I've been asked by my employer to implement a log-in system for our web application using users' GitHub accounts. I've looked around online but I haven't been able to find a clear explanation of how to go about doing this with GitHub accounts (as opposed to with Facebook or Google accounts).
I just spent about a week's worth of effort figuring out how to do this, so I thought I'd write up an explanation to save future developers time.
The short(er) answer
You'll want to follow this guide in GitHub's docs ("Authorizing OAuth Apps"), with some additions (explained below) to allow it to work as a method of user authentication.
I implemented the "web application flow" for when our application will be deployed on our company's servers (where we can keep our company's GitHub app's "client secret" a secret), and the "device flow" for when our application will be deployed on our client's computers (because in that situation we won't be able to keep our "client secret" a secret).
GitHub's guide doesn't mention the steps below (because that guide is not intended specifically for implementing social login), but to get social login working I also did the following:
I created a users database table, with the idea being that each GitHub account used to log in would have its own corresponding row in this table.
Example users table schema:
id - INTEGER
email - VARCHAR
name - VARCHAR
github_id - VARCHAR
I created an oauth_tokens database table to store a copy of all of the GitHub access tokens that our back-end receives from GitHub.
This is needed to prevent other malicious websites from impersonating our users with valid GitHub access tokens generated by the user authorizing an access token for the malicious website.
Example oauth_tokens table schema:
id - INTEGER
user_id - INTEGER
access_token - VARCHAR
expires_at - DATETIME
refresh_token - VARCHAR
refresh_token_expires_at - DATETIME
device_code - VARCHAR <-- Used for the "device flow". I have the back-end send the
front-end the device code immediately upon starting the device flow, and I then
have the front-end poll the back-end with it until the back-end has received
the access token from GitHub, at which point the front-end discards the device
code and uses the access token as its authentication token.
I had the back-end send the front-end (the user) the GitHub access token for it to present with future requests as its authentication mechanism.
The front-end should store the token in localStorage if you want the user to remain logged in even after they close the browser tab they logged in with.
I added middleware on the back-end that--for each incoming request--looks up the provided access token in our database to see if it's expired, and if so, attempts to refresh it. If it succeeds in refreshing the token, it proceeds with the request as normal and includes the new access token in the response to the front-end in a custom response header the front-end is keeping an eye out for (I named it x-updated-access-token). If it fails to refresh the token, it aborts the request and sends a 401 response that the front-end takes as a signal to redirect the user to the login page.
Setting up your app to only allow unexpired access tokens to serve as a method of authentication is necessary to make it possible for the user to sign out of the application remotely from their settings page at GitHub.com.
I added front-end code to handle the saving / updating / removing of the GitHub access token, both to/from localStorage as well as to all requests to the back-end, as well as redirecting to a /login route if the front-end doesn't find an "access_token" localStorage variable set.
The code is further below if you want an in-depth explanation, but basically I used this article as a rough guide for how the front-end code should work for the "web application flow": OpenID Connect Client by Example - Codeburst.io
More information
To clarify some vocabulary: The goal here is to do user authentication via social login. Social login is a type of single-sign on.
The first thing you should understand is that--as of the time I'm writing this--GitHub has not set itself up to be a provider of social login in the way Facebook and Google have.
Facebook and Google both have developed special JavaScript libraries that you can use to implement social login without needing to write any(?) login-specific back-end code. GitHub has no such library, and from what I can tell it's not even possible for a third party to develop such a library because GitHub's API doesn't offer the functionality required to make such a library possible (specifically, they seem to support neither the "implicit flow" nor OpenID Connect).
The next thing you should understand is that--as of the time I'm writing this--GitHub's API does not seem to support the use of OpenID Connect to implement social login using GitHub accounts.
When I started doing research into how to implement social login I was confused by the fact that the most-recent online guides were saying that OpenID Connect was the current best-practice way to do it. And this is true, if the Identity Provider (e.g. GitHub) you're using supports it (i.e. their API can return OpenID Connect ID tokens). As far as I can tell, GitHub's API doesn't currently have the ability to return OpenID Connect ID tokens from the endpoints we'd need to request them from, although it does seem they support the use of OpenID Connect tokens elsewhere in their API.
Thus, the way web apps will generally want to implement social login with GitHub accounts is to use the OAuth 2.0 flow that most websites used before OpenID Connect, which most online resources call the "authorization code flow", but which GitHub's docs refer to as the "web application flow". It's just as secure but requires some more work/code than the other methods to implement properly. The takeaway is that implementing social login with GitHub is going to take more time than using an Identity Provider like Facebook or Google that have streamlined the process for developers.
