All of lore.kernel.org
 help / color / mirror / Atom feed
* [B.A.T.M.A.N.] bug - unregister_netdevice: usage_count = 1
@ 2011-05-11  0:17 Antonio Quartulli
  2011-05-11 12:44 ` [B.A.T.M.A.N.] [RFC PATCH] batman-adv: Add missing hardif_free_ref in forw_packet_free Sven Eckelmann
  0 siblings, 1 reply; 5+ messages in thread
From: Antonio Quartulli @ 2011-05-11  0:17 UTC (permalink / raw)
  To: B.A.T.M.A.N

Hi all,

I found a bug playing with my interfaces while bat0 was active.
The bug is described here: http://www.open-mesh.org/issues/148
The steps to reproduce it are reported too.

In a few words, when trying to unregister a device by unloading its module,
the modprobe hangs and in the dmesg I can see this message:

unregister_netdevice: waiting for wlan0 to become free. Usage count = 1

I tried to do some tests and what I observed is that
hard_iface->refcount doesn't reach 0. This means that
hard_iface_free_rcu() is never invoked and dev_put() too.

I hope the information are enough..

Regards,

-- 
Antonio Quartulli

..each of us alone is worth nothing..
Ernesto "Che" Guevara

^ permalink raw reply	[flat|nested] 5+ messages in thread

* [B.A.T.M.A.N.] [RFC PATCH] batman-adv: Add missing hardif_free_ref in forw_packet_free
  2011-05-11  0:17 [B.A.T.M.A.N.] bug - unregister_netdevice: usage_count = 1 Antonio Quartulli
@ 2011-05-11 12:44 ` Sven Eckelmann
  2011-05-11 18:59   ` [B.A.T.M.A.N.] [RFC PATCHv2] " Sven Eckelmann
  2011-05-13 13:05   ` [B.A.T.M.A.N.] [RFC PATCH] " Antonio Quartulli
  0 siblings, 2 replies; 5+ messages in thread
From: Sven Eckelmann @ 2011-05-11 12:44 UTC (permalink / raw)
  To: b.a.t.m.a.n

add_bcast_packet_to_list increases the refcount for if_incoming but the
reference count is never decreased. This is an regression introduced by
89079aec4c0c2c60a3a2561afcce62538cebd71d.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
---
This patch is not ready to be applied - just a first note. And before
someone asks - no, it doesn't fix the problem.

 aggregation.c |   14 +++++++++++---
 send.c        |    2 ++
 2 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/aggregation.c b/aggregation.c
index 9b94590..a8c3203 100644
--- a/aggregation.c
+++ b/aggregation.c
@@ -23,6 +23,7 @@
 #include "aggregation.h"
 #include "send.h"
 #include "routing.h"
+#include "hard-interface.h"
 
 /* calculate the size of the tt information for a given packet */
 static int tt_len(struct batman_packet *batman_packet)
@@ -105,12 +106,15 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
 	struct forw_packet *forw_packet_aggr;
 	unsigned char *skb_buff;
 
+	if (!atomic_inc_not_zero(&if_incoming->refcount))
+		return;
+
 	/* own packet should always be scheduled */
 	if (!own_packet) {
 		if (!atomic_dec_not_zero(&bat_priv->batman_queue_left)) {
 			bat_dbg(DBG_BATMAN, bat_priv,
 				"batman packet queue full\n");
-			return;
+			goto out;
 		}
 	}
 
@@ -118,7 +122,7 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
 	if (!forw_packet_aggr) {
 		if (!own_packet)
 			atomic_inc(&bat_priv->batman_queue_left);
-		return;
+		goto out;
 	}
 
 	if ((atomic_read(&bat_priv->aggregated_ogms)) &&
@@ -133,7 +137,7 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
 		if (!own_packet)
 			atomic_inc(&bat_priv->batman_queue_left);
 		kfree(forw_packet_aggr);
-		return;
+		goto out;
 	}
 	skb_reserve(forw_packet_aggr->skb, sizeof(struct ethhdr));
 
@@ -164,6 +168,10 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
 	queue_delayed_work(bat_event_workqueue,
 			   &forw_packet_aggr->delayed_work,
 			   send_time - jiffies);
+
+	return;
+out:
+	hardif_free_ref(if_incoming);
 }
 
 /* aggregate a new packet into the existing aggregation */
diff --git a/send.c b/send.c
index f30d0c6..6e83e23 100644
--- a/send.c
+++ b/send.c
@@ -377,6 +377,8 @@ static void forw_packet_free(struct forw_packet *forw_packet)
 {
 	if (forw_packet->skb)
 		kfree_skb(forw_packet->skb);
+	if (forw_packet->if_incoming)
+		hardif_free_ref(forw_packet->if_incoming);
 	kfree(forw_packet);
 }
 
