As part of an Exchange Online migration I was re-running a Public Folder sync that I initially ran two years ago when we still had basic authentication enabled from internal networks for Exchange Online. When I went to re-run I realized it was not going to work because we completely disabled basic authentication. I had recently created an unattended script that connects to Exchange Online using MSAL and integrated Windows authentication (IWA) and was able to use this methodology in this script without making any modifications to the script.
To make this work we first need to install the MSAL.PS module:
Install-Module -Name 'MSAL.PS'
Next, we need to get an OAuth token for Exchange Online PowerShell, create a bearer token secure string, and create a basic credential from it along with our UPN. We will need to specify the client ID for Exchange Online PowerShell (612e1698-a44c-49a4-b66b-4188cc69cbaa) along with our tenant ID (I’m entering a fake one here). When prompted you will login with an account that has EXO administrative access:
We recently came across an application that uses Graph API and we wanted to start using it for some our on-prem mailboxes. Hybrid Graph API only supports delegated authentication flows and not application authentication flows. Just because something isn’t “supported” doesn’t mean you can’t make it work! There are two things that we’ll need to do to make this work.
First, any internet-facing Exchange server will need to have ‘V1S2SAppOnly‘ OAuth support added. You can do this by adding V1S2SAppOnly to the OAuthHttpModule.Profiles key in the REST web.config (ex. …\Program Files\Microsoft\Exchange Server\V15\FrontEnd\HttpProxy\rest). Once you add this value the key should look like this:
After this has been added either perform an iisreset or restart the ‘MSExchangeRestFrontEndAppPool‘ app pool in IIS for each server where you did this modification.
The next step is to add the appropriate ‘AppOnlyPermissions‘ to the Microsoft Graph partner application in Exchange/AD. First we’ll take a look at our Graph partner application:
The ‘AppOnlyPermissions‘ value should be blank. We need this to match the ‘ActAsPermissions‘ value. You’d think (and some other articles say) that you could just run Set-PartnerApplication -AppOnlyPermissions… but this was not a supported parameter for me. To set this we’ll have to edit AD directly. You’ll need to fire up something like ADSIEdit, load the AD configuration partition, and drill down to your Exchange org and partner application object. The path should be something like:
Once here you’ll need to open the ‘Microsoft Graph‘ object and copy the ‘msExchConfigurationXML‘ attribute value to the clipboard:
Next, we’ll use Notepad++ and the XML Tools plugin to manipulate this. MAKE SURE YOU BACKUP THIS VALUE. We want to convert this to a nicely formatted XML output so that it is easy to work with. To do this we use the ‘Pretty print’ option in XML Tools.
Once we have this we’ll need to duplicate all of the ‘ActAsPermissions’ lines and then use find and replace to convert those tags to be ‘AppOnlyPermissions’. Doing this will create a set of application permissions based on our delegated permissions (ActAsPermissions).
Once completed we need to linearize the output again so that we can copy it back into AD. We can use the ‘Linearize’ option in XML Tools for this:
Once we have the XML in the proper format we can put it back in the AD object:
Now that we’ve updated the AD object we can verify everything looks good by checking Exchange again (AppOnlyPermissions should have the same values as ActAsPermissions):
I recently had a call with another company attempting to setup Autopilot following my previous post (Windows Autopilot with User-Driven Hybrid Azure AD Domain Join using Palo Alto GlobalProtect VPN). While speaking to them I learned that are currently using basic credentials (LDAP+RADIUS) with GlobalProtect and are only attempting to setup certificate authentication to get Autopilot working. They were still planning on having the user perform two-factor basic authentication after the Autopilot-based deployment. This configuration was the perfect use-case for GlobalProtect’s new “Use Connect Before Logon” functionality. This functionality was introduced version 5.2 and works by registering a Pre-Login Access Provider (PLAP). With PLAP you now have interactive access to the GlobalProtect client at the logon screen. A huge plus with this method is that it requires NO back-end changes to your existing GlobalProtect configuration. The functionality is completely client-side and only really requires an additional step during installation. This PLAP functionality works with basic credentials, certificates, and even SAML! I will be using basic two factor credentials below.
