Build your own Thin-ish client with Windows 10 LTSB

After some positive user feedback from the launch of our new Server 2016-powered RDS setup I started wondering if it could have a wider use that just the remote access concept we initially wanted to address. One thought in mind was making use of old \ low-spec devices that would be a bit too clunky for running a modern OS but where the physical hardware itself was in good condition.

Chrome-OS esque distributions such as CloudReady sound nice but come at cost so I set up a little side-project to see if there’s anything that could be done with what we have on our licensing agreement or anything in the open-source space.

Looking around there do seem to be various thin-client “converter” products but again they all seem to be commercial e.g. https://www.igel.com/desktop-converter-udc/

The only other option I found was ThinStation which may also be worth a look when I have more time as it seems a bit more involved to get set up and I wanted to stick to the Microsoft RDP client for now for maximum compatibility.

Windows options

Going back some time I remember Microsoft released cut-down versions of Windows for RDS-type scenarios; going back to the XP days it was called Windows Fundamentals for Legacy PCs and morphed into Windows 7 Thin PC in its next incarnation. Effectively all I want the OS to do is boot up, log in quickly then pass the credentials to a pre-configured RDP file using the standard mstsc.exe application.

However building any solutions on a Windows 7 base going forward seems to be a false economy so I decided to have a look around to see what was available on the Windows 10 codebase – the results were interesting…

IoT is name of the day

Going forward it seems Microsoft have changed the branding for this kind of cut-down devices to Windows IoT. In fact there’s a free edition which sounds ideal but it only runs on certain devices and isn’t really geared for UI use:

Ref: https://www.theregister.co.uk/2015/05/21/first_look_windows_10_iot_core_on_raspberry_pi_2/
Ref: http://blogs.perficient.com/microsoft/2016/01/windows-10-iot-editions-explained/

Reading a bit further it appears Microsoft license an edition called Windows 10 IoT Enterprise for new thin client devices. Now it gets interesting… it seems that the OS itself is Windows 10 Enterprise LTSB but with some special OEM licensing. It just so happens the edu customers get Enterprise LTSB on EES licensing so it’s time to take a closer look!

What this does mean is that Windows 10 Enterprise LTSB gets features from the old Windows Embedded products such as Unified Write Filter, perfect for a locked down device that shouldn’t need to experience configuration changes to the base OS.

Ref: https://msdn.microsoft.com/en-us/windows/hardware/commercialize/customize/enterprise/unified-write-filter

All these features are available in Enterprise LTSB simply by going into Add \ Remove Windows Features window, look for the Device Lockdown section and add whichever ones meet your needs (more on this later).

Image & GPOs

After downloading the latest ISO the LTSB 2016 WIM was imported into MDT. I made a quick task sequence to get it up and running and deployed the OS to a Hyper-V VM.

Boot and logon speeds are very quick given the lack of any Modern Apps which usually need to be provisioned at each new login. The performance gain explains why quite a few people within education have used LTSB for their desktop builds against MS’ wishes; however they’ll miss out on new features such as the much-needed OneDrive Files on Demand that will only be provided to the Current Branch release.

In theory setting up a Mandatory Profile could speed up login even further but haven’t got round to trying that yet.

RDS domain SSO

Upon logging in with domain credentials the next aim is to seamlessly drop users into the RDS farm without any further prompts. After doing a bit of research this can be achieved by setting a couple of GPOs:

  • allow credential delegation
  • trust SHA1 signature of signed RDP file

The need to allow delegation of credentials is fairly commonly mentioned but a lot of the articles are old and don’t mention where this needs to be set in a 2016 farm. In fact you only need to allow the delegation on the FQDN of the Connection Broker based on the results of my testing so far.

Computer Configuration > Administrative Templates > System > Credentials Delegation

To avoid any unwanted prompts about trusting the signature of a signed RDP file populate the GPO mentioned above and copy \ paste the signature from the RDP file that is provided by RDWeb for whatever RDS Collection you want to connect to.