-- 
1.7.5.1


^ permalink raw reply related	[flat|nested] 5+ messages in thread

* [B.A.T.M.A.N.] [RFC PATCHv2] batman-adv: Add missing hardif_free_ref in forw_packet_free
  2011-05-11 12:44 ` [B.A.T.M.A.N.] [RFC PATCH] batman-adv: Add missing hardif_free_ref in forw_packet_free Sven Eckelmann
@ 2011-05-11 18:59   ` Sven Eckelmann
  2011-05-14 15:26     ` Marek Lindner
  2011-05-13 13:05   ` [B.A.T.M.A.N.] [RFC PATCH] " Antonio Quartulli
  1 sibling, 1 reply; 5+ messages in thread
From: Sven Eckelmann @ 2011-05-11 18:59 UTC (permalink / raw)
  To: b.a.t.m.a.n

add_bcast_packet_to_list increases the refcount for if_incoming but the
reference count is never decreased. The reference count must be
increased for all kinds of forwarded packets which have the primary
interface stored and forw_packet_free must decrease them. Also
purge_outstanding_packets has to invoke forw_packet_free when a work
item was really cancelled.

This was a regression introduced by
89079aec4c0c2c60a3a2561afcce62538cebd71d.

Signed-off-by: Sven Eckelmann <sven@narfation.org>
---
I will leave the hard work to someone else (cleanup, fixing commit
message, ...)

 aggregation.c |   14 +++++++++++---
 send.c        |   10 +++++++++-
 2 files changed, 20 insertions(+), 4 deletions(-)

diff --git a/aggregation.c b/aggregation.c
index 9b94590..a8c3203 100644
--- a/aggregation.c
+++ b/aggregation.c
@@ -23,6 +23,7 @@
 #include "aggregation.h"
 #include "send.h"
 #include "routing.h"
+#include "hard-interface.h"
 
 /* calculate the size of the tt information for a given packet */
 static int tt_len(struct batman_packet *batman_packet)
@@ -105,12 +106,15 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
 	struct forw_packet *forw_packet_aggr;
 	unsigned char *skb_buff;
 
+	if (!atomic_inc_not_zero(&if_incoming->refcount))
+		return;
+
 	/* own packet should always be scheduled */
 	if (!own_packet) {
 		if (!atomic_dec_not_zero(&bat_priv->batman_queue_left)) {
 			bat_dbg(DBG_BATMAN, bat_priv,
 				"batman packet queue full\n");
-			return;
+			goto out;
 		}
 	}
 
@@ -118,7 +122,7 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
 	if (!forw_packet_aggr) {
 		if (!own_packet)
 			atomic_inc(&bat_priv->batman_queue_left);
-		return;
+		goto out;
 	}
 
 	if ((atomic_read(&bat_priv->aggregated_ogms)) &&
@@ -133,7 +137,7 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
 		if (!own_packet)
 			atomic_inc(&bat_priv->batman_queue_left);
 		kfree(forw_packet_aggr);
-		return;
+		goto out;
 	}
 	skb_reserve(forw_packet_aggr->skb, sizeof(struct ethhdr));
 
@@ -164,6 +168,10 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
 	queue_delayed_work(bat_event_workqueue,
 			   &forw_packet_aggr->delayed_work,
 			   send_time - jiffies);
+
+	return;
+out:
+	hardif_free_ref(if_incoming);
 }
 
 /* aggregate a new packet into the existing aggregation */
diff --git a/send.c b/send.c
index f30d0c6..76f70c7 100644
--- a/send.c
+++ b/send.c
@@ -377,6 +377,8 @@ static void forw_packet_free(struct forw_packet *forw_packet)
 {
 	if (forw_packet->skb)
 		kfree_skb(forw_packet->skb);
+	if (forw_packet->if_incoming)
+		hardif_free_ref(forw_packet->if_incoming);
 	kfree(forw_packet);
 }
 
@@ -539,6 +541,7 @@ void purge_outstanding_packets(struct bat_priv *bat_priv,
 {
 	struct forw_packet *forw_packet;
 	struct hlist_node *tmp_node, *safe_tmp_node;
+	bool pending;
 
 	if (hard_iface)
 		bat_dbg(DBG_BATMAN, bat_priv,
@@ -591,8 +594,13 @@ void purge_outstanding_packets(struct bat_priv *bat_priv,
 		 * send_outstanding_bat_packet() will lock the list to
 		 * delete the item from the list
 		 */
-		cancel_delayed_work_sync(&forw_packet->delayed_work);
+		pending = cancel_delayed_work_sync(&forw_packet->delayed_work);
 		spin_lock_bh(&bat_priv->forw_bat_list_lock);
+
+		if (pending) {
+			hlist_del(&forw_packet->list);
+			forw_packet_free(forw_packet);
+		}
 	}
 	spin_unlock_bh(&bat_priv->forw_bat_list_lock);
 }
-- 
1.7.5.1


