[PATCH v2 2/4] wifi: libertas: only add RSN/WPA IE in lbs_add_wpa_tlv

[Doug Brown]

Hi Simon,

On 1/13/2023 1:52 AM, Simon Horman wrote:
> On Sat, Jan 07, 2023 at 05:30:14PM -0800, Doug Brown wrote:
>> The existing code only converts the first IE to a TLV, but it returns a
>> value that takes the length of all IEs into account. When there is more
>> than one IE (which happens with modern wpa_supplicant versions for
>> example), the returned length is too long and extra junk TLVs get sent
>> to the firmware, resulting in an association failure.
>>
>> Fix this by finding the first RSN or WPA IE and only adding that. This
>> has the extra benefit of working properly if the RSN/WPA IE isn't the
>> first one in the IE buffer.
>>
>> While we're at it, clean up the code to use the available structs like
>> the other lbs_add_* functions instead of directly manipulating the TLV
>> buffer.
>>
>> Signed-off-by: Doug Brown <doug at schmorgal.com>
>> ---
>>   drivers/net/wireless/marvell/libertas/cfg.c | 28 +++++++++++++--------
>>   1 file changed, 18 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/net/wireless/marvell/libertas/cfg.c b/drivers/net/wireless/marvell/libertas/cfg.c
>> index 3e065cbb0af9..5cd78fefbe4c 100644
>> --- a/drivers/net/wireless/marvell/libertas/cfg.c
>> +++ b/drivers/net/wireless/marvell/libertas/cfg.c
>> @@ -416,10 +416,20 @@ static int lbs_add_cf_param_tlv(u8 *tlv)
>>   
>>   static int lbs_add_wpa_tlv(u8 *tlv, const u8 *ie, u8 ie_len)
>>   {
>> -	size_t tlv_len;
>> +	struct mrvl_ie_data *wpatlv = (struct mrvl_ie_data *)tlv;
>> +	const struct element *wpaie;
>> +
>> +	/* Find the first RSN or WPA IE to use */
>> +	wpaie = cfg80211_find_elem(WLAN_EID_RSN, ie, ie_len);
>> +	if (!wpaie)
>> +		wpaie = cfg80211_find_vendor_elem(WLAN_OUI_MICROSOFT,
>> +						  WLAN_OUI_TYPE_MICROSOFT_WPA,
>> +						  ie, ie_len);
>> +	if (!wpaie || wpaie->datalen > 128)
>> +		return 0;
>>   
>>   	/*
>> -	 * We need just convert an IE to an TLV. IEs use u8 for the header,
>> +	 * Convert the found IE to a TLV. IEs use u8 for the header,
>>   	 *   u8      type
>>   	 *   u8      len
>>   	 *   u8[]    data
>> @@ -428,14 +438,12 @@ static int lbs_add_wpa_tlv(u8 *tlv, const u8 *ie, u8 ie_len)
>>   	 *   __le16  len
>>   	 *   u8[]    data
>>   	 */
>> -	*tlv++ = *ie++;
>> -	*tlv++ = 0;
>> -	tlv_len = *tlv++ = *ie++;
>> -	*tlv++ = 0;
>> -	while (tlv_len--)
>> -		*tlv++ = *ie++;
>> -	/* the TLV is two bytes larger than the IE */
>> -	return ie_len + 2;
>> +	wpatlv->header.type = wpaie->id;
>> +	wpatlv->header.len = wpaie->datalen;
> 
> Hi Doug,
> 
> For correctness should type and len be converted to little endian,
> f.e. using cpu_to_le16() ?
> 
> Likewise in patch 4/4.

Thank you for reviewing my changes. You are absolutely right -- this
would be broken on big-endian systems. That was a major oversight on my
part. Not sure how I missed it because all of the other functions do it
correctly. Nice catch! I'll fix it in the next version of the series.

> 
>> +	memcpy(wpatlv->data, wpaie->data, wpaie->datalen);
>> +
>> +	/* Return the total number of bytes added to the TLV buffer */
>> +	return sizeof(struct mrvl_ie_header) + wpaie->datalen;
>>   }
>>   
>>   /*
>> -- 
>> 2.34.1
>>