According to official spring documentation:
WebSockets reuse the same authentication information that is found in
the HTTP request when the WebSocket connection was made. This means
that the Principal on the HttpServletRequest will be handed off to
WebSockets. If you are using Spring Security, the Principal on the
HttpServletRequest is overridden automatically.
More concretely, to ensure a user has authenticated to your WebSocket
application, all that is necessary is to ensure that you setup Spring
Security to authenticate your HTTP based web application.
If I understood it correctly, this means that WebSocket is using the same channel for communication since the handshake, and thus the authentication should be made on the first connection.
However nowhere is stated how to actually authenticate the handshake in a standard secure way. As far as I am aware HTTP doesn't send an Authentication header while upgrading to the WebSockets so how it is done?
Do I really need to send authentication token in connection query, e.g
localhost:8080/ws?Auth=...
and leave the security to HTTPS
Or do I need to authenticate the WebSocket after the connection is made e.g create my own handshake?
Is there any proper formal way to do it? I am using RAW websockets.
Thanks for the ideas/help.
I personally use STOMP but with STOMP (basically a framework for communications on raw WebSocks), the session cookie (from spring security) is send with any message down the socket.
You can use the StompAccessorHeader like:
#MessageMapping("/agents/start")
public void start(StompHeaderAccessor stompHeaderAccessor) {
log.info("Subscriber Start! {}-{}", stompHeaderAccessor.getUser() != null ? stompHeaderAccessor.getUser().getName() : "ANON", stompHeaderAccessor.getSessionId());
mysessionstore.addSessionId(stompHeaderAccessor.getSessionId());
}
With not using the STOMP framework there may be a way to read the SessionCookie sent per request on the raw socket?
I am not 100% sure but I am guessing you are using the TextWebSocketHandler implmentation with the:
#Override
protected void handleTextMessage(WebSocketSession session, TextMessage textMessage)
I can see in the source code for WebSocketSession you should be able to get your principal authenticate user there:
https://github.com/spring-projects/spring-framework/blob/0de2833894c24c1e70bde991bad171435c6ecac2/spring-websocket/src/main/java/org/springframework/web/socket/WebSocketSession.java#L37
So you authenticate like normal REST like POST "/login" and then that session should be valid for the websockets as well.
You may be able to auth via the socket? You'd have to like make your own socket endpoint to take their credentials and do SecurityContextHolder.getContext().setAuthentication(myAuthUserToken) but maybe that will then pass a session cookie back? You'd have to test this ofc as I am unsure if it would work shrug.
I personally then make a "store" (a singleton or redis) that holds the user principal and the socketSessionId so I can then match a user to a socket.
You could say store them in a singleton with a HashMap<String,String> userPrincipalNameToSocketSessionId as a crude way store which socket session belongs to which user.
eg.
A bit like
#Override
protected void handleTextMessage(WebSocketSession session, TextMessage textMessage){
MySessionStore.addSessionToMap(session.getPrincipal(),session.getId());
log.info("Added user {} websocket session {} to the store.",session.getPrincipal(),session.getId());
}
public */MySingletonClass*/ MySessionStore{
#Getter
public static volatile HashMap<String,String> userPrinciapalToSocketMap = new HashMap<>();
//Method to add to map here
public synchronized static addToMap(String principalName,String webSocketSessionId){
...Adds to the map.
}
JWT Stateless Auth System wanting a Socket Session?
As far as I can guess with this one...unless there is a lot of overriding and practically forking/extending lots of Spring classes...
You could make a controller:
Http GET => "/websocket-ticket" which would return a signed token with the user's principal/username/id for the UX to then pass as a first message after websocket connect.
The socket handler TextMessageHandler can check the signaute of the token and add it to your HashMap<String,String> principalUserToSessionId store.
The security issue (unlikely but it is there):
An attacker with XSS could snoop that token and hijack that websocket session. Maybe you win on a race condition (i.e. the MITM takes longer and the token is single use...more wonderful implementation...you now also need a "websocket-ticket-consumed" store...).
https://devcenter.heroku.com/articles/websocket-security
I am feeling this is all heading to the X/Y problem.
