I'm enjoying developing cross-browser web extensions, the main target being Chrome, so much that I started to think to develop one for my company. I find a chrome extension quite a cheap and efficient way to deploy internal apps. The main purpose is to host a couple of dynamic dashboards that fetch data from various APIs by using cross-domain ajax in background scripts. I finalized the app and I was also able to implement the authentication via chrome.identity and Azure AD.
However, I am struggling to find a safe way to customise the content.
I mean, when the extension is installed it requires to login to azure via the chrome.identity flow. Then I get a token that I use to query ms graph and get the user ID, name, email and basic info.
Until I get this information I want the browser action (popup) to be unavailable to the user as well as any other extension pages. After a successful login I would like to show the content on the pop up and to let the user access the pages, but here I want to customize the experience.
I know how to use the user id retrieved from the api call to customize the extension, but I think it is not safe because all the code is in the client.
If I code something like
if (user === logged) show something
it will be damn easy for a malicious user to look at the code and bypass it, or even to impersonate another user. And chrome extension cannot be obfuscated.
Any help?
Thanks
I am trying to authenticate a user inside a desktop application using the web api. I am not using a browser, I am using straight up GET and PUSH calls to the endpoints of the Spotify servers. Immediately I ran into some problems. It appears that upon the initial GET command to "accounts.spotify.com", the returned response includes HTML with a javascript function that runs and is responsible for dynamically generating HTML that you see on the initial login page. If you look at the Javascript function, it is clear that this is what is going on, however, you can also see this code is obfuscated and not meant to be used by us, the developers! (Link to Javascript code here for reference: Javascript function)
So my question is, while I can probably reverse engineer the code to get this working, would this be against the Spotify developer TOS?
Thanks!
Spotify's authentication happens through oauth, and a big part of user authentication as per the oauth rfc is where the user delegates permissions to your app to carry out API calls that affect their account, or return information about them. That's the web page you're seeing - it must be presented to your users so that they can delegate permissions so that Spotify can give your app an access token. It doesn't necessarily need to happen in a browser - it can happen in a web view inside your desktop application - but it does need to be loaded over https, and your application must not alter or reverse engineer the Spotify permissions delegations page.
As you correctly guessed, reverse engineering any Spotify APIs is against terms of service.
For more information on authorization on the Spotify platform, I'd recommend having a look at this guide.
Hope that helps! Please ping me if you have any more questions.
Hugh
Spotify Developer Support
I've asked a Google Dev Advocate for help as I'm struggling to implement Account Linking on my Google Actions app, he sent me a link to a documentation article I already had read and suggested I also consult Stackoverflow. Having already done the second suggestion too and having struggled to find the exact answers to my questions, I've deiced to link to the doc article here, add all my queries and send this back to the Dev Advocate in the hope to get more clarification, especially as a reminder that Documentation could be read by complete newbies on the topic and that nothing should be given for granted.
This is the article I am referring to https://developers.google.com/actions/identity/account-linking
My queries below:
What is the difference between implicit and authorization code flow. In the article "Authorization code" is chosen, why?
Although I have found on Stackoverflow where to get your Client ID and secret, don't you think it would be good to add a link in the article?
Authorization URL - this is something instead I haven't found a clear guide for. Some Stackoverflow tickets report 2 Google OAUth URLs can be used (For the Authorization URL, enter https://accounts.google.com/o/oauth2/v2/auth
For the Token URL, enter https://www.googleapis.com/oauth2/v4/token), but a recent change to google policy suggests
When implementing account linking using OAuth, you must own your OAuth endpoint
So I'm now extremely confused at what I should put in the Aiuthorization URL and Token URL - why isn't this documented in a more basic and clear way? I've also read it needs to be served over HTTPS, what if you're working on local and on a pet project which isn't commercial and you won't be able to pay for HTTPS?
What is Seamless Account Linking and why isn't this explained and documented?
If your app supports seamless account linking
Where should we whitelist this?
Whitelist the following redirect URI: https://oauth-redirect.googleusercontent.com/r/
What are your OAuth 2.0 client configuration details? Where can they be found?
