Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.debian.user > #166730 > unrolled thread

Incorrect arp reply when bonding and 3rd interface are on same LAN.

Back to article view | Back to linux.debian.user

Contents

  Incorrect arp reply when bonding and 3rd interface are on same LAN. Mimiko <vbvbrj@gmail.com> - 2016-04-15 10:40 +0200
    Re: Incorrect arp reply when bonding and 3rd interface are on same  LAN. Mimiko <vbvbrj@gmail.com> - 2016-04-15 11:40 +0200
    Re: Incorrect arp reply when bonding and 3rd interface are on same LAN. Sven Hartge <sven@svenhartge.de> - 2016-04-15 12:40 +0200
      Re: Incorrect arp reply when bonding and 3rd interface are on same  LAN. Mimiko <vbvbrj@gmail.com> - 2016-04-15 14:20 +0200

#166730 — Incorrect arp reply when bonding and 3rd interface are on same LAN.

Hello.

A server has 3 interfaces: eth0, eth1, eth2. I've setup a bond12 with 
mode adaptive-alb with eth1 and eth2. Now interfaces have:

auto eth0
iface eth0 inet static
address x.x.x.1
netmask 255.255.255.0

auto eth1
iface eth1 inet manual

auto eth2
iface eth2 inet manual

auto bond12
iface bond12 inet static
address x.x.x.2
netmask 255.255.255.0
bond-slaves eth1 eth2
bond-mode balance-alb

All interfaces are connected to 3 different switches which are on same 
LAN (switches are interconnected).

The problem is that when I do from another computer:

arping -c 1 -I eth0 x.x.x.1

or

arping -c 1 -I eth0 x.x.x.2

in both cases I receive two mac addresses: one for eth0 and another for 
currently active interface in the bond. This confuses clients and there 
are problem when accessing servicing on incorrect interface regarding 
the IP.

mac for x.x.x.1 must be of the eth0
mac for x.x.x.2 must be of the currently active interface in the bond

What could be the problem?

Also on start-up, bond0 is also created with no slaves. Why it is created?

Thank you.

[toc] | [next] | [standalone]

#166731 — Re: Incorrect arp reply when bonding and 3rd interface are on same LAN.

On 15.04.2016 11:33, Mimiko wrote:
> Hello.
>
> A server has 3 interfaces: eth0, eth1, eth2. I've setup a bond12 with
> mode adaptive-alb with eth1 and eth2. Now interfaces have:
>
> auto eth0
> iface eth0 inet static
> address x.x.x.1
> netmask 255.255.255.0
>
> auto eth1
> iface eth1 inet manual
>
> auto eth2
> iface eth2 inet manual
>
> auto bond12
> iface bond12 inet static
> address x.x.x.2
> netmask 255.255.255.0
> bond-slaves eth1 eth2
> bond-mode balance-alb
>
> All interfaces are connected to 3 different switches which are on same
> LAN (switches are interconnected).
>
> The problem is that when I do from another computer:
>
> arping -c 1 -I eth0 x.x.x.1
>
> or
>
> arping -c 1 -I eth0 x.x.x.2
>
> in both cases I receive two mac addresses: one for eth0 and another for
> currently active interface in the bond. This confuses clients and there
> are problem when accessing servicing on incorrect interface regarding
> the IP.
>
> mac for x.x.x.1 must be of the eth0
> mac for x.x.x.2 must be of the currently active interface in the bond
>
> What could be the problem?
>
> Also on start-up, bond0 is also created with no slaves. Why it is created?
>
> Thank you.

I've just found the problem. I read this site:

http://kb.linuxvirtualserver.org/wiki/Using_arp_announce/arp_ignore_to_disable_ARP

And use:

echo 1 > /proc/sys/net/ipv4/conf/all/arp_ignore


This resolves the problem. On windows by default is not to reply with 
the mac of the not received interface.

[toc] | [prev] | [next] | [standalone]

#166733

Mimiko <vbvbrj@gmail.com> wrote:

> A server has 3 interfaces: eth0, eth1, eth2. I've setup a bond12 with 
> mode adaptive-alb with eth1 and eth2. Now interfaces have:

> iface eth0 inet static
> address x.x.x.1
> netmask 255.255.255.0

> iface bond12 inet static
> address x.x.x.2
> netmask 255.255.255.0

> All interfaces are connected to 3 different switches which are on same 
> LAN (switches are interconnected).

This is bad.

> The problem is that when I do from another computer:

> arping -c 1 -I eth0 x.x.x.1

> or

> arping -c 1 -I eth0 x.x.x.2

> in both cases I receive two mac addresses: one for eth0 and another for 
> currently active interface in the bond. This confuses clients and there 
> are problem when accessing servicing on incorrect interface regarding 
> the IP.

This is the way Linux Networking works when multiple interfaces are
connected to the same LAN. "Weak Host Model" is your research hint.

> What could be the problem?

The problem is, that Linux does not behave in the way you expect it to
behave. The simplest solution would be to have all interfaces in one
ALB bond and configure both IPs on that.

Grüße,
Sven.

-- 
Sigmentation fault. Core dumped.

[toc] | [prev] | [next] | [standalone]

#166737 — Re: Incorrect arp reply when bonding and 3rd interface are on same LAN.

On 15.04.2016 13:37, Sven Hartge wrote:
>> >All interfaces are connected to 3 different switches which are on same
>> >LAN (switches are interconnected).
> This is bad.
>

Well, I need the eth0 for administration, as bond I can disconnect, 
connect restart and will lose connection.

 > This is the way Linux Networking works when multiple interfaces are
 > connected to the same LAN. "Weak Host Model" is your research hint.

Thank you, I've found how to resolve this. See my other post.

If you could help identifying why bond0 is created with no slaves.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.debian.user

csiph-web