Consider I'm creating a RESTful API (like Twitter API) and I need only a single key token to access the endpoint GET /messages.
The token needs to be sent via header access_token for example.
If I don't pass the token I'll get an 4xx error.
In the RESTful API scenario, is this all needed to protect against CSRF attacks? I mean, in RESTful APIs I'm not using Cookies to manager a user session, since the idea here is to not be a session (Stateless is one of the principles of REST).
Am I forgetting something or is this correct? I don't see why I would need a anti CSRF token besides my access token.
Thanks in advance!
Short answer: if your access_token is not a cookie, yes, it's enough to protect you against CSRF.
Long answer:
A CSRF attack can only happen when HTTP client automatically sends some authentication data(usually cookies) to server. The key point is attacker do not need to know the Cookie, it's is automatically sent if web user accesses 2 sites(your site and the attacker's site) using the same browser.
To anti CSRF, server sends to client a private token that browser does not automatically send it to server(so it's must not be a cookie). The developer has to write code to attach the token to each subsequent request.
CSRF is an issue for using Cookie by Web browser, if you are developing restful API that is not used by web browser, usually, you do not need to care about CSRF
#Bart Firstly CSRF attacks are usually from a trusted user. Which means that the client has successfully authenticated himself by maybe passing a valid token in case of a rest API. The attack could be by passing let's say a javascript which can change the state of a resource. So just having a single key token won't really help to avoid CSRF attacks.
Please refer to below URL for more reference on CSRF :-
https://en.wikipedia.org/wiki/Cross-site_request_forgery
Most of the frameworks like Spring have inbuilt support for CSRF.
I'm struggling with setting up reliable and performant solution to communicate frontend with different microservices. I do not really now how to maintain (maybe not need) CSRF between my frontend and end services
Solutions stack: PHP, Laravel Passport, JWT, oAuth 2.0, Axios
Current approach:
Actually I've started up with approach from Laravel's passport
https://laravel.com/docs/5.4/passport#consuming-your-api-with-javascript
Using oAuth 2.0 to authorize user from website A to service B.
JWT token is returned for further communication.
Token is saved in cookie within website A
Once user is authorized website A uses JWT token to manage requests without additional to oAuth server, by sending JWT token as cookie using HTTP headers (withCredentials) to authorize user.
For each website A's request there was CSRF token created from service B since user is authorized and cookie could be applied by another unauthorized website to access service B. That was killing my performance since it has to retrieve CSRF for each request made. (that what I actually assume from laravel passport approach and need to create CSRF with JWT token - maybe that was mistake)
My concerns:
Regarding to of James Ward post:
http://www.jamesward.com/2013/05/13/securing-single-page-apps-and-rest-services
The easiest way to do authentication without risking CSRF
vulnerabilities is to simply avoid using cookies to identify the user.
Cookies themselves are not the cause of CSRF vulnerabilities. It’s
using the cookies on the server to validate a user that is the cause
of CSRF. Just putting an authentication token into a cookie doesn’t
mean it must be used as the mechanism to identify the user.
From my understanding setting JWT with website A's cookie with its domain set could not be accessed via any other site from outside. Since that there is no possible way to make request to service B without accessing JWT.
So do we really need CSRF then to secure potential attack to service B while using JWT?
If so, how could I achieve the best (in term of performant) way to generate CSRF through different services to be sure that communication would not be vulnerable for attack from different sites?
Any advice will be appreciated!
I am developing a REST application with its own authentication and authorization mechanism. I want to use JSON Web Tokens for authentication. Is the following a valid and safe implementation?
A REST API will be developed to accept username and password and do the authentication. The HTTP method to be used is POST so that there is no caching. Also, there will be SSL for security at the time of transit
At the time of authentication, two JWTs will be created - access token and refresh token. Refresh token will have longer validity. Both the tokens will be written in cookies, so that they are sent in every subsequent requests
On every REST API call, the tokens will be retrieved from the HTTP header. If the access token is not expired, check the privileges of the user and allow access accordingly. If the access token is expired but the refresh token is valid, recreate new access token and refresh token with new expiry dates (do all necessary checks to ensure that the user rights to authenticate are not revoked) and sent back through Cookies
Provide a logout REST API that will reset the cookie and hence subsequent API calls will be rejected until login is done.