Why are you using JWTs for Auth?
Related
I'm trying to enable multi-tenancy for a previously single-user system. The application used to run on a local server and had a relatively simple frontend baked in.
Now I want to allow multiple users to simultaneously use it in a cloud environment. I went ahead and implemented Auth2 with OIDC and PKCE to redirect users to an external Auth Provider. What I want now is that for every request, the user sends his Access token with the request in order for me to decide what data to provide with the answer.
I could not figure out how to obtain that data, as it seems that the spring framework (by default) only sends the ID token with the request. I suspect the fact that my software would simultaneously be the client and the resource server has something to do with it.
This is my first question, so I'm very happy to modify or extend my question if I've forgotten anything.
What I've tried to far:
I've used Postman to verify that the three tokens, ID token, refresh token and access token are issued correctly and can be retrieved with my credentials.
I tried getting the access token from the request itself. Any parameters (like #AuthenticationPrincipal OidcUser oidcUser) in the controller that include the token, however, are only showing the ID token and not the access token.
Getting the token via the OAuth2AuthorizedClientService does not work either, same problem, as I can only get the ID token, but not the access token.
Update #1, 13.12.2022/11:40: I am using PingOne by PingIdentity as the authentication provider and the following dependencies are or might be related or helpful to this matter:
spring-boot-starter-web
spring-boot-starter-security
spring-boot-starter-thymeleaf
spring-boot-starter-web-services
spring-boot-starter-oauth2-resource-server
Split your security in two with http.securityMatcher in the first picked SecurityFilterChain bean to restrict the endpoints it applies to (use #Order to control that), and configure WebClient to add access-token as Authorization header to its requests to resource-server (REST endpoints). This is demoed in the tutorial I already linked in my comment to your question. This tutorial matches exactly what you are asking, the fact that it uses Keycloak instead of PingOne as authorization server is a detail (solved by editing a property or two).
Resource-server security filter-chain
As reminder, a resource-server is a REST API secured with OAuth2. All end-points of #RestController and #Controller with #ResponseBody should be secured with a SecurityFilterChain containing http.oauth2ResourceServer(). The easiest way to configure a resource-server with Spring is using spring-boot-starter-oauth2-resource-server (or one of the very thin wrappers I wrote around it which enable to configure it from properties with 0 java conf)
By default with such filter-chains, Spring populates the security-context with a sub-class of AbstractOAuth2TokenAuthenticationToken<?>. You can retrieve the access-token from it. At least 2 options to access this Authentication instance:
SecurityContextHolder.getContext().getAuthentication()
have Spring auto-magically inject AbstractOAuth2TokenAuthenticationToken<?> auth as controller method parameter
You can also have the original authorization header injected as controller method parameter with #RequestHeader(value = HttpHeaders.AUTHORIZATION) String authorizationHeader.
I expose various ways to configure resource-servers security filter-chain in this tutorials.
Client security filter-chain
End-points of #Controller returning a template name and secured with OAuth2 are clients. The easiest way to configure a Spring client is with spring-boot-starter-oauth2-client and http.oauth2Login().
Note that in this configuration, the request between the browser and the Spring client is not OAuth2 (it is most frequently secured with a session cookie, not a Bearer access-token in Authorization header). Only requests sent by the Spring client (on the server) to resource-servers are secured with OAuth2.
With client security filter-chain, security-context is populated with OAuth2AuthenticationToken which, on purpose, exposes ID-token and not access-token. As reminder, ID tokens are holding user identity data and are intended to be used by clients when access-tokens audience is resource-servers and is designed for access-control. Clients should consider access-tokens as black box and use it only to authorize their requests to resource-servers (set Bearer Authorization header).
