All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH 1/4] mac80211: Avoid filling up mesh preq queue with redundant requests
@ 2011-10-29  5:05 Thomas Pedersen
  2011-10-29  5:05 ` [PATCH 2/4] mac80211: QoS multicast frames have No Ack policy Thomas Pedersen
                   ` (3 more replies)
  0 siblings, 4 replies; 12+ messages in thread
From: Thomas Pedersen @ 2011-10-29  5:05 UTC (permalink / raw)
  To: linux-wireless; +Cc: devel, Javier Cardona, johannes, linville

From: Javier Cardona <javier@cozybit.com>

Don't accept redundant PREQs for a given destination. This fixes a
problem under high load:

kernel: [20386.250913] mesh_queue_preq: 235 callbacks suppressed
kernel: [20386.253335] Mesh HWMP (mesh0): PREQ node queue full
kernel: [20386.253352] Mesh HWMP (mesh0): PREQ node queue full
(...)

The 802.11s protocol has a provision to limit the rate of path requests
(PREQs) are transmitted (dot11MeshHWMPpreqMinInterval) but there was no
limit on the rate at which PREQs were being queued up.  There is a valid
reason for queuing PREQs: this way we can even out PREQ bursts.  But
queueing multiple PREQs for the same destination is useless.

Reported-by: Pedro Larbig <pedro.larbig@carhs.de>
Signed-off-by: Javier Cardona <javier@cozybit.com>
---
v2: fix merge conflict

 net/mac80211/mesh.h      |    3 +++
 net/mac80211/mesh_hwmp.c |   15 +++++++++++++--
 2 files changed, 16 insertions(+), 2 deletions(-)

diff --git a/net/mac80211/mesh.h b/net/mac80211/mesh.h
index 8c00e2d..b3745e8 100644
--- a/net/mac80211/mesh.h
+++ b/net/mac80211/mesh.h
@@ -31,6 +31,8 @@
  * @MESH_PATH_FIXED: the mesh path has been manually set and should not be
  * 	modified
  * @MESH_PATH_RESOLVED: the mesh path can has been resolved
+ * @MESH_PATH_REQ_QUEUED: there is an unsent path request for this destination
+ * already queued up, waiting for the discovery process to start.
  *
  * MESH_PATH_RESOLVED is used by the mesh path timer to
  * decide when to stop or cancel the mesh path discovery.
@@ -41,6 +43,7 @@ enum mesh_path_flags {
 	MESH_PATH_SN_VALID =	BIT(2),
 	MESH_PATH_FIXED	=	BIT(3),
 	MESH_PATH_RESOLVED =	BIT(4),
+	MESH_PATH_REQ_QUEUED =	BIT(5),
 };
 
 /**
diff --git a/net/mac80211/mesh_hwmp.c b/net/mac80211/mesh_hwmp.c
index 9a1f8bb..b7d9dfd 100644
--- a/net/mac80211/mesh_hwmp.c
+++ b/net/mac80211/mesh_hwmp.c
@@ -867,9 +867,19 @@ static void mesh_queue_preq(struct mesh_path *mpath, u8 flags)
 		return;
 	}
 
+	spin_lock_bh(&mpath->state_lock);
+	if (mpath->flags & MESH_PATH_REQ_QUEUED) {
+		spin_unlock_bh(&mpath->state_lock);
+		spin_unlock_bh(&ifmsh->mesh_preq_queue_lock);
+		return;
+	}
+
 	memcpy(preq_node->dst, mpath->dst, ETH_ALEN);
 	preq_node->flags = flags;
 
+	mpath->flags |= MESH_PATH_REQ_QUEUED;
+	spin_unlock_bh(&mpath->state_lock);
+
 	list_add_tail(&preq_node->list, &ifmsh->preq_queue.list);
 	++ifmsh->preq_queue_len;
 	spin_unlock_bh(&ifmsh->mesh_preq_queue_lock);
@@ -921,6 +931,7 @@ void mesh_path_start_discovery(struct ieee80211_sub_if_data *sdata)
 		goto enddiscovery;
 
 	spin_lock_bh(&mpath->state_lock);
+	mpath->flags &= ~MESH_PATH_REQ_QUEUED;
 	if (preq_node->flags & PREQ_Q_F_START) {
 		if (mpath->flags & MESH_PATH_RESOLVING) {
 			spin_unlock_bh(&mpath->state_lock);
@@ -1028,8 +1039,7 @@ int mesh_nexthop_lookup(struct sk_buff *skb,
 			mesh_queue_preq(mpath, PREQ_Q_F_START);
 		}
 
-		if (skb_queue_len(&mpath->frame_queue) >=
-				MESH_FRAME_QUEUE_LEN)
+		if (skb_queue_len(&mpath->frame_queue) >= MESH_FRAME_QUEUE_LEN)
 			skb_to_free = skb_dequeue(&mpath->frame_queue);
 
 		info->flags |= IEEE80211_TX_INTFL_NEED_TXPROCESSING;
@@ -1062,6 +1072,7 @@ void mesh_path_timer(unsigned long data)
 		++mpath->discovery_retries;
 		mpath->discovery_timeout *= 2;
 		spin_unlock_bh(&mpath->state_lock);
+		mpath->flags &= ~MESH_PATH_REQ_QUEUED;
 		mesh_queue_preq(mpath, 0);
 	} else {
 		mpath->flags = 0;
-- 
1.7.5.4


^ permalink raw reply related	[flat|nested] 12+ messages in thread
end of thread, other threads:[~2011-11-02 17:54 UTC | newest]

Thread overview: 12+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2011-10-29  5:05 [PATCH 1/4] mac80211: Avoid filling up mesh preq queue with redundant requests Thomas Pedersen
2011-10-29  5:05 ` [PATCH 2/4] mac80211: QoS multicast frames have No Ack policy Thomas Pedersen
2011-10-29  9:40   ` Christian Lamparter
2011-10-31  7:03     ` Thomas Pedersen
2011-10-31  9:16       ` Christian Lamparter
2011-11-02 10:17   ` Johannes Berg
2011-11-02 17:35     ` Thomas Pedersen
2011-11-02 17:54       ` Thomas Pedersen
2011-10-29  5:05 ` [PATCH 3/4] mac80211: check if frame is really part of this BA Thomas Pedersen
2011-11-02 10:19   ` Johannes Berg
2011-10-29  5:05 ` [PATCH 4/4] mac80211: Populate QoS header on mesh frames sent to gates Thomas Pedersen
2011-11-02 10:18 ` [PATCH 1/4] mac80211: Avoid filling up mesh preq queue with redundant requests Johannes Berg

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.