My understanding of refresh token here is:
Due to the presence of refresh token, we can keep shorter validity period for access token and check frequently (at the expiry of access token) that the user is still authorized to login.
Please correct me if I am wrong.
A REST API will be developed to accept username and password and do
the authentication. The HTTP method to be used is POST so that there
is no caching. Also, there will be SSL for security at the time of
transit
This is the way most do it, so you're good here.
At the time of authentication, two JWTs will be created - access token
and refresh token. Refresh token will have longer validity. Both the
tokens will be written in cookies so that they are sent in every
subsequent requests
Storing the tokens in cookies I not dangerous in itself, but if you somehow get you JWT module on your server to read them from there you vulnerable to CSRF attacks where any webpage can trigger a users browser to send a form + you sites cookie to your server unless you use CSRF tokens. So generally they are stored in localStorage and "manually" added to request headers every time.
On every REST API call, the tokens will be retrieved from the HTTP
header. If the access token is not expired, check the privileges of
the user and allow access accordingly. If the access token is expired
but the refresh token is valid, recreate new access token and refresh
token with new expiry dates (do all necessary checks to ensure that
the user rights to authenticate are not revoked) and sent back through
Cookies
Apart from the cookie dangers, it seems safe.
Provide a logout REST API that will reset the cookie and hence
subsequent API calls will be rejected until login is done.
You don't even need to make an API call, you can simply just purge the cookies or the localStorage object and make sure your client doesn't break on missing tokens.
The standard for the express-jwt module expects the tokens to be in its own "Authorization: Bearer [Token]" header, which I would strongly recommend over cookies. The localStorage API is available all the way back to IE8 so you should be good.
Edit:
First, it's important to know the difference between XSS and CSRF attacks since they're often believed to be the same thing.
XSS is when users get unsafe JS running on your domain in other users browsers when that happens neither JWT in localStorage or sessions and JWT in cookies are safe. With httpOnly flag on cookies, you can't directly access them, but the browser will still send them with AJAX requests to your server. If this happens you generally out of luck. To prevent this, make sure to escape all user input if it's sent to the browser.
If you load 3rd party JS with script tags or iframes this might compromise localStorage unless you are careful, but I haven't worked enough with this to help you here.
CSRF is only when other domains are trying to send normal HTML forms to your server by getting the browser to send cookies automatically. Frameworks prevent this by inserting unique random strings as hidden fields and checking them again when it's submitted. JWT's in localStorage is safe from this since each domain gets its own separate localStorage area.
But ultimately all this depends on if your service will be using one single domain, in which case httpOnly cookies will be plenty secure and easier to set up, but if you wanna spread your service out on multiple domains like api.domain.com + app.domain.com or add a native app you're forced to store you're JWTs in localStorage or some other native storage area.
Hope this helps!
I asked this question two years back and also accepted the answer. However, based on my experience and study in the last two years, I'd like to answer this just in case someone stumbles on this thread with the same question.
The approach mentioned in the question is similar to the "Resource Owner Password Credentials" grant type of OAuth 2.0. However, I think it is better to use the "Authorization Code Grant" type instead and Cookie to store the tokens instead of browser localStorage or sessionStorage. I have detailed my reasons, implementation points, security considerations and references in this StackOverlow answer.
Like OP I been using resource owner password grant.
I learned so much from Saptarshi Basu's other answer in a different post I think anyone looking into OAuth Code Flow should take a look at it, it has outlined a very solid approach to auth SPA and resource servers. It primarily relies on your backend(resource server) to handle authentication with the auth provider as a private client.
However, I will just add that people looking at implementing authentication with SPA should also consider OAuth Code Flow with PKCE. The main goal of PKCE is to allow public client such as SPA to authenticate directly with auth provider. All PKCE adds, is that when a SPA app initiates authentication, a hashed value is sent to the auth provider when the user is authenticated. And after user authenticate with the authorization provider, it redirects the user back to SPA with that hashed value as well as authorization code. Now, for the next part where the SPA calls auth provider to exchange code for tokens, instead of providing client secret, it has to provide the key that was originally used to create the hashed value. This mechanism guarantees the code cannot be used by someone who intercepted the code, and the SPA doesnt need to store a client secret like a server-side app does.
