Big-IP and ADFS Part 5 – “Working with ADFS 3.0 and SNI”
Can you believe it? It’s true, it’s true! There’s a part 5. What can I say? Times change; people change; software changes. Active Directory Federation Services, (ADFS) is no exception. While the BIG-IP with SAML 2.0 can alleviate the need for and ADFS infrastructure in many use cases, there are still organizations that need/want to continue utilizing ADFS. Fortunately, regardless of which way you go, F5 can help. So, in the spirit of free will, collaboration, and serving the greater good, (too much?), let’s talk about load balancing ADFS 3.0 with the BIG-IP.
As you may, or may not, recall the previous posts around BIG-IP and ADFS revolved around load balancing ADFS 2.0 and ADFS Proxy, replacing the ADFS Proxy with Access Policy Manager, and replacing the entire ADFS infrastructure with APM and SAML. The good news is that these posts are still relevant with regards to ADFS 3.0 and the ADFS proxy replacement, (WAP); well for the most part anyway.
ADFS and SNI
While there are numerous differences between ADFS 3.0 and previous versions, the most significant change with respect to providing HA and scalability for the ADFS 3.0 infrastructure is its use of Server Name Indication, (SNI). To Successfully integrate a load balancing solution, ( including full reverse proxy), into the ADFS environment the device must support SNI. The load balancing device must be able to present the server name to the backend host as part of the initial Client Hello. Fortunately, the BIG-IP, (ver. 11.1.0 and later) supports this TLS protocol extension.
The rest of this post will provide guidance on enabling SNI support for ADFS 3.0 integration. For overall guidance refer to parts one thru three of this series as well as the recently published ADFS 2.0 Deployment Guide.
SNI and the Server Profile
The BIG-IP provides a virtual server, (listener), that receives client SSL connections and subsequently intelligently passes traffic into a pool of ADFS/WAP servers. Depending upon the organization’s infrastructure and security requirements, the BIG-IP can simply receive encrypted client connections and pass them through to the backend ADFS farm, (aka SSL tunneling). However, the preferred method, (SSL bridging), receives encrypted clients connections; terminates and decrypts the traffic. The traffic is then re-encrypted and sent to the backend application servers. This method adds an additional layer of security since external traffic never directly connects to the internal domain-joined machines as well affording the ability to perform additional deep packet inspection.
SSL bridging back to the ADFS farm requires associating a server SSL profile to the virtual server. Enabling SNI is simply a matter of specifying the server name on the associated server SSL profile, (see below).
1. Navigate to the appropriate profile;
2. Select ‘Advanced’ configuration and enter the FQDN of the backend ADFS service hostname. The hostname will now be provided during the TLS negotiation. In the example below, the server name is ‘fs.f5demo.net’, (refer to the highlighted field). Like I said, simple!
Health Monitoring and SNI
Effectively monitoring the backend ADFS/WAP farm members is a little trickier but very doable. Since the built-in HTTP monitors do not provide the server name as part of the TLS negotiation, using them will result in the being backend servers being incorrectly marked as down, (not good).
You could simply use a non-HTTP monitor, (ICMP being the most common), but that doesn’t provide a reasonable guarantee that the actual ADFS service is functioning. Better than that, what we can do is create an external custom SNI enabled monitor that validates the service metadata and associate it to the pool. It’s as easy as 1,2,3,…..um.. 4,5,6.
1. Download the script: http://www.f5.com/pdf/deployment-guides/sni-eav.zip
2. Upload the previously downloaded file into the BIG-IP via the web interface. Navigate to ‘System’ –> ‘External Monitor Program List’ –> ‘Import’;
IMPORTANT!!!
If ADFS proxy server is configured to accept SSL/TLS connections only using TLSv1.1 or better , the monitor will not work.
If have come up with this one-liner to replace the “curl” based command in the script. Thanks to Jerry Tower for helping fix the actual HTTP request as well as testing the script.
