{ config , pkgs , notnft , lib , router-lib , server-config , ... }: let cfg = config.router-settings; hapdConfig = { inherit (cfg) country_code wpa_passphrase; he_su_beamformer = true; he_su_beamformee = true; he_mu_beamformer = true; he_spr_sr_control = 3; he_default_pe_duration = 4; he_rts_threshold = 1023; he_mu_edca_qos_info_param_count = 0; he_mu_edca_qos_info_q_ack = 0; he_mu_edca_qos_info_queue_request = 0; he_mu_edca_qos_info_txop_request = 0; he_mu_edca_ac_be_aifsn = 8; he_mu_edca_ac_be_aci = 0; he_mu_edca_ac_be_ecwmin = 9; he_mu_edca_ac_be_ecwmax = 10; he_mu_edca_ac_be_timer = 255; he_mu_edca_ac_bk_aifsn = 15; he_mu_edca_ac_bk_aci = 1; he_mu_edca_ac_bk_ecwmin = 9; he_mu_edca_ac_bk_ecwmax = 10; he_mu_edca_ac_bk_timer = 255; he_mu_edca_ac_vi_ecwmin = 5; he_mu_edca_ac_vi_ecwmax = 7; he_mu_edca_ac_vi_aifsn = 5; he_mu_edca_ac_vi_aci = 2; he_mu_edca_ac_vi_timer = 255; he_mu_edca_ac_vo_aifsn = 5; he_mu_edca_ac_vo_aci = 3; he_mu_edca_ac_vo_ecwmin = 5; he_mu_edca_ac_vo_ecwmax = 7; he_mu_edca_ac_vo_timer = 255; preamble = true; country3 = "0x49"; # indoor }; # routing tables wan_table = 1; # vpn table, assign an id but don't actually add a rule for it, so it is the default vpn_table = 2; vpn_mtu = config.networking.wireguard.interfaces.wg0.mtu; vpn_ipv4_mss = vpn_mtu - 40; vpn_ipv6_mss = vpn_mtu - 60; dnatRuleMode = rule: if rule.mode != "" then rule.mode else if rule.target4.address or null == netAddresses.lan4 || rule.target6.address or null == netAddresses.lan6 then "rule" else "mark"; dnatRuleProtos = rule: let inherit (notnft.inetProtos) tcp udp; in if rule.tcp && rule.udp then notnft.dsl.set [ tcp udp ] else if rule.tcp then tcp else if rule.udp then udp else throw "Invalid rule: either tcp or udp must be set"; setIfNeeded = arr: if builtins.length arr == 1 then builtins.head arr else notnft.dsl.set arr; # nftables rules generator # selfIp4/selfIp6 = block packets from these addresses # extraInetEntries = stuff to add to inet table # extraNetdevEntries = stuff to add to netdev table # wans = external interfaces (internet) # lans = internal interfaces (lan) # netdevIngressWanRules = additional rules for ingress (netdev) # inetInboundWanRules = additional rules for input from wan (inet) # inetInboundLanRules = same for lan # inetForwardRules = additional forward rules besides allow lan->wan forwarding # inetSnatRules = snat rules (changing source address, usually just called nat) # inetDnatRules = dnat rules (changing destination address, i.e. port forwarding) # logPrefix = log prefix for drops mkRules = { selfIp4, selfIp6, extraInetEntries ? {}, extraNetdevEntries ? {}, wans, lans, netdevIngressWanRules ? [], inetInboundWanRules ? [], inetInboundLanRules ? [], inetForwardRules ? [], inetSnatRules ? [], inetDnatRules ? [], logPrefix ? "", }: let logIfWan = prefix: lib.optional (logPrefix == "wan") (notnft.dsl.log prefix); in with notnft.dsl; with payload; ruleset { filter = add table.netdev ({ ingress_common = add chain ([(is.eq (bit.and tcp.flags (f: bit.or f.fin f.syn)) (f: bit.or f.fin f.syn))] ++ logIfWan "${logPrefix}fin+syn drop " ++ [drop]) ([(is.eq (bit.and tcp.flags (f: bit.or f.syn f.rst)) (f: bit.or f.syn f.rst))] ++ logIfWan "${logPrefix}syn+rst drop " ++ [drop]) [(is.eq (bit.and tcp.flags (f: with f; bit.or fin syn rst psh ack urg)) 0) (log "${logPrefix}null drop ") drop] [(is tcp.flags (f: f.syn)) (is.eq tcpOpt.maxseg.size (range 0 500)) (log "${logPrefix}maxseg drop ") drop] # reject requests with own saddr # log if they are meant for us... [(is.eq ip.saddr selfIp4) (is.eq (fib (f: with f; [ daddr iif ]) (f: f.type)) (f: f.local)) (log "${logPrefix}self4 ") drop] [(is.eq ip6.saddr selfIp6) (is.eq (fib (f: with f; [ daddr iif ]) (f: f.type)) (f: f.local)) (log "${logPrefix}self6 ") drop] # ...but ignore if they're multicast/broadcast [return]; ingress_lan_common = add chain # there are some issues with this, disable it for lan # [(is.eq (fib (f: with f; [ saddr mark iif ]) (f: f.oif)) missing) (log "${logPrefix}oif missing ") drop] inetInboundLanRules [(jump "ingress_common")]; ingress_wan_common = add chain netdevIngressWanRules [(jump "ingress_common")] # [(is.ne (fib (f: with f; [ daddr iif ]) (f: f.type)) (f: set [ f.local f.broadcast f.multicast ])) (log "${logPrefix}non-{local,broadcast,multicast} ") drop] # separate limits for echo-request and all other icmp types [(is.eq ip.protocol (f: f.icmp)) (is.eq icmp.type (f: f.echo-request)) (limit { rate = 50; per = f: f.second; }) accept] [(is.eq ip.protocol (f: f.icmp)) (is.ne icmp.type (f: f.echo-request)) (limit { rate = 100; per = f: f.second; }) accept] [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq icmpv6.type (f: f.echo-request)) (limit { rate = 50; per = f: f.second; }) accept] [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.ne icmpv6.type (f: f.echo-request)) (limit { rate = 100; per = f: f.second; }) accept] # always accept destination unreachable, time-exceeded, parameter-problem, packet-too-big [(is.eq ip.protocol (f: f.icmp)) (is.eq icmp.type (f: with f; set [ destination-unreachable time-exceeded parameter-problem ])) accept] [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq icmpv6.type (f: with f; set [ destination-unreachable time-exceeded parameter-problem packet-too-big ])) accept] # don't log echo-request drops [(is.eq ip.protocol (f: f.icmp)) (is.eq icmp.type (f: f.echo-request)) drop] [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq icmpv6.type (f: f.echo-request)) drop] [(is.eq ip.protocol (f: f.icmp)) (log "${logPrefix}icmp flood ") drop] [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (log "${logPrefix}icmp6 flood ") drop]; } // extraNetdevEntries // builtins.listToAttrs (map (name: { name = "ingress_${name}"; value = add chain { type = f: f.filter; hook = f: f.ingress; dev = name; prio = -500; policy = f: f.accept; } [(jump "ingress_lan_common")]; }) lans) // builtins.listToAttrs (map (name: { name = "ingress_${name}"; value = add chain { type = f: f.filter; hook = f: f.ingress; dev = name; prio = -500; policy = f: f.accept; } [(jump "ingress_wan_common")]; }) wans)); global = add table { family = f: f.inet; } ({ inbound_wan_common = add chain [(vmap ct.state { established = accept; related = accept; invalid = drop; })] [(is ct.status (f: f.dnat)) accept] ([(is.eq (bit.and tcp.flags (f: f.syn)) 0) (is.eq ct.state (f: f.new))] ++ logIfWan "${logPrefix}new non-syn " ++ [drop]) # icmp: only accept ping requests [(is.eq ip.protocol (f: f.icmp)) (is.eq icmp.type (f: f.echo-request)) accept] # icmpv6: accept no-route info from link-local addresses [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq ip6.saddr (cidr "fe80::/10")) (is.eq icmpv6.code (f: f.no-route)) (is.eq icmpv6.type (f: with f; set [ mld-listener-query mld-listener-report mld-listener-done mld2-listener-report ])) accept] # icmpv6: accept commonly useful stuff [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq icmpv6.type (f: with f; set [ destination-unreachable time-exceeded echo-request echo-reply ])) accept] # icmpv6: more common stuff [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq icmpv6.code (f: f.no-route)) (is.eq icmpv6.type (f: with f; set [ packet-too-big parameter-problem ])) accept] [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq icmpv6.code (f: f.admin-prohibited)) (is.eq icmpv6.type (f: f.parameter-problem)) accept] inetInboundWanRules; # trust the lan inbound_lan_common = add chain [accept]; inbound = add chain { type = f: f.filter; hook = f: f.input; prio = f: f.filter; policy = f: f.drop; } [(vmap meta.iifname ({ lo = accept; } // lib.genAttrs lans (_: jump "inbound_lan_common") // lib.genAttrs wans (_: jump "inbound_wan_common") ))]; #[(log "${logPrefix}inbound drop ")]; forward = add chain { type = f: f.filter; hook = f: f.forward; prio = f: f.filter; policy = f: f.drop; } [(vmap ct.state { established = accept; related = accept; invalid = drop; })] [(is ct.status (f: f.dnat)) accept] # accept lan->wan fw [(is.eq meta.iifname (setIfNeeded lans)) (is.eq meta.oifname (setIfNeeded wans)) accept] # accept lan->lan fw [(is.eq meta.iifname (setIfNeeded lans)) (is.eq meta.oifname (setIfNeeded lans)) accept] # accept wan->lan icmpv6 forward [(is.eq meta.iifname (setIfNeeded wans)) (is.eq icmpv6.type (f: with f; set [ destination-unreachable time-exceeded echo-request echo-reply ])) accept] [(is.eq meta.iifname (setIfNeeded wans)) (is.eq icmpv6.code (f: f.no-route)) (is.eq icmpv6.type (f: with f; set [ packet-too-big parameter-problem ])) accept] [(is.eq meta.iifname (setIfNeeded wans)) (is.eq icmpv6.code (f: f.admin-prohibited)) (is.eq icmpv6.type (f: f.parameter-problem)) accept] inetForwardRules [(log "${logPrefix}forward drop ")]; snat = add chain { type = f: f.nat; hook = f: f.postrouting; prio = f: f.srcnat; policy = f: f.accept; } # masquerade ipv6 because my isp doesn't provide it and my vpn gives a single ipv6 [(is.eq meta.protocol (f: set [ f.ip f.ip6 ])) (is.eq meta.iifname (setIfNeeded lans)) (is.eq meta.oifname (setIfNeeded wans)) masquerade] inetSnatRules; dnat = add chain { type = f: f.nat; hook = f: f.prerouting; prio = f: f.dstnat; policy = f: f.accept; } inetDnatRules; } // extraInetEntries); }; netAddressesWithPrefixLen = { lan4 = cfg.network; lan6 = cfg.network6; netns4 = cfg.netnsNet; netns6 = cfg.netnsNet6; wg4 = cfg.wgNetwork; wg6 = cfg.wgNetwork6; }; # parse a.b.c.d/x into { address, prefixLength } netParsedCidrs = builtins.mapAttrs (_: router-lib.parseCidr) netAddressesWithPrefixLen; # generate network cidr from device address # (normalizeCidr applies network mask to the address) netCidrs = builtins.mapAttrs (_: v: router-lib.serializeCidr (router-lib.normalizeCidr v)) netParsedCidrs; netAddresses = builtins.mapAttrs (_: v: v.address) netParsedCidrs // { netnsWan4 = cfg.wanNetnsAddr; netnsWan6 = cfg.wanNetnsAddr6; }; parsedGatewayAddr4 = router-lib.parseIp4 netAddresses.lan4; parsedGatewayAddr6 = router-lib.parseIp6 netAddresses.lan6; addToIp' = ip: n: lib.init ip ++ [ (lib.last ip + n) ]; addToIp = ip: n: router-lib.serializeIp (addToIp' ip n); # server serverAddress4 = addToIp