If you (or your boss) still want to use GitHub for social login even after understanding it's going to take more time, it's worth spending some time to watch some explanations of how the OAuth 2.0 flow works, why OpenID Connect was developed (even though GitHub doesn't seem to support it), and become familiar with some key technical terms, as it'll make it easier to understand the GitHub guide.
OAuth 2.0
The best explanation of OAuth 2.0 that I found was this one by Okta: An Illustrated Guide to OAuth and OpenID Connect
The most important technical terms:
Identity Provider - This is GitHub, Facebook, Google, etc.
Client - This is your app; specifically, the back-end part of your app.
Authorization Code - "A short-lived temporary code the Client gives the [Identity Provider] in exchange for an Access Token."
Access Token: This is what lets your app ask GitHub for information about the user.
You may also find this graph helpful:
The slide title is "OIDC Authorization Code Flow" but the same flow is used for a non-OIDC OAuth 2.0 authorization code flow, with the only difference being that step 10 doesn't return an ID token, just the access token and refresh token.
The fact that step 11 is highlighted in green isn't significant; it's just the step the presenter wanted to highlight for this particular slide.
The graph shows the "Identity Provider" and "Resource Server" as separate entities, which might be confusing. In our case they're both GitHub's API; the "Identity Provider" is the part of GitHub's API that gets us an access token, and the "Resource Server" is the part of GitHub's API that we can send the access token to to take actions on behalf of the user (e.g. asking about their profile).
Source: Introduction to OAuth 2.0 and OpenID Connect (PowerPoint slides) - PragmaticWebSecurity.com
OpenID Connect (OIDC)
Again, GitHub doesn't seem to support this, but it's mentioned a lot online, so you may be curious to know what's going on here / what problem it solves / why GitHub doesn't support it.
The best explanation I've seen for why OpenID Connect was introduced and why it would be preferred over plain OAuth 2.0 for authentication is my own summary of a 2012 ThreadSafe blog post: Why use OpenID Connect instead of plain OAuth2?.
The short answer is that before OIDC existed, pure-frontend social login JavaScript libraries (like Facebook's) were using plain OAuth 2.0, but this method was open to an exploit where a malicious web app could have a user sign into their site (for example, using Facebook login) and then use the generated (Facebook) access token to impersonate that user on any other site that accepted that (Facebook) access token as a method of authentication. OIDC prevents that exploit.
This particular exploit is what people are referring to when they say "OAuth 2.0 is an authorization protocol, not an authentication protocol...OAuth says absolutely nothing about the user, nor does it say how the user proved their presence or even if they're still there.", which I saw mentioned over and over again while doing research on how to use OAuth 2.0 to implement social login, and which had me initially thinking that I needed to use OpenID Connect.
But GitHub doesn't have a pure-frontend social login JavaScript library, so it doesn't need to support OpenID Connect to address that exploit. You just need to make sure your app's back-end is keeping track of which GitHub access tokens it has generated rather than just trusting any valid GitHub access token it receives.
While doing research I came across HelloJS and wondered if I could use it to implement social login. From what I can tell, the answer is "not securely".
The first thing to understand is that when you use HelloJS, it is using the same authentication code flow I describe above, except HelloJS has its own back-end ("proxy") server set up to allow you to skip writing the back-end code normally needed to implement this flow, and the HelloJS front-end library allows you to skip writing all the front-end code normally needed.
The problem with using HelloJS for social login is the back-end server/proxy part: there seems to be no way to prevent the kind of attack that OpenID Connect was created to prevent: the end result of using HelloJS seems to be a GitHub access token, and there seems to be no way for your app's back-end to tell whether that access token was created by the user trying to log into your app or if it was created when the user was logging into some other malicious app (which is then using that access token to send requests to your app, impersonating the user).
If your app doesn't use a back-end then you could be fine, but most apps do rely on a back-end to store user-specific data that should only be accessible to that user.
You could get around this problem if you were able to query the proxy server to double-check which access tokens it had generated, but HelloJS doesn't seem to have a way to do this out-of-the-box, and if you decide to create your own proxy server so that you can do this, you seem to be ending up in a more-complicated situation than if you'd just avoided HelloJS from the beginning.
HelloJS instead seems to be intended for situations where your front-end just wants to query the GitHub API on behalf of the user to get information about their account, like their user details or their list of repositories, with no expectation that your back-end will be using the user's GitHub access token as a method for that user to access their private information on your back-end.
