mac80211: fix key vs. sta locking problems

Up to now, key manipulation is supposed to run under RTNL to
avoid concurrent manipulations and also allow the set_key()
hardware callback to sleep. This is not feasible because STA
structs are rcu-protected and thus a lot of operations there
cannot take the RTNL. Also, key references are rcu-protected
so we cannot do things atomically.

This patch changes key locking completely:
 * key operations are now atomic
 * hardware crypto offload is enabled and disabled from
   a workqueue, due to that key freeing is also delayed
 * debugfs code is also run from a workqueue
 * keys reference STAs (and vice versa!) so during STA
   unlink the STAs key reference is removed but not the
   keys STA reference, to avoid races key todo work is
   run before STA destruction.
 * fewer STA operations now need the RTNL which was
   required due to key operations

This fixes the locking problems lockdep pointed out and also
makes things more light-weight because the rtnl isn't required
as much.

Note that the key todo lock/key mutex are global locks, this
is not required, of course, they could be per-hardware instead.

Signed-off-by: Johannes Berg <johannes@sipsolutions.net>
Signed-off-by: John W. Linville <linville@tuxdriver.com>

1 files changed, 264 insertions, 144 deletions

diff --git a/net/mac80211/key.c b/net/mac80211/key.c
index 5df9e0c..711e36e 100644
--- a/net/mac80211/key.c
+++ b/net/mac80211/key.c
@@ -2,7 +2,7 @@
  * Copyright 2002-2005, Instant802 Networks, Inc.
  * Copyright 2005-2006, Devicescape Software, Inc.
  * Copyright 2006-2007	Jiri Benc <jbenc@suse.cz>
- * Copyright 2007	Johannes Berg <johannes@sipsolutions.net>
+ * Copyright 2007-2008	Johannes Berg <johannes@sipsolutions.net>
  *
  * This program is free software; you can redistribute it and/or modify
  * it under the terms of the GNU General Public License version 2 as
@@ -33,17 +33,78 @@
  * There is currently no way of knowing this except by looking into
  * debugfs.
  *
- * All operations here are called under RTNL so no extra locking is
- * required.
+ * All key operations are protected internally so you can call them at
+ * any time.
  *
- * NOTE: This code requires that sta info *destruction* is done under
- *	 RTNL, otherwise it can try to access already freed STA structs
- *	 when a STA key is being freed.
+ * Within mac80211, key references are, just as STA structure references,
+ * protected by RCU. Note, however, that some things are unprotected,
+ * namely the key->sta dereferences within the hardware acceleration
+ * functions. This means that sta_info_destroy() must flush the key todo
+ * list.
+ *
+ * All the direct key list manipulation functions must not sleep because
+ * they can operate on STA info structs that are protected by RCU.
  */
 
 static const u8 bcast_addr[ETH_ALEN] = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
 static const u8 zero_addr[ETH_ALEN];
 
+/* key mutex: used to synchronise todo runners */
+static DEFINE_MUTEX(key_mutex);
+static DEFINE_SPINLOCK(todo_lock);
+static LIST_HEAD(todo_list);
+
+static void key_todo(struct work_struct *work)
+{
+	ieee80211_key_todo();
+}
+
+static DECLARE_WORK(todo_work, key_todo);
+
+/**
+ * add_todo - add todo item for a key
+ *
+ * @key: key to add to do item for
+ * @flag: todo flag(s)
+ */
+static void add_todo(struct ieee80211_key *key, u32 flag)
+{
+	if (!key)
+		return;
+
+	spin_lock(&todo_lock);
+	key->flags |= flag;
+	/* only add if not already added */
+	if (list_empty(&key->todo))
+		list_add(&key->todo, &todo_list);
+	schedule_work(&todo_work);
+	spin_unlock(&todo_lock);
+}
+
+/**
+ * ieee80211_key_lock - lock the mac80211 key operation lock
+ *
+ * This locks the (global) mac80211 key operation lock, all
+ * key operations must be done under this lock.
+ */
+static void ieee80211_key_lock(void)
+{
+	mutex_lock(&key_mutex);
+}
+
+/**
+ * ieee80211_key_unlock - unlock the mac80211 key operation lock
+ */
+static void ieee80211_key_unlock(void)
+{
+	mutex_unlock(&key_mutex);
+}
+
+static void assert_key_lock(void)
+{
+	WARN_ON(!mutex_is_locked(&key_mutex));
+}
+
 static const u8 *get_mac_for_key(struct ieee80211_key *key)
 {
 	const u8 *addr = bcast_addr;
@@ -70,26 +131,23 @@ static void ieee80211_key_enable_hw_accel(struct ieee80211_key *key)
 	int ret;
 	DECLARE_MAC_BUF(mac);
 