The first step will be to create a new GlobalProtect package in Intune. I am using the newest version below, 5.2.7. You can use the same steps for creating the package that I laid out in my first post, but we will be using an alternate wrapper script, InstallGlobalProtect_PLAP.ps1. InstallGlobalProtect_PLAP.ps1, will install GlobalProtect, set our default GlobalProtect portal, and register the Pre-Login Access Provider (PLAP). Everything else non-certificate related in my original post will still apply (ex. IntuneHybridJoinHelperInstaller.ps1).
Once the machine has been deployed you will notice an extra button in the lower right. This is the PLAP.
When clicked, GlobalProtect will attempt to connect to the portal configured in the wrapper script and you will be presented with a screen like the one below. The prompts here will vary based on your authentication method. Here I am being prompted for my LDAP credentials to authenticate to the portal.
Once I passed the correct credentials here (and the correct second set of credentials at a second screen for two-factor) I was connected.
At this point you can click the ‘Back’ button and continue to log in to the device. That’s all there is to it! This is a great option for those of you who are lacking the desire to use certificates in your existing GlobalProtect configuration, but want to start using Autopilot.
We recently came across an issue where after a recent Zoom app update anytime a user clicked a tel: link from within the work profile the Zoom app was launched instead of the native dialer. We first looked into what changed within the app in the release notes. It was obvious that a change in 5.6.3 caused this:
Support for tel: protocol links - Android
Users can tap a telephone number in an email or web page to call that number with Zoom Phone.
What is happening here is Zoom is now registered to handle tel: links and because it exists inside of the work profile it is taking precedence (the dialer is not really part of the work profile). This would have happened if we introduced ANY application that was registered to handle tel: links, but Zoom was the only one for us. What we need to do is bring the dialer into the work profile to give it precedence and to do that we use the ‘Android Enterprise system app‘ option to create instances of the native dialers in the environment. In our case we created one for the native (Google) Android dialer and one for the Samsung dialer. You may have other vendors in your environment so there may be others you need to add.
First, we need a target group of devices. We could simply deploy the applications to all devices and it will take effect where applicable, but it is better to be more targeted and specific when possible. We already have two Azure AD dynamic device groups that target our Android Enterprise devices. You can also get more specific using properties like deviceManufacturer to target specific vendor devices.
The first group is for our personally-owned work profile devices:
The second group is for our corporate-owned devices (this technically will also target non-work profile Android Enterprise devices, but there is no harm in that):
Next, we need to create our dialer apps in Intune and assign them to our dynamic device groups. We are going to create one for the native Android dialer (com.google.android.dialer) and one for the Samsung dialer (com.samsung.android.dialer). The first time I attempted this I had an issue with the Samsung app installing because I had a capital ‘S’ in Samsung. Make sure the app id is all lowercase.
Once the new apps have have applied to the devices you will notice that the ‘Phone’ app shows inside the work profile app launcher. Once the ‘Phone’ app shows there you will be able to open tel: links in the dialer from within the work profile again.
Some months back I was faced with an interesting challenge in Microsoft Teams. How do we prevent chats for users that have access to an external tenant? The background here is that certain industries like mine require you to capture and archive electronic communications. Depending on how you interpret the requirement this could include chats taking place in a tenant other than your own. An administrator naturally has zero access to the data in a tenant other than their own. You can use tenant restrictions to prevent a user (who you have some network or proxy control over) from accessing a tenant other than your own, but there are cases where a user needs this access for external collaboration. The challenge here was finding a way to prevent certain actions while working in another tenant… and so came TeamsFirewall.
I started analyzing Teams traffic in my free time and got a feel for how it works. I quickly realized that I could approach the issue with a scriptable proxy server that supported HTTPS. I chose mitmproxy for this task. In the first iteration of the product I broke traffic down to actions (like sending a message or deleting a message) and location (internal tenant or external tenant). After I had that working I wanted to expand the functionality to control actions based on not only location, but on other things like conversation type and participants. After more development I came up with a process that learns the environment by looking at action -> conversation -> participants. The system caches API tokens from the users it supervises to make requests on their behalf in order to learn what it does not know about the environment. The product does not need any credentials or direct access to a tenant to function. This data is then saved in the cache database (teamsfirewall_cache.db). The cache is used for lookups so that faster decisions can be made on the fly. Cache lifetime of both the user table and the conversation table can be configured via the config file.
