Deep Dive into Azure AD Domain Services – Part 3

Posted on February 28, 2018 by mattfeltonma

Well folks, it’s time to wrap up this series on Azure Active Directory Domain Services (AAD DS). In my first post I covered the basic configurations of the managed domain and in my second post took a look at how well Microsoft did in applying security best practices and complying with NIST standards. In this post I’m going to briefly cover the LDAPS over the Internet capability, summarize some key findings, and list out some improvements I’d like to see made to the service.

One of the odd features Microsoft provides with the AAD DS service is the ability to expose the managed domain over LDAPS to the Internet. I really am lost as to the use case that drove the feature. LDAP is very much a legacy on-premises protocol that has no place being exposed to risks of the public Internet. It’s the last thing that should the industry should be encouraging. Just because you can, doesn’t mean you should. Now let me step off the soap box and let’s take a look at the feature.

As I covered in my last post LDAPS is not natively enabled in the managed domain. The feature must be configured and enabled through the Azure Portal. The configuration consists of uploading the private key and certificate the service will use in the form of a PKCS12 file (*.PFX). The certificate has a few requirements that are outlined in the instructions above. After the certificate is validated, it takes about 10-15 minutes for the service to become available. Beyond enabling the service within the VNet, you additionally have the option to expose the LDAPS endpoint to the Internet.

Microsoft provides instructions on how to restrict access to the endpoint to trusted IPs via a network security group (NSG) because yeah, exposing an LDAP endpoint to the Internet is just a tad risky. To lock it down you simply associate an NSG with the subnet AAD DS is serving. Once that is done enable the service via the option in the image above and wait about 10 minutes. After the service is up, register a external DNS record for the service that points to the IP address noted under the properties section of the AADS blade and you’re good to go.

For my testing purposes, I locked the external LDAPS endpoint down to the public IP address my Azure VM was SNATed to. After that I created an entry in the host file of the VM that matched the external DNS name I gave the service (whyldap.geekintheweeds.com) to the public IP address of the LDAPS endpoint in order to bypass the split-brain DNS challenge. Initiating a connection from LDP.EXE was a success.

Now that we know the service is running, let’s check out what the protocol support and cipher suite looks like.

Again we see the use of deprecated cipher suites. Here the risk is that much greater since a small mistake with an NSG could expose this endpoint directly to the Internet. If you’re going to use this feature, please just don’t. If you’re really determined to, don’t screw up your NSGs.

This series was probably one of the more enjoyable series I’ve done since I knew very little about the AAD DS offering. There were a few key takeaways that are worth sharing:

The more objects in the directory, the more expensive the service.
Users and groups can be created directly in managed domain after a new organizational unit is created.
Password and lockout policy is insanely loose to the point where I can create an account with a three character password (just need to meet complexity requirements) and accounts never lockout. The policy cannot be changed.
RC4 encryption ciphers are enabled and cannot be disabled.
NTLMv1 is enabled and cannot be disabled.
The service does not support smart-card enforced users. Yes, that includes both the users synchronized from Azure AD as well as any users you create directly in the managed domain. If I had to guess, it’s probably due to the fact that you’re not a Domain Admin so hence you can’t add to the NTAuth certificate store.
LDAPS is not enabled by default.
Schema extensions are not supported.
Account-Based Kerberos Delegation is not supported.
If you are syncing identities to Azure AD, you’ll also need to synchronize your passwords.
The managed domain is very much “out of the box” defaults.
Microsoft creates a “god” account which is a permanent member of every privileged group in the forest
Recovery of deleted objects created directly in the managed domain is not possible. The rights have not been delegated to the AADC Administrator.
The service does not allow for Active Directory trusts
SIDHistory attribute of users and groups sourced from Azure AD is populated with Primary Group from on-premises domain

My verdict on AAD DS is it’s not a very useful service in its current state. Beyond small organizations, organizations that have very little to no requirements on legacy infrastructure, organizations that don’t have strong security requirements, and dev/qa purposes I don’t see much of a use for it right now. It comes off as a service in its infancy that has a lot of room to grow and mature. Microsoft has gone a bit too far in the standardization/simplicity direction and needs to shift a bit in the opposite direction by allowing for more customization, especially in regards to security.

I’d really like to see Microsoft introduce the capabilities below. All of them should exposed via the resource blade in the Azure Portal if at all possible. It would provide a singular administration point (which seems to be the strategy given the move of Azure AD and Intune to the Azure Portal) and would allow Microsoft to control how the options are enabled in the managed domain. This means no more administrators blowing up their Active Directory forest because they accidentally shut off all the supported cipher suites for Kerberos.