parsedGatewayAddr4 1; serverAddress6 = addToIp parsedGatewayAddr6 1; # robot vacuum (valetudo) vacuumAddress4 = addToIp parsedGatewayAddr4 2; vacuumAddress6 = addToIp parsedGatewayAddr6 2; # light bulb (tasmota) lightBulbAddress4 = addToIp parsedGatewayAddr4 3; lightBulbAddress6 = addToIp parsedGatewayAddr6 3; # server in initrd serverInitrdAddress4 = addToIp parsedGatewayAddr4 4; serverInitrdAddress6 = addToIp parsedGatewayAddr6 4; hosted-domains = builtins.filter (domain: domain != "localhost") (builtins.concatLists (builtins.attrValues (builtins.mapAttrs (k: v: [ k ] ++ v.serverAliases) server-config.services.nginx.virtualHosts))); in { imports = [ ./options.nix ./metrics.nix ]; system.stateVersion = "22.11"; boot.kernel.sysctl = { "net.ipv4.conf.all.src_valid_mark" = true; "net.ipv4.conf.default.src_valid_mark" = true; "net.ipv4.conf.all.forwarding" = true; "net.ipv6.conf.all.forwarding" = true; }; services.openssh.enable = true; services.fail2ban = { enable = true; ignoreIP = [ netCidrs.lan4 netCidrs.lan6 ]; maxretry = 10; }; router-settings.dhcpReservations = [ { ipAddress = serverAddress4; macAddress = cfg.serverMac; } { ipAddress = vacuumAddress4; macAddress = cfg.vacuumMac; } { ipAddress = lightBulbAddress4; macAddress = cfg.lightBulbMac; } { ipAddress = serverInitrdAddress4; macAddress = cfg.serverInitrdMac; } ]; router-settings.dhcp6Reservations = [ { ipAddress = serverAddress6; duid = cfg.serverDuid; macAddress = cfg.serverMac; } { ipAddress = vacuumAddress6; macAddress = cfg.vacuumMac; } { ipAddress = lightBulbAddress6; macAddress = cfg.lightBulbMac; } { ipAddress = serverInitrdAddress6; macAddress = cfg.serverInitrdMac; } ]; # dnat to server, take ports from its firewall config router-settings.dnatRules = let bannedPorts = [ 631 9100 # printing 5353 # avahi # pass it through to VPN rather than WAN server-config.services.qbittorrent-nox.torrent.port ]; inherit (server-config.networking.firewall) allowedTCPPorts allowedTCPPortRanges allowedUDPPorts allowedUDPPortRanges; tcpAndUdp = builtins.filter (x: !builtins.elem x bannedPorts && builtins.elem x allowedTCPPorts) allowedUDPPorts; tcpOnly = builtins.filter (x: !builtins.elem x (bannedPorts ++ allowedUDPPorts)) allowedTCPPorts; udpOnly = builtins.filter (x: !builtins.elem x (bannedPorts ++ allowedTCPPorts)) allowedUDPPorts; rangesTcpAndUdp = builtins.filter (x: builtins.elem x allowedTCPPortRanges) allowedUDPPortRanges; rangesTcpOnly = builtins.filter (x: !builtins.elem x allowedUDPPortRanges) allowedTCPPortRanges; rangesUdpOnly = builtins.filter (x: !builtins.elem x allowedTCPPortRanges) allowedUDPPortRanges; in lib.optional (tcpAndUdp != [ ]) { port = setIfNeeded tcpAndUdp; tcp = true; udp = true; target4.address = serverAddress4; target6.address = serverAddress6; } ++ lib.optional (tcpOnly != [ ]) { port = setIfNeeded tcpOnly; tcp = true; udp = false; target4.address = serverAddress4; target6.address = serverAddress6; } ++ lib.optional (udpOnly != [ ]) { port = setIfNeeded udpOnly; tcp = false; udp = true; target4.address = serverAddress4; target6.address = serverAddress6; } ++ lib.flip map rangesTcpAndUdp (range: { port = notnft.dsl.range range.from range.to; tcp = true; udp = true; target4.address = serverAddress4; target6.address = serverAddress6; }) ++ lib.flip map rangesTcpOnly (range: { port = notnft.dsl.range range.from range.to; tcp = true; udp = false; target4.address = serverAddress4; target6.address = serverAddress6; }) ++ lib.flip map rangesUdpOnly (range: { port = notnft.dsl.range range.from range.to; tcp = false; udp = true; target4.address = serverAddress4; target6.address = serverAddress6; }) ++ [ { port = 24; tcp = true; udp = true; target4.port = 22; target6.port = 22; target4.address = serverInitrdAddress4; target6.address = serverInitrdAddress6; } { inVpn = true; port = server-config.services.qbittorrent-nox.torrent.port; tcp = true; udp = true; target4.address = serverAddress4; target6.address = serverAddress6; } ]; router.enable = true; # 2.4g ap router.interfaces.wlan0 = { bridge = "br0"; hostapd.enable = true; hostapd.settings = { inherit (cfg) ssid; hw_mode = "g"; channel = 3; chanlist = [ 3 ]; supported_rates = [ 60 90 120 180 240 360 480 540 ]; basic_rates = [ 60 120 240 ]; ht_capab = "[LDPC][SHORT-GI-20][SHORT-GI-40][TX-STBC][RX-STBC1][MAX-AMSDU-7935]"; } // hapdConfig; }; # 5g ap router.interfaces.wlan1 = { bridge = "br0"; hostapd.enable = true; hostapd.settings = { ssid = "${cfg.ssid}_5G"; ieee80211h = true; hw_mode = "a"; channel = 60; chanlist = [ 60 ]; tx_queue_data2_burst = 2; ht_capab = "[HT40+][LDPC][SHORT-GI-20][SHORT-GI-40][TX-STBC][RX-STBC1][MAX-AMSDU-7935]"; vht_oper_chwidth = 1; # 80mhz ch width vht_oper_centr_freq_seg0_idx = 42; vht_capab = "[RXLDPC][SHORT-GI-80][SHORT-GI-160][TX-STBC-2BY1][SU-BEAMFORMER][SU-BEAMFORMEE][MU-BEAMFORMER][MU-BEAMFORMEE][RX-ANTENNA-PATTERN][TX-ANTENNA-PATTERN][RX-STBC-1][SOUNDING-DIMENSION-4][BF-ANTENNA-4][VHT160][MAX-MPDU-11454][MAX-A-MPDU-LEN-EXP7]"; } // hapdConfig; }; # Unfortunately, this router's networking hardware is highly prone