Difference between revisions of "Router"

Revision as of 11:50, 15 February 2014

This article is a tutorial for turning a computer into an internet gateway/router. It focuses on security, since the gateway is connected directly to the Internet. It should not run any services available to the outside world. Towards the LAN, it should only run gateway specific services. It should not run httpd, ftpd, samba, nfsd, etc. as those belong on a server in the LAN as they introduce security flaws.

This article does not attempt to show how to set up a shared connection between 2 PCs using cross-over cables. For a simple internet sharing solution, see Internet sharing.

Hardware Requirements

• At least 1 GB of hard drive space. The base install will take up around 500MB of space and if you want to use a caching web proxy, you will need to reserve space for the cache as well.
• At least two physical network interfaces: a gateway connects two networks with each other. You will need to be able to connect those networks to the same physical computer. One interface must connect to the external network, while the other connects to the internal network.
• A hub, switch or UTP cable: You need a way to connect the other computers to the gateway

Conventions

Conventions in this guide will be to use non-realistic interface names, to avoid confusion about which interface is which.

• intern0: the network card connected to the LAN. On an actual computer it will probably have the name enp2s0, enp1s1, etc.
• extern1: the network card connected to the external network (or WAN). It will probably have the name enp2s0, enp1s1, etc.

Installation

A fresh install of Arch Linux is the easiest to start from, as no configuration changes have been made and there is a minimal amount of packages installed. This is helpful when attempting to reduce security risk.

Partitioning

For security purposes, /var, /tmp and /home should be separate from the / partition. This prevents disk space from being completely used up by log files, daemons or the unprivileged user. It also allows different mount options for those partitions. If you have already partitioned your drive, the gparted livecd can be used to resize, move, or create new partitions.

Your home and root partitions can be much smaller than a regular install since this is not a desktop machine. /var should be the largest partition—it is where databases, logs and long-term caches are stored. If you have a lot of RAM, mounting /tmp as tmpfs is a good idea, so making a disk partition for it during the initial install is unnecessary. Note that /tmp is mounted as tmpfs by default in Arch.

Post-Installation

After creation of non-root account you are recommended to install sudo and disable root login.

Network interface configuration

Persistent naming and Interface renaming

Systemd automatically chooses unique interface names for all your interfaces. These are persistent and will not change when you reboot. If you would like to rename interface to user friendlier names read Network_Configuration#Device_names.

IP configuration

Now you will need to configure the network interfaces. The best way to do so is using netctl profiles. You will need to create two profiles.

• /etc/netctl/extern0-profile

Description='Public Interface.'
Interface=extern0
Connection=ethernet
IP='dhcp'

• /etc/netctl/intern0-profile

Description='Private Interface'
Interface=intern0
Connection=ethernet
IP='static'
Address=('10.0.0.1/24')

Next up is to set up the interfaces with netctl.

# netctl enable extern0-profile
# netctl enable intern0-profile

ADSL connection/PPPoE

Using rp-pppoe, we can connect an ADSL modem to the extern1 of the firewall and have Arch manage the connection. Make sure you put the modem in bridged mode though (either half-bridge or RFC1483), otherwise the modem will act as a router too.

# pacman -S rp-pppoe

It should be noted that if you use only PPPoE to connect to the internet (ie. you do not have other WAN port, except for the one that connects to your modem) you do not need to set up the extern0-profile as the external pseudo-interface will be ppp0.

PPPoE configuration

You can use netctl to setup the pppoe connection. To get started

# cp /etc/netctl/examples/pppoe /etc/netctl/

and start editing. For the interface configuration choose the interface that connects to the modem. If you only connect to the internet through PPPoE this will probably be extern0. Fill in the rest of the fields with your ISP information. See the pppoe section in netctl.profile man page for more information on the fields.

DNS and DHCP

We will use dnsmasq, a DNS and DHCP daemon for the LAN. It was specifically designed for small sites. To get it, install dnsmasq from the official repositories.

Dnsmasq needs to be configured to be a DHCP server. To do this:

Edit /etc/dnsmasq.conf:

interface=intern0 # make dnsmasq listen for requests only on intern0 (our LAN)
expand-hosts      # add a domain to simple hostnames in /etc/hosts
domain=foo.bar    # allow fully qualified domain names for DHCP hosts (needed when
                  # "expand-hosts" is used)
dhcp-range=10.0.0.2,10.0.0.255,255.255.255.0,1h # defines a DHCP-range for the LAN: 
                  # from 10.0.0.2 to .255 with a subnet mask of 255.255.255.0 and a
                  # DHCP lease of 1 hour (change to your own preferences)

Somewhere below, you will notice you can also add "static" DHCP leases, i.e. assign an IP-address to the MAC-address of a computer on the LAN. This way, whenever the computer requests a new lease, it will get the same IP. That is very useful for network servers with a DNS record. You can also deny certain MAC's from obtaining an IP.

