[PATCH 0/7] cfg80211/mac80211: Add 802.11d support

[PATCH 0/7] cfg80211/mac80211: Add 802.11d support

[Luis R. Rodriguez]

This patch series adds 802.11d support to mac80211 and
cfg80211. I've tried to split the required changes as much
as possible to make it easier for review. I've tested this
on iwlagn with OLD_REG enabled and disable and with /sbin/crda
missing and not. I'm sure I've missed a few corner cases, but
I think this is a good start.

  Luis

[PATCH 1/7] cfg80211: mark negative frequencies as invalid

[Luis R. Rodriguez]

Regulatory rules with negative frequencies are now
marked as invalid in is_valid_reg_rule().

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
---
 net/wireless/reg.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index 4c7e39d..7a2ce27 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -330,7 +330,7 @@ static bool is_valid_reg_rule(const struct ieee80211_reg_rule *rule)
 	const struct ieee80211_freq_range *freq_range = &rule->freq_range;
 	u32 freq_diff;
 
-	if (freq_range->start_freq_khz == 0 || freq_range->end_freq_khz == 0)
+	if (freq_range->start_freq_khz <= 0 || freq_range->end_freq_khz <= 0)
 		return false;
 
 	if (freq_range->start_freq_khz > freq_range->end_freq_khz)
-- 
1.5.6.3

[PATCH 2/7] cfg80211: remove switch from __regulatory_hint()

[Luis R. Rodriguez]

We have complete control over REGDOM_SET_BY_* enum passed
down to __regulatory_hint() as such there is no need to
account for unexpected REGDOM_SET_BY_*'s, lets just remove
the switch statement as this code does not change and
won't change even when we add 802.11d support.

This patch has no functional changes.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
---
 net/wireless/reg.c |   43 +++++++++++++++++--------------------------
 1 files changed, 17 insertions(+), 26 deletions(-)

diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index 7a2ce27..e44f005 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -698,34 +698,25 @@ int __regulatory_hint(struct wiphy *wiphy, enum reg_set_by set_by,
 	else if (r)
 		return r;
 
-	switch (set_by) {
-	case REGDOM_SET_BY_CORE:
-	case REGDOM_SET_BY_COUNTRY_IE:
-	case REGDOM_SET_BY_DRIVER:
-	case REGDOM_SET_BY_USER:
-		request = kzalloc(sizeof(struct regulatory_request),
-				  GFP_KERNEL);
-		if (!request)
-			return -ENOMEM;
-
-		request->alpha2[0] = alpha2[0];
-		request->alpha2[1] = alpha2[1];
-		request->initiator = set_by;
-		request->wiphy = wiphy;
-		request->intersect = intersect;
-
-		kfree(last_request);
-		last_request = request;
-		r = call_crda(alpha2);
+	request = kzalloc(sizeof(struct regulatory_request),
+			  GFP_KERNEL);
+	if (!request)
+		return -ENOMEM;
+
+	request->alpha2[0] = alpha2[0];
+	request->alpha2[1] = alpha2[1];
+	request->initiator = set_by;
+	request->wiphy = wiphy;
+	request->intersect = intersect;
+
+	kfree(last_request);
+	last_request = request;
+	r = call_crda(alpha2);
+
 #ifndef CONFIG_WIRELESS_OLD_REGULATORY
-		if (r)
-			printk(KERN_ERR "cfg80211: Failed calling CRDA\n");
+	if (r)
+		printk(KERN_ERR "cfg80211: Failed calling CRDA\n");
 #endif
-		break;
-	default:
-		r = -ENOTSUPP;
-		break;
-	}
 
 	return r;
 }
-- 
1.5.6.3

[PATCH 3/7] cfg80211: remove switch from __set_regdom()

[Luis R. Rodriguez]

We have control over the REGDOM_SET_BY_* macros passed
so remove the switch.

This patch has no functional changes.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
---
 net/wireless/reg.c |   21 ++++++---------------
 1 files changed, 6 insertions(+), 15 deletions(-)

diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index e44f005..290ecec 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -829,21 +829,12 @@ static int __set_regdom(const struct ieee80211_regdomain *rd)
 	reset_regdomains();
 
 	/* Country IE parsing coming soon */
-	switch (last_request->initiator) {
-	case REGDOM_SET_BY_CORE:
-	case REGDOM_SET_BY_DRIVER:
-	case REGDOM_SET_BY_USER:
-		if (!is_valid_rd(rd)) {
-			printk(KERN_ERR "cfg80211: Invalid "
-				"regulatory domain detected:\n");
-			print_regdomain_info(rd);
-			return -EINVAL;
-		}
-		break;
-	case REGDOM_SET_BY_COUNTRY_IE: /* Not yet */
-		WARN_ON(1);
-	default:
-		return -EOPNOTSUPP;
+
+	if (!is_valid_rd(rd)) {
+		printk(KERN_ERR "cfg80211: Invalid "
+			"regulatory domain detected:\n");
+		print_regdomain_info(rd);
+		return -EINVAL;
 	}
 
 	if (unlikely(last_request->intersect)) {
-- 
1.5.6.3

[PATCH 4/7] cfg80211: separate intersection section in __set_regdom()

[Luis R. Rodriguez]

So far the __set_regdom() code is pretty generic as the
intersection case is fairly straight forward; this will however
change when 802.11d support is added so lets separate intersection
code for now in preparation for 802.11d support.

This patch only has slight functional changes.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
---
 net/wireless/reg.c |   29 +++++++++++++++++++++--------
 1 files changed, 21 insertions(+), 8 deletions(-)

diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index 290ecec..845e2d3 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -826,10 +826,6 @@ static int __set_regdom(const struct ieee80211_regdomain *rd)
 	if (WARN_ON(!reg_is_valid_request(rd->alpha2)))
 		return -EINVAL;
 
-	reset_regdomains();
-
-	/* Country IE parsing coming soon */
-
 	if (!is_valid_rd(rd)) {
 		printk(KERN_ERR "cfg80211: Invalid "
 			"regulatory domain detected:\n");
@@ -837,16 +833,33 @@ static int __set_regdom(const struct ieee80211_regdomain *rd)
 		return -EINVAL;
 	}
 
-	if (unlikely(last_request->intersect)) {
+	if (!last_request->intersect) {
+		reset_regdomains();
+		cfg80211_regdomain = rd;
+		return 0;
+	}
+
+	/* Intersection requires a bit more work */
+
+	if (last_request->initiator != REGDOM_SET_BY_COUNTRY_IE) {
+
 		intersected_rd = regdom_intersect(rd, cfg80211_regdomain);
 		if (!intersected_rd)
 			return -EINVAL;
+
+		/* We can trash what CRDA provided now */
 		kfree(rd);
-		rd = intersected_rd;
+		rd = NULL;
+
+		reset_regdomains();
+		cfg80211_regdomain = intersected_rd;
+
+		return 0;
 	}
 
-	/* Tada! */
-	cfg80211_regdomain = rd;
+	/* Country IE parsing coming soon */
+	reset_regdomains();
+	WARN_ON(1);
 
 	return 0;
 }
-- 
1.5.6.3

[PATCH 5/7] cfg80211: expect different rd in cfg80211 when intersecting

[Luis R. Rodriguez]

When intersecting it is possible that set_regdom() was called
with a regulatory domain which we'll only use as an aid to
build a final regulatory domain.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
---
 net/wireless/reg.c |    5 +++--
 1 files changed, 3 insertions(+), 2 deletions(-)

diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index 845e2d3..f062b8b 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -880,12 +880,13 @@ int set_regdom(const struct ieee80211_regdomain *rd)
 	}
 
 	/* This would make this whole thing pointless */
-	BUG_ON(rd != cfg80211_regdomain);
+	if (!last_request->intersect)
+		BUG_ON(rd != cfg80211_regdomain);
 
 	/* update all wiphys now with the new established regulatory domain */
 	update_all_wiphy_regulatory(last_request->initiator);
 
-	print_regdomain(rd);
+	print_regdomain(cfg80211_regdomain);
 
 	return r;
 }
-- 
1.5.6.3

[PATCH 6/7] cfg80211: initialize last_request

[Luis R. Rodriguez]

last_request was not being initialized. By initializing it
we don't fix a bug as we are not checking for it before prior
to a regulatory hint however we may do so at a later time.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
---
 net/wireless/reg.c |    2 ++
 1 files changed, 2 insertions(+), 0 deletions(-)

diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index f062b8b..3754d0d 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -899,6 +899,8 @@ int regulatory_init(void)
 	if (IS_ERR(reg_pdev))
 		return PTR_ERR(reg_pdev);
 
+	last_request = NULL;
+
 #ifdef CONFIG_WIRELESS_OLD_REGULATORY
 	cfg80211_regdomain = static_regdom(ieee80211_regdom);
 
-- 
1.5.6.3

[PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Luis R. Rodriguez]

This adds country IE parsing to mac80211 and enables its usage
within the new regulatory infrastructure in cfg80211. We parse
the country IEs only on management beacons (when your associated)
and disregard the IEs when the country matches the country already
picked up by the device through IEs.

To avoid following misinformed or outdated APs we build and use
a regulatory domain out of the intersection between what the AP
provides us on the country IE and what CRDA is aware is allowed
on the same country.

A secondary device is allowed to follow only the same country IE
as it make no sense for two devices on a system to be in two
different countries.

In the case the AP is using country IEs for an incorrect country
the user may help compliance further by setting the regulatory
domain and in that case another intersection will be performed.

CONFIG_WIRELESS_OLD_REGULATORY is supported but requires CRDA
present.

Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
---
 include/linux/ieee80211.h |   62 ++++++++++++
 include/net/wireless.h    |   60 ++++++++++++
 net/mac80211/Kconfig      |    8 ++
 net/mac80211/mlme.c       |  236 +++++++++++++++++++++++++++++++++++++++++++++
 net/wireless/Kconfig      |   11 ++
 net/wireless/core.c       |   16 +++
 net/wireless/nl80211.c    |    2 +-
 net/wireless/reg.c        |  235 ++++++++++++++++++++++++++++++++++++++++-----
 net/wireless/reg.h        |    4 +-
 9 files changed, 608 insertions(+), 26 deletions(-)

diff --git a/include/linux/ieee80211.h b/include/linux/ieee80211.h
index 56b0eb2..5e2bf16 100644
--- a/include/linux/ieee80211.h
+++ b/include/linux/ieee80211.h
@@ -1042,6 +1042,68 @@ enum ieee80211_spectrum_mgmt_actioncode {
 	WLAN_ACTION_SPCT_CHL_SWITCH = 4,
 };
 
+/*
+ * IEEE 802.11-2007 7.3.2.9 Country information element
+ *
+ * Minimum length is 8 octects, ie len must be evenly
+ * divisible by 2
+ */
+
+/* Although the spec says 8 I'm seeing 6 in practice */
+#define IEEE80211_COUNTRY_IE_MIN_LEN	6
+
+/*
+ * Regulatory extension stuff is for 802.11j, ammeneded to
+ * IEEE 802.11-2007, see Annex I (page 1141) and Annex J
+ * (page 1147). Also review 7.3.2.9.
+ *
+ * When dot11RegulatoryClassesRequired is true and the
+ * first_channel/reg_extension_id is >= 201 then the IE
+ * compromises of the 'ext' struct represented below:
+ *
+ *  - Regulatory extension ID - when generating IE this just needs
+ *    to be monotonically increasing for each triplet passed in
+ *    the IE
+ *  - Regulatory class - index into set of rules
+ *  - Coverage class - index into air propagation time (Table 7-27),
+ *    in microseconds, you can compute the air propagation time from
+ *    the index by multiplying by 3, so index 10 yields a propagation
+ *    of 10 us. Valid values are 0-31, values 32-255 are not defined
+ *    yet. A value of 0 inicates air propagation of <= 1 us.
+ *
+ *  See also Table I.2 for Emission limit sets and table
+ *  I.3 for Behavior limit sets. Table J.1 indicates how to map
+ *  a reg_class to an emission limit set and behavior limit set.
+ */
+#define IEEE80211_COUNTRY_EXTENSION_ID 201
+
+/*
+ *  Channels numbers in the IE must be monntonically increasing
+ *  if dot11RegulatoryClassesRequired is not true.
+ *
+ *  If dot11RegulatoryClassesRequired is true consecutive
+ *  subband triplets following a regulatory triplet shall
+ *  have monotonically increasing first_channel number fields.
+ *
+ *  Channel numbers shall not overlap.
+ *
+ *  Note that max_power is signed.
+ */
+struct ieee80211_country_ie_triplet {
+	union {
+		struct {
+			u8 first_channel;
+			u8 num_channels;
+			s8 max_power;
+		} __attribute__ ((packed)) chans;
+		struct {
+			u8 reg_extension_id;
+			u8 reg_class;
+			u8 coverage_class;
+		} __attribute__ ((packed)) ext;
+	};
+} __attribute__ ((packed));
+
 /* BACK action code */
 enum ieee80211_back_actioncode {
 	WLAN_ACTION_ADDBA_REQ = 0,
diff --git a/include/net/wireless.h b/include/net/wireless.h
index 17d4b58..0dcd15c 100644
--- a/include/net/wireless.h
+++ b/include/net/wireless.h
@@ -181,6 +181,13 @@ struct ieee80211_supported_band {
  * struct wiphy - wireless hardware description
  * @idx: the wiphy index assigned to this item
  * @class_dev: the class device representing /sys/class/ieee80211/<wiphy-name>
+ * @country_ie_alpha2: ISO / IEC 3166 alpha2 for which this wiphy is receiving
+ * 	country IEs on, this should help drivers and the wireless
+ * 	stack disregard country IEs from APs quickly on the same alpha2.
+ * 	The alpha2 may differ from cfg80211_regdomain's alpha2 when
+ * 	an intersection has occurred. If the AP is reconfigured this
+ * 	can also be used to tell us if the country on the country
+ * 	IE changed.
  * @reg_notifier: the driver's regulatory notification callback
  */
 struct wiphy {
@@ -201,6 +208,8 @@ struct wiphy {
 
 	struct ieee80211_supported_band *bands[IEEE80211_NUM_BANDS];
 
