Introduction

I recently had a bit of a … run in … with deploying Named Values in APIm. These values are connected to a keyvault. This is a very good strategy for keeping keys and passwords centralized and secret. I have done this before, but was trying to use Bicep this time, but in doing so I got an error.

The error

This is the error: Can not find Managed Identity with provided IdentityClientId : <Principal ID goes here>

What I was trying to do

I was trying to add the Named Value and KeyVault connection using this bicep:

resource ApimNamedValue 'Microsoft.ApiManagement/service/namedValues@2021-04-01-preview' =  {
  parent: ApimInstance
  name: 'BackEndAPIKey'
  properties: {
    displayName: 'BackEndAPIKey'
    secret: true
    keyVault:{
      secretIdentifier: '${MyKeyVault.properties.vaultUri}secrets/BackEndAPIKey'
      identityClientId: APIminstance.identity.principalId
    }
  }
}

resource ApimInstance 'Microsoft.ApiManagement/service@2021-04-01-preview' existing ={
  name: 'ApimInstanceName'
}

resource MyKeyVault 'Microsoft.KeyVault/vaults@2019-09-01'  existing = {
  scope: resourceGroup(KeyVaultRGName)
  name: keyVaultName
}

I am using “references” to existing instances of a KeyVault (row 19) and an APIm (row 15).
There is just one thing wrong with it: You cannot suppliy the ClientID by submitting the PrincipalID (sometimes also called ObjectID). That is what triggers the error.

The solution

I got the solution directly from Adrian Hall (thank you).
It is down to a misunderstanding of the documentation. You are supposed to supply a ClientID if you want to use another identity than the APIm identity. If you want to use the identity of the APIm you are trying to deploy a named value for: Simply omit the property.

TLDR; remove row 9 from the Bicep code above and everything will work.

Just a short thing about access

To get the APIm instance to read values from the KeyVault, you need to add it to your Access Policies for the KeyVault. You can achieve this by adding the following Bicep code to your deployment.

// APIm reference
resource APIminstance 'Microsoft.ApiManagement/service@2021-04-01-preview' existing = {
  name: 'APImInstanceName'
  scope: resourceGroup('APImRGName')
}

// Keyvault config
resource MyKeyvault 'Microsoft.KeyVault/vaults@2021-06-01-preview' existing = {
  name: 'KeyVaultName'
}

resource KeyVaultSetting 'Microsoft.KeyVault/vaults/accessPolicies@2021-06-01-preview' = {
  name: 'add'
  parent: MyKeyvault
  properties: {
    accessPolicies: [
      {
        objectId: APIminstance.identity.principalId
        tenantId: APIminstance.identity.tenantId
        permissions: {
          'keys': []
          'secrets': [
            'list'
            'get'
          ]
          'certificates': []
        }
      }
    ]
  }
}

Introduction

I asked on Twitter if, using API Management, there is any way of not allowing the user to send the API-key as a query parameter. The feedback and discussion made me create a blog post.

As ever, if you are looking for the solution, just scroll to the end.

The problem

TLS or HTTPs encryption

You are using an API-key to protect your API. That key is what security people calls a “shared private secret”. This means that anyone getting their hands on the key can use it, regardless of who you are and where. For most scenarios this is really good, but you do not want the key to fall into the wrong hands, you need to protect it.

You can protect it using TLS (called HTTPs) here is a good and detailed overview of the communication process. In HTTPs almost everything is protected using encryption. A very good and clear explaination of this can be found here.

Unencrypted

Encrypted

So what is the problem with sending an API-key as a query? It is encrypted, right?

Logging

The issue is not that anyone can intercept a message and look at the querystring, it is down to logging. When hosting your API (in Azure API manager or not) you might be behind a firewall, an application gateway or similar. The people running that log usually hosts the TLS endpoint, meaning they can decrypt the message. These logging tools usually surface some metadata about the message, including the querystring. The Azure Application Gateway is a great example of this.

This is an example of a call to an API. See how the subscription-key is logged?