To implement the "web application flow" I used the following article as a reference, although it didn't perfectly map to what I needed to do with GitHub: OpenID Connect Client by Example - Codeburst.io
Keep in mind that this guide is for implementing the OpenID Connect authentication flow, which is similar-to-but-not-the-same-as the flow we need to use for GitHub.
The code here was especially helpful for getting my front-end code working properly.
GitHub does not allow for the use of a "nonce" as described in this guide, because that is a feature specific to (some implementations of?) OpenID Connect, and GitHub's API does not support the use of a nonce in the same way that Google's API does.
To implement the "device flow" I used the following article as inspiration: Using the OAuth 2.0 device flow to authenticate users in desktop apps
The key quote is this: "Basically, when you need to authenticate, the device will display a URL and a code (it could also display a QR code to avoid having to copy the URL), and start polling the identity provider to ask if authentication is complete. You navigate to the URL in the browser on your phone or computer, log in when prompted to, and enter the code. When you’re done, the next time the device polls the IdP, it will receive a token: the flow is complete."
Example code
The app I'm working on uses Vue + Quasar + TypeScript on the front-end, and Python + aiohttp on the back-end. Obviously you may not be able to use the code directly, but hopefully using it as a reference will give you enough of an idea of what the finished product should look like that you can more-quickly get your own code working.
Because of Stack Overflow's post length limits, I can't include the code in the body of this answer, so instead I'm linking the code in individual GitHub Gists.
App.vue
This is the 'parent component' which the entire front-end application is contained within. It has code that handles the situation during the "web application flow" where the user has been redirected by GitHub back to our application after authorizing our application. It takes the authorization code from the URL query parameters and sends it to our application's back-end, which in turn sends the authorization code to GitHub in exchange for the access token and refresh token.
axios.ts
This is most of the code from axios.ts. This is where I put the code that adds the GitHub access token to all requests to our app's back-end (if the front-end finds such a token in localStorage), as well as the code that looks at any responses from our app's back-end to see if the access token has been refreshed.
auth.py
This is the back-end file that contains all the routes used during the login process for both the "web application flow" and the "device flow". If the route URL contains "oauth" it's for the "web application flow", and if the route URL contains "device" it's for the "device flow"; I was just following GitHub's example there.
middleware.py
This is the back-end file that contains the middleware function that evaluates all incoming requests to see if the presented GitHub access token is one in our app's database, and hasn't yet expired. The code for refreshing the access token is in this file.
Login.vue
This is the front-end component that displays the "Login page". It has code for both the "web application flow" as well as the "device flow".
Summary of the two login flows as implemented in my application:
The web application flow
The user goes to http://mywebsite.com/
The front-end code checks whether there's an access_token localStorage variable (which would indicate the user has already logged in), and doesn't find one, so it redirects the user to the /login route.
See App.vue:mounted() and App.vue:watch:authenticated()
At the Login page/view, the user clicks the "Sign in with GitHub" button.
The front-end sets a random state localStorage variable, then redirects the user to GitHub's OAuth app authorization page with our app's client ID and the random state variable as URL query parameters.
See Login.vue:redirectUserToGitHubWebAppFlowLoginLink()
The user signs into GitHub (if they're not already signed in), authorizes our application, and is redirected back to http://mywebsite.com/ with an authentication code and the state variable as URL query parameters.
The app is looking for those URL query parameters every time it loads, and when it sees them, it makes sure the state variable matches what it stored in localStorage, and if so, it POSTs the authorization code to our back-end.
See App.vue:mounted() and App.vue:sendTheBackendTheAuthorizationCodeFromGitHub()
Our app's back-end receives the POSTed authorization code and then very quickly:
Note: the steps below are in auth.py:get_web_app_flow_access_token_and_refresh_token()
It sends the authorization code to GitHub in exchange for the access token and refresh token (as well as their expiration times).
It uses the access token to query GitHub's "/user" endpoint to get the user's GitHub id, email address, and name.
It looks in our database to see if we have a user with the retrieved GitHub id, and if not, creates one.
It creates a new "oauth_tokens" database record for the newly-retrieved access tokens and associates it with the user record.
Finally, it sends the access token to the front-end in the response to the front-end's request.
The front-end receives the response, sets an access_token variable in localStorage, and sets an authenticated Vue variable to true, which the app is constantly watching out for, and which triggers the front-end to redirect the user from the "login" view to the "app" view (i.e. the part of the app that requires the user to be authenticated).