Expose Domain Controller Event Logs to Azure Portal/Graph API and add support for AAD DS Power BI Dashboards
Support for Active Directory trusts
Out of the box provide a Red Forest model (get rid of that “god” account)
Option to disable risky cipher suites for both Kerberos and LDAPS
Option to harden the password and lockout policy
Option to disable NTLMv1
Option to turn on LDAP Debug Logging
Option to direct Domain Controller event logs to a SIEM
Option to restore deleted users and groups that were created directly in the managed domain. If you’re allow creation, you need to allow for restoration.
Removal of Internet-accessible LDAPS endpoint feature or at least somehow incorporate the NSG lockdown feature directly into the AAD DS blade.

While the service has a lot of room for improvement the direction of a managed Windows AD offering is spot on. In the year 2018, there is no reason Windows AD shouldn’t be offered as a managed service. The direction Microsoft has gone by sourcing the identities and credentials from Azure AD is especially creative. It’s a solid step in the direction of creating a singular centralized identity service that provides both legacy and modern protocols. I’ll be watching this service closely as Microsoft builds upon it for the next few months.

Thanks and see you next post!

Deep Dive into Azure AD Domain Services – Part 2

Posted on February 25, 2018 by mattfeltonma

Welcome back to part 2 of my series on Microsoft’s managed services offering of Azure Active Directory Domain Services (AAD DS). In my first post I covered so some of the basic configuration settings of the a default service instance. In this post I’m going to dig a bit deeper and look at network flows, what type of secure tunnels are available for LDAPS, and examine the authentication protocols and supporting cipher suites are configured for the service.

To perform these tests I leveraged a few different tools. For a port scanner I used Zenmap. To examine the protocols and cipher suites supported by the LDAPS service I used a custom openssl binary running on an Ubuntu VM in Azure. For examination of the authentication protocol support I used Samba’s smbclient running on the Ubuntu VM in combination with WinSCP for file transfer, tcpdump for packet capture, and WireShark for packet analysis.

Let’s start off with examining the open ports since it takes the least amount of effort. To do that I start up Zenmap and set the target to one of the domain controllers (DCs) IP addresses, choose the intense profile (why not?), and hit scan. Once the scan is complete the results are displayed.

Navigating to the Ports / Hosts tab displays the open ports. All but one of them are straight out of the standard required ports you’d see open on a Windows Server functioning as an Active Directory DC. An opened port 443 deserves more investigation.

Let’s start with the obvious and attempt to hit the IP over an HTTPS connection but no luck there.

Let’s break out Fiddler and hit it again. If we look at the first session where we build the secure tunnel to the website we see some of the details for the certificate being used to secure the session. Opening the TextView tab of the response shows a Subject of CN=DCaaS Fleet Dc Identity Cert – 0593c62a-e713-4e56-a1be-0ef78f1a2793. Domain Controller-as-a-Service, I like it Microsoft. Additionally Fiddler identifies the web platform as the Microsoft HTTP Server API (HTTP.SYS). Microsoft has been doing a lot more that API since it’s much more lightweight than IIS. I wanted to take a closer look at the certificate so I opened the website in Dev mode in Chrome and exported it. The EKUs are normal for a standard use certificate and it’s self-signed and untrusted on my system. The fact that the certificate is untrusted and Microsoft isn’t rolling it out to domain-joined members tells me whatever service is running on the port isn’t for my consumption.

So what’s running on that port? I have no idea. The use of the HTTP Server API and a self-signed certificate with a subject specific to the managed domain service tells me it’s providing access to some type of internal management service Microsoft is using to orchestrate the managed domain controllers. If anyone has more info on this, I’d love to hear it.

Let’s now take a look at how Microsoft did at securing LDAPS connectivity to the managed domain. LDAPS is not enabled by default in the managed domain and needs to be configured through the Azure AD Domain Services blade per these instructions. Oddly enough Microsoft provides an option to expose LDAPS over the Internet. Why any sane human being would ever do this, I don’t know but we’ll cover that in a later post.

I wanted to test SSLv3 and up and I didn’t want to spend time manipulating registry entries on a Windows client so I decided to spin up an Ubuntu Server 17.10 VM in Azure. While the Ubuntu VM was spinning up, I created a certificate to be used for LDAPS using the PowerShell command referenced in the Microsoft article and enabled LDAPS through the Azure AD Domain Services resource in the Azure Portal. I did not enable LDAPS for the Internet for these initial tests.