Now the only thing I'm not certain at this point is which is technically more secure, server-side authentication using standard Code Flow without PKCE or SPA authenticating directly using PKCE? Most resources I could find online currently describes and recommends the latter . However I feel that letting a private server side client handle authentication (as Saptarshi Basu described) might still be more secure. I would love to hear his opinion on this as well.
My understanding of refresh token here is:
Due to the presence of refresh token, we can keep shorter validity period for access token and check frequently (at the expiry of access token) that the user is still authorized to login.
Please correct me if I am wrong.
Assuming you're talking about using JWT as Bearer-token in OAuth (and I would strongly advice to follow the OAuth 2.0 protocol), that's right.
With an additional auth-time (timestamp of authentication) claim in your JWT, you could even drop the second token and sent your access- as a refresh-token (the auth-server could then issue a new access-token if token is valid & auth-time within allowed range)... but sure, it's also good to follow the standard ;)
Anyway, there are certain additional aspects (that tend to get difficult or are even against the fundamental ideas of JWT) you should consider before using JWTs as refresh-token, as this basically means you introduce long-living JWT:
do you need to have something like forced user logout/ token revocation by subject (e.g. if user got identified as fraudulent)?
do you need to have something like revocation of a specific token (e.g. if a user looses a device)?
...
Dependent on your use-case you should consider all the possible implications, long-living tokens have as they usually require you to introduce some kind of state on your server-side (e.g. to allow revocation/ blacklisting). Keep in mind the beauty and security of the JWT concept lies within JWTs being short-lived.
We would like to implement "Sign-in with LinkedIn" in our app. Since the app has JS fronted and RESt-based backend, we decided to exchange JSAPI tokens for REST API OAuth tokens as described here.
If a user successfully signs in, the frontend sends credentials cookie with client-side bearer token and member ID to the backend. On the backend we check if a user with such a member ID already exists and if not, we exchange JSAPI token for REST API OAuth token, retrieve user details from LinkedIn a store it in our database.
Now the question is if we can use that cookie to authenticate each user's request to our REST backend. After a user successfully signed in via JSAPI, the cookie should be automatically passed to our backend on all subsequent requests so we can check member ID. Are there any drawbacks that we missed? Or is this idea as a whole wrong?
Should we rather authenticate a user only once by means of the cookie and then issue our own authentication token and send it back to the client?
The way cookies work in general is they are passed on every request to the domain they belong to. LinkedIn is setting a credentials cookie to your domain.
As long as you are validating those credentials on every request it's perfectly acceptable to use their tokens as authentication.
Personally I don't find that to be a great idea and would prefer to validate their credentials once and create my own auth token to use from there on out. You can always set that token to expire at some-point and re-validate the LinkedIn credentials (which will still be getting sent on every request anyway). This limits the amount of times you're checking with LinkedIn and should increase the responsiveness of your app.
Either way could work.
If you are using the LinkedIn cookie to validate a user by member id, you should validate the cookie's signature on each request per section 2 of the doc you linked and question 2 of the FAQ.
Using your own token could make it easier to implement an account which belongs to your app and is not necessarily connected to LinkedIn, assuming there's the potential to either connect solely with some other service(s) or no 3rd part(y/ies). Still should validate any time you trust the member id in the cookie though.
The doc provides a validation example in PHP, and if you're interested in improving a ruby version, I have a shameless plug.
The flow that you've outlined in your latest comment of going straight for the OAuth tokens is the best way to go if you were only signing in to convert the JSAPI tokens to OAuth tokens and then not using the JSAPI further. If you were planning to actually use both the JSAPI tokens within your front-end app and the OAuth tokens on your back-end, then it's better to take the conversion route.
Why do you need both a "code" and a "token" in the Facebook OAuth2 authentication flow as described here: https://developers.facebook.com/docs/authentication/ ?
