Wireless network problem after upgrade to 8.04

[Derek Broughton]

Paul Johnson wrote:

> This is leading back to some basic questions about Upstart and service
> management.  See below.
> 
> On Thu, Jun 5, 2008 at 12:31 PM, Derek Broughton <news at pointerstop.ca>
> wrote:
>> Paul Johnson wrote:
>>
>>> Au contraire, mon brother! :)
>>>
>>> At least for me, there is there is a script in /etc/init.d
>>
>> OK, then, "the network-manager package no longer installs anything
>> in /etc/init.d". :-)  If you have "dlocate" installed, it would be
>> informative to run "dlocate /etc/init.d/network-manager" to see just
>> where
>> yours came from.  Anyway, Ken probably doesn't have it, as it moved into
>> dbus a release or two back.
> 
> Well, recall I'm upstream from you with the NetworkManager 0.7svn  (
> http://ubuntuforums.org/showthread.php?t=797059 ) , and that's where
> it comes from.  Look at the last page worth of output from dpkg -L
> network-manager:
> 
> /etc/init.d
> /etc/init.d/network-manager

Yeah, but that doesn't exist in Ubuntu hardy.

> The two daemon files, /etc/init.d/network-manager and
> /etc/dbus-1/even.d/25NetworkManager are almost identical, except for
> top matter. So I think it is just a packaging difference / mistake.

It's not a mistake (though I'm not trying to say it can't be wrong, just
that it was intentional).

> Here's a puzzler, though.  /etc/sbus-1/event.d/25NetowrkManager has
> this comment at the top:

I assume "sbus" is just a type like "Netowrk" :-)

> #This file should be placed in /etc/init.d.

Indeed it does!
> 
> Why put that precious thing under /etc/dbus-1/events.d, not
> /etc/init.d ?  Upstart magic?

Well, it needs to be under dbus, so that "/etc/init.d/dbus restart" will
trigger it, but that shouldn't exclude symlinking it in /etc/init.d/ so
that we can find it where we'd expect it.

> I don't see the wmaster0 errors now, however.
> 
> But, the "networking" daemon probably has nothing to do with wireless
> control. 

Not directly, but stopping and restarting it makes it retry the things that
appear to be Ken's problem.

> In 
> reading Ubuntu /etc/init.d/networking, it appears to me that this
> service should not be activated at start time, because it will
> frustrate NetworkManager.  But I'm still new here :<

It shouldn't because ubuntu (or maybe Debian - but certainly not upstream of
Debian) modified NM to ignore anything specified /etc/network/interfaces. 
So one or the other should work, but of any given interface not both.

> I've been doing a little reading and I see that Ubuntu created
> "Upstart" as a replacement for sysvinit.  

I really don't see upstart being at all related to this.

> My Ubuntu system starts in runlevel 2, and the 
> start scripts are in in /etc/rc2.d, a list of symbolic links.

That's standard Debian stuff...

> Observe:
> 
> .$ ls -la /etc/rc2.d/ | grep net
> lrwxrwxrwx   1 root root    25 2008-05-27 14:52 S26network-manager ->
> ../init.d/network-manager
> 
> Note, there is no link to "networking".  Of course, my NetworkManager
> package is different from yours.
>
> The only runlevel that does seem to activate networking is the sudo
> runlevel:
> 
> /etc/rcS.d/S40networking

"sudo runlevel"?  I've never heard that term.  I think that's "System" -
rcS.d gets invoked _before_ rc1.d (single-user mode - sort of "sudo") or
rc2.d.

> Maybe it gets called implicitly by some other service, I don't know.

Or just by virtue of being in rcS.d :-)

> Anyway, lets suppose we install the kernel module backports and the
> connection still fails.  Lets try to find out why.

Yeah, I don't have that wireless, so if you say using the backports is the
way to go, I'd have to bow to your experience.

> "is it true that service 'networking' should not be 
> running if we are counting on NetworkManager?"

"networking" has never been strictly a "service".  "networking start"
runs "ifup <IFACE>" for every interface (or only "auto" ones?  I can't
recall) and exits.

> "Why did Hardy Herron put the NetworkManager script in
> /etc/dbus-1/event.d, rather than /etc/init.d?"

Because it needs to be there... but there would have been no harm in having
a link from /etc/init.d/.  btw, that wasn't hardy that did that.  I believe
it was feisty, but at latest gutsy.
-- 
derek