- nftables is default to `ip` family which only applies to IPv4. Setting it to `inet` will allow rules to apply to both IPv4 & 6; or `ip6` for IPv6 only. You can skip NAT rules, usually.
- dnsmasq: in addition to DNS and DHCP, turns on router advertisement with SLAAC. Some devices can get IPv6 address from stateful DHCPv6 server, others (e.g. Android) only work with SLAAC.
This is an excellent post and great reference material. I’ve done this a few times before and the information was scattered all over the place. I appreciate the clear and concise writing here. I even added it to my HN favorites - a rare accolade!
One thing I’d add, is that the best explanation I’ve ever seen for this, is the famous diagram [0] on Wikipedia of the netfilter API — I remember when I saw that, everything clicked into place. I’m not sure how up to date it is now, but it’s really good.
Are there any preconfigured images/installers available for a major Linux distro to turn them a router with safe and sensible defaults?
I know there is OpenWrt, but my experience is that is more geared toward running on embedded wifi hardware than an x86 machine. The x86 install comes with a tiny root partition that's actually pretty difficult to resize, for example, and upgrades are quite brittle compared to standard Linux distros.
And there's also pfSense and OPNsense, but these run on FreeBSD which seems to lag behind Linux for hardware support. There's no support for the Aquantia AQC113 NIC, for example (although it looks like this may finally have been added in the last month or so).
Something like an Ubuntu Appliance [1] would be quite nice.
Modified Ubuntu LTS server image will work, and a minimal Debian kernel will have far less bloat. Note pfSense/FreeBSD is fairly robust, and a mature project.
Keep in mind most network appliances have dedicated hardware hand-off adapters, and so the CPU isn't involved in routing once the connection is setup. It is why people can use a $30 SoC, and still be able to saturate several 10Gb/100Gb ports. =3
The Linux box instantly turns into a router as soon as you run `sysctl net.ipv4.ip_forward=1`, because the default policy for FORWARD table is ACCEPT.
You need to explicitly reconfigure the iptables/nftables to prevent that from happening.
Some software, say LXD/Incus, enable forwarding automatically upon installation/startup, and do not configure firewall to block non-their traffic, making the machine an open router. I've reported that, the developers said that's by design (despite other virtualization/containerization systems block forwarding if they happen to enable the sysctl).
Respectfully- I don’t think this statement applies to the scenario I presented.
“The Linux box instantly turns into a router as soon as you run `sysctl net.ipv4.ip_forward=1`, because the default policy for FORWARD table is ACCEPT.”
In the setup I presented, we are bridging an Ethernet and a WiFi network. This would be desirable if you wanted to use an upstream dhcp server for your WiFi clients- or if you wanted to avoid double nat’ing.
In 802.11 infrastructure mode, a station can only send frames with its own MAC address. The AP won’t accept or forward frames from unknown MACs. So you can’t transparently bridge Ethernet devices’ MAC addresses through a WiFi client interface. This is why we need hostapd.
In every other circumstance- I think your statement holds.
I tried to do some weird alerting on new MAC addresses and ran into this weirdness. Bridging WiFi and Ethernet gets weird.
The risk is minimal if you control or trust both networks. A network boundary is a natural choke point for access control, so that's where it's usually implemented. For an ipv4 boundary router (as is the topic of the post) you almost certainly need to configure Network Address Translation because your internal network addresses are non-routable on the Internet (at uni my dorm had public IP addresses for each student computer, fun times).
As for the GP's example, running VM's or containers* on your own machine? I'd say the default ACCEPT policy is fine. However, silently changing such a setting on software installation is a problem because if the machine is multi-homed (i.e. has more than one network interface), you've now created a network route outside of the network admin's control.
* The default for docker and podman is to use a private network, not a bridge anyway.
Basically you're introducing a hole. For example, if you have some devices in your network (like a dodgy TV box) that are not supposed to reach the internet or other parts of the network, the computer with net.ipv4.ip_forward=1 could be used as a pivot. Depending on the routing tables you probably would also need to enable IP masquerading (NAT) to allow bidirectional communication.
Used to run a virtualized firewall setup. And then one day discovered that somewhere along the lines I had made a change (or an update changed something) that meant proxmox admin interface was being served publicly. That's despite confirming during initial setup that it isn't.
