So, I'm developing an API using slim/slim and league/oauth2-server to manage the OAuth2 connection. OAuth2 will be useful because I will need to use Client Credentials grant between services.
Then, I'm also developing an hybrid app with React Native. This app will requires user login by using e-mail and password or connecting with another services (such as Facebook, Google, Twitter, etc).
And I'm confused about what OAuth2 flow to use for this case. Across the web are a lot of articles saying that Resource Owner Password Credentials is not safe anymore, and we should use instead Authentication Code with PKCE.
But I can't discover or understand how to apply Authentication Code with PKCE in a first party app, because all documentation talks about you will need the uses a browser to get authentication code in redirect_uri.
The flow I imagine is something like that:
User open the app, then insert your credentials username and password;
This screen will connect to API /request_token URI sending { 'grant_type': 'password', 'username': username, 'password': password, 'client_id': CLIENT_ID }, considering it as a public app we can't send client_secret;
The API validates credentials and returns some data such as { "access_token": access_token, "token_type": "JWT", "expires_in": LIFE_SPAN }, here we will use JWT to gerenate the access_token based in public/private key;
Authentication done, the app will store the access_token while it's alive and when it expires will do the flow to refresh_token.
My question: is it safe? Scott Brady did some "aggressive" article talking it's NEVER safe.
How apps does this things? When I use Instagram app, for example, they own the app and the API, I don't need a browser in the User Experience flow. Are modern apps using "Resource Owner Password Credentials" or "Authentication Code with PKCE"? There is a away to avoid insert browser in the flow while using "Authentication Code with PKCE"?
[EDIT] Possible Solution
As Gary Archer said "Auth Code flow with PKCE is recommended - along with logging on via the system browser", but we are not talking about grant permissions to access users data or third-party apps.
As a designer I don't agree that loggin in the first-party app owned by the same API owner requires a browser this is the not the User Experience we are looking for. And all apps we see such as Instagram, Facebook, Uber... we just put your username and password and we have access to your account.
What I will do is create a custom version of Authentication Code with PKCE removing the required_uri.
[EDIT:2] The New Flow
After a lot of search, I found some answers I think was interesting to adapt. As above, I removed redirect_url from flow. Look:
The flow starts in a login screen, when user give your credentials;
Client generates a code_verifier then hashes code_verifier to code_challenge and sends it to Authorization Server with following parameters:
response_type=code : indicates that your server expects to receive an authorization code.
client_id=xxxx : the client id.
client_integrity=xxxx : app integrity check for first-party app.
code_challenge=xxxx : the code challenge generated as previously described.
code_challenge_method=S256 : either plain or S256, depending on whether the challenge is the plain verifier string or the SHA256 hash of the string. If this parameter is omitted, the server will assume plain.
username=xxxx : username to authenticate.
password=xxxx : hashed version of password.
state=xxxx : a random string generated by your application (CSRF protection).
Authorization Server will validates user authentication, stores code_challenge and return the authorization_code with a client_token;
After receive the aauthorization_code and client_token, Client saves the client_token and immediately send authorization_code back to Authorization Server with following parameters:
grant_type=authorization_code : ndicates the grant type of this token request.
code=xxxx : the client will send the authorization code it obtained.
client_id=xxxx : the client id.
code_verifier=xxxx : the code verifier for the PKCE request, that the client originally generated before the authorization request.
Authorization Server will validates all data and, if everything is right, will return the access_token;
Client will set Authorization header with the access_token and always send client_token to every request, it will be only accepted with both values are right;
If access_token expires, then Client will do a request to refresh access_token and get a new one.
Now, I will reproduce this logic to PHP language. If everything goes right and I hope it does, I will be back with definitive answer.
[EDIT] Clarifications
I'm using OAuth2 to user connect with your third-party accounts (Google, Facebook, etc). But user also can log to a local account in my database. For this case, user doesn't need to grant anything at all. So, no makes sense send user to a browser to him does your login.
I wondering if, to this case, local accounts, we can use Resource Owner Password Credentials or it's more safe Authentication Code with PKCE (we already conclude it's a better approuch). But Authentication Code with PKCE requires redirect_uri, do I need uses this redirection to log users into a local account where they don't need to grant access?
