I need to implement a system where the frontend is a pure SPA made with react (by pure I mean it is only hosted as a bunch of static files) and a nodejs REST API made with express. I can not use any 3rd party authentication providers like Auth0, etc. so everything must work locally.
The app must have authentication and it must be more or less up to modern standards. I am looking to implement PKCE flow with short-lived JWT auth tokens and long-lived refresh tokens.
There is one thing I am confused about. Every guide on PKCE flow always mentions redirects from one URL to another during the auth flow. In my setup, however, redirects are highly unwelcome.
I want to check and find out whether it is OK to implement PKCE flow using pure XHR requests to communicate between the SPA and API. Is it appropriate to do so? What major risks are there compared to a setup with redirects?
EDIT: to make this question more specific, what should be the PKCE authentication flow if you can only use SPA and XHR (and must avoid redirects)?
There is one thing I am confused about. Every guide on PKCE flow always mentions redirects from one URL to another during the auth flow. In my setup, however, redirects are highly unwelcome.
The whole point with OAuth 2.0 is the delegation (redirect) to the ID Provider, such that the user always send his password to the ID Provider instead of sending the password to each application. So you will always need to redirect the user to sign in - but hopefully the user is already signed in, so he immediately becomes redirected back to your app. There are techniques to check if the user is signed-in using XHR, that can be used until he need to sign in.
what should be the PKCE authentication flow if you can only use SPA and XHR (and must avoid redirects)?
In the Authorization Code flow, the client typically does two request to the ID provider. The request to authorization endpoint typically is done through a redirect (e.g. in case the user need to authenticate), but the second request, to token endpoint can typically be done in the background using XHR or fetch using appropriate CORS configurations.
Related
I am building a React-based SPA that communicates with a spring-boot backend via a REST API. I need the user to be able to log into their Microsoft account on the browser client (the SPA) and I need the backend service (spring-boot app) to be able to query Microsoft's Graph API on behalf of that user.
After reading up on the Oauth2 flows, the authorization code flow (not the PKCE flow, just the regular authorization code flow) seems the most appropriate. The browser client could let the user log into their Microsoft account, retrieve an authorization code, and send the authorization code to our backend service via HTTP request. The backend service (which is trusted and can safely store a client secret) can then request an access token, make requests to the Graph API directly (meaning that the SPA would never need to make any requests to the Graph API), and silently refresh the token as needed.
However, I cannot see any examples of anyone using this flow to access Microsoft's Graph API.
Looking at Microsoft's documentation, it seems like they recommend using the on-behalf-of flow. But this flow requires the browser client to request an access token and then use that to communicate with the backend service (which in turn can communicate with the Graph API). It doesn't make sense to me why the access token cannot be requested on the backend using a client secret. Wouldn't this be a more secure and preferred method than having the client retrieve the access token, as is done in the on-behalf-of flow?
The Oauth2.0 site, recommends that SPAs should either use the authorization code with PKE or the implicit flow, but I do not see an option to use the standard authentication code flow for SPAs. Should I take this as an indication that SPAs should not be using the standard authorization code flow as I described earlier?
Despite not finding a clear-cut example of the standard authorization code flow in Microsoft's documentation for a react frontend + java backend, I tried to go about doing this myself. However, using the #microsoft/mgt-react and #microsoft/mgt-element libraries to do this are not straight forward. For example, the #microsoft/mgt-element notion of a Provider supports a call to retrieve an access token, but doesn't clearly expose the authorization code. If I wanted to do the authorization code flow described earlier, it seems like I would need to use raw HTTP requests, which I know is not a recommended way of accomplishing this.
Summarizing my questions:
What OAuth2.0 flow should I be using: 1) authorization code (access token is retrieved by backend service using client secret), 2)
authorization code with PKE (access token is retrieved by client), or
3) on-behalf-of flow (access token is retrieved by client, seems to be an extension of PKE flow)?
If using the on-behalf-of flow, does the SPA just include the access token in the header (marked as 'bearer') and the backend service just
includes that same header to query the Graph API, or does the backend
service need to request another token before querying the Graph API?
Agree with #ch4mp to call graph api directly in SPA if it's allowed. If not, then I recommend you using on-behalf-flow or client credential flow based on your requirement.
