wpa_supplicant for ad-hoc mode

[Shawn Rutledge]

My goal is to use wpa_supplicant to take care of all wireless
configuration, including both ad-hoc and AP modes, because it's very
nice to have one config file and have wpa_supplicant maintain it, and
use the socket to instruct wpa_supplicant about changes... I can build
a nice embedded configuration UI without having to write conf files
myself, and without ever having to fork a process just to run iwconfig
or ifup or the like (which I currently have had to do).

This page says there are some kernel patches that are necessary, which
were to be merged into 2.6.26:

http://lists.shmoo.com/pipermail/hostap/2008-July/018036.html

On my laptop I run kernel 2.6.27 (Gentoo version) with rtl8187 for my
Alfa AWUS036H adapter.  (That is because ath5k is not working quite
well enough for the built-in Atheros chip, which lspci identifies as
AR5006EG... and I've read elsewhere that the ability to identify which
Atheros chip it is exactly, is not reliable.  With the ath5k driver,
iwlist scan does not return any results, and wpa_supplicant also
cannot scan.)  With the RTL driver, I cannot get wpa_supplicant to
connect in ad-hoc mode.  I already added the line

mode=1

to the relevant section.  (And IMO it's a bug that wpa_supplicant does
not do that itself.  It can tell that the ESSID is an ad-hoc one, when
it does the scan, right?  If you can be sure that the essid is only
ad-hoc, can't be an AP, then why not write that mode flag?  If not,
then wpa_gui needs a setting for ad-hoc.)

Yet, to fix it, all I need to do (while wpa_supplicant is still
running - I don't even have to kill it!) is

ifconfig wlan1 down
iwconfig wlan1 mode ad-hoc essid my-essid

and when I do that, via strace I can see that iwconfig only does 2
ioctls - that's enough to get it working.

ioctl(3, SIOCSIWMODE, 0xbfb52efc)       = 0
ioctl(3, SIOCSIWESSID, 0xbfb52ef8)      = 0

I'm having almost the same experience on the embedded system I'm
working on (with different drivers).

wpa_supplicant on the other hand does much more (when I watch it via
strace as it's trying to configure the interface in ad-hoc mode) so my
thought was all the excessive frobbing got the driver confused... so I
commented out a lot of ioctls in the wpa_supplicant "driver" and got
it down to just those same two ioctls.  Still no luck.  If it's so
easy for iwconfig to get it working, wpa_supplicant could do the
initialization more or less the same way, right?

What are these kernel patches?  Did they make it into 2.6.27?  Anyway
the embedded system is running 2.6.18, so if I need to apply the same
patches to that kernel as well, anybody know where to get them?

(BTW I have been doing most of the experiments with wpa_supplicant
0.6.7, and some with 0.5.7)