Let's go then. After a lot research, I found some approaches that I will apply and may work correctly. So, first of all, here is the challenges:
You must never trust in clients running in client side. There's a lot of concerns about, your applications can be decomplied, modified, the users devices can be with a malware or connection may suffer with a man in the middle attacking (MITM)...
An API Server, even using OAuth2, will be able to only identify WHO is accessing the resources, but not WHAT is accessing. Therefore, any sensitive information will be dangerous, anything can steal it and uses it.
Resource Owner Password Credentials makes part of OAuth2 protocol for authorize resource owner to access your resources. So, it doesn't make part of authentication process and you will your ruin if you treat it like that;
By using ROPC grant type there is no way to know if resource owner is really making that request, what make "easy" a phishing attack. Reminds about "you know WHO and not WHAT". For last, this kind of grant makes easy for whatever thing assumes the user identity;
This grant type also goes against OAuth2 propourse, since OAuth seeks to avoid the password use to access resources. That why many people say to don't use it;
For reinforce, it's important to highlight ROPC is not authenticating user, but it just authorizing him to access the resource server.
And yes, ROPC allows for refresh tokens, but there are two issues: first, client needs resupply credentials each time needed to get a new token; second, if using a long-term access code, then things get more dangerous.
To prevent a malicious thing from arbitrarily using user credentials there are access tokens. They replace passwords and needed to be refreshed in short amount of time. That's why they are so much better than HTTP Basic Authentication.
That's why is recommended to use in modern apps the Authentication Code with PKCE, it provides all features and benefits of using OAuth2 protocol. But, here cames a long discussion and, even, problem for developer community:
To get an Authentication Code some user needs to make your login in a browser, grant access, redirect back to client and, soon, client will receive a code to exchange for an access token.
This scenario works good and NEEDS to be used for third-party apps. But, what if it is a first-party app? When you own the database with user data and you own the "trusted" app, redirect user doesn't make any sense. Right?
At this moment, my question is: how can I use the AuthCode (PKCE) flow without redirect user? And, again, it's important to highlight that talking about OAuth2 protocol is always the same that "to grant client to access resource server" (authorization, not authentication).
So the real question is: why Authorization Code needs a redirection at all? Then, I came with the following answer:
This flow requires to know client credentials and user consensus to turn back an authorization code.
That's why I was wrong in my edits. There's no change needed in OAuth2 protocol (sorry me for think different). For this reason, what OAuth2 needs is a authorization mediator, above your layer. Thus, the authorization code not will turn back to client, but to authorization mediator that, finally, will return it to client. Makes sense?
How it gonna work? Well, will be need 4 different "cores":
Authentication Server: will be responsible to authenticate user credentials and client identity. The main objective is to prove "WHO is the user and WHAT is connecting to get authentication";
Authorization Mediator (one layer above OAuth2): will validate client unique identity to ensure client/user is "know" and can get an access token;
Authorization Server: makes part of OAuth2 implementation, nothing change. Will authorize a client to get your authorization code, access tokens an refresh tokens;
Resource Server: will allow access resources through an access token.
And, then, security techniques we may consider:
API Key: each application (client) will have your own API Key with permissions scopes associated with those keys. By using it, you can gather basic statistics about API usage. Most API services use statistics to enforce rate limits per application to provide different tiers of service or reject suspiciously high frequency calling patterns;
Mutual SSL Authentication: by using this technique client and server exchange and verify each other's public keys. Once the keys are verified, the client and server negotiate a shared secret, a message authentication code (MAC) and encryption algorithms;
HMAC: API Key will be separeted into an ID and a shared secret. Then, as before, the ID is passed with each HTTP request, but the shared secret is used to sign, validates and/or encrypt the information in transit. The client and server will exchange the shared secret with algorithm such as HMAC SHA-256;
Protecting code application: using code obfuscators will make harder to locate and extract sensitive data from app, such as secret shared, api keys, public keys...
Handle user credentials: providing a simple method to user login and prove your identity. After insert valid credentials, server can return a user token (JWT) and emulates a user session with this.
