Today I don’t have a managed switch. I had one years ago, but it was user hostile - it would try phone home without asking me, it also wouldn’t let me lock down access to the admin web interface as well as I’d have liked - I presume that was to make it harder to brick.
I’ve grown up building networks out of OpenBSD/OpenWRT/Debian (routers/APs/firewalls) with dumb switches.
I don’t use VLANs, since dumb switches are so cheap I just have (real) RLANS.
But maybe it’s time I joined the mainstream.
If I look at the switch architecture that 1.3 Gb/s link between the switch and “CPU” is my main concern. I’d like the “CPU” to be effectively off the network, just there to configure and monitor the switch. If that can all happen over I2C or SPI that’s cool.
The ideal position would be not having to update the firmware, because the “CPU” side isn’t easily able to be exposed to malicious input. Also not having the firmware in a position to be trying to phone home.
In such a setup I could, say, replace two dumb switches with 1 switch and 2 VLANs without increasing the vulnerability surface much. (It will increase it somewhat as the switch chips have their own bugs, eg: malicious packets may take advantage of errors in parsing logic, but I like to think those are largely going to cause crashes or DoS type issues and not RCEs).
What do people think? Is this asking too much? Other paths I could try are OpenWRT switches (at least I can trust the firmware) or DIP switch configured switches (limited but a step above regular dumb switches).
For years the big networking brands have been great targets for hacking, take over a router or switch and a malicious actor has a very useful asset - people don’t pay much attention or scan such network devices much for malware, they by definition have a lot of power over the network and often have a lot of bandwidth. And take over they have been, such that I’ve become sceptical of them, like a lot of other embedded devices.

