Re: One more clue

[Dan Williams <dcbw redhat com>]

On Thu, 2006-09-07 at 19:18 -0400, Darren Albers wrote:
> On 9/7/06, David Abrahams <dave boost-consulting com> wrote:
> > > It's checked into NM CVS
> >
> > Where's that repository?
> >
> > > on both HEAD and stable branches as test/nm-supplicant-test.c. Build
> > > NM (or grab it here [1]),
> >
> > Sorry, Firefox tells me that's a BIN file.  What am I supposed to do
> > with it?
> >
> Unfortunately the version Dan built requires libdbus-1.so.3 which is
> not available on Ubuntu, you might be able to create a symlink from
> libdbus-1.so but I am not what that would do.

Sorry about that; there's really nothing in there that should use dbus.
A less-heavy version has been uploaded to the same place that requires
only:

        linux-gate.so.1 =>  (0xb7f9d000)
        libglib-2.0.so.0 => /lib/libglib-2.0.so.0 (0x00c12000)
        libc.so.6 => /lib/libc.so.6 (0x00cca000)
        librt.so.1 => /lib/librt.so.1 (0x00f12000)
        /lib/ld-linux.so.2 (0x00e84000)
        libpthread.so.0 => /lib/libpthread.so.0 (0x00b54000)

Dan

> 
> I am building it on Ubuntu and I can email it to you or you can
> checkout it from CVS with:
> cvs -d :pserver:anonymous anoncvs gnome org:/cvs/gnome co NetworkManager
> 
> grab the required dependencies:
> sudo apt-get build-dep network-manager
> sudo apt-get install automake cvs
> 
> and build Network-Manager with:
> edit configure.in to remove -Werror
> and then do:
> ./autogen.sh
> make (Do not do a make install)
> 
> Stop Network Manager with:
> sudo /etc/dbus-1/event.d/25NetworkManager stop
> 
> and do then the command Dan described below pointing the test tool to
> you wpa_supplicant.conf that is working properly.
> 
> > > construct a wpa_supplicant config file, then run:
> > >
> > > sudo test/nm-supplicant-test eth1 /path/to/wpa_supplicant.conf
> > >
> > > It will use the network config for the first _uncommented_ network.  To
> > > diagnose the issue, I'll need the output from the supplicant tester (be
> > > _sure_ to remove any "wep_key0" or "psk" lines) as well as the output
> > > from a run of NetworkManager itself
> 
> > Sorry, I was under the impression that NM doesn't use config options I
> > set in wpa_supplicant.conf.  It certainly worked (as well as it has)
> > whether or not I had such a file.  Are there NM config options somewhere?
> 
> It does not, you described that the problem does not happen when you
> run wpa_supplicant manually, this tool is going to take your config
> and use it to send all the options to wpa_supplicant the same way
> NetworkManager normally does in an attempt to debug why it works with
> wpa_supplicant but not with Network Manager.

Exactly right; if you can connect with this tool and the wpa_supplicant
file you posted a while back in the "Does anyone understand NM" thread,
then we need to see the output of this tool and NM and compare them.  If
there are no material differences between their output, and the tool
works but not NM, then something else about NM interacting with
wpa_supplicant is an issue.

Some things I noticed today testing with a Netgear WGR614 v5 and Linksys
WRT54G v2.2:

a) Sometimes wpa_supplicant can't react fast enough to control interface
input; saw this only once in about 20 runs

b) Non-broadcasting SSIDs are trouble; they screw up a plain
wpa_supplicant at ap_scan=1 and even at ap_scan=2 sometimes.  I can't
quite tell what makes it work and what doesn't.

c) I guess we should be using ap_scan=1 where we can, since it's a lot
less picky and connects a lot faster.  But ap_scan=2 is needed for
non-SSID-broadcasting access points.

d) Even using the bssid= option for wpa_supplicant didn't reliably make
it connect to non-SSID-broadcasting access points with ap_scan=2

So, clearly more investigation is needed here for non-SSID-broadcasting
access points.

dan