eyJhb changed the topic of #nixos-on-your-router to: NixOS on your Router || https://logs.nix.samueldr.com/nixos-on-your-router
<betawaffle> not exactly, no. i haven't figured out what filters to use
<andi-> ip6 ;-)
<betawaffle> doesn't look like there's anything, but i can't bounce the network right now
<betawaffle> this is all i see: 2019-12-12 20:08:16.629537 88:96:4e:6d:8f:21 > 33:33:00:00:83:84, ethertype IPv6 (0x86dd), length 136: (flowlabel 0xd71c1, hlim 1, next-header UDP (17) payload length: 82) fe80::8a96:4eff:fe6d:8f21.48739 > ff12::8384.21027: [bad udp cksum 0xe8a0 -> 0xa686!] UDP, length 74
<betawaffle> oh, there's a router solicitation, but no response
cransom has quit [Ping timeout: 276 seconds]
cransom has joined #nixos-on-your-router
<thefloweringash> also maybe interesting in my config: I use systemd-networkd for everything except ipv6 configuration, for which I use the standard dhcpcd+radvd pair. I'm sure this is a terrible idea, but it does seem to work.
pie__ has quit [Ping timeout: 265 seconds]
<andi-> thefloweringash: on my current setup I do the same.. I do think that the way systemd networkd handles DHCPv6 and especially PD is not great. It is a very simple approach to do N PD requests (one for each interface where you need a prefix) but that might not really be compatible outside of some corporate DC or some very nice ISP
<andi-> I also see why they are going that route because otherwise you are in the subnetting hell with networkd. Not sure if there is a nice way to fix this.
<andi-> If you have many PD prefixes per customer it also is very expensive on your routers. You no longer need ~3 routes (1x v3, 1x NA, 1x PD) per customer but 3+(N-1) many.. When I was still building those systems the entire discussion was around the amounts of routes and what a customer costs. Eventuallya all those routes must fit in some kind of hardware RIB and not just the linux kernel.
<thefloweringash> I'm not confident my isp would respond if I requested anything other than my /56. They don't give out addresses over dhcp.
<andi-> Yeah same here. In the setups I have built before (ISP side) I always saw (and expected) the customer to just request on prefix and if they request more that would be denied.
<andi-> Which is also how the average Fitz!Box and cheap plastic routers behave.
xwvvvvwx has quit [Ping timeout: 265 seconds]
xwvvvvwx has joined #nixos-on-your-router
<betawaffle> thefloweringash: awesome, that's what i was gonna ask about today
<betawaffle> (the nftables dependencies on device units
<andi-> thefloweringash: really enjoying reading your MAP-E code :)
<andi-> finally a use-case for BPF that I can some-what support ;)
<betawaffle> what's MAP-E?
<betawaffle> oh interesting
<thefloweringash> the bit that's hard with linux is doing the nat and keeping endpoint independence
<thefloweringash> my approach is to make linux map to a contiguous set of 256 ports, and then shuffle the port bits with bpf to them into 16 groups of 16 ports
<thefloweringash> other approaches are multiple nat rules and some logic to decide which one triggers, like round-robin with nth, or limiting by connections by destination addr (seen in openwrt)
<betawaffle> gchristensen: i successfully flashed my apu2e4 with the *latest* firmware, built just days ago
<gchristensen> ohh nice!
<betawaffle> it was crazy easy
<gchristensen> how?
<gchristensen> I've never tried
<betawaffle> install flashrom (available in nixpkgs), download the firmware you want from their site: https://pcengines.github.io/
<gchristensen> oh cool
<betawaffle> it ends up being either flashrom -w whatever.rom -p internal (or internal:boardmismatch=force the first time, depending)
<betawaffle> then unplug the power, and plug it back in
<betawaffle> (needs to be a cold reset)
<betawaffle> i'd recommend also doing `poweroff` in the os before
pie_ has joined #nixos-on-your-router
mcfrank has joined #nixos-on-your-router
Ox4A6F has joined #nixos-on-your-router