The rule engine of the product allows for extremely granular rulesets. You can get as granular as saying user A cannot edit messages sent to user B or as broad as user A/B/C cannot communicate with anyone @companyA.com. I would like to note that M365 Information Barriers perform some basic ethical wall functions, but it does not have much granularity and does not address examples like the the one above with external tenants.
Some next steps are improving the scalability of the product and developing an easily deployable package. I am currently looking at adding an option to use a central database for the cache database and using Docker containers with a load balancer to add more workers. You can see some of this upcoming work in the TODO file within the project repository (below).
UPDATE (11/2/22): I just noticed a parameter in OWA policies in EXO PowerShell the other day that allows you to set the default domain for Bookings calendars. I was able to set the default domain to bookings.company.com and it is taking effect for all newly created Bookings calendars. You need to run this for all OWA mailbox policies that apply to users who have access to create calendars. The command I used was:
During testing of Microsoft Bookings we found there was no way to adjust the sender email address for a Bookings company. This was frustrating as our default address was our default tenant address (which ended in onmicrosoft.com). This led me to ask how does this system even work? What is generating the emails? What hosts the email address?
Let’s start by looking for the object(s) hosting this service from an email perspective. To do this we can issue the command below in an Exchange Online PowerShell session. For each ‘company’ created in Bookings there is an associated mailbox created with the ‘RecipientTypeDetails‘ of ‘SchedulingMailbox‘. These mailboxes are not visible in the Exchange Online admin console.
Now that we have the new sub-domain we can use this in our Bookings company mailbox. The easiest way to do this is to adjust the UPN of the user account associated with the mailbox. This will automatically update the address.
After the Exchange Online directory syncs up with Azure AD the email address should be updated.
Now we need to unpublish and re-publish the company booking page so that it starts using the new address.
After re-publishing the page we need to wait a little bit for it to pick up the new email address. Eventually Bookings will start sending emails from the new email address. In testing it took around 10-15 minutes to take effect.
One more thing to note is that Bookings does not give you a method to delete companies after creating them. If you ever need to delete the company you simply need to delete the mailbox/user using the Remove-Mailbox cmdlet.
After implementing Teams last year we were faced with a number of issues to work through. One of the issues we were having was the inability of mailbox delegates to schedule meetings on behalf of mailbox owners. If you are a hybrid Exchange user, the very first step is making sure you have OAuth configured between your on-premises environment and Exchange online. The Hybrid Configuration Wizard should be taking care of this for you. I’ve also written an article on properly securing it externally. You will also need create and configure a service principal to allow Teams to interact with your on-premises Exchange environment. Steps 2-3 in this article must be preformed to set this up. If one of these items are not configured, you will receive the error ‘Sorry, but we can’t connect to the server right now. Please try again later.‘
Once these pre-requisites are configured, you may find users are still receiving errors when trying to schedule Teams meetings on behalf of other users. An error that many of our users were experiencing was ‘Looks like you don’t have permission to schedule meetings for this account. Talk to the owner to get permission and try again.‘
During testing we found that Teams was looking for a specific set of permissions on the owner’s mailbox when generating the Teams meeting in Outlook. In order to create a meeting as a delegate, the delegate must be a true ‘delegate’ under the owner’s mailbox AND have either ‘Author‘ or ‘Editor‘ access to the calendar like the example below. If the user is not present in this list or if they are present but show a calendar permission such as ‘Custom‘ in this dialog, they will receive an error like the one above when creating the Teams portion of the meeting even if they technically have all the required permissions to create a meeting on behalf of the organizer. Many times permissions are granted either by giving full access (Add-MailboxPermission) to the mailbox or by granting folder level permissions (Add-MailboxFolderPermission) and by granting Send-On-Behalf/SendAs permissions.