+	char country_ie_alpha2[2];
+
 	/* Lets us get back the wiphy on the callback */
 	int (*reg_notifier)(struct wiphy *wiphy, enum reg_set_by setby);
 
@@ -373,4 +382,55 @@ ieee80211_get_response_rate(struct ieee80211_supported_band *sband,
  * for a regulatory domain structure for the respective country.
  */
 extern void regulatory_hint(struct wiphy *wiphy, const char *alpha2);
+
+/**
+ * regulatory_hint_ie - wireless stack hints a regulatory domain
+ * @wiphy: the wireless device giving the hint (used only for reporting
+ *	conflicts)
+ * @rd: a regulatory domain built from the received country IE
+ * @country_ie_checksum: checksum of country IE of fields we are interested in,
+ * 	we can remove this if its determined the IE never should change
+ * 	for an AP, keep in mind dot11RegulatoryClassesRequired.
+ *
+ * We will intersect the rd with the what CRDA tells us should apply
+ * for the alpha2 this country IE belongs to, this prevents APs from
+ * sending us incorrect or outdated information against a country.
+ * If CONFIG_WIRELESS_OLD_REGULATORY is enabled and CRDA is not present
+ * we will intersect immediately against what is currently set in the
+ * wireless core.
+ */
+extern int regulatory_hint_ie(struct wiphy *wiphy,
+	struct ieee80211_regdomain *rd,
+	u32 country_ie_checksum);
+
+/**
+ * cfg80211_is_country_change - tells us if wiphy moves to new country
+ * @wiphy: the wireless device on which the IE was received on
+ * @alpha2: the alpha2 for the country IE the wiphy received
+ *
+ * This can be used to determine if a country IE a wiphy receives
+ * has changed the alpha2. The structure of the IEs generated on
+ * the AP should not change. If the the country changes we should
+ * disassociate and re-associate as the AP was probably reconfigured
+ * and rebooted, hopefully with a correct country setting.
+ *
+ * We don't support two wiphys being present in two different
+ * countries, however it might be possible a different wiphy
+ * already set the current regulatory domain; in that case we
+ * want to process this wiphy's country IE in case it may be
+ * different even if on the same alpha2.
+ */
+extern bool cfg80211_is_country_change(struct wiphy *wiphy,
+	const char *alpha2);
+
+/**
+ * country_ie_integrity_changes - tells us if the country IE has changed
+ * @checksum: checksum of country IE of fields we are interested in
+ *
+ * If the country IE has not changed you can ignore it safely. This is
+ * useful to determine if two wiphys are seeing two different country IEs
+ * even on the same alpha2. Note that this will return false if no IE has
+ * been set on the wireless core yet.
+ */
+extern bool country_ie_integrity_changes(u32 checksum);
 #endif /* __NET_WIRELESS_H */
diff --git a/net/mac80211/Kconfig b/net/mac80211/Kconfig
index 7f710a2..219fb7b 100644
--- a/net/mac80211/Kconfig
+++ b/net/mac80211/Kconfig
@@ -203,6 +203,14 @@ config MAC80211_DEBUG_COUNTERS
 
 	  If unsure, say N.
 
+config MAC80211_COUNTRY_IE_DEBUG
+	bool "Country IE parsing debug (IEEE 802.11d) debugging"
+	depends on MAC80211_DEBUG_MENU
+	---help---
+	  Say Y here to print out verbose country IE parsing (IEEE 802.11d)
+	  debug messages. This generates a lot of messages. You should only
+	  enable this option if no regulatory changes are being made.
+
 config MAC80211_VERBOSE_SPECT_MGMT_DEBUG
 	bool "Verbose Spectrum Management (IEEE 802.11h)debugging"
 	depends on MAC80211_DEBUG_MENU
diff --git a/net/mac80211/mlme.c b/net/mac80211/mlme.c
index 12064fb..08675b4 100644
--- a/net/mac80211/mlme.c
+++ b/net/mac80211/mlme.c
@@ -648,6 +648,211 @@ static void ieee80211_sta_send_apinfo(struct ieee80211_sub_if_data *sdata,
 	wireless_send_event(sdata->dev, SIOCGIWAP, &wrqu, NULL);
 }
 
+/* Converts a country IE to a regulatory domain. A regulatory domain
+ * structure has a lot of information which the IE doesn't yet have,
+ * so for the other values we use upper max values as we will intersect
+ * with our userspace regulatory agent to get lower bounds. */
+static struct ieee80211_regdomain *ieee80211_country_ie_2_rd(
+				u8 *country_ie,
+				u8 country_ie_len,
+				u32 *checksum)
+{
+	struct ieee80211_regdomain *rd = NULL;
+	unsigned int i = 0;
+	char alpha2[2];
+	u32 flags = 0;
+	u32 num_rules = 0, size_of_regd = 0;
+	u8 *triplets_start = NULL;
+	u8 len_at_triplet = 0;
+	/* Used too often */
+	int triplet_size = sizeof(struct ieee80211_country_ie_triplet);
+	/* the last channel we have registered in a subband (triplet) */
+	int last_sub_max_channel = 0;
+
+	*checksum = 0xDEADBEEF;
+
+	/* Country IE requirements */
+	BUG_ON(country_ie_len < IEEE80211_COUNTRY_IE_MIN_LEN ||
+		country_ie_len & 0x01);
+
+	alpha2[0] = country_ie[0];
+	alpha2[1] = country_ie[1];
+
+	/*
+	 * Third octet can be:
+	 *    'I' - Indoor
+	 *    'O' - Outdoor
+	 *
+	 *  anything else we assume is no restrictions
+	 */
+	if (country_ie[2] == 'I')
+		flags = NL80211_RRF_NO_OUTDOOR;
+	else if (country_ie[2] == 'O')
+		flags = NL80211_RRF_NO_INDOOR;
+
+	country_ie += 3;
+	country_ie_len -= 3;
+
+	triplets_start = country_ie;
+	len_at_triplet = country_ie_len;
+
+	*checksum ^= flags ^ alpha2[0] ^ alpha2[1];
+
+	/* We need to build a reg rule for each triplet, but first we must
+	 * calculate the number of reg rules we will need. We will need one
+	 * for each channel subband */
+	while (country_ie_len >= triplet_size) {
+		struct ieee80211_country_ie_triplet *triplet =
+			(struct ieee80211_country_ie_triplet *) country_ie;
+		int cur_sub_max_channel = 0, cur_channel = 0;
+
+		if (triplet->ext.reg_extension_id >=
+				IEEE80211_COUNTRY_EXTENSION_ID) {
+			country_ie += triplet_size;
+			country_ie_len -= triplet_size;
+			continue;
+		}
+
+		cur_channel = triplet->chans.first_channel;
+		cur_sub_max_channel = ieee80211_channel_to_frequency(
+			cur_channel + triplet->chans.num_channels);
+
+		/* Basic sanity check */
+		if (cur_sub_max_channel < cur_channel)
+			return NULL;
+
+		/* Do not allow overlapping channels. Also channels
+		 * passed in each subband must be monotonically
+		 * increasing */
+		if (last_sub_max_channel) {
+			if (cur_channel <= last_sub_max_channel)
+				return NULL;
+			if (cur_sub_max_channel <= last_sub_max_channel)
+				return NULL;
+		}
+
+		/* When dot11RegulatoryClassesRequired is supported
+		 * we can throw ext triplets as part of this soup,
+		 * for now we don't care when those change as we
+		 * don't support them */
+		*checksum ^= cur_channel ^ cur_sub_max_channel ^
+			triplet->chans.max_power;
+
+		last_sub_max_channel = cur_sub_max_channel;
+
+		country_ie += triplet_size;
+		country_ie_len -= triplet_size;
+		num_rules++;
+
+		/* Note: this is not a IEEE requirement but
+		 * simply a memory requirement */
+		if (num_rules > NL80211_MAX_SUPP_REG_RULES)
+			return NULL;
+	}
+
+	/* This will be true if a second wiphy receives a different
+	 * country IE than the one we last processed and accepted,
+	 * for now we simply disregard these. */
+	if (country_ie_integrity_changes(*checksum))
+		return NULL;
+
+	country_ie = triplets_start;
+	country_ie_len = len_at_triplet;
+
+	size_of_regd = sizeof(struct ieee80211_regdomain) +
+		(num_rules * sizeof(struct ieee80211_reg_rule));
+
+	rd = kzalloc(size_of_regd, GFP_KERNEL);
+	if (!rd)
+		return NULL;
+
+	rd->n_reg_rules = num_rules;
+	rd->alpha2[0] = alpha2[0];
+	rd->alpha2[1] = alpha2[1];
+
+	/* This time around we fill in the rd */
+	while (country_ie_len >= triplet_size) {
+		struct ieee80211_country_ie_triplet *triplet =
+			(struct ieee80211_country_ie_triplet *) country_ie;
+		struct ieee80211_reg_rule *reg_rule = NULL;
+		struct ieee80211_freq_range *freq_range = NULL;
+		struct ieee80211_power_rule *power_rule = NULL;
+
+		/* Must parse if dot11RegulatoryClassesRequired is true,
+		 * we don't support this yet */
+		if (triplet->ext.reg_extension_id >=
+				IEEE80211_COUNTRY_EXTENSION_ID) {
+			country_ie += triplet_size;
+			country_ie_len -= triplet_size;
+			continue;
+		}
+
+		reg_rule = &rd->reg_rules[i];
+		freq_range = &reg_rule->freq_range;
+		power_rule = &reg_rule->power_rule;
+
+		reg_rule->flags = flags;
+
+		/* The +10 is since the regulatory domain expects
+		 * the actual band edge, not the center of freq for
+		 * its start and end freqs, assuming 20 MHz bandwidth on
+		 * the channels passed */
+		freq_range->start_freq_khz =
+			MHZ_TO_KHZ(ieee80211_channel_to_frequency(
+				triplet->chans.first_channel) + 10);
+		freq_range->end_freq_khz =
+			MHZ_TO_KHZ(ieee80211_channel_to_frequency(
+				triplet->chans.first_channel +
+					triplet->chans.num_channels) + 10);
+
+		/* Large arbitrary values, we intersect later */
+		/* Increment this if we ever support >= 40 MHz channels
+		 * in IEEE 802.11 */
+		freq_range->max_bandwidth_khz = MHZ_TO_KHZ(40);
+		power_rule->max_antenna_gain = DBI_TO_MBI(100);
+		power_rule->max_eirp = DBM_TO_MBM(100);
+
+		country_ie += triplet_size;
+		country_ie_len -= triplet_size;
+		i++;
+
+		BUG_ON(i > NL80211_MAX_SUPP_REG_RULES);
+	}
+
+	return rd;
+}
+
+static int ieee80211_sta_process_country_ie(struct ieee80211_local *local,
+				u8 *country_ie, u8 country_ie_len)
+{
+	char alpha2[2];
+	struct ieee80211_regdomain *rd = NULL;
+	u32 checksum = 0;
+
+	/* IE len must be evenly divisible by 2 */
+	if (country_ie_len & 0x01)
+		return -EINVAL;
+
+	if (country_ie_len < IEEE80211_COUNTRY_IE_MIN_LEN)
+		return -EINVAL;
+
+	/* Note if the AP is reconfigured to outdoor this doesn't pick
+	 * that up, but that assumes we can keep the state alive while it
+	 * does that */
+	alpha2[0] = country_ie[0];
+	alpha2[1] = country_ie[1];
+
+	/* Ignore IEs coming in on this wiphy which with the same alpha2 */
+	if (!cfg80211_is_country_change(local->hw.wiphy, alpha2))
+		return -EALREADY;
+
+	rd = ieee80211_country_ie_2_rd(country_ie, country_ie_len, &checksum);
+	if (!rd)
+		return -EINVAL;
+
+	return regulatory_hint_ie(local->hw.wiphy, rd, checksum);
+}
+
 static void ieee80211_sta_send_associnfo(struct ieee80211_sub_if_data *sdata,
 					 struct ieee80211_if_sta *ifsta)
 {
@@ -1681,6 +1886,29 @@ static void ieee80211_rx_mgmt_probe_resp(struct ieee80211_sub_if_data *sdata,
 	}
 }
 