Let's look at flow:
Part one: autheticating user and client;
User will type your credentials and be asked to prove your identity using your e-mail or mobile number, after Client will send user credentials (such as { email, mobile_number, hash ( password ), verification_method }) to Authentication Server route /login;
Authentication Server will validate user credentials and send a one-time password to user confirm your identity (for e-mail or mobile number as choose by user);
Then, user will insert the OTP received and client will send back to Authentication Server route /login-otp including the verification method (such as { otp, verification_method });
At the end, Authentication Server will return a { hash ( shared_secret ) } to be used soon.
Part two: authorizing API access;
When receive shared_secret Client will stores securely at mobile app, then it will ask for a authorization code using PKCE calling /auth with { response_type, client_id, scope, state, code_challenge, code_challenge_method }, Authorization Server will validate credentials and return an authorization code with no redirects;
Later, Client will exchange received code to an access token accessing /token, but it will need to send some extra data: { payload: { grant_type, code, client_id, code_verifier }, timestamp, hash ( some_user_data + timestamp + shared_secret ) };
Authorization Mediator will receive this request and validate trying to generate the same hash generated by user. And redirect all data to Authorization Server that will validate client_id, code and code_verifier responding with an access token;
This new access_token will return to Authorization Mediator and, after, to client granting access to API resources.
Part three: accessing resource server;
Client will each time needs send a call to API /api containing the Authorization header and some extradata with { timestamp, hash ( some_user_data + timestamp + shared_secret ) };
Authorization Mediator will validates the shared_secret hashes, call Resource Server validating access_token and return data.
Part four: refreshing access token;
After access token expires, Client will send a call to /refresh-token containing the Authorization header and some extradata with { payload: { grant_type, refresh_token, client_id, scope }, timestamp, hash ( some_user_data + timestamp + shared_secret ) };
Authorization Mediator will validates the shared_secret hashes, call Authorization Server and return a new fresh token access.
A visual image for this flow:
I don't think it is a perfect strategy, but it replaces Resource Owner Password Credentials to Authentication Code with PKCE and gives some extra security techniques. It's way better then a single and simple authentication method, preserves the OAuth2 protocol and mantaein a lit bit more hard to compromise user data.
Some references and support:
How do popular apps authenticate user requests from their mobile app to their server?
Why does your mobile app need an API key?
Mobile API Security Techniques
Secure Yet Simple Authentication System for Mobile Applications: Shared Secret Based Hash Authentication
Auth Code flow with PKCE is recommended - along with logging on via the system browser. Also the AppAuth pattern is recommended.
https://curity.io/resources/develop/sso/sso-for-mobile-apps-with-openid-connect/
It is tricky and time consuming to implement though - so you need to think about it - sometimes using a cheaper option is good enough. Depends on the sensitivity of data being exposed.
If it helps here are some notes for an Android demo app of mine, which also focuses on usability - and links to a code sample you can run:
https://authguidance.com/2019/09/13/android-code-sample-overview/
First of all, do not invent a OAuth grant simply because you need to adopt it in your application. It will make tings complex to maintain.
In your scenario you need to provide social login (ex:- Login via Google, facebook). This of course a desired functionality one must support. But it doesn't limit you from obtaining end user credentials through a custom registration process. There are many reasons for this, for example not everyone use social media or a Google account. And sometims people prefer to register than sharing user identifier of some other service (yes, this is the opposite end of social login).
So go ahead, provide social login. Store user identifiers when first login through external identity server (ex:- Google). But also, have a good old registration step with password and an email.
My use case is something like this.
I'm developing a rest api and single page web application.
But I don't want to store my user credentials (email, password) with me.
I want to store it in more secure place. From that place I need to verify credentials and issue tokens, as well as first time user register with the system that user's email ID should be verified and also If an user forgot his password there should be a way to reset it as well.
Finally in my node.js back-end I need to protect my routes from unauthorized accesses.
Do I can achieve all this things using a authentication service provider.
I go through the firebase docs and found It is little bit harder to implement my what I want using the firebase.
Is authentication services capable to provide reset password and email verification and store user credentials.