to breakage # Many people have reported their routers' TCP offloading being faulty, with an error # like this being thrown at random (it can be shortly after boot, or in a few days): # NETDEV WATCHDOG: eth0 (mtk_soc_eth): transmit queue 3 timed out 5388 ms # My hardware broke after a few months of use as well, so here's a potential fix router.interfaces.eth0.extraInitCommands = '' ${pkgs.ethtool}/bin/ethtool --offload eth0 tso off ''; # ethernet lan0-3 router.interfaces.lan0 = { bridge = "br0"; systemdLink.linkConfig.MACAddressPolicy = "persistent"; }; router.interfaces.lan1 = { bridge = "br0"; systemdLink.linkConfig.MACAddressPolicy = "persistent"; }; router.interfaces.lan2 = { bridge = "br0"; systemdLink.linkConfig.MACAddressPolicy = "persistent"; }; router.interfaces.lan3 = { bridge = "br0"; systemdLink.linkConfig.MACAddressPolicy = "persistent"; }; # sfp lan4 router.interfaces.lan4 = { bridge = "br0"; systemdLink.linkConfig.MACAddressPolicy = "persistent"; }; /* # sfp lan5 router.interfaces.lan5 = { bridge = "br0"; # i could try to figure out why this doesn't work... but i don't even have sfp to plug into this systemdLink.matchConfig.OriginalName = "eth1"; systemdLink.linkConfig.MACAddressPolicy = "persistent"; }; */ # ethernet wan router.interfaces.wan = { dependentServices = [ { service = "wireguard-wg0"; inNetns = false; } { service = "wireguard-wg1"; inNetns = false; } ]; systemdLink.linkConfig.MACAddressPolicy = "none"; systemdLink.linkConfig.MACAddress = cfg.routerMac; dhcpcd = { enable = true; # technically this should be assigned to br0 instead of veth-wan-b # however, br0 is in a different namespace! # Considering this doesn't work at all because my ISP doesn't offer IPv6, # I'd say this is "good enough" since it might still work in the wan # namespace, though I can't test it. extraConfig = '' interface wan ipv6rs ia_na 0 ia_pd 1 veth-wan-b/0 ''; }; networkNamespace = "wan"; }; # disable default firewall as it uses iptables # (and we have our own firewall) networking.firewall.enable = false; # br0, which bridges all lan devices # this is "the" lan device router.interfaces.br0 = { dependentServices = [ { service = "unbound"; bindType = "wants"; } ]; ipv4.addresses = lib.toList (netParsedCidrs.lan4 // { dns = [ netAddresses.lan4 ]; keaSettings.reservations = map (res: { hw-address = res.macAddress; ip-address = res.ipAddress; }) cfg.dhcpReservations; }); ipv6.addresses = lib.toList (netParsedCidrs.lan6 // { dns = [ netAddresses.lan6 ]; gateways = [ netAddresses.lan6 ]; radvdSettings.AdvAutonomous = true; coreradSettings.autonomous = true; # don't allocate addresses for most devices keaSettings.pools = [ ]; # just assign the reservations keaSettings.reservations = map (res: (if res.duid != null then { duid = res.duid; } else { hw-address = res.macAddress; }) // { ip-addresses = [ res.ipAddress ]; }) cfg.dhcp6Reservations; }); ipv4.routes = [ { extraArgs = [ netCidrs.lan4 "dev" "br0" "proto" "kernel" "scope" "link" "src" netAddresses.lan4 "table" wan_table ]; } ]; ipv6.routes = [ { extraArgs = [ netCidrs.lan6 "dev" "br0" "proto" "kernel" "metric" "256" "pref" "medium" "table" wan_table ]; } ]; ipv4.kea.enable = true; ipv6.corerad.enable = true; ipv6.kea.enable = true; }; router.networkNamespaces.default = { # set routing table depending on packet mark rules = [ { ipv6 = false; extraArgs = [ "fwmark" wan_table "table" wan_table ]; } { ipv6 = true; extraArgs = [ "fwmark" wan_table "table" wan_table ]; } # below is dnat config ] ++ builtins.concatLists (map (rule: let table = if rule.inVpn then 0 else wan_table; forEachPort = func: port: if builtins.isInt port then [ (func port) ] else if port?set then builtins.concatLists (map (forEachPort func) port.set) else if port?range.min then let inherit (port.range) min max; in [ (func "${toString min}-${toString max}") ] else if port?range then let max = builtins.elemAt port.range 1; min = builtins.head port.range; in [ (func "${toString min}-${toString max}" ) ] else throw "Unsupported expr: ${builtins.toJSON port}"; gen = len: proto: tgt: forEachPort (port: [ "from" "${tgt.address}/${toString len}" "ipproto" proto "sport" port "table" table ]) (if tgt.port == null then rule.port else tgt.port); in lib.optionals (rule.tcp && rule.target4 != null) (map (x: { ipv6 = false; extraArgs = x; }) (gen 32 "tcp" rule.target4)) ++ lib.optionals (rule.udp && rule.target4 != null) (map (x: { ipv6 = false; extraArgs = x; }) (gen 32 "udp" rule.target4)) ++ lib.optionals (rule.tcp && rule.target6 != null) (map (x: { ipv6 = true; extraArgs = x; }) (gen 128 "tcp" rule.target6)) ++ lib.optionals (rule.udp && rule.target6 != null) (map (x: { ipv6 = true; extraArgs = x; }) (gen 128 "udp" rule.target6)) ) (builtins.filter (x: (x.tcp || x.udp) && dnatRuleMode x == "rule") cfg.dnatRules)); # nftables rules # things to note: this has the code for switching between rtables # otherwise, boring stuff nftables.jsonRules = let lanSet = notnft.dsl.set [ "br0" "wg1" ]; in mkRules { selfIp4 = netAddresses.lan4; selfIp6 = netAddresses.lan6; lans = [ "br0" "wg1" ]; wans = [ "wg0" "veth-wan-a" ]; logPrefix = "lan "; netdevIngressWanRules = with notnft.dsl; with payload; [ # check oif only from vpn # dont check it from veth-wan-a because of dnat fuckery and because we already check packets coming from wan there [(is.eq meta.iifname (set [ "wg0" "wg1" ])) (is.eq (fib (f: with f; [ saddr mark iif ]) (f: f.oif)) missing) (log "lan oif missing ") drop] ]; inetDnatRules = builtins.concatLists (map (rule: let protocols = dnatRuleProtos rule; rule4 = rule.target4; rule6 = rule.target6; in with notnft.dsl; with payload; lib.optional (rule4 != null) [ (is.eq meta.iifname "wg0") (is.eq ip.protocol protocols) (is.eq th.dport rule.port) (if rule4.port == null then dnat.ip rule4.address else dnat.ip rule4.address rule4.port) ] ++ lib.optional (rule6 != null) [ (is.eq meta.iifname "wg0") (is.eq ip6.nexthdr protocols) (is.eq th.dport rule.port) (if rule6.port == null then dnat.ip6 rule6.address else dnat.ip6 rule6.address rule6.port) ] ) (builtins.filter (x: x.inVpn && (x.tcp || x.udp)) cfg.dnatRules)) ++ (with notnft.dsl; with payload; [ # hijack Microsoft DNS server hosted on Cloudflare [(is.eq meta.iifname lanSet) (is.eq ip.daddr "162.159.36.2") (is.eq ip.protocol (f: set [ f.tcp f.udp ])) (dnat.ip netAddresses.lan4)] ] ++ lib.optionals (cfg.naughtyMacs != []) [ [(is.eq meta.iifname lanSet) (is.eq ether.saddr (setIfNeeded cfg.naughtyMacs)) (is.eq ip.protocol (f: set [ f.tcp f.udp ])) (is.eq th.dport (set [ 53 853 ])) (dnat.ip netAddresses.lan4)] [(is.eq meta.iifname lanSet) (is.eq ether.saddr (setIfNeeded cfg.naughtyMacs)) (is.eq ip6.nexthdr (f: set [ f.tcp f.udp ])) (is.eq th.dport (set [ 53 853 ])) (dnat.ip6 netAddresses.lan6)] ]); inetForwardRules = with notnft.dsl; with payload; [ # allow access to lan from the wan namespace [(is.eq meta.iifname "veth-wan-a") (is.eq meta.oifname lanSet) accept] # allow dnat ("ct status dnat" doesn't work) ]; inetInboundWanRules = with notnft.dsl; with payload; [ [(is.eq ip.saddr (cidr netCidrs.netns4)) accept] [(is.eq ip6.saddr (cidr netCidrs.netns6)) accept] ]; extraInetEntries = with notnft.dsl; with payload; { block4 = add set { type = f: f.ipv4_addr; flags = f: with f; [ interval ]; } [ (cidr "194.190.137.0" 24) (cidr "194.190.157.0" 24) (cidr "194.190.21.0" 24) (cidr "194.226.130.0" 23) # no idea what this IP is, but it got a port 53 connection from one of the devices in this network - so off it goes "84.1.213.156" ]; block6 = add set { type = f: f.ipv6_addr; flags = f: with f; [ interval ]; }; # those tables get populated by unbound force_unvpn4 = add set { type = f: f.ipv4_addr; flags = f: with f; [ interval ]; }; force_unvpn6 = add set { type = f: f.ipv6_addr; flags = f: with f; [ interval ]; }; force_vpn4 = add set { type = f: f.ipv4_addr; flags = f: with f; [ interval ]; }; force_vpn6 = add set { type = f: f.ipv6_addr; flags = f: with f; [ interval ]; }; allow_iot4 = add set { type = f: f.ipv4_addr; flags = f: with f; [ interval ]; }; allow_iot6 = add set { type = f: f.ipv6_addr; flags = f: with f; [ interval ]; }; # TODO: is type=route hook=output better? It might help get rid of the routing inconsistency # between router-originated and forwarded traffic. The problem is type=route isn't supported # for family=inet, so I don't care enough to test it right now. prerouting = add chain { type = f: f.filter; hook = f: f.prerouting; prio = f: f.filter; policy = f: f.accept; } ([ [(mangle meta.mark ct.mark)] [(is.ne meta.mark 0) accept] # ban requests to/from block4/block6 # (might as well do this in ingress but i'm lazy) [(is.eq ip.daddr "@block4") (log "block4 ") drop] [(is.eq ip6.daddr "@block6") (log "block6 ") drop] [(is.eq ip.saddr "@block4") (log "block4/s ") drop] [(is.eq ip6.saddr "@block6") (log "block6/s ") drop] # default to vpn... [(mangle meta.mark vpn_table)] # ...but unvpn traffic to/from force_unvpn4/force_unvpn6 [(is.eq ip.daddr "@force_unvpn4") (mangle meta.mark wan_table)] [(is.eq ip6.daddr "@force_unvpn6") (mangle meta.mark wan_table)] [(is.eq ip.saddr "@force_unvpn4") (mangle meta.mark wan_table)] [(is.eq ip6.saddr "@force_unvpn6") (mangle meta.mark wan_table)] # ...force vpn to/from force_vpn4/force_vpn6 # (temporarily disable this because it breaks codeforces.org) # [(is.eq ip.daddr "@force_vpn4") (mangle meta.mark vpn_table)] # [(is.eq ip6.daddr "@force_vpn6") (mangle meta.mark vpn_table)] # [(is.eq ip.saddr "@force_vpn4") (mangle meta.mark vpn_table)] # [(is.eq ip6.saddr "@force_vpn6") (mangle meta.mark vpn_table)] # block requests to port 25 from hosts other than the server so they can't send mail pretending to originate from my domain # only do this for lans since traffic from other interfaces isn't forwarded to wan [(is.eq meta.iifname lanSet) (is.ne ether.saddr cfg.serverMac) (is.eq meta.l4proto (f: f.tcp)) (is.eq tcp.dport 25) (log "smtp ") drop] # don't vpn smtp requests so spf works fine (and in case the vpn blocks requests over port 25, which it usually does) [(is.eq meta.l4proto (f: f.tcp)) (is.eq tcp.dport 25) (mangle meta.mark wan_table)] ] ++ # 