I don't care about the L2 interface. I imagine I could twist my ISPs arm and spend a bunch of money to end up with the fiber line directly connected to a card in my Linux box... but why? The ISP's upstream equipment is just as vulnerable as a modem, and there's a lot more of it. That's not the point for me.

When I had DSL (and later DOCSIS), I just bought my own modems that weren't routers, and used ethernet to connect the Linux router (a raspberry pi 3 originally, yes the DSL was so slow it wasn't a bottleneck!). It was economical because Time Warner charged me something like $5/month to use their modem.

There's a lot of reasons I like having a Linux router, a short list:

* I have 10G internet from Sonic, in 2020 building a small Linux machine with ixgbe cards was cheaper than any commercially available 10G router with as many ports as I needed (4+). I get line rate through it, no problem.

* It runs an open source NAT implementation that is actually correct (although this is less of a problem with commercial routers now than it was back in the day...). Sadly, I don't have IPv6, but the same argument would apply there.

* I can run whatever DHCP server implementation I want.

* No web GUIs: I SSH to the thing and write config files.

* If it blows up, it's just Linux, and all the config files are in git: any machine will do, I don't have to worry about some commercial product disappearing and forcing me to waste an afternoon reinventing the wheel.

* I can build and run recent vanilla kernels with all the hardening options enabled. A Linux router spends 99% of its time in the kernel.

* The machine is actually fast enough to scan 10G traffic at line rate in software when I want to do that (it's doing all the routing in software, after all).

It's really easy, I'd estimate I spend 1-2 hours a year maintaining it.