Help with testing the AP+Client Bug

[DoesItMatter]

Eric,

How can I help test the AP+Client bug?

Attaching an image of my home network - It's definitely NOT your standard home config !!

My Fonera is purely for testing, so is open to whatever help you need with troubleshooting.

Let me know if you need logs or what type of tests I could help you run.

[Eric]

Configure the Buffalo Wireless G router (doesn't look like it's connected to anything) as an access point. Then configure the Fon to use AP+Client mode to connect with the Buffalo router (select connect via dhcp wireless), and configure it to have an access point with a separate SSID from the buffalo router. Connect from a laptop/pc to the SSID of the Fonera AP (if you can). If you do manage to connect the connection will fail in less than an hour, likely much sooner. It will fail whether you have encryption enabled or not, though the symptoms seem to be slightly different.

One of the problems is that everyone agrees that it fails, but the symptoms of failure seem to be different, time to failure differs, as well as seeing different messages from the system log (output of logread after sshing in). So, it would be helpful if you could run some tests in this configuration and provide: time to failure, output of logread after failure (connect another pc to the LAN port on the Fonera to ssh into it and get the output), and encryption/lack thereof used on both the client and AP. Multiple tests in the same configuration may yield different results too, so the more times you test the better. To reset things to a base state reboot the router after each failure to make sure we recover from the failure properly. I know this is tedious so if you don't want to worry about it I understand, but it would be helpful.

Right now the exact symptoms of this thing seem to be all over the map, other than that the connection dies. It would be very helpful to better characterize these symptoms so that I have a better idea where to look in the mountain of driver code that makes up the madwifi driver.

[crisman]

I think problem is in a patch of this changeset:

https://dev.openwrt.org/changeset/13708 ... i/patches/

In this changeset, many patches from trunk were merged in 8.09.

So, the first attempt could be:

try openwrt compiled from branch 8.09 with 13707 revision;
if problem is present I'm wrong

If not, try openwrt compiled from branch 8.09 with 13708 revision;
if problem is present, the bug is in one or more patches of this changeset. Else we must look in other later changesets.

Could you make this test?
If you don't know how to compile openwrt (or you haven't time to compile), tell me and I will compile the 2 different version for you.

[Eric]

To start with, I've been trying to pinpoint the problem when encryption is not active. I think this problem starts later, at r15465. At least, this changeset makes the problem a lot worse. However, I think the problems with encryption are older, and you may very well be right that 13708 is responsible.

Compiling is easy... it just takes a bit of time and installation takes another 45 minutes. So it takes a while to test one version and my time has been limited lately. It would be really helpful if you could also run this test... I'm a bit worried because everyone is reporting different symptoms, and there are a large number of variables that could be responsible including (but not limited to): wireless chip in AP client is connecting to, distance to that AP, distance of connecting computer to AP and random chance (if it's some sort of race/deadlock condition which of several sections of code gets there first).

[crisman]

Could you make this test?
If you don't know how to compile openwrt (or you haven't time to compile), tell me and I will compile the 2 different version for you.

This was referred to anyone want help to find and to resolve this bug.

I'm not at home right now and I will be at home only next weekend, so I cannot make tests before.

If anyone want to make tests, and want the two revisions compiled, just ask!

[DoesItMatter]

I'm not a Unix guy. I know Unix and am familiar with it, but I would need to get pointers.

If you can tell me where/how to put those 2 compiled versions and how to activate them, I can do that.

Also Eric - What directory could I get the logs from on the Fonera?

[crisman]

OK, I compiled both versions ( revision 13707 and revision 13708 ).

You can download them here : http://www.megaupload.com/?d=R2AE49TO

In the tar.gz there are file for fonera and for ubnt ns2.

How to put them in your fonera?
Search the forum and you will find the answer.

What test to do?

Try first the revision 13707 in client mode ( you need to connect to an encrypted network) with a virtual ap ( set encryption also in virtual ap, different from your main ap). Connect with a pc to fonera in wireless mode, and browse internet,download files,ecc. After a while ssh in the fonera and post the output of "logread".

Make same tests for the other revision.

[DoesItMatter]

OK, will try more tests this weekend with these re-builds.

--------------------------

Eric, did testing as you advised.

Tried for about 3-4 hours, and I could only connect once to the Fonera with Gargoyle 1.0.0 setup as Client + AP.

Jul 15 11:51:14 OpenWrt user.notice root: adding wan (br-wan) to firewall zone w an
Jul 15 11:51:15 OpenWrt daemon.info dnsmasq[1411]: reading /tmp/resolv.conf.auto
Jul 15 11:51:15 OpenWrt daemon.info dnsmasq[1411]: using nameserver 192.168.1.1# 53
Jul 15 11:51:15 OpenWrt daemon.info dnsmasq[1411]: using local addresses only fo r domain lan
Jul 15 11:51:53 OpenWrt daemon.info dnsmasq[1411]: DHCPDISCOVER(br-lan) 00:14:a5 :d9:82:cf
Jul 15 11:51:53 OpenWrt daemon.info dnsmasq[1411]: DHCPOFFER(br-lan) 192.168.1.1 90 00:14:a5:d9:82:cf
Jul 15 11:51:53 OpenWrt daemon.info dnsmasq[1411]: DHCPREQUEST(br-lan) 192.168.1 .190 00:14:a5:d9:82:cf
Jul 15 11:51:53 OpenWrt daemon.info dnsmasq[1411]: DHCPACK(br-lan) 192.168.1.190 00:14:a5:d9:82:cf small-laptop

Is the only relevant info I saw. My wireless test laptop connected for about 10 secs, then connection dropped.

---------------

I use a program called Xirrus Wi-Fi Inspector to see the wireless networks around and also signal strengths, etc.

With the Fon 2201+ and Gargoyle 1.0, setup as a normal router DHCP Wired access point, in Xirrus - it shows the vendor type of router as FON

When I setup the Fon 2201+ as DHCP Wireless, in the Client + AP mode, in Xirrus, it shows the vendor type as "Allied Telesyn Internaional"

What is that Allied Telesyn? Is that a wrapper driver that allows the Atheros to both be a client + access point? I wonder if there is something in that driver/wrapper that is causing the issues?

It won't let me attach the logread file, but I can e-mail it if necessary.

[crisman]

When I setup the Fon 2201+ as DHCP Wireless, in the Client + AP mode, in Xirrus, it shows the vendor type as "Allied Telesyn Internaional"

I think it happens because when setting ap+sta, the mac ap of fonera become 06:18... So, I think your program identify that as Allied.. I don't think this is related to the issue, but who know?

It won't let me attach the logread file, but I can e-mail it if necessary.

Upload it using a file hosting server, like megaupload.com or similar.

Best regards

[Eric]

Ok, I have a version that seems to be working for me... but since everyone seems to report different symptoms I'm not sure this is a definitive solution. I'm hoping you guys can help test.

What I did (and I admit this is a horrible, ugly hack) is revert madwifi, hostapd and wpa_supplicant back to revision 13000, just after the 8.09 branch of openwrt was created, and make a couple necessary config changes for compatibility. I have this version installed on my Fon+ and configured to ap+client mode with wpa2 psk encryption. It's been running fine for the past 4 hours, without any interruption in the connection.

So, here's the new version (Gargoyle 1.0.1 + older madwifi & related packages). Please, if you can, help test this:
gargoyle-madwifi-testA.squashfs
gargoyle-madwifi-testA.lzma

Thanks in advance to all who help test!