If you look at the OAuth dialog reference (https://developers.facebook.com/docs/reference/dialogs/oauth/), it seems like you only ever use the token to fetch information about the user, and if you specify the response_type parameter as token or code,token, then you get the token on the first time.
Why do you need to get a "code" and then use the code to get a "token" as opposed to getting the token directly?
I guess I'm misunderstanding something basic about how OAuth works, but it seems you avoid the request to https://graph.facebook.com/oauth/access_token entirely if you get the token the first time with the dialog.
Let us take a simple example to differentiate authentication code vs access token.
You as a user want to try a new Facebook app called Highjack.
So you click on the application and the Highjack app asks you to log into your Facebook account. When you are done, Facebook generates an authentication code for you.
This code is then passed to the Highjack server which uses its own FB client id, FB secret and your authentication code to get an access token.
In the above example the authentication code is confirming you as a user is a valid FB user. But the second steps says "you as a FB user is giving access to the Highjack app for certain resources".
If the Highjack app wanted implicit grant (i.e direct access token), then the access token would be visible to you also since it is being exchanged with the browser. This means you can now call all Facebook APIs on behalf of Highjack using the access token. (You can only use the access token to get your personal information but Facebook has no way of knowing who is calling their APIs.)
Since we have 2 parties (You and Highjack) authenticating with Facebook we have this 2 fold mechanism.
Borrowed shamelessly from Salesforce Documentation:
Authorization Code
An authorization code is a short-lived token representing the user's access grant, created by the authorization server and passed to the client application via the browser. The client application sends the authorization code to the authorization server to obtain an access token and, optionally, a refresh token.
Access Token
The access token is used by the client to make authenticated requests on behalf of the end user. It has a longer lifetime than the authorization code, typically on the order of minutes or hours. When the access token expires, attempts to use it will fail, and a new access token must be obtained via a refresh token.
From the OAuth 2.0 Spec:
The authorization code provides a few important security benefits
such as the ability to authenticate the client, and the transmission
of the access token directly to the client without passing it through
the resource owner's user-agent, potentially exposing it to others,
including the resource owner.
So, basically - the main reason is to limit the # of actors getting the access token.
"token" response is intended primarily for clients that live in the browser (e.g.: JavaScript client).
Answer) You need/want both the code and token for extra security.
According to Nate Barbettini we want the extra step of exchanging the authentication code for the access token, because the authentication code can be used in the front channel (less secure), and the access token can be used in the back channel (more secure).
Thus, the security benefit is that the access token isn't exposed to the browser, and thus cannot be intercepted/grabbed from a browser. We trust the web server more, which communicates via back channels. The access token, which is secret, can then remain on the web server, and not be exposed to the browser (i.e. front channels).
For more information, watch this fantastic video:
OAuth 2.0 and OpenID Connect (in plain English)
https://youtu.be/996OiexHze0?t=26m30s (Start 26 mins)
If you look at the flow of Authorization Code OAuth type, yes, there are actuary two steps:
<user_session_id, client_id> => authorization_code
<client_id, redirect_uri, authorization_code, client_secret> => access_token, refresh_token
In step1: the user tells the OAuth Server that "I want to auth this client (client_id) to access my resource. Here is my authentication (user_session_id or what else)"
In step2: the client (client_id) tells the OAuth server that "I've got the user the authorization (authorization_code), please give me an access token for later access. And this is my authentication (client_id & client_secret)"
You see, if we omit step 2, then there is no guarantee for client authentication. Any client can invoke step1 with a different client_id and get an access token for that client_id instead of its own. That's why we need step2.
If you really want to combine step1 and step2, you can do something like this:
<client_id, redirect_uri, client_secret> => access_token, refresh_token
We use this approach in our Open API Platform, and we haven't find any security problem yet.
BTW, there is actually an Implicit Grant type, that is:
<client_id, redirect_uri> => access_token, refresh_token
It is generally applicable to client only application which have no server backend. In that case, the OAuth server must ensure that the redirect URI belongs to that client (same with the register redirect_uri, for example).