+#ifdef CONFIG_MAC80211_COUNTRY_IE_DEBUG
+static void ieee80211_country_ie_results_debug(int r)
+{
+	if (likely(r)) {
+		if (likely(r == -EALREADY)) {
+			printk(KERN_DEBUG "mac80211: disgarding"
+				" already processed country "
+				"IE\n");
+		} else {
+			printk(KERN_DEBUG "mac80211: Unable "
+				"to parse country IE\n");
+		}
+	} else {
+		printk(KERN_DEBUG "mac80211: Parsed "
+			"country IE correctly\n");
+	}
+}
+#else
+static inline void ieee80211_country_ie_results_debug(int r)
+{
+	return;
+}
+#endif
 
 static void ieee80211_rx_mgmt_beacon(struct ieee80211_sub_if_data *sdata,
 				     struct ieee80211_mgmt *mgmt,
@@ -1694,6 +1922,7 @@ static void ieee80211_rx_mgmt_beacon(struct ieee80211_sub_if_data *sdata,
 	u32 changed = 0;
 	bool erp_valid;
 	u8 erp_value = 0;
+	int r = 0;
 
 	/* Process beacon from the current BSS */
 	baselen = (u8 *) mgmt->u.beacon.variable - (u8 *) mgmt;
@@ -1753,6 +1982,13 @@ static void ieee80211_rx_mgmt_beacon(struct ieee80211_sub_if_data *sdata,
 					       ap_ht_cap_flags);
 	}
 
+	if (elems.country_elem) {
+		/* Note we are only reviewing this on management beacons */
+		r = ieee80211_sta_process_country_ie(local,
+			elems.country_elem, elems.country_elem_len);
+		ieee80211_country_ie_results_debug(r);
+	}
+
 	ieee80211_bss_info_change_notify(sdata, changed);
 }
 
diff --git a/net/wireless/Kconfig b/net/wireless/Kconfig
index ae7f226..c40a0f2 100644
--- a/net/wireless/Kconfig
+++ b/net/wireless/Kconfig
@@ -1,6 +1,15 @@
 config CFG80211
         tristate "Improved wireless configuration API"
 
+config CFG80211_REG_DEBUG
+        bool "cfg80211 regulatory debugging"
+	depends on CFG80211
+	default n
+	---help---
+	  You can enable this if you want to debug regulatory changes.
+
+	  If unsure, say N.
+
 config NL80211
 	bool "nl80211 new netlink interface support"
 	depends on CFG80211
@@ -40,6 +49,8 @@ config WIRELESS_OLD_REGULATORY
 	  ieee80211_regdom module parameter. This is being phased out and you
 	  should stop using them ASAP.
 
+	  Note: You will need CRDA if you want 802.11d support
+
 	  Say Y unless you have installed a new userspace application.
 	  Also say Y if have one currently depending on the ieee80211_regdom
 	  module parameter and cannot port it to use the new userspace
diff --git a/net/wireless/core.c b/net/wireless/core.c
index 72825af..a9e40c3 100644
--- a/net/wireless/core.c
+++ b/net/wireless/core.c
@@ -134,6 +134,22 @@ cfg80211_get_dev_from_ifindex(int ifindex)
 	return drv;
 }
 
+bool cfg80211_is_country_change(struct wiphy *wiphy,
+	const char *alpha2)
+{
+	if (!wiphy->country_ie_alpha2) {
+		if (alpha2)
+			return true;
+		else
+			return false;
+	}
+	if (wiphy->country_ie_alpha2[0] == alpha2[0] &&
+			wiphy->country_ie_alpha2[1] == alpha2[1])
+		return false;
+	return true;
+}
+EXPORT_SYMBOL(cfg80211_is_country_change);
+
 void cfg80211_put_dev(struct cfg80211_registered_device *drv)
 {
 	BUG_ON(IS_ERR(drv));
diff --git a/net/wireless/nl80211.c b/net/wireless/nl80211.c
index e3e1494..a7edd1e 100644
--- a/net/wireless/nl80211.c
+++ b/net/wireless/nl80211.c
@@ -1760,7 +1760,7 @@ static int nl80211_req_set_reg(struct sk_buff *skb, struct genl_info *info)
 		return -EINVAL;
 #endif
 	mutex_lock(&cfg80211_drv_mutex);
-	r = __regulatory_hint(NULL, REGDOM_SET_BY_USER, data);
+	r = __regulatory_hint(NULL, REGDOM_SET_BY_USER, data, 0);
 	mutex_unlock(&cfg80211_drv_mutex);
 	return r;
 }
diff --git a/net/wireless/reg.c b/net/wireless/reg.c
index 3754d0d..6d32684 100644
--- a/net/wireless/reg.c
+++ b/net/wireless/reg.c
@@ -60,12 +60,15 @@
  * @intersect: indicates whether the wireless core should intersect
  * 	the requested regulatory domain with the presently set regulatory
  * 	domain.
+ * @country_ie_checksum: checksum of the last processed and accepted
+ * 	country IE
  */
 struct regulatory_request {
 	struct wiphy *wiphy;
 	enum reg_set_by initiator;
 	char alpha2[2];
 	bool intersect;
+	u32 country_ie_checksum;
 };
 
 /* Receipt of information from last regulatory request */
@@ -85,6 +88,11 @@ static u32 supported_bandwidths[] = {
  * information to give us an alpha2 */
 static const struct ieee80211_regdomain *cfg80211_regdomain;
 
+/* We use this as a place for the rd structure built from the
+ * last parsed country IE to rest until CRDA gets back to us with
+ * what it thinks should apply for the same country */
+static const struct ieee80211_regdomain *country_ie_regdomain;
+
 /* We keep a static world regulatory domain in case of the absence of CRDA */
 static const struct ieee80211_regdomain world_regdom = {
 	.n_reg_rules = 1,
@@ -264,6 +272,18 @@ static bool is_unknown_alpha2(const char *alpha2)
 	return false;
 }
 
+static bool is_intersected_alpha2(const char *alpha2)
+{
+	if (!alpha2)
+		return false;
+	/* Special case where regulatory domain is the
+	 * result of an intersection between two regulatory domain
+	 * structures */
+	if (alpha2[0] == '9' && alpha2[1] == '8')
+		return true;
+	return false;
+}
+
 static bool is_an_alpha2(const char *alpha2)
 {
 	if (!alpha2)
@@ -660,16 +680,14 @@ static int ignore_request(struct wiphy *wiphy, enum reg_set_by set_by,
 					return -EOPNOTSUPP;
 				return -EALREADY;
 			}
-			/* Two consecutive Country IE hints on the same wiphy */
-			if (!alpha2_equal(cfg80211_regdomain->alpha2, alpha2))
+			/* Two consecutive Country IE hints on the same wiphy.
+			 * This should be picked up early by the driver/stack */
+			if (WARN_ON(!alpha2_equal(cfg80211_regdomain->alpha2,
+					alpha2)))
 				return 0;
 			return -EALREADY;
 		}
-		/*
-		 * Ignore Country IE hints for now, need to think about
-		 * what we need to do to support multi-domain operation.
-		 */
-		return -EOPNOTSUPP;
+		return REG_INTERSECT;
 	case REGDOM_SET_BY_DRIVER:
 		if (last_request->initiator == REGDOM_SET_BY_DRIVER)
 			return -EALREADY;
@@ -677,6 +695,11 @@ static int ignore_request(struct wiphy *wiphy, enum reg_set_by set_by,
 	case REGDOM_SET_BY_USER:
 		if (last_request->initiator == REGDOM_SET_BY_COUNTRY_IE)
 			return REG_INTERSECT;
+		/* If the user knows better the user should set the regdom
+		 * to their country before the IE is picked up */
+		if (last_request->initiator == REGDOM_SET_BY_USER &&
+				last_request->intersect)
+			return -EOPNOTSUPP;
 		return 0;
 	}
 
@@ -685,7 +708,8 @@ static int ignore_request(struct wiphy *wiphy, enum reg_set_by set_by,
 
 /* Caller must hold &cfg80211_drv_mutex */
 int __regulatory_hint(struct wiphy *wiphy, enum reg_set_by set_by,
-		      const char *alpha2)
+			const char *alpha2,
+			u32 country_ie_checksum)
 {
 	struct regulatory_request *request;
 	bool intersect = false;
@@ -708,15 +732,22 @@ int __regulatory_hint(struct wiphy *wiphy, enum reg_set_by set_by,
 	request->initiator = set_by;
 	request->wiphy = wiphy;
 	request->intersect = intersect;
+	request->country_ie_checksum = country_ie_checksum;
 
 	kfree(last_request);
-	last_request = request;
-	r = call_crda(alpha2);
 
-#ifndef CONFIG_WIRELESS_OLD_REGULATORY
-	if (r)
-		printk(KERN_ERR "cfg80211: Failed calling CRDA\n");
-#endif
+	last_request = request;
+	call_crda(alpha2);
+
+	/* Note: When CONFIG_WIRELESS_OLD_REGULATORY is enabled
+	 * AND if CRDA is NOT present nothing will happen, if someone
+	 * wants to bother with 11d with OLD_REG you can add a timer.
+	 * If after x amount of time nothing happens you can call:
+	 *
+	 * return set_regdom(country_ie_regdomain);
+	 *
+	 * to intersect with the static rd
+	 */
 
 	return r;
 }
@@ -726,11 +757,72 @@ void regulatory_hint(struct wiphy *wiphy, const char *alpha2)
 	BUG_ON(!alpha2);
 
 	mutex_lock(&cfg80211_drv_mutex);
-	__regulatory_hint(wiphy, REGDOM_SET_BY_DRIVER, alpha2);
+	__regulatory_hint(wiphy, REGDOM_SET_BY_DRIVER, alpha2, 0);
 	mutex_unlock(&cfg80211_drv_mutex);
 }
 EXPORT_SYMBOL(regulatory_hint);
 
+bool country_ie_integrity_changes(u32 checksum)
+{
+	/* If no IE has been set then the checksum doesn't change */
+	if (unlikely(!last_request->country_ie_checksum))
+		return false;
+	if (unlikely(last_request->country_ie_checksum != checksum))
+		return true;
+	return false;
+}
+EXPORT_SYMBOL(country_ie_integrity_changes);
+
+/* The only thing this should allow through without a complaint
+ * is the case where a second device received a country IE
+ * with a different checksum. We provide helpers to avoid
+ * the other cases very early on to not waste resources */
+static bool reg_same_country_ie_hint(struct wiphy *wiphy,
+			u32 country_ie_checksum)
+{
+	if (!last_request->wiphy)
+		return false;
+	if (last_request->wiphy != wiphy) {
+		/* The stack/drivers should check for this earlier and
+		 * not let these through */
+		if (WARN_ON(!country_ie_integrity_changes(country_ie_checksum)))
+			return true;
+		return false;
+	}
+	/* The stack/drivers should not let these through */
+	if (WARN_ON(!country_ie_integrity_changes(country_ie_checksum)))
+		return true;
+	return false;
+}
+
+int regulatory_hint_ie(struct wiphy *wiphy, struct ieee80211_regdomain *rd,
+			u32 country_ie_checksum)
+{
+	int r = -EINVAL;
+
+	mutex_lock(&cfg80211_drv_mutex);
+
+	/* Pending country IE processing, this can happen after we
+	 * call CRDA and wait for a response if a beacon was received before
+	 * we were able to process the last regulatory_hint_ie() call */
+	if (country_ie_regdomain)
+		goto out;
+
+	if (unlikely(reg_same_country_ie_hint(wiphy, country_ie_checksum)))
+		goto out;
+
+	/* We keep this around for when CRDA comes back with
+	 * a response, so we can interesct with that */
+	country_ie_regdomain = rd;
+
+	r = __regulatory_hint(wiphy, REGDOM_SET_BY_COUNTRY_IE,
+		country_ie_regdomain->alpha2, country_ie_checksum);
+
+out:
+	mutex_unlock(&cfg80211_drv_mutex);
+	return r;
+}
+EXPORT_SYMBOL(regulatory_hint_ie);
 
 static void print_rd_rules(const struct ieee80211_regdomain *rd)
 {
@@ -770,7 +862,24 @@ static void print_rd_rules(const struct ieee80211_regdomain *rd)
 static void print_regdomain(const struct ieee80211_regdomain *rd)
 {
 
