Skip to contentSkip to conversation

Proxy Arp with same left/right IP address.

LARTC Adrian Chung 2 participants 5 messages Feb 13, 2002

lore
Share & tools
lore
#0 lore raw

Hey everyone! I came across this link today:

http://www.sjdjweis.com/linux/proxyarp/

Which I thought was very interesting. But I don't quite understand how, or what effect assigning the same IP address to both the left and right side interfaces of the pseudo-bridge has.

In the article, the author describes a mechanism for being able to shape/filter traffic to an entire ISP assigned subnet.

Basically, use proxy arp, assign the same ISP IP to both interfaces on the router, and use iproute2 to create routes to the router on one interface, and the rest of the network on the other.

It's almost identical to the Proxy-Arp/Pseudo Bridge part of the LARTC HOWTO.

Except that I'm confused as to how the kernel deals with the fact that both interfaces have the same IP address.

I guess logically it doesn't matter to the kernel, because the routes clearly delineate which interface to send packets to, proxy-arp connects the two LANs, and both interfaces have different MAC addresses.

It just seems wrong. :)

--
Adrian Chung (adrian at enfusion-group dot com)
http://www.enfusion-group.com/~adrian
GPG Fingerprint: C620 C8EA 86BA 79CC 384C E7BE A10C 353B 919D 1A17 [toad.enfusion-group.com] up 1:44, 1 user, load average: 0.00

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/lartc/

#1 lore raw
Quoted 8 lines · Adrian Chung

Hey everyone! I came across this link today:

http://www.sjdjweis.com/linux/proxyarp/

Which I thought was very interesting. But I don't quite understand how, or what effect assigning the same IP address to both the left and right side interfaces of the pseudo-bridge has.

Quoted 6 lines

In the article, the author describes a mechanism for being able to shape/filter traffic to an entire ISP assigned subnet.

Basically, use proxy arp, assign the same ISP IP to both interfaces on the router, and use iproute2 to create routes to the router on one interface, and the rest of the network on the other.

It does not matter which ip addresses you asign to which interfaces. You can also assing the official ip to lo.

Quoted 5 lines

It's almost identical to the Proxy-Arp/Pseudo Bridge part of the LARTC HOWTO.

Except that I'm confused as to how the kernel deals with the fact that both interfaces have the same IP address.

The kernel sees one IP address, and two nics, with each having some part of a network directly attached to it.

Quoted 4 lines

I guess logically it doesn't matter to the kernel, because the routes clearly delineate which interface to send packets to, proxy-arp connects the two LANs, and both interfaces have different MAC addresses.

Yep.

Quoted 1 line

It just seems wrong. :)

Nope.
If you think about it: it is CLEAN!

I've attached an e-mail which should give some more examples, and a copy of the mail I send to the maintainer of that site.
(Eh, the update to the last mail: I tested it, and it works.)

Some advice, eh to everybody:
Really forget about route and ifconfig, try to switch to the iproute utility...

Signature · 4 lines
<[email protected]> Telegraaf Elektronische Media  http://wwwijzer.nl
http://leerquoten.monster.org/ http://www.faqs.org/rfcs/rfc1855.html 
Let your government know you value your freedom. Sign the petition:
http://petition.eurolinux.org/
Quoted 5 lines · Carlos Carvalho

I'm about to try ethernet tags for the first time. I need to use the same ip in all vlan interfaces. Is this possible?

I can't put each vlan in a different IP subnet because there will be tens of them.

I am not sure of what you are trying to do, but this is possible:

Give all the networks the idea of a virtual 192.168.0.0/16 f.i. . You can divide your network in c classes by f.i.:

vconfig add eth0 2
ip addr add 127.0.0.1/8 dev vlan2
ip route add 192.168.2.0/24 dev vlan2
echo 1 > /proc/sys/net/ipv4/conf/vlan2/proxy_arp

vconfig add eth0 3
ip addr add 127.0.0.1/8 dev vlan3
ip route add 192.168.3.0/24 dev vlan3
echo 1 > /proc/sys/net/ipv4/conf/vlan2/proxy_arp

vconfig add eth0 4
ip addr add 127.0.0.1/8 dev vlan4
ip route add 192.168.4.0/24 dev vlan4
echo 1 > /proc/sys/net/ipv4/conf/vlan2/proxy_arp

vconfig add eth0 5
ip addr add 127.0.0.1/8 dev vlan5
ip route add 192.168.5.0/24 dev vlan5
echo 1 > /proc/sys/net/ipv4/conf/vlan2/proxy_arp

ip addr add 192.168.0.1/32 dev lo
echo 1 > /proc/sys/net/ipv4/ip_forward

If 192.168.5.2 tries to ping 192.168.3.9, it will arp for it, since it thinks it is in the same subnet (/16). Your box knows how to route it, and therefore will reply to the arp. 192.168.5.2 will then send the packet to your box, which will happily forward it.
The default gw in this example is 192.168.0.1, put on lo.