The mix-up came because the user on behalf of himself and not the client app authenticate against the authorization server (i.e. facebook).
Its much simple to secure the client app (with https) then the user-agent (browser).
Here is the original formulation from IETF-oauth (https://datatracker.ietf.org/doc/html/draft-ietf-oauth-v2-threatmodel-08#section-3.4):
3.4. Authorization Code
An authorization code represents the intermediate result of a
successful end-user authorization process and is used by the client
to obtain access and refresh token. Authorization codes are sent to
the client's redirection URI instead of tokens for two purposes.
Browser-based flows expose protocol parameters to potential
attackers via URI query parameters (HTTP referrer), the browser
cache, or log file entries and could be replayed. In order to
reduce this threat, short-lived authorization codes are passed
instead of tokens and exchanged for tokens over a more secure
direct connection between client and authorization server.
It is much simpler to authenticate clients during the direct
request between client and authorization server than in the
context of the indirect authorization request. The latter would
require digital signatures.
Theoretically,
Access Tokens cannot tell us if the user has authenticated but auth code does.
Auth code should not be used to gain access to an API but access token should be.
If you have a single page application or mobile application with no or minimum backend, your application may want to access user's FB data directly at frontend. Hence the access token is provided.
In another case, you may want a user to register/login to your app using some external auth service provider like Facebook, Google etc. In this case, your frontend will send the auth code to the backend that can be used to get access token from Facebook at serverside. Now your server becomes enabled to access user's FB data from the server.
Basically, as an extension of Lix's answer, the access code route allows a Resource Owner (i.e. the Facebook User) to revoke authorization for their User Agent (i.e. their browser), e.g. by logging off, without revoking authorization for an offline Client (i.e. Your Application).
If this is not important, then there is no need to use the access code route.
Furthermore, the access code is provided to ensure that the Token provided to a server is actually registered to the Resource Owner (i.e. the Facebook User), and not the User Agent (or a Man-in-the-Middle).
This seems similar to the question of either choosing the implicit vs authorization code grant flow. In fact, here is what looks like an opposite view point?!.
Also, as Drew mentioned,
When the access token expires, attempts to use it will fail, and a new access token must be obtained via a refresh token.
another piece is the refresh token, but I don't see that being explained too well in the FB Docs. If I'm correct, the implicit grant (the direct token) should be really short lived, but that is to-be-enforced and FB.js seems to hide a lot of that (this one I have not looked as deep into).
If I'm correct, the code%20token is an optimization allowing both the User Agent to have a token and allowing for the server to initiate the token exchange process in a single request (as anything over Network IO is considered expensive, especially to a User Agent).
In OAuth 2.0 with facebook, the overall concept is simple as follows.
Step 1. Obtain "Authorization Code" by a GET request
request URI: https://www.facebook.com/dialog/oauth
Params:
response_type=code
client_id={add your "App id" got by registering app}
redirect_uri={add redirect uri defined at the registration of app}
scope={add the scope needed in your app}
Headers: None
Step 2. Obtain the "Access Token" by sending the authorization code as a POST request
URI: https://graph.facebook.com/oauth/access_token
Params:
grant_type=authorization_code
client_id=<add your "App id" got by registering app>
redirect_uri=<add redirect uri defined at the registration of app>
code=<obtained authorization code from previous step>
Headers:
Authorization:Basic encode <App Id:App Secret> with base64
Content-Type:application/json
Step 3. Use the access token got from above step and retrieve user resources
It’s because the access token is given to an AUTHENTICATED client (third-party app) using a shared secret that only FB and the client knows. The only way that the user could directly request the access token is by knowing the shared secret, which would make the secret public and could lead to a man-in-the-middle attack. Further, while FB can guarantee a secure connection to the user, FB can’t guarantee the handoff of the token to the client is secure. However, FB (and OAuth2) does require a secure connection between the client and FB. The access token is tied to the client public ID (usually hashed), which means only the original client application can use it to request the token because the secret is sent along with the authorization code to get the access token.
You recieve a token when the user logs in. But you might want to change the token when you are performing other actions. EG posting as your app/page or posting as a user with offline_access.