See App.vue:sendTheBackendTheAuthorizationCodeFromGitHub() and App.vue:watch:authenticated()
The device flow
The user goes to http://mywebsite.com/
The front-end code checks whether there's an access_token localStorage variable (which would indicate the user has already logged in), and doesn't find one, so it redirects the user to the /login route.
See App.vue:mounted() and App.vue:watch:authenticated()
At the Login page/view, the user clicks the "Sign in with GitHub" button.
The front-end sends a request to our app's back-end asking for the user code that the user will enter while signed into their GitHub account.
See Login.vue:startTheDeviceLoginFlow()
The back-end receives this request and:
See auth.py:get_device_flow_user_code()
Sends a request to GitHub asking for a new user_code.
Creates an asynchronous task polling GitHub to see if the user has entered the user_code yet.
Sends the user a response with the user_code and device_code that it got from GitHub.
The front-end receives the response from our app's back-end and:
It stores the user_code and device_code in Vue variables.
See Login.vue:startTheDeviceLoginFlow()
The device_code is also saved to localStorage so that if the user closes the browser window that has the "log in" page open and then opens up a new one, they won't need to restart the login process.
It displays the user_code to the user.
See Login.vue in the template code block starting <div v-if="deviceFlowUserCode">
It shows a button that will open the GitHub URL where the user can enter the user_code (it will open the page in a new tab).
It shows a QR code that links to the same GitHub link, so that if the user is using the application on a computer and wants to enter the code on their phone, they can do that.
The app uses the received device_code to set a deviceFlowDeviceCode variable. A separate part of the code in the app is constantly checking to see if that variable has been set, and when it sees that it has, it begins polling the back-end to see if the back-end has received the access_token yet from GitHub.
See Login.vue:watch:deviceFlowDeviceCode() and Login.vue:repeatedlyPollTheBackEndForTheAccessTokenGivenTheDeviceCode()
The user either clicks the aforementioned button or scans the QR code with their phone, and enters the user code at https://github.com/login/device while logged into their GitHub account, either on the same device this application is running on or some other device (like their phone).
The back-end, while polling GitHub every few seconds as previously mentioned, receives the access_token and refresh_token, and as mentioned while describing the "web app flow", sends a request to GitHub's "/user" endpoint to get user data, then gets or creates a user db record, and then creates a new oauth_tokens db record.
See auth.py:_repeatedly_poll_github_to_check_if_the_user_has_entered_their_code()
The front-end, while polling our application's back-end every few seconds, finally receives a response from the back-end with the access_token, sets an access_token variable in localStorage, redirects the user to the "app" view (i.e. the part of the app that requires the user to be authenticated).
See Login.vue:repeatedlyPollTheBackEndForTheAccessTokenGivenTheDeviceCode()
I am encountering a strange situation with MobileFirst 7.1 where users are occasionally unable to authenticate/login. The only indication that something is awry is a message in the console.log
[AUDIT ] CWWKS1100A: Authentication did not succeed for user ID . An invalid user ID or password was specified.
My custom login module uses com.worklight.core.auth.ext.LdapLoginModule (so to clarify I have a login module which authenticates using LDAP). Like I say everything seems to work most of the time but occasionally users end up in a situation where they are unable to authenticate. I suspect that it is probably related to the session in some way, but that is only a guess based on my investigation.
I have added some logging to my 'secret' adapter which prints the session state to the console log, and obviously this appears in the logs just before the failed authentication message above, but it is empty ie. the session contains nothing.The user is obviously trying to access a secure adapter at this point, and because they are not authenticated they end up at the login page (form based authentication I should say also).
Anyway, I noticed that although there appears to be no session data, the jsessionid is there and has not changed i.e. it does not change even if I refresh the browser. This may not be an issue in itself of course, but interestingly if I remove this entry and refresh my browser I am able to login successfully.
I am pretty sure that my handler code calls the relevant success/failure methods in the correct places but of course there is nothing to stop the user refreshing their browser, which causes them to be re-directed to the login page (the app has been developed using AngularJS so is effectively a single-page navigation model).
The only reproducible test I have been able to come up with is when I login to the MobileFirst console and then try to login to our MF 'desktopbrowser' app. I have read that this situation causes a session-related conflict, but as I say the occasional issue I am seeing is not caused by this (though it may be related).
So the problem seems to have been more related to the flow of logic in our application after successfully logging in, than any inherent issue with the MF Platform.