After adding the certificate used by LDAPS to the trusted certificate store on the Windows Server, I opened LDP.EXE and tried establishing LDAPS connection over port 636 and we get a successful connection.

Once I verified the managed domain was now supporting LDAPS connections I switched over to the Ubuntu box via an SSH session. Ubuntu removed SSLv3 support in the OpenSSL binary that comes pre-packaged with Ubuntu so to test it I needed to build another OpenSSL binary. Thankfully some kind soul out there on the Interwebz documented how to do exactly that without overwriting the existing version. Before I could build a new binary I had to add re-install the Make package and add the Gnu Compiler Collection (GCC) package using the two commands below.

sudo apt-get install –reinstall make
sudo apt-get install gcc

After the two packages were installed I built the new binary using the instructions in the link, tested the command, and validated the binary now includes SSLv3.

After Poodle hit the news back in 2014, Microsoft along with the rest of the tech industry advised SSLv3 be disabled. Thankfully this basic well known vulnerability has been covered and SSLv3 is disabled.

SSLv3 is disabled, but what about TLS 1.0, 1.1, and 1.2? How about the cipher suites? Are they aligned with NIST guidance? To test that I used a tool named TestSSLServer by Thomas Pornin. It’s a simple command line tool which makes cycling through the available cipher suites quick and easy.

The options I chose perform the following actions:

-all -> Perform an “exhaustive” search across cipher suites
-t 1 -> Space out the connections by one second
-min tlsv1 -> Start with TLSv1

The command produces the output below.

TLSv1.0:
server selection: enforce server preferences
3f- (key: RSA) ECDHE_RSA_WITH_AES_256_CBC_SHA
3f- (key: RSA) ECDHE_RSA_WITH_AES_128_CBC_SHA
3f- (key: RSA) DHE_RSA_WITH_AES_256_CBC_SHA
3f- (key: RSA) DHE_RSA_WITH_AES_128_CBC_SHA
3– (key: RSA) RSA_WITH_AES_256_CBC_SHA
3– (key: RSA) RSA_WITH_AES_128_CBC_SHA
3– (key: RSA) RSA_WITH_3DES_EDE_CBC_SHA
3– (key: RSA) RSA_WITH_RC4_128_SHA
3– (key: RSA) RSA_WITH_RC4_128_MD5
TLSv1.1: idem
TLSv1.2:
server selection: enforce server preferences
3f- (key: RSA) ECDHE_RSA_WITH_AES_256_CBC_SHA384
3f- (key: RSA) ECDHE_RSA_WITH_AES_128_CBC_SHA256
3f- (key: RSA) ECDHE_RSA_WITH_AES_256_CBC_SHA
3f- (key: RSA) ECDHE_RSA_WITH_AES_128_CBC_SHA
3f- (key: RSA) DHE_RSA_WITH_AES_256_GCM_SHA384
3f- (key: RSA) DHE_RSA_WITH_AES_128_GCM_SHA256
3f- (key: RSA) DHE_RSA_WITH_AES_256_CBC_SHA
3f- (key: RSA) DHE_RSA_WITH_AES_128_CBC_SHA
3– (key: RSA) RSA_WITH_AES_256_GCM_SHA384
3– (key: RSA) RSA_WITH_AES_128_GCM_SHA256
3– (key: RSA) RSA_WITH_AES_256_CBC_SHA256
3– (key: RSA) RSA_WITH_AES_128_CBC_SHA256
3– (key: RSA) RSA_WITH_AES_256_CBC_SHA
3– (key: RSA) RSA_WITH_AES_128_CBC_SHA
3– (key: RSA) RSA_WITH_3DES_EDE_CBC_SHA
3– (key: RSA) RSA_WITH_RC4_128_SHA
3– (key: RSA) RSA_WITH_RC4_128_MD5

As can be seen from the bolded output above, Microsoft is still supporting the RC4 cipher suites in the managed domain. RC4 has been known to be a vulnerable algorithm for years now and it’s disappointing to see it still supported especially since I haven’t seen any options available to disable within the managed domain. While 3DES still has a fair amount of usage, there have been documented vulnerabilities and NIST plans to disallow it for TLS in the near future. While commercial customers may be more willing to deal with the continued use of these algorithms, government entities will not.

