If no, where are these CSRF generated tokens stored at: JCR Repository or Objects in the application heap? Also how does it validate the received token at very high level?
If yes, is this not a scalabilty issue?
I tried to follow these links: https://datatracker.ietf.org/doc/html/draft-ietf-oauth-json-web-token-16#section-7 & https://datatracker.ietf.org/doc/html/draft-ietf-jose-json-web-signature-41#appendix-A.4.2 and seems like they use a sort of public-private key, along with user, user-agent and other info to build a key-pair and a signature, and validate it in a similar fashion, where the token is deciphered in a sense, but not exactly matching it to another stored token. But not sure, hence the question.
Short answer: Yes the AEM CSRF Authentication/protection framework is stateless.
Details
The tokens are not persisted and all the information is in the token encrypted using a Symmetric Algorithm. As long as all your instances share the same Crypto Key, any instance can decrypt and decode the CSRF token issued within the cluster. More details on this can be found in the official CSRF documentation.
Related
I need to pass JSON-encoded signed (and sometimes additionally encrypted) objects between multiple instances of my software. The obvious choice here is JWT.
Yet, JWT allows apparently to both sign and encrypt a token (JWS and JWE) or nest JWS into a JWE (nested JWE).
While both approaches seem reasonable for me, is there a "standard" way of doing this? I haven't found any specifics on this.
Short answer
When both signing and encryption are necessary, you should first sign the message and then encrypt the result. That is, nesting a JWS into a JWE is a valid approach.
Long answer
JSON Web Token (JWT) is an open standard that defines a compact and self-contained way for securely transmitting information between parties as a JSON object. JWT is a generic name for the following types of token:
JSON Web Signature (JWS): The payload is encoded and signed so the integrity of the claims can be verified.
JSON Web Encryption (JWE): They payload is encrypted so the claims are hidden from other parties.
The image was extracted from this page.
JWT allows apparently to both sign and encrypt a token (JWS and JWE) or nest JWS into a JWE (nested JWE).
While both approaches seem reasonable for me, is there a "standard" way of doing this? I haven't found any specifics on this.
The concept of nested JWT is defined in the RFC 7519:
A JWT in which nested signing and/or encryption are employed. In
Nested JWTs, a JWT is used as the payload or plaintext value of an
enclosing JWS or JWE structure, respectively.
Regarding the order of the operations, it's advisable to first sign the message and then encrypt the result, as mentioned in the same document:
11.2. Signing and Encryption Order
While syntactically the signing and encryption operations for Nested
JWTs may be applied in any order, if both signing and encryption are
necessary, normally producers should sign the message and then
encrypt the result (thus encrypting the signature). This prevents
attacks in which the signature is stripped, leaving just an encrypted
message, as well as providing privacy for the signer. Furthermore,
signatures over encrypted text are not considered valid in many
jurisdictions.
AFAIK it's not possible to set a header field when a form is submited, it can only be done in ajax requests
This questions also points that it's not possible:
How to set a Header field on POST a form?
Custom HTTP Request headers in HTML
But reading Cross-Site Request Forgery (CSRF) Prevention Cheat Sheet it's mentioned that:
Encrypted Token Pattern
Overview
The Encrypted Token Pattern leverages an encryption, rather than
comparison, method of Token-validation. After successful
authentication, the server generates a unique Token comprised of the
user's ID, a timestamp value and a nonce, using a unique key available
only on the server. This Token is returned to the client and embedded
in a hidden field. Subsequent AJAX requests include this Token in the
request-header, in a similar manner to the Double-Submit pattern.
Non-AJAX form-based requests will implicitly persist the Token in its hidden field, although I recommend persisting this data in a
custom HTTP header in such cases. On receipt of this request, the
server reads and decrypts the Token value with the same key used to
create the Token.
This sentence confuses me:
I recommend persisting this data in a custom HTTP header in such cases
Could anyone shed some light on it?
Yes, the sentence implies that the POST operation is invoked by a custom JavaScript handler in order to inject the AUTH header. I've corrected the OWASP description to reflect this oversight.
