[pkg-wpa-devel] Bug#471858: wpasupplicant: option "wpa_iface" half works - but does IP config on the wpa_iface not the actual interface (normally a bridge)

[Kel Modderman]

Hi Paul,

On Wednesday 26 March 2008 19:11:52 Paul Hedderly wrote:
> Kel Modderman wrote:
> > I think this patch works, but modifies wpa_action in a way it is clearly not
> > intended to be.
> >
> > When wpa_supplicant experiences a CONNECTED|DISCONNECTED event it executes
> > an action script like: <script> <interface> <event>, and wpa_cli sets a few env
> > variables at that time too.
> >
> > When <interface> is in a bridge do we always want to act on the bridge and not
> > the interface? If this is always the case should wpa_supplicant supply the
> > bridge interface name as argument when it is started with the bridge option?
> > If not, we should make this somehow adjustable.
> >
> > One way would be to have WPA_BRIDGE variable set by wpa_cli at script execution
> > time, as it does for WPA_ID and WPA_CTRL_DIR, rather than to rely on
> > environment variables inherited from daemon start time. Otherwise, the script
> > could check IF_WPA_BRIDGE before ignoring the arguments wpa_cli gave it and
> > using IFACE environment variable that is set only at daemon start time.
> >
> > Thanks, Kel.
> >   
> You are right that my patch it is a totally brute force approach and
> assumes that the wireless interface is the primary bridge interface - I 
> submitted it only to demonstrate what I wanted to be able to do and show
> that it is possible - I have actually changed most references to
> WPA_IFACE to IFACE in the script for myself in the down and check subs.
> But that is not a complete solution.
> 
> So yes having an interfaces variable like "wpa_control_bridge -" sounds
> like a great idea.
> 
> One thing I think should be clear, is that if the wireless interface is
> part of a bridge, it's own IP should never be set, and there should
> never be any routes on it.

Please try attached patch to /sbin/wpa_action, without your other changes,
see if it works.

> 
> So if the wireless is part of a bridge (and setting "wpa_iface" gives
> that expectation,) then wpa should either control IP/routes on the
> bridge or do nothing with IP  at all. So perhaps the option should be
> "wpa_ignore_ip -" having the default to do IP/route/dhcp setting on the
> bridge, giving room for those who might be using their laptop as
> effectively a roaming access point. (That would be a weird setup - using
> an AP to control the SSID, and using a station to route traffic... :O)
> 
> Hmm second thought - it may be more logical to set "wpa_iface <blah>"
> and "wpa_ip_iface <foO>" or "wpa_ip_iface none".
> 
> Anyway, my hack is making me very happy here - What I havent got round
> to testing properly is if it will still work fine with a normal setup.
> Oh hang it, I'll do a quick check now.... Yes - it all appears to work
> just fine.
> 
> By the way - are we much closer to a fix for 373180
> <mailto:373180 at bugs.debian.org>? When testing I keep forgetting to kill
> wpa_supplicant and I wonder why nothing works...

No. See my last comment on that thread.

Thanks, Kel.

---
--- debian/ifupdown/wpa_action.sh       (revision 1157)
+++ debian/ifupdown/wpa_action.sh       (working copy)
@@ -36,6 +36,10 @@
        exit 0
 fi

+if [ "$IF_WPA_BRIDGE" ]; then
+       WPA_IFACE="$IF_WPA_BRIDGE"
+fi
+
 case "$WPA_ACTION" in
        "CONNECTED")
                wpa_log_init
---