Sending the API-key as a header is much more secure, from the logging standpoint. You should also consider sending other sensitive data such as PII as headers and not query for the same reason.

Best practise

Since the person hosting the private key (the pfx-cert) theoretically can decrypt and look at the whole message, they could access the data anyway. However this is not a built in functionallity in any Azure logging product (that I am aware of). I would consider not sending the API-key as a query, being a best practise; a should and not a must.

The solution

I added a policy to the Global policy in APIm. The Global policy is always executed before any call and I think you should consider putting more stuff in there.

<choose>
    <when condition="@(context.Request.OriginalUrl.Query.GetValueOrDefault("subscription-key","") != "")">
        <return-response>
            <set-status code="400" reason="Bad request" />
            <set-header name="content-type" exists-action="override">
                <value>application/json</value>
            </set-header>
            <set-body>{"Message":"Use the header Auth option."}</set-body>
        </return-response>
    </when>
</choose>

A flaw

This means that the call was sent all the way to APIm and was logged. However, it can create better habits in your callers from now on.

Another flaw

This solution has one obvious flaw: If someone renames the API-key query from subscription-key to something-else, then the policy will accept the incoming call, but remember this is a recommendation. A should.

A simple self-reminding how to

First of all, thank you to @ToonVanhoutte for writing the original post on how to get the bearer token using Fiddler (super sneaky!) and also to @splaxi (Mötz Jensen) who provided the PowerShell solution.

What are you trying to do?

Hopefully you found this post while googling “logging into azure as a user from postman” or “postman azure as user”. Most posts (including my own) are about how to access Azure APIs from Postman using an App Registration.

Sometimes that is not an option, and your account might actually have enough access-rights to get the information you want. The problem is that sometimes, the portal is not up to the task.

Using Postman to log into Azure as yourself

Rumor has it that it is possible to do this, using just postman but I was not able to, but a simple PowerShell script does the trick.

If you run the following script, you will log in and copy the bearer token to the clipboard. Then you paste it into the token textbox in Postman and you are good for about an hour untill your token has expired.

The powershell script

# Install-Module Az.Accounts

# Login as a user
Connect-AzAccount

# Get the token, and store it in the clipboard.
Get-AzAccessToken | Select-Object -ExpandProperty Token | Set-Clipboard

Note that you need to install the Az.Accounts module for this to work.

When you run this script you will get the old Login Window(tm), you login and when the script is done, the bearer is in the clipboard.

Postman

Return to Postman and find the Authorization tab. Select Bearer Token and paste the copied token into the textbox. That is it.

A strange experience and a bug

You can read this post in two ways, either get the full story or scroll down to the solution. My guess is that you found this post because you searched for this error message:

"Definitions/body parameters are not exported since this API references multiple schemas with OpenAPI components as opposed to one. Schemas referenced: [number],[another number]"

And you have spent some time being frustrated because removing the whole API and starting over did not work.

Backstory

This is not only filler, but it also describes how the error occured. Perhaps you recognize yourself in this.

The error from my point of view

I had developed an API for a colleague. The API had a demand for using Definitions of both requests and responses. Honestly, we as API developers, do not use that as much as we should. The colleague supplied me with examples and I got to working using the portal interface to enter definitions.

Things started going south when the portal stopped responding, but after a refresh everything seemed ok. I added all the definitions and then connected them to operations within the API. When I was done I exported the API as OpenAPI v2 (compatability reasons). This is when the error occured.

Looking at the OpenAPI file I could see that all my operations was missing their content properties, meaning that all definitions I had put in, had been omitted. Scrolling down in the OpenAPI file I found the error above.

Redeploy will not solve it

I deleted the entire API and started over, but the same thing happened again! How could APIm remeber “schemas” from an API that I had deleted? Time to get creative and provide the solution.

The solution – short version

You need to delete the “schemas” and you cannot do that from the portal, you need to access your APIm instance using the RestAPI. Find the offending schema using Api Schema – List By Api and the delete it using Api Schema – Delete

The solution – how to version

I will try to explain every aspect of this operation, to make you feel confident about deleting things from your APIm instance.