-	if (is_world_regdom(rd->alpha2))
+	if (is_intersected_alpha2(rd->alpha2)) {
+		struct wiphy *wiphy = NULL;
+
+
+		if (last_request->initiator == REGDOM_SET_BY_COUNTRY_IE) {
+			if (last_request->wiphy) {
+				wiphy = last_request->wiphy;
+				printk(KERN_INFO "cfg80211: Current regulatory "
+					"domain updated by AP to: %c%c\n",
+					wiphy->country_ie_alpha2[0],
+					wiphy->country_ie_alpha2[1]);
+			} else
+				printk(KERN_INFO "cfg80211: Current regulatory "
+					"intersected: \n");
+		} else
+				printk(KERN_INFO "cfg80211: Current regulatory "
+					"intersected: \n");
+	} else if (is_world_regdom(rd->alpha2))
 		printk(KERN_INFO "cfg80211: World regulatory "
 			"domain updated:\n");
 	else {
@@ -793,10 +902,39 @@ static void print_regdomain_info(const struct ieee80211_regdomain *rd)
 	print_rd_rules(rd);
 }
 
+#ifdef CONFIG_CFG80211_REG_DEBUG
+static void reg_country_ie_process_debug(
+	const struct ieee80211_regdomain *rd,
+	const struct ieee80211_regdomain *country_ie_regdomain,
+	const struct ieee80211_regdomain *intersected_rd)
+{
+	printk(KERN_DEBUG "cfg80211: Received country IE:\n");
+	print_regdomain_info(country_ie_regdomain);
+	printk(KERN_DEBUG "cfg80211: CRDA thinks this should applied:\n");
+	print_regdomain_info(rd);
+	if (intersected_rd) {
+		printk(KERN_DEBUG "cfg80211: We intersect both of these "
+			"and get:\n");
+		print_regdomain_info(rd);
+		return;
+	}
+	printk(KERN_DEBUG "cfg80211: Intersection between both failed\n");
+}
+#else
+static inline void reg_country_ie_process_debug(
+	const struct ieee80211_regdomain *rd,
+	const struct ieee80211_regdomain *country_ie_regdomain,
+	const struct ieee80211_regdomain *intersected_rd)
+{
+	return;
+}
+#endif
+
 /* Takes ownership of rd only if it doesn't fail */
 static int __set_regdom(const struct ieee80211_regdomain *rd)
 {
 	const struct ieee80211_regdomain *intersected_rd = NULL;
+	struct wiphy *wiphy = NULL;
 	/* Some basic sanity checks first */
 
 	if (is_world_regdom(rd->alpha2)) {
@@ -813,10 +951,18 @@ static int __set_regdom(const struct ieee80211_regdomain *rd)
 	if (!last_request)
 		return -EINVAL;
 
-	/* allow overriding the static definitions if CRDA is present */
-	if (!is_old_static_regdom(cfg80211_regdomain) &&
-	    !regdom_changed(rd->alpha2))
-		return -EINVAL;
+	/* Lets only bother proceeding on the same alpha2 if the current
+	 * rd is non static (it means CRDA was present and was used last)
+	 * and the pending request came in from a country IE */
+	if (last_request->initiator != REGDOM_SET_BY_COUNTRY_IE) {
+		/* If someone else asked us to change the rd lets only bother
+		 * checking if the alpha2 changes if CRDA was already called */
+		if (!is_old_static_regdom(cfg80211_regdomain) &&
+		    !regdom_changed(rd->alpha2))
+			return -EINVAL;
+	}
+
+	wiphy = last_request->wiphy;
 
 	/* Now lets set the regulatory domain, update all driver channels
 	 * and finally inform them of what we have done, in case they want
@@ -857,9 +1003,44 @@ static int __set_regdom(const struct ieee80211_regdomain *rd)
 		return 0;
 	}
 
-	/* Country IE parsing coming soon */
+	/*
+	 * Country IE requests are handled a bit differently, we intersect
+	 * the country IE rd and what CRDA believes that country should have
+	 */
+
+	BUG_ON(!country_ie_regdomain);
+
+	if (rd != country_ie_regdomain) {
+		/* Intersect what CRDA returned and our what we
+		 * had built from the Country IE received */
+
+		intersected_rd = regdom_intersect(rd, country_ie_regdomain);
+
+		reg_country_ie_process_debug(rd, country_ie_regdomain,
+			intersected_rd);
+
+		kfree(country_ie_regdomain);
+		country_ie_regdomain = NULL;
+	} else {
+		/* This would happen when CRDA was not present and
+		 * OLD_REGULATORY was enabled. We intersect our Country
+		 * IE rd and what was set on cfg80211 originally */
+		intersected_rd = regdom_intersect(rd, cfg80211_regdomain);
+	}
+
+	if (!intersected_rd)
+		return -EINVAL;
+
+	wiphy->country_ie_alpha2[0] = rd->alpha2[0];
+	wiphy->country_ie_alpha2[1] = rd->alpha2[1];
+
+	BUG_ON(intersected_rd == rd);
+
+	kfree(rd);
+	rd = NULL;
+
 	reset_regdomains();
-	WARN_ON(1);
+	cfg80211_regdomain = intersected_rd;
 
 	return 0;
 }
@@ -900,6 +1081,7 @@ int regulatory_init(void)
 		return PTR_ERR(reg_pdev);
 
 	last_request = NULL;
+	country_ie_regdomain = NULL;
 
 #ifdef CONFIG_WIRELESS_OLD_REGULATORY
 	cfg80211_regdomain = static_regdom(ieee80211_regdom);
@@ -912,11 +1094,11 @@ int regulatory_init(void)
 	 * that is not a valid ISO / IEC 3166 alpha2 */
 	if (ieee80211_regdom[0] != 'E' || ieee80211_regdom[1] != 'U')
 		err = __regulatory_hint(NULL, REGDOM_SET_BY_CORE,
-					ieee80211_regdom);
+					ieee80211_regdom, 0);
 #else
 	cfg80211_regdomain = cfg80211_world_regdom;
 
-	err = __regulatory_hint(NULL, REGDOM_SET_BY_CORE, "00");
+	err = __regulatory_hint(NULL, REGDOM_SET_BY_CORE, "00", 0);
 	if (err)
 		printk(KERN_ERR "cfg80211: calling CRDA failed - "
 		       "unable to update world regulatory domain, "
@@ -932,6 +1114,11 @@ void regulatory_exit(void)
 
 	reset_regdomains();
 
+	if (unlikely(country_ie_regdomain)) {
+		kfree(country_ie_regdomain);
+		country_ie_regdomain = NULL;
+	}
+
 	kfree(last_request);
 
 	platform_device_unregister(reg_pdev);
diff --git a/net/wireless/reg.h b/net/wireless/reg.h
index c9b6b63..b10d88c 100644
--- a/net/wireless/reg.h
+++ b/net/wireless/reg.h
@@ -15,6 +15,8 @@ int set_regdom(const struct ieee80211_regdomain *rd);
  *	is required to be set to the wiphy that received the information
  * @alpha2: the ISO/IEC 3166 alpha2 being claimed the regulatory domain
  *	should be in.
+ * @country_ie_checksum: checksum of processed country IE, set this to 0
+ * 	if the hint did not come from a country IE
  *
  * The Wireless subsystem can use this function to hint to the wireless core
  * what it believes should be the current regulatory domain by
@@ -26,6 +28,6 @@ int set_regdom(const struct ieee80211_regdomain *rd);
  *
  */
 extern int __regulatory_hint(struct wiphy *wiphy, enum reg_set_by set_by,
-			     const char *alpha2);
+			     const char *alpha2, u32 country_ie_checksum);
 
 #endif  /* __NET_WIRELESS_REG_H */
-- 
1.5.6.3

Re: [PATCH 1/7] cfg80211: mark negative frequencies as invalid

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 952 bytes --]

On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
> Regulatory rules with negative frequencies are now
> marked as invalid in is_valid_reg_rule().
> 
> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>

Acked-by: Johannes Berg <johannes@sipsolutions.net>

> ---
>  net/wireless/reg.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/net/wireless/reg.c b/net/wireless/reg.c
> index 4c7e39d..7a2ce27 100644
> --- a/net/wireless/reg.c
> +++ b/net/wireless/reg.c
> @@ -330,7 +330,7 @@ static bool is_valid_reg_rule(const struct ieee80211_reg_rule *rule)
>  	const struct ieee80211_freq_range *freq_range = &rule->freq_range;
>  	u32 freq_diff;
>  
> -	if (freq_range->start_freq_khz == 0 || freq_range->end_freq_khz == 0)
> +	if (freq_range->start_freq_khz <= 0 || freq_range->end_freq_khz <= 0)
>  		return false;
>  
>  	if (freq_range->start_freq_khz > freq_range->end_freq_khz)

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 2/7] cfg80211: remove switch from __regulatory_hint()

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 2166 bytes --]

On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
> We have complete control over REGDOM_SET_BY_* enum passed
> down to __regulatory_hint() as such there is no need to
> account for unexpected REGDOM_SET_BY_*'s, lets just remove
> the switch statement as this code does not change and
> won't change even when we add 802.11d support.
> 
> This patch has no functional changes.
> 
> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>

Acked-by: Johannes Berg <johannes@sipsolutions.net>

> ---
>  net/wireless/reg.c |   43 +++++++++++++++++--------------------------
>  1 files changed, 17 insertions(+), 26 deletions(-)
> 
> diff --git a/net/wireless/reg.c b/net/wireless/reg.c
> index 7a2ce27..e44f005 100644
> --- a/net/wireless/reg.c
> +++ b/net/wireless/reg.c
> @@ -698,34 +698,25 @@ int __regulatory_hint(struct wiphy *wiphy, enum reg_set_by set_by,
>  	else if (r)
>  		return r;
>  
> -	switch (set_by) {
> -	case REGDOM_SET_BY_CORE:
> -	case REGDOM_SET_BY_COUNTRY_IE:
> -	case REGDOM_SET_BY_DRIVER:
> -	case REGDOM_SET_BY_USER:
> -		request = kzalloc(sizeof(struct regulatory_request),
> -				  GFP_KERNEL);
> -		if (!request)
> -			return -ENOMEM;
> -
> -		request->alpha2[0] = alpha2[0];
> -		request->alpha2[1] = alpha2[1];
> -		request->initiator = set_by;
> -		request->wiphy = wiphy;
> -		request->intersect = intersect;
> -
> -		kfree(last_request);
> -		last_request = request;
> -		r = call_crda(alpha2);
> +	request = kzalloc(sizeof(struct regulatory_request),
> +			  GFP_KERNEL);
> +	if (!request)
> +		return -ENOMEM;
> +
> +	request->alpha2[0] = alpha2[0];
> +	request->alpha2[1] = alpha2[1];
> +	request->initiator = set_by;
> +	request->wiphy = wiphy;
> +	request->intersect = intersect;
> +
> +	kfree(last_request);
> +	last_request = request;
> +	r = call_crda(alpha2);
> +
>  #ifndef CONFIG_WIRELESS_OLD_REGULATORY
> -		if (r)
> -			printk(KERN_ERR "cfg80211: Failed calling CRDA\n");
> +	if (r)
> +		printk(KERN_ERR "cfg80211: Failed calling CRDA\n");
>  #endif
> -		break;
> -	default:
> -		r = -ENOTSUPP;
> -		break;
> -	}
>  
>  	return r;
>  }

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 3/7] cfg80211: remove switch from __set_regdom()

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 1395 bytes --]

On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
> We have control over the REGDOM_SET_BY_* macros passed
> so remove the switch.
> 
> This patch has no functional changes.
> 
> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>

Acked-by: Johannes Berg <johannes@sipsolutions.net>

> ---
>  net/wireless/reg.c |   21 ++++++---------------
>  1 files changed, 6 insertions(+), 15 deletions(-)
> 
> diff --git a/net/wireless/reg.c b/net/wireless/reg.c
> index e44f005..290ecec 100644
> --- a/net/wireless/reg.c
> +++ b/net/wireless/reg.c
> @@ -829,21 +829,12 @@ static int __set_regdom(const struct ieee80211_regdomain *rd)
>  	reset_regdomains();
>  
>  	/* Country IE parsing coming soon */
> -	switch (last_request->initiator) {
> -	case REGDOM_SET_BY_CORE:
> -	case REGDOM_SET_BY_DRIVER:
> -	case REGDOM_SET_BY_USER:
> -		if (!is_valid_rd(rd)) {
> -			printk(KERN_ERR "cfg80211: Invalid "
> -				"regulatory domain detected:\n");
> -			print_regdomain_info(rd);
> -			return -EINVAL;
> -		}
> -		break;
> -	case REGDOM_SET_BY_COUNTRY_IE: /* Not yet */
> -		WARN_ON(1);
> -	default:
> -		return -EOPNOTSUPP;
> +
> +	if (!is_valid_rd(rd)) {
> +		printk(KERN_ERR "cfg80211: Invalid "
> +			"regulatory domain detected:\n");
> +		print_regdomain_info(rd);
> +		return -EINVAL;
>  	}
>  
>  	if (unlikely(last_request->intersect)) {

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 4/7] cfg80211: separate intersection section in __set_regdom()

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 2016 bytes --]

