The basic procedure is as follows:
Throughout this procedure, I will assume a setup with two ethernet (NIC) cards, an outside link via DSL (where a DSL modem connects to one of the NIC's), and a local net that connects to the other NIC. I will arbitrarily call the NIC to the DSL modem "eth1" and the local net NIC "eth0". The device naming by the kernel of the NIC's depends on what slot they are in.
I will assume that you have been assigned a subnet of IP addresses at 192.168.2.128-191, i.e. a netmask of 255.255.255.192, and the router provided by the DSL company is at 192.168.2.129. These are all arbitrary fictional examples to illustrate the setup. I will use the address 192.168.2.130 for the firewall machine (both NIC's), though it turns out you can also use distinct IP addresses for each NIC if you want.
You will need two ethernet cards to make this work. The biggest problem I had was that I randomly picked a slot in my motherboard for the second NIC and it turned out that that slot (PCI) shared an interrupt with the first NIC. I did not know that this was a problem (in fact there is little information about this, and I thought it should work fine). It caused both cards to shut down quietly (no error indication) and stop sending and receiving packets. Naturally when you are doing all sort of configuration changes, this is the last thing you need. I do not know if this is a problem with all PCI NIC cards or just ours, but I would advise against sharing interrupts. The tulip driver, which we use, reports the IRQ for each NIC in syslog when you boot. There is a bunch of information out there (see the Ethernet-HOWTO section Using More than one Ethernet Card per Machine) about making the kernel recognize two ethernet cards using boot arguments; however, I did not need this (my kernel recognized both cards with no arguments).
Next, you need to hook the second NIC to the DSL modem (or whatever links you to the outside world) and make sure that it is working. You should be able to ifconfig the second ethernet card to a proper IP address and ping the router on the other end of your outside link. This verifies that you can send and receive packets over the DSL link. For instance, for the sample net you would do:
ifconfig eth1 192.168.2.130 netmask 255.255.255.192 broadcast 192.168.2.191
to configure the NIC. And then
ifconfig eth0 down # just to make sure it does not interfere with things ping 192.168.2.129
to test that you can get to the router. For good measure, you should also test that you can get to the machines on your local network through the other NIC:
ifconfig eth1 down # just to make sure it does not interfere with things ifconfig eth0 up ping 192.168.2.x # where x is the address for a machine on your local net
At this point, you have verified that all the hardware is working.
Depending upon your kernel version you will need either the old bridge configuration utility (BRCFG) for kernels before 2.2.14, or the new bridge configuration utility (bridgex) for later kernels; these utilities allow you to control the bridging in your kernel when CONFIG_BRIDGE is turned on. BRCFG is distributed as source with pre-compiled executables. I do not know what kernel the executable was compiled with, but I got different results after I recompiled it with my kernel (2.2.13) include files. Unfortunately, to do this I had to patch them slightly. Here are the patches:
diff -C 3 -r /tmp/BRCFG/brcfg.c ./brcfg.c *** /tmp/BRCFG/brcfg.c Wed Feb 21 19:11:59 1996 --- ./brcfg.c Wed Dec 8 12:52:23 1999 *************** *** 1,6 **** ! #include <sys/types.h> ! #include <sys/socket.h> #include <skbuff.h> #include "br.h" --- 1,6 ---- ! #include <types.h> ! #include <socket.h> #include <skbuff.h> #include "br.h"
Apply the patch, recompile brcfg and install it somewhere appropriate (I chose /usr/sbin).
For kernels later than 2.2.13 you definitely want to use the newer bridge configuration utility bridgex. I am not sure if it works with earlier kernels or not. Not that the URL for this utility is found in the kernel configuration help file /usr/src/linux/Documentation/Configure.help, so if the URL mentioned here is not correct, look in the help file (it is the help for the CONFIG_BRIDGE kernel configuration item. The bridgex tarball contains an already compiled executable, but you should probably remake it using the included Makefile. Note that the bridgex utility takes slightly different arguments than does the BRCFG package (that will be covered later when I talk about configuring the bridge).
You will need to patch and configure your kernel for bridging and the bridging filter (as well as firewalling, networking, etc. if you do not already have it). The following kernel configuration items will be needed (at least):
CONFIG_EXPERIMENTAL=y CONFIG_BRIDGE=y CONFIG_FIREWALL=y CONFIG_IP_FIREWALL=y
You should grab the Bridge Filter Patch and apply it to your kernel. Recompile and install your kernel and then reboot.
