revert e3c5de1

[zacknewman]

Fixes #407.

[zacknewman]

Alternatively, we can revert the commit just for OpenBSD (i.e., I can remove defined(__FreeBSD__) || in the preprocessor directive).

[perkelix]

I'd just revert the whole change completely. It has broken expected operation on Linux too, as I reported in #392.

[zacknewman]

I'd just revert the whole change completely. It has broken expected operation on Linux too, as I reported in #392.

That's unrelated. The reverted commit only impacts OpenBSD and FreeBSD as you can see with the preprocessor directive. I don't use FreeBSD, so I can't say for certain the commit is invalid for FreeBSD. I just know it's invalid for OpenBSD. The commit also only impacts IPv6. OpenBSD does not allow IPv6 sharing; thus when a route is attempted to be added for the same interface, it forbids it causing /var/log/daemon to be spammed. Before the problematic commit, dhcpcd only attempted it if it was not an IPv6 route on OpenBSD and FreeBSD.

[rsmarples]

I haven't been able to replicate it, but I'll revert as is.
I'll see if I can submit a FreeBSD kernel change at least so we can drop that test at some point.

[zacknewman]

Based on the original comment, you were unable to trigger a kernel panic in FreeBSD when you first wrote that code as well. Not sure why you are unable to trigger it. Perhaps it's a case of running in a VM vs. bare-metal? The lease file I e-mailed was of no help?

[rsmarples]

Ah, I lost the lease file. Do you still have it?

[zacknewman]

Re-sent. Did you receive it? [email protected], correct?

[rsmarples]

Yes thanks.
Note for future self, I was able to re-create it on OpenBSD-7.6 by ensuring a default route was not present.
I still feel like this is a kernel bug that should be fixed.

[zacknewman]

I'm glad you were able to re-create it. You can try submitting a kernel patch, but IPv6 is unfortunately still treated with disdain by some of the community so there may be pushback. I know OpenBSD (ab)uses the fact that currently only the fe80::/64 portion of the fe80::/10 link-local block can be used; thus they use the second octet pair to embed the interface index. Point being there are likely other issues with the IPv6 stack.