TTL is one - prevents multicast routing

[Neil Wilson]

Binary package hint: corosync

corosync uses a TTL of 1 in its multicast packets - restricting its use to a local segment and making it unusable in a multicast routed environment (such as with routed virtual machines).

ProblemType: Bug
DistroRelease: Ubuntu 10.04
Package: corosync 1.2.0-0ubuntu1
ProcVersionSignature: Ubuntu 2.6.32-22.36-generic 2.6.32.11+drm33.2
Uname: Linux 2.6.32-22-generic i686
Architecture: i386
Date: Mon Sep 13 13:17:59 2010
ProcEnviron:
 PATH=(custom, no user)
 LANG=en_GB
 SHELL=/bin/bash
SourcePackage: corosync

[Andres Rodriguez (andreserl)]

Hi Neil,

Could you please provide further information on how the network is configured and any other thing necessary to test this please?

Thank you.

[Andres Rodriguez (andreserl)]

Neil,

I've just spoken with upstream about the issue. He would be happy to make a tuneable option for this. However, this requires you to file a "Enhancement Request" as per [1], to test the changes.

Could you please do as required so that this gets fixed in upstream?

Thank you!!

[1]: http://www.corosync.org/doku.php?id=support

[Neil (neil-redhat-bugs)]

Description of problem:

Multicast TTL is 1, so it can't be used on a routed network.

Version-Release number of selected component (if applicable):

corosync 1.2.0-0ubuntu1

How reproducible:

Hardcoded into system.

Steps to Reproduce:
1. Two hosts linked by an ethernet
2. Libvirt in routed mode.
3. Virtual machines sat on a 10.x.x.x/30 subnet.
4. Unicast routing in place (I'm using bird and OSPF).
5. Multicast routing in place (pimd).
6. Libvirt configured so that it doesn't filter multicast!
7. Tested as working with ssmping

Actual results:

Packet won't get beyond the first host as the corosync packet has a TTL of one (local subnet).

Expected results:

TTL should really be 64 for the address range corosync is using.

Additional info:

This is an old issue from openais days by the look of it:

http://<email address hidden>/msg00548.html

[Neil Wilson (neil-aldur)]

Bug reported upstream - thanks for the lead on the bug tracker. Couldn't find it earlier.

The network is configured as follows:

Steps to Reproduce:
1. Two hosts linked by an ethernet
2. Libvirt in routed mode.
3. Virtual machines sat on a 10.x.x.x/30 subnet.
4. Unicast routing in place (I'm using bird and OSPF).
5. Multicast routing in place (pimd).
6. Libvirt configured so that it doesn't filter multicast!
7. Tested as working with ssmping

[Neil Wilson (neil-aldur)]

Just in case it wasn't clear, each virtual machine sits on their own 10.x.x.x/30 subnet and there is one VM on each host

[Neil (neil-redhat-bugs)]

Just in case it wasn't clear, each virtual machine has their own 10.x.x.x/30 subnet and there is one on each host

[Jan (jan-redhat-bugs)]

This is fixed in flatiron d3b983953d43dd17162be04de405d223fb21cd26 and will be included in 1.4.

[Rafael David Tinoco (rafaeldtinoco)]

Correct flag for this bug is: Fix Released

per upstream fix:

commit d3b983953d43dd17162be04de405d223fb21cd26
Author: Angus Salkeld <email address hidden>
Date: Wed Nov 24 14:35:56 2010 +1100

    Add totem/interface/ttl config option.

    This adds a per-interface config option to
    adjust the TTL.

    Signed-off-by: Angus Salkeld <email address hidden>
    Reviewed-by: Steven Dake <email address hidden>