You can get the access-token string from OAuth2AuthorizedClient: authorizedClient.getAccessToken().getTokenValue(), which is itself retrieved from the OAuth2AuthorizedClientService you can auto-wire in your controllers: authorizedClientService.loadAuthorizedClient("your-client-id", auth.getName()) (auth being the OAuth2AuthenticationToken instance retrieved from security-context via SecurityContextHolder or controller method parameter injection)
If you need to authorize a WebClient request from client to resource-server, you can do simpler than retrieve access token and position authorization header: do as in the UiController of the tutorial already linked in my comment to your question:
final var authorizedClient = authorizedClientService.loadAuthorizedClient("spring-addons-public", auth.getName());
final var response = webClient.get().uri("http://localhost:8080/api/greet")
.attributes(ServerOAuth2AuthorizedClientExchangeFilterFunction.oauth2AuthorizedClient(authorizedClient))
...
With WebClient configured as follow:
#Bean
WebClient webClient(ClientRegistrationRepository clientRegistrationRepository,
OAuth2AuthorizedClientService authorizedClientService) {
var oauth = new ServletOAuth2AuthorizedClientExchangeFilterFunction(
new AuthorizedClientServiceOAuth2AuthorizedClientManager(clientRegistrationRepository,
authorizedClientService));
oauth.setDefaultClientRegistrationId("spring-addons-public");
return WebClient.builder().apply(oauth.oauth2Configuration()).build();
}
Thanks to those who tried to help me, but eventually I figured it out myself.
I extended my Controllers by two attributes: OAuth2AuthenticationToken authentication and HttpServletRequest request.
Also, I #Autowired in the OAuth2AuthorizedClientRepository oAuth2AuthorizedClientRepository.
This then allows the following call returning the value of the accessToken:
(oAuth2AuthorizedClientRepository.loadAuthorizedClient(myClientRegistrationId, authentication, request)).client.getAccessToken().getTokenValue();.
After that, it's just parsing the token and retrieving the values using JWTParser.parse() and provided methods from the JWT-result.
Personal note: Don't do the parsing and retrieving value parts in your controller to keep any more complex logic out of it.
I hope this helps somebody!
I am very much new to web services. I have exposed some REST services using Jersey 2 in integration with Spring. Now I need to secure those rest services using authentication with username/password. I am told not to use Spring Security.
I have no idea of how to do this. I did search on the net but various links show various implementation and I am unable to decide how to proceed with it.
A common way for authenticating with username and password is to use Basic Authentication. Basically the client needs to send a request header Authorization, with the the header value as Basic Base64Encoded(username:password). So is my username is peeskillet and my password is pass, I, as a client, should set the header as
Authorization: Basic cGVlc2tpbGxldDpwYXNz
In a servlet environment, the container should have support for Basic authentication. You would configure this support on the web.xml. You can see an example in 48.2 Securing Web Applications of the Java EE tutorial. You will also notice in an example
<transport-guarantee>CONFIDENTIAL</transport-guarantee>
That is for SSL support. This is recommended for Basic Authentication.
If you don't want to deal with the hassle of working with security domains and login modules, realm, and such, that would be required to customize the servlet support, or if you're just not in a servlet environment, implementing Basic Auth in a ContainerRequestFilter is really not too difficult.
You can see a complete example of how this could be done at jersey/examples/https-clientserver-grizzly. You should focus on the SecurityFilter
The basic flow in the filter goes something like this
Get the Authorization header. If it doesn't exist, throw an AuthenticationException. In which case the AuthenticationExceptionMapper will send out the header "WWW-Authenticate", "Basic realm=\"" + e.getRealm() + "\", which is part of the Basic Auth protocol
Once we have the header, we parse it just to get the Base64 encoded username:password. Then we decode it, then split it, then separate the user name and password. If any of this process fails, again throw the WebApplicationException that maps to a 400 Bad Request.
Check the username and password. The example source code just checks if the username is user and the password is password, but you will want to use some service in the filter to verify this information. If either of these fail, throw an AuthenticationException
If all goes well, a User is created from the authenticate method, and is injected into an Authorizer (which is a SecurityContext). In JAX-RS, the SecurityContext is normally used for authorization`.