+	assert_key_lock();
+	might_sleep();
+
 	if (!key->local->ops->set_key)
 		return;
 
-	/*
-	 * This makes sure that all pending flushes have
-	 * actually completed prior to uploading new key
-	 * material to the hardware. That is necessary to
-	 * avoid races between flushing STAs and adding
-	 * new keys for them.
-	 */
-	__ieee80211_run_pending_flush(key->local);
-
 	addr = get_mac_for_key(key);
 
 	ret = key->local->ops->set_key(local_to_hw(key->local), SET_KEY,
 				       key->sdata->dev->dev_addr, addr,
 				       &key->conf);
 
-	if (!ret)
+	if (!ret) {
+		spin_lock(&todo_lock);
 		key->flags |= KEY_FLAG_UPLOADED_TO_HARDWARE;
+		spin_unlock(&todo_lock);
+	}
 
 	if (ret && ret != -ENOSPC && ret != -EOPNOTSUPP)
 		printk(KERN_ERR "mac80211-%s: failed to set key "
@@ -98,26 +156,24 @@ static void ieee80211_key_enable_hw_accel(struct ieee80211_key *key)
 		       key->conf.keyidx, print_mac(mac, addr), ret);
 }
 
-static void ieee80211_key_mark_hw_accel_off(struct ieee80211_key *key)
-{
-	if (key->flags & KEY_FLAG_UPLOADED_TO_HARDWARE) {
-		key->flags &= ~KEY_FLAG_UPLOADED_TO_HARDWARE;
-		key->flags |= KEY_FLAG_REMOVE_FROM_HARDWARE;
-	}
-}
-
 static void ieee80211_key_disable_hw_accel(struct ieee80211_key *key)
 {
 	const u8 *addr;
 	int ret;
 	DECLARE_MAC_BUF(mac);
 
+	assert_key_lock();
+	might_sleep();
+
 	if (!key || !key->local->ops->set_key)
 		return;
 
-	if (!(key->flags & KEY_FLAG_UPLOADED_TO_HARDWARE) &&
-	    !(key->flags & KEY_FLAG_REMOVE_FROM_HARDWARE))
+	spin_lock(&todo_lock);
+	if (!(key->flags & KEY_FLAG_UPLOADED_TO_HARDWARE)) {
+		spin_unlock(&todo_lock);
 		return;
+	}
+	spin_unlock(&todo_lock);
 
 	addr = get_mac_for_key(key);
 
@@ -131,8 +187,72 @@ static void ieee80211_key_disable_hw_accel(struct ieee80211_key *key)
 		       wiphy_name(key->local->hw.wiphy),
 		       key->conf.keyidx, print_mac(mac, addr), ret);
 