So now I do not do any funky stuff with firewalls anymore. Separate appliance with opnsense bare metal.
My router is a 16GB n150 mini PC with dual NICs. The actual router OS is within openwrt VM managed by Incus (VM/Container hypervisor) that has both NICs passed through.
One of the NICs is connected to another OpenWrt wifi access point, and the other is connected to the ISP modem.
The n150 also has a wifi card that I setup as an additional AP I can connect to if something goes wrong with the virtualization setup.
Been running this for at least 6 months and has been working pretty well.
Both port specific firewall rules, and web-server IP permissions are important.
For example, bandwidth rate-limiting may be inhibited for admin SSH or package updates, and LAN IPv4 private ranges for your host address pool are set.
Finally, your internal DHCP should statically bind your admin computer MAC to a fixed LAN host IP to further reduce issues.
Personally, I always build my NAS from scratch, as I have lost count of the number of problems web-GUI have caused over the years. =3
Dedicated appliances are the way to go for the most important parts of your setup. I've always had my router as an appliance because I don't like the idea of my network failing due to something going wrong with the server that runs a bunch of other things that are less important. I also have Home Assistant running on a dedicated machine because that's also important.
Btw you do also need to be careful with opnsense. I was years behind on updates for mine because every time I updated I assumed that it would bring me up to date with the latest version. But opnsense has to install the upgrades in order. After you reboot you need to check again for updates and repeat until there's no more to install.
I don't bother with virtualization, and use the machine at the edge of my network as router, email server, Web server, DNS server, and countless or other things such as hostapd.
An x86 mini PC can run all this without breaking a sweat; using separate appliances seems very wasteful. That being said, I configure everything in DIY mode, and don't rely on GUIs or other similar things that increase the attack surface considerably.
I used to try to combine everything, but now I don't. Separate appliances isolates issues to a subset of services. If everything is on a single PC and that one dies or even just needs a reboot, everything goes down.
Fair enough and I think you have done the right thing - opnsense is pretty decent - and the clear delineation between collision domains helps avoid showing too much ankle to the internet 8)
I think your initial setup was perfectly valid. Then you diagnosed a fault and fixed it with aplomb, in a way that you could verify. The key point is: "in a way you could verify" and you failed safe. Well played.
Proxmox itself has a useful firewall implementation too, although it takes a bit of getting used to because you can set it at the cluster, host and VM levels. I personally love it because it is easier to manage than individual host based firewalls, which I also do, but I'm a masochist! For smaller systems I generally use the cluster level to keep all the rules in one place.
Stories like these make me sad when I think about chat control and age verification. Kids in the future may no longer be allowed to talk to random helpful strangers on the internet about computers and other technical topics, because apparently the internet is too dangerous for children.
Aside from the fact that "a CPE" is grammatically incorrect, you are also semantically wrong. A router is any device connected to multiple networks that can forward packets between them; and consumer-premises equipment includes everything that's directly connected or consumes a service from a telecom provider. Landline phones, set-top boxes and satellite decoders are also examples of CPE.
It's like me stating "you're not a man, you're a human!" and then expecting you to be in awe of my profound wisdom.
Technically it's an IPv4 router once you enable net.ipv4.ip_forward in step 1, the rest is enabling a whole lot of supplementary services and operations not intrinsic to the definition of a router.
Thank you for informing me that a novel definition of the term "router" has come along since the last time I turned a Linux box into a router. The world changes in strange ways sometimes!
What is "CPE" in this context? It's probably not "Common Platform Enumeration" (my top results for "cpe linux") or "Customer-Premises Equipment." ("cpe networking")
> CPE generally refers to devices such as telephones, routers, network switches, residential gateways (RG), set-top boxes, fixed mobile convergence products, home networking adapters and Internet access gateways that enable consumers to access providers' communication services and distribute them in a residence or enterprise with a local area network (LAN).
I think a CPE could (be/include) a router, but usually it refers to the demarc between the provider's network and the customer's (no matter who owns/manages it).
For a Linux box to be a true CPE you'd likely need somewhat of a specialized card, one that can communicate directly to the next device up the line (e.g, take commercial fiber or cable in, ISDN modem, etc).
If it just shoots out ethernet into some other box next to it, it's likely not a CPE.
Plenty of isps that provide internet over regular ethernet. But it's a ye olde telecom provider term that referred to the phone, that you also didnt own yourself. Doesn't always apply cleanly these days.