1. dnat non-vpn: change rttable to wan builtins.concatLists (map (rule: let protocols = dnatRuleProtos rule; rule4 = rule.target4; rule6 = rule.target6; in with notnft.dsl; with payload; lib.optionals (rule4 != null) [ [ (is.eq meta.iifname lanSet) (is.eq ip.protocol protocols) (is.eq ip.saddr rule4.address) (is.eq th.sport (if rule4.port != null then rule4.port else rule.port)) (mangle meta.mark wan_table) ] [ (is.eq meta.iifname "veth-wan-a") (is.eq ip.protocol protocols) (is.eq ip.daddr rule4.address) (is.eq th.dport (if rule4.port != null then rule4.port else rule.port)) (mangle meta.mark wan_table) ] ] ++ lib.optionals (rule6 != null) [ [ (is.eq meta.iifname lanSet) (is.eq ip6.nexthdr protocols) (is.eq ip6.saddr rule6.address) (is.eq th.sport (if rule6.port != null then rule6.port else rule.port)) (mangle meta.mark wan_table) ] [ (is.eq meta.iifname "veth-wan-a") (is.eq ip6.nexthdr protocols) (is.eq ip6.daddr rule6.address) (is.eq th.dport (if rule6.port != null then rule6.port else rule.port)) (mangle meta.mark wan_table) ] ]) (builtins.filter (x: !x.inVpn && (x.tcp || x.udp) && dnatRuleMode x == "mark") cfg.dnatRules)) ++ # 2. dnat vpn: change rttable to vpn builtins.concatLists (map (rule: let protocols = dnatRuleProtos rule; rule4 = rule.target4; rule6 = rule.target6; in with notnft.dsl; with payload; lib.optionals (rule4 != null) [ [ (is.eq meta.iifname lanSet) (is.eq ip.protocol protocols) (is.eq ip.saddr rule4.address) (is.eq th.sport (if rule4.port != null then rule4.port else rule.port)) (mangle meta.mark vpn_table) ] [ (is.eq meta.iifname "wg0") (is.eq ip.protocol protocols) (is.eq ip.daddr rule4.address) (is.eq th.dport (if rule4.port != null then rule4.port else rule.port)) (mangle meta.mark vpn_table) ] ] ++ lib.optionals (rule6 != null) [ [ (is.eq meta.iifname lanSet) (is.eq ip6.nexthdr protocols) (is.eq ip6.saddr rule6.address) (is.eq th.sport (if rule6.port != null then rule6.port else rule.port)) (mangle meta.mark vpn_table) ] [ (is.eq meta.iifname "wg0") (is.eq ip6.nexthdr protocols) (is.eq ip6.daddr rule6.address) (is.eq th.dport (if rule6.port != null then rule6.port else rule.port)) (mangle meta.mark vpn_table) ] ]) (builtins.filter (x: x.inVpn && (x.tcp || x.udp) && dnatRuleMode x == "mark") cfg.dnatRules)) ++ [ # for the robot vacuum, only allow traffic to/from allow_iot4/allow_iot6 [(is.eq ether.saddr cfg.vacuumMac) (is.ne ip.daddr (cidr netCidrs.lan4)) (is.ne ip.daddr "@allow_iot4") (log "iot4 ") drop] [(is.eq ether.saddr cfg.vacuumMac) (is.ne ip6.daddr (cidr netCidrs.lan6)) (is.ne ip6.daddr "@allow_iot6") (log "iot6 ") drop] [(is.eq ether.daddr cfg.vacuumMac) (is.ne ip.saddr (cidr netCidrs.lan4)) (is.ne ip.saddr "@allow_iot4") (log "iot4/d ") drop] [(is.eq ether.daddr cfg.vacuumMac) (is.ne ip6.saddr (cidr netCidrs.lan6)) (is.ne ip6.saddr "@allow_iot6") (log "iot6/d ") drop] # MSS clamping - since VPN reduces max MTU # We only do this for the first packet in a connection, which should be enough [(is.eq meta.nfproto (f: f.ipv4)) (is.eq meta.mark vpn_table) (is.gt tcpOpt.maxseg.size vpn_ipv4_mss) (mangle tcpOpt.maxseg.size vpn_ipv4_mss)] [(is.eq meta.nfproto (f: f.ipv6)) (is.eq meta.mark vpn_table) (is.gt tcpOpt.maxseg.size vpn_ipv6_mss) (mangle tcpOpt.maxseg.size vpn_ipv6_mss)] # warn about dns requests to foreign servers # TODO: check back and see if I should forcefully redirect DNS requests from certain IPs to router [(is.eq meta.iifname lanSet) (is.ne ip.daddr (netAddresses.lan4)) (is.eq ip.protocol (f: set [ f.tcp f.udp ])) (is.eq th.dport (set [ 53 853 ])) (log "dns4 ")] [(is.eq meta.iifname lanSet) (is.ne ip6.daddr (netAddresses.lan6)) (is.eq ip6.nexthdr (f: set [ f.tcp f.udp ])) (is.eq th.dport (set [ 53 853 ])) (log "dns6 ")] # finally, preserve the mark via conntrack [(mangle ct.mark meta.mark)] ]); }; }; }; # veths are virtual ethernet cables # veth-wan-a - located in the default namespace # veth-wan-b - located in the wan namespace # this allows routing traffic to wan namespace from default namespace via veth-wan-a # (and vice versa) router.veths.veth-wan-a.peerName = "veth-wan-b"; router.interfaces.veth-wan-a = { ipv4.addresses = [ netParsedCidrs.netns4 ]; ipv6.addresses = [ netParsedCidrs.netns6 ]; ipv4.routes = [ # default config duplicated for wan_table { extraArgs = [ netCidrs.netns4 "dev" "veth-wan-a" "proto" "kernel" "scope" "link" "src" netAddresses.netns4 "table" wan_table ]; } # default all traffic to wan in wan_table { extraArgs = [ "default" "via" netAddresses.netnsWan4 "table" wan_table ]; } ]; ipv6.routes = [ # default config duplicated for wan_table { extraArgs = [ netCidrs.netns6 "dev" "veth-wan-a" "proto" "kernel" "metric" "256" "pref" "medium" "table" wan_table ]; } # default all traffic to wan in wan_table { extraArgs = [ "default" "via" netAddresses.netnsWan6 "table" wan_table ]; } ]; }; router.interfaces.veth-wan-b = { networkNamespace = "wan"; ipv4.addresses = [ { address = netAddresses.netnsWan4; inherit (netParsedCidrs.netns4) prefixLength; } ]; ipv6.addresses = [ { address = netAddresses.netnsWan6; inherit (netParsedCidrs.netns6) prefixLength; } ]; # allow wan->default namespace communication ipv4.routes = [ { extraArgs = [ netCidrs.lan4 "via" netAddresses.netns4 ]; } ]; ipv6.routes = [ { extraArgs = [ netCidrs.lan6 "via" netAddresses.netns6 ]; } ]; }; router.networkNamespaces.wan = { # this is the even more boring nftables config nftables.jsonRules = mkRules { selfIp4 = netAddresses.netnsWan4; selfIp6 = netAddresses.netnsWan6; lans = [ "veth-wan-b" ]; wans = [ "wan" ]; netdevIngressWanRules = with notnft.dsl; with payload; [ [(is.eq (fib (f: with f; [ saddr mark iif ]) (f: f.oif)) missing) (log "wan oif missing ") drop] ]; inetDnatRules = builtins.concatLists (map (rule: let protocols = dnatRuleProtos rule; rule4 = rule.target4; rule6 = rule.target6; in with notnft.dsl; with payload; lib.optionals (rule4 != null) [ [ (is.eq meta.iifname "wan") (is.eq ip.protocol protocols) (is.eq th.dport rule.port) (if rule4.port == null then dnat.ip rule4.address else dnat.ip rule4.address rule4.port) ] ] ++ lib.optionals (rule6 != null) [ [ (is.eq meta.iifname "wan") (is.eq ip6.nexthdr protocols) (is.eq th.dport rule.port) (if rule6.port == null then dnat.ip6 rule6.address else dnat.ip6 rule6.address rule6.port) ] ]) (builtins.filter (x: !x.inVpn && (x.tcp || x.udp)) cfg.dnatRules)); inetSnatRules = # historically, i needed this, now i switched to ip rules # if i ever need this again, i have it right here builtins.concatLists (map (rule: let protocols = dnatRuleProtos rule; rule4 = rule.target4; rule6 = rule.target6; in with notnft.dsl; with payload; lib.optionals (rule4 != null) [ [ (is.eq meta.iifname "wan") (is.eq meta.oifname "veth-wan-b") (is.eq ip.protocol protocols) (is.eq th.dport (if rule4.port != null then rule4.port else rule.port)) (is.eq ip.daddr rule4.address) masquerade ] ] ++ lib.optionals (rule6 != null) [ [ (is.eq meta.iifname "wan") (is.eq meta.oifname "veth-wan-b") (is.eq ip6.nexthdr protocols) (is.eq th.dport (if rule6.port != null then rule6.port else rule.port)) (is.eq ip6.daddr rule6.address) masquerade ] ]) (builtins.filter (x: !x.inVpn && (x.tcp || x.udp) && dnatRuleMode x == "snat") cfg.dnatRules)); logPrefix = "wan "; inetInboundWanRules = with notnft.dsl; with payload; [ # DHCP [(is.eq meta.nfproto (x: x.ipv4)) (is.eq udp.dport 68) accept] [(is.eq meta.nfproto (x: x.ipv6)) (is.eq udp.dport 546) accept] # igmp, used for setting up multicast groups [(is.eq ip.protocol (f: f.igmp)) accept] # accept router solicitation stuff [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq icmpv6.type (f: with f; set [ nd-router-solicit nd-router-advert ])) accept] # accept neighbor solicitation stuff [(is.eq ip6.nexthdr (f: f.ipv6-icmp)) (is.eq icmpv6.code (f: f.no-route)) (is.eq icmpv6.type (f: with f; set [ nd-neighbor-solicit nd-neighbor-advert ])) accept] # SSH [(is.eq meta.l4proto (f: f.tcp)) (is.eq tcp.dport 23) accept] # wg1 [(is.eq meta.l4proto (f: with f; set [ udp tcp ])) (is.eq th.dport (set [ 854 855 ])) accept] ]; }; }; # vpn socket is in wan namespace, meaning traffic gets sent through the wan namespace # vpn interface is in default namespace, meaning it can be used in the default namespace networking.wireguard.interfaces.wg0 = cfg.wireguard // { socketNamespace = "wan"; interfaceNamespace = "init"; }; # use main netns's address instead of 127.0.0.1 # this ensures all network namespaces can access it networking.resolvconf.extraConfig = '' name_servers="${netAddresses.netns4} ${netAddresses.netns6}" ''; users.users.${config.common.mainUsername}.extraGroups = [ config.services.unbound.group ]; services.unbound = { enable = true; package = pkgs.unbound-full; localControlSocketPath = "/run/unbound/unbound.ctl"; # we override resolvconf above manually resolveLocalQueries = false; settings = { server = rec { interface = [ netAddresses.netns4 netAddresses.netns6 netAddresses.lan4 netAddresses.lan6 ]; access-control = [ "${netCidrs.netns4} allow" "${netCidrs.netns6} allow" "${netCidrs.lan4} allow" "${netCidrs.lan6} allow" ]; aggressive-nsec = true; do-ip6 = true; module-config = ''"validator python iterator"''; local-zone = [ # incompatible with avahi resolver # ''"local." static'' ''"${server-config.server.domainName}." typetransparent'' ]; local-data = builtins.concatLists (map (domain: [ ''"${domain}. A ${serverAddress4}"'' ''"${domain}. AAAA ${serverAddress6}"'' ]) hosted-domains); # incompatible with avahi resolver # ++ [ # ''"retracker.local. A ${netAddresses.lan4}"'' # ''"retracker.local. AAAA ${netAddresses.lan6}"'' # ]; # performance tuning num-threads = 4; # cpu core count msg-cache-slabs = 4; # nearest power of 2 to num-threads rrset-cache-slabs = msg-cache-slabs; infra-cache-slabs = msg-cache-slabs; key-cache-slabs = msg-cache-slabs; so-reuseport = true; msg-cache-size = "50m"; # (default 4m) rrset-cache-size = "100m"; # msg*2 (default 4m) # timeouts unknown-server-time-limit = 752; # default=376 }; # normally it would refer to the flake path, but then the service changes on every flake update # instead, write a new file in nix store python.python-script = builtins.toFile "avahi-resolver-v2.py" (builtins.readFile ./avahi-resolver-v2.py); remote-control.control-enable = true; }; }; environment.etc."unbound/iot_ips.json".text = builtins.toJSON [ # local multicast "224.0.0.0/24" # local broadcast "255.255.255.255" ]; environment.etc."unbound/iot_domains.json".text = builtins.toJSON [ # ntp time sync "pool.ntp.org" # valetudo update check "api.github.com" "github.com" "*.githubusercontent.com" ]; networking.hosts."