Or it just a token generator only?
If you are asking if Firebase Auth provides the ability to generate tokens for verified email/pass credentials it securely stores with email verification and password reset, the answer is yes. Learn more from their official documentation: https://firebase.google.com/docs/auth/web/password-auth
They also provide the ability to issue session cookies better suited for a Node.js server side managed sessions: https://firebase.google.com/docs/auth/admin/manage-cookies
You don't need to store the credentials. Firebase Auth will store them for you using industry best practices.
I am new to Hashicorp Vault. I was hoping to secure my client id and secret for an OAuth2 Password Flow using Hashicorp Vault. Each time my backend REST API is called, it requires the client id and secret, as well as the user credentials of username and password. How would I do this in a secure way and only let my app pass this without disclosing this in my javascript client?
Thanks.
John
If I read your question correctly, you have a Javascript application that calls your own (REST) backend service. That call is secured using a client id, client secret, username ánd password. That raises a couple of issues:
client id and secret, as well as the user credentials of username and password
That seems like the wrong approach to take: an OAuth-secured resource (your REST backend service) should not require a username and password. Logging in the user is done in the authorization server.
Try starting with reading the OAuth2-spec (RFC 6749) or the DigitalOcean tutorial for a comprehensive overview.
How would I do this in a secure way and only let my app pass this without disclosing this in my javascript client?
You can't: client secrets cannot be protected in a client-controlled application because an evil client can reverse-engineer your application (or read your javascript). What you have is called a "public client", i.e. a client that cannot keep its secret confidential. In this case, you don't use client secrets. Try starting with this question or the introduction to oauth2 client types.
I am dealing with Google api(oauth2.0) and DropBox api (oauth1.0),
just wondering how can I protect the redirect_uri that can only call by google server
and
oauth_callback by Dropbox server only.
Do I check their ip ? Since the url is always public, if there are no protection for it, there could be someone find out the uri and do the attack without anything notice from the server.
are there any guidelines that I have missed?
[edited]
I was wrong that redirect_uri and oauth_callback are actually called by the clients, not the auth server. So I should check the end user ip to make sure they are the same one requesting the token.
I don't think you're preventing any attack by checking IPs. Normal users won't disclose their tokens, so you shouldn't get bogus secrets for tokens that you obtained. If an attacker controls your user's computer and obtains the token, they can also use the computer to make requests to your redirect_uri, so the IP check will go through.
On the other hand, you should make sure that you use https for the OAuth flow, so tokens are protected from network sniffing.
I'm trying to use two legged oauth to allow a mobile client to log into an api I've created, however I can't quite grok the proper workflow for this and all the tutorials seem to say something different.
From what I've read in the two legged version the oauth consumer key and consumer secret are specifically assigned to a user, and the tokens aren't used. So when a user logs in they (or their device) would have to present their consumer key and secret and we can use that to verify their identity. But then what? Does the client device receive some token that they use to access the API, or do they send the consumer information with every request?
And the user can only be expected to remember a username and password, how do we get from username and password on the client device to a consumer key and secret to send to the server?
You shouldn't have a consumer key/secret pair for each client device. The OAuth notion of "consumer" is a particular site or developer using the API to authenticate to you. Who is creating the username/password pairs? Are these specifically your user accounts, or are you looking for users to be able to log into you with Yahoo, Google, etc. accounts?
At any rate, I would expect the users to have a username and password, not a consumer key and consumer secret.
2-legged OAuth removes a separate authN/authZ server that talks directly to the client that is otherwise present in 3-legged OAuth. It certainly does involve (access) tokens. The client device would receive a token and could use that until it expires.
The advantage of this setup is that you do not need to worry about the security of the client_id/secret on every API call. Sending client_id/secret on every call is basic authentication, and it is not recommended. Instead, by using OAuth, you only need to worry about the security of client_id/secret on the API call used to get the token (e.g., once for the life of each token). And if a token is compromised, it has a TTL, whereas client_id/secret do not.
The client_id/secret are not known to the end-user who provides their own user credentials. The client app is expected to handle the negotiation of client_id/secret for token.