Let's come back to your requirement -- call ms graph api in a springboot api project. First, let's see one of the graph api getting user api. You can see permission types here: Delegated which means call graph api on behalf of the user, Application which means calling api on behalf of the application(your spingboot api project) itself. If you want to call api behalf of the user, then you have to use on-behalf-of flow. This is because the api project which will be considered as a daemon application, so the project itself doesn't have a UI page to let users enter username/password to sign in and get authenticated.
You can certainly use ROPC flow which have to pass the username/password to api but I really think it unsafe, so I don't recommend.
If it's not necessary for you to call graph api on behalf of user, you can certainly take client credential flow into consideration. But pls note here, application type api permission is a "large" api permission which always have name like User.ReadWrite.All, Mail.ReadWrite.All and it always means the application can not only query user information but also be able to modify user information.
If you want to use on-behalf-flow, then you may review this answer and it explained the whole progress...
I would use authorization-code flow (with PKCE) to get an access-token and then refresh-token flow to "maintain" this token, both from client.
Authorizing the request to your resource-server with this token only makes sense if Microsoft authorization-server is your main authorization-server. Otherwise (user also logged in with an authorization-server of your own or not using OAuth2 betwean React and backend), you can still send Microsoft access-token in request body.
In any case, when issuing requests in the name of the user from the backend, do as you suggest: just set the access-token sent by the client as Bearer Authorization header (token is retrieved either from Spring security context or request body). Backend fetches a new access-token (using client-credentials flow) when issuing requests in its own name (without the context of a user).
Side note: have you considered calling Microsoft API directly from React client? If you don't have to store the result of that call on your resource-server (i.e. call graph API to display data and store only what user selected from that data), that would save quite some latency on the client and costs (network and CPU) on the backend.
I have a client who is using OAuth2 for single sign on with their own login page. Once user is logged in they will be redirected back to my React SPA i'm building.
What I'm trying to find out is a way for my app to integrate with OAuth2 to keep check if user is still signed in. For example say they refresh the app I need a way to check OAuth2 if user is still authenticated.
Should I use an express server to manage the AUTHORIZATION_CODE returned by the redirect from OAuth2 ?
From what I understand OAuth2 returns a AUTHORIZATION_CODE and STATE in the redirect url.
If so how would I manage this on the express side ? Is there a express plugin to manage this process.
Or could I just bypass using express and just use the Auth0 React SDK to check if user is Authenticated ? I found this example on the auth0 docs. https://auth0.com/docs/libraries/auth0-react
Not sure if this would work because of my setup where the login happens from the client login page and redirects back to my app ?
Authorization codes are meant to be exchanged with access tokens. After you exchanged the code with an access token, what you technically need to do is to inspect the token to check whether it's active or not, with active status meaning that the user is logged in. This can be done in a couple of ways, the following are just the ones I can think of:
Most conventional way I know is to send an API request to the introspection endpoint while including the access token. The response will tell whether the token is valid and/or active.
If the token if in JWT format, it is usually possible to verify the token signature and verify the expiry time after decoding it. Depending on the authorization server, it might be possible to send this token to the introspection endpoint as well.
Unfortunately I can't help with the specifics as the OAuth2 server being used is not mentioned. But if you actually use Auth0 as your authorization server, you can probably use their libraries to help with their OAuth 2.0 flow, such as logins and auth status checks. Otherwise you might need to find another library like node-openid-client or do it manually as explained.
We are developing an application that uses a React front end website hosted on AWS using Amplify. This communicates with a .NET Core 3.1 Web API running on EC2 / Elastic Beanstalk. Cognito is used for user authentication with the Web API configured to use JWT tokens.
It works OK, but we have noticed that the Cognito provider stores the JWT access token in the browser local storage. This is what we see using F12 in Chrome and inspecting local storage.
From what we have read, storing access tokens in local storage is not advised as it makes the application susceptible to XSS attacks. Strange then, that the Cognito identity provider chooses to store sensitive information here.
If this approach is not considered safe, can the provider be configured to store this information elsewhere, such as cookies?
Alternatively, as we control both front and back ends, is there an alternative method that can be used to secure the API that does not involve tokens? Obviously the API needs to know which user is logged on to the web application in order to perform authorization checks. [Note authorization in the application is record level and defined in database tables, so it goes beyond simple user profile attributes.]
Many thanks in advance for your advice.
Doug
Security is a spectrum not a feature so it really depends on your appetite for risk vs effort. Amplify is not a particularly nice codebase, it has 500+ issues and if you look at the code you might be fairly shocked at the quality of it.
