I have successfully configured qn Azure AD App Registration, allowing a client_credentials based OAuth 2.0 flow to work. This allows a third party application to access Microsoft Graph API. The app has "Calendar.Read" permission. Meaning the app can pretty much read any of the calendars (including CEO's).
I now have a conversation with security. What is out there in the Microsoft world, that I can use to lock down usage of API access via this Application Permission? Is there ability to do things like:
Restrict IP ranges the App can be accessed from?
Restrict users that can access the App? (However in Client_credentials, there is no user context)
Log traffic / activity happening via the App?
The only thing i can think of now is to say the Redirect URL configuration on the app means, no other application can get an access token using the Client_credentials, even if the application id & passkey get compromised
Any advice on further security controls that can be put in place?
Restricting access: You would need to do this in your application. The Client Credential flow doesn't allow for restricting what users as you point out. However there is nothing stopping you from adding user authentication to your application, possibly using a delegated graph auth flow to determine who they are.
IP Ranges: This is not possible currently.
Logging Traffic: This is not possible on the graph side currently, however you could/should log traffic on your applications side.
Redirect urls will not help you because they are not used int eh client credential flow.
In general application only auth (client credential flow) + a broad authorization scope is very powerful, but must be managed correctly. You don't inadvertently want to build a totally new users/permissions model over the top of the graph :)
There are very few options available currently available to offer these controls at token issuance (in Azure AD) or at API access (in Microsoft Graph). However, you can achieve similar results by carefully managing access to the app's credentials. Here are a couple steps you can take (not exhaustive):
App credentials: keep them secret, keep them safe
Use Key Vault. You can configure many of the restrictions you mention for access to data in Key Vault, including IP ranges and which users access. Key Vault also offers auditing of access to secrets. Don't forget to also be careful about which users have management access to the Key Vault (e.g. other users with access to the same Azure subscription).
Use certificates (public/private key pair), rather than client secrets (passwords), to authenticate the app. People tend to manage certificates much more carefully than they manage shared passwords, and developers are much less likely to hard-code the secret into scripts/code.
Be careful and deliberate about which users can manage the app's credentials
This is often overlooked. A user (or another app) who can access existing credentials, or add a new authorized credential to an app can act as the app and (mis)use all the permissions the app has been granted. This includes:
Users (and apps) in the "Company Administrator", "Application Administrator" and "Cloud Application Administrator" directory roles.
Users who are set as owners of the app registration (Application object) and enterprise app (ServicePrincipal object) for the app.
Users (or systems) who have access to the server or service the application resides on (which will have, or have access to, the credentials).
For all of these cases, ensure this is the smallest possible number of users, and they actually have a legitimate need. For users who do need access, wherever possible enforce just-in-time, time-limited access (not persistent access), such as with Azure AD Privileged Identity Management, for time-bound, just-in-time access for Azure AD directory roles and Azure resources.
Related
For my current ASP.NET Core MVC application I authenticate directly with a web app registered in Azure AD Portal. This provides me with an access token so on the backend of my web application I can use MS Graph with my users specific account (ie add files to their onedrive , email, etc). However, my organization also has Okta which a lot of applications authenticate against. So I was trying to determine to authenticate through Okta (which has a much cleaner sign in process IMO) as well as authenticate against Azure AD and get an access token. Through my research I found something in my web application registration in Azure AD called Workload Identity Federation. This led me to this useful video
https://www.youtube.com/watch?v=wZ0gCJYMUKI
and also microsofts info site:
https://learn.microsoft.com/en-us/azure/active-directory/develop/workload-identity-federation
This seems to answer what I want which is to use Okta but allow me to still use MS Graph for my users since it will authenticate against Azure AD (correct me if I am wrong and this is for something else). My issue is none of these resources really go into depth regarding how the access token is passed to my application so I can use MS Graph. My research this is called client credential flow since my application only has delegated permissions so it requires the users to log in and it basically allows my web app to act on their behalf when using MS Graph. So I am trying to understand and fill this void of information regarding how client credential flow fits into Workload Identity Federation and is this the solution to my problem.
As shown by many samples I have two AAD application registrations, one for my javascript-based front end, and one for my JSON-only web APIs.
If I fully trust my client AAD application, why does AAD require me to create a second AAD application for my web APIs?