On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
> So far the __set_regdom() code is pretty generic as the
> intersection case is fairly straight forward; this will however
> change when 802.11d support is added so lets separate intersection
> code for now in preparation for 802.11d support.
> 
> This patch only has slight functional changes.
> 
> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>

Acked-by: Johannes Berg <johannes@sipsolutions.net>

> ---
>  net/wireless/reg.c |   29 +++++++++++++++++++++--------
>  1 files changed, 21 insertions(+), 8 deletions(-)
> 
> diff --git a/net/wireless/reg.c b/net/wireless/reg.c
> index 290ecec..845e2d3 100644
> --- a/net/wireless/reg.c
> +++ b/net/wireless/reg.c
> @@ -826,10 +826,6 @@ static int __set_regdom(const struct ieee80211_regdomain *rd)
>  	if (WARN_ON(!reg_is_valid_request(rd->alpha2)))
>  		return -EINVAL;
>  
> -	reset_regdomains();
> -
> -	/* Country IE parsing coming soon */
> -
>  	if (!is_valid_rd(rd)) {
>  		printk(KERN_ERR "cfg80211: Invalid "
>  			"regulatory domain detected:\n");
> @@ -837,16 +833,33 @@ static int __set_regdom(const struct ieee80211_regdomain *rd)
>  		return -EINVAL;
>  	}
>  
> -	if (unlikely(last_request->intersect)) {
> +	if (!last_request->intersect) {
> +		reset_regdomains();
> +		cfg80211_regdomain = rd;
> +		return 0;
> +	}
> +
> +	/* Intersection requires a bit more work */
> +
> +	if (last_request->initiator != REGDOM_SET_BY_COUNTRY_IE) {
> +
>  		intersected_rd = regdom_intersect(rd, cfg80211_regdomain);
>  		if (!intersected_rd)
>  			return -EINVAL;
> +
> +		/* We can trash what CRDA provided now */
>  		kfree(rd);
> -		rd = intersected_rd;
> +		rd = NULL;
> +
> +		reset_regdomains();
> +		cfg80211_regdomain = intersected_rd;
> +
> +		return 0;
>  	}
>  
> -	/* Tada! */
> -	cfg80211_regdomain = rd;
> +	/* Country IE parsing coming soon */
> +	reset_regdomains();
> +	WARN_ON(1);
>  
>  	return 0;
>  }

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 5/7] cfg80211: expect different rd in cfg80211 when intersecting

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 1087 bytes --]

On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
> When intersecting it is possible that set_regdom() was called
> with a regulatory domain which we'll only use as an aid to
> build a final regulatory domain.
> 
> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
> ---
>  net/wireless/reg.c |    5 +++--
>  1 files changed, 3 insertions(+), 2 deletions(-)
> 
> diff --git a/net/wireless/reg.c b/net/wireless/reg.c
> index 845e2d3..f062b8b 100644
> --- a/net/wireless/reg.c
> +++ b/net/wireless/reg.c
> @@ -880,12 +880,13 @@ int set_regdom(const struct ieee80211_regdomain *rd)
>  	}
>  
>  	/* This would make this whole thing pointless */
> -	BUG_ON(rd != cfg80211_regdomain);
> +	if (!last_request->intersect)
> +		BUG_ON(rd != cfg80211_regdomain);
>  
>  	/* update all wiphys now with the new established regulatory domain */
>  	update_all_wiphy_regulatory(last_request->initiator);
>  
> -	print_regdomain(rd);
> +	print_regdomain(cfg80211_regdomain);

Maybe the printing here should be removed? It seems confusing now.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 6/7] cfg80211: initialize last_request

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 885 bytes --]

On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
> last_request was not being initialized. By initializing it
> we don't fix a bug as we are not checking for it before prior
> to a regulatory hint however we may do so at a later time.

It's a static variable, it always starts out as NULL, this isn't
necessary.

> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
> ---
>  net/wireless/reg.c |    2 ++
>  1 files changed, 2 insertions(+), 0 deletions(-)
> 
> diff --git a/net/wireless/reg.c b/net/wireless/reg.c
> index f062b8b..3754d0d 100644
> --- a/net/wireless/reg.c
> +++ b/net/wireless/reg.c
> @@ -899,6 +899,8 @@ int regulatory_init(void)
>  	if (IS_ERR(reg_pdev))
>  		return PTR_ERR(reg_pdev);
>  
> +	last_request = NULL;
> +
>  #ifdef CONFIG_WIRELESS_OLD_REGULATORY
>  	cfg80211_regdomain = static_regdom(ieee80211_regdom);
>  

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 12974 bytes --]

On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:


> 
> +/*
> + * IEEE 802.11-2007 7.3.2.9 Country information element
> + *
> + * Minimum length is 8 octects, ie len must be evenly
> + * divisible by 2
> + */

That seems odd, I thought there are triplets in there? That mean it
should be 2+N*3, no?

> +/*
> + * Regulatory extension stuff is for 802.11j, ammeneded to
> + * IEEE 802.11-2007, see Annex I (page 1141) and Annex J
> + * (page 1147). Also review 7.3.2.9.

I think you should drop the 802.11j mention, it's just confusing when
that document no longer exists and you also mention annex J.

> + * When dot11RegulatoryClassesRequired is true and the
> + * first_channel/reg_extension_id is >= 201 then the IE
> + * compromises of the 'ext' struct represented below:
> + *
> + *  - Regulatory extension ID - when generating IE this just needs
> + *    to be monotonically increasing for each triplet passed in
> + *    the IE
> + *  - Regulatory class - index into set of rules
> + *  - Coverage class - index into air propagation time (Table 7-27),
> + *    in microseconds, you can compute the air propagation time from
> + *    the index by multiplying by 3, so index 10 yields a propagation
> + *    of 10 us. Valid values are 0-31, values 32-255 are not defined
> + *    yet. A value of 0 inicates air propagation of <= 1 us.

indicates

> + *
> + *  See also Table I.2 for Emission limit sets and table
> + *  I.3 for Behavior limit sets. Table J.1 indicates how to map
> + *  a reg_class to an emission limit set and behavior limit set.

heh. my spell checker doesn't like behavior (rather than behaviour) but
I guess that's because it's set to proper English ;)

> + */
> +#define IEEE80211_COUNTRY_EXTENSION_ID 201

Should that be something like EXTENSION_ID_OFFSET? Clearly they must
mean to subtract this?

> +/*
> + *  Channels numbers in the IE must be monntonically increasing

monotonically

> @@ -181,6 +181,13 @@ struct ieee80211_supported_band {
>   * struct wiphy - wireless hardware description
>   * @idx: the wiphy index assigned to this item
>   * @class_dev: the class device representing /sys/class/ieee80211/<wiphy-name>
> + * @country_ie_alpha2: ISO / IEC 3166 alpha2 for which this wiphy is receiving
> + * 	country IEs on, this should help drivers and the wireless
> + * 	stack disregard country IEs from APs quickly on the same alpha2.
> + * 	The alpha2 may differ from cfg80211_regdomain's alpha2 when
> + * 	an intersection has occurred. If the AP is reconfigured this
> + * 	can also be used to tell us if the country on the country
> + * 	IE changed.


>   * @reg_notifier: the driver's regulatory notification callback
>   */
>  struct wiphy {
> @@ -201,6 +208,8 @@ struct wiphy {
>  
>  	struct ieee80211_supported_band *bands[IEEE80211_NUM_BANDS];
>  
> +	char country_ie_alpha2[2];

Is that needed in the external interface? It seems it should be part of
cfg80211_registered_device?

> +
> +/**
> + * regulatory_hint_ie - wireless stack hints a regulatory domain
> + * @wiphy: the wireless device giving the hint (used only for reporting
> + *	conflicts)
> + * @rd: a regulatory domain built from the received country IE
> + * @country_ie_checksum: checksum of country IE of fields we are interested in,
> + * 	we can remove this if its determined the IE never should change
> + * 	for an AP, keep in mind dot11RegulatoryClassesRequired.
> + *
> + * We will intersect the rd with the what CRDA tells us should apply
> + * for the alpha2 this country IE belongs to, this prevents APs from
> + * sending us incorrect or outdated information against a country.
> + * If CONFIG_WIRELESS_OLD_REGULATORY is enabled and CRDA is not present
> + * we will intersect immediately against what is currently set in the
> + * wireless core.
> + */
> +extern int regulatory_hint_ie(struct wiphy *wiphy,
> +	struct ieee80211_regdomain *rd,
> +	u32 country_ie_checksum);

Ok this I don't understand. Why the checksum thing? Why pass the
regdomain struct in? Can't cfg80211 do the work of parsing the IE and
you just parse the IE in, possibly "debouncing" when it hasn't changed
(or do the debouncing in cfg80211 too, for the overhead of an extra
function call)

> +/**
> + * cfg80211_is_country_change - tells us if wiphy moves to new country
> + * @wiphy: the wireless device on which the IE was received on
> + * @alpha2: the alpha2 for the country IE the wiphy received
> + *
> + * This can be used to determine if a country IE a wiphy receives
> + * has changed the alpha2. The structure of the IEs generated on
> + * the AP should not change. If the the country changes we should
> + * disassociate and re-associate as the AP was probably reconfigured
> + * and rebooted, hopefully with a correct country setting.
> + *
> + * We don't support two wiphys being present in two different
> + * countries, however it might be possible a different wiphy
> + * already set the current regulatory domain; in that case we
> + * want to process this wiphy's country IE in case it may be
> + * different even if on the same alpha2.
> + */
> +extern bool cfg80211_is_country_change(struct wiphy *wiphy,
> +	const char *alpha2);

How would you call this?

> +/**
> + * country_ie_integrity_changes - tells us if the country IE has changed
> + * @checksum: checksum of country IE of fields we are interested in
> + *
> + * If the country IE has not changed you can ignore it safely. This is
> + * useful to determine if two wiphys are seeing two different country IEs
> + * even on the same alpha2. Note that this will return false if no IE has
> + * been set on the wireless core yet.
> + */
> +extern bool country_ie_integrity_changes(u32 checksum);

Heh. Ok if you're going to have this function here, let cfg80211 do the
debouncing completely.

>  #endif /* __NET_WIRELESS_H */
> diff --git a/net/mac80211/Kconfig b/net/mac80211/Kconfig
> index 7f710a2..219fb7b 100644
> --- a/net/mac80211/Kconfig
> +++ b/net/mac80211/Kconfig
> @@ -203,6 +203,14 @@ config MAC80211_DEBUG_COUNTERS
>  
>  	  If unsure, say N.
>  
> +config MAC80211_COUNTRY_IE_DEBUG
> +	bool "Country IE parsing debug (IEEE 802.11d) debugging"
> +	depends on MAC80211_DEBUG_MENU
> +	---help---
> +	  Say Y here to print out verbose country IE parsing (IEEE 802.11d)
> +	  debug messages. This generates a lot of messages. You should only
> +	  enable this option if no regulatory changes are being made.

No, you should never enable an option that might print messages if
somebody sends you frames, even beacons can trivially be faked. This may
lull people into a false sense of security ("who'd change the regulatory
settings??")

> --- a/net/mac80211/mlme.c
> +++ b/net/mac80211/mlme.c
> @@ -648,6 +648,211 @@ static void ieee80211_sta_send_apinfo(struct ieee80211_sub_if_data *sdata,
>  	wireless_send_event(sdata->dev, SIOCGIWAP, &wrqu, NULL);
>  }
>  
> +/* Converts a country IE to a regulatory domain. A regulatory domain
> + * structure has a lot of information which the IE doesn't yet have,
> + * so for the other values we use upper max values as we will intersect
> + * with our userspace regulatory agent to get lower bounds. */
> +static struct ieee80211_regdomain *ieee80211_country_ie_2_rd(
> +				u8 *country_ie,
> +				u8 country_ie_len,
> +				u32 *checksum)

I think this is misplaced. You can live with a single new exported
function that takes the country IE and parses it in cfg80211, put this
function into cfg80211, and the checksumming and everything else too.


> +	if (elems.country_elem) {
> +		/* Note we are only reviewing this on management beacons */

"management beacon" sounds weird. beacons are management frames :)

> +config CFG80211_REG_DEBUG
> +        bool "cfg80211 regulatory debugging"
> +	depends on CFG80211

there seems to be a space indentation there

> @@ -40,6 +49,8 @@ config WIRELESS_OLD_REGULATORY
>  	  ieee80211_regdom module parameter. This is being phased out and you
>  	  should stop using them ASAP.
>  
> +	  Note: You will need CRDA if you want 802.11d support
> +

Why is this? Not that I really mind, but I haven't seen an explanation
yet.

> + * @country_ie_checksum: checksum of the last processed and accepted
> + * 	country IE
>   */
>  struct regulatory_request {
>  	struct wiphy *wiphy;
>  	enum reg_set_by initiator;
>  	char alpha2[2];
>  	bool intersect;
> +	u32 country_ie_checksum;

I definitely don't like this, it seems out of place, but it'll go away
once you push the parsing functionality up from mac80211 to cfg80211, I
think.