User Configuration > Administrative Templates > Windows Components > Remote Desktop Services > Remote Desktop Connection Client > Specify SHA1 thumbprints of certificates representing trusted .rdp Publishers

Custom shell

Now with the credentials side sorted out the final piece of the puzzle was to cleanly launch the session and (here’s the tricky bit) made a seamless logout once the RDS connection is closed. Now there’s a few ways to achieve the first part:

  • use the IoT Embedded Shell Launcher feature \ enable Kiosk Mode via System Image Manager
  • use the Custom User Interface User GPO

Ref: https://social.technet.microsoft.com/Forums/en-US/b4552957-45c2-4cc4-a13d-6397f06ee62e/windows-10-kiosk-build-embedded-shell-launcher-vs-custom-user-interface?forum=win10itprosetup

Ref: https://docs.microsoft.com/en-us/windows/configuration/set-up-a-kiosk-for-windows-10-for-desktop-editions

One thing to bear in mind with Shell Launcher is what happens when the shell i.e. mstsc.exe closes, you only have the choice of

  • Restart the shell.
  • Restart the device.
  • Shut down the device.
  • Do nothing

For the sake of speed logging off would be better so I decided to go with the Custom User Interface GPO – seeing as the Windows 10 device would be domain-joined anyway it also seemed a quicker more efficient way to configure multiple clients too.

Seeing as the Custom User Interface is a User GPO it goes without saying that Loopback Policy Processing needs to be enabled for the OU where the client resides. That also comes in handy for a few additional personalisation settings later on too.

The User GPO settings are summarised in the screenshot below, you can add more lock-down policies as you see fit:

Auto log-out on disconnect

Seeing as I wanted to automate the process as much as possible and all the devices would be domain managed anyway the GPO method seems to be the quickest way to achieve what I want. Also avoids needing to do an Add \ Remove Features step for each endpoint device.

Another important point is that the Shell Launcher method only provides options to relaunch the program, shut down or restart the machine. For speed I was aiming to log off the “client” when the RDS session is done so definitely going down the GPO route as a result.

In the GPO settings I initially tried the standard string you’d expect to launch a Remote Desktop session i.e. mstsc.exe C:\Default.rdp but noticed some strange behaviour:

  • Windows logs in
  • RDP file launched
  • connection starts
  • before the green bar completes i.e. handshake still in progress
  • host session logs out

This seemed like a behaviour I’ve seen with some other programs in the past where they appear to terminate mid-way through actions actually occurring. To check I tried manually with the “start” command with the same result. It appears mstsc.exe doesn’t play nicely so we need another way…

Plan b) was to monitor the mstsc.exe process then log out from the client once RDS disconnected and therefore the process was no longer running. After looking around and trying a few scripts out I settled on one I found here:

Ref: https://www.experts-exchange.com/questions/24218998/Check-if-a-process-is-running-in-vbs.html

Just add the logout command as the action to run when the desired process terminates and we have the desired behaviour. It takes a second or two to react to the process closing but there doesn’t seem to be a way to speed that up as far as I can see.

Final steps

Now just some finishing touches required to give the solution a bit of polish 🙂

  • set logon and desktop wallpaper
  • disable Task Manager and related lockdown setings

When the machine boots users see this login screen, easily customised via GPO…

After login connection to RDS is pretty much immediate and no further credential \ security prompts appear…

UWF

The final piece of the puzzle is tidying up after the client has been in use for a while. That’s where the Unified Write Filter from earlier comes in handy:

Enable-WindowsOptionalFeature -Online -FeatureName Client-UnifiedWriteFilter

Then enable the filter;

uwfmgr.exe filter enable

Ref: https://docs.microsoft.com/en-us/windows-hardware/customize/enterprise/unified-write-filter
Ref: https://developer.microsoft.com/en-us/windows/iot/docs/uwf
Ref: https://deploymentresearch.com/Research/Post/632/Using-the-Unified-Write-Filter-UWF-feature-in-Windows-10

And there you have it, a locked down RDS client that will run on older hardware (Windows 10 works on pretty much anything from the last 10 years) which can be managed through your standard AD infrastructure, all using stuff you already have access to via your Campus agreement… enjoy!

Server 2016 RDS via Azure AD Application Proxy end-to-end guide

remote_desktop_blueOne of our priorities for this year was to improve our remote access offering to staff to enable more flexible working whilst outside of college. Office 365 helps greatly and has already improved functionality in many ways but there’s still some legacy applications and classic file shares that need to be provided remotely too. If at all possible we prefer the files not to leave the network so some form of virtual desktop looked the way to go.

After discounting VMware and Citrix offerings on cost grounds the improvements to Microsoft’s RDS offering in Server 2016 seemed to come at a perfect time.

Even more so now we’ve implemented Azure AD Application Proxy (more on that shortly!) We’ve also recently decommissioned some services that freed up a bit of physical hardware resource to “play” with so away we went!