I've set up a MediaWiki server on an Azure website with the PluggableAuth and OpenID Connect extensions. The latter uses the PHP OpenID Connect Basic Client library. I am an administrator in the Azure AD domain example.com, wherein I've created an application with App ID URI, sign-on URL and reply URL all set to https://wiki.azurewebsites.net/. When I navigate to the wiki, I observe the following behavior (cookie values omitted for now):
Client Request
GET https://wiki.azurewebsites.net/ HTTP/1.1
RP Request
GET https://login.windows.net/example.com/.well-known/openid-configuration
IP Response
(some response)
RP Response
HTTP/1.1 302 Moved Temporarily
Location: https://login.windows.net/{tenant_id}/oauth2/authorize?response_type=code&redirect_uri=https%3A%2F%2Fwiki.azurewebsites.net%2F&client_id={client_id}&nonce={nonce}&state={state}
Client Request
(follows redirect)
IP Response
HTTP/1.1 302 Found
Location: https://wiki.azurewebsites.net/?code={code}&state={state}&session_state={session_state}
Client Request
(follows redirect)
RP Request (also repeats #2 & #3)
POST https://login.windows.net/{tenant_id}/oauth2/token
grant_type=authorization_code&code={code}&redirect_uri=https%3A%2F%2Fwiki.azurewebsites.net%2F&client_id={client_id}&client_secret={client_secret}
IP Response
(As interpreted by MediaWiki; I don't have the full response logged at this time)
AADSTS50001: Resource identifier is not provided.
Note that if I change the OpenID PHP client to provide the 'resource' parameter in step 8, I get the following error response from AAD instead:
RP Request
POST https://login.windows.net/{tenant_id}/oauth2/token
grant_type=authorization_code&code={code}&redirect_uri=https%3A%2F%2Fwiki.azurewebsites.net%2F&resource=https%3A%2F%2Fwiki.azurewebsites.net%2F&client_id={client_id}&client_secret={client_secret}
IP Response
AADSTS90027: The client '{client_id}' and resource 'https://wiki.azurewebsites.net/' identify the same application.
(This has come up before.)
Update
I've made some progress based on #jricher's suggestions, but after working through several more errors I've hit one that I can't figure out. Once this is all done I'll submit pull requests to the affected libraries.
Here's what I've done:
I've added a second application to the example.com Azure AD domain, with the App ID URI set to mediawiki://wiki.azurewebsites.net/, as a dummy "resource". I also granted the https://wiki.azurewebsites.net/ application delegated access to this new application.
Passing in the dummy application's URI as the resource parameter in step #8, I'm now getting back the access, refresh, and ID tokens in #9!
The OpenID Connect library requires that the ID token be signed, but while Azure AD signs the access token it doesn't sign the ID token. It comes with the following properties: {"typ":"JWT","alg":"none"}. So I had to modify the library to allow the caller to specify that unsigned ID tokens are considered "verified". Grrr.
Okay, next it turns out that the claims can't be verified because the OpenID Provider URL I specified and the issuer URL returned in the token are different. (Seriously?!) So, the provider has to be specified as https://sts.windows.net/{tenant_id}/, and then that works.
Next, I found that I hadn't run the MediaWiki DB upgrade script for the OpenID Connect extension yet. Thankfully that was a quick fix.
After that, I am now left with (what I hope is) the final problem of trying to get the user info from AAD's OpenID Connect UserInfo endpoint. I'll give that its own section.
Can't get the user info [Updated]
This is where I am stuck now. After step #9, following one or two intermediate requests to get metadata and keys for verifying the token, the following occurs:
RP Request:
(Updated to use GET with Authorization: Bearer header, per MSDN and the spec.)
GET https://login.windows.net/{tenant_id}/openid/userinfo
Authorization: Bearer {access_token}
IP Response:
400 Bad Request
AADSTS50063: Credential parsing failed. AADSTS90010: JWT tokens cannot be used with the UserInfo endpoint.