> +/* We use this as a place for the rd structure built from the
> + * last parsed country IE to rest until CRDA gets back to us with
> + * what it thinks should apply for the same country */
> +static const struct ieee80211_regdomain *country_ie_regdomain;
> +

How are you handling this variable when two beacons come in while crda
is being called? Locking rules?

> @@ -660,16 +680,14 @@ static int ignore_request(struct wiphy *wiphy, enum reg_set_by set_by,
>  					return -EOPNOTSUPP;
>  				return -EALREADY;
>  			}
> -			/* Two consecutive Country IE hints on the same wiphy */
> -			if (!alpha2_equal(cfg80211_regdomain->alpha2, alpha2))
> +			/* Two consecutive Country IE hints on the same wiphy.
> +			 * This should be picked up early by the driver/stack */
> +			if (WARN_ON(!alpha2_equal(cfg80211_regdomain->alpha2,
> +					alpha2)))
>  				return 0;

I don't like warning here, but I suppose pushing the stuff into cfg80211
assures this won't ever happen.

>  	case REGDOM_SET_BY_USER:
>  		if (last_request->initiator == REGDOM_SET_BY_COUNTRY_IE)
>  			return REG_INTERSECT;
> +		/* If the user knows better the user should set the regdom
> +		 * to their country before the IE is picked up */
> +		if (last_request->initiator == REGDOM_SET_BY_USER &&
> +				last_request->intersect)
> +			return -EOPNOTSUPP;

Can you use a more customary indentation of the second if line to just
after the parenthesis?

> @@ -685,7 +708,8 @@ static int ignore_request(struct wiphy *wiphy, enum reg_set_by set_by,
>  
>  /* Caller must hold &cfg80211_drv_mutex */
>  int __regulatory_hint(struct wiphy *wiphy, enum reg_set_by set_by,
> -		      const char *alpha2)
> +			const char *alpha2,
> +			u32 country_ie_checksum)

Same there, why this obsession with tabs :)
Besides that, I don't like the addition of the checksum, it's weird,
calculated by mac80211 and then used up here... But that'll go away when
you push the functionality into cfg80211 too.

When I suggested to use a checksum I was thinking of an easy way to
debounce. But a checksum is never perfect. Hence, the WARN_ON you added
might trigger despite the debounce checksum. Or we might fail to follow
changes. Maybe we should just keep a copy of the IE. When you change the
code to be parsing in cfg80211, please keep in mind that all this
doesn't need to be part of the wiphy struct, public API users don't need
it.

>  	kfree(last_request);
> -	last_request = request;
> -	r = call_crda(alpha2);
>  
> -#ifndef CONFIG_WIRELESS_OLD_REGULATORY
> -	if (r)
> -		printk(KERN_ERR "cfg80211: Failed calling CRDA\n");
> -#endif
> +	last_request = request;
> +	call_crda(alpha2);
> +
> +	/* Note: When CONFIG_WIRELESS_OLD_REGULATORY is enabled
> +	 * AND if CRDA is NOT present nothing will happen, if someone
> +	 * wants to bother with 11d with OLD_REG you can add a timer.
> +	 * If after x amount of time nothing happens you can call:
> +	 *
> +	 * return set_regdom(country_ie_regdomain);
> +	 *
> +	 * to intersect with the static rd
> +	 */

Ah, I see. I suppose it'll still work as before so that's fine.

> +	mutex_lock(&cfg80211_drv_mutex);
> +
> +	/* Pending country IE processing, this can happen after we
> +	 * call CRDA and wait for a response if a beacon was received before
> +	 * we were able to process the last regulatory_hint_ie() call */
> +	if (country_ie_regdomain)
> +		goto out;

Oh ok, that answers my other question. Might still want to indicate
locking rules.

> +static inline void reg_country_ie_process_debug(
> +	const struct ieee80211_regdomain *rd,
> +	const struct ieee80211_regdomain *country_ie_regdomain,
> +	const struct ieee80211_regdomain *intersected_rd)
> +{
> +	return;
> +}

No need for return, heh.

> +	if (rd != country_ie_regdomain) {
> +		/* Intersect what CRDA returned and our what we
> +		 * had built from the Country IE received */

our what we?

>  	last_request = NULL;
> +	country_ie_regdomain = NULL;

no need
 
> +	if (unlikely(country_ie_regdomain)) {
> +		kfree(country_ie_regdomain);
> +		country_ie_regdomain = NULL;
> +	}

no need, kfree(null) is fine, and setting to NULL useless

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Luis R. Rodriguez]

On Wed, Nov 5, 2008 at 4:23 AM, Johannes Berg <johannes@sipsolutions.net> wrote:
> On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
>
>
>>
>> +/*
>> + * IEEE 802.11-2007 7.3.2.9 Country information element
>> + *
>> + * Minimum length is 8 octects, ie len must be evenly
>> + * divisible by 2
>> + */
>
> That seems odd, I thought there are triplets in there? That mean it
> should be 2+N*3, no?

Forgot the environment 'I', 'O' or ' '.

alpha+environment (3) + n * 3

>> +/*
>> + * Regulatory extension stuff is for 802.11j, ammeneded to
>> + * IEEE 802.11-2007, see Annex I (page 1141) and Annex J
>> + * (page 1147). Also review 7.3.2.9.
>
> I think you should drop the 802.11j mention, it's just confusing when
> that document no longer exists and you also mention annex J.

Sure.

>> + */
>> +#define IEEE80211_COUNTRY_EXTENSION_ID 201
>
> Should that be something like EXTENSION_ID_OFFSET? Clearly they must
> mean to subtract this?

Sure but just to determine if it is a new one as they seem to come in
increasing.



>>  struct wiphy {
>> @@ -201,6 +208,8 @@ struct wiphy {
>>
>>       struct ieee80211_supported_band *bands[IEEE80211_NUM_BANDS];
>>
>> +     char country_ie_alpha2[2];
>
> Is that needed in the external interface? It seems it should be part of
> cfg80211_registered_device?

That seems like a better home for it.

>> +
>> +/**
>> + * regulatory_hint_ie - wireless stack hints a regulatory domain
>> + * @wiphy: the wireless device giving the hint (used only for reporting
>> + *   conflicts)
>> + * @rd: a regulatory domain built from the received country IE
>> + * @country_ie_checksum: checksum of country IE of fields we are interested in,
>> + *   we can remove this if its determined the IE never should change
>> + *   for an AP, keep in mind dot11RegulatoryClassesRequired.
>> + *
>> + * We will intersect the rd with the what CRDA tells us should apply
>> + * for the alpha2 this country IE belongs to, this prevents APs from
>> + * sending us incorrect or outdated information against a country.
>> + * If CONFIG_WIRELESS_OLD_REGULATORY is enabled and CRDA is not present
>> + * we will intersect immediately against what is currently set in the
>> + * wireless core.
>> + */
>> +extern int regulatory_hint_ie(struct wiphy *wiphy,
>> +     struct ieee80211_regdomain *rd,
>> +     u32 country_ie_checksum);
>
> Ok this I don't understand. Why the checksum thing? Why pass the
> regdomain struct in? Can't cfg80211 do the work of parsing the IE and
> you just parse the IE in, possibly "debouncing" when it hasn't changed
> (or do the debouncing in cfg80211 too, for the overhead of an extra
> function call)

It seems wrong to offload onto cfg80211 some IE parsing work even if
its related to regulatory. Are you sure you want that?

So we want two things:

1. A very quick and efficient way to determine if we can ignore the IE
processing in case the IE has already been processed
2.  A quick way to determine if a new device can ignore IE processing

The alpha2 on the registered interface solves the first
The checksum solves the second

If we leave the checksumming outside we need the helpers to check the
checksum exported. Otherwise then definitely we can stuff it into
cfg80211 but that means for a second device you'll pass the IE pointer
and len as part of the regulatory_hint_ie() and it *can* fail for
several reasons (-EALREADY, -EINVAL, -ENOMEM).

Come to think of it maybe we can remove the checksum completely if we
just check the BSSID the STA is associated to. That is a second device
can assume the IE will be the same if the alpha2 matches *and* the
BSSID is the same as the one who found it. This of course assumes the
IE will not change if the alpha2 remains the same which I think is
reasonable unless the AP's 11d settings are reconfigured and it
doesn't let off STAs after doing it.

Another thing we should consider here is the case of an Indoor AP and
an Outdoor AP. According to Felix this is very possible and he's dealt
with it. I doubt 11d is being used in those cases but regardless it
seems we should account for it and deal with it. In that case it seems
two different registered devices can be using one country but
different environments. I have kept this in mind but admit not
thoroughly. I think its good to see now what changes we'll need. Maybe
we can deal with this later on though.

> No, you should never enable an option that might print messages if
> somebody sends you frames, even beacons can trivially be faked. This may
> lull people into a false sense of security ("who'd change the regulatory
> settings??")

Fine by me. Also I think its easier to just add a damn debugfs counter
for beacons if we don't have one (I think we don't) and maybe a a 3
octect string which indicates the alph2 and the environment. If
nothing IE is set in the IE we can leave it empty.


>>       case REGDOM_SET_BY_USER:
>>               if (last_request->initiator == REGDOM_SET_BY_COUNTRY_IE)
>>                       return REG_INTERSECT;
>> +             /* If the user knows better the user should set the regdom
>> +              * to their country before the IE is picked up */
>> +             if (last_request->initiator == REGDOM_SET_BY_USER &&
>> +                             last_request->intersect)
>> +                     return -EOPNOTSUPP;
>
> Can you use a more customary indentation of the second if line to just
> after the parenthesis?

Like what

if (foo ||
<TAB><SPACE><SPACE>bar)
<TAB><TAB>bar;

?

> When I suggested to use a checksum I was thinking of an easy way to
> debounce. But a checksum is never perfect. Hence, the WARN_ON you added
> might trigger despite the debounce checksum. Or we might fail to follow
> changes.

Let me know what you think about the alpha2 + BSSID thing above. I
guess it should also consider the environment (I, O, ' ') in case the
AP is reconfigured to be outdoor (without being moved). I don't think
we should care about supporting these cases on the fly but we should
ensure the change won't break things.

> Maybe we should just keep a copy of the IE.

Ehh.

> When you change the
> code to be parsing in cfg80211, please keep in mind that all this
> doesn't need to be part of the wiphy struct, public API users don't need
> it.

Sure.

>>       kfree(last_request);
>> -     last_request = request;
>> -     r = call_crda(alpha2);
>>
>> -#ifndef CONFIG_WIRELESS_OLD_REGULATORY
>> -     if (r)
>> -             printk(KERN_ERR "cfg80211: Failed calling CRDA\n");
>> -#endif
>> +     last_request = request;
>> +     call_crda(alpha2);
>> +
>> +     /* Note: When CONFIG_WIRELESS_OLD_REGULATORY is enabled
>> +      * AND if CRDA is NOT present nothing will happen, if someone
>> +      * wants to bother with 11d with OLD_REG you can add a timer.
>> +      * If after x amount of time nothing happens you can call:
>> +      *
>> +      * return set_regdom(country_ie_regdomain);
>> +      *
>> +      * to intersect with the static rd
>> +      */
>
> Ah, I see. I suppose it'll still work as before so that's fine.

Heh yeah.. I should have read kobject_uevent_env() more carefully, it
only fails if it cannot build the uevent thing, it never returns what
the uvent result is from userspace, which makes sense. I should send
this as a separate patch too to clarify this.

  Luis

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 4551 bytes --]

On Wed, 2008-11-05 at 08:57 -0800, Luis R. Rodriguez wrote:

> >> +/*
> >> + * IEEE 802.11-2007 7.3.2.9 Country information element
> >> + *
> >> + * Minimum length is 8 octects, ie len must be evenly
> >> + * divisible by 2
> >> + */
> >
> > That seems odd, I thought there are triplets in there? That mean it
> > should be 2+N*3, no?
> 
> Forgot the environment 'I', 'O' or ' '.
> 
> alpha+environment (3) + n * 3

But then it should be evenly divisible by three, not two, right? And 8
doesn't make sense as a minimum length either then.

> >> + */
> >> +#define IEEE80211_COUNTRY_EXTENSION_ID 201
> >
> > Should that be something like EXTENSION_ID_OFFSET? Clearly they must
> > mean to subtract this?
> 
> Sure but just to determine if it is a new one as they seem to come in
> increasing.

Oh ok, I thought it was ok to have multiple in the same IE or something.

> > Ok this I don't understand. Why the checksum thing? Why pass the
> > regdomain struct in? Can't cfg80211 do the work of parsing the IE and
> > you just parse the IE in, possibly "debouncing" when it hasn't changed
> > (or do the debouncing in cfg80211 too, for the overhead of an extra
> > function call)
> 
> It seems wrong to offload onto cfg80211 some IE parsing work even if
> its related to regulatory. Are you sure you want that?

Why not? My scanning patch for cfg80211 also offloaded all of the
parsing to it.