You need to have at least 1 ip address on a device to get the ip stack bound to that device. If you don't have an ip address on a device (no matter what ip address), you will not be able to use the kernel-ip stack on that device. Not binding the kernel ip-stack to the device is a good idea if you want to do some network sniffing for intrusion detection etc...

(Thanks to Pauline Middelink for pointing me at the proxy_arp idea)

DISCLAIMER: I already tested the concept using 127.0.0.1 addresses and proxy-arp in a sort of production environment. I did not test the example above :)

Signature · 8 lines
<[email protected]> Telegraaf Elektronische Media  http://wwwijzer.nl
http://leerquoten.monster.org/ http://www.faqs.org/rfcs/rfc1855.html 
Let your government know you value your freedom. Sign the petition:
http://petition.eurolinux.org/
_______________________________________________
VLAN mailing list  -  [email protected]
http://www.WANfear.com/mailman/listinfo/vlan
VLAN Page:  http://scry.wanfear.com/~greear/vlan.html

In
http://www.sjdjweis.com/linux/proxyarp/index.html

You say:
You will need a 2.4 series kernel, the iproute2 utility, and a recent iptables userspace program.

Allright, then these are my comments :) :
Everything except for the iptables already work on a 2.2 kernel, there is nothing new there. Yes, I already used proxy-arp on a 2.2 kernel. Hmmm, ok, you are right that in this case you definitely need the connection tracking to prevent from unwanted garbage from entering your network, which is not the case in the classical (masquerading) firewalling approach. But that is the "only" reason to have 2.4...

Another thing I am now busy to find out (I am not sure if this also works on 2.2, but I have no reason to believe it does not work):

ip link set up dev eth0
ip addr add 127.0.0.1/32 dev eth0
echo 1 > /proc/sys/net/ipv4/conf/eth0/proxy_arp
ip link set up dev eth1
ip addr add 127.0.0.1/32 dev eth1
echo 1 > /proc/sys/net/ipv4/conf/eth1/proxy_arp
ip route add x.x.x.96/28 dev eth0
ip route add x.x.x.97/32 dev eth1

The theory behind it: the IP protocol driver needs to be bound to the device. This is done by giving any (bogus) ip address to an interface. Only after the IP protocol driver is bound to the interface you can use it for IP.
/proc/sys/net/ipv4/conf will only show interfaces bound to the IP protocol... :)

Of course: you cannot reach the firewall, and the firewall can't reach you. You need a non-bogus ip address for that... (Hmmmm: ssh 127.0.0.1 ... Hah! I hacked the firewall.... shutdown -h now ... Uh? ;) )

Signature · 4 lines
<[email protected]> Telegraaf Elektronische Media  http://wwwijzer.nl
http://leerquoten.monster.org/ http://www.faqs.org/rfcs/rfc1855.html 
Let your government know you value your freedom. Sign the petition:
http://petition.eurolinux.org/
#2 lore raw
In reply to Adrian Chung
Quoted 1 line · Ard van Breemen

On Wed, Feb 13, 2002 at 10:35:15AM -0500, Adrian Chung wrote:

[...]

Quoted 5 lines
Quoted 2 lines

Except that I'm confused as to how the kernel deals with the fact that both interfaces have the same IP address.

The kernel sees one IP address, and two nics, with each having some part of a network directly attached to it.

Quoted 8 lines
Quoted 4 lines

I guess logically it doesn't matter to the kernel, because the routes clearly delineate which interface to send packets to, proxy-arp connects the two LANs, and both interfaces have different MAC addresses.

Yep.

Quoted 1 line

It just seems wrong. :)

Nope.
If you think about it: it is CLEAN!

You're definitely right about that. It's a *very* cool solution, I'd been looking for a way to do this.

[...]