Let’s now jump over to Kerberos and check out what cipher suites are supported by the managed DC. For that we pull up ADUC and check the msDS-SupportedEncryptionTypes attribute of the DC’s computer object. The attribute is set to a value of 28, which is the default for Windows Server 2012 R2 DCs. In ADUC we can see that this value translates to support of the following algorithms:

• RC4_HMAC_MD5
• AES128_CTS_HMAC_SHA1
• AES256_CTS_HMAC_SHA1_96

Again we see more support for RC4 which should be a big no no in the year 2018. This is a risk that orgs using AAD DS will need to live with unless Microsoft adds some options to harden the managed DCs.

Last but not least I was curious if Microsoft had support for NTLMv1. By default Windows Server 2012 R2 supports NTLMv1 due to requirements for backwards compatibility. Microsoft has long recommended disabling NTLMv1 due to the documented issues with the security of the protocol. So has Microsoft followed their own advice in the AAD DS environment?

To check this I’m going use Samba’s smbclient package on the Ubuntu VM. I’ll use smbclient to connect to the DC’s share from the Ubuntu box using the NTLM protocol. Samba has enforced the use NTLMV2 in smbclient by default so I needed to make some modifications to the global section of the smb.conf file by adding client ntlmv2 auth = no. This option disables NTLMv2 on smbclient and will force it to use NTLMv1.

After saving the changes to smb.conf I exit back to the terminal and try opening a connection with smbclient. The options I used do the following:

-L -> List the shares on my DC’s IP address
-U -> My domain user name
-m -> Use the SMB2 protocol

While I ran the command I also did a packet capture using tcpdump which I moved over to my Windows box using WinSCP. I then opened the capture with WireShark and navigated to the packet containing the Session Setup Request. In the parsed capture we don’t see an NTLMv2 Response which means NTLMv1 was used to authenticate to the domain controller indicating NTLMv1 is supported by the managed domain controllers.

Based upon what I’ve observed from poking around and running these tests I’m fairly confident Microsoft is using a very out-of-the-box configuration for the managed Windows Active Directory domain. There doesn’t seem to be much of an attempt to harden the domain against some of the older and well known risks. I don’t anticipate this offering being very appealing to organizations with strong security requirements. Microsoft should look to offer some hardening options that would be configurable through the Azure Portal. Those hardening options are going to need to include some type of access to the logs like I mentioned in my last post. Anyone who has tried to rid their network of insecure cipher suites or older authentication protocols knows the importance of access to the domain controller logs to the success of that type of effort.

My next post will be the final post in this series. I’ll cover the option Microsoft provides to expose LDAPS to the Internet (WHY OH WHY WOULD YOU DO THAT?), summarize my findings, and mention a few other interesting things I came across during the study for this series.

Thanks!

Deep Dive into Azure AD Domain Services – Part 1

Posted on February 19, 2018 by mattfeltonma

Hi everyone. In this series of posts I’ll be doing a deep dive into Microsoft’s Azure AD Domain Services (AAD DS). AAD DS is Microsoft’s managed Windows Active Directory service offered in Microsoft Azure Infrastructure-as-a-Service intended to compete with similar offerings such as Amazon Web Services’s (AWS) Microsoft Active Directory. Microsoft’s solution differs from other offerings in that it sources its user and group information from Azure Active Directory versus an on-premises Windows Active Directory or LDAP.

Like its competitors Microsoft realizes there are still a lot of organizations out there who are still very much attached to legacy on-premises protocols such as NTLM, Kerberos, and LDAP. Not every organization (unfortunately) is ready or able to evolve its applications to consume SAML, Open ID Connect, OAuth, and Rest-Based APIs (yes COTS vendors I’m talking to you and your continued reliance on LDAP authentication in the year 2018). If the service has to be there, it makes sense to consume a managed service so staff can focus less on maintaining legacy technology like Windows Active Directory and focus more on a modern Identity-as-a-Service (IDaaS), Software-as-a-Service (SaaS), and Platform-as-a-Service (Paas) solutions.

Sounds great right? Sure, but how does it work? Microsoft’s documentation does a reasonable job giving the high level details of the service so I encourage you to read through it at some point. I won’t be covering information included in that documentation unless I notice a discrepancy or an area that could use more detail. Instead, I’m going to focus on the areas which I feel are important to understand if you’re going to attempt to consume the service in the same way you would a traditional on-premises Windows Active Directory.

With that introduction, let’s dig in.

The first thing I did was to install the Remote Server Administration Tools (RSAT) for Active Directory Domain Services and Group Policy Management tools. I used these tools to explore some of the configuration choices Microsoft made in the managed service. I also installed Microsoft Network Monitor 3.4 to review packet captures captured using the netsh.

