Re: NM 0.9.8 doesn't realise that DHCP worked

[Dan Williams <dcbw redhat com>]

On Wed, 2013-02-27 at 12:28 +0100, Bastien Nocera wrote:
> On Tue, 2013-02-26 at 16:52 -0500, Pavel Simerda wrote:
> > ----- Original Message -----
> > > From: "Bastien Nocera" <hadess hadess net>
> > > Heya,
> > >
> > > I'm using rawhide on my laptop, and I can get the network to work in
> > > sporadic 30 second bursts, as NM doesn't realise that dhclient got an
> > > IP
> > > address:
> > > Feb 26 13:35:16 sirocco dhclient[1177]: bound to 192.168.0.24 --
> > > renewal in 17055 seconds.
> > > Feb 26 13:35:59 sirocco NetworkManager[576]: <warn> (wlan0): DHCPv4
> > > request timed out.
> >
> > Looks wrong. 
> >
> > > Feb 26 13:35:59 sirocco NetworkManager[576]: <info> (wlan0): canceled
> > > DHCP transaction, DHCP client pid 1177
> > > Feb 26 13:35:59 sirocco NetworkManager[576]: <info> Activation
> > > (wlan0) Stage 4 of 5 (IPv4 Configure Timeout) scheduled...
> > > Feb 26 13:35:59 sirocco NetworkManager[576]: <info> Activation
> > > (wlan0) Stage 4 of 5 (IPv4 Configure Timeout) started...
> > > Feb 26 13:35:59 sirocco kernel: [  211.996738] wlan0:
> > > deauthenticating from f4:ca:e5:94:aa:60 by local choice (reason=3)
> > > Feb 26 13:35:59 sirocco NetworkManager[576]: <info> (wlan0): device
> > > state change: ip-config -> failed (reason 'ip-config-unavailable')
> > > [70 120 5]
> > >
> > > Any ideas what might be causing this? I get the same problem on wired
> > > and wireless networks.
> >
> > Not really. Did you check if it isn't a selinux issue?
>
> I have SELinux in permissive mode, it's not SELinux.
>
> >  We changed quite a lot of stuff but DHCP was always working for me.
> > We didn't modify the dhclient action binary for time, so it might be
> > theoretically somewhere in the dhcp plugin.
> >
> > Btw you said the network works for you sometimes? Does that mean
> > dhclient sets up the address directly?
>
> There's an IP address bound, as seen in the logs. Might be explained by
> what's below.
>
> >  That would suggest it's not running NM's dhcp action script at all
> > but instead runs the default one.
>
> I symlink'ed /etc/dhclient-script to /usr/sbin/dhclient-script after
> seeing dhclient complain about /etc/dhclient-script being missing.
>
> It didn't work before either, doesn't work when I remove that symlink,
> doesn't work when I 
>
> > Maybe also try to killall dhclient (and NetworkManager) before you
> > test once again.
>
> The machine's been rebooted a million times already. I doubt that a
> killall would fix things anymore.
>
> I've finally found the problem however.
>
> In my /etc I had a copy of the example /etc/dhclient.conf, dating back
> from the 25th July 2012 (when I debugged dhcp to find why we couldn't
> connect to the GUADEC wireless network).
>
> That file never caused a problem with older versions of NetworkManager
> or dhclient but it seems that recent version of dhclient or the way it's
> being called by NM makes it not be ignored anymore.
>
> I'm pretty sure that your commit
> 1a7e7c9031526b0c8973965c826ad7d1b7cc22f9 is the one causing the
> regression:
> http://cgit.freedesktop.org/NetworkManager/NetworkManager/commit/?id=1a7e7c9031526b0c8973965c826ad7d1b7cc22f9
>
> It's basically loading anything that lies around, even if it's not
> relevant for the distribution used.

Oh fun.  I suppose we should strip out any script thats specified in
that file when copying its configuration to the one that we actually
send to dhclient.  Would something like this fix it?

diff --git a/src/dhcp-manager/nm-dhcp-dhclient-utils.c b/src/dhcp-manager/nm-dhcp-dhclient-utils.c
index 8dc7902..d311d50 100644
--- a/src/dhcp-manager/nm-dhcp-dhclient-utils.c
+++ b/src/dhcp-manager/nm-dhcp-dhclient-utils.c
@@ -168,6 +168,10 @@ nm_dhcp_dhclient_create_config (const char *interface,
                                        continue;
                        }
 
+                       /* Ignore 'script' since we pass our own */
+                       if (g_str_has_prefix (p, "script "))
+                               continue;
+
                        /* Check for "also require" */
                        if (!strncmp (p, ALSOREQ_TAG, strlen (ALSOREQ_TAG))) {
                                in_alsoreq = TRUE;


Dan