Server installation

The physical hardware for now is running on some reclaimed Dell PowerEdge R610 servers; 64GB RAM, dual CPU and 6 x 15k disks in RAID10. Should be plenty to get us up and running with the RDS roles eventually split across two hosts. For now we’re running on just the one but even that’s plenty to get up and running with.

We installed Server 2016 Core running the Hyper-V role, which was simple enough. The Core role looks to be a tad more polished in Server 2016, although not new the sconfig tool got the main settings entered with fairly minimal fuss.

r610
yes it will go back in the rack once we’re done with it!

Getting the OS to update correctly wasn’t so simple due to Microsoft doing something silly to the update mechanism in the initial release of Windows 10 1607 and its equivalent Server 2016 release. Update status was stuck on “Downloading” showing no signs of progressing. In the end manually installing the latest Cumulative update release from the Microsoft Update Catalog did the trick e.g.

wusa.exe windows10.0-kb3213986-x64_a1f5adacc28b56d7728c92e318d6596d9072aec4.msu /quiet /norestart

Server roles

With Hyper-V up and running the next stage was to install our guests. We went with 3 VMs set up as follows:

  • Connection Broker \ RD Licensing
  • RD Web Access \ RD Gateway
  • RD Session Host

The original plan was to try and embrace the Server Core concept and only install the GUI where absolutely necessary. With that in mind we made the first two servers with Core and only the Session Host with a GUI. More on that soon… (!)

add-roles-wizard
RDS deployment wizard Role Services

Running the deployment through Server Manager on my desktop was easy going, Microsoft have done good work with this and the deployment doesn’t seem too far removed from the 2012 R2 guides I’ve been looking at online. We added each server to the roles as per above, got to the final screen and hit the magic Deploy button then…

"Unable to install RD Web Access role service on server"

Role service... Failed
Deployment... Cancelled

Well that didn’t go to plan! We had a look online, trying to find reasons for the failures and went through some initial troubleshooting to make sure all recent updates were installed and each server’s patches matched exactly, also enabled Powershell remoting…

Enable-PSRemoting -force

…still no joy until we found this little nugget of information…

Ref: https://social.technet.microsoft.com/Forums/Sharepoint/en-US/b5c2bae3-0e3b-4d22-b64d-a51d27f0b0e4/deploying-rds-2012-r2-unable-to-install-rd-web-access-role-service-on-server?forum=winserverTS

So it appears the RD Gateway \ RD Web Access role isn’t supported on Server Core. Of course we wouldn’t want the web-facing part of the deployment running on a server with reduced attack surface would we Microsoft… not impressed!

Ref: https://technet.microsoft.com/en-us/library/jj574158(v=ws.11).aspx

To confirm the hypothesis running Get-WindowsFeature on Server 2016 Core gives this…

server-core-available-rds-roles
Server Core

and on Server 2016 with GUI gives this…

server-gui-available-rds-roles
Server with GUI

Published names & certificate fun and games

After begrudgingly re-installing one of the VMs with a GUI (seemed quicker than trying to convert the Core install) we managed to get past the final Deploy page with 3 success bars 🙂

The first key setting we were asked for was the external FQDN for the RD Gateway, which was added to our ISP-hosted DNS records. We use a wildcard certificate to cover our external facing SSL needs, nothing out the ordinary there and went on to apply it to each of the four roles specified by the RDS Deployment wizard. A Session Collection was created for a test group and pointed at the new Session Host. All looking promising.

The RD Gateway FQDN naming in itself wasn’t a problem but led us to an interesting part of the setup relating to SSL certificates and domains. Once we had the RDS services accessible from outside the network (see below) I fired up my 4G tethering to give it a test.

The connection worked but threw up a certificate warning and it was obvious to see why. Our wildcard certificate is for *.domain.ac.uk but the Connection Broker’s published FQDN is servername.subdomain.domain.ac.uk and therefore isn’t covered.

Fortunately a Powershell script called Set-RDPublishedName exists to change this published name and works a treat! Grab it from https://gallery.technet.microsoft.com/Change-published-FQDN-for-2a029b80

You’ll also need to ensure that you can access the new published name internally, depending on what form your internal domain is vs. your external you may need to do a bit of DNS trickery with zones to get the records you need. More on that can be found at:

Ref: https://msfreaks.wordpress.com/2013/12/09/windows-2012-r2-remote-desktop-services-part-1
Ref: https://msfreaks.wordpress.com/2013/12/23/windows-2012-r2-remote-desktop-services-part-2

set-rdpublishedname
Set-RDPublishedName script in action

External access via Azure AD Application Proxy

We published the RD Gateway and RD Web Access via our new shiny Azure AD Application Proxy for a few reasons…

  • simplicity, no firewall rules or DMZ required
  • security, leverages Azure to provide the secure tunnel
  • SSO, use Kerberos Delegation to sign into RD Web Access as part of the user’s Office 365 login

I followed the excellent guides from Arjan Vroege’s blog for this, in particular the section regarding how to edit the RD Web Access webpage files… nice work Arjan!