Schemas?

Looking back at the error message, it mentions schemas, which might seem strange as APIm never seem to use schemas but rather definitions. The reason is simple: OpenAPI uses schemas and a collection of definition is a schema. APIm usually puts all your defintions into one single schema. The problem was that something went wrong and my API had multiple schemas which messed up the export.

Since the schema property is never surfaced in APIm you cannot delete them. Even if I deleted the entire API, the schema was retained and since you cannot delete it (you can only delete all values/definitions within a schema) I was stuck.

The APIm Rest API?

One of the most confusing things is that there is an API to manage your Azure API management. A manager’s manager if you will. This is really well documented an you can find the basics on how to activate and access it here.

You need to activate the API via the portal and also generate an accesskey in order to execute any APIs. You can use Azure OAuth if you want to but I find it easy to just generate a key and supply it using an Authentication header. I use this in all my calls.

Using the APIm Rest API

This API can be used for all sorts of operations on your APIm instance and I encourage you to explore it further. For now you need to execute two calls: one to find the extra schema(s), and one to delete.

There is one important exception to the documentation, and that is the API base-address. If you want to use the easy, key-based, authentication you need to update the address given in the documentation. Lets look at the first API, Api Schema – ListByApi. The documentation gives the following address:

https://management.azure.com/... etc

You must use the direct API path:

https://serviceinstancename.management.azure-api.net.management.azure-api.net/... etc

If you use the first one, you will get an authentication error.

Find the extra schemas

The schema identifiers are given in the error message, but you need to know which one is the wrong one. Therefore you need to list the schemas for your API. Issue this call:

GET https://serviceinstancename.management.azure-api.net/subscriptions/[subscriptionID]/resourceGroups/[resourcegroupname]/providers/Microsoft.ApiManagement/service/[serviceinstancename]/apis/[ApiID]?api-version=2020-06-01-preview

If you use Postman you can easily handle the path variables by using this:

GET https://{{api-servicename}}.management.azure-api.net/subscriptions/:subscriptionID/resourceGroups/:resourceGroupName/providers/Microsoft.ApiManagement/service/:serviceName/apis/:apiId/schemas?api-version=2020-06-01-preview?api-version=2020-06-01-preview

The API-ID can be found on the Settings-page for the API, but there it is called Name and is located below Display Name.

The response contains the schemas, in an array and a count property at the end. If this property is > 1, you have too many schemas.

Find the schema ID

The schema object is defined with four base properties, id, type, name and properties. Confusingly the ID you are looking for is located in name. The ID is a string of numbers and sometimes also letters, but it is not a GUID.

As to finding which schema you need to delete? I cannot tell you. It depends on your data model. In my case it was easy.

Deleting the schema

To be extra sure that you will be deleting the correct schema; find the schema first. Issue this call:

GET https://serviceinstancename.management.azure-api.net/subscriptions/[subscriptionID]/resourceGroups/[resourcegroupname]/providers/Microsoft.ApiManagement/service/[serviceinstancename]/apis/[ApiID]/schemas/[SchemaID]?api-version=2020-06-01-preview

The schema ID is the one you got from the call above.

Here is a Postman call for you:

GET https://{{api-servicename-neu}}.management.azure-api.net/subscriptions/:subscriptionId/resourceGroups/:resourceGroup/providers/Microsoft.ApiManagement/service/:serviceName/apis/:apiId/schemas/:schemaId?api-version=2020-06-01-preview

If you get back the schema you want to delete, you simply issue the same call again, just change the VERB from GET to DELETE.

Conclusion

I have NO IDEA how the problem happened to begin with, and I do not like that schemas cannot be deleted from the portal, but is is easy to do using the API Management REST.

Yes I know, not the snappiest title.
In my previous post Secure communication with APIm and Functions using Managed Identity, I showed how easy it is to setup OAUTH-based authentication in front of your Azure Functions, and how to configure an APIm policy to call that function, thereby uping the security level of your Azure Function communication.