(If I change #10 to be either a POST request, with access_token in the body, or a GET request with access_token in the query string, AAD returns the error: AADSTS70000: Authentication failed. UserInfo token is not valid. The same occurs if I use the value of the id_token in place of the access_token value that I received.)
Help?
Update
I'm still hoping someone can shed light on the final issue (the UserInfo endpoint not accepting the bearer token), but I may split that out into a separate question. In the meantime, I'm adding some workarounds to the libraries (PRs coming soon) so that the claims which are already being returned in the bearer token can be used instead of making the call to the UserInfo endpoint. Many thanks to everyone who's helped out with this.
There's also a nagging part of me that wonders if the whole thing would not have been simpler with the OpenID Connect Basic Profile. I assume there's a reason why that was not implemented by the MediaWiki extension.
Update 2
I just came across a new post from Vittorio Bertocci that includes this helpful hint:
...in this request the application is asking for a token for itself! In Azure AD this is possible only if the requested token is an id_token...
This suggests that just changing the token request type in step 8 from authorization_code to id_token could remove the need for the non-standard resource parameter and also make the ugly second AAD application unnecessary. Still a hack, but it feels like much less of one.
Justin is right. For authorization code grant flow, your must specify the resource parameter in either the authorization request or the token request.
Use &resource=https%3A%2F%2Fgraph.windows.net%2F to get an access token for the Azure AD Graph API.
Use &resource=https%3A%2F%2Fmanagement.core.windows.net%2F to get a token for the Azure Service Management APIs.
...
Hope this helps
Microsoft's implementation of OpenID Connect (and OAuth2) has a known bug where it requires the resource parameter to be sent by the client. This is an MS-specific parameter and requiring it unfortunately breaks compatibility with pretty much every major OAuth2 and OpenID Connect library out there. I know that MS is aware of the issue (I've been attempting to do interoperability testing with their team for quite a while now), but I don't know of any plans to fix the problem.
So in the mean time, your only real path is to hack your client software so that it sends a resource parameter that the AS will accept. It looks like you managed to make it send the parameter, but didn't send a value that it liked.
I had issues getting this running on Azure, even though I got something working locally. Since I was trying to setup a private wiki anyway, I ended up enabling Azure AD protection for the whole site by turning on:
All Settings -> Features -> Authentication / Authorization
From within the website in https://portal.azure.com
This made it so you had to authenticate to Azure-AD before you saw any page of the site. Once you were authenticated a bunch of HTTP Headers are set for the application with your username, including REMOTE_USER. As a result I used the following plugin to automatically log the already authenticated user into Azure:
https://www.mediawiki.org/wiki/Extension:Auth_remoteuser
I got a (Flask) backend powering an API that serves JSON to an Angular app.
I love the fact that my backend (algorithms, database) is totally disconnected from my frontend (design, UI) as it could literally run from two distinct servers. However since the view is entirely generated client side everyone can access the JSON data obviously. Say the application is a simple list of things (the things are stored in a JSON file).
In order to prevent direct access to my database through JSON in the browser console I found these options :
Encrypting the data (weak since the decrypting function will be freely visible in the javascript, but not so easy when dealing with minified files)
Instead of $http.get the whole database then filtering with angular, $http.get many times (as the user is scrolling a list for example) so that it is programmatically harder to crawl
I believe my options are still weak. How could I make it harder for a hacker to crawl the whole database ? Any ideas ?
As I understand this question - the user should be permitted to access all of the data via your UI, but you do not want them to access the API directly. As you have figured out, any data accessed by the client cannot be secured but we can make accessing it a little more of PITA.
One common way of doing this is to check the HTTP referer. When you make a call from the UI the server will be given the page the request is coming from. This is typically used to prevent people creating mashups that use your data without permission. As with all the HTTP request headers, you are relying on the caller to be truthful. This will not protect you from console hacking or someone writing a scraper in some other language. #see CSRF
Another idea is to embed a variable token in the html source that bootstraps your app. You can specify this as an angular constant or a global variable and include it in all of your $http requests. The token itself could be unique for each session or be a encrypted expiration date that only the server can process. However, this method is flawed as well as someone could parse the html source, get the code, and then make a request.