For the authorization, if you want to secure certain areas for certain resources, you can use the #RolesAllowed annotation for your classes or methods. Jersey has support for this annotation, by registering the RolesAllowedDynamicFeature.
What happens under the hood is that the SecurityContext will be obtained from the request. With the example I linked to, you can see the Authorizer, it has an overridden method isUserInRole. This method will be called to check against the value(s) in #RolesAllowed({"ADMIN"}). So when you create the SecurityContext, you should make sure to include on the overridden method, the roles of the user.
For testing, you can simply use a browser. If everything is set up correctly, when you try and access the resource, you should see (in Firefox) a dialog as seen in this post. If you use cURL, you could do
C:/>curl -v -u username:password http://localhost:8080/blah/resource
This will send out a Basic Authenticated request. Because of the -v switch, you should see all the headers involved. If you just want to test with the client API, you can see here how to set it up. In any of the three cases mentioned, the Base64 encoding will be done for you, so you don't have to worry about it.
As for the SSL, you should look into the documentation of your container for information about how to set it up.
So this is really a matter what you would like to achieve. My case was to get this thing running with mobile and a One-Page-App JavaScript.
Basically all you need to do is generate some kind of header that value that will be needed in every consecutive request you client will make.
So you do a endpoint in which you wait for a post with user/password:
#Path("/login")
public class AuthenticationResource {
#POST
#Consumes("application/json")
public Response authenticate(Credentials credential) {
boolean canBeLoggedIn = (...check in your DB or anywher you need to)
if (canBeLoggedIn) {
UUID uuid = UUID.randomUUID();
Token token = new Token();
token.setToken(uuid.toString());
//save your token with associated with user
(...)
return Response.ok(token).type(MediaType.APPLICATION_JSON_TYPE).build();
} else {
return Response.status(Response.Status.UNAUTHORIZED).build();
}
}
}
Now you need to secure resource with need for that token:
#Path("/payment")
#AuthorizedWithToken
public class Payments {
#GET
#Produces("application/json")
public Response sync() {
(...)
}
}
Notice the #AuthorizedWithToken annotation. This annotaation you can create on your own using special meta annotation #NameBinding
#NameBinding
#Target({ElementType.METHOD, ElementType.TYPE})
#Retention(RetentionPolicy.RUNTIME)
public #interface AuthorizedWithToken {}
And now for the filter that implements checking of the header:
#AuthorizedWithToken
#Provider
public class XAuthTokenFilter implements ContainerRequestFilter {
private static String X_Auth_Token = "X-Auth-Token";
#Override
public void filter(ContainerRequestContext crc) throws IOException {
String headerValue = crc.getHeaderString(X_Auth_Token);
if (headerValue == null) {
crc.abortWith(Response.status(Response.Status.FORBIDDEN).entity("Missing " + X_Auth_Token + " value").build());
return;
}
if(! TOKEN_FOUND_IN_DB) {
crc.abortWith(Response.status(Response.Status.UNAUTHORIZED).entity("Wrong " + X_Auth_Token + " value").build());
return;
}
}
}
You can create any number of your own annotations checking for various things in the http request and mix them. However you need to pay attention to Priorities but that actually easy thing to find. This method needs using https but that is obvious.
Security comes in two main flavours :
Container Based
application based
the standard way to secure spring applications is to use Spring Security (formerly Acegi).
It would be interesting to know why you're not being allowed to use that.
You could use container based security, but I'm guessing that your use of spring precludes that option too.
Since the choice of Spring is usually to obviate the need for the use of a full J2EE container (Edit : though as pointed out below by others, most ordinary servlet containers do allow you to implement various container based security methods)
This really only leaves you with one option which is to roll your own security.
Your use of Jersey suggests that this might be a REST application.
In which case you really ought to stick with standard HTTP Authentication methods that
comes in the following flavours in reverse order of strength :
BASIC
Digest
Form
Certificate
REST applications are usually supposed to be 'stateless', which essentially rules out form based authentication (because you'd require the use of Session)
leaving you with BASIC, Digest and Certificate.