In the expanded OAuth 2.0 form, fill out the fields with your OAuth 2.0 client configuration. When filling in scopes, ensure they are space delimited.
I don't see the Discovery tab on my Oneplus 3T Google App, where else can I find it?
Open the Google app and go to the Discover tab.
This is where I get stuck - as many other people on Stackoverflow I get "The account is not linked yet" error. Maybe resolving the issues above will resolve the Account Linking error?
Invoke your app. Since it's the first time invoking the app with your Google account, the Assistant notifies you that you must link your account.
In addition to those questions, I also have the following:
I would like to get access to the user calendar and user basic info so I've added profile, email and https://www.googleapis.com/auth/calendar could you confirm these are correct?
Thanks and please remember documentation should be for everyone!
Documentation is for all developers. However, keep in mind that some of the tasks might require you, as a developer, to learn more than you currently know. Coming to SO is one of the ways to do that, but there are many other avenues that supplement that.
Good original documentation does, however, help. Google's docs are currently just bad - they used to be terrible.
Update - Before we begin, let me answer a question you suggest, but don't actually ask.
Why do I need an OAuth server at all?
First of all - you don't.
Think of your service like a website and the Assistant as a browser. For lots of websites, they don't need to know who the user is in order to use the website. There are lots of things the website can do without a user account at all.
In some cases, it is useful to know that the user visiting your website has visited you before. Frequently, you'll use a cookie to do track users like this.
The Assistant has an equivalent to this, although it is slightly different. The Assistant sends an anonymous UserID with each message to you. This UserID is only for this user and for your Action - it isn't re-used for any other Action or any other user. So if you track it, you'll know when the user returns. Like cookies, users can reset or clear it, but for the most part, this is durable.
But sometimes, you might need a person to log in to an account on your website. This is what the OAuth server is meant to accomplish - give users a way to log into your Action. OAuth is a pretty standard way to let people log into services these days, although the intent is really to authorize a client to act on your behalf.
The latter is really what OAuth is doing in this case - your user is authorizing the Assistant to act on the user's behalf when talking to your Action.
(Update - There are now ways to avoid having to setup an OAuth server at all in some circumstances. See the update at the bottom of this answer.)
Now back to your questions
But... let's go over your questions.
What is the difference between implicit and authorization code flow.
These are two terms that are more carefully defined by the OAuth2 standard, but in short - both of them let a client (a remote server from yours - the Assistant in this case) to get a user to give certain rights on your server.
The Implicit flow is simpler, both in what you need to setup and what the two servers exchange, but assume that once you issue a token, it is indefinitely valid. This brings with it a slightly higher risk that someone can get this token and use it to impersonate the Assistant.
The Auth Code flow is more complex (although not a lot) and addresses the risks in several ways. One way is that some transactions are done server-to-server instead of including the client, and that those transactions include a shared secret. Another way is that the auth token has a limited lifetime, and therefore a limited window of exposure, but that there is a refresh token which can be used to get a new auth token.
In the article "Authorization code" is chosen, why?
Most likely because it is more secure for a minimal level of extra work. Most of the security issues it addresses, however, are most visible in more open environments such as browser and mobile - they're not as big a risk with the Assistant. However, for places that need to setup an auth server, going with the more secure route has benefits in other areas.
Most Google APIs use the Auth Code flow or variants of it. (Although most use it from the client side - not the server side. Which is what Account Linking for Actions requires.)
Although I have found on Stackoverflow where to get your Client ID and secret, don't you think it would be good to add a link in the article?
Well... except that SO answer is no longer valid. (And, apparently, was never intended to be valid.) As you noted in your next question, Google has clarified their policy that requires you own the OAuth endpoints you use for an Action. They have, furthermore, made technical changes that prevent you from using Google's endpoints. (And I've updated the answer to say so.)
While the "Configure cloud project" part is correct, and describes how you setup credentials to be used with the Calendar API, you cannot use Google's OAuth endpoints to do the auth for your own project.
So I'm now extremely confused at what I should put in the Authorization URL and Token URL - why isn't this documented in a more basic and clear way?
Because this is a point where they're making an assumption that isn't very clear in the documentation. It is suggested where they say "Step 1. Configure your server" that you have an OAuth server. If you have an OAuth server already, then you should know what your server's Authorization and Token URLs are.