Publish your RDS Environment with Azure and AD Proxy – Part 1 – http://www.vroege.biz/?p=2462
Publish your RDS Environment with Azure and AD Proxy – Part 2 – http://www.vroege.biz/?p=2563
Publish your RDS Environment with Azure and AD Proxy – Part 3 – http://www.vroege.biz/?p=2647

As per my previous post on Azure AD Application Proxy & Kerberos delegation use the command below to add the SPN record (replace the FQDN and server name as appropriate)

setspn -s HTTP/servername.subdomain.domain.ac.uk servername

When done the end result is a seamless login to RD Web Access via the Azure AD login page. In our case the link will eventually end up as a button on our Office 365-based Staff Intranet, therefore not requiring any further logins to get to the RDWeb app selection screen.

I particularly wanted to avoid the RDWeb login screen, which I’m amazed in 2017 still requires DIY hacks to avoid the requirement to login with the DOMAIN\username format. Thought Microsoft would’ve improved that in the Server 2016 release but evidently not.

One more gotcha

So having done all the hard work above preparing the login all that was left was to click the Remote Desktop icon and enjoy, right? Wrong.

After running the Set-RDPublishedName script the certificate warning went away and I could see the change to the new wildcard-friendly name, however the connection attempt now failed with the error “Remote Desktop can’t connect to the remote computer *connectionbrokername* for one of these reasons”

remote-desktop-cant-connect
connection failure after changing Published Name

Neither explanation made any sense as the connection was working perfectly fine until changing the Published Name. Indeed changing it back to the original FQDN of the Connection Broker restored service so it had to be something to do with that. After being stumped initially I came back after food (always helps!) then after a bit more research found this very helpful post:

Ref: https://social.technet.microsoft.com/Forums/windowsserver/en-US/4fa952bc-6842-437f-8394-281823b0e7ad/change-published-fqdn-for-2012-r2-rds?forum=winserverTS

It turns out the new FQDN we added when changing the Published Name needs to be added to RDG_RDAllConnectionBrokers Local Computer Group.

This group is used to approve connections in the Resource Authorization Policies (RD-RAP) section of RD Gateway Manager. By default only the server’s domain FQDN is present in the list (as you’d expect) so it appears unless you add the new Published Name in there the connection attempt gets denied.

To add your external published name follow these steps:

  • Server Manager > Tools > Remote Desktop Services > Remote Desktop Gateway Manager
  • expand your RD Gateway server > Policies > Resource Authorization Policies
  • Click Manage Local Computer Groups on the right hand pane
  • Select RDG_RDConnectionBrokers > Properties
  • Click the Network Resources tab
  • type the FQDN of the Published Name you supplied to the Powershell script earlier then click Add
  • OK all the way out then try your connection again

manage-locally-stored-computer-groups
RD Gateway Manager

The example below replaces the real server names with dummy entries but should illustrate the concept. The same scenario applies if your servers exist in a .local Active Directory domain (which will be the top entry) and your external domain is something different (again remember to sort out internal DNS zone entries to suit)

add-external-name-to-rdcbcomputers-group
Manage RDG_RDCBComputers group

Finishing touches

Once all the above is done you should then get a connection, there is one seemingly unavoidable credential prompt due to Microsoft persisting with using an ActiveX control to start the RDP session but perhaps one day they’ll update it (we live in hope). It seems you can use the UPN style format here which is handy as it keeps things consistent. In a way it’s a bit of a security measure so not the end of the world.

Now the connection itself is sorted out all that’s left is to tweak the Session Host to our requirements. This guide gives some nice pointers on locking down the server via GPO:

Ref: http://www.it.ltsoy.com/windows/lock-down-remote-desktop-services-server-2012

We also push out a custom Start Menu using the newer Windows 10 1607 GPO settings along with the Export-StartLayout command. Finally install any programs required, remember to change the mode of the server first:

Ref: https://technet.microsoft.com/en-us/library/ff432698.aspx

change user /install

Then once done

change user /execute

Now enjoy 🙂

rds-screenshot
Connection to Server 2016 RDS Session Based desktop via RD Web Access \ RD Gateway