When using Azure Managed Identity in the App Service/Functions, the endpoint and header are available as environment variables.
When using Azure Managed Identity in Virtual Machine, the endpoint is published as the Azure Instance Metadata service (IMDS) (http://169.254.169.254/metadata/identity/oauth2/token), accessible only from within the VM.
On a VM, is the endpoint set as an environment variable (MSI_ENDPOINT)?
On a VM, is there an indication whether managed identity is enabled?
Ideally, my code can simply check for that variable. Otherwise, I need to check the variable and make a call to the IMDS.
The Azure Managed Identity is different between App Service/Functions and VM. As you know, the App Service/Functions use the Managed Identity through the environment variables MSI_ENDPOINT and MSI_SECRET as it shows here.
But for VM, it does not set the environment variables, you will get the access token from a server inside the Azure and it cannot access outside. Here are ways to get the access token for the VMS.
The indication for the Managed Identity of the VM, I think you can get the VM identity property and it will show you if it enables the Managed Identity and which type is used. For example, I use the Azure CLI and it shows like this:
Related
I have created a very simple API with FastAPI to access text files on the filesystem.
It runs in a docker container that has only access to a volume, not the host filesystem and is accessible via a private network in docker from other containers.
The whole app is simple, just 112 lines, and should just store some test files for other containers to access.
I would like to add a simple Token bases authentication to the API. I have found many examples, but they assume that we have users accessing the API. Hence they require to setup a user DB etc.
In my case there is only a machine to machine communication with other apps in docker containers.
Is there any way to have a simple Token based authentication added without the overhead of users? Maybe set a single Token for all access that can be read form the environment variable at the start of the container.
I tried Grant your VM access to a Resource Group in Azure Resource Manager
but it does not list any VM, althought I have plenty of VMs in my subscription:
No virtual machine managed identities found in this subscription
See screenshot below:
I suppose you have not enabled the system-assigned managed identity for your VM, navigate to your VM in the portal -> Identity -> System assigned -> turn the status on, see this link.
Note: Per my test, it seems to have some delay. If so, just use the filter with Azure AD user, group, or service principal, search your VM name, refer to the screenshot.
When we store the Service principal certificate/appKey in the VM (to access the keyvault), we could limit access to that file to just the user account running the program. Other users or accounts wouldn't have access to the secrets in keyvault.
When we use Azure Managed Service Identity to access keyvault from an IaaS VM, my understanding is that any user logged into the VM or any program running on the machine can access the keyvault secrets - is this true?
And if it is, doesn't that decrease the security in case one of the user accounts is compromised?
According to the article access Azure Key Vault, it seems this is true. If MSI is enabled, just need to invoke web request in the VM without e.g. appKey.
And if it is, doesn't that decrease the security in case one of the user accounts is compromised?
It should be, but the prerequisites of the access to the secret in the keyvault is the VM service principal was added as a role in Access control (IAM) and Access policies.
If you want to increase the security, you may need to remove the VM service principal in the Access policies, then it will not be able to access the secret, if you want to the service principal does not have the access to the keyvault at all, remove its role in Access control (IAM).
For more details, you could refer to: Secure your key vault.
Update:
From the doc #Arturo mentioned, it is the fact.
Any code running on that VM, is able to call the managed identities for Azure resources endpoint and request tokens.
I have an application that runs in a Service Fabric(SF) cluster and I wan't to access Key Vault from it.
The cluster hosts a number of applications and I want to give access to a Key Vault for my application without giving access to the other applications. By default an application runs under the same user as the SF cluster, but each applicatiuon has it's own unique name, mine has the name fabric:/application1.
My question is, is it possible to create an Active Directory application account for fabric:/application1 and grant access to the key vault?
I know it is possible to use the RunAs options in the SF manifest, but that requires me storing an encrypted password in the manifest/source code and I want to try and avoid this if possible.
AFAIK,
The only way to have this flexibility is using ClientID & Secret or Service Principal certificates and each application manage their own credentials.
Service Principal Certificate is already integrated to AD, but does not require the application, the user or the Host to be part of the domain, the only requirement is setup an user on AD to grant the permissions on Keyvault.
There are other solutions using AD integration, like Managed identities for Azure resources(Former: Managed Service Identity) but I am not sure if you are able to restrict access per application like you described, because the MI add this as a service in the node, so technically other applicaitons would have access as well, worth a try to validate if you can restrict this.
If you want to try this approach, you can use with Microsoft.Azure.Services.AppAuthentication for implicit authentication of the services running in your cluster, where the nodes are setup with Managed Identities extension like described here.