Your next question is, who am I authenticating. If you can expect to know the username AND the password of the user based on what URL they requested (say if it's one set of credentials for all users) then Digest is the best bet since the password is never sent, only a hash.
If you cannot know the Password (because you ask a third party system to validate it etc.) then you are stuck with BASIC.
But you can enhance the security of BASIC by using SSL, or better yet, combining BASIC with client certificate authentication.
In fact BASIC authentication over HTTPS is the standard technique for securing most REST applications.
You can easily implement a Servlet Filter that looks for the Authentication Header and validates the credentials yourself.
There are many examples of such filters, it's a single self contained class file.
If no credentials are found the filter returns 401 passing a prompt for basic auth in the response headers.
If the credentials are invalid you return 403.
App security is almost an entire career in itself, but I hope this helps.
As the former posts say, you could go with different options, with a varying overhead for implementation. From a practical view, if you're going to start with this and are looking for a comfortable way for a simple implementation, I'd recommend container-based option using BASIC authentication.
If you use tomcat, you can setup a realm, which is relatively simple to implement. You could use JDBCRealm, which gets you a user and password from specified columns in your database, and configure it via server.xml and web.xml.
This will prompt you for credentials automatically, everytime you are trying to access your application. You don't have any application-side implementation to do for that.
What I can tell you now is that you already did most of the 'dirty' job integrating Jersey with Spring. I recommend to you to go an Application-based solution, is it does not tie you to a particular container. Spring Security can be intimidating at first, but then when you tame the beast, you see it was actually a friendly puppy.
The fact is that Spring Security is hugely customizable, just by implementing their interfaces. And there is a lot of documentation and support. Plus, you already have a Spring based application.
As all you seek is guidance, I can provide you with some tutorials. You can take advantage from this blog.
http://www.baeldung.com/rest-with-spring-series/
http://www.baeldung.com/2011/10/31/securing-a-restful-web-service-with-spring-security-3-1-part-3/
Hi I am looking for a way to authenticate users when they make WebSocket connection and simply if they are not authenticated close the connection. I am using Dropwizard framework and Atmosphere for the WebSocket connections. Here is the example that I use.
It would be great if I could use '#Auth' annotation provided by Dropwizard for authentication when the connection is establishing.
How are you exposing this? Is it through a javascript frontend?
You are using the servlet based approach as described in your link and here: https://cvwjensen.wordpress.com/2014/08/02/websockets-in-dropwizard/, and not the jersey atmosphere extension?
If you are using the servlet based approach would recommend using a token-based approach putting the token in the http header an then access this header in the #Ready handler method, like this:
#Ready
public String onReady(final AtmosphereResource resource) {
String AuthHeader = resource.getRequest().getHeader("Authorization");
...DO AUTH HERE
logger.info("Resource {} connected ", resource.uuid());
return "Connect " + resource.uuid();
}
Then, you could also easily close the connection if auth fails. However, depending on your client side implementation, you might want to think about this. If the client auto-reconnects on close, you have a scenario of constant closing and opening of sockets that may cause a resource drain. You could, if auth is not successful, store a private variable that states whether or not this instance is authenticated and just drop sending any messages to or process any messages from it. That would also be a "obscure" way of letting an attacker know that auth failed, it simply is in a limbo state; connected, but not failed or closed. Just not receiving any data. But again, this is specific to your implementation.
Jwt auth is an option, check this out: https://github.com/ToastShaman/dropwizard-auth-jwt. I have, although not made it public, ported this implementation to dropwizard 0.8rc3-SNAPSHOT. If you need this, please let me know and I can post it to github.
I know this has been asked already, but I am not able to get it to work.
Here is what I would like to get accomplished:
I am using Spring Security 3.2 to secure a REST-like service. No server side sessions.