Now start dnsmasq:

# systemctl start dnsmasq.service

Connection sharing

Time to tie the two network interfaces to each other.

iptables

Simple stateful firewall documents the setup of an iptables firewall and NAT.

Shorewall

Shorewall, an iptables frontend, can be used as an easier alternative. See Shorewall for detailed configuration.

IPv6

Useful reading: IPv6 and the Wikipedia IPv6 entry.

You can use your router in IPv6 mode even if you do not have an IPv6 address from your ISP. Unless you disable IPv6 all interfaces should have been assigned a unique fe80::/10 address.

For internal networking the block fc00::/7 has been reserved. These addresses are guaranteed to be unique and non-routable from the open internet. Addresses that belong to the fc00::/7 block are called Unique Local Addresses. To get started generate a ULA /64 block to use in your network. For this example we will use fd00:aaaa:bbbb:cccc::/64. Firstly we must assign a static IPv6 on the internal interface. Modify the intern0-profile we created above to include the following line

IPCustom=('-6 addr add fd00:aaaa:bbbb:cccc::1/64 dev intern0')

This will add the ULA to the internal interface. As far as the router goes, this is all you need to configure.

Router Advertisement and Stateless Autoconfiguration (SLAAC)

To properly hand out IPv6s to the network clients we will need to use an advertising daemon. The standard tool for this job is radvd and is available in official repositories. Configuration of radvd is fairly simple. Edit /etc/radvd.conf to include

interface intern0 {
  AdvSendAdvert on;
  MinRtrAdvInterval 3;
  MaxRtrAdvInterval 10;
  prefix fd00:aaaa:bbbb:cccc::/64 {
    AdvOnLink on;
    AdvAutonomous on;
    AdvRouterAddr on;
  };
};

The above configuration will tell clients to autoconfigure themselves using addresses from the specified /64 block. Addresses on the clients are uniquely generated using the MAC address of the connected interface and are optionally mangled for security reasons if privacy extensions are enabled (which is recommended to do). On the client side you need to enable IP6=stateless in your active netctl profile. If you want a static IP as well add

IPCustom=('-6 addr add fd00:aaaa:bbbb:cccc::2/64 dev eth0')

Don't forget to enable radvd.service

Firewall tweaks

Stateless autoconfiguration works on the condition that IPv6 icmp packets are allowed throughout the network. So some firewall tweaks are required on both ends of the network for it to work properly. On the client side all you need to do is allow the ipv6-icmp protocol on the INPUT chain. If you are using Simple Stateful Firewall you only need to add

-A INPUT -p ipv6-icmp -j ACCEPT

You can limit it to internal network using -s fd00:aaaa:bbbb:cccc::/64 and/or -s fe80::/10 if you feel it is a security threat. Additionally you must add the same rules to your router firewall but extending it to the OUTPUT and FORWARD chains as well.

-A INPUT -p ipv6-icmp -j ACCEPT
-A OUTPUT -p ipv6-icmp -j ACCEPT
-A FORWARD -p ipv6-icmp -j ACCEPT

Again, you can limit it to the internal network for the INPUT chain.

Global Unicast Addresses

Static WAN IPv6

If your ISP or WAN network can access the IPv6 Internet you can assign global link addresses to your router and propagate them through SLAAC to your internal network. If you can use a Static IPv6 all you must do is add it to your external profile and enable it the advertisement of the global unicast block in radvd.conf.

In /etc/netctl/extern0-profile simply add the IPv6 and the IPv6 prefix (usually /64) you have been provided

IPCustom=('-6 addr add 2002:1:2:3:4:5:6:7/64 dev extern0')

and edit /etc/radvd.conf to include the new advertisement block.

interface intern0 {
  AdvSendAdvert on;
  MinRtrAdvInterval 3;
  MaxRtrAdvInterval 10;
  prefix fd00:aaaa:bbbb:cccc::/64 {
    AdvOnLink on;
    AdvAutonomous on;
    AdvRouterAddr on;
  };
  prefix 2002:1:2:3::/64 {
    AdvOnLink on;
    AdvAutonomous on;
    AdvRouterAddr on;
  };
};

In that way your internal network clients will also get a Global IPv6 address. This IP is routable from the open internet, so adjust your firewalls. Please note that global and local IPv6s can co-exist on the same interface without further configuration.

Acquiring WAN IPv6 via DHCPv6-PD

If your ISP handles out IPv6s using DHCPv6-PD you will need to use a DHCPv6 client to get the IP from your ISP. Common such programs are dibbler^AUR, wide-dhcpv6^AUR and dhcpcd. ISC's dhclient should also work, but documentation on prefix delegation is scarce.

