I'm using directus to grant users access to ressources required by an SPA written in Angular. To authenticate users I created an auth service and interceptor to handle sessions and attach the "Authorization" header. Those services work fine and login as intended. But here comes the problem:
Directus session times are configured with default values (15 min validity for access_token, 7d for refresh_token) but as soon as the access_token expires I cannot retrieve a new one using the refresh token. This bugs me, because the goal is to keep users logged in for the next 7d (refresh_token lifespan) or until logout if they check this option.
My attempts at achieving this:
Since i'm using graphQL, i tried the "auth_refresh" mutation from the authentication documentation. While the access token is still valid, refreshing works fine. After the access token expired there is no way to retrieve a new one via a valid refresh token.
Alternatively I tried to achieve a refresh via the POST request specified by the docs (to double check if it was some sort of config error with graphql) but I encounter exactly the same problems as with graphQL. Directus returns either "401 unauthorized : Token expired."
if i extend the lifespan of the access token for longer than the server defined lifetime,
Response: Sending a token with prolonged life
or "401 unauthorized : Invalid user credentials." if I request a new token without an
"Authorization" header.
Response: Sending no access token
The refresh token is correctly loaded and sent to the server as specified in the docs in both
cases.
Now my questions are:
Am I missing something? I haven't found any further specification in the docs and the Auth0 protocol specifies that a new access token should be retrievable with a valid refresh token.
If this feature is not intended: How could I achieve a "keep me signed in" option with directus? I would like to keep user rights management in one place and do not really want to handle user auth redundantly for my current use case.
2b. Why is the lifespan of the refresh token so much longer than the lifespan of the access token if this isn't intended?
One of my thoughts is, that it has to do with access rights of the "public" role on the "directus_sessions" table. But I can't think of a way to grant only read rights for owned/received tokens, since there are no payload variables available inside the filters. Could this be the cause? Would there be a way to achieve this?
Thx&Greetz
I'm building an installed application that will have features requiring the Google Drive REST API using Qt and C++. I understand Qt is now releasing new libraries to support OAuth flows but let's assume I'm a student and learning to use OAuth at this layer is a requirement for this project.
In my application, I have a working OAuth flow for installed applications that ends with an Access Token and Refresh Token being stored using QSettings (I'm open to input on whether this is a disastrously bad idea too). The application requires no authentication/login for its own sake/data, but it does need authentication to Google for calling API's using an Access Token. This application has no associated web backend being hosted; its simple and should be deployable completely locally (I've written and included a simple TCP server that will receive the authorization redirect_uri and will run and close when called from within the application).
As such, I'm curious about the best way to make sure that, when a user opens my application and wants to use the Google Drive features, they are appropriately authenticated on Google's end. Say, if I maintain an access token in the registry, and this access token is granted per-user/per-application basis (right?), then how can I make sure only the user the token belongs to is able to make calls to the API with it?
Here's my understanding and approach; feel free to correct me or educate me if I've got the wrong interpretation.
If an Access Token is found, perform the following:
Open a browser page to a Google login domain and have the user authenticate there (this could prohibit a user from being able to use a cached login session that would have access to a token they otherwise shouldn't have access to)
If user has correctly authenticated with a Google account, return control to the application and make a test call to an API using the stored token.
If the call fails (responds with an invalid_credentials) I should be able to be sure its because the access token has expired and the application will go through the flow to renew an Access Token from a Refresh Token.
If no Access Token is initially found:
Start a normal OAuth installed application flow
Get the tokens and store them so that when the user opens the application next time the former procedure is used
My issue then is the first two steps if the Access Token is found. Nominally this could be done by the typical OAuth flow but it appears that when using a localhost as the redirect uri, Google will always prompt for consent, regardless of settings for prompt and access_type authorization query parameters.
What can be done to accomplish these first two steps in a way that my application can control (i.e. not a solution that relies on a backend server being hosted somewhere)?
If this question is too open-ended for SO requirements I can make some more restrictions/assumptions to limit the problem domain but I'd rather not do that yet in case I unknowingly rope off a good viable solution.
Thanks for reading! Sorry if its a verbose; I wanted to ensure my problem domain was fully fleshed out!
If you are using an installed application, I wouldn't recommend using or storing refresh tokens. Storing refresh tokens on the client side means that if an intruder gains access to the client's application, they have infinite access to the user's application without ever having to enter the user's credentials. If you do insist on having a refresh token, ensure you follow the Google's installed app flow, with the code_verifier parameter included in your requests.
