I encountered a problem that’s sort of funny. It turns out that Starlink uses addresses in thie RFC6598 space for numbering devices within their network. My default router on Starlink (I’m using my own router and talking to the user terminal directly in pass-through mode) is 100.64.0.1. This hasn’t been an issue until recently since I’m not running Tailscale on the router at the edge of my network.
However, just today I installed Tailscale on an Ubuntu 20.04 Linux server at home, running a dozen or more docker containers (mostly related to Home Automation stuff). One of the containers that I have continuously runs fping in the background, to monitor connectivity and packet loss over each my ISPs (Starlink and a cable MSO). I had been pinging the Starlink default router (100.64.0.1) as one measurement point… and of course that stopped working when Tailscale came up since it grabbed those addresses.
My tactical question is if there’s a way I can have it ignore that address, or a prefix that covers it? I see there’s some cleverness going on with IP tables rules and it’s further complicated by all the wackyness that Docker does with IP tables, too.
Longer term, it might be nice to be able to choose a narrower prefix so I can have my Tailscale network addresses come out of some non-overlapping/conflicting range of addresses. A /10 prefix is way more addresses that I’ll need. Though maybe having a sufficiently sparse allocation in the larger range reduces probability of conflicts when access to my tailnet is shared with some external parties?