So you should have your two NIC's working, a newly configured kernel, and brcfg installed. Now you need to construct a startup script to put it all together. I did this using the RedHat type startup scripts (/etc/rc.d). I put specific network addresses and masks in /etc/sysconfig/network:
GATEWAY=192.168.2.129 # the address of the DSL router
GATEWAYDEV=eth1 # the NIC that the router is connected to
ETH0_ADDR=192.168.2.130 # the IP address for the NIC on our LAN
ETH0_MASK=255.255.255.192 # the netmask of our LAN
ETH0_BROAD=192.168.2.191 # the broadcast address of our LAN
ETH1_ADDR=192.168.2.130 # the IP address for the NIC on the DSL side
# can be different from ETH0_ADDR if you want
ETH1_MASK=$ETH0_MASK # the DSL side netmask, should be the same as eth0
ETH1_BROAD=$ETH1_BROAD # ditto for the broadcast address
Next I created a script in /etc/rc.d/init.d/bridge to setup the bridge. I include two scripts here. The first script is used with the old BRCFG utility, the second for the newer bridgex. First the one for the older BRCFG:
#!/bin/sh
#
# bridge This shell script takes care of installing bridging for dsl with BRCFG
#
# description: Uses brcfg to start bridging and ifconfigs eths
# processname: bridge
# config:
# Source function library.
. /etc/rc.d/init.d/functions
# Source networking configuration.
. /etc/sysconfig/network
# See how we were called.
case "$1" in
start)
echo -n "Configuring bridge: "
ifconfig eth0 $ETH0_ADDR netmask $ETH0_MASK broadcast $ETH0_BROAD
ifconfig eth1 $ETH1_ADDR netmask $ETH1_MASK broadcast $ETH1_BROAD
route add $GATEWAY dev $GATEWAYDEV
route add default gw $GATEWAY dev $GATEWAYDEV
ifconfig eth0 promisc
ifconfig eth1 promisc
brcfg -enable
echo
;;
stop)
# Stop daemons.
brcfg -disable
ifconfig eth0 down
ifconfig eth1 down
;;
restart)
$0 stop
$0 start
;;
status)
ifconfig eth0
ifconfig eth1
brcfg
;;
*)
echo "Usage: bridge {start|stop|restart|status}"
exit 1
esac
exit 0
The next script is the one to use with the newer bridge configuration utility bridgex. Note that bridgex is much more configurable than the older BRCFG and so you may want to look man page included with the bridgex tarball and custom configure this script:
#!/bin/sh
#
# bridge This shell script takes care of installing bridging for dsl with BRCFG
#!/bin/sh
#
# bridge This shell script takes care of installing bridging for dsl with bridgex
#
# description: Uses brcfg to start bridging and ifconfigs eths
# processname: bridge
# config:
# Source function library.
. /etc/rc.d/init.d/functions
# Source networking configuration.
. /etc/sysconfig/network
# See how we were called.
case "$1" in
start)
echo -n "Configuring bridge: "
ifconfig eth0 $ETH0_ADDR netmask $ETH0_MASK broadcast $ETH0_BROAD
ifconfig eth1 $ETH1_ADDR netmask $ETH1_MASK broadcast $ETH1_BROAD
route add default gw $GATEWAY dev $GATEWAYDEV
ifconfig eth0 promisc
ifconfig eth1 promisc
brcfg start
brcfg device eth0 enable
brcfg device eth1 enable
echo
;;
stop)
# Stop daemons.
brcfg stop
ifconfig eth0 down
ifconfig eth1 down
;;
restart)
$0 stop
$0 start
;;
status)
ifconfig eth0
ifconfig eth1
brcfg
;;
*)
echo "Usage: bridge {start|stop|restart|status}"
exit 1
esac
exit 0
The script is run during bootup. It assigns addresses to each NIC, adds a default route that goes to the DSL router, adds a specific route direct to the DSL router, puts each NIC in "promiscuous" mode, and then enables bridging. I linked this script into the following directories in /etc/rc.d:
/etc/rc.d/rc0.d/K90bridge /etc/rc.d/rc1.d/K90bridge /etc/rc.d/rc2.d/S11bridge /etc/rc.d/rc3.d/S11bridge /etc/rc.d/rc4.d/S11bridge /etc/rc.d/rc5.d/S11bridge /etc/rc.d/rc6.d/K90bridge
This makes it run right after the network start script. You should disable other configuration of eth0 (or eth1) such as done in the /etc/rc.d/init.d/network script (in RedHat by removing files ifcfg-eth? from /etc/sysconfig/network-scripts/).
To try things out, I suggest rebooting in single user mode (specify "single" as an arg to the kernel, e.g. in lilo "lilo: linux single") and running the startup scripts in /etc/rc.d/rc3.d one at a time until you get to the bridge startup. Startup the bridge and then see if you can reach some machines (you probably want to use "ping -n" for this to keep the nameserver out of the equation):
If you can ping all those places, there is a good chance that things are working. Note that the bridge takes a few moments to startup. You can monitor the status of the bridge by issuing the command brcfg with no arguments.
You still need to setup your firewall (assuming you want one) to prevent unauthorized access. The Bridge Filter Patch that you applied allows you to use a new built-in rule "bridgein" with ipchains. This rule is used whenever a packet is going to be forwarded either from eth0 to eth1 or vice versa. The bridgein rule is not used when a packet is destined for the firewall itself; you will want to use the input rule for that. I will not attempt to delve into the firewall setup in detail; please see the IPCHAINS HOWTO for that.
For each of your local machines, you simply have to setup the proper IP address and netmask and use the DSL router for the gateway (default route). The firewall/bridge will bridge the packets to/from the DSL router.