Usually it's "something else" that turns into RJ45 (as ethernet has a maximum length) - now if you're in a datacenter you likely can get raw RJ45 Internet).
> CPE generally refers to devices such as telephones, routers, network switches, residential gateways (RG), set-top boxes, fixed mobile convergence products, home networking adapters and Internet access gateways that enable consumers to access providers' communication services
From my understanding any type of device that is used to extend or facilitate provider services is a CPE. So a router just acting as an extender would still be a cpe, as would a modem, as would anything that is on the customer side and facilitates provider services. Only situation a router wouldn't be a cpe is if it was just for a local lan network.
I didn't see in TFA --although I may have missed it-- where it said it was replacing the ISP's router/CPE. Anything routing traffic is a router.
At home I've got both a CPE given by my ISP and my own router that routes and bridges traffic between two LANs of mine (192. and 10.).
Moreover the lack of IPv6 inside our own LANs is, for many of us, a feature. It doesn't mean we don't have an IPv6 address: it just means we have the choice and did choose to have our own LANs on IPv4 only. And, no, I don't care that it makes some programmers at some megacorp' lives more difficult to "reach" inside my networks.
I'm the boss at my home and my router is IPv4 only.
- net.ipv6.conf.all.forwarding=1
- nftables is default to `ip` family which only applies to IPv4. Setting it to `inet` will allow rules to apply to both IPv4 & 6; or `ip6` for IPv6 only. You can skip NAT rules, usually.
- dnsmasq: in addition to DNS and DHCP, turns on router advertisement with SLAAC. Some devices can get IPv6 address from stateful DHCPv6 server, others (e.g. Android) only work with SLAAC.
One thing I’d add, is that the best explanation I’ve ever seen for this, is the famous diagram [0] on Wikipedia of the netfilter API — I remember when I saw that, everything clicked into place. I’m not sure how up to date it is now, but it’s really good.
[0] https://commons.wikimedia.org/wiki/File:Netfilter-packet-flo...
I know there is OpenWrt, but my experience is that is more geared toward running on embedded wifi hardware than an x86 machine. The x86 install comes with a tiny root partition that's actually pretty difficult to resize, for example, and upgrades are quite brittle compared to standard Linux distros.
And there's also pfSense and OPNsense, but these run on FreeBSD which seems to lag behind Linux for hardware support. There's no support for the Aquantia AQC113 NIC, for example (although it looks like this may finally have been added in the last month or so).
Something like an Ubuntu Appliance [1] would be quite nice.
[1] https://ubuntu.com/appliance
Keep in mind most network appliances have dedicated hardware hand-off adapters, and so the CPU isn't involved in routing once the connection is setup. It is why people can use a $30 SoC, and still be able to saturate several 10Gb/100Gb ports. =3
You need to explicitly reconfigure the iptables/nftables to prevent that from happening.
Some software, say LXD/Incus, enable forwarding automatically upon installation/startup, and do not configure firewall to block non-their traffic, making the machine an open router. I've reported that, the developers said that's by design (despite other virtualization/containerization systems block forwarding if they happen to enable the sysctl).
- router
- NAT gateway
- DHCP server
In a typical scenario, turning IP forwarding on will do nothing unless:
- DHCP has given the devices on the 'inside' IP addresses and told them the gateway address, and
- the router is set up to do IP masquerading
“The Linux box instantly turns into a router as soon as you run `sysctl net.ipv4.ip_forward=1`, because the default policy for FORWARD table is ACCEPT.”
In the setup I presented, we are bridging an Ethernet and a WiFi network. This would be desirable if you wanted to use an upstream dhcp server for your WiFi clients- or if you wanted to avoid double nat’ing.
In 802.11 infrastructure mode, a station can only send frames with its own MAC address. The AP won’t accept or forward frames from unknown MACs. So you can’t transparently bridge Ethernet devices’ MAC addresses through a WiFi client interface. This is why we need hostapd.
In every other circumstance- I think your statement holds.
I tried to do some weird alerting on new MAC addresses and ran into this weirdness. Bridging WiFi and Ethernet gets weird.
As for the GP's example, running VM's or containers* on your own machine? I'd say the default ACCEPT policy is fine. However, silently changing such a setting on software installation is a problem because if the machine is multi-homed (i.e. has more than one network interface), you've now created a network route outside of the network admin's control.