-	key->flags &= ~(KEY_FLAG_UPLOADED_TO_HARDWARE |
-			KEY_FLAG_REMOVE_FROM_HARDWARE);
+	spin_lock(&todo_lock);
+	key->flags &= ~KEY_FLAG_UPLOADED_TO_HARDWARE;
+	spin_unlock(&todo_lock);
+}
+
+static void __ieee80211_set_default_key(struct ieee80211_sub_if_data *sdata,
+					int idx)
+{
+	struct ieee80211_key *key = NULL;
+
+	if (idx >= 0 && idx < NUM_DEFAULT_KEYS)
+		key = sdata->keys[idx];
+
+	rcu_assign_pointer(sdata->default_key, key);
+
+	if (key)
+		add_todo(key, KEY_FLAG_TODO_DEFKEY);
+}
+
+void ieee80211_set_default_key(struct ieee80211_sub_if_data *sdata, int idx)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&sdata->local->sta_lock, flags);
+	__ieee80211_set_default_key(sdata, idx);
+	spin_unlock_irqrestore(&sdata->local->sta_lock, flags);
+}
+
+
+static void __ieee80211_key_replace(struct ieee80211_sub_if_data *sdata,
+				    struct sta_info *sta,
+				    struct ieee80211_key *old,
+				    struct ieee80211_key *new)
+{
+	int idx, defkey;
+
+	if (new)
+		list_add(&new->list, &sdata->key_list);
+
+	if (sta) {
+		rcu_assign_pointer(sta->key, new);
+	} else {
+		WARN_ON(new && old && new->conf.keyidx != old->conf.keyidx);
+
+		if (old)
+			idx = old->conf.keyidx;
+		else
+			idx = new->conf.keyidx;
+
+		defkey = old && sdata->default_key == old;
+
+		if (defkey && !new)
+			__ieee80211_set_default_key(sdata, -1);
+
+		rcu_assign_pointer(sdata->keys[idx], new);
+		if (defkey && new)
+			__ieee80211_set_default_key(sdata, new->conf.keyidx);
+	}
+
+	if (old) {
+		/*
+		 * We'll use an empty list to indicate that the key
+		 * has already been removed.
+		 */
+		list_del_init(&old->list);
+	}
 }
 
 struct ieee80211_key *ieee80211_key_alloc(enum ieee80211_key_alg alg,
@@ -160,6 +280,7 @@ struct ieee80211_key *ieee80211_key_alloc(enum ieee80211_key_alg alg,
 	key->conf.keylen = key_len;
 	memcpy(key->conf.key, key_data, key_len);
 	INIT_LIST_HEAD(&key->list);
+	INIT_LIST_HEAD(&key->todo);
 
 	if (alg == ALG_CCMP) {
 		/*
@@ -168,7 +289,7 @@ struct ieee80211_key *ieee80211_key_alloc(enum ieee80211_key_alg alg,
 		 */
 		key->u.ccmp.tfm = ieee80211_aes_key_setup_encrypt(key_data);
 		if (!key->u.ccmp.tfm) {
-			ieee80211_key_free(key);
+			kfree(key);
 			return NULL;
 		}
 	}
@@ -176,56 +297,14 @@ struct ieee80211_key *ieee80211_key_alloc(enum ieee80211_key_alg alg,
 	return key;
 }
 
-static void __ieee80211_key_replace(struct ieee80211_sub_if_data *sdata,
-				    struct sta_info *sta,
-				    struct ieee80211_key *key,
-				    struct ieee80211_key *new)
-{
-	int idx, defkey;
-
-	if (new)
-		list_add(&new->list, &sdata->key_list);
-
-	if (sta) {
-		rcu_assign_pointer(sta->key, new);
-	} else {
-		WARN_ON(new && key && new->conf.keyidx != key->conf.keyidx);
-
-		if (key)
-			idx = key->conf.keyidx;
-		else
-			idx = new->conf.keyidx;
-
-		defkey = key && sdata->default_key == key;
-
-		if (defkey && !new)
-			ieee80211_set_default_key(sdata, -1);
-
-		rcu_assign_pointer(sdata->keys[idx], new);
-		if (defkey && new)
-			ieee80211_set_default_key(sdata, new->conf.keyidx);
-	}
-
-	if (key) {
-		ieee80211_key_mark_hw_accel_off(key);
-		/*
-		 * We'll use an empty list to indicate that the key
-		 * has already been removed.
-		 */
-		list_del_init(&key->list);
-	}
-}
-
 void ieee80211_key_link(struct ieee80211_key *key,
 			struct ieee80211_sub_if_data *sdata,
 			struct sta_info *sta)
 {
 	struct ieee80211_key *old_key;
+	unsigned long flags;
 	int idx;
 
-	ASSERT_RTNL();
-	might_sleep();
-
 	BUG_ON(!sdata);
 	BUG_ON(!key);
 
@@ -234,11 +313,7 @@ void ieee80211_key_link(struct ieee80211_key *key,
 	key->sdata = sdata;
 	key->sta = sta;
 
-	ieee80211_debugfs_key_add(key->local, key);
-
 	if (sta) {
-		ieee80211_debugfs_key_sta_link(key, sta);
-
 		/*
 		 * some hardware cannot handle TKIP with QoS, so
 		 * we indicate whether QoS could be in use.
@@ -249,7 +324,10 @@ void ieee80211_key_link(struct ieee80211_key *key,
 		if (sdata->vif.type == IEEE80211_IF_TYPE_STA) {
 			struct sta_info *ap;
 