If the access token is found, you should try to verify it, and if verified then use it at the google api, otherwise force the user to login again (or refresh it if you chose to still use refresh tokens).
If no access token is found, your flow sounds fine.
A few notes on loggin in with Google:
Google will only return a refresh token if you specify access_type=offline in your auth request.
Google will only return a refresh token on the user's first authorization request, unless you always specify prompt=consent in your query params.
In my experience, when leaving out the prompt query param, the user is not prompted for their consent again. If they are logged in to google, you will get a new access token, but no refresh token, unless you have prompt=consent.
I think the idea is you use prompt=consent if you have no record of the user ever using your application. Otherwise if they have used it before, you may prefer to use prompt=select_account to allow the user to select which account he wants to use in case he has more then one, or you can just use prompt=none.
This is just my understanding of it all.
My approach I ended up using was just to deploy with an SQLite db that will be stored in the AppData roaming directory. The db schema includes a field for the user's Name (from the OpenID IDToken field if it exists), the user's picture URL (again from IDToken if it exists), the refresh and access token strings (will be stored as encrypted strings when I get around to it), the user's UID/sub string, and a field for a user name and password.
These latter two fields are authentication fields for within my own application, which, again, I wanted to avoid but it seems impossible to do so. So the user will be prompted to enter a username and password into a form, and these credentials will be checked against the existing SQLite db file mentioned previously.
If they exist and are correct, the user gets logged in and will have access to their respective access and refresh token.
If the user has forgotten their password, they'll be asked for reconsent (going through the installed app flow again), and whatever password they provided during initial login will be used as the reset password. It is considered, for my purposes, that logging into Google for the installed app flow is proof enough that the user account belongs to them and they should have authorization to reset the password.
If the user is a new user and doesn't have a record in the local SQLite db file, then they can also click a button to "Create New Account" - which effectively goes through the authorization flow as well but this time a whole new record is posted to the SQLite db with the appropriate fields filled.
There's still more optimization that could be done but at least I am getting closer to the level of security and control of access to Google user accounts that I want.
I'm not marking this as an answer because I feel like this solution is still not desired and that there should be an easier way. So if someone has evidence or experience of providing an equivalent level of authentication control without needing to maintain a local user account database then I would be more than happy to mark such a method as the solution!
Thanks again!
I've successfully installed and tested the OAuth 2 workflow with Symfony 2 and FOSOAuthServerBundle.
I can request a code, and get a pair of access/refresh token successfully from a "login with" button on a third party test page i've set up and retrieve a user from my API through a custom API call. Pretty cool here.
However, each time I test the flow from the beginning, my oAuth server keeps on redirecting the user on the authorization page. Here are my questions.
Once a user has authorized and app, shouldn't the authorization part of the process be skipped with OAuth 2.0 ?
Is that fixed on the server side or should i change the way i request the credentials on client site ?
And finally... could i debug and fix this ?
If anyone struggles again, the solution is here :
https://github.com/FriendsOfSymfony/FOSOAuthServerBundle/blob/master/Resources/doc/the_oauth_event_class.md
This feature is not default in FOSOAuthServer bundle. You have to create and EventListener and check client or user status, save the user's choice, and [quote:even bypass the authorization process].
I searched through similar questions but none seem to answer this particular concern:
Assuming I am building a simple oauth user login system, when the oauth flow is done, should I pass the obtained access token through GET query, such as /login/provider?access_token=xxxx (assuming HTTPS), or should I store it in the session and delete it when my REST api call are done (assuming HTTPS + Signed cookies)
From what I can see: session approach doesn't expose token to even the user himself, and prevent logfile from recording ?access_token=xxxx, appear to be the safer bet.
Are there reasons for choosing GET query? Does my argument apply to OAuth 1 token and secret as well?
I am working on a WordPress plugin that requires the use of OAuth 2. So far I've included the necessary OAuth files in an include folder in my plugin, put in the ClientId, ClientSecret I've obtained, and was able to authorize the app. After I authorized the app I received a callback code that I am supposed to use to get the token I assume. I put that code in my plugin, refreshed the page, and saw the information that is supposed to be outputted. When I go to a different page, it tells me that I am not authorized to view the information because the access token may be missing. So essentially it feels like the code that is sent to retrieve the token is only able to be used once. Do I need to store the token somehow? Any suggestions would be greatly appreciated.
In OAuth2 three-pass procedure the auth_code must be used only to obtain the access token
invoking a specific URL of the authentication server. The access token obtained can be used to access the pages later, as long as it is valid.