[PATCH] cfg80211: check correct maximum bandwidth for quarter and half rate.

[PATCH] cfg80211: check correct maximum bandwidth for quarter and half rate.

[Matthias May]

When using quarter and half rates we might want to use self defined
frequencies with self defined country codes closer to the border.
To avoid these frequencies to be disabled, we need to check if
the frequency fits the band with the actual bandwidth.

Signed-off-by: Matthias May <matthias.may@neratec.com>
---
 net/wireless/reg.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index fc2f135..c8fabda 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -1016,6 +1016,7 @@ freq_reg_info_regd(struct wiphy *wiphy, u32 center_freq,
 	for (i = 0; i < regd->n_reg_rules; i++) {
 		const struct ieee80211_reg_rule *rr;
 		const struct ieee80211_freq_range *fr = NULL;
+		u32 max_bw = MHZ_TO_KHZ(20);
 
 		rr = &regd->reg_rules[i];
 		fr = &rr->freq_range;
@@ -1028,8 +1028,10 @@ freq_reg_info_regd(struct wiphy *wiphy, u32 center_freq,
 		 */
 		if (!band_rule_found)
 			band_rule_found = freq_in_rule_band(fr, center_freq);
+		if (fr->max_bandwidth_khz < max_bw)
+			max_bw = fr->max_bandwidth_khz;
 
-		bw_fits = reg_does_bw_fit(fr, center_freq, MHZ_TO_KHZ(20));
+		bw_fits = reg_does_bw_fit(fr, center_freq, max_bw);
 
 		if (band_rule_found && bw_fits)
 			return rr;
-- 
2.1.4

Re: [PATCH] cfg80211: check correct maximum bandwidth for quarter and half rate.

[Johannes Berg]

On Tue, 2015-06-09 at 17:27 +0200, Matthias May wrote:
> When using quarter and half rates we might want to use self defined
> frequencies with self defined country codes closer to the border.
> To avoid these frequencies to be disabled, we need to check if
> the frequency fits the band with the actual bandwidth.

> +++ b/net/wireless/reg.c
> @@ -1016,6 +1016,7 @@ freq_reg_info_regd(struct wiphy *wiphy, u32 center_freq,
>  	for (i = 0; i < regd->n_reg_rules; i++) {
>  		const struct ieee80211_reg_rule *rr;
>  		const struct ieee80211_freq_range *fr = NULL;
> +		u32 max_bw = MHZ_TO_KHZ(20);
>  
>  		rr = &regd->reg_rules[i];
>  		fr = &rr->freq_range;
> @@ -1028,8 +1028,10 @@ freq_reg_info_regd(struct wiphy *wiphy, u32 center_freq,
>  		 */
>  		if (!band_rule_found)
>  			band_rule_found = freq_in_rule_band(fr, center_freq);
> +		if (fr->max_bandwidth_khz < max_bw)
> +			max_bw = fr->max_bandwidth_khz;
>  
> -		bw_fits = reg_does_bw_fit(fr, center_freq, MHZ_TO_KHZ(20));
> +		bw_fits = reg_does_bw_fit(fr, center_freq, max_bw);

So the old code here assumes 20 MHz channel bandwidth, which was
reasonable until 5/10 MHz were supported.

However, your change looks very odd.

Consider a situation where for some reason you have a regulatory domain
without 20 MHz channels at all, only allowing a max bandwidth of 10 MHz.
Then, this code will cause all checks for "channels" to be erroneously
successful, since you're not really checking the request against the
regd.

What's needed instead is to actually pass in the requested bandwidth
from the caller. Additionally, it seems that at least the caller in
handle_channel_custom() must loop through the available bandwidths
(5/10/20) and disable those that don't fit, instead of disabling the
channel if 20 MHz doesn't fit.

johannes

Re: [PATCH] cfg80211: check correct maximum bandwidth for quarter and half rate.

[Matthias May]

On 09/06/15 22:29, Johannes Berg wrote:
> On Tue, 2015-06-09 at 17:27 +0200, Matthias May wrote:
>> When using quarter and half rates we might want to use self defined
>> frequencies with self defined country codes closer to the border.
>> To avoid these frequencies to be disabled, we need to check if
>> the frequency fits the band with the actual bandwidth.
>> +++ b/net/wireless/reg.c
>> @@ -1016,6 +1016,7 @@ freq_reg_info_regd(struct wiphy *wiphy, u32 center_freq,
>>   	for (i = 0; i < regd->n_reg_rules; i++) {
>>   		const struct ieee80211_reg_rule *rr;
>>   		const struct ieee80211_freq_range *fr = NULL;
>> +		u32 max_bw = MHZ_TO_KHZ(20);
>>   
>>   		rr = &regd->reg_rules[i];
>>   		fr = &rr->freq_range;
>> @@ -1028,8 +1028,10 @@ freq_reg_info_regd(struct wiphy *wiphy, u32 center_freq,
>>   		 */
>>   		if (!band_rule_found)
>>   			band_rule_found = freq_in_rule_band(fr, center_freq);
>> +		if (fr->max_bandwidth_khz < max_bw)
>> +			max_bw = fr->max_bandwidth_khz;
>>   
>> -		bw_fits = reg_does_bw_fit(fr, center_freq, MHZ_TO_KHZ(20));
>> +		bw_fits = reg_does_bw_fit(fr, center_freq, max_bw);
> So the old code here assumes 20 MHz channel bandwidth, which was
> reasonable until 5/10 MHz were supported.
>
> However, your change looks very odd.
>
> Consider a situation where for some reason you have a regulatory domain
> without 20 MHz channels at all, only allowing a max bandwidth of 10 MHz.
> Then, this code will cause all checks for "channels" to be erroneously
> successful, since you're not really checking the request against the
> regd.
>
> What's needed instead is to actually pass in the requested bandwidth
> from the caller. Additionally, it seems that at least the caller in
> handle_channel_custom() must loop through the available bandwidths
> (5/10/20) and disable those that don't fit, instead of disabling the
> channel if 20 MHz doesn't fit.
>
> johannes
>
Hi
The scenario you describe, is actually what i'm working on.
I have some self defined country codes and some self defined frequencies.
Constructed scenario:
I run a single 5Mhz wide channel on the frequency 5175.
I created a dummy countrycode XS which only allows 5MHz wide channels.
root@RM2:~# iw reg get
country XS: DFS-UNSET
     (5172 - 5247 @ 5), (N/A, 15), (N/A)

