Re: IPv6, tentative addresses, bind(), wheezy

[Don Armstrong <don@debian.org>]

On Fri, 02 Nov 2012, martin f krafft wrote:
> The reason is that by the time bind() is called, the IPv6 address
> (configured with /e/n/i inet6 static, which unbound should listen
> on) is not yet ready, but "tentative", so the bind() call fails.

I've actually seen users run into a similar class of problem twice in
the past two days on #debian. [One was easily fixed by adding setting
ASYNCMOUNTNFS="no" to /etc/default/rcS, the other by adding a look
checking for the appropriate device to come up to the init script.]

I think we're going to have little choice but to switch to a real
dependency system which allows for init scripts to be called when a
much broader set of dependencies have been satisfied.

> I can now disable DAD, or insert "sleep 10" at the top of
> /etc/init.d/unbound, but neither is an acceptable solution.

Seems like the right solution is to wait for the network device to be
fully up in the init script with a sleep loop... though that's
certainly not optimal.


Don Armstrong

-- 
I now know how retro SCOs OSes are. Riotous, riotous stuff. How they
had the ya-yas to declare Linux an infant OS in need of their IP is
beyond me. Upcoming features? PAM. files larger than 2 gigs. NFS over
TCP. The 80's called, they want their features back.
 -- Compactable Dave http://www3.sympatico.ca/dcarpeneto/sco.html

http://www.donarmstrong.com              http://rzlab.ucr.edu