Last week I had to setup a small Active Directory Federation Services (ADFS) farm that will be used to allow Single Sign-On (SSO) with Office 365.
Active Directory Federation Services (ADFS) is a solution developed by Microsoft to provide users an authenticated access to applications, that are not capable of using Integrated Windows Authentication (IWA).
Required by the customer was a two node ADFS farm located on the internal network, and a two node ADFS Proxy farm located at the DMZ.
An ADFS Proxyserver acts as a reverse proxy and it is typically located in your organizations perimeter network (DMZ).
This picture shows a typical ADFS/ ADFS Proxy setup:
My customer has decided to use Citrix ADC (former NetScaler) to load balance the requests for the ADFS farm and the ADFS Proxy farm. In addition to load balancing, this offers high availability in case of a failed ADFS server or ADFS Proxy server. Please note that Citrix ADC can act as a ADFS Proxy, but this requires the Advanced Edition license. My customer “only” had a Standard License, so we had to setup dedicated ADFS Proxy servers on the DMZ network.
Citrix ADC setup
The ADFS service name is typically something like adfs.customer.tld. This farm name has to be the same for internal and external access. For internal access, the ADFS service name must be resolved to the VIP of the Citrix ADC. The same applies to external accesss. So you have to setup split DNS.
ADFS uses HTTP and HTTP, so my first attempt was to use this Citrix ADC Content Switch based setup:
add server srv_adfs1 x.x.x.x add server srv_adfs2 x.x.x.y add cs vserver cs_vsrv_adfs SSL x.x.x.x 443 -cltTimeout 180 -caseSensitive OFF add lb vserver lb_vsrv_adfs SSL 0.0.0.0 0 -persistenceType SSLSESSION -cltTimeout 180 add cs action cs_action_adfs -targetLBVserver lb_vsrv_adfs add cs policy cs_pol_adfs -rule "HTTP.REQ.URL.SET_TEXT_MODE(IGNORECASE).CONTAINS(\"adfs.customer.tld\")" -action cs_action_adfs bind cs vserver cs_vsrv_adfsL -policyName cs_pol_adfs -priority 100 add serviceGroup svcgrp_adfs SSL -maxClient 0 -maxReq 0 -cip ENABLED X-MS-Forwarded-Client-IP -usip NO -useproxyport YES -cltTimeout 180 -svrTimeout 360 -CKA NO -TCPB NO -CMP YES -appflowLog DISABLED add lb monitor mon_adfs HTTP-ECV -send "GET /federationmetadata/2007-06/federationmetadata.xml" -recv "adfs.customer.tld/adfs/services/trust" -LRTM ENABLED -secure YES bind serviceGroup svcgrp_adfs srv_gk-adfs1 443 -CustomServerID "\"None\"" bind serviceGroup svcgrp_adfs srv_gk-adfs2 443 -CustomServerID "\"None\"" bind serviceGroup svcgrp_adfs -monitorName mon_adfs bind lb vserver lb_vsrv_adfs svcgrp_adfs bind ssl vserver lb_vsrv_adfs -certkeyName cert-key-pair bind ssl vserver cs_vsrv_adfs -certkeyName cert-key-pair set ssl vserver lb_vsrv_adfs -ssl3 DISABLED set ssl vserver cs_vsrv_adfs -ssl3 DISABLED
This is a pretty common setup for HTTP/ HTTPS based services. But it doesn’t work… Mainly because the monitor was not getting the required response. So the monitored service was down for the ADC, and therefore the service group, the load balancing virtual server and the content switch won’t came up.
The reason for this is Server Name Indication (SNI), an extension to Transport Layer Security (TLS). SNI is enabled and required since ADFS 3.0. The monitor tries to access the URL http://x.x.x.x/federationmetadata/2007-06/federationmetadata.xml, but the ADFS service won’t answer to those requests, because it includes the ip address, and not the ADFS service name.
But there is a workaround for everything on the Internet! You can change the binding on the ADFS server nodes using netsh.
netsh http add sslcert ipport=<IPAddress:port> certhash=<certhash> appid=<appid> certstorename=MY
I will not add the necessary options to this command, because: DON’T DO THIS!
Yes, the service group, the load balancing virtual server and the content switch will come up after this change. But you will not be able to enable a trust between your ADFS Proxy servers and the ADFS farm.
Microsofts requirements on Load Balancing ADFS
Microsoft offers a nice overview about the requirements when deploying ADFS. There is a section about the Network requirements. Below this, Microsoft clearly documents the requirements when load balancing ADFS servers and ADFS Proxy servers.