Without the patch the channel 5175 and 5180 would be disabled because 
the check sees that 20MHz won't fit.
I'm not sure what exactly you mean that the handle_channel_custom needs 
to loop through 5/10/20.
The freq_reg_info_regd sets the disabled flag on the channel at init.
This is not while trying to start operation on a channel, because the 
channel is already disabled.
Or do you mean to actually have different sets of flags for the 
different bandwidths?

I see that it's erroneously possible to run a 10/20MHz channel on 5175, 
even though this shouldn't be allowed.
However setting the proper flags bw_flags should fix this.

Expanding the patch with the patch below ensures that one can't start 
operation on a channel which it's not allowed on.
Or is there a better way?

diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index c8fabda..4f96563 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -1139,9 +1139,12 @@ static void handle_channel(struct wiphy *wiphy,
      /* Check if auto calculation requested */
      if (reg_rule->flags & NL80211_RRF_AUTO_BW)
          max_bandwidth_khz = reg_get_max_bandwidth(regd, reg_rule);
-
+    if (max_bandwidth_khz < MHZ_TO_KHZ(10))
+        bw_flags = IEEE80211_CHAN_NO_10MHZ;
+    if (max_bandwidth_khz < MHZ_TO_KHZ(20))
+        bw_flags |= IEEE80211_CHAN_NO_20MHZ;
      if (max_bandwidth_khz < MHZ_TO_KHZ(40))
-        bw_flags = IEEE80211_CHAN_NO_HT40;
+        bw_flags |= IEEE80211_CHAN_NO_HT40;
      if (max_bandwidth_khz < MHZ_TO_KHZ(80))
          bw_flags |= IEEE80211_CHAN_NO_80MHZ;
      if (max_bandwidth_khz < MHZ_TO_KHZ(160))
@@ -1575,8 +1578,12 @@ static void handle_channel_custom(struct wiphy 
*wiphy,
      if (reg_rule->flags & NL80211_RRF_AUTO_BW)
          max_bandwidth_khz = reg_get_max_bandwidth(regd, reg_rule);

+    if (max_bandwidth_khz < MHZ_TO_KHZ(10))
+        bw_flags = IEEE80211_CHAN_NO_10MHZ;
+    if (max_bandwidth_khz < MHZ_TO_KHZ(20))
+        bw_flags |= IEEE80211_CHAN_NO_20MHZ;
      if (max_bandwidth_khz < MHZ_TO_KHZ(40))
-        bw_flags = IEEE80211_CHAN_NO_HT40;
+        bw_flags |= IEEE80211_CHAN_NO_HT40;
      if (max_bandwidth_khz < MHZ_TO_KHZ(80))
          bw_flags |= IEEE80211_CHAN_NO_80MHZ;
      if (max_bandwidth_khz < MHZ_TO_KHZ(160))

Matthias

Re: [PATCH] cfg80211: check correct maximum bandwidth for quarter and half rate.

[Johannes Berg]

On Wed, 2015-06-10 at 14:26 +0200, Matthias May wrote:

> I'm not sure what exactly you mean that the handle_channel_custom needs 
> to loop through 5/10/20.
> The freq_reg_info_regd sets the disabled flag on the channel at init.

That's kinda my point - this isn't true. freg_req_info_regd() *doesn't*
change any channel flags, it just returns some information, and through
freq_reg_info() this is even exported to drivers.

> This is not while trying to start operation on a channel, because the 
> channel is already disabled.
> Or do you mean to actually have different sets of flags for the 
> different bandwidths?

I mean basically what you did below, except handle it explicitly at the
outer level.

> I see that it's erroneously possible to run a 10/20MHz channel on 5175, 
> even though this shouldn't be allowed.
> However setting the proper flags bw_flags should fix this.
> 
> Expanding the patch with the patch below ensures that one can't start 
> operation on a channel which it's not allowed on.
> Or is there a better way?

I was thinking you should extend freq_reg_info_regd() to get the desired
bandwidth, in which case you'd have to call it three times (5, 10, 20).

However, you could also make the API so that freq_reg_info_regd() gets
the *minimum* bandwidth [that the caller is willing to handle].

That way, you could set the argument to 20 MHz (for now) in
freq_reg_info(), and to 5 MHz in the other callers that you're modifying
(with the proposed additional patch) to actually handle the returned
bandwidth lower than 20 MHz.



This is really the crux of the problem - with your patch,
freq_reg_info_regd() and thus freq_reg_info() can return something that
doesn't fit the requested channel. You're changing the semantics of this
function to no longer mean "return the rule that fits a 20MHz channel
around the given center frequency" but to mean "return a rule around the
given center frequency" - no matter how wide that rule can allow
channels to be.

If you change the function to pass in a "minimum required bandwidth"
argument *and* change the checking as suggested with the combination, I
think it'll be OK.

johannes