On Fri, 17 Aug 2007 03:04:08 am Michael Renzmann wrote:
quoted text
> Hi all.
>
> Ticket #275 [1] deals with a patch that was expected to fix issues when
> scanning for and/or associating with Access Points that do not broadcast
> their ESSID.
>
> Part of the solution for this problem is the introduction of the function
> preempt_scan(). This gets called in case a scan is requested (by
> wpa_supplicant, for example) while a scan is still active. The old scan
> will be canceled, so that the probably changed scanning parameters can be
> honoured with the new scan request.
> The original patch has undergone some modifications and finally was
> committed to trunk in r1664.
>
> There is evidence, however, that preempt_scan() is the cause for new
> association issues. In #275 a patch is provided that removes
> preempt_scan() and it has been confirmed to solve association issues for
> some users.
>
> preempt_scan() brought us out of the frying pan into the fire, backing
> that patch out brings us back into the frying pan... no good deal, and the
> same for "let's ignore the situation". The best option obviously would be
> to fix the original issue without the side-effects that were introduced
> along with preempt_scan().
>
> Comments or suggestions, anyone?

Most of the complaintents so far have been using network-manager to
control their madwifi device.

The impact of this problem is best shown by the widespread effort to workaround
madwifi wext/scanning weaknesses by major distributions. Below is a patch from
the network-manager (version 0.6.5-0ubuntu9) package targetted for the next
big Ubuntu release, "gutsy". It seems that the origin of the patch can be traced
back to fedora community.

The most important observation that can be made, is the use of "madwifi" driver
backend in preference to "wext" for madwifi driven devices.

debian/patches/13-rml-wpa-workarounds.patch:

diff -Nur bzr.debian.0.9.5/src/nm-device-802-11-wireless.c bzr.debian.0.9.5.new/src/nm-device-802-11-wireless.c
--- bzr.debian.0.9.5/src/nm-device-802-11-wireless.c	2007-06-25 17:32:45.000000000 +0200
+++ bzr.debian.0.9.5.new/src/nm-device-802-11-wireless.c	2007-06-25 17:32:57.000000000 +0200
@@ -2764,6 +2764,9 @@
 	const char *		iface = nm_device_get_iface (NM_DEVICE (self));
 	gboolean			success = FALSE;
 	int				tries = 0;
+	const char *            wpa_driver;
+	const char *            kernel_driver;
+
 
 	/* Try to open wpa_supplicant's global control socket */
 	for (tries = 0; tries < WPA_SUPPLICANT_NUM_RETRIES && !ctrl; tries++)
@@ -2777,10 +2780,22 @@
 		nm_info ("Error opening supplicant global control interface.");
 		goto exit;
 	}
+     
+	kernel_driver = nm_device_get_driver (NM_DEVICE (self));
+
+       /*
+        * We want to work with the generic "wext" wpa_supplicant driver, but some kernel drivers
+        * are just utter junk.  For those losers, we use a specific wpa_supplicant driver.
+        */
+       
+	if (!strcmp (kernel_driver, "ath_pci"))
+               wpa_driver = "madwifi";
+       else
+               wpa_driver = "wext";
 
 	/* wpa_cli -g/var/run/wpa_supplicant-global interface_add eth1 "" wext /var/run/wpa_supplicant */
 	if (!nm_utils_supplicant_request_with_check (ctrl, "OK", __func__, NULL,
-			"INTERFACE_ADD %s\t\twext\t" WPA_SUPPLICANT_CONTROL_SOCKET "\t", iface))
+			"INTERFACE_ADD %s\t\t%s\t" WPA_SUPPLICANT_CONTROL_SOCKET "\t", iface, wpa_driver))
 		goto exit;
 	wpa_ctrl_close (ctrl);
 
@@ -2816,7 +2831,8 @@
 	struct wpa_ctrl *	ctrl;
 	gboolean			is_adhoc;
 	char *		hex_essid = NULL;
-	const char *		ap_scan = "AP_SCAN 1";
+	const char *		ap_scan;
+	const char *		kernel_driver;
 	guint32			caps;
 	gboolean			supports_wpa;
 
@@ -2845,6 +2861,46 @@
 	if (!nm_ap_get_broadcast (ap) || is_adhoc || !supports_wpa)
 		ap_scan = "AP_SCAN 2";
 
+	kernel_driver = nm_device_get_driver (NM_DEVICE (self));
+
+        /*
+         * We want to use "AP_SCAN 1", which tells wpa_supplicant to perform scanning.  This seems
+         * to work better with some drivers.  But we want "AP_SCAN 2", telling wpa_supplicant that
+         * we will do our own scanning, if
+         *
+         *      - The driver is orinoco_cs.  It chokes on "AP_SCAN 1".
+         *      - The AP is non-broadcast or Ad-Hoc.  Unless the driver is madwifi.
+         */
+        
+	if (!strcmp (kernel_driver, "orinoco_cs"))
+               	ap_scan = "AP_SCAN 2";
+        else if ((!nm_ap_get_broadcast (ap) || is_adhoc) && strcmp (kernel_driver, "ath_pci"))
+               	ap_scan = "AP_SCAN 2";
+        else if (!nm_ap_get_broadcast (ap) || is_adhoc || !supports_wpa)
+		ap_scan = "AP_SCAN 2";
+	else
+               	ap_scan = "AP_SCAN 1";
+
+	kernel_driver = nm_device_get_driver (NM_DEVICE (self));
+
+        /*
+         * We want to use "AP_SCAN 1", which tells wpa_supplicant to perform scanning.  This seems
+         * to work better with some drivers.  But we want "AP_SCAN 2", telling wpa_supplicant that
+         * we will do our own scanning, if
+         *
+         *      - The driver is orinoco_cs.  It chokes on "AP_SCAN 1".
+         *      - The AP is non-broadcast or Ad-Hoc.  Unless the driver is madwifi.
+         */
+        
+	if (!strcmp (kernel_driver, "orinoco_cs"))
+               	ap_scan = "AP_SCAN 2";
+        else if ((!nm_ap_get_broadcast (ap) || is_adhoc) && strcmp (kernel_driver, "ath_pci"))
+               	ap_scan = "AP_SCAN 2";
+        else if (!nm_ap_get_broadcast (ap) || is_adhoc || !supports_wpa)
+		ap_scan = "AP_SCAN 2";
+	else
+               	ap_scan = "AP_SCAN 1";
+
 	/* Tell wpa_supplicant that we'll do the scanning */
 	if (!nm_utils_supplicant_request_with_check (ctrl, "OK", __func__, NULL, ap_scan))
 		goto out;

-------------------------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc.
Still grepping through log files to find problems?  Stop.
Now Search log events and configuration files using AJAX and a browser.
Download your FREE copy of Splunk now >>  http://get.splunk.com/