To understand why Teams requires such a specific configuration, let’s look at what is happening in the background… When you are creating a new Teams meeting via Outlook, the Teams back-end actually does a OAuth-authenticated EWS (Exchange Web Services) ‘GetDelegate‘ call to your on-premises Exchange environment to verify that the user creating the meeting actually has access to the mailbox. The request look like this:
Exchange will respond with delegate list and associated permissions (similar to the view you see in Outlook).
<?xml version="1.0" encoding="utf-8"?>
<!-- Note: EwsEditor has replaced the "utf-16" text in the first line with"utf-8" in order for the XML to render in the response web control. -->
<h:ServerVersionInfo MajorVersion="15" MinorVersion="1" MajorBuildNumber="1979" MinorBuildNumber="3" Version="V2017_07_11" xmlns:h="http://schemas.microsoft.com/exchange/services/2006/types" xmlns="http://schemas.microsoft.com/exchange/services/2006/types" xmlns:xsd="http://www.w3.org/2001/XMLSchema" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" />
<s:Body xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xmlns:xsd="http://www.w3.org/2001/XMLSchema">
<GetDelegateResponse ResponseClass="Success" xmlns="http://schemas.microsoft.com/exchange/services/2006/messages">
The important things here are that the user is present and has either ‘Author‘ or ‘Editor‘ permissions to the calendar. Remember, even having ‘Custom‘ permissions to the calendar will result in an error during meeting creation.
Another issue you can run into is one where you have already added the user as a delegate with the appropriate calendar permissions, but still receive the error ‘Looks like you don’t have permission to schedule meetings for this account. Talk to the owner to get permission and try again.‘ To troubleshoot this, let’s look at the mailbox from the perspective of Teams and perform the EWS ‘GetDelegate‘ call ourselves. We can use EwsEditor to do this.
In this example we receive an empty response from EWS even though we have properly configured the delegate permissions. This is indicative of mailbox corruption and to resolve this we have to understand how delegates are actually stored in the mailbox. Delegates actually translate to a number of things, but at the center of all of this is the ‘LocalFreebusy‘ object in the mailbox which can only be viewed with a tool like MFCMapi. The reason EWS is returning a blank list is because the object in the mailbox is corrupt and missing the delegate information. In order to resolve this we will use MFCMapi to delete the object. You can perform this on behalf of the user if you have full mailbox permissions and write access to their AD account, but in this example we will run as the mailbox owner using their default MAPI profile. To reset/repair the delegates list we will perform the following steps:
Before anything, record all of the users delegates from Outlook along with all the permissions for each one
Go to QuickStart -> Open Folder -> Local Freebusy
Choose the item with the subject ‘LocalFreebusy‘
Choose ‘Delete message‘ and when prompted with options choose ‘Permanent delete passing DELETE_HARD_DELETE…’
Go back into Outlook and re-create the user’s delegate(s)
After recreating the user’s delegates we can issue the EWS call again and see if they now show properly. The response should contain every delegate user along with their ‘Author‘ or ‘Editor‘ calendar permissions. Once you have verified EWS is returning the correct information you can try creating a new Teams meeting request.
Back in April, at the beginning of the pandemic, I started putting a lot of focus into getting Windows Autopilot to work with Hybrid Join clients and Microsoft Always On VPN. I was looking at both for different reasons but also looking at them as a combined solution. The issue with Autopilot was that technically you were still required to have line of sight to a domain controller even though the domain join happened via an offline blob using the on-prem Intune connector. Sifting through logs I could see the only thing holding back a successful enrollment was a little function at the end of enrollment that was simply looking for a domain controller. I was able to sometimes get an enrollment to work via device tunnel MS VPN policies, but success wasn’t consistent and relied on policies/certificates coming down in a timely manner. In the logging I also saw references to a configuration parameter that would disable the DC check. Soon after, I found a post from Microsoft saying that they had this setting in private beta and would be releasing it in the coming months. After this I decided to put everything on the backburner and abandon MS VPN (I found the MS VPN solution using RRAS to be clunky and inconsistent with a lot to be desired).