For dibbler edit /etc/dibbler/client.conf

log-mode short
log-level 7
iface "extern0" {
  ia
  pd
}

For wide-dhcpv6 edit /etc/wide-dhcpv6/dhcp6c.conf

interface extern0 {
  send ia-pd 0;
};
 
id-assoc pd 0 {
  prefix-interface intern0 {
    sla-id 1;
    sla-len 8;
  };
};

To enable/start wide-dhcpv6 client use the command

# systemctl enable/start dhcp6c@extern0.service

For dhcpcd edit /etc/dhcpcd.conf. You might already be using dhcpcd for IPv4 so just update your existing configuration. If you would like to use it for IPv6 only uncomment the third line.

duid
noipv6rs
#ipv6only
interface extern0
ia_pd 1 intern0

This configuration will ask for a prefix from WAN (interface extern0) and delegate it to the internal interface (intern0).

Because the IPv6 prefix is now dynamic, we need to change radvd to advertize any subnet instead of specific ones. With this configuration radvd will pick any /64 prefix available on the internal interface and propagate SLAAC IPv6s to the clients. Simply change /etc/radvd.conf to

interface intern0 {
  AdvSendAdvert on;
  MinRtrAdvInterval 3;
  MaxRtrAdvInterval 10;
  prefix ::/64 {
    AdvOnLink on;
    AdvAutonomous on;
    AdvRouterAddr on;
    DeprecatePrefix on;
  };
};

PPPoE and IPv6

If your ISP provides IPv6 via PPPoE you can enable it in your pppoe netctl profile. Just add this to pppoe netctl profile

PPPoEIP6=yes

and restart it. Also you must change any extern0 references to the configuration files above to ppp0 instead since IPv6 is assigned to ppp pseudo-interface instead of a real ethernet interface. Please note, that depending on your modem IPv6 might not be available through half-bridge so switch to full RFC1483 bridging instead.

Cleanup

Now that the installation has been performed, it is necessary to remove as many packages as possible. Since we are making a gateway, keeping unneeded packages only "bloats" the system, and increases the number of security risks.

First, check for obsolete/deprecated packages (likely after a fresh install and massive series of updates):

$ pacman -Qm

Review the list of explicitly installed packages that are not dependencies and remove any that are unneeded. Having only needed packages installed is an important security consideration.

$ pacman -Qet

Completely remove the packages you do not need along with their configuration files and dependencies:

# pacman -Rsn package1 package2 package3

Logrotate

You should review the logrotate configuration to make sure the box is not brought down by lack of diskspace due to logging.

Logrotate is installed by default, so you will not have to install it.

Optional additions

UPnP

The above configuration of shorewall does not include UPnP support. Use of UPnP is discouraged as it may make the gateway vulnerable to attacks from within the LAN. However, some applications require this to function correctly.

To enable UPnP on your router, you need to install an UPnP Internet gateway daemon (IGD). To get it, install miniupnpd from the official repositories.

Read the Shorewall guide on UPnP for more information

Remote administration

OpenSSH can be used to administer your router remotely. This is useful for running it "headless" (no monitor or input devices).

Caching web proxy

See Squid or Polipo for the setup of a web proxy to speed up browsing and/or adding an extra layer of security.

Time server

To use the router as a time server, see Network Time Protocol.

Then, configure shorewall or iptables to allow NTP traffic in and out.

Content filtering

Install and configure DansGuardian or Privoxy if you need a content filtering solution.

Traffic shaping

Traffic shaping is very useful, especially when you are not the only one on the LAN. The idea is to assign a priority to different types of traffic. Interactive traffic (ssh, online gaming) probably needs the highest priority, while P2P traffic can do with the lowest. Then there is everything in between.

Traffic shaping with shorewall

Read Shorewall's Traffic Shaping/Control guide.

Here is my config as an example:

• /etc/shorewall/tcdevices : here is where you define the interface you want to have shaped and its rates. I have got a ADSL connection with a 4MBit down/256KBit up profile.

ppp0        4mbit        256kbit 

• /etc/shorewall/tcclasses : here you define the minimum (rate) and maximum (ceil) throughput per class. You will assign each one to a type of traffic to shape.

# interactive traffic (ssh)
ppp0            1       full    full    0
# online gaming
ppp0            2       full/2  full    5
# http
ppp0            3       full/4  full    10
# rest
ppp0            4       full/6  full    15              default

• /etc/shorewall/tcrules : this file contains the types of traffic and the class it belongs to.

1       0.0.0.0/0       0.0.0.0/0       tcp     ssh
2       0.0.0.0/0       0.0.0.0/0       udp     27000:28000
3       0.0.0.0/0       0.0.0.0/0       tcp     http
3       0.0.0.0/0       0.0.0.0/0       tcp     https

I have split it up my traffic in 4 groups:

1. interactive traffic or ssh: although it takes up almost no bandwidth, it is very annoying if it lags due to leechers on the LAN. This get the highest priority.
2. online gaming: needless to say you ca not play when your ping sucks. ;)
3. webtraffic: can be a bit slower
4. everything else: every sort of download, they are the cause of the lag anyway.

Intrusion detection and prevention with snort

See Snort.

See also

• Simple stateful firewall
• Internet sharing