After the tools were installed I started a persistent network capture using netsh using an elevated command prompt. This is an incredibly useful feature of Windows when you need to debug issues that occur prior or during user or system logon. I’ve used this for years to troubleshoot a number of Windows Active Directory issues including slow logons and failed logons. The only downfall of this is you’re forced into using Microsoft Network Monitor or Microsoft Message Analyzer to review the packet captures it creates. While Microsoft Message Analyzer is a sleek tool, the resources required to run it effectively are typically a non-starter for a lab or traditional work laptop so I tend to use Network Monitor.

After the packet capture was started I went through the standard process of joining the machine to the domain and rebooting the computer. After reboot, I logged in an account in the AAD DC Administrators Azure Active Directory group, started an elevated command prompt as the VM’s local administrator and stopped the packet capture. This provided me with a capture of the domain join, initial computer authentication, and initial user authentication.

While I know you’re as eager to dig into the packet capture as I am, I’ll cover that in a future post. Instead I decided to break out the RSAT tools and poke around at configuration choices an administrator would normally make when building out a Windows Active Directory domain.

Let’s first open the tool everyone who touches Windows Active Directory is familiar with, Active Directory Users and Computers (ADUC). The data layout (with Advanced Features option on) for organizational units (OUs) and containers looks very similar to what we’re used to seeing with the exception of the AADC Computers, AADC Users, AADDSDomainAdmin OUs, and AADDSDomainConfig container. I’ll get into these containers in a minute.

If we right-click the domain node and go to properties we see that the domain and forest are running in Windows Server 2012 R2 domain and forest functional level with no trusts defined. Examining the operating system tab of the two domain controllers in the Domain Controllers OU shows that both boxes run Windows Server 2012 R2. Interesting that Microsoft chose not to use Windows Server 2016.

Navigating to the Security tab and clicking the Advanced button shows that the AAD DC Administrator group has only been granted the Create Organizational Unit objects permission while the AAD DC Service Accounts group has been granted Replicating Directory Changes. As you can see from these permissions the base of the directory tree is very locked down.

Let me circle back to the OUs and Containers I talked about above. The AADDC Computers and AADDC Users OUs are the default OUs Microsoft creates for you. Newly joined machines are added to the AADDC Computers OU and users synchronized from the Azure AD tenant are placed in the AADDC Users OU. As we saw from the permissions above, we could use an account in the AAD DC Administrators group to create additional OUs under the domain node to delegate control to another set of more restricted admins, for the purposes of controlling GPOs if security filtering doesn’t meet our requirements, or for creating additional service accounts or groups for the workloads we deploy in the environment. The permissions within the default OUs are very limited. In the AADDC Computer OU GPOs can be applied and computer objects can be added and removed. In the AADC Users OU only GPOs can be applied which makes sense considering the user and group objects stored there are sourced from your authoritative Azure AD tenant.

The AADDSDomainAdmin OU contains a single security group named AADDS Service Administrators Group (pre-Windows 2000 name of AADDSDomAdmGroup). The group contains a single member names dcaasadmin which is the renamed built-in Active Directory administrator account. The group is nested into a number of highly privileged built-in Active Directory groups including Administrators, Domain Admins, Domain Users, Enterprise Admins, and Schema Admins. I’m very uncomfortable with Microsoft’s choice to make a “god” group and even a “god” user of the built-in administrator. This directly conflicts with security best practices for Active Directory which would see no account being a permanent member of these highly privileged groups or at the least divvying up the privileges among separate security principals. I would have liked to see Microsoft leverage a Red Forest Red Forest design here. Hopefully we’ll see some improvements as the service matures. I’m unsure as to the purpose of the OU and this group at this time.

The AADDSDomainConfig container contains a single container object named SchemaUpdate. I reviewed the attributes of both containers hoping to glean some idea of the purpose of the containers and the only thing I saw of notice was the revision attribute was set to 2. Maybe Microsoft is tracking the schema of their standard managed domain image via this attribute? In a future post in this series I’ll do a comparison of this managed domain’s schema with a fresh Windows Server 2012 R2 schema.

Opening Active Directory Sites and Services shows that Microsoft has chosen to leave the domain with a single site. This design choice makes sense given that a limitation of AAD DS is that it can only serve a single region. If that limitation is ever lifted, Microsoft will need to revisit this choice and perhaps include a site for each region. Expanding the Default-First-Site-Name site and the Servers node shows the two domain controllers Microsoft is using to provide the Windows Active Directory service to the VNet.