Quoted 19 lines

Another thing I am now busy to find out (I am not sure if this also works on 2.2, but I have no reason to believe it does not work):

ip link set up dev eth0
ip addr add 127.0.0.1/32 dev eth0
echo 1 > /proc/sys/net/ipv4/conf/eth0/proxy_arp
ip link set up dev eth1
ip addr add 127.0.0.1/32 dev eth1
echo 1 > /proc/sys/net/ipv4/conf/eth1/proxy_arp
ip route add x.x.x.96/28 dev eth0
ip route add x.x.x.97/32 dev eth1

The theory behind it: the IP protocol driver needs to be bound to the device. This is done by giving any (bogus) ip address to an interface. Only after the IP protocol driver is bound to the interface you can use it for IP.
/proc/sys/net/ipv4/conf will only show interfaces bound to the IP protocol... :)

Quoted 3 lines

Of course: you cannot reach the firewall, and the firewall can't reach you. You need a non-bogus ip address for that... (Hmmmm: ssh 127.0.0.1 .. Hah! I hacked the firewall.... shutdown -h now ... Uh? ;) )

Is this for real? So you bind 127.0.0.1 to both eth0 and eth1, not using an ISP assigned IP at all on the bridging box, and you'd still be able to route from the ISP's network through to machines on the ISP assigned network?

That's way too neat. :)

Have you tried this under 2.4?

--
Adrian Chung (adrian at enfusion-group dot com)
http://www.enfusion-group.com/~adrian
GPG Fingerprint: C620 C8EA 86BA 79CC 384C E7BE A10C 353B 919D 1A17 [toad.enfusion-group.com] up 2:11, 1 user, load average: 0.00

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/lartc/

#3 lore raw
In reply to Adrian Chung
Quoted 8 lines · Adrian Chung
Quoted 7 lines · Ard van Breemen

The theory behind it: the IP protocol driver needs to be bound to the device. This is done by giving any (bogus) ip address to an interface. Only after the IP protocol driver is bound to the interface you can use it for IP.
/proc/sys/net/ipv4/conf will only show interfaces bound to the IP protocol... :)

Quoted 7 lines
Quoted 3 lines

Of course: you cannot reach the firewall, and the firewall can't reach you. You need a non-bogus ip address for that... (Hmmmm: ssh 127.0.0.1 .. Hah! I hacked the firewall.... shutdown -h now ... Uh? ;) )

Is this for real? So you bind 127.0.0.1 to both eth0 and eth1, not using an ISP assigned IP at all on the bridging box, and you'd still be able to route from the ISP's network through to machines on the ISP assigned network?

Be careful however, you don't want to send icmp_host_unreachables with source ip 127.0.0.1, I don't think they get far.
Anyway, 127.0.0.1 is as far as I know a special adres, also for the kernel. I haven't experimented with that...
If you want to have any sane outgoing source ip addresses, you also have to set the default source ip for the outgoing routes.

Signature · 7 lines
<[email protected]> Telegraaf Elektronische Media  http://wwwijzer.nl
http://leerquoten.monster.org/ http://www.faqs.org/rfcs/rfc1855.html 
Let your government know you value your freedom. Sign the petition:
http://petition.eurolinux.org/
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/lartc/
#4 lore raw
In reply to Adrian Chung
Quoted 1 line · Ard van Breemen

On Wed, Feb 13, 2002 at 11:00:35AM -0500, Adrian Chung wrote:

[...]

Quoted 10 lines
Quoted 4 lines

Is this for real? So you bind 127.0.0.1 to both eth0 and eth1, not using an ISP assigned IP at all on the bridging box, and you'd still be able to route from the ISP's network through to machines on the ISP assigned network?

Be careful however, you don't want to send icmp_host_unreachables with source ip 127.0.0.1, I don't think they get far.
Anyway, 127.0.0.1 is as far as I know a special adres, also for the kernel. I haven't experimented with that...
If you want to have any sane outgoing source ip addresses, you also have to set the default source ip for the outgoing routes.

True enough... However, in my case anyways, it's not an issue. I already have a box that's using an ISP assigned IP as a gateway, but using aspects of the above solutions, I can now have all my DMZ servers funnelled through the one machine.

That means that traffic control/shaping and bandwidth monitoring should become real easy, as it means one access point into/out of the network, but still with IP assigned IPs on the DMZ machines.

Sorry if this is not news to anyone else, it's an amazing revelation to me. :)

--
Adrian Chung (adrian at enfusion-group dot com)
http://www.enfusion-group.com/~adrian
GPG Fingerprint: C620 C8EA 86BA 79CC 384C E7BE A10C 353B 919D 1A17 [toad.enfusion-group.com] up 2:24, 1 user, load average: 0.00

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://ds9a.nl/lartc/