For example when a user refreshes the browser they are effectively still logged in, but because the app (based on logic we have developed) takes the user to the login page on refresh, the user is effectively re-logging in to the same session. If this failed every time it would of course have been easier to pinpoint but it does not. The solution was to force logout on refresh (when the app initialises), thus cleaning up any session data. In future iterations it may of course be better to re-establish the application based on the authenticated session after refresh, but at present that was a step too far.
Another example of this was post login if the subsequent adapter calls failed (e.g. we authenticate and then retrieve profile data from a database), then we were also not logging the successfully authenticated user out.
We are trying to build OAuth2 Authorization with IdentityServer3.
So we downloaded the Bytes from nuget and connected it with our database.
The database was initialized with the default scopes and the sample clients from Thinktecture self.
Then we connected AD FS as IDP via OWIN and made an simple ExternalUserService.
So far everything worked fine and the permissions page of the IdSrv could be opened, showing the username and that no application has consent up to now.
Then we tried to connect Xamarin.Auth to that and got an error Cannot determine application to sign in to and in the logs an error Signin Id not present (after logon at the ADFS IDP).
To reduce complexity, we decided to go back to the InMemoryUserService and created one InMemoryUser. This worked for the permissions page (at least for a short period of time - time is over now), but it did not allow OAuth2 Authorization Code Flow, which ended up in showing the login page again and again and again. And there is no evidence of any error in the logs.
How can we debug, what is happening? Is there any way to see, why a user gets redirected to the login page again despite being logged in?
--
We reduced the complexity even further by creating a new empty MVC application, which just uses a simple InMemoryUserFactory.
Now it's getting a little bit confusing: one user was able to logon from his machine - other machines (same user - since we created only one) are not able to login and get prompted with the login over and over again.
If using IdentityServer3 and you use own external login methods you should really pay close attention to the API of the IdSrv3.
We tried to create a login resutl with just the subject - this is made for local login on the server. If this is switched off at the same time, you will end up having problems.
So if you use an own external login provider and switch off local login, make sure to call the right overload for the authenticate method (3 Parameters in our case).
I am trying to integrate my iPad app with Box. I am having an issue with the Box API where files in the account of one user are returned for some other user. Here are the steps to reproduce this issue:
Make the authorization calls and get the access token as mentioned in this guide. For login, I am opening the Box login page in Safari. I have the specified a custom url scheme for the redirect url, which opens up my app after the user logs in.
Once you get the access token, make a call to list the contents of the root folder. This succeeds.
Delete the app from the iPad and rebuild it.
Again go to the login process (as in step 1), but this time use a different Box account to login. You get a new access code and OAuth token this time.
If you make the call to list the files using the new token, you will get the response from the earlier account. Ideally it should return the files for the currently authorized user.
Does Box use just OAuth to return response or does it use cookies as well? Because after authentication and receiving the access token, I also see a cookie from Box (verified using [[NSHTTPCookieStorage sharedStorage] cookies]).
I have tried repeating the above process by deleting all Box cookies before starting the authentication flow. Also, I am not saving the OAuth token on disk and retrieving it. I am not saving/caching the response in any way.
One more thing that I have noticed is that there can be two Box users logged in at once in Safari. Also, if I make the authentication request, get the access token and again make the authentication request, it shows the login page again (instead of showing the allow/deny access page). Is this intentional?
I am using the Box v2 API and iOS 5/6
Upon further inspection, the problem seems to be with Box servers caching response. I did a quick test with curl using two different access tokens created from the iPad app. I made a call to fetch the user files for the root folder using both tokens. The results were correct, i.e. I got the correct files for each account.
When I did the same test on the iPad app, the files for one user were returned for the other user. If I maintained a considerable gap between the two logins, I got the correct files.
To permanently fix this, I am setting the Cache-Control header to no-cache for the request to fetch the user's files.
But it is strange that I have to do this. Box needs to check their cache validation logic IMHO.
I'm trying to implement a way to actually logout from my application that is using the Valence API. I can obviously clear the session on my end, but is there a way through the API to actually log out of the Desire2Learn site as well? I've looked through the docs and didn't see anything.
No, there is currently no route to explicitly log out, or log in. You can, however, use the Valence auth process to generate credentials for a new user. What you need to do in that case is use a browser to interact with the user that doesn't have an open session with the LMS: as long as the LMS thinks that the browser doing the user part of the authentication has an open session, it will pass back the user credentials for that user instead of asking the user to re-authenticate.
Typically an inactive session with the LMS expires after a short time and then the LMS will force the user to re-authenticate if your app initiates the auth process.