Re: dhclient and wpa_supplicant

From: Stijn Hoop (stijn_at_win.tue.nl)
Date: 07/31/05


Date: Sun, 31 Jul 2005 23:45:38 +0200
To: Sam Leffler <sam@errno.com>


On Sun, Jul 31, 2005 at 11:31:58AM -0700, Sam Leffler wrote:
> Stijn Hoop wrote:
> > On Fri, Jul 29, 2005 at 06:08:53PM -0700, Sam Leffler wrote:
> > > I just ran tests with wpa_supplicant and found that it does appear to
> > > handle open ap's properly.
> >
> > I cannot get this to work, using 2 different MiniPCI cards and
> > a Linksys WRT54G, on a 7-CURRENT from yesterday.
> >
> > The cards:
> >
> > - an Intel Pro Wireless 2200B/G (model WM3A2200BG, Dell branded) using iwi
> > - a Dell Truemobile 1300 WLAN (bcm4306) using ndis
> >
> > For details, log files, etc, see
> >
> > http://sandcat.nl/~stijn/freebsd/20050731-wlan/
>
> Sigh, well for iwi you do say:
>
> "Maybe the problem is not with wpa_supplicant though; I cannot get
> if_iwi to associate by using ifconfig manually either."
>
> Er, yeah, if you cannot associate w/o wpa_supplicant I'm not sure why
> you'd think it'd be more capable? (unless you're using wpa)

Wasn't sure myself either; but maybe wpa_supplicant set some modes
differently then 'plain regular' ifconfig. At least now I know I have
to get it working first, so I'll try that in the future.

> The log shows:
>
> Jul 31 11:04:17 pounce kernel: iwi0: macaddr bssid
> chan rssi rate flag wep essid
> Jul 31 11:04:17 pounce kernel: - 00:50:fc:d6:85:82 00:50:fc:d6:85:82
> 11 38 11M ess no "default"!
>
> which means you found one ap w/ an ssid of "default" and it was rejected
> because you configured the device for a different ssid. However I see
> probe resp frames from LZ111 in your log but it's never added to the
> scan list for some reason. Seems like someone with iwi devices should
> investigate.

OK, so this confirms a bug in if_iwi. I'll send a PR.

> The ndis driver shows nothing in the log that indicates it hooks up
> properly to the net80211 layer so wpa_supplicant is unlikely to work
> with it. Unfortunately there are many drivers that bypass the net80211
> layer causing applications written to the common api to not work.

:-(

> OTOH your dhclient.conf file is so trivial that I'm not surprised it works;
> you could just as well have used:
>
> ifconfig_ndis0="ssid LZ111 DHCP"

Didn't know that either, good idea. One less config file to edit.

> The issue with dhclient using dhclient.conf to set wireless parameters
> is that it does so, then sends it's DHCP protocol msg, then waits a
> fixed amount of time for a response from a DHCP server. This is prone
> to fail as different configurations take different amounts of time to
> complete the scan+associate work. Also dhclient (now) processes link
> state and assoc/reassoc/disassoc msgs which confuses this approach. I
> did some changes to have it ignore these messages while setting the
> media parameters but it was complicated and still routinely failed
> because the timeouts were too small. In the end I concluded that it
> wasn't worth supporting this mechanism any more; that wpa_supplicant is
> the better approach since it will scan ONCE and then select the
> appropriate network (while dhclient had to scan once for each possible
> network). In general I have advocated that dhclient do less; in this
> case I want dhclient to let someone else setup the wireless networking
> parameters and have it do just the dhcp protocol. The fact that
> dhclient messes with media parameters for an interface is more a
> byproduct of configuring wired networks (and not having better tools to
> configure wireless networks).

OK, noted. So 'all it takes' is someone converting ndis to use all of the
802.11 api? (no idea how large a task this is).

> wpa_supplicant depends on a couple features in the net80211 layer to
> work properly: manual roaming (forcing the driver to not automtically do
> things like associate after scanning so it can instead decide which ap
> to use), scan (scan for ap's and report results), and associate to a
> particular ap (as identified in the scan results). Past that it uses
> the common ioctl's defined to set keys and other parameters that you
> normally get to via ifconfig. For wpa and 802.1x is needs some more
> stuff. Drivers that bypass the net80211 layer for some or all of this
> stuff may not work correctly and should be fixed.

Is there a list of these drivers? How about ral/ural? I read about
other problems with those. About the only useful wireless option then
is an Atheros card?

Thanks for the detailed reply, some of this stuff makes more sense now.

--Stijn

-- 
I really hate this damned machine
I wish that they would sell it.
It never does quite what I want
But only what I tell it.




Relevant Pages

  • Group Policy loading
    ... connected in to the network. ... wireless newsgroup) that Group Policy is just timing out. ... >>latest Dell and Buffalo firmware and drivers but to no ... >>file to match that of the Dell TrueMobile card. ...
    (microsoft.public.win2000.group_policy)
  • Linux 2.6.19.2: maybe a bug inside the r8169 network driver (was Re: Linux 2.6.19.2: Freeze with CIF
    ... It seems possible that my problem is related to a bug inside the r8169 network ... only partially support the current CIFS Unix Extensions). ... GigE drivers loaded on your system to determine if there is an easy way to ...
    (Linux-Kernel)
  • Re: Home network sharing a USB printer
    ... printers and the corresponding drivers only work when the driver can ... getting them to work over a network is impossible or very difficult. ... port - but I can't because its connected to the XP box in another room!!!! ... > When you look at the properties of a connection you will actually ...
    (microsoft.public.windowsxp.print_fax)
  • 2.6.13-rc2+ - problem with DHCP
    ... Network card is based on RTL8139C chip, ... # Firmware Drivers ... # ACPI (Advanced Configuration and Power Interface) Support ...
    (Linux-Kernel)
  • Re: No Local Area Connection
    ... Glad you found out about the network drivers. ... Warning 1116: Adding unreleased component to configuration: "Broadcom 440x ... The estimated configuration size may not be accurate. ...
    (microsoft.public.windowsxp.embedded)