I am not using basic auth, because that would mean that I need to store the user's password in a cookie on client side. Otherwise the user would need to login with each page refresh/ change. Storing a token is I guess the lesser evil.
A web client (browser, mobile app) calls a REST-like URL to login "/login" with username and password
The server authenticates the user and sends a token back to the client
The client stores the token and adds it to the http request header with each api call
The server checks the validity of the token and sends a response accordingly
I did not even look at the token generation part yet. I know it is backwards, but I wanted to get the token validation part implemented first.
I am trying to get this accomplished by using a custom filer (implementation of AbstractAuthenticationProcessingFilter), however I seem to have the wrong idea about it.
Defining it like this:
public TokenAuthenticationFilter() {
super("/");
}
will only trigger the filter for this exact URL.
I am sticking to some sample implementation, where it calls AbstractAuthenticationProcessingFilter#requiresAuthentication which does not accept wildcards.
I can of course alter that behavior, but this somehow makes me think that I am on the wrong path.
I also started implementing a custom AuthenticationProvider. Maybe that is the right thing?
Can someone give me a push into the right direction?
I think pre-auth filter is a better fit for your scenario.
Override AbstractPreAuthenticatedProcessingFilter's getPrincipal and getCredentials methods.
In case the token is not present in the header, return null from getPrincipal.
Flow:
User logs in for the first time, no header passed, so no
authentication object set in securityContext, normal authentication
process follows i.e. ExceptionTranslation filter redirtects the user
to /login page based on form-logon filter or your custom authenticationEntryPoint
After successful authentication, user requests secured url, pre-auth filter gets token from header authentication object set in
securityContext, if user have access he is allowed to access secured
url
I have found numerous academic answers to this question, but I would like one from practitioners in the field.
Background
I would like to create a Java-based RESTful API, using the Grails framework for a variety of mobile clients (iOS and Android) to access protected resources through my service. I require authentication on certain requests, and I already have SSL setup over the wire (so all requests occur over https). My web API will eventually be exposed as a service to other web applications.
Problem
What authentication method do people recommend for a web service that is to be consumed by mobile devices, and eventually other web applications?
These are what I see as my choices. Can you tell me when would be appropriate use cases for each one?
I can do HTTP Basic authentication
I can do HTTP Digest authentication
I can implement OAuth authentication (1.0 or 2.0)?
I can pass the credentials as parameters in my request
I can use an authentication method above, and then pass a delegate/token around for authentication
I can implement my own custom HTTP authentication headers
I can use cookies and pass those to the server on each request
Other...?
Need
If you have one leaning one way or another, I'd like to know why you would choose that method. Better yet, if you're doing this in Grails, I'm very interested.
I already know...
I've already read through the excellent answers here and Richardson and Ruby's book, Restful Web Services.
REST is stateless protocol, thus using "work sessions" - I mean login/work/(auto)logout concept is somewhat questionable. Sending credentials as a parameter with each request seems to be the most often used method due to its simplicity. Just keep in mind that
1) The api url must be SSL only - it makes sense to use a dedicated domain AND ip address, e.g. api.example.com and configure your web server to handle SSL only for this address and domain. Just to avoid accidental disclosure of credentials.
2) Avoid using login/password with the request if possible, use "API key" (a shared secret) instead. You always may use "API key" instead of login/password if authentication is all you need, that is you do not need authorization (all users share the same permissions) and no need for logging (who did what).
3) Rather than sending the API key with each request, it is better to "sign" the request using shared secret and supply the signature. With the key long enough you may use this technique over plain unencrypted http.
=== responding to a comment ===
If you need authorization, just use basic authentication:
HTTPBuilder builder = new HTTPBuilder("https://api.example.com/v1/foo/bar")
builder.auth.basic(login, password)
builder.headers.put('Accept', 'application/json')
def result = builder.request(POST, JSON) { req ->
body = [
....
]
response.'201' = { resp, json ->
....
}
response.success = { resp, json ->
....
}
response.failure = { resp ->
log.error "failure, ${resp.statusLine}"
}
}