${serverAddress4}" = hosted-domains; networking.hosts."${serverAddress6}" = hosted-domains; systemd.services.unbound = lib.mkIf config.services.unbound.enable { environment.PYTHONPATH = let unbound-python = pkgs.python3.withPackages (ps: with ps; [ pydbus dnspython requests pytricia nftables ]); in "${unbound-python}/${unbound-python.sitePackages}"; environment.MDNS_ACCEPT_NAMES = "^(.*\\.)?local\\.$"; # resolve retracker.local to whatever router.local resolves to # we can't add a local zone alongside using avahi resolver, so we have to use hacks like this environment.DOMAIN_NAME_OVERRIDES = "retracker.local->router.local"; # load vpn_domains.json and vpn_ips.json, as well as unvpn_domains.json and unvpn_ips.json # resolve domains and append it to ips and add it to the nftables sets # TODO: allow changing family/table name environment.NFT_QUERIES = "vpn:force_vpn4,force_vpn6;unvpn!:force_unvpn4,force_unvpn6;iot:allow_iot4,allow_iot6"; serviceConfig.EnvironmentFile = "/secrets/unbound_env"; # it needs to run after nftables has been set up because it sets up the sets after = [ "nftables-default.service" "avahi-daemon.service" ]; wants = [ "nftables-default.service" "avahi-daemon.service" ]; # allow it to call nft serviceConfig.AmbientCapabilities = [ "CAP_NET_ADMIN" ]; }; systemd.services.update-rkn-blacklist = { # fetch vpn_ips.json and vpn_domains.json for unbound script = '' BLACKLIST=$(${pkgs.coreutils}/bin/mktemp) || exit 1 ${pkgs.curl}/bin/curl "https://reestr.rublacklist.net/api/v2/ips/json/" -o "$BLACKLIST" || exit 1 ${pkgs.jq}/bin/jq ".[0:0]" "$BLACKLIST" && chown unbound:unbound "$BLACKLIST" && mv "$BLACKLIST" /var/lib/unbound/vpn_ips.json ${pkgs.curl}/bin/curl "https://reestr.rublacklist.net/api/v2/domains/json/" -o "$BLACKLIST" || exit 1 ${pkgs.jq}/bin/jq ".[0:0]" "$BLACKLIST" && chown unbound:unbound "$BLACKLIST" && mv "$BLACKLIST" /var/lib/unbound/vpn_domains.json ''; serviceConfig = { Type = "oneshot"; }; }; systemd.timers.update-rkn-blacklist = { wantedBy = [ "timers.target" ]; partOf = [ "update-rkn-blacklist.service" ]; # slightly unusual time to reduce server load timerConfig.OnCalendar = [ "*-*-* 00:00:00" ]; # every day timerConfig.RandomizedDelaySec = 43200; # execute at random time in the first 12 hours }; # run an extra sshd so we can connect even if forwarding/routing between namespaces breaks # (use port 23 because 22 is forwarded to the server) systemd.services.sshd-wan = { description = "SSH Daemon (WAN)"; wantedBy = [ "multi-user.target" ]; after = [ "network.target" "netns-wan.service" ]; bindsTo = [ "netns-wan.service" ]; stopIfChanged = false; path = with pkgs; [ gawk config.programs.ssh.package ]; environment.LD_LIBRARY_PATH = config.system.nssModules.path; restartTriggers = [ config.environment.etc."ssh/sshd_config".source ]; preStart = config.systemd.services.sshd.preStart; serviceConfig = { ExecStart = "${config.programs.ssh.package}/bin/sshd -D -f /etc/ssh/sshd_config -p 23"; KillMode = "process"; Restart = "always"; Type = "simple"; NetworkNamespacePath = "/var/run/netns/wan"; }; }; services.printing = { enable = true; allowFrom = [ "localhost" netCidrs.lan4 netCidrs.lan6 ]; browsing = true; clientConf = '' ServerName router.local ''; defaultShared = true; drivers = [ pkgs.hplip ]; startWhenNeeded = false; }; # share printers (and allow unbound to resolve .local) services.avahi = { enable = true; hostName = "router"; allowInterfaces = [ "br0" ]; publish = { enable = true; addresses = true; domain = true; userServices = true; }; }; services.iperf3 = { enable = true; bind = netAddresses.lan4; }; services.opentracker = { enable = true; extraOptions = "-i ${netAddresses.lan4} -p 6969 -P 6969 -p 80"; }; networking.wireguard.interfaces.wg1 = let parsedAddr4 = router-lib.parseIp4 netAddresses.wg4; parsedAddr6 = router-lib.parseIp6 netAddresses.wg6; in { ips = [ "${addToIp parsedAddr4 0}/24" "${addToIp parsedAddr6 1}/64" ]; listenPort = 854; privateKeyFile = "/secrets/wg1/wg_key"; socketNamespace = "wan"; peers = lib.flip lib.imap0 cfg.wgPubkeys (i: publicKey: { inherit publicKey; allowedIPs = [ "${addToIp parsedAddr4 (1 + i)}/32" "${addToIp parsedAddr6 (1 + i)}/128" ]; presharedKeyFile = "/secrets/wg1/wg_psk${toString i}"; }); }; # I only have 2GB RAM, so Unbound is killed during peak system load without this option zramSwap.enable = true; impermanence.directories = [ # for wireguard key { directory = /secrets; mode = "0000"; } # my custom impermanence module doesnt detect it { directory = /var/db/dhcpcd; mode = "0755"; } { directory = /var/lib/private/kea; mode = "0750"; } ]; }