For example, assuming I add specific roles to my client AAD application, if client signs in with AAD and gets an id token and access token containing my roles, it only needs to send the access token to my APIs. The API only needs to crack the JWT, validate the audience, issuer, tenant, roles permissions, and signature. In this world, no client secret is needed in the web APIs, a second AAD application registration not needed, and still no call to AAD from my APIs. Unfortunately, without two AAD applications, I cannot figure out a way to have AAD include roles into my access token.
If I didn't fully trust the issuer from mucking with claims, I can see why I would need two AAD applications and a client secret. But since I do trust my AAD application and the signature of the JWT, why the extra complexity? Or maybe there is a way to do this that I haven't found?
Thanks!
Responding to Marc here because just not enough characters in the comments field -- The sample you referenced is an excellent sample, specifically the JavaScript one calling the Web API. It is what I am doing right now in fact. However, the problem is that Web API in the sample is open to anybody who has authenticated on the tenant. I need to secure the Web API down to certain individuals in the tenant, and simply checking the client/app id is not sufficient as anybody who can create an AAD app can fake it.
So what I need to do is have roles added to the the access token so I know that my application authenticated the user, and that user has been granted the required roles. For example, here is a Microsoft sample. And even here a Microsoft video walking through the process.
If I don't have two AAD applications w/client secret, the roles claims is never provided in the access token. It is always provided in the id token, but not the access token.
I feel like I am missing something obvious here. If AAD would just put the roles I requested into the JWT when I authenticated against it, and I validated its signature, audience, issuer, and roles, I wouldn't need any of this extra complexity?
Ah, I think I understand where you are going: you would like to control which users can access an API, no matter what client app they are using to access the API with. That's a function of the API - you cannot control that through AAD. In AAD you can control which users can access which applications (UI) using either user access restrictions (enterprise tab) or role-based access. However, access to an API is controlled in AAD at the calling application level via scopes. APIs are never accessed directly by users but only by other apps so controlling access permissions at user level would cause admin havoc. So, you can control what permissions a user has in the app they are using and you can control what permissions that application (client) has in other applications (APIs, resource servers) it is using.
In other words: role is about user access to UI, scope is about one apps' access to another.
App secrets provide added security for getting tokens - they have no bearing on what permissions are included in the token.
Can you provide a link showing that two apps are needed? That should only be the case if the API you want to call is not provided by the web app which served the JS to the browser. None of the 'official' samples require you to register two apps (Graph API, used in some of these samples is a separate API and it is already registered). A problem with tokens passed from the browser is that they were acquired by a public client, not using any secrets apart from user creds. Therefore, they are easier to steal and re-use. Your own back-end app may want to use a secret to get its own token (extension grant) to call yet another API using a token that does not reside in a public client.
My organization is taking a look at the security of registered applications within Azure Active Directory (AAD) and have concerns around the ability of individuals to add client secrets and certificates for applications that are using the "application permissions" model. I'm working to help narrow the roles of individuals within the organization to restrict this, but this investigation begged the question of what a malicious insider could do if he or she could add a client secret to this application.
I've looked through the 30 Days of Microsoft Graph blog series, which is excellent, but wanted to clarify what else can be done to prevent an insider from gaining access to the permissions this application would allow.
Does the redirect URL itself protect against this kind of scenario, provided the organization retains control of all registered URLs (meaning, for example, that https://localhost isn't registered)? Based on this post under Step 3, I assume the answer is yes but wanted to make sure this is the case.
Is it technically correct to say that without the redirect URL being secured/owned by the organization, a malicious insider who could add client secrets could exploit the permissions granted by the application?
If you are able to add a client secret to an app that already has been granted application permissions to something, then this user can use the new secret to get tokens and access those resources as the app.
Redirect URL is not used with application permissions, only delegated permissions.
This is because there are no redirects in the client credentials grant flow, which is used when acquiring a token with app permissions.
It's just an HTTP request.
So you are correct in your assumption that being able to add a new secret to an app that already has permissions can be a security issue.
There are audit logs though, and I believe adding a secret/certificate is logged.
I created a "Web app / API" app in our organization's "xxx.onmicrosoft.com" Azure Active Directory. The app's "Multi-tenanted" property has been set to "Yes".
We configured OpenID Connect (we use https://github.com/mitreid-connect/) to use the following URLs:
https://login.microsoftonline.com/common/oauth2/authorize
https://login.microsoftonline.com/common/oauth2/token
Please note that we used "common" in the URLs and we didn't use "xxx.onmicrosoft.com" because we want people from outside "xxx.onmicrosoft.com" to be able to authenticate and access our app.
