Thread (22 messages) flat view 22 messages, 8 authors, 2009-08-12

Re: Libertas: Association request to the driver failed

From: Jonathan Cameron <hidden>
Date: 2009-08-12 14:47:18
Also in: lkml

Hi All, 

After applying this patch I've been receiving 0x12 response from
an access point (association failed: not all rates supported)
to association requests.

See below for queries on what is happening,
quoted hunk ↗ jump to hunk
Several arrays were read before checking whether the index was within
bounds. ARRAY_SIZE() should be used to determine the size of arrays.

rates->rates has an arraysize of 1, so calling get_common_rates()
with a rates_size of MAX_RATES (14) was causing reads out of bounds.

tmp_size can increment at most to MAX_RATES * ARRAY_SIZE(lbs_bg_rates),
so that should be the number of elements of tmp[].

Signed-off-by: Roel Kluin <redacted>
---
quoted
| Is it a good idea to use dynamic stack arrays in the kernel?
| What about kmalloc for dynamic allocations?
| 
| -- 
| Greetings, Michael.

I saw one pattern in trace code (not sure if it's
still there) but personally don't like dynamic
stack arrays (though at moment the max value
being passed into routine is known maybe just
use MAX_RATES instead of (*rates_size)?). Hmm?
Good point.
quoted
	-- Cyrill
Thanks,

I think there was another problem in lbs_associate(),
the memcpy already affected rates->rates.

Also in get_common_rates() I think we can safely move the
memset/memcpy, originally after label done, upwards.

The patch below, if correct, is to be applied after the revert

Roel
diff --git a/drivers/net/wireless/libertas/assoc.c b/drivers/net/wireless/libertas/assoc.c
index b9b3741..ba0164a 100644
--- a/drivers/net/wireless/libertas/assoc.c
+++ b/drivers/net/wireless/libertas/assoc.c
@@ -1,6 +1,7 @@
 /* Copyright (C) 2006, Red Hat, Inc. */
 
 #include <linux/types.h>
+#include <linux/kernel.h>
 #include <linux/etherdevice.h>
 #include <linux/ieee80211.h>
 #include <linux/if_arp.h>
@@ -43,41 +44,41 @@ static int get_common_rates(struct lbs_private *priv,
 	u16 *rates_size)
 {
 	u8 *card_rates = lbs_bg_rates;
-	size_t num_card_rates = sizeof(lbs_bg_rates);
-	int ret = 0, i, j;
-	u8 tmp[30];
+	int i, j;
+	u8 tmp[MAX_RATES * ARRAY_SIZE(lbs_bg_rates)];
 	size_t tmp_size = 0;
 
 	/* For each rate in card_rates that exists in rate1, copy to tmp */
-	for (i = 0; card_rates[i] && (i < num_card_rates); i++) {
-		for (j = 0; rates[j] && (j < *rates_size); j++) {
+	for (i = 0; i < ARRAY_SIZE(lbs_bg_rates) && card_rates[i]; i++) {
+		for (j = 0; j < *rates_size && rates[j]; j++) {
 			if (rates[j] == card_rates[i])
 				tmp[tmp_size++] = card_rates[i];
 		}
 	}
 
 	lbs_deb_hex(LBS_DEB_JOIN, "AP rates    ", rates, *rates_size);
-	lbs_deb_hex(LBS_DEB_JOIN, "card rates  ", card_rates, num_card_rates);
+	lbs_deb_hex(LBS_DEB_JOIN, "card rates  ", card_rates,
+			ARRAY_SIZE(lbs_bg_rates));
 	lbs_deb_hex(LBS_DEB_JOIN, "common rates", tmp, tmp_size);
 	lbs_deb_join("TX data rate 0x%02x\n", priv->cur_rate);
 
+	memset(rates, 0, *rates_size);
+	*rates_size = min_t(u16, tmp_size, *rates_size);
+	memcpy(rates, tmp, *rates_size);
+
 	if (!priv->enablehwauto) {
 		for (i = 0; i < tmp_size; i++) {
 			if (tmp[i] == priv->cur_rate)
-				goto done;
+				break;
+		}
+		if (i == tmp_size) {
+			lbs_pr_alert("Previously set fixed data rate %#x isn't "
+					"compatible with the network.\n",
+					priv->cur_rate);
+			return -1;
 		}
-		lbs_pr_alert("Previously set fixed data rate %#x isn't "
-		       "compatible with the network.\n", priv->cur_rate);
-		ret = -1;
-		goto done;
 	}
-	ret = 0;
-
-done:
-	memset(rates, 0, *rates_size);
-	*rates_size = min_t(int, tmp_size, *rates_size);
-	memcpy(rates, tmp, *rates_size);
-	return ret;
+	return 0;
 }
 
 
@@ -321,8 +322,8 @@ static int lbs_associate(struct lbs_private *priv,
 
 	rates = (struct mrvl_ie_rates_param_set *) pos;
 	rates->header.type = cpu_to_le16(TLV_TYPE_RATES);
-	memcpy(&rates->rates, &bss->rates, MAX_RATES);
-	tmplen = MAX_RATES;
+	tmplen = min_t(u16, ARRAY_SIZE(rates->rates), MAX_RATES);
Isn't this always going to be 1? Switching back to original version
allows association to work for me.

As is, it only allows one rate to be tested as ARRAY_SIZE(rates->rates)
is always 1 as it stands.  

If this is the desired behaviour please explain why?
I'll admit I'm not really sure what should be happening, I've merely
been bisecting looking for what was causing a regression for me.
quoted hunk ↗ jump to hunk
+	memcpy(&rates->rates, &bss->rates, tmplen);
 	if (get_common_rates(priv, rates->rates, &tmplen)) {
 		ret = -1;
 		goto done;
@@ -598,7 +599,7 @@ static int lbs_adhoc_join(struct lbs_private *priv,
 
 	/* Copy Data rates from the rates recorded in scan response */
 	memset(cmd.bss.rates, 0, sizeof(cmd.bss.rates));
-	ratesize = min_t(u16, sizeof(cmd.bss.rates), MAX_RATES);
+	ratesize = min_t(u16, ARRAY_SIZE(cmd.bss.rates), MAX_RATES);
 	memcpy(cmd.bss.rates, bss->rates, ratesize);
 	if (get_common_rates(priv, cmd.bss.rates, &ratesize)) {
 		lbs_deb_join("ADHOC_JOIN: get_common_rates returned error.\n");
_______________________________________________
libertas-dev mailing list
libertas-dev@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/libertas-dev
  
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help