^ permalink raw reply related	[flat|nested] 5+ messages in thread

* Re: [B.A.T.M.A.N.] [RFC PATCH] batman-adv: Add missing hardif_free_ref in forw_packet_free
  2011-05-11 12:44 ` [B.A.T.M.A.N.] [RFC PATCH] batman-adv: Add missing hardif_free_ref in forw_packet_free Sven Eckelmann
  2011-05-11 18:59   ` [B.A.T.M.A.N.] [RFC PATCHv2] " Sven Eckelmann
@ 2011-05-13 13:05   ` Antonio Quartulli
  1 sibling, 0 replies; 5+ messages in thread
From: Antonio Quartulli @ 2011-05-13 13:05 UTC (permalink / raw)
  To: The list for a Better Approach To Mobile Ad-hoc Networking

I tried the patch and it fixes the issu I reported!

Thanks Sven!

On Wed, May 11, 2011 at 02:44:37PM +0200, Sven Eckelmann wrote:
> add_bcast_packet_to_list increases the refcount for if_incoming but the
> reference count is never decreased. This is an regression introduced by
> 89079aec4c0c2c60a3a2561afcce62538cebd71d.
> 
> Signed-off-by: Sven Eckelmann <sven@narfation.org>
> ---
> This patch is not ready to be applied - just a first note. And before
> someone asks - no, it doesn't fix the problem.
> 
>  aggregation.c |   14 +++++++++++---
>  send.c        |    2 ++
>  2 files changed, 13 insertions(+), 3 deletions(-)
> 
> diff --git a/aggregation.c b/aggregation.c
> index 9b94590..a8c3203 100644
> --- a/aggregation.c
> +++ b/aggregation.c
> @@ -23,6 +23,7 @@
>  #include "aggregation.h"
>  #include "send.h"
>  #include "routing.h"
> +#include "hard-interface.h"
>  
>  /* calculate the size of the tt information for a given packet */
>  static int tt_len(struct batman_packet *batman_packet)
> @@ -105,12 +106,15 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
>  	struct forw_packet *forw_packet_aggr;
>  	unsigned char *skb_buff;
>  
> +	if (!atomic_inc_not_zero(&if_incoming->refcount))
> +		return;
> +
>  	/* own packet should always be scheduled */
>  	if (!own_packet) {
>  		if (!atomic_dec_not_zero(&bat_priv->batman_queue_left)) {
>  			bat_dbg(DBG_BATMAN, bat_priv,
>  				"batman packet queue full\n");
> -			return;
> +			goto out;
>  		}
>  	}
>  
> @@ -118,7 +122,7 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
>  	if (!forw_packet_aggr) {
>  		if (!own_packet)
>  			atomic_inc(&bat_priv->batman_queue_left);
> -		return;
> +		goto out;
>  	}
>  
>  	if ((atomic_read(&bat_priv->aggregated_ogms)) &&
> @@ -133,7 +137,7 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
>  		if (!own_packet)
>  			atomic_inc(&bat_priv->batman_queue_left);
>  		kfree(forw_packet_aggr);
> -		return;
> +		goto out;
>  	}
>  	skb_reserve(forw_packet_aggr->skb, sizeof(struct ethhdr));
>  
> @@ -164,6 +168,10 @@ static void new_aggregated_packet(unsigned char *packet_buff, int packet_len,
>  	queue_delayed_work(bat_event_workqueue,
>  			   &forw_packet_aggr->delayed_work,
>  			   send_time - jiffies);
> +
> +	return;
> +out:
> +	hardif_free_ref(if_incoming);
>  }
>  
>  /* aggregate a new packet into the existing aggregation */
> diff --git a/send.c b/send.c
> index f30d0c6..6e83e23 100644
> --- a/send.c
> +++ b/send.c
> @@ -377,6 +377,8 @@ static void forw_packet_free(struct forw_packet *forw_packet)
>  {
>  	if (forw_packet->skb)
>  		kfree_skb(forw_packet->skb);
> +	if (forw_packet->if_incoming)
> +		hardif_free_ref(forw_packet->if_incoming);
>  	kfree(forw_packet);
>  }
>  
> -- 
> 1.7.5.1

-- 
Antonio Quartulli

..each of us alone is worth nothing..
Ernesto "Che" Guevara

^ permalink raw reply	[flat|nested] 5+ messages in thread

* Re: [B.A.T.M.A.N.] [RFