[quagga-dev 7736] Re: [PATCH] [zebra] fix bug #486 (lingering after IP address deletion)

paul at jakma.org paul at jakma.org
Tue Feb 2 01:01:09 GMT 2010

On Tue, 2 Feb 2010, David Lamparter wrote:

> Please don't reject a good fix for one small issue because you want 
> a (most likely new bug introducing) change for the big picture.

I think this was working before. Least, I had tested on Solaris, BSD 
and Linux. But it's been a while. I'm not 100% sure, but I think the 
netlink change to filter out self-originated commands is what's 
broken this (unintentionally) - I didnt know till recently the kernel 
send was synchronous and couldn't fail.

> P.S.: http://www.diac24.net/files/quagga/address.png

Nice. How did you generate that?

ip_address_install is still a monster.

> - you're completely right that this entire thing is a huge mess. 
> but maybe we should fix the small issues on broken code first, then 
> go on to fix the big issues.

Right. I'm trying to point you toward a small fix that is inline with 
the big issues.

> P.P.S.: we can nicely find the places that need fixing by looking for
> [UN]SET_FLAG(..., ZEBRA_IFC_REAL); that should only happen in
> from-kernel call path according to your model.

Note that there is a unit test in zebra/. I think it's missing some 
stubs to properly test vty IP add/delete, but..

Paul Jakma	paul at jakma.org	Key ID: 64A2FF6A
  * At first I thought these guys were on crack, but then I discovered the
  * LART.

 	- comment from include/linux/mtd/cfi_endian.h

More information about the Quagga-dev mailing list