The load balancer MUST NOT terminate SSL. AD FS supports multiple use cases with certificate authentication which will break when terminating SSL. Terminating SSL at the load balancer is not supported for any use case.Requirements for deploying AD FS/ microsoft.com
Okay, with this in mind, the you can’t use a ADC Content Switch as described above. Because it will terminate SSL. You have to switch to a load balancing virtual server and a service group with SSL bridge . Citrix describes SSL bridge as follows:
A SSL bridge configured on the NetScaler appliance enables the appliance to bridge all secure traffic between the SSL client and the SSL server. The appliance does not offload or accelerate the bridged traffic, nor does it perform encryption or decryption. Only load balancing is done by the appliance. The SSL server must handle all SSL-related processing. Features such as content switching, SureConnect, and cache redirection do not work, because the traffic passing through the appliance is encrypted.
But there is a second, very interesting statement:
It is recommended to use the HTTP (not HTTPS) health probe endpoints to perform load balancer health checks for routing traffic. This avoids any issues relating to SNI. The response to these probe endpoints is an HTTP 200 OK and is served locally with no dependence on back-end services. The HTTP probe can be accessed over HTTP using the path ‘/adfs/probe’http://<Web Application Proxy name>/adfs/probeRequirements for deploying AD FS/ microsoft.com
http://<ADFS server name>/adfs/probe
http://<Web Application Proxy IP address>/adfs/probe
http://<ADFS IP address>/adfs/probe
This is pretty interesting, because it addresses the above described issue with the monitor. The solution to this is a HTTP-ECV monitor with on port 80, a GET to “/adfs/probe” and the check for a HTTP/200.
A working Citrix ADC setup
This setup is divided into two parts: One for the ADFS farm, and a second one for the ADFS Proxy farm. It uses SSL bridge and HTTP for the service monitor.
Load balancing the ADFS farm
add server srv_adfs1 x.x.x.x add server srv_adfs2 x.x.x.y add serviceGroup svcgrp_adfs SSL_BRIDGE -maxClient 0 -maxReq 0 -cip DISABLED -usip NO -useproxyport YES -cltTimeout 180 -svrTimeout 360 -CKA NO -TCPB NO -CMP NO add lb vserver lb_vsrv_adfs SSL_BRIDGE x.x.x.z 443 -persistenceType SSLSESSION -cltTimeout 180 add lb monitor mon_adfs_http HTTP -respCode 200 -httpRequest "GET /adfs/probe" -LRTM ENABLED -destPort 80 bind serviceGroup svcgrp_adfs srv_adfs1 443 bind serviceGroup svcgrp_adfs srv_adfs2 443 bind serviceGroup svcgrp_adfs -monitorName mon_adfs_http bind lb vserver lb_vsrv_adfs svcgrp_adfs set ssl vserver lb_vsrv_adfsproxy -ssl3 DISABLED
Load balancing the ADFS Proxy farm
add server srv_adfsproxy1 y.y.y.y add server srv_adfsproxy2 y.y.y.x add serviceGroup svcgrp_adfsproxy SSL_BRIDGE -maxClient 0 -maxReq 0 -cip DISABLED -usip NO -useproxyport YES -cltTimeout 180 -svrTimeout 360 -CKA NO -TCPB NO -CMP NO add lb vserver lb_vsrv_adfsproxy SSL_BRIDGE y.y.y.z 443 -persistenceType SSLSESSION -cltTimeout 180 add lb monitor mon_adfs_proxy_http HTTP -respCode 200 -httpRequest "GET /adfs/probe" -LRTM ENABLED -destPort 80 bind serviceGroup svcgrp_adfsproxy srv_adfsproxy1 443 bind serviceGroup svcgrp_adfsproxy srv_adfsproxy2 443 bind serviceGroup svcgrp_adfs -monitorName mon_adfs_proxy_http bind lb vserver lb_vsrv_adfsproxy svcgrp_adfsproxy set ssl vserver lb_vsrv_adfsproxy -ssl3 DISABLED
I have implemented it on a NetScaler 12.1 with a Standard license. If you have feedback or questions, please leave a comment. :)
- Configure VMware Horizon View client device certificate authentication - March 11, 2021
- VMware vCenter 7.0 U2 deployment fails at stage 2 - March 10, 2021
- Veeam B&R backup failes with “No scale-out repository extents are available” - February 24, 2021