fix: address some security concerns (ENG-5229)

function-app-v1/ProviderRelay/__init__.py

@@ -32,6 +32,7 @@ def get_session(client_id, audience, role_arn):
         managed_identity_client=client_id, exclude_environment_credential=True
     )
     token = creds.get_token(audience)
+    logging.info(f"token: {token}")
     try:
         res = client.assume_role_with_web_identity(
             WebIdentityToken=token.token,

function.tf

@@ -20,6 +20,7 @@ resource "azurerm_application_insights" "stacklet" {
   resource_group_name = azurerm_resource_group.stacklet_rg.name
   application_type    = "web"
   tags                = local.tags
+  retention_in_days   = 90
 }
 
 resource "azurerm_service_plan" "stacklet" {
@@ -84,38 +85,40 @@ data "archive_file" "function_app" {
   output_path = "${path.module}/function-app.zip"
 }
 
-# azurerm_linux_function_app's zip_deploy_file will only redeploy the function
-# when the path changes, so copy the zip to a versioned filename.
-resource "local_file" "function_app_versioned" {
-  filename = "${path.module}/function-app-${data.archive_file.function_app.output_sha256}.zip"
-  source = "${path.module}/function-app.zip"
-}
-
 resource "azurerm_linux_function_app" "stacklet" {
-  name                = "stacklet-${var.prefix}-function-app-${substr(random_string.storage_account_suffix.result, 0, 15)}"
+  name                = "${var.prefix}-relay-app"
   resource_group_name = azurerm_resource_group.stacklet_rg.name
   location            = azurerm_resource_group.stacklet_rg.location
   service_plan_id     = azurerm_service_plan.stacklet.id
 
   storage_account_name       = azurerm_storage_account.stacklet.name
   storage_account_access_key = azurerm_storage_account.stacklet.primary_access_key
+  # storage_uses_managed_identity = true
+  # After the function has been deployed, this value can be turned on, and
+  # the storage_account_access_key removed. However, doing this also stops
+  # future redeployment of the function code from working as the azure cli
+  # looks for the `AzureWebJobsStorage` application setting. When using the
+  # managed identity, this value is not set, and instead
+  # `AzureWebJobsStorage__accountKey` is set. This causes the azure cli to
+  # fail to deploy the function code.
 
   # Enforce HTTPS on the HTTP endpoint even though the data plane aspect of it
   # is unused, to avoid showing up in security checks.
   https_only                    = true
-
-  # Deploy from zip file
-  zip_deploy_file = local_file.function_app_versioned.filename
+  client_certificate_enabled    = false
+  public_network_access_enabled = false
 
   site_config {
-    application_insights_key = azurerm_application_insights.stacklet.instrumentation_key
+    application_insights_key               = azurerm_application_insights.stacklet.instrumentation_key
+    application_insights_connection_string = azurerm_application_insights.stacklet.connection_string
 
     application_stack {
       python_version = "3.12"
     }
 
     # More somewhat pointless HTTP security; the HTTP data plane is unused.
     minimum_tls_version = "1.3"
+    http2_enabled       = true
   }
 
   app_settings = {
@@ -150,3 +153,37 @@ resource "azurerm_linux_function_app" "stacklet" {
     ]
   }
 }
+
+# In order to get the underlying function in the application to be redeployed
+# when the zip hash changes, we need use fork out to the azure cli.
+# We also need to temporarily enable public access to allow the zip file to be deployed.
+resource "null_resource" "deploy_function_app" {
+  depends_on = [azurerm_linux_function_app.stacklet]
+
+  provisioner "local-exec" {
+    command = <<-EOT
+      # Temporarily enable public access
+      az functionapp update \
+        --name ${azurerm_linux_function_app.stacklet.name} \
+        --resource-group ${azurerm_resource_group.stacklet_rg.name} \
+        --set publicNetworkAccess=Enabled
+
+      # Deploy the code
+      az functionapp deployment source config-zip \
+        --name ${azurerm_linux_function_app.stacklet.name} \
+        --resource-group ${azurerm_resource_group.stacklet_rg.name} \
+        --src ${data.archive_file.function_app.output_path} \
+        --build-remote true
+
+      # Disable public access again
+      az functionapp update \
+        --name ${azurerm_linux_function_app.stacklet.name} \
+        --resource-group ${azurerm_resource_group.stacklet_rg.name} \
+        --set publicNetworkAccess=Disabled
+    EOT
+  }
+
+  triggers = {
+    build_hash = data.archive_file.function_app.output_sha256
+  }
+}

main.tf

@@ -28,7 +28,7 @@ data "azurerm_role_definition" "builtin" {
 resource "random_uuid" "app_role_uuid" {}
 
 locals {
-  resource_group_name = coalesce(var.resource_group_name, var.prefix)
+  resource_group_name = coalesce(var.resource_group_name, "${var.prefix}-stacklet-relay")
 }
 
 resource "azurerm_resource_group" "stacklet_rg" {
@@ -44,6 +44,27 @@ resource "azurerm_user_assigned_identity" "stacklet_identity" {
   tags                = local.tags
 }
 
+# Role assignment for storage queue access
+resource "azurerm_role_assignment" "storage_queue_data_contributor" {
+  scope                = azurerm_storage_account.stacklet.id
+  role_definition_name = "Storage Queue Data Contributor"
+  principal_id         = azurerm_user_assigned_identity.stacklet_identity.principal_id
+}
+
+# Role assignment for storage blob access (for function app packages)
+resource "azurerm_role_assignment" "storage_blob_data_contributor" {
+  scope                = azurerm_storage_account.stacklet.id
+  role_definition_name = "Storage Blob Data Contributor"
+  principal_id         = azurerm_user_assigned_identity.stacklet_identity.principal_id
+}
+
+# Role assignment for Application Insights
+resource "azurerm_role_assignment" "monitoring_contributor" {
+  scope                = azurerm_application_insights.stacklet.id
+  role_definition_name = "Monitoring Contributor"
+  principal_id         = azurerm_user_assigned_identity.stacklet_identity.principal_id
+}
+
 resource "azuread_application" "stacklet_application" {
   count           = var.azuread_application == null ? 1 : 0
   display_name    = "${var.prefix}-application"

provider.tf

@@ -34,4 +34,7 @@ provider "azurerm" {
   }
 
   subscription_id = var.subscription_id
+
+  # Use Azure AD authentication for storage operations (required when access keys are disabled)
+  storage_use_azuread = true
 }

storage.tf

@@ -33,7 +33,26 @@ resource "azurerm_storage_account" "stacklet" {
   location                 = azurerm_resource_group.stacklet_rg.location
   account_tier             = "Standard"
   account_replication_type = "LRS"
-  tags                     = local.tags
+
+  # Defaults to https only, and min tls version is 1.2
+
+  # allow_nested_items_to_be_public = false
+  # Setting this value to false prevents the deployment of the function app
+  # using the azure cli.
+
+  # public_network_access_enabled = true
+  # Setting this value stops terraform being able to query the state of the
+  # storage account. It seems that the control plane and data plane are
+  # effectively configured together. In order to turn this value on, the
+  # storage account needs to configure either selective IP allow lists or
+  # selective VPCs.
+
+  # shared_access_key_enabled = false
+  # Turning this value on seems to prevent the servers running the function
+  # app from acquiring some form of internal lock that is necessary, so the
+  # azure function wrapping code is never able to query the storage queue.
+
+  tags = local.tags
 }
 
 resource "azurerm_storage_queue" "stacklet" {