CI/CD using ARM

I like ARM and after some serious digging I found the ARM template definition for how to secure your Azure Function. I prefer to use version V2, but you should look at V1 as well, as it has better documentation.

You can find the V2 template definition here, and here is V1.

When you deploy your function app you are using CI/CD and ARM. In order to add the functionality of the authourization setting you need to add a resourse called Microsoft.Web/sites/config and set its name to yourfunctionappname/authsettingsV2. You can either add this to your existing ARM-template or as a separate template. If you add it to your existing ARM template, do not forget to add the dependsOn property.

Minimal ARM Template

Here is my ARM-template that will deploy the settings addressed in the other post.

{
    "$schema": "https://schema.management.azure.com/schemas/2019-04-01/deploymentTemplate.json#",
    "contentVersion": "1.0.0.0",
    "parameters": {
      "appClientId": {
        "type": "string",
        "metadata": {
          "description": "The Client ID of the AppReg that the function is a part of."
        }
      },
      "sites_function_name": {
        "type": "string",
        "metadata": {
          "description": "The function app name"
        }
      },
      "tenantID": {
        "type": "string",
        "metadata": {
          "description": "The ID of the tenant providing auth token"
        }
      }
    },
    "resources": [
      {
        "name": "[concat(parameters('sites_function_name'), '/authsettingsV2')]",
        "type": "Microsoft.Web/sites/config",
        "apiVersion": "2020-12-01",
        "properties": {
          "platform": {
            "enabled": true
          },
          "globalValidation": {
            "requireAuthentication": true,
            "unauthenticatedClientAction": "Return401"
          },
          "identityProviders": {
            "azureActiveDirectory": {
              "enabled": true,
              "registration": {
                "openIdIssuer": "[concat('https://sts.windows.net/',parameters('tenantID'),'/')]",
                "clientId": "[parameters('appClientId')]"
              }
            },
            "isAutoProvisioned": false
          },
          "login": {
            "tokenStore": {
              "enabled": true
            }
          },
          "httpSettings": {
            "requireHttps": true
          }
        }
      }
    ],
    "outputs": {}
  }

I will go thru some of the properties
– properties/platform/enabled This basically enables the setting. You can switch this between true or false if you, for instance, have different rules for TEST and PROD. Simply turn it off for TEST and enable it for PROD.
– globalValidation/requireAuthentication Should be set to true but in some cases you might want to be able to handle unauthorized calls as well.
– globalValidation/unauthenticatedClientAction Set this to Return401 if the function should be called as APIs. For other scenarios, consult the documentation.
– identityProviders/azureActiveDirectory This object contains all the settings needed for the AAD type authentication.
– identityProviders/azureActiveDirectory/registration/openIdIssuer This points to your tenant using its GUID. If you do not know your tenant GUID you can visit whatismytenantid.com. This setting makes sure that the authenticating token is issued by your Azure tenant.
– identityProviders/azureActiveDirectory/registration/clientId This is the ID of Application Registration that was created when you setup the function authentication. If this is the first run, create an application registration first and use it with this function. I made an easy walkthru post a while back.
– properties/httpSettings/requireHttps always require HTTPs.

Additional settings

Consider reading the documentation for additional settings. There are a lot of properties relating to other authentication providers, but also some settings relating to authentication and AAD that you might find useful .
jwtClaimChecks/allowedGroups If you are using claims and groups/roles, you can make sure that the caller has the right credentials (I have not validated this functionality with APIm)
allowedClientApplications Stop callers that are not APIm simply by adding the clientID of your APIm (I think, I have not tried this)
allowedAudiences Might be useful to add other clients as audiences as well. The connected Application Registration ID (the clientID in the ARM-template) is always allowed, even if it is not added as allowed.

Next step

You should now add your version of the ARM template to your CI/CD process, and let me know how it works out for you. You can find me on Twitter.