Fast forward a few months and Microsoft finally released the new ‘functionality‘. At its core it is really just a flag telling OOBE not to perform a DC connectivity check. After enrollment is completed you are on your own to establish pre-login connectivity to facilitate an initial logon to your domain as there are no cached credentials yet on the machine. We are already a Palo Alto GlobalProtect customer and have been happy with the solution, so getting the two to work together just made sense. At the same time there has also been a push to implement a proper Always On VPN configuration. I’ll be writing a post dedicated to the full technical and security architecture around a cert-based Palo Alto Always On VPN configuration, so I’ll only briefly touch on the relevant parts here. Please refer to Palo Alto documentation on the missing pieces. There are a number of security aspects that should be taken into account like revocation, key storage, etc., and you should already have a proper certificate authority. I am not going over those in this article. You should also already have configured your Autopilot profiles, Intune Connector for Active Directory, etc. as per this document.
In this configuration I use a certificate-only approach (only using certificate profiles and no other authentication methods) for both the portal and the gateway. Remember, our first GlobalProtect connection after an Autopilot enrollment will be a pre-logon connection via certificate. There are no other authentication methods available for this first connection and the portal -> gateway authentication flow needs to support this. Before configuration of the portal and gateway you need to configure zones, interfaces, policies, and a certificate profile. These steps are documented here (steps 1-3 and 5-6).
Below is the portal config. Notice how there are no client authentication methods present. If you were to add any method here, it would be layered on top of the certificate authentication and would prevent the pre-logon connection. PAN-OS 9.0 has implemented mixed authentication support so that you can implement an either/or type of configuration here. The other important note is that your user connection (post-logon) will be connecting to the same portal/gateway and because of this we will be using certificates for the user as well.
The next step is to configure the agent settings within the portal config. Our config is being configured as Always On, but this is not technically required for Autopilot to work. If you do not want an Always On user connection, set ‘Connect Method‘ to ‘Pre-logon then On-Demand‘. Some Palo-Alto documents mention using multiple agent configurations for pre-logon and post-logon that use different connect methods, but this is not necessary here (and will not always work as expected due to the order of operations). The other important thing is to set ‘Client Certificate Store Lookup‘ to ‘User and Machine‘ so that the client will be able to use user and device certificate. The client seems to do a good job at using the proper certificate depending on if the connection is pre-logon or post-logon.
After the portal you will configure the gateway. Authentication will be identical to the portal to allow for a seamless authentication flow.
To deliver a device certificate to the device we will use an Intune PKCS certificate profile. I won’t go into great detail here as Microsoft has done a good job of documenting the steps involved. The profile will need to be assigned to the device properly, and the easiest way to do this is by using an Azure AD dynamic group. I am using a custom EKU value in the screenshot below (Extended Key Usage) which you do not need to replicate. I had a specific reason for doing this.
You will also need to deliver a user certificate to the device. You can use another Intune PKCS certificate profile to do this or you can use GPO/User Certificate Autoenrollment. I chose the latter because I like the granular control it provides. If you use an Intune profile, but target the machine, every user that logs on to the machine will get a certificate and VPN access. If I target the user, every Intune-managed machine they log on to will get a user certificate. Both of these were undesirable. With autoenrollment I’m only enabling autoenrollment for the computers I want (using a user policy and loopback processing), and I’m controlling the users that can enroll/autoenroll via the ACL on the certificate template itself.
The first wrapper script is InstallGlobalProtect.ps1. This is the one responsible for installing the MSI and pre-configuring some registry values. I thought this part would be very straightforward, but I had trouble getting the pre-logon credential provider to kick in initially when installing the client via Intune. If I manually installed the client, it worked the first time without an issue. It was only acting this way when being deployed via Intune during Autopilot. After a few hours of procmon traces and some reverse engineering of the client I figured out the issue. There is a post-setup process that runs that doesn’t process some registry changes correctly until the client is executed in the context of a user at least once. I use the wrapper to stage these two registry values (LogonFlag + LogonState) along with the others needed to make this configuration work. I also enable User-initiated Pre-Logon (via the ShowPrelogonButton value), so it gives the user a chance to verify they have internet connectivity and so that they can perform a retry of the pre-logon connection on demand. I later turn this off via GPO making pre-logon completely automatic after the first successful login. This value is totally optional.
