autoipd breakage when it sees its own ARPs

[npmccallum]

In the case where you have a system connected to a hub which returns packets to the sender, autoipd always lists every IP address as taken, even when it isn't. The attached patch fixes this problem by ignoring conflicts when the packet is generated by the local address. I cannot think of any downsides to this approach.

[lennart]

Hmm, sounds reasonable. But, hmm, what kind of weird hub is this?

[lennart]

Using memcmp() for that comparison is probably a better idea though. Also it might be an idea log about weird responses like that.

[npmccallum]

updated patch

[npmccallum]

I've updated the patch with your suggestions. Thanks for the hint on memcpy, I'm not quite sure how I never learned about that function. Oh well.

[lennart]

memcmp, not memcpy!

[npmccallum]

Oops! Can you fix my typo? or do you need a new patch?

[lennart]

Thanks, I'll fix it when I merge it. It's easy enough.

[npmccallum]

Any chance this can be merged? Its holding up a project I'm working on. It would be great if we could get it into the Fedora RPMs as well.

[npmccallum]

Any update? This should be a simple merge.

[lennart]

Yes, I will merge this. Please be patient. I am waiting for more fixes however before I will do a new release. And usually before I sit down and do the release I will merge them all.

[lennart]

BTW, I am still wondering what kind of weird hw you have that responds all packets back?

[npmccallum]

Its an old no-name hub I bought at a computer fair back in the day...

[aboyer]

Chiming in in support of this patch: I am experiencing a similar situation in which avahi-autoipd is incorrectly detecting conflicts on FreeBSD. It picks an address and moves to the background. When I try to ping it, Avahi sees the arp reply and changes addresses.

I'm not privy to the whole network configuration, but the system is connected directly to a 10Gbit Arista switch. We've got changes in that force all loopback packets out onto the wire, but I don't know if that's related.

The patch attached to this bug (well, with memcmp()) solves the problem. Please incorporate it into your next release.

Thank you!

[lennart]

Fixed and applied. Thank you!