So the layout is simple, what about the group policy objects (GPO)? Opening up the Group Policy Management Console displays five GPOs which are included in every managed domain.

The AADDC Users GPO is empty of settings while the AADC Computers GPO has a single Preference defined that adds the AAD DC Administrators group to the built-in Administrators group on any member servers added to the OU. The Default Domain Controllers Policy (DDCP) GPO is your standard out of the box DDCP with nothing special set. The Default Domain Policy (DDP) GPO on the other hand has a number of settings applied. The password policy is interesting… I get that you have the option to source all the user accounts within your AAD DS domain from Azure AD, but Microsoft is still giving you the ability to create user accounts in the managed domain as I covered above which makes me uncomfortable with the default password policy. Microsoft hasn’t delegated the ability to create Fine Grained Password Policies (FGPPs) either, which means you’re stuck with this very lax password policy. Given the lack of technical enforcement, I’d recommend avoiding creating user accounts directly in the managed domain for any purpose until Microsoft delegates the ability to create FGPPs. The remaining settings in the policy are standard out of the box DDP.

The GPO named Event Log GPO is linked to the Domain Controllers OU and executes a startup script named EventLogRetentionPolicy.PS1. Being the nosy geek I am, I dug through SYSVOL to find the script. The script is very simple in that it sets each event log to overwrite events over 31 days old. It then verifies the results and prints the results to the console. Event logs are an interesting beast in AAD DS. An account in the AAD DC Administrators group doesn’t have the right to connect to the Event Logs on the DCs remotely and I haven’t come across any options to view those logs. I don’t see any mention of them in the Microsoft documentation, so my assumption is you don’t get access to them at this time. I have to imagine this is a show stopper for some organizations considering the critical importance of Domain Controller logs. If anyone knows how to access these logs, please let me know. I’d like to see Microsoft incorporate an option to send the logs to a syslog agent via a configuration option in the Azure AD Domain Services blade in the Azure Portal.

I’m going to stop here today. In my next post I’ll do some poking around by running a port scan against the managed domain controllers to see what network flows are open, enable LDAPS to see what the SSL/TLS landscape looks like, and examine authentication protocols and algorithms supported (NTLMv1,v2, Kerberos DES, etc). Thanks for reading!

Integrating Azure AD and G-Suite – Automated Provisioning

Posted on February 13, 2018 by mattfeltonma

Today I’ll wrap up my series on Azure Active Directory’s (Azure AD) integration with Google’s G-Suite. In my first entry I covered the single-sign on (SSO) integration and in my second and third posts I gave an overview of Google’s Cloud Platform (GCP) and demonstrated how to access a G-Suite domain’s resources through Google’s APIs. In this post I’m going to cover how Microsoft provides automated provisioning of user, groups, and contacts . If you haven’t read through my posts on Google’s API (part 1, part 2) take a read through so you’re more familiar with the concepts I’ll be covering throughout this post.

SSO using SAML or Open ID Connect is a common capability of most every cloud solutions these days. While that solves the authentication problem, the provisioning of users, groups, and other identity-relates objects remains a challenge largely due to the lack of widely accepted standards (SCIM has a ways to go folks). Vendors have a variety of workarounds including making LDAP calls back to a traditional on-premises directory (YUCK), supporting uploads of CSV files, or creating and updating identities in its local databases based upon the information contained in a SAML assertion or Open ID Connect id token. A growing number of vendors are exposing these capabilities via a web-based API. Google falls into this category and provides a robust selection of APIs to interact with its services from Gmail to resources within Google Cloud Platform, and yes even Google G-Suite.

If you’re a frequent user of Azure AD, you’ll have run into the automatic provisioning capabilities it brings to the table across a wide range of cloud services. In a previous series I covered its provisioning capabilities with Amazon Web Services. This is another use case where Microsoft leverages a third party’s robust API to simplify the identity management lifecycle.

In the SSO Quickstart Guide Microsoft provides for G-Suite it erroneously states:

“Google Apps supports auto provisioning, which is by default enabled. There is no action for you in this section. If a user doesn’t already exist in Google Apps Software, a new one is created when you attempt to access Google Apps Software.”

This simply isn’t true. While auto provisioning via the API can be done, it is a feature you need to code to and isn’t enabled by default. When you enable SSO to G-Suite and attempt to access it using an assertion containing the claim for a user that does not exist within a G-Suite domain you receive the error below.