* The default for docker and podman is to use a private network, not a bridge anyway.
Setting that up with safe/fair bandwidth-sharing requires intermediate IT skill level. Still a great hobby project =3
So now I do not do any funky stuff with firewalls anymore. Separate appliance with opnsense bare metal.
My router is a 16GB n150 mini PC with dual NICs. The actual router OS is within openwrt VM managed by Incus (VM/Container hypervisor) that has both NICs passed through.
One of the NICs is connected to another OpenWrt wifi access point, and the other is connected to the ISP modem.
The n150 also has a wifi card that I setup as an additional AP I can connect to if something goes wrong with the virtualization setup.
Been running this for at least 6 months and has been working pretty well.
For example, bandwidth rate-limiting may be inhibited for admin SSH or package updates, and LAN IPv4 private ranges for your host address pool are set.
Finally, your internal DHCP should statically bind your admin computer MAC to a fixed LAN host IP to further reduce issues.
Personally, I always build my NAS from scratch, as I have lost count of the number of problems web-GUI have caused over the years. =3
Btw you do also need to be careful with opnsense. I was years behind on updates for mine because every time I updated I assumed that it would bring me up to date with the latest version. But opnsense has to install the upgrades in order. After you reboot you need to check again for updates and repeat until there's no more to install.
An x86 mini PC can run all this without breaking a sweat; using separate appliances seems very wasteful. That being said, I configure everything in DIY mode, and don't rely on GUIs or other similar things that increase the attack surface considerably.
I think your initial setup was perfectly valid. Then you diagnosed a fault and fixed it with aplomb, in a way that you could verify. The key point is: "in a way you could verify" and you failed safe. Well played.
Proxmox itself has a useful firewall implementation too, although it takes a bit of getting used to because you can set it at the cluster, host and VM levels. I personally love it because it is easier to manage than individual host based firewalls, which I also do, but I'm a masochist! For smaller systems I generally use the cluster level to keep all the rules in one place.
As a kid with no AI, no google, it was quite a feat and I’m still very proud of it
Was my introduction into how the internet works and I’ll never forget working with ipchains
I remember enduring a lot of people in forums calling me a noob, but only after spending collective hours answering my dumb questions
I credit a big part of my moderate success in tech, to being familiar with stuff at just a tad bit lower of a level than the average bear
To my friend Sam who I haven’t talked to in 20 years, thanks for the idea
It's like me stating "you're not a man, you're a human!" and then expecting you to be in awe of my profound wisdom.
IPv6 support is not required for a router. You'll note they also fail to offer IPX/SPX or ATM and many more.
People who use this term are in telco.
https://en.wikipedia.org/wiki/Customer-premises_equipment
Given that the Wikipedia definition of CPE includes routers, I don't see how calling it CPE precludes it being a router, as the poster claimed:
> That's not a router, that's a CPE, and one without IPv6 support
For a Linux box to be a true CPE you'd likely need somewhat of a specialized card, one that can communicate directly to the next device up the line (e.g, take commercial fiber or cable in, ISDN modem, etc).
If it just shoots out ethernet into some other box next to it, it's likely not a CPE.
I've never heard of that before. How does that work? Your ISP would always have to have some infrastructure within 100 meters of your router then.
> CPE generally refers to devices such as telephones, routers, network switches, residential gateways (RG), set-top boxes, fixed mobile convergence products, home networking adapters and Internet access gateways that enable consumers to access providers' communication services
From my understanding any type of device that is used to extend or facilitate provider services is a CPE. So a router just acting as an extender would still be a cpe, as would a modem, as would anything that is on the customer side and facilitates provider services. Only situation a router wouldn't be a cpe is if it was just for a local lan network.
At home I've got both a CPE given by my ISP and my own router that routes and bridges traffic between two LANs of mine (192. and 10.).
Moreover the lack of IPv6 inside our own LANs is, for many of us, a feature. It doesn't mean we don't have an IPv6 address: it just means we have the choice and did choose to have our own LANs on IPv4 only. And, no, I don't care that it makes some programmers at some megacorp' lives more difficult to "reach" inside my networks.
I'm the boss at my home and my router is IPv4 only.
And I've got that in addition to my ISP's CPE.
IMO, it’s a plain old router/switch/bridge.