Something link this:
When you use the Microsoft.Azure.Services.AppAuthentication, the Step 2 will be handled by the library and you won't have to add much changes to your key vault auth logic.
When you run your code on an Azure App Service or an Azure VM with a
managed identity enabled, the library automatically uses the managed
identity. No code changes are required.
The following docs describe other options you can use for KeyVault Authentication.
PS: I've done other KeyVault integrations using Client Secrets and Certificates and they are secure enough, With Certificates you can store it on the managed store or with the application, I would recommend MI only if is a requirement for your solution.
I am using .NET Core 2.0 and ASP.NET Core 2.0 for application development. The "test" application is a .NET Core Console application. The core code I am writing is a class library. Once proper testing. I choose to do this since I won't be putting this to use for awhile (it's replacing older ASPNET code).
Anyway, since I have to work with a LOT of API keys for various services I decided to use Microsoft Azure Key Vault for storing the keys. I have this all setup and understand how this works. The test application uses a test Azure account so it's not critical. And since this is replacing legacy code and it's in the infancy, I am the sole developer.
Basically, I'm running into this issue. There's not too much information on Azure Key Vault from what I can see. A lot of examples are storing the Client ID and Secret in a plain text json file (for example: https://www.humankode.com/asp-net-core/how-to-store-secrets-in-azure-key-vault-using-net-core). I really don't understand how this can be secure. If someone were to get those keys they could easily access stored information Azure, right?
The Microsoft MSDN has a powershell command that grants access (I lost the original link, this is closest I can find: https://www.red-gate.com/simple-talk/cloud/platform-as-a-service/setting-up-and-configuring-an-azure-key-vault/) My development operating system is Windows 10 and my primary server operating system is Debian.
How would I approach this?
Yes, you are right, the plain text config file could be used only during development, not for production purpose. And in general, available options depend on where and how you host an App.
If you have an Azure Web App, you have at least next built-in options (from the documentation):
add the ClientId and ClientSecret values for the AppSettings in the Azure portal. By doing this, the actual values will not be in the web.config but protected via the Portal where you have separate access control capabilities. These values will be substituted for the values that you entered in your web.config. Make sure that the names are the same.
authenticate an Azure AD application is by using a Client ID and a Certificate instead of a Client ID and Client Secret. Following are the steps to use a Certificate in an Azure Web App:
Get or Create a Certificate
Associate the Certificate with an Azure AD application
Add code to your Web App to use the Certificate
Add a Certificate to your Web App
You may also find an approach that uses env variables to store credentials. This may be OK only if you can guarantee that it's not possible to do a snapshot of env variable on prod machine. Look into Environment Variables Considered Harmful for Your Secrets for more details.
And the last one thing: there is also a technic that based on the idea, that you need to store/pass only a ClientSecret value while ClientId should be constructed based on machine/container details where the App is hosted (e.g. docker container id). I have found an example for Hashicorp Vault and an App hosted on AWS, but the general idea is the same: Secret management with Vault
In addition to the first answer, with the context of running applications on Azure VM, instead of using client_secret to authenticate, you can use client certificate authentication as explained in this documentation: Authenticate with a Certificate instead of a Client Secret.
In the picture above:
Application is authenticating to AAD by proving that it has the private key of the certificate (which is basically stored in CNG if you are using Windows).
Application get back the access_token and then use it to access the Key Vault.
The developer does not need to know the private key value of the certificate in order for their app to be successfully authenticated. Instead, they only need to know the location of the imported pfx (a container for private key and its certificate) in the Certificate Store.
At least on Windows, you as secret administrator can convert the private key and the certificate into pfx format which is password protected, and then deploy it into the Windows Certificate store. This way no one could know the private key unless they know the password of the pfx file.
The other approach specifics for Azure Compute, is to use Azure Managed Service Identity. Using Azure MSI, Azure will automatically assign your resources such as VM with an identity / Service Principal, and you can fire requests at a specific endpoint that are only accessible by your resource to get the access_token. But be wary that Azure MSI are still under public preview, so please review the known issues before using it.
The picture above explain how Azure Resource Manager assign a Service Principal identity to your VM.
When you enable MSI in a VM, Azure will create a service principal in your AAD.
Azure will then deploy a new MSI VM extension to your VM. This provides an endpoint at http://localhost:50432/oauth2/token to be used to get the access_token for the service principal.
You can then use the access_token to access the resources such as Key Vault which authorize the service principal access.