net/interfaces: handle iOS network transitions when exit node in use (ENG-2507)

[agottardo]

Fixes ENG-2507
Updates #8022
Updates #6075

On iOS, we currently rely on delegated interface information to figure out the default route interface. When an exit node is being use, our utun* interface is the default route, so we ask the system for the delegated interface for the utun* interface we use. The NetworkExtension framework in iOS seems to set the delegate interface only once, upon the creation of the VPN tunnel. If a network transition (e.g. from Wi-Fi to Cellular) happens while the tunnel is connected, it will be ignored and we will still try to set Wi-Fi as the default route because the delegated interface is not getting updated as connectivity transitions.

Here, we special-case iPhones and iPads with a simpler logic that doesn't look at the routing table: we try finding a hardcoded Wi-Fi interface called en0, if it doesn't have an address, we fall back to cellular, which is pdp_ip0. This logic can safely run every time a link change is detected.

I tested this on iPhones and iPads running iOS 17.1 and it appears to work. Switching between different cellular plans on a dual SIM configuration also works (the interface name remains pdp_ip0).

To-Dos

• handle USB Ethernet
• verify behaviour on other Apple platforms
• verify behaviour on devices with no cellular interface
• verify behaviour on device with cellular data available, but disabled
• verify behaviour with Wireless CarPlay

[bobbyl140]

This may be a very specific use case, but what would happen in the case of a USB to Ethernet adapter? I don't frequently use one, but I have done it before. If there was a way to incorporate support for those without much additional effort, that would be awesome. If not, I understand, this seems to be an Apple problem in the first place.

[agottardo]

This may be a very specific use case, but what would happen in the case of a USB to Ethernet adapter? I don't frequently use one, but I have done it before. If there was a way to incorporate support for those without much additional effort, that would be awesome. If not, I understand, this seems to be an Apple problem in the first place.

Thanks for the suggestion. We lack logic for that. We should look at all en* interfaces (USB to Ethernet adapters appear to use en4 on my iPhone 15).

[agottardo]

I feel like this is re-inventing the wheel. Might be better to observe NWPathHandler updates from the network extension and forward those updates to Go.

[bobbyl140]

Thanks for the suggestion. We lack logic for that. We should look at all en* interfaces (USB to Ethernet adapters appear to use en4 on my iPhone 15).

Bluetooth PAN is probably somewhere in there too, which isn't a problem on iPhones (as originators of Personal Hotspot), but on iPads where people might use Bluetooth to connect to their iPhone it could be prevalent.

[agottardo]

Bluetooth PAN is probably somewhere in there too, which isn't a problem on iPhones (as originators of Personal Hotspot), but on iPads where people might use Bluetooth to connect to their iPhone it could be prevalent.

Precisely. I spent the morning looking into how we can ask iOS for the right interface to use instead of making a decision on our end, because I'm concerned about edge cases like that. I'll likely scrap this entire PR and see if we can observe NWPathMonitor updates from our Swift code instead, which is the Apple-sanctioned way to do this.

We couldn't use this before because we had to support older versions of macOS and iOS where that wasn't available. But it is no longer an issue now.

[bobbyl140]

That would make sense... Thank you again for looking into this!

[skaeight]

I’m seeing this issue. Any eta when this change will go live?

[agottardo]

I refactored the logic here. We now have a UpdateLastKnownDefaultRouteInterface(string) function that the iOS app can call when it notices a network transition via NWPathMonitor. Once Swift has given us an interface name, we always use that, and that should handle all edge cases including USB Ethernet adapters and other similar setups.

The logic here that hard-codes en0 and pdp_ip0 will only be used as a fall back when Swift hasn't delivered an update yet, or has delivered an interface name that is invalid because the interface is down.

[knyar]

Looks good, just a couple style nits.