If you don't, however, this does get further explained where they talk about determining what the endpoints will be for an OAuth service you're creating.
I've also read it needs to be served over HTTPS, what if you're working on local and on a pet project which isn't commercial and you won't be able to pay for HTTPS?
Yes, it has to be HTTPS. This is a requirement of OAuth, and good practice when you're sending tokens that can be used to do things authorized by a user. It sounds like you want to be able to issue API calls to a Google server, and if those tokens got out (or tokens that could be used to access the same resources), then your Google Account could be compromised.
You have a lot of options here for your local or pet project development. Just to list a few:
You can use Firebase Functions. For projects on a "pet" level, they're free. (And if your Action gets a little popular, Google Assistant will give you credits that should pay for a modest level of use.)
You can get SSL certificates for your server for free using Let's Encrypt.
Since your server has to have a public address, you can create a tunnel using ngrok, which also provides a public HTTPS address you can use. This probably isn't good once your project gets out of the "personal testing" stage, but is a good tool to start with.
There are other approaches, of course, but these are a few good tools that you can use depending on your needs.
What is Seamless Account Linking and why isn't this explained and documented?
It is. Except in the documentation they confuse things by also calling it "Streamlined Identity Flow".
On the Account Linking Overview page it says "For more information, see Streamlined Identity Flows about how to configure your OAuth server to support the seamless identity experiences on the Google Assistant."
This takes you to a page talking about how this flow builds on top of the other two identity flows and has some additional requirements, but should make the user's experience better.
However... don't worry so much about this. If you're just doing this for fun, the normal identity flows aren't that much of a burden. If you're doing this for a commercial product - get the normal flows working first.
Where should we whitelist this?
Whitelist the following redirect URI: https://oauth-redirect.googleusercontent.com/r/
This is one of the underlying concepts of OAuth - as part of the communication between the client server and your server, it will say to redirect to a particular URL when you're done authenticating the user and getting their permission to issue a token.
The OAuth spec requires you to compare that redirect URL to a URL that has already been setup for that client. It does not specify how you set that up. So Google is saying "When you setup the OAuth server for our client - here is the URL that we will ask you to redirect to."
Google can't answer where to whitelist this except "in your OAuth server". Most OAuth servers have a way to configure multiple clients, and this is one of the values you'll set for that client. (The ClientID and ClientSecret are other values, but Google lets you determine these values and tell it as part of the configuration for Account Linking in the Action Console. Which is your next question.)
What are your OAuth 2.0 client configuration details? Where can they be found?
Again, this depends on your OAuth server and your requirements for what you want to prompt the user when they try to login to your server. The ClientID and ClientSecret are two such parameters. The OAuth scopes that the Assistant should request access to are other parameters. But these are up to you - because it is your server they are trying to get access to.
I don't see the Discovery tab on my Oneplus 3T Google App, where else can I find it?
That documentation looks incorrect. I think that should say that you should open the Google Home app on your mobile device.
It is also possible that it does mean the Google app, in which case your phone may not support the Google Assistant as part of the Google app. You can download the Google Assistant separately, if necessary.
However - use the simulator to test initially. Although it requires a few manual steps, they are easy to follow and help you trace things.
This is where I get stuck - as many other people on Stackoverflow I get "The account is not linked yet" error. Maybe resolving the issues above will resolve the Account Linking error?
Well, your account isn't linked yet. {:
It sounds like you haven't set an auth server for your Action. Until you get an auth server working, the rest isn't going to work.
I would like to get access to the user calendar and user basic info so I've added profile, email and https://www.googleapis.com/auth/calendar could you confirm these are correct?
First of all, keep in mind that this whole process is to link the user's Assistant account to their account on your service. You may have information in their account (on your service) that you use to do things - such as access Google resources or access other things that you know about them.
This is not directly a way that you gain access to the Google account that they're using to talk to the Assistant.
In order to get a user's permission to access their resources on Google's servers, you'll need to get them to authorize your server permission to access that. That is done using OAuth, again, but this time you're the client. User's will need to go to your server, you'll redirect them to Google's server to authorize you, and they'll be redirected back to your server with codes that you will need to store. This is all done outside of the Assistant and it's Account Linking system.
