It has been known for a while that iPhones and other iDevices do not play well with Exchange ActiveSync when roaming between a public network (such as 3G) and a private internal network to which the Exchange Server is connected. In particular, push email often does not work, which seems to be a bug in the iOS software. It’s a known issue, according to Apple. However, it caught me out recently, because the problem seemed to go away for a long time with the release of the iPhone 4.
At work, we have a set-up which is quite common for organisations of our size. We have two distinct networks: the internal network, which is reserved only for trusted devices owned and managed by us (the PCs, laptops, printers, switching gear, servers and now, thin clients). With 1000s of devices on this network, it is VLANed quite heavily to increase manageability, although I will admit this project was something I only completed fairly recently. Before last year, it was a single broadcast domain… but that is another story.
However, we also have a guest network. The guest network is isolated into its own VLAN, and is for wired clients which cannot authenticate as domain members (via 802.1x authentication) or for wireless clients connecting via a “Guest” SSID issued by our wireless controller. The guest network is still restricted to internal use – users authenticate to our RADIUS servers from their phones or laptops. Provided they provide valid credentials, they are provided with restricted access to the Internet.
All of the networks are linked together by our Forefront TMG deployment. This is driven by our inbound ISP connection and exposes several interfaces to the network – the internal network has two, teamed interfaces (for redundancy and throughput for data from cache) and the guest network has a further interface. The TMG deployment is the gateway for the guest network, and the internal network has a 0.0.0.0 default route for unresolved traffic crossing the VLANs.
When the Forefront TMG was provisioned last year, I initially configured the guest network both for internet access, but also with an internal set of “relay” rules, if you like, for access to certain resources on the internal network – OWA, RD Web Access, our management system, internal websites and, crucially, DNS lookups via our internal name servers. In effect, guest traffic was not NATed onto its own public IP. When it matched a firewall rule for one of our internal services, it was simply routed into the internal network. This made the deployment much simpler, and meant the internal IP addresses returned by internal DNS nameservers would still work for guest clients. Upshot: I don’t need more nameservers!
At the time, this did not pose a problem, even with the iPhone and iPad devices used by our staff. These phones could have been on 3G and wifi simultaneously, and we never had an issue with the mismatched IP addresses on the two networks stopping ActiveSync working.
That is, however, until someone upgraded to the iPhone 4S.
As noted in the blog post linked above:
“push” may stop working if your company’s Exchange ActiveSync server has a different IP address for intranet and Internet clients. Make sure the DNS for your network returns a single, externally-routable address to the Exchange ActiveSync server for both intranet and Internet clients
The problem experienced with this one iPhone 4S user went beyond push email. The user’s phone worked perfectly when away from the network. However, the moment it roaming onto our wifi, it seemed to have an adverse effect on the Exchange account configuration. Almost immediately, the phone would report a password error on a manual email check. The Exchange account would then refuse to work at all – on any network – until the user deleted the device from the Exchange Control Panel (ECP), switched back to 3G and re-created his connection.
I was not convinced the issue was with Exchange – all manner of other devices, even the iPhone 4, were still working. Nothing tested incorrectly. The problem was not a user issue, as I had him configure a test user account for a few days. Same problem.
Eventually, after a lot of painstaking troubleshooting (and waiting for feedback), it started to become very clear the issue was present only on the iPhone 4S and only in certain circumstances. However, it was much more serious than before – when the issue occurred, it did not just stop the iPhone from working until it roamed off-site again. It essentially wrote the email capability on the device off.
The resolution was a simple one, and one I should probably have implemented in the first place. The Forefront TMG deployment was re-configured. No routing was permitted between the internal network and the guest network. Instead, I added a network rule for guest network traffic to be NATed to its own public IP. I built a new cluster of standalone DNS servers, which serve two purposes – recursive lookups from the internal network (they are, effectively, caching servers) and hosting of the public DNS zones which return public IP addresses for all our network services.
When the guest network was given access to these nameservers, the iPhone 4S problem immediately went away. As detailed by Apple, it seems their devices are once again having issues with multiple IP addresses being issued by DNS for the same service. I thought this inconvenience had been resolved, but it would appear this design strategy will be going back into my network design methodology in the future. In any event, it did allow me to streamline and simplify our guest network configuration, which is always a good thing!
Watch out for Apple devices and the problem with issuing different internal and external IPs if they are used on your internal wifi. Either make the public IP routable internally and use that for internal access, or – a very common solution – don’t use them on wifi at all.