-			rcu_read_lock();
+			/*
+			 * We're getting a sta pointer in,
+			 * so must be under RCU read lock.
+			 */
 
 			/* same here, the AP could be using QoS */
 			ap = sta_info_get(key->local, key->sdata->u.sta.bssid);
@@ -258,11 +336,11 @@ void ieee80211_key_link(struct ieee80211_key *key,
 					key->conf.flags |=
 						IEEE80211_KEY_FLAG_WMM_STA;
 			}
-
-			rcu_read_unlock();
 		}
 	}
 
+	spin_lock_irqsave(&sdata->local->sta_lock, flags);
+
 	if (sta)
 		old_key = sta->key;
 	else
@@ -270,108 +348,150 @@ void ieee80211_key_link(struct ieee80211_key *key,
 
 	__ieee80211_key_replace(sdata, sta, old_key, key);
 
-	if (old_key) {
-		synchronize_rcu();
-		ieee80211_key_free(old_key);
-	}
+	spin_unlock_irqrestore(&sdata->local->sta_lock, flags);
+
+	/* free old key later */
+	add_todo(old_key, KEY_FLAG_TODO_DELETE);
 
+	add_todo(key, KEY_FLAG_TODO_ADD_DEBUGFS);
 	if (netif_running(sdata->dev))
-		ieee80211_key_enable_hw_accel(key);
+		add_todo(key, KEY_FLAG_TODO_HWACCEL);
 }
 
 void ieee80211_key_free(struct ieee80211_key *key)
 {
-	ASSERT_RTNL();
-	might_sleep();
+	unsigned long flags;
 
 	if (!key)
 		return;
 
+	/*
+	 * Replace key with nothingness if it was ever used.
+	 */
 	if (key->sdata) {
-		/*
-		 * Replace key with nothingness.
-		 *
-		 * Because other code may have key reference (RCU protected)
-		 * right now, we then wait for a grace period before freeing
-		 * it.
-		 * An empty list indicates it was never added to the key list
-		 * or has been removed already. It may, however, still be in
-		 * hardware for acceleration.
-		 */
-		if (!list_empty(&key->list))
-			__ieee80211_key_replace(key->sdata, key->sta,
-						key, NULL);
+		spin_lock_irqsave(&key->sdata->local->sta_lock, flags);
+		__ieee80211_key_replace(key->sdata, key->sta,
+					key, NULL);
+		spin_unlock_irqrestore(&key->sdata->local->sta_lock, flags);
+	}
 
-		/*
-		 * Do NOT remove this without looking at sta_info_destroy()
-		 */
-		synchronize_rcu();
+	add_todo(key, KEY_FLAG_TODO_DELETE);
+}
 
-		/*
-		 * Remove from hwaccel if appropriate, this will
-		 * only happen when the key is actually unlinked,
-		 * it will already be done when the key was replaced.
-		 */
-		ieee80211_key_disable_hw_accel(key);
-	}
+void ieee80211_enable_keys(struct ieee80211_sub_if_data *sdata)
+{
+	struct ieee80211_key *key;
 
-	if (key->conf.alg == ALG_CCMP)
-		ieee80211_aes_key_free(key->u.ccmp.tfm);
-	ieee80211_debugfs_key_remove(key);
+	might_sleep();
 
-	kfree(key);
+	if (WARN_ON(!netif_running(sdata->dev)))
+		return;
+
+	ieee80211_key_lock();
+
+	list_for_each_entry(key, &sdata->key_list, list)
+		ieee80211_key_enable_hw_accel(key);
+
+	ieee80211_key_unlock();
 }
 
-void ieee80211_set_default_key(struct ieee80211_sub_if_data *sdata, int idx)
+void ieee80211_disable_keys(struct ieee80211_sub_if_data *sdata)
 {
-	struct ieee80211_key *key = NULL;
+	struct ieee80211_key *key;
 
-	if (idx >= 0 && idx < NUM_DEFAULT_KEYS)
-		key = sdata->keys[idx];
+	might_sleep();
 
