Multicast fails when interface changed

We are running an application that uses multicasting
on Linux kernel 2.2.19. The application is
connected to two network interfaces for redundancy purposes - only one interface is active at a time.
When the application starts up on the first interface, the application can send and receive multicast messages. We then use ifconfig to bring down the first interface, and use ifconfig to bring up the second interface. The application works with the exception that it cannot receive multicast packets (it can still send multicast packets).

It appears that the multicast information is not propagated when switching from one interface to the other. After the switch, the information in /proc/dev/mcast is still set up to work with the first interface. Also, it looks like the device list of multicast information (checked by calling ip_check_mc()) is lost when switching from the first to second interface.

Is there a fix or work around to keep multicasting working when switching from one interface to another? We looked at the 2.4 kernel and it appears that it will have the same problem.

Also, if there is no fix or patch, we are considering whether we should come up with a kernel patch. If we submitted a kernel patch that propagates the multicast information when the interface is changed, would it be accepted into the kernel?

I am not on the list so please CC with any answers/comments.

Thanks.

Christopher ý:.žË›±Êâmçë¢kaŠÉb²ßìzwm…éb�ïîžË›±Êâmébžìÿ‘êçz_âžØ^n‡r¡ö¦zË�ëh™¨è­Ú&£ûàz¿äz¹Þ—ú+€Ê+zf£¢·hšˆ§~†­†Ûiÿÿï�êÿ‘êçz_è®æj:+v‰¨þ)ߣømšSåy«­æ�¶…
­†ÛiÿÿðÃí»è®å’i

does your app use INADDR_ANY for imr_interface when you join the multicast group?

-chris

the app (vocal 1.2) does not use INADDR_ANY for imr_interface when joining the multicast group

-Christopher ý:.žË›±Êâmçë¢kaŠÉb²ßìzwm…éb�ïîžË›±Êâmébžìÿ‘êçz_âžØ^n‡r¡ö¦zË�ëh™¨è­Ú&£ûàz¿äz¹Þ—ú+€Ê+zf£¢·hšˆ§~†­†Ûiÿÿï�êÿ‘êçz_è®æj:+v‰¨þ)ߣømšSåy«­æ�¶…
­†ÛiÿÿðÃí»è®å’i

Hello!

Hmm... and what does it use?

As soon as /proc/net/dev_mcast does not show membership on the second interface, it is not difficult to conclude that the applciation just forgot to request mmembership on it.

Alexey

When joining the multicast group, imr_interface is set to the hostname IP address, in this case 192.168.25.113. This was verified by by looking at the Vocal code, and by putting printfs in Vocal code to print out imr_interface when joining the multicast group.

More background on app and problem: We are developing a high availability solution where the application can run on one of two redundant networks. The application uses either eth1 (connected to network1)
or eth2 (connected to network2). Only one interface is up at a time. When the application comes up, it uses eth1. Another process (other than the application) monitors the health of network1. If network1 goes down, then the
monitoring process uses ifconfig to take down eth1, and ifconfig to bring up eth2 (using the same configuration information as used for eth1). This switch is completely transparent to the application: the application does
not know about the switch or do anything to make the switch; the app does not set
multicast membership on the new interface.

It was our expectation that the switch from the first to second interface should

work without any involvement from the application because the second interface is configured
exactly the same as the first interface. After the switch, everything seems to work with the
exception of multicasting: the multicast membership information is not propagated to the second
interface, it stays with the first interface.

Christopher
(I'm not on list so please CC with answers/comments)

ý:.žË›±Êâmçë¢kaŠÉb²ßìzwm…éb�ïîžË›±Êâmébžìÿ‘êçz_âžØ^n‡r¡ö¦zË�ëh™¨è­Ú&£ûàz¿äz¹Þ—ú+€Ê+zf£¢·hšˆ§~†­†Ûiÿÿï�êÿ‘êçz_è®æj:+v‰¨þ)ߣømšSåy«­æ�¶…
­†ÛiÿÿðÃí»è®å’i

Hello!

The expectation was wrong.

Alexey

i don't think this is a valid expectation. joining a multicast group is a device specific action. when you join, you either specify an interface to join on (imr_interface=dev_ip_addr) or let the kernel choose (imr_interface=INADDR_ANY). in either case, you are telling some hardware to adjust its multicast filter and identifying that hardware by a unique device index. as alexey mentioned, it sounds like your app has never told the second interface that is should even care about multicast packets. did you try joining on both interfaces? (you may find using a unique service ip addr on each interface, and failing over an application ip addr using aliases will help)

cheers,
-chris