So really, you can make it harder for someone, but it is hardly foolproof.
If users should only be able to access some of the data, you can try something like firebase. It allows you to define rules for who can access what.
Security Considerations When designing web applications, consider
security threats from:
JSON vulnerability XSRF Both server and the client must cooperate in
order to eliminate these threats. Angular comes pre-configured with
strategies that address these issues, but for this to work backend
server cooperation is required.
JSON Vulnerability Protection A JSON vulnerability allows third party
website to turn your JSON resource URL into JSONP request under some
conditions. To counter this your server can prefix all JSON requests
with following string ")]}',\n". Angular will automatically strip the
prefix before processing it as JSON.
For example if your server needs to return:
['one','two'] which is vulnerable to attack, your server can return:
)]}', ['one','two'] Angular will strip the prefix, before processing
the JSON.
Cross Site Request Forgery (XSRF) Protection XSRF is a technique by
which an unauthorized site can gain your user's private data. Angular
provides a mechanism to counter XSRF. When performing XHR requests,
the $http service reads a token from a cookie (by default, XSRF-TOKEN)
and sets it as an HTTP header (X-XSRF-TOKEN). Since only JavaScript
that runs on your domain could read the cookie, your server can be
assured that the XHR came from JavaScript running on your domain. The
header will not be set for cross-domain requests.
To take advantage of this, your server needs to set a token in a
JavaScript readable session cookie called XSRF-TOKEN on the first HTTP
GET request. On subsequent XHR requests the server can verify that the
cookie matches X-XSRF-TOKEN HTTP header, and therefore be sure that
only JavaScript running on your domain could have sent the request.
The token must be unique for each user and must be verifiable by the
server (to prevent the JavaScript from making up its own tokens). We
recommend that the token is a digest of your site's authentication
cookie with a salt for added security.
The name of the headers can be specified using the xsrfHeaderName and
xsrfCookieName properties of either $httpProvider.defaults at
config-time, $http.defaults at run-time, or the per-request config
object.
Please Kindly refer the below link,
https://docs.angularjs.org/api/ng/service/$http
From AngularJS DOCs
JSON Vulnerability Protection
A JSON vulnerability allows third party website to turn your JSON resource URL into JSONP request under some conditions. To counter this your server can prefix all JSON requests with following string ")]}',\n". Angular will automatically strip the prefix before processing it as JSON.
There are other techniques like XSRF protection and Transformations which will further add security to your JSON communications. more on this can be found in AngularJS Docs https://docs.angularjs.org/api/ng/service/$http
You might want to consider using JSON Web Tokens for this. I'm not sure how to implement this in Flask but here is a decent example of how it can be done with a Nodejs backend. This example at least shows how you can implement it in Angularjs.
http://www.kdelemme.com/2014/03/09/authentication-with-angularjs-and-a-node-js-rest-api/
Update: JWT for Flask:
https://github.com/mattupstate/flask-jwt
In regards to this Haacked blog, I'm hesitant to implement the proposed anti-JSON GET hijacking solutions since
The recommended solutions to mitigating JSON hijacking involve non-REST-full JSON POSTs to GET data
The alternate solution (object wrapping) causes problems with 3rd party controls I don't have source-code access to.
I can't find a community-vetted implementation that implements the Alternative Solution (listed below) on how to compose the security token, or securely deliver it within the webpage. I also won't claim to be enough of an expert to roll my own implementation.
Referrer headers can't be relied upon
Background
This blog describes a CSRF issue regarding JSON Hijacking and recommends using JSON POSTs to GET data. Since using a HTTP POST to GET data isn't very REST-full, I'd looking for a more RESTfull solution that enables REST actions per session, or per page.
Another mitigation technique is to wrap JSON data in an object as described here. I'm afraid this may just delay the issue, until another technique is found.
Alternative Implementation
To me, it seems natural to extend the use ASP.NET MVC's AntiForgeryToken with jQuery HTTP GETs for my JSON.