> So we want two things:
> 
> 1. A very quick and efficient way to determine if we can ignore the IE
> processing in case the IE has already been processed
> 2.  A quick way to determine if a new device can ignore IE processing
> 
> The alpha2 on the registered interface solves the first
> The checksum solves the second

Ok, but why put anything of that outside cfg80211?

> If we leave the checksumming outside we need the helpers to check the
> checksum exported. Otherwise then definitely we can stuff it into
> cfg80211 but that means for a second device you'll pass the IE pointer
> and len as part of the regulatory_hint_ie() and it *can* fail for
> several reasons (-EALREADY, -EINVAL, -ENOMEM).

Except nobody will care about the return value, so you might as well
leave it off, no?

> Another thing we should consider here is the case of an Indoor AP and
> an Outdoor AP. According to Felix this is very possible and he's dealt
> with it. I doubt 11d is being used in those cases but regardless it
> seems we should account for it and deal with it. In that case it seems
> two different registered devices can be using one country but
> different environments. I have kept this in mind but admit not
> thoroughly. I think its good to see now what changes we'll need. Maybe
> we can deal with this later on though.

We may well need different indoor/outdoor settings at some point and a
way to tell the kernel which device is out/indoors, but that seems to
only affect TX power so I wouldn't worry about it too much for now.

> >>       case REGDOM_SET_BY_USER:
> >>               if (last_request->initiator == REGDOM_SET_BY_COUNTRY_IE)
> >>                       return REG_INTERSECT;
> >> +             /* If the user knows better the user should set the regdom
> >> +              * to their country before the IE is picked up */
> >> +             if (last_request->initiator == REGDOM_SET_BY_USER &&
> >> +                             last_request->intersect)
> >> +                     return -EOPNOTSUPP;
> >
> > Can you use a more customary indentation of the second if line to just
> > after the parenthesis?
> 
> Like what
> 
> if (foo ||
> <TAB><SPACE><SPACE>bar)
> <TAB><TAB>bar;
> 
> ?

at least that's what I do, yeah, sorry if this seems too trivial, if you
wouldn't have to respin the patch anyway I wouldn't even mention it.

> > When I suggested to use a checksum I was thinking of an easy way to
> > debounce. But a checksum is never perfect. Hence, the WARN_ON you added
> > might trigger despite the debounce checksum. Or we might fail to follow
> > changes.
> 
> Let me know what you think about the alpha2 + BSSID thing above. I
> guess it should also consider the environment (I, O, ' ') in case the
> AP is reconfigured to be outdoor (without being moved). I don't think
> we should care about supporting these cases on the fly but we should
> ensure the change won't break things.

Well I haven't checked in detail which things you care about in the IE
now, but shouldn't you check if those changed?

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Luis R. Rodriguez]

On Wed, Nov 5, 2008 at 10:10 AM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Wed, 2008-11-05 at 08:57 -0800, Luis R. Rodriguez wrote:
>
>> >> +/*
>> >> + * IEEE 802.11-2007 7.3.2.9 Country information element
>> >> + *
>> >> + * Minimum length is 8 octects, ie len must be evenly
>> >> + * divisible by 2
>> >> + */
>> >
>> > That seems odd, I thought there are triplets in there? That mean it
>> > should be 2+N*3, no?
>>
>> Forgot the environment 'I', 'O' or ' '.
>>
>> alpha+environment (3) + n * 3
>
> But then it should be evenly divisible by three, not two, right?

Nope, the end of the IE requires a pad to make it divisible by 2.

> And 8
> doesn't make sense as a minimum length either then.

Agreed :) but that's exactly why I put those comments there.

>> >> + */
>> >> +#define IEEE80211_COUNTRY_EXTENSION_ID 201
>> >
>> > Should that be something like EXTENSION_ID_OFFSET? Clearly they must
>> > mean to subtract this?
>>
>> Sure but just to determine if it is a new one as they seem to come in
>> increasing.
>
> Oh ok, I thought it was ok to have multiple in the same IE or something.

Well you can pass multiple triplets for extensions it seems but I
think this thing remains constant across them. The spec doesn't
elaborate much on this. Jouni mentioned the AP just increments this
when generating the IEs and its used, that's all.

>> > Ok this I don't understand. Why the checksum thing? Why pass the
>> > regdomain struct in? Can't cfg80211 do the work of parsing the IE and
>> > you just parse the IE in, possibly "debouncing" when it hasn't changed
>> > (or do the debouncing in cfg80211 too, for the overhead of an extra
>> > function call)
>>
>> It seems wrong to offload onto cfg80211 some IE parsing work even if
>> its related to regulatory. Are you sure you want that?
>
> Why not? My scanning patch for cfg80211 also offloaded all of the
> parsing to it.

OK fine, just wanted to double check :)

>> So we want two things:
>>
>> 1. A very quick and efficient way to determine if we can ignore the IE
>> processing in case the IE has already been processed
>> 2.  A quick way to determine if a new device can ignore IE processing
>>
>> The alpha2 on the registered interface solves the first
>> The checksum solves the second
>
> Ok, but why put anything of that outside cfg80211?

Alright, I'll respin.

>> If we leave the checksumming outside we need the helpers to check the
>> checksum exported. Otherwise then definitely we can stuff it into
>> cfg80211 but that means for a second device you'll pass the IE pointer
>> and len as part of the regulatory_hint_ie() and it *can* fail for
>> several reasons (-EALREADY, -EINVAL, -ENOMEM).
>
> Except nobody will care about the return value, so you might as well
> leave it off, no?

Sure.

>> Another thing we should consider here is the case of an Indoor AP and
>> an Outdoor AP. According to Felix this is very possible and he's dealt
>> with it. I doubt 11d is being used in those cases but regardless it
>> seems we should account for it and deal with it. In that case it seems
>> two different registered devices can be using one country but
>> different environments. I have kept this in mind but admit not
>> thoroughly. I think its good to see now what changes we'll need. Maybe
>> we can deal with this later on though.
>
> We may well need different indoor/outdoor settings at some point and a
> way to tell the kernel which device is out/indoors, but that seems to
> only affect TX power so I wouldn't worry about it too much for now.

Sure.

>> >>       case REGDOM_SET_BY_USER:
>> >>               if (last_request->initiator == REGDOM_SET_BY_COUNTRY_IE)
>> >>                       return REG_INTERSECT;
>> >> +             /* If the user knows better the user should set the regdom
>> >> +              * to their country before the IE is picked up */
>> >> +             if (last_request->initiator == REGDOM_SET_BY_USER &&
>> >> +                             last_request->intersect)
>> >> +                     return -EOPNOTSUPP;
>> >
>> > Can you use a more customary indentation of the second if line to just
>> > after the parenthesis?
>>
>> Like what
>>
>> if (foo ||
>> <TAB><SPACE><SPACE>bar)
>> <TAB><TAB>bar;
>>
>> ?
>
> at least that's what I do, yeah, sorry if this seems too trivial, if you
> wouldn't have to respin the patch anyway I wouldn't even mention it.

OK

>> > When I suggested to use a checksum I was thinking of an easy way to
>> > debounce. But a checksum is never perfect. Hence, the WARN_ON you added
>> > might trigger despite the debounce checksum. Or we might fail to follow
>> > changes.
>>
>> Let me know what you think about the alpha2 + BSSID thing above. I
>> guess it should also consider the environment (I, O, ' ') in case the
>> AP is reconfigured to be outdoor (without being moved). I don't think
>> we should care about supporting these cases on the fly but we should
>> ensure the change won't break things.
>
> Well I haven't checked in detail which things you care about in the IE
> now, but shouldn't you check if those changed?

We only care about the channel triplet and that has:

a. Channel start
b. Number of channels
c. Power

But I think that if the alpha2+environment is the same that's enough
for the STA to assume the BSSID has not changed the IE. At least with
openwrt I need to bring down the interface of the AP to change 11d.
I'd guess most home APs act the same way as in that they require a
reboot or something. On some APs I use I use iwpriv to change it but I
haven't checked to see internally what that will do while its on.

The checksum really only should be useful if a secondary device tries
to associate to an AP and it gets a country IE. It needs a way to know
if the IE is different than the one an already-present device is
using. What I'm suggesting is that if these things are the same we can
assume it doesn't change: BSSID+alpha2+environment.

  Luis

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 3241 bytes --]

On Wed, 2008-11-05 at 11:24 -0800, Luis R. Rodriguez wrote:
> On Wed, Nov 5, 2008 at 10:10 AM, Johannes Berg
> <johannes@sipsolutions.net> wrote:
> > On Wed, 2008-11-05 at 08:57 -0800, Luis R. Rodriguez wrote:
> >
> >> >> +/*
> >> >> + * IEEE 802.11-2007 7.3.2.9 Country information element
> >> >> + *
> >> >> + * Minimum length is 8 octects, ie len must be evenly
> >> >> + * divisible by 2
> >> >> + */
> >> >
> >> > That seems odd, I thought there are triplets in there? That mean it
> >> > should be 2+N*3, no?
> >>
> >> Forgot the environment 'I', 'O' or ' '.
> >>
> >> alpha+environment (3) + n * 3
> >
> > But then it should be evenly divisible by three, not two, right?
> 
> Nope, the end of the IE requires a pad to make it divisible by 2.

Ah, that's weird, I suspect nobody cares and that's why your Broadcom
router has only 6 bytes.

> > And 8
> > doesn't make sense as a minimum length either then.
> 
> Agreed :) but that's exactly why I put those comments there.

right :)

> >> > When I suggested to use a checksum I was thinking of an easy way to
> >> > debounce. But a checksum is never perfect. Hence, the WARN_ON you added
> >> > might trigger despite the debounce checksum. Or we might fail to follow
> >> > changes.
> >>
> >> Let me know what you think about the alpha2 + BSSID thing above. I
> >> guess it should also consider the environment (I, O, ' ') in case the
> >> AP is reconfigured to be outdoor (without being moved). I don't think
> >> we should care about supporting these cases on the fly but we should
> >> ensure the change won't break things.
> >
> > Well I haven't checked in detail which things you care about in the IE
> > now, but shouldn't you check if those changed?
> 
> We only care about the channel triplet and that has:
> 
> a. Channel start
> b. Number of channels
> c. Power
> 
> But I think that if the alpha2+environment is the same that's enough
> for the STA to assume the BSSID has not changed the IE. At least with
> openwrt I need to bring down the interface of the AP to change 11d.
> I'd guess most home APs act the same way as in that they require a
> reboot or something. On some APs I use I use iwpriv to change it but I
> haven't checked to see internally what that will do while its on.
> 
> The checksum really only should be useful if a secondary device tries
> to associate to an AP and it gets a country IE. It needs a way to know
> if the IE is different than the one an already-present device is
> using. What I'm suggesting is that if these things are the same we can
> assume it doesn't change: BSSID+alpha2+environment.

Ok, checksums are nice to avoid having to memcmp large things, but
afaict we're talking about maybe 12 bytes, right? If so, why don't we
just keep a copy of the country IE and compare it fully every time?

The way I see it, driver/mac80211 calls cfg80211's
	regulatory_11d(wiphy, ie, ie_length)
function and that sorts out whether to do anything at all, ignore it
because it already knows etc. So regardless of what we do we can just do
it all in here and not bother anything elsewhere with checksums,
debouncing etc, and will be free to change it easily.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 6/7] cfg80211: initialize last_request

[Luis R. Rodriguez]

On Wed, Nov 5, 2008 at 4:02 AM, Johannes Berg <johannes@sipsolutions.net> wrote:
> On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
>> last_request was not being initialized. By initializing it
>> we don't fix a bug as we are not checking for it before prior
>> to a regulatory hint however we may do so at a later time.
>
> It's a static variable, it always starts out as NULL, this isn't
> necessary.

You're right but this is not true in userspace, why is this different BTW?

  Luis

Re: [PATCH 5/7] cfg80211: expect different rd in cfg80211 when intersecting

[Luis R. Rodriguez]

On Wed, Nov 5, 2008 at 4:01 AM, Johannes Berg <johannes@sipsolutions.net> wrote:
> On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
>> When intersecting it is possible that set_regdom() was called
>> with a regulatory domain which we'll only use as an aid to
>> build a final regulatory domain.
>>
>> Signed-off-by: Luis R. Rodriguez <lrodriguez@atheros.com>
>> ---
>>  net/wireless/reg.c |    5 +++--
>>  1 files changed, 3 insertions(+), 2 deletions(-)
>>
>> diff --git a/net/wireless/reg.c b/net/wireless/reg.c
>> index 845e2d3..f062b8b 100644
>> --- a/net/wireless/reg.c
>> +++ b/net/wireless/reg.c
>> @@ -880,12 +880,13 @@ int set_regdom(const struct ieee80211_regdomain *rd)
>>       }
>>
>>       /* This would make this whole thing pointless */
>> -     BUG_ON(rd != cfg80211_regdomain);
>> +     if (!last_request->intersect)
>> +             BUG_ON(rd != cfg80211_regdomain);
>>
>>       /* update all wiphys now with the new established regulatory domain */
>>       update_all_wiphy_regulatory(last_request->initiator);
>>
>> -     print_regdomain(rd);
>> +     print_regdomain(cfg80211_regdomain);
>
> Maybe the printing here should be removed? It seems confusing now.

