Why I built my own homebrew Linux router | Hacker News Reader

Why I built my own homebrew Linux router

opensource.com

73 pointsalxsanchez30 comments

I went through building a home network recently. I went into the project planning to build my own pfsense box, but after researching it I decided that I couldn't piece together a custom solution that would both fit my needs and beat an off the shelf ubiquiti edge router in more than one of: price, performance, reliability, power usage, and form factor.

As to the "reliability" point: I love messing with my own networking stuff, but I've learnt the hard way that if I build my infrastructure from scratch, I'll inevitably play around with it too much and it'll break as soon as I desperately need it to just work.

Is Linux the best choice for a router, or would one of the BSDs work better? Obviously the author chose Linux because he was most familiar with it, and that's an excellent reason, but I'm curious what the choice would be if you were starting with a clean slate.

I found pf's syntax to be much clearer and easier to write, so I chose OpenBSD for my home router.

CentOS 7 (and probably some other distributions) contain firewalld which is much easier to configure than iptables. You can easily setup NAT, open and forward ports, so most of typical router tasks are very easy. Though if you encounter untypical task, you have to learn iptables, of course.

For reference it's at least RHEL 7, CentOS 7, Fedora 21+, Debian, Ubuntu, Arch, and Gentoo.

Source: http://www.firewalld.org/documentation/

I've used ferm <http://ferm.foo-projects.org/> to have iptables behave more like pf

It doesn't really matter that much the problem with most homebrew routers will be the hardware you need a pretty decent desktop and some good network cards to match the performance of mid range home routers.

I had a router like that for a couple of years running pfsense which is based on BSD but since London got a 1gbit ISP I had to switch to a normal router unless I wanted to invest like a 1000$ in hardware.

Which ISP is that? (not Virgin, I hope)

Hyperoptic

It doesn't take $1k of hw to run pfsense at 1gbps. I have 1gbps / 1 gbps in Austin, own the company behind pfsense, and run just fine on a 4 core Atom.

Probably doesn't make any difference now but there was a time when some of the operating systems didn't have PPPoE support in the kernel.

Smallwall (http://smallwall.org), the successor of M0n0wall: not many 'moving parts', rock-solid. Saves a lot of time comparing to building it yourself (software-wise).

How can it be faster than two lines

iptables -t nat -A POSTROUTING -o eth0 -j MASQUERADE

iptables -A FORWARD -i eth1 -j ACCEPT

and how does it get security patches more often than Debian or FreeBSD or OpenBSD?

As an answer to the second, I checked the site and the last release was 51 weeks ago and the last beta was 3 weeks ago. Compare that to Debian archive runs a couple times a day, has a couple hundred devs, and a couple mirrors, and full security teams.

Here's a list to security related bugs that Debian has patched since the last beta of smallwall.

https://www.debian.org/security/

There's 27 debian security patches since the most recent smallwall beta, not all related to a firewall installation of course.

Maybe it doesn't need that many patches?

You're comparing the numbers of patches for a distro that's an 11 MB download to one whose total archive weighs in at about a 1.5 TB.

Latency vs bandwidth.

One has large formal teams and procedures and FAST updates.

The other ... doesn't.

If there's a problem found and released today, the first will have a patch in hours. The second will have a new beta every couple weeks, or maybe a stable release every couple years. That's a long time to wait.

Specialization vs. generalization.

One aims to do one thing and do it well (in this case, to provide a means to use an embedded PC as a secure firewall), while the other aims to provide a complete solution to make use of a vast repository of free software for many platforms.

The developer of SmallWall follows the same philosophy put in place by the developer of m0n0wall that preceeded it:

> SmallWall is a firewall, and the purpose of a firewall is to provide security. The more functionality is added, the greater the chance that a vulnerability in that additional functionality will compromise the security of the firewall.

SmallWall is barebones FreeBSD with about 10 utilities strung together to provide the functionality one would expect of a commercial-grade firewall. That's not to say that SmallWall is impenetrable, of course, but it does mean that its attack surface is a few orders of magnitude smaller than Debian's, which attempts to make possible virtually anything that can be done today with free software, on multiple architectures.

You don't have to rush to patch the latest vulnerability in a piece of software if you've made a conscious decision to avoid using that software in the first place. For instance, while people on Debian's security team scrambled to patch Shellshock, m0n0wall was unaffected because it doesn't provide shell access.

For years I was running a Linux “router” for a DSL Line on an old Pentium 3. It was running http://grml.org/ of a Compact-Flash card in an IDE Adapted, mounted read-only.

Worked very, very stable. And e machine came for free from a dumpster.

Running services were pppd, dnsmasqd, NTP, misc cron-jobs and a IPv6 tunnel.

My college roommates and I used to run m0n0wall (which pfSense was forked from) on a Pentium 3, running from a 400MB hard drive from a stack of very old computers in my parents' garage.

I turned it on, assigned the network interfaces, and left it alone for about 9 months at a time. Security wasn't really the point. One device per student was allowed on the dorm network. We had three computer science students living in that room, so obviously that wouldn't fly for long.

I did my own, too. There's more to it than you think.

Firstly my internet provider BT also provides premium rate TV channels via multicast, so that was fun. igmpproxy isn't in the latest couple of Ubuntu releases - i'm currently rewriting it in Golang so to spare others the pain.

Secondly, there's other things you take for granted - upnp for example, which is not great for security by default but my son would kill me if he didn't get decent XBox Live online play.

Then you've got the usual suspects - DNS, DHCP, traffic shaping per device, a DMZ for friends devices without access to my LAN, Unifi controller, the generally shitty quality of Ubuntu packages etc etc.

It's a useful exercise for learning and i'm glad I did it, but it is definitely more work than sticking a Ubuntu install on and enabling ip_forward in sysctl.

I use VyOS (www.vyos.net) for my home router. Linux based, but has a command line interface that something you would expect in a dedicated router platform. Your entire configuration is easily backed up to a text config file that makes setting up the same configuration on new hardware easy.

Still I wished they had a html interface wrapped around it, for convenience.

Ubiquiti maintains a fork of Vyatta (which is related to VyOS in ways I'm admittedly not quite clear on) that also has a web interface. It's for their hardware, but the EdgeRouter series is quite decent in terms of both performance and cost. (The Lite platform will do gigabit line speed for about $100.)

I've used Vyos and was generally impressed. The part where we used Vyos was not exercised to capacity. One thing that confused me was how most (all?) networking hardware of today makes use of custom ASICs (e.g. chips from Broadcom) to get line speed in routing etc. How does Vyos/Vyatta compare? If you had your VMs on the same host as the software router, I understand it is going to be really fast. But if you have a separate box, don't you see a significant slowdown?

That depends on if your network card supports TOE. If your network card supports full TCP offload then your routing latency can be as low as .6-1.4 us. If you have to go through the entire TCP software stack it's around 20-40 us, on 10Gb hardware. Full TCP offload puts it in the same ballpark a hardware router.

Originally discussed at https://news.ycombinator.com/item?id=11514221.

I would do something like this, but I can't seem to find any cheap VDSL2 cards that would work with Centurylink's DSL.

Even if you could, you'd want to keep that blobby PoS away from your PCIe bus. AFAIK, DSL chipsets suffer from the same insecurities as cellphone basebands, from similar desires for secrecy.

You're better off just getting a modem that speaks ethernet or using your current router in bridge mode. Consider the ethernet your demarc point. This will get you some additional galvanic isolation as well.

nice and that's what I do, c1037u is a bit out of date as far as cpu goes though.

why the down vote? you can find better CPUs these days for the same purposes be it gen8 or j1900.

Reply on news.ycombinator.com