-	if (sdata->default_key != key) {
-		ieee80211_debugfs_key_remove_default(sdata);
+	ieee80211_key_lock();
 
-		rcu_assign_pointer(sdata->default_key, key);
+	list_for_each_entry(key, &sdata->key_list, list)
+		ieee80211_key_disable_hw_accel(key);
 
-		if (sdata->default_key)
-			ieee80211_debugfs_key_add_default(sdata);
-	}
+	ieee80211_key_unlock();
 }
 
-void ieee80211_free_keys(struct ieee80211_sub_if_data *sdata)
+static void __ieee80211_key_free(struct ieee80211_key *key)
 {
-	struct ieee80211_key *key, *tmp;
-	LIST_HEAD(tmp_list);
+	if (!key)
+		return;
 
-	ASSERT_RTNL();
-	might_sleep();
+	ieee80211_key_disable_hw_accel(key);
 
-	list_for_each_entry_safe(key, tmp, &sdata->key_list, list)
-		ieee80211_key_free(key);
+	if (key->conf.alg == ALG_CCMP)
+		ieee80211_aes_key_free(key->u.ccmp.tfm);
+	ieee80211_debugfs_key_remove(key);
+
+	kfree(key);
 }
 
-void ieee80211_enable_keys(struct ieee80211_sub_if_data *sdata)
+static void __ieee80211_key_todo(void)
 {
 	struct ieee80211_key *key;
+	bool work_done;
+	u32 todoflags;
 
-	ASSERT_RTNL();
-	might_sleep();
+	/*
+	 * NB: sta_info_destroy relies on this!
+	 */
+	synchronize_rcu();
+
+	spin_lock(&todo_lock);
+	while (!list_empty(&todo_list)) {
+		key = list_first_entry(&todo_list, struct ieee80211_key, todo);
+		list_del_init(&key->todo);
+		todoflags = key->flags & (KEY_FLAG_TODO_ADD_DEBUGFS |
+					  KEY_FLAG_TODO_DEFKEY |
+					  KEY_FLAG_TODO_HWACCEL |
+					  KEY_FLAG_TODO_DELETE);
+		key->flags &= ~todoflags;
+		spin_unlock(&todo_lock);
+
+		work_done = false;
+
+		if (todoflags & KEY_FLAG_TODO_ADD_DEBUGFS) {
+			ieee80211_debugfs_key_add(key);
+			work_done = true;
+		}
+		if (todoflags & KEY_FLAG_TODO_DEFKEY) {
+			ieee80211_debugfs_key_remove_default(key->sdata);
+			ieee80211_debugfs_key_add_default(key->sdata);
+			work_done = true;
+		}
+		if (todoflags & KEY_FLAG_TODO_HWACCEL) {
+			ieee80211_key_enable_hw_accel(key);
+			work_done = true;
+		}
+		if (todoflags & KEY_FLAG_TODO_DELETE) {
+			__ieee80211_key_free(key);
+			work_done = true;
+		}
 
-	if (WARN_ON(!netif_running(sdata->dev)))
-		return;
+		WARN_ON(!work_done);
 
-	list_for_each_entry(key, &sdata->key_list, list)
-		ieee80211_key_enable_hw_accel(key);
+		spin_lock(&todo_lock);
+	}
+	spin_unlock(&todo_lock);
 }
 
-void ieee80211_disable_keys(struct ieee80211_sub_if_data *sdata)
+void ieee80211_key_todo(void)
 {
-	struct ieee80211_key *key;
+	ieee80211_key_lock();
+	__ieee80211_key_todo();
+	ieee80211_key_unlock();
+}
 
-	ASSERT_RTNL();
-	might_sleep();
+void ieee80211_free_keys(struct ieee80211_sub_if_data *sdata)
+{
+	struct ieee80211_key *key, *tmp;
+	LIST_HEAD(tmp_list);
 
-	list_for_each_entry(key, &sdata->key_list, list)
-		ieee80211_key_disable_hw_accel(key);
+	ieee80211_key_lock();
+
+	ieee80211_debugfs_key_remove_default(sdata);
+
+	list_for_each_entry_safe(key, tmp, &sdata->key_list, list)
+		ieee80211_key_free(key);
+
+	__ieee80211_key_todo();
+
+	ieee80211_key_unlock();
 }