That said, for what you want, profile and email are fairly normal scopes to request. The Calendar API Documentation confirms that the https://www.googleapis.com/auth/calendar scope is what you need to access that API. (Keep in mind that this URL is not one that you'd use in a browser or that you'd go to to access anything - it is a uniquely identifying name only.)
Update to reflect API Changes. Since this answer was originally written, Google has introduced Google Sign In for Assistant, which lets you avoid having to setup your own OAuth server when you are willing to tie operations to the same Google account they use on the Assistant. If the user permits, you can get simple user profile information this way, and you can then leverage this to get access to other APIs (again, with the user's permission). See this SO answer that discusses how to use this to access Google's other APIs.
Thanks and please remember documentation should be for everyone!
From my conversation with Google's Assistant team, they are looking to make documentation easier, and hopefully they will take many of your suggestions to heart. I hope these clarifications have helped you (and anyone else who gets here with similar problems.)
We have a mobile web based app at my company. Due to the nature of the application we do not want the browser on the users phone to prompt the user to save the passwords on the form a.k.a the autocomplete feature.
We managed to do that for IE and Firefox by setting the autocomplete tag to "off" but that doesnt seem to work for Opera mini (and i am guessing opera in general). I know user's can set it to off in their settings but for security reason we rather have it disabled?
Is there a workaround for this through code? the app is Java app using faces components based on an Jboss/Apache architecture.
In general, Opera lets the user configure whether it should respect autocomplete=off. On principle, users should be able to configure the password storage feature, and web sites should not be able to affect the configuration at a whim.
However, I can certainly see that for specific scenarios, like "send one-time passwords by SMS to the device Opera remembered the regular password on", this sucks. If you have stored password for a high-value site + use SMS one-time passwords as an "out of band" authentication, a lost phone becomes a major risk. The root of this problem is the assumption that an SMS constitutes "three-factor" authentication - if the stored login and the SMS is on the same device it's no longer "three-factor"..
It is tricky to try to leave users in control, while yielding to the web site when it's a really good idea to do so. Sadly, I think this is an unsolved question for now.
If you have a good use case for disabling password storage and are working on an important site, perhaps Opera Mini server admins could be persuaded to disable password manager on a site-specific basis? I don't know, but if you report it as a bug to Opera it would at least give the internal discussions some more momentum. Feel free to contact me with a reference to the bug report because I'm in a position where I could keep an eye on it ;-)
GMail can used as a SMTP server. I've written the code that does it. But as we all know GMail may occasionally authenticate using captcha (image verification as they call it). The same thing may be the cause to reject SMTP authentication.
As I've seen google shows image verification when you try to log-in for the first time from some machine. All consecutive log-ins from the same machine (to the same account) use regular login. I'm a bit afraid this captcha may also come up again for some other reasons that I can't control.
So. Is it possible to still authenticate when special measures are needed? And how?
I should also mention that logging in from the machine via web browser also enables programmatic SMTP authentication.
Try:
http://www.google.com/accounts/DisplayUnlockCaptcha
Or for Google Apps for your domain:
https://www.google.com/a/yourdomain.com/UnlockCaptcha
I can bet it's an IP based solution, so if your app if deployed somewhere else, It doesn't help.
Google doesn't seem to be particularly clear about what prompts them to block a user's account until he has successfully entered a captcha phrase. However, it's likely that this is a mechanism which is triggered when Google sees what it considers to be unusual or suspicious activity associated with your account. As a result, I would expect it to be unlikely that they will supply an option to opt out of a mechanism which is protecting both them and you, although others are welcome to find evidence to the contrary.
Various discussions of when and how this happens throw up some suggestions which you might like to try to see if they help, such as choosing a stronger password or simply changing your password. Good luck!
IMHO The right way to do this, without implementing the user consent workflow or "hacking your account's security settings", is obtaining a refresh token with the Google OAuth2.0 Playground for Offline Access. After that, you can authenticate using the OAuth 2.0 API.
I got it working after reading Nodemailer/Gmail - What exactly is a refresh token and how do I get one?