For example if I GET some sensitive data, according to the Haacked link above, the following code is vulnerable:
$.getJSON('[url]', { [parameters] }, function(json) {
// callback function code
});
I agree that it isn't RESTfull to GET data using the recommended POST workaround. My thought is to send a validation token in the URL. That way the CSRF-style attacker won't know the complete URL. Cached, or not cached, they won't be able to get the data.
Below are two examples of how a JSON GET query could be done. I'm not sure what implementation is most effective, but may guess that the first one is safer from errant proxies caching this data, thus making it vulnerable to an attacker.
http://localhost:54607/Home/AdminBalances/ENCODEDTOKEN-TOKEN-HERE
or
http://localhost:54607/Home/AdminBalances?ENCODEDTOKEN-TOKEN-HERE
... which might as well be MVC3's AntiForgeryToken, or a variant (see swt) thereof. This token would be set as an inline value on whatever URL format is chosen above.
Sample questions that prevent me from rolling my own solution
What URL format (above) would you use to validate the JSON GET (slash, questionmark, etc) Will a proxy respond to http://localhost:54607/Home/AdminBalances with http://localhost:54607/Home/AdminBalances?ENCODEDTOKEN-TOKEN-HERE data?
How would you deliver that encoded token to the webpage? Inline, or as a page variable?
How would you compose the token? Built in AntiforgeryToken, or by some other means?
The AntiForgeryToken uses a cookie. Would a backing cookie be used/needed in this case? HTTP Only? What about SSL in conjunction with HTTP Only?
How would you set your cache headers? Anything special for the Google Web Accelerator (for example)
What are the implications of just making the JSON request SSL?
Should the returned JSON array still be wrapped in an object just for safety's sake?
How will this solution interop with Microsoft's proposed templating and databinding features
The questions above are the reasons I'm not forging ahead and doing this myself. Not to mention there likely more questions I haven't thought of, and yet are a risk.
The Asp.net MVC AntiForgeryToken won't work through HTTP GET, because it relies on cookies which rely on HTTP POST (it uses the "Double Submit Cookies" technique described in the OWASP XSRF Prevention Cheat Sheet). You can also additionally protect the cookies sent to the client by setting the as httponly, so they cannot be spoofed via a script.
In this document you can find various techniques that can be used to prevent XSRF. It seems the you described would fall into the Approach 1. But we have a problem on how to retrieve the session on the server when using Ajax HTTP GET request since the cookies are not sent with the request. So you would also have to add a session identifier to you action's URL (aka. cookieless sessions, which are easier to hijack). So in order to perform an attack the attacker would only need to know the correct URL to perform the GET request.
Perhaps a good solution would be to store the session data using some key from the users SSL certificate (for example the certs thumb-print). This way only the owner of the SSL certificate could access his session. This way you don't need to use cookies and you don't need to send session identifiers via query string parameters.
Anyway, you will need to roll out your own XSRF protection if you don't want to use HTTP POST in Asp.net MVC.
I came to this problem and the solution was not so trivial however there is a fantastic blog to get you started this can be used with get and post ajax.
http://johan.driessen.se/posts/Updated-Anti-XSRF-Validation-for-ASP.NET-MVC-4-RC
If you place the following in the global name space all your post/gets can take advantage having an anti forgery token and you don't have to modify your ajax calls. Create an input element in a common page.
<form id="__AjaxAntiForgeryForm" action="#" method="post">#Html.AntiForgeryToken()</form>
The following javascript will read the anti forgery tokken and add it to the request header.
// Wire up the global jQuery ajaxSend event handler.
$(document).ajaxSend(namespace.ajax.globalSendHandler);
// <summary>
// Global handler for all ajax send events.
// </summary>
namespace.ajax.globalSendHandler = function (event, xhr, ajaxOptions) {
// Add the anti forgery token
xhr.setRequestHeader('__RequestVerificationToken', $("#__AjaxAntiForgeryForm input[name=__RequestVerificationToken]").val());
};
I think it is legitimate to use AntiforgeryToken (AFT) within an ajax http GET request provided that it is embedded in a form that already provides the AFT and associated cookie. The ajax handler can then do the validate on the server just how it would in a normal form post.