If you are using Hosted-UI then you can write code to manage the tokens yourself rather than using amplify, although you will need to learn a bit about OAuth grants and OIDC.
Be aware that the Hosted UI lacks a huge amount of features, so if you are going to use it make sure you are happy with it. Off the top of my head
no silent refresh capability in the hosted UI, so no safe way to store the refresh token.
no support for custom auth flow in the hosted UI
no passwordless support in the hosted UI
no ability to pre-populate a field in the hosted UI (e.g. username)
no ability to customise the plethora of obscure error messages in the custom UI
fixed now, but for years the email addresses were case sensitive!
An alternative is also to just use the AWS SDK to get tokens directly using cognito-idp but this also has a bunch of issues:
no code/PKCE/nonce capability so insecure in a mobile authsession
no ability to set oauth scopes, so can't use them
consequently not possible to use for OIDC
the SRP implementation is bananas and so far off spec
if you make device registration mandatory, it will deliver a working access tokens before the device is registered! (allows invisible devices for malicious logins)
We were using auth0 which was leagues ahead but we had to move to Cognito because of SMS OTP cost (min $25k per year at auth0).
I have been using AWS for over a decade now, Cognito is by far the worst service I have used, and I have used a lot! If you can avoid it, do so.
To answer the original question, yeah it's insecure. The best you can probably do is keep them in memory. If you wanted to you could probably put the hosted UI behind a cloudfront and use an lambda#edge to transform the token into a cookie instead. This has now opened you up to CSRF attacks though.
answering the original question: no, it is not safe at all.
Storing refreshtoken in any local storage accessable to any local app/script is not secure. So, the best way would be to store the refreshoten (and also the access token) in an httponly cookie or even better to store a one-time session token in httponly secure cookie could be used to get new access and refresh cookies - similarly as it is made by cognito hosted ui with XSRF-TOKEN.
See below how I would solve (and plan to solve) this issue:
Some background:
Due to GDPR regulations I think I can not use the cognito hosted ui - I have to make sure users read and accept the general terms and conditions (giving clear and auditable consent) and can review and accept cookie policies as well before they type in any user data for sign up. Nevertheless the built in hosted ui design is quite outdated and unflexible. I have an SPA website where I want to manage users, secure endpoints, etc.
So I have the following idea which is still not super secure but I think it is more secure one if you want to use js and ampify sdk and which also might answer your question:
I'll use amplify javascript sdk to let users sign up, change psw and log in (get tokenid, access token and refresh token), will make my own "hosted ui". I'll store the access token in memory only (not in local cookies and not in localstorage for sure). Access token will be used in header (bearer) to access apiGW endpoints. Access tokens will have very short expire dates. (I'd also use httponly secure cookies sent back by the apigw, as well as in the body.., then compare at BE side..)
And here comes the trick: I'd cut the refresh token into two. (Don't forget it is just a string.) I'd store the first part of the string in a local cookie (javascript can read it, if browser is closed and opened again it will be still there) and will send the other half of the refresh token to an apiGW endpoint (accessable without authentication) which will store it in a dynamoDB table (with TTL) and will send back an httponly secure cookie to the browser with a randomly generated "storagetoken" in it (which will be a key in dynamodb). There will be another unauthenticated apigw endpoint which will be called by the client whenever the client needs the full refresh token. Calling this endpoint the browser will send in the httponly secure cookie as well (same domain), so the backend will get it. As it is issued by the BE and available only in the given browser it can not be stolen so the backend will send back the stored half refreshtoken. The other half part of refreshtoken is stored in a simple cookie in the browser.
If the browser is closed and opened again client checks if there is any valid accesstoken and if not it checks if there is a half refeshtoken stored as cookie. Then ask the other part of refreshtoken assuming there is a httponlycookie also stored and it will get back the other part of the refreshtoken from the BE. In case of success the client tries to use the full refreshtoken to renew/get access token from cognito, in case of failure it will pop up the login screen.
Whenever refreshtoken is not in use it is deleted from the memory.
I know this is still not supersecure but might be a better solution than storing refresh token in localstorage.
Alternatively, as we control both front and back ends, is there an alternative method that can be used to secure the API that does not involve tokens?
