST support PKCE¶
-12¶
Added updated attacker model¶
-11¶
Adapted section 2.1.2 to outcome of consensus call¶
more text on refresh token inactivity and implementation note on refresh token replay detection via refresh token rotation¶
-10¶
incorporated feedback by Joseph Heenan¶
changed occurrences of SHALL to MUST¶
added text on lack of token/cert binding support tokens issued in the authorization response as justification to not recommend issuing tokens there at all¶
added requirement to authenticate clients during code exchange (PKCE or client credential) to 2.1.1.¶
added section on refresh tokens¶
editorial enhancements to 2.1.2 based on feedback¶
-09¶
changed text to recommend not to use implicit but code¶
added section on access token injection¶
reworked sections 3.1 through 3.3 to be more specific on implicit grant issues¶
-08¶
added recommendations re implicit and token injection¶
uppercased key words in Section 2 according to RFC 2119¶
-07¶
incorporated findings of Doug McDorman¶
added section on HTTP status codes for redirects¶
added new section on access token privilege restriction based on comments from Johan Peeters¶
-06¶
reworked section 3.8.1¶
incorporated Phil Hunt's feedback¶
reworked section on mix-up¶
extended section on code leakage via referrer header to also cover state leakage¶
added Daniel Fett as author¶
replaced text intended to inform WG discussion by recommendations to implementors¶
modified example URLs to conform to RFC 2606¶
-05¶
Completed sections on code leakage via referrer header, attacks in browser, mix-up, and CSRF¶
Reworked Code Injection Section¶
Added reference to OpenID Connect spec¶
removed refresh token leakage as respective considerations have been given in section 10.4 of RFC 6749¶
first version on open redirection¶
incorporated Christian Mainka's review feedback¶
-04¶
-03¶
-02¶
Folded Mix up and Access Token leakage through a bad AS into new section for dynamic OAuth threats¶
reworked dynamic OAuth section¶
-01¶
Added references to mitigation methods for token leakage¶
Added reference to Token Binding for Authorization Code¶
incorporated feedback of Phil Hunt¶
fixed numbering issue in attack descriptions in section 2¶
-00 (WG document)¶