The next wrapper script is a batch script that launches the script above. It is called InstallGlobalProtect.cmd. This script is needed because Intune will launch the installer in 32-bit mode and we want everything kicking off in 64-bit mode (mainly for the registry work above).
Once the three files are ready, we can create our package which generates a .intunewin file.
We then take this file and upload it to Intune to create our application. The application will need to be assigned to a group. Again, I am using a dynamic group that targets my Autopilot devices. As you can see below some of the MSI info is pulled in automatically because it was read by the Microsoft-Win32-Content-Prep-Tool utility. We must also change the ‘Install command‘ to point to the batch file we created earlier.
In testing I came across multiple issues due to machine GPO not being applied before the first login. One of these was loopback processing not applying which caused multiple user GPOs not to apply. Others were trusted root certs not installing (used for things like SSL decryption) and User Certificate Autoenrollment not working (I touched on this earlier). The trusted root issue actually caused my hybrid join to get stuck (SSL decryption is being used here). I decided to create IntuneHybridJoinHelperInstaller.ps1 to solve all of this.
Modifies the SACL of the directory to remove modify access from ‘Authenticated Users‘ (someone could use this directory to execute malicious code in the context of ‘LOCAL SYSTEM‘ if you do not do this)
Create a script in the directory above called IntuneHybridJoinHelper.ps1 with an accompanying scheduled task that executes at any user logon in the context of ‘LOCAL SYSTEM‘
At the next logon, this newly deployed script is triggered by the scheduled task, checking to see if the computer group policy cache has ever been provisioned (has ever received computer policies) and if not it will do the following:
Perform a gpupdate for computer policies
Get the interactive logged on user
Create a task to run gpupdate as the currently logged on user which will perform a gupdate of their user policies
Re-run ‘Automatic-Device-Join‘ task to complete the device registration in case it failed at logon
It is best to deploy this as a Win32 app, like the GlobalProtect client, so that we can ensure it is on the machine before the first logon. Like GlobalProtect, we are using a batch wrapper (IntuneHybridJoinHelperInstaller.cmd) to launch the PowerShell script as a 64-bit process. I used the same dynamic group that I used for the GlobalProtect client as the target here. I also used a dummy uninstall command since we never need to ever uninstall this. For install detection I am just using the script path (C:\Scripts\IntuneHybridJoinHelperInstaller.ps1). We will build our package using the utility like we did for GlobalProtect.
Now that we have everything in place we can test an enrollment. If everything is configured properly, you’ll be asked to sign-in to your corporate environment right after establishing network connectivity. After everything completes you should wind up at a logon screen. Because I am using User-initiated Pre-Logon I will need to switch to the GlobalProtect logon provider, click ‘Start GlobalProtect Connection’, and wait for the status to change to ‘Connected’.
After logging on you are presented with the User ESP (Enrollment Status Page). This is when our helper script kicks in to resolve GPO issues and moves our device registration along. This process can take a bit because after the ‘Automatic-Device-Join‘ completes you still have to wait for the on-prem computer object to sync up to Azure AD via AD Connect. Steve Prentice came up with a little script to help speed this up called SyncNewAutoPilotComputersandUsersToAAD.ps1. It just forces an AD Connect sync after computer object has its ‘userCertificate‘ attribute populated.
Once this is completed you should be left at a functioning desktop and GlobalProtect should have switched over to a full tunnel using the user certificate. At this point I would be using my primary endpoint management product, Ivanti Endpoint Manager, to perform any additional application installs/configurations. I have its agent being deployed via Win32 app as part of my Autopilot process.
In the past many organizations completely blocked or limited external access to on-premises Exchange servers because of the lack of multi-factor authentication. Protocols like OutlookAnywhere (also known as RPC-over-HTTP, now MAPI-over-HTTP) and EWS had no native methods to accomplish multi-factor authentication. Failure to protect these protocols from external exposure has led to many breaches like FIN4 and London Blue.