With those settings, the people from xxx.onmicrosoft.com can properly authenticate and access the app.
However, when I use my personal live.com account (with username xxx#gmail.com) to access the app, I get AADSTS50020 error. I am able to properly authenticate with my xxx#gmail.com account, but I do not get redirected to the Reply URL. I'm stuck on Microsoft's Web page with the following error msg:
AADSTS50020: User account 'xxx#gmail.com' from identity provider
'live.com' does not exist in tenant 'xxx.onmicrosoft.com' and cannot
access the application '391e7103-ZZZZ-zz87-xxxx-7xxxxxd5xxxx' in that
tenant. The account needs to be added as an external user in the
tenant first. Sign out and sign in again with a different Azure Active
Directory user account.
What configuration do I need to change if I want people from any identity provider to be able to access my app ?
Like it has been stated here, I expected that people from anywhere could access my app without requiring more configuration on my side.
I'm asking this question because I'm in the process of getting certified for AppSource and this currently blocks me from being able to do so.
AppSource only requires work accounts to sign-in. You are using an #gmail account - which is a personal account - and because you are using the Azure Active Directory v1 endpoint in addition to common (https://login.microsoftonline.com/common), it can't accept personal accounts to sign-in directly - only work accounts.
You have three options:
If sign-in personal accounts is not a requirement for your application, then you can continue using the v1 endpoint and use a work account to sign-in/test your application. This will make you ready for AppSource certification.
If you need/ want to allow personal accounts in your application in addition to work accounts, then you can consider using the v2 endpoint (https://login.microsoftonline.com/common/v2.0) for Azure Active Directory. The v2 endpoint allow both personal accounts and work accounts to sign-in with no effort.A note is the v2 endpoint has some limitations: if you can live with these limitations (for example, your application only needs to sign-in users and eventually make queries against Graph API), then in general it should be fine to use, but if you need extra features like protecting your own Web API with scopes, then this feature is not released at this point (as November 2017). Please read this document for an updated list of limitations of the v2 endpoint.
A third (but less recommended option for AppSource) is to keep using the v1 endpoint and make your application to be single tenant - which is to change the endpoint from https://login.microsoftonline.com/common to https://login.microsoftonline.com/{yourtenantid}, and then use B2B invitations API to invite every external users (including work and personal accounts) to be part of your Azure AD tenant/organization. More information about B2B here as well.
The option '3' above have some consequences for management as well for AppSource: by using this option, you are required to have one Azure Active Directory tenant (if you don't have a tenant already, you can get one using these instructions), and the users being invited will be guests accounts of this tenant - this mean that you need to invite every external user to your application/ tenant. A multi-tenant application allows any user from any organization to sign-in to your application with less management on your side. In general for SaaS applications, multi-tenant configuration is recommended.
For AppSource, also the option '3' leads to a less-immersive user experience (Partner led trial), where the end user won't be able to access your application's demo right away - mainly because that they have to wait for the invitation's email and accept it (user has to accept being guest of your tenant) so that they can access your application.
For more information about AppSource requirements and trial options - please see this article.
We have a bit of a dilemma that we are running into with a couple applications that are trying to read a given users email without user interaction to authorize. The key to this approach is that we want no user interaction, and want to load the client server application with the proper JSON credentials downloaded from the Google Developer Console.
I have this approach working for programs where we create a service account in the Developer Console, and then delegate domain wide authority to that account with the proper scope access. However what we are hoping is that we don't have to delegate domain wide authority, and just read the users email who created this developer console project. I have tried many different types of solutions for this, but always run into the same limitation that I have to grant domain wide access.
What I am wondering is if there is any way to gain access to a single users mailbox using a server to server type approach and not have to grant domain wide access?
I appreciate your help with this issue!
There is no supported authorization flow for what you want to do. You must either use a service account that has been delegated domain-wide authority, or you must use a 3LO flow that involves user consent.
It seems you're looking for OAuth for Server to Server Application. You will also be using a service account. But, granting of domain-wide authority for service accounts is an optional thing. You don't have to enable it if you don't want to.
To support server-to-server interactions, first create a service
account for your project in the Developers Console. If you want to
access user data for users in your Google Apps domain, then delegate
domain-wide access to the service account.
Then, your application prepares to make authorized API calls by using
the service account's credentials to request an access token from the
OAuth 2.0 auth server.
Finally, your application can use the access token to call Google
APIs.