I don't know anything of Amplify but in AWS Cognito what you describe is the Implicit grant OAuth flow. In AWS Cognito it is possible to use Authorization code grant where you instead of the token get a code which you in the backend can exchange for a user pool token.
https://docs.aws.amazon.com/cognito/latest/developerguide/cognito-user-pools-configuring-app-integration.html
I am not sure if I should create it here on StackOverflow or another stackexchange channel, but let's try here.
We have a web api made in asp.net core which uses the basic authentication where another web app post some login data to the api and it respond a token for the next requests. The client app stores this token and the next request to get/post data uses this token key for authentication. It works fine from the api perspective.
The point here is our web app. We are building it using react.js and the point how to keep the authentication token safe. We store the token on the current app (which is executed in a web browser). We have a feeling about store it on the browser because bad users can access the console on devTools and investigate how to to get the token from the global variables on the react app (just a sample). Given this point the questions are: How to deal with it to keep the back-end and front-end safe? How make sure the users cannot get the auth token and use it on another apps?
We were thinking in creating a kind of server-side channel just to store the authentication token like the picture bellow:
The web browser app make requests to server-side channel to get/post some data;
The server-side channel make a new request the API defining the authentication token and repassing the get/post data;
The api process the request and respond;
The server-side channel get a response from api and send it to the browser;
The web browser app get the final response.
It could work and keep it safe, but another question is: How computationally expensive is that?
Is there any possible solutions or suggestions how to deal with it?
Thank you.
Use JWT access tokens against your API and authenticate your SPA with an identity provider using an Open ID Connect flow (OIDC).
https://www.ubisecure.com/single-sign-on/single-page-application-and-openid-connect/
There are lots of examples of this, Identity Server is a common OIDC implementation with examples, http://docs.identityserver.io/en/latest/quickstarts/6_javascript_client.html
Once you've gone through the OIDC flow and acquired an access token for the user, you can store this client side safely, as
The access token has an inbuilt lifetime and once it's expired can no longer be used. Good practice is to keep this lifetime short (thus limiting the attack vector) and provide some sort of token refresh logic to automatically keep the user working against the API, as long as they keep the SPA open.
Your netcore web api has all the libraries it needs to do token validation / lifetime valdiation etc. This has been made very simple at the API layer
NB: I mention safely as there is still an attack vector, someone who acquires the JWT can act as that user for the lifetime of the token against your API, they are the bearer of the token. It's up to you to make sure the lifetimes of your tokens are sane and the the process for acquiring a new token is as secure as possible
There are a lot of examples on how to implement this, and whether or not you want to use your own Identity Server or use a solution such as Auth0.
Don't try and roll your own security solution. Stick to the specs and standards and adhere to all the industry best practices, making use of battle-tested libraries.
store token in local storage in web browser in encrypted form
I'm considering using AWS Cognito as a user management system for a single page web app I'm building using React along with a Python REST API backend (Pyramid). I'm struggling to see how all the pieces fit together for my architecture (the docs don't seem to be helping me). There are many great examples of how to implement authentication into the frontend using JS. My issue is how to integrate this authentication into my backend REST API.
In my current hand rolled user management system, the frontend calls the REST API on sign-in and is given a token which is passed to API again for every subsequent request. I'm then able to use ACL's on my API functions, check permissions to access resources, etc. If I were to use Cognito and perform the authentication on the frontend (as many examples do) how will my backend know if the token is valid when it receives it with a request? Surely I wont have to call Coginto from the backend to verify this for every request? Also how can I perform checks for information such as 'is this user in the admin group' if that group is defined within Cognito? Again, calling out to Cognito for every request seems very heavyweight and cumbersome.
I did see one example where a list of valid tokens was exported from Cognito as a JSON file and kept on the backend. This seems horribly static when users could be added and removed regularly.
Is Cognito really suitable for my use case? Some high level guidance and pointers to any relevant examples and docs would be greatly appreciated!
When authenticating with Cognito, the user can have 3 tokens:
Refresh
Access
ID
For python, boto3 can interface now with Cognito. There's also this python lib wrapper: warrant, to make it easier.
Once you have the token, it is possible to pass it to the API (eg: access) and it can be checked on the server side with python-jose, as per AWS docs
To pass the token, an example pyramid /login implementation can keep the information in the session before setting the request response:
request.session['my_token'] = str(a_token)
The default cookie session factory works, though it warns that the token is not sent encrypted.