HMA to the rescue… In 2017 Microsoft finally answered this deficiency with Hybrid Modern Authentication. I briefly touched on modern authentication in two previous articles (here and here). With Hybrid Modern Authentication Microsoft gave you the ability to use new technologies like modern authentication and conditional access for on-premises Exchange. Clients will connect using modern authentication by default once Exchange is on a supported version, supported clients are implemented, and the configuration is implemented. The issue here is that legacy Windows authentication is still available. You can simply disable modern authentication in the client or use a different client and you are now connected to on-premises Exchange with a simple username and password completely bypassing conditional access. Conditional access is only invoked when you are authenticating with modern authentication. Exchange 2019 implemented Authentication Policies which allow you turn off legacy authentication methods. If you are using Exchange 2019, you can use these to lock down your environment.
We were in the situation where we wanted to allow secure external access to Exchange (mainly for OutlookAnywhere, but also Outlook Mobile), but we couldn’t have any legacy authentication exposure. The solution we came up with was creating a set of externally facing Exchange 2016 mailbox servers (think Client Access Servers from the pre-Ex2016 days) that have all legacy authentication methods disabled (only OAuth available). These servers are the only ones exposed to the internet. The protocols we want to expose but lock down are ActiveSync (needed for Outlook Mobile), EWS (Exchange Web Services), MAPI, and OAB (Offline Address Book). To lock these down we ran the following against the externally facing servers:
After this is completed, Windows and basic authentication should now fail for these virtual directories.
IMPORTANT: It is VERY important to regularly check that these settings are still in place. You should always re-run these commands after any kind of Exchange update. If you do not do this, you could inadvertently expose your Exchange environment. A simple script could be run on a schedule to check and report on any changes to the authentication configuration of these virtual directories.
The second step is disabling or blocking the other virtual directories that do not need to be accessed externally. For us, these were ECP, OWA, PowerShell, and RPC. We have an on-premises load balancer with SSL bridging configured for our Exchange environment, so we used that to block access to these virtual directories. Another option is to use IP restrictions in IIS on these virtual directories. A third option is to disable the virtual directories via PowerShell. For those of you who want to allow secure access to OWA (Outlook Web Access) you can use Azure App Proxy to accomplish this or an ADC like NetScaler or F5 Big-IP.
The final step in this configuration is allowing the O365 servers to reach an unaltered version of EWS for the IntraOrganizationConnector used for Exchange Online to pull free/busy data (and other data like photos) from your on-premises environment. I found that for some reason the IntraOrganizationConnector fails to authenticate from EXO->on-premises when it uses the modified virtual directory even though all OAuth tests pass. I also use this configuration for my MRS endpoint when doing mailbox migrations since MRS wants to do traditional Windows authentication to EWS. If you are using the Microsoft Hybrid Agent, you shouldn’t have to do this since Azure App Proxy is taking care of the MRS and free/busy communication. I have still have an ongoing ticket open with Microsoft to understand the root cause of this. The workaround is fairly simple:
Create a namespace that can be used for EXO->on-premises communications. (Ex. exocomm.domain.com)
Configure this namespace to point to your regular INTERNAL and unaltered mailbox servers
I decided to do a Fiddler trace to get to the bottom of this and this is where the issue started becoming clearer. In the trace you see Outlook reaching out to autodiscover.domainname.com (which is on-prem), getting a 401 response, reaching out to login.windows.net/login.microsoftonline.com, and looping in this manner. This part of the capture aligned exactly with the mysterious white box.
In my case this specific set of users had a different primary SMTP address (and UPN) than the other users we had already enabled ADAL for and their autodiscover.domain.com URL was never added to our Azure AD service principals for the ‘Office 365 Exchange Online‘ application ID. Microsoft documentation talks about this in Step 5 of the link I added at the beginning of this post. Using the ‘MSOnline‘ PowerShell module I was able to add the URL to the service principal list.