Multicast group not appearing on Rendezvous Point router

[stormshield-rlibaert]

Hello,

We may have found a bug where Pimd as RP does not get the information of a new route. We use pimd version 3.0-beta1.

We have the following topology :

• pc_1 is the receiver
• pc_2 is the source
• firewall_2 is the RP (Rendezvous Point) router
• firewall_3 is the DR (Designated Router)

When starting emission of traffic with iperf, we see in pimctl show mrt that a new multicast route is created on the DR, but not on the RP. Similarly when the receiver is started, a new route is shown on both firewall_1 and the RP. However as the RP can not match the requested multicast group, no data is received on pc_1.

We investigated and found this commit may be introducing the bug : https://github.com/troglobit/pimd/commit/edb0aac7326be7020a4da18b0291fb0fdd777960 Issue related : https://github.com/troglobit/pimd/pull/67

We tried the following patch : https://github.com/stormshield-rlibaert/pimd/commit/cd3f7e6332c1b4433ebe3ccb87fad7c85ba6be12 and it seems to be fixing the issue.

We would like to know your advice on whether or not this is a correct way to fix the problem. If so, should we make a merge request ? Also, we are not sure if it is related, but it seems that the "interval" field of spt-threshold is ignored. What do you think ?

[troglobit]

Hi,

you've stumbled into the last remaining blocker issue before 3.0 GA. I still have a few commits in my patch queue that I haven't pushed yet, and even some unstable fixes for what I believe is very close to your issue. I've been meaning to finalize this, and add a test case for it, but have not have the time. Hoping for Christmas break ...

A few questions:

• Do you get this early after starting up the pimd instances, or does it happen after "a while"?
• If it's the former, I believe you're seeing what I am, which seems to be lack of route updates on RP election changes
• What multicast group(s) are you using for testing? The code you changed should only affect the reserved PIM-SSM group range 232.0.0.0/8

[stormshield-rlibaert]

Hi,

Thank you for your fast answer! :smiley:

Indeed we get this as soon as pimd is started (actually after a few second, just the time to switch terminals and launch commands). Our tests were performed in PIM-SM. I realize that this makes the fix kinda strange indeed. I'll go and take a look in the meantime.

[stormshield-rlibaert]

Looks like I was mistaken, the fix was not working on our side. For now, just reverting the suspected commit makes pimd to work as we expect (for reference https://github.com/stormshield-rlibaert/pimd/commit/203b2a8e6df26aa8be52f1e6119ec6f367259a18). And it makes more sense too.

Edit: in our case find_routes returns mrt == NULL, so the call to switch_shortest_path must be done before checking for return, otherwise it doesn't work.

[troglobit]

Yeah that's not right either, the SPT switchover should follow its established config. I've been sniffing around that same code myself.

The problem seem to be exactly what I've seen; when booting up the routers try to elect an RP and any PIM Joins sent before that has stabilized are prone to this problem. So my theory right now, if I remember correctly from a few months back when I was debugging this the last time, is that pimd doesn't resend all PIM Joins (and Leaves) after a new RP election. When that works, then we can have a closer look at reducing the convergence time.

[stormshield-rlibaert]

Hello Joachim,

I further investigated the issue and as you were expecting the proposed revert is not good. I found out the RP indeed created the SG as I wanted but it also started to send registers more than needed, exactly as described in https://github.com/troglobit/pimd/issues/128.

After a few researches, I came with another solution which is basically a fix for the issue I just mentioned. https://github.com/stormshield-rlibaert/pimd/commit/099748bb7476d1e12e19b587f9ca69b31c9b390c

This works pretty well for my test case. However, I am pretty certain that it has some unwanted side effects (PMBR, BorderBit, Null Register, ...). That being said, trying to further fix the issue would imply much intrusive changes. I was thinking about reworking the receive_pim_register function to make it more like what we can find in FRrouting.

[troglobit]

Nice! Yeah I actually have a backlog of things I haven't pushed yet, some of it addressing issues you've mentioned here. Been busy with lots of other projects, however, so haven't got back to this in a while. I'd like to set up a test case for verifying convergence and such bits.

Please send a PR for stormshield-rlibaert@099748b that looks much better than what we have now :)

Meanwhile, I'll reopen #128 so we can close it properly with your PR.