This establishes what we already knew in that identities representing our users attempting SSO to G-Suite need to be created before the users can authenticate. Microsoft provides a Quickstart for auto provisioning into G-Suite. The document does a good job telling you were to click and giving some basic advice but really lacks in the detail into what’s happening in the background and describing how it works.

Let’s take a deeper look shall we?

If you haven’t already, add the Google Apps application from the Azure AD Application Gallery. Once the application is added navigate to the blade for the application and select the Provisioning page. Switch the provisioning mode from manual to automatic.

Right off the bat we see a big blue Authorize button which tells us that Microsoft is not using the service accounts pattern for accessing the Google API. Google’s recommendation is to use the service account pattern when accessing project-based data rather than user specific data. The argument can be made that G-Suite data doesn’t fall under project-based data and the service account credential doesn’t make sense. Additionally using a service account would require granting the account domain-wide delegation for the G-Suite domain allowing the account to impersonate any user in the G-Suite domain. Not really ideal, especially from an auditing perspective.

By using the Server-side Web Apps pattern a new user in G-Suite can be created and assigned as the “Azure AD account”. The downfall with of this means you’re stuck paying Google $10.00 a month for a non-human account. The price of good security practices I guess.

Microsoft documentation states that the account must be granted the Super Admin role. I found this surprising since you’re effectively giving the account god rights to your G-Suite domain. It got me wondering what authorization scopes is Microsoft asking for? Let’s break out Fiddler and walk through the process that kicks off after clicking on the Authorization button.

A new window pops up from Google requesting me to authenticate. Here Azure AD, acting as the OAuth client, has made an authorization request and has sent me along with the request over to the Google which is acting as the authorization server to authenticate, consent to the access, and take the next step in the authorization flow.

When I switch over to Fiddler I see a number of sessions have been captured. Opening the WebForms window of the first session to accounts.google.com a number of parameters that were passed to Google.

The first parameter gives us the three authorization scopes Azure AD is looking for. The admin.directory.group and admin.directory.user are scopes are both related to the Google Directory API, which makes sense if it wants to manage users and groups. The /m8/feeds scope grants it access to manage contacts via the Google Contacts API. This is an older API that uses XML instead of JSON to exchange information and looks like it has been/is being replaced by the Google People API.

Management of contacts via this API is where the requirement for an account in the Super Admin role originates. Google documentation states that management of domain shared contacts via the /m8/feeds API requires an administrator username and password for Google Apps. I couldn’t find any privilege in G-Suite which could be added to a custom Admin role that mentioned contacts. Given Google’s own documentation along the lack of an obvious privilege option, this may be a hard limitation of G-Suite. Too bad too because there are options for both Users and Groups. Either way, the request for this authorization scope drives the requirement for Super Admin for the account Azure AD will be using for delegated access.

The redirect_uri is the where Google sends the user after the authorization request is complete. The response_type tells us Azure AD and Google are using the OAuth authorization code grant type flow. The client_id is the unique identifier Google has assigned to Azure AD in whatever project Microsoft has it built in. The approval_prompt setting of force tells Google to display the consent window and the data Azure AD wants to access. Lastly, the access_type setting of offline allows Azure AD to access the APIs without the user being available to authenticate via a refresh token which will be issued along with the access token. Let’s pay attention to that one once the consent screen pops up.

I plug in valid super user credentials to my G-Suite domain and authenticate and receive the warning below. This indicates that Microsoft has been naughty and hasn’t had their application reviewed by Google. This was made a requirement back in July of 2017… so yeah… Microsoft maybe get on that?

To progress to the consent screen I hit the Advanced link in the lower left and opt to continue. The consent window then pops up.

Here I see that Microsoft has registered their application with a friendly name of azure.com. I’m also shown the scopes that the application wants to access which jive with the authorization scopes we saw in Fiddler. Remember that offline access Microsoft asked for? See it mentioned anywhere in this consent page that I’m delegating this access to Microsoft perpetually as long as they ask for a refresh token? This is one of my problems with OAuth and consent windows like this. It’s entirely too vague as to how long I’m granting the application access to my data or to do things as me. Expect to see this OAuth consent attacks continue to grow in in use moving forward. Why worry about compromising the user’s credentials when I can display a vague consent window and have them grant me access directly to their data? Totally safe.

Hopping back to the window, I click the Allow button and the consent window closes. Looking back at Fiddler I see that I received back an authorization code and posted it back to the reply_uri designated in the original authorization request.

Switching back to the browser window for the Azure Portal the screen updates and the Test Connection button becomes available. Clicking the button initiates a quick check where Azure AD obtains an access token for the scopes it requires unseen to the user. After the successful test I hit the Save button.

Switching to the browser window for the Google Admin Portal let’s take a look at the data that’s been updated for the user I used to authorize Microsoft its access. For that I select the user, go to the Security section and I now see that the Azure Active Directory service is authorized to the contacts, user, and group management scopes.

Switching back to the browser window for the Azure Portal I see some additional options are now available.

The mappings are really interesting and will look familiar to you if you’ve ever done anything with an identity management tool like Microsoft Identity Manager (MIM) or even Azure AD Sync. The user mappings for example show which attributes in Azure AD are used to populate the attributes in G-Suite.

The attributes that have the Delete button grayed out are required by Google in order to provision new user accounts in a G-Suite domain. The options available for deletion are additional data beyond what is required that Microsoft can populate on user accounts it provisions into G-Suite. Selecting the Show advanced options button, allow you to play with the schema Microsoft is using for G-Suite. What I found interesting about this schema is it doesn’t match the resource representation Google provides for the API. It would have been nice to match the two to make it more consumable, but they’re probably working off values used in the old Google Provisioning API or they don’t envision many people being nerdy enough to poke around the schema.

Next up I move toggle the provisioning status from Off to On and leave the Scope option set to sync only the assigned users and groups.

I then hit the Save button to save the new settings and after a minute my initial synchronization is successful. Now nothing was synchronized, but it shows the credentials correctly allowed Azure AD to hit my G-Suite domain over the appropriate APIs with the appropriate access.

So an empty synchronization works, how about one with a user? I created a new user named dutch.schaefer@geekintheweeds.com with only the required attributes of display name and user principal name populated, assigned the new user to the Google Apps application and give Azure AD a night to run another sync. Earlier tonight I checked the provisioning summary and verified the sync grabbed the new user.

Review of the audit logs for the Google Apps application shows that the new user was exported around 11PM EST last night. If you’re curious the synch between Azure AD and G-Suite occurs about every 20 minutes.

Notice that the FamilyName and GivenName attributes are set to a period. I never set the first or last name attributes of the user in Azure AD, so both attributes are blank. If we bounce back to the attribute mapping and look at the attributes for Google Apps, we see that FamilyName and GivenName are both required meaning Azure AD had to populate them with something. Different schemas, different requirements.

Switching over to the Google Admin Console I see that the new user was successfully provisioned into G-Suite.

Pretty neat overall. Let’s take a look at what we learned:

Azure AD supports single sign-on to G-Suite via SAML using a service provider-initiated flow where Azure AD acts as the identity provider and G-Suite acts as the service provider.
A user object with a login id matching the user’s login id in Azure Active Directory must be created in G-Suite before single sign-on will work.
Google provides a number of libraries for its API and the Google API Explorer should be used for experimentation with Google’s APIs.
Google’s Directory API is used by Azure AD to provision users and groups into a G-Suite domain.
Google’s Contacts API is used by Azure AD to provision contacts into a G-Suite domain.
A user holding the Super Admin role in the G-Suite domain must be used to authorize Azure AD to perform provisioning activities. The Super Admin role is required due to the usage of the Google Contact API.
Azure AD’s authorization request includes offline access using refresh tokens to request additional access tokens to ensure the sync process can be run on a regular basis without requiring re-authorization.
Best practice is to dedicate a user account in your G-Suite domain to Azure AD.
Azure AD uses the Server-side Web pattern for accessing Google’s APIs.
The provisioning process will populate a period for any attribute that is required in G-Suite but does not have a value in the corresponding attribute in Azure AD.
The provisioning process runs a sync every 20 minutes.

Even though my coding is horrendous, I absolutely loved experimenting with the Google API. It’s easy to realize why APIs are becoming so critical to a good solution. With the increased usage of a wide variety of products in a business, being able to plug and play applications is a must. The provisioning aspect Azure AD demonstrates here is a great example of the opportunities provided when critical functionality is exposed for programmatic access.

I hope you enjoyed the series, learned a bit more about both solutions, and got some insight into what’s going on behind the scenes.

Journey Of The Geek

The chronicles of a Bostonian tech geek navigating through life, technology, and general geekiness.

Monthly Archives: February 2018

Deep Dive into Azure AD Domain Services – Part 2

Deep Dive into Azure AD Domain Services – Part 1

Integrating Azure AD and G-Suite – Automated Provisioning