(echo -e "GET $URI HTTP/1.1\r\nHost: $HOST\r\nConnection: Close\r\n\r\n"; sleep 2) | openssl s_client –quiet –servername $HOST -connect $NODE:$PORT 2> /dev/null| grep -i "$RECV" 2>&1> /dev/null
The script line that this one-liner should replace is the following:
curl-apd -k -v -i --resolve $HOST:$PORT:$NODE https://$HOST$URI | grep -i "${RECV}" 2>&1 > /dev/null
3. Browse to and select the file. Provide a name for the file and select ‘Import’;
4. Create a new external monitor utilizing the associate external file. Navigate to ‘Local Traffic’ –> ‘+’ sign;
5. Provide a name and select ‘External’ for the type. Select the previously created external program. The script provided requires three, (3) variables entered as name/value pairs. The variables are listed below. Select ‘Finished’;
Name | Value |
RECV | HTTP/1.1 200 |
URI | /FederationMetadata/2007-06/FederationMetadata.xml |
HOST |
6. Associate the newly created monitor to the ADFS pool and/or the WAP pool. Select ‘Local Traffic’ –> ‘Pools’ –> ‘Pool List’. Move the monitor into the active pane and select ‘Update’.
Additional Links:
Big-IP and ADFS Part 1 – “Load balancing the ADFS Farm”
Big-IP and ADFS Part 2 – “APM–An Alternative to the ADFS Proxy”
Big-IP and ADFS Part 3 – “ADFS, APM, and the Office 365 Thick Clients”
Big-IP and ADFS Part 4 – “What about Single Sign-Out?”
BIG-IP Access Policy Manager (APM) Wiki Home - DevCentral Wiki
- DaleLeBlancNimbostratus
"If ADFS proxy server is configured to accept SSL/TLS connections only using TLSv1.1 or better , the monitor will not work."
My configuration only permits TLSv1.2. Is there a workaround? - TobiasNimbostratus
Unfortunately this does not work with Windows Server 2019 / ADFS 5.0. Any updates how to use a proper monitor for this setup?
- t_vanderkrogtAltostratus
For some reason we keep getting a lot of "EAV exceeded runtime needed to kill..." messages in our logs using this script. The curl command works fine, but the $PIDFILE removal doesn't really seem to work every time.
I already tried to increase the timeouts, but don't really want make them to big.
Anyone any idea what to do about this?
- Aussie_DaveNimbostratus
I had a requirement where we needed to authenticate together with using SNI, I ended up modifying the script to allow input of the username and password into the custom monitor variables. However because this password is stored in plain text I looked at encrypting it using the default RSA key on the F5.
Replace:
curl-apd -k -v --resolve $HOST:$PORT:$NODE https://$HOST$URI 2>&1 >
With:
encrypted_pass= openssl rsautl -inkey /config/httpd/conf/ssl.key/server.key -decrypt
curl-apd -k -v -u "${USER}:${encrypted_pass}" --resolve $HOST:$PORT:$NODE https://$HOST$URI 2>&1 > You will need to create an encrypted file using the below command from F5 ssh: echo "password" | openssl rsautl -inkey /config/httpd/conf/ssl.key/server.key -encrypt >/home/sp2016mon.bin After you have created the file the monitor will then decrypt the password and login using curl. If the default key pair on the F5 device ever expires or changes you will need to rerun this command to re-create the file. If the password changes you will also need to rerun this command to re-create the file. When creating the monitor use the "USER" variable to add your username.
- Peter_Baumann_5Nimbostratus
Very nice stuff, thanks Greg!
Does someone have any idea to solve this without a EAV monitor or get better performance for the monitoring?
- Jad_Tabbara__J1Cirrostratus
Thanks Greg it works well !
- t_vanderkrogtAltostratus
We've slightly changed the curl command in our script in order to have multiple options in the RECV string.
curl-apd -k -i --resolve $HOST:$PORT:$NODE https://$HOST$URI | grep -i -P "${RECV}" > /dev/null 2>&1
We can now work with a receive string as HTTP/1.1\s[2|3]0[0-7]|
- rmd1023Nimbostratus
I think you'll need different scripts. So, for example, you'd have external monitor script "sni_monkey.sh" where the "HOST" command is hardcoded as "monkey.example.com", and another external monitor "sni_banana.sh" where "HOST" is hardcoded as "banana.example.com". The hardcoded URI would be the same in both scripts.
- MattMillson_213Nimbostratus
What is the best way to add multiple HOST entries with the same URI
- Paul_Taylor_110Nimbostratus
rmd1023 - Your comment above was very timely, as I was looking into this today! Thanks for posting it. It looks like variables aren't passed properly in the version I'm running, or something, but with your host specific modifications, it's working great.