Well IMHO its correct to print it here as *this* is after its been set
and passed through all wiphys. ACK?

  Luis

Re: [PATCH 6/7] cfg80211: initialize last_request

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 722 bytes --]

On Wed, 2008-11-05 at 11:53 -0800, Luis R. Rodriguez wrote:
> On Wed, Nov 5, 2008 at 4:02 AM, Johannes Berg <johannes@sipsolutions.net> wrote:
> > On Tue, 2008-11-04 at 18:50 -0800, Luis R. Rodriguez wrote:
> >> last_request was not being initialized. By initializing it
> >> we don't fix a bug as we are not checking for it before prior
> >> to a regulatory hint however we may do so at a later time.
> >
> > It's a static variable, it always starts out as NULL, this isn't
> > necessary.
> 
> You're right but this is not true in userspace, why is this different BTW?

TBH, I'm not entirely sure. Must be something about the kernel
initialising memory all around, and userspace not doing that.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 5/7] cfg80211: expect different rd in cfg80211 when intersecting

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 832 bytes --]

On Wed, 2008-11-05 at 11:55 -0800, Luis R. Rodriguez wrote:

> >>       /* This would make this whole thing pointless */
> >> -     BUG_ON(rd != cfg80211_regdomain);
> >> +     if (!last_request->intersect)
> >> +             BUG_ON(rd != cfg80211_regdomain);
> >>
> >>       /* update all wiphys now with the new established regulatory domain */
> >>       update_all_wiphy_regulatory(last_request->initiator);
> >>
> >> -     print_regdomain(rd);
> >> +     print_regdomain(cfg80211_regdomain);
> >
> > Maybe the printing here should be removed? It seems confusing now.
> 
> Well IMHO its correct to print it here as *this* is after its been set
> and passed through all wiphys. ACK?

Ah, hmm, rd is intersected before this, right? I think I just got
confused with the code flow, yeah, seems fine.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Luis R. Rodriguez]

On Wed, Nov 5, 2008 at 11:52 AM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Wed, 2008-11-05 at 11:24 -0800, Luis R. Rodriguez wrote:
>> On Wed, Nov 5, 2008 at 10:10 AM, Johannes Berg
>> <johannes@sipsolutions.net> wrote:

>> >> > When I suggested to use a checksum I was thinking of an easy way to
>> >> > debounce. But a checksum is never perfect. Hence, the WARN_ON you added
>> >> > might trigger despite the debounce checksum. Or we might fail to follow
>> >> > changes.
>> >>
>> >> Let me know what you think about the alpha2 + BSSID thing above. I
>> >> guess it should also consider the environment (I, O, ' ') in case the
>> >> AP is reconfigured to be outdoor (without being moved). I don't think
>> >> we should care about supporting these cases on the fly but we should
>> >> ensure the change won't break things.
>> >
>> > Well I haven't checked in detail which things you care about in the IE
>> > now, but shouldn't you check if those changed?
>>
>> We only care about the channel triplet and that has:
>>
>> a. Channel start
>> b. Number of channels
>> c. Power
>>
>> But I think that if the alpha2+environment is the same that's enough
>> for the STA to assume the BSSID has not changed the IE. At least with
>> openwrt I need to bring down the interface of the AP to change 11d.
>> I'd guess most home APs act the same way as in that they require a
>> reboot or something. On some APs I use I use iwpriv to change it but I
>> haven't checked to see internally what that will do while its on.
>>
>> The checksum really only should be useful if a secondary device tries
>> to associate to an AP and it gets a country IE. It needs a way to know
>> if the IE is different than the one an already-present device is
>> using. What I'm suggesting is that if these things are the same we can
>> assume it doesn't change: BSSID+alpha2+environment.
>
> Ok, checksums are nice to avoid having to memcmp large things, but
> afaict we're talking about maybe 12 bytes, right?

Depends on the number of triplets, but since we only support 32 reg
rules this means we only support 32 triplets and therefore we only
support a max IE of size of 99 bytes (32 * 3 + 3). Realistically in JP
I expect there to be around 3, 4, or 5 or maybe even 6 triplets which
is 21 bytes.

> If so, why don't we
> just keep a copy of the country IE and compare it fully every time?

Eh. I'm inclined to leave the checksum then. We'd then have to bother
with freeing yet another thing later.

> The way I see it, driver/mac80211 calls cfg80211's
>        regulatory_11d(wiphy, ie, ie_length)
> function and that sorts out whether to do anything at all, ignore it
> because it already knows etc. So regardless of what we do we can just do
> it all in here and not bother anything elsewhere with checksums,
> debouncing etc, and will be free to change it easily.

You have sold on this.

  Luis

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Luis R. Rodriguez]

On Wed, Nov 5, 2008 at 12:10 PM, Luis R. Rodriguez
<lrodriguez@atheros.com> wrote:
> On Wed, Nov 5, 2008 at 11:52 AM, Johannes Berg
> <johannes@sipsolutions.net> wrote:
>> On Wed, 2008-11-05 at 11:24 -0800, Luis R. Rodriguez wrote:
>>> On Wed, Nov 5, 2008 at 10:10 AM, Johannes Berg
>>> <johannes@sipsolutions.net> wrote:
>
>>> >> > When I suggested to use a checksum I was thinking of an easy way to
>>> >> > debounce. But a checksum is never perfect. Hence, the WARN_ON you added
>>> >> > might trigger despite the debounce checksum. Or we might fail to follow
>>> >> > changes.
>>> >>
>>> >> Let me know what you think about the alpha2 + BSSID thing above. I
>>> >> guess it should also consider the environment (I, O, ' ') in case the
>>> >> AP is reconfigured to be outdoor (without being moved). I don't think
>>> >> we should care about supporting these cases on the fly but we should
>>> >> ensure the change won't break things.
>>> >
>>> > Well I haven't checked in detail which things you care about in the IE
>>> > now, but shouldn't you check if those changed?
>>>
>>> We only care about the channel triplet and that has:
>>>
>>> a. Channel start
>>> b. Number of channels
>>> c. Power
>>>
>>> But I think that if the alpha2+environment is the same that's enough
>>> for the STA to assume the BSSID has not changed the IE. At least with
>>> openwrt I need to bring down the interface of the AP to change 11d.
>>> I'd guess most home APs act the same way as in that they require a
>>> reboot or something. On some APs I use I use iwpriv to change it but I
>>> haven't checked to see internally what that will do while its on.
>>>
>>> The checksum really only should be useful if a secondary device tries
>>> to associate to an AP and it gets a country IE. It needs a way to know
>>> if the IE is different than the one an already-present device is
>>> using. What I'm suggesting is that if these things are the same we can
>>> assume it doesn't change: BSSID+alpha2+environment.
>>
>> Ok, checksums are nice to avoid having to memcmp large things, but
>> afaict we're talking about maybe 12 bytes, right?
>
> Depends on the number of triplets, but since we only support 32 reg
> rules this means we only support 32 triplets and therefore we only
> support a max IE of size of 99 bytes (32 * 3 + 3). Realistically in JP
> I expect there to be around 3, 4, or 5 or maybe even 6 triplets which
> is 21 bytes.
>
>> If so, why don't we
>> just keep a copy of the country IE and compare it fully every time?
>
> Eh. I'm inclined to leave the checksum then. We'd then have to bother
> with freeing yet another thing later.
>
>> The way I see it, driver/mac80211 calls cfg80211's
>>        regulatory_11d(wiphy, ie, ie_length)
>> function and that sorts out whether to do anything at all, ignore it
>> because it already knows etc. So regardless of what we do we can just do
>> it all in here and not bother anything elsewhere with checksums,
>> debouncing etc, and will be free to change it easily.
>
> You have sold on this.

I meant you have *me* sold on this. :)

  Luis

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 962 bytes --]

On Wed, 2008-11-05 at 12:10 -0800, Luis R. Rodriguez wrote:

> > Ok, checksums are nice to avoid having to memcmp large things, but
> > afaict we're talking about maybe 12 bytes, right?
> 
> Depends on the number of triplets, but since we only support 32 reg
> rules this means we only support 32 triplets and therefore we only
> support a max IE of size of 99 bytes (32 * 3 + 3). Realistically in JP
> I expect there to be around 3, 4, or 5 or maybe even 6 triplets which
> is 21 bytes.

Alright, not so bad either though.

> > If so, why don't we
> > just keep a copy of the country IE and compare it fully every time?
> 
> Eh. I'm inclined to leave the checksum then. We'd then have to bother
> with freeing yet another thing later.

Yeah, ok, I don't really care much, it just seemed the checksum
complicated the code a bit. I guess much of that was due to the somewhat
unnatural split between cfg80211 and mac80211 though.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH 5/7] cfg80211: expect different rd in cfg80211 when intersecting

[Luis R. Rodriguez]

On Wed, Nov 5, 2008 at 11:59 AM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Wed, 2008-11-05 at 11:55 -0800, Luis R. Rodriguez wrote:
>
>> >>       /* This would make this whole thing pointless */
>> >> -     BUG_ON(rd != cfg80211_regdomain);
>> >> +     if (!last_request->intersect)
>> >> +             BUG_ON(rd != cfg80211_regdomain);
>> >>
>> >>       /* update all wiphys now with the new established regulatory domain */
>> >>       update_all_wiphy_regulatory(last_request->initiator);
>> >>
>> >> -     print_regdomain(rd);
>> >> +     print_regdomain(cfg80211_regdomain);
>> >
>> > Maybe the printing here should be removed? It seems confusing now.
>>
>> Well IMHO its correct to print it here as *this* is after its been set
>> and passed through all wiphys. ACK?
>
> Ah, hmm, rd is intersected before this, right? I think I just got
> confused with the code flow, yeah, seems fine.

Yeap, OK thanks.

  Luis

Re: [PATCH 0/7] cfg80211/mac80211: Add 802.11d support

[Luis R. Rodriguez]

On Tue, Nov 4, 2008 at 6:49 PM, Luis R. Rodriguez
<lrodriguez@atheros.com> wrote:
> This patch series adds 802.11d support to mac80211 and
> cfg80211. I've tried to split the required changes as much
> as possible to make it easier for review. I've tested this
> on iwlagn with OLD_REG enabled and disable and with /sbin/crda
> missing and not. I'm sure I've missed a few corner cases, but
> I think this is a good start.

OK John here's the review, in case it helps

patch 1-5 are ACKed
patch 6 is unnecessary
I'll respin patch 7, and actually I'll split it in two to account for
the call_crda() return value thing.

  Luis

Re: [PATCH 7/7] cfg80211/mac80211: Add Country IE parsing/802.11d support

[Luis R. Rodriguez]

On Wed, Nov 5, 2008 at 12:14 PM, Johannes Berg
<johannes@sipsolutions.net> wrote:
> On Wed, 2008-11-05 at 12:10 -0800, Luis R. Rodriguez wrote:
>
>> > Ok, checksums are nice to avoid having to memcmp large things, but
>> > afaict we're talking about maybe 12 bytes, right?
>>
>> Depends on the number of triplets, but since we only support 32 reg
>> rules this means we only support 32 triplets and therefore we only
>> support a max IE of size of 99 bytes (32 * 3 + 3). Realistically in JP
>> I expect there to be around 3, 4, or 5 or maybe even 6 triplets which
>> is 21 bytes.
>
> Alright, not so bad either though.
>
>> > If so, why don't we
>> > just keep a copy of the country IE and compare it fully every time?
>>
>> Eh. I'm inclined to leave the checksum then. We'd then have to bother
>> with freeing yet another thing later.
>
> Yeah, ok, I don't really care much, it just seemed the checksum
> complicated the code a bit. I guess much of that was due to the somewhat
> unnatural split between cfg80211 and mac80211 though.

Yeah good point, let me respin and we can see how it looks then with
this merged to cfg80211.

  Luis

Re: [PATCH 0/7] cfg80211/mac80211: Add 802.11d support

[John W. Linville]

On Wed, Nov 05, 2008 at 12:16:57PM -0800, Luis R. Rodriguez wrote:
> On Tue, Nov 4, 2008 at 6:49 PM, Luis R. Rodriguez
> <lrodriguez@atheros.com> wrote:
> > This patch series adds 802.11d support to mac80211 and
> > cfg80211. I've tried to split the required changes as much
> > as possible to make it easier for review. I've tested this
> > on iwlagn with OLD_REG enabled and disable and with /sbin/crda
> > missing and not. I'm sure I've missed a few corner cases, but
> > I think this is a good start.
> 
> OK John here's the review, in case it helps
> 
> patch 1-5 are ACKed
> patch 6 is unnecessary
> I'll respin patch 7, and actually I'll split it in two to account for
> the call_crda() return value thing.

Hey, thanks for the recap!

-- 
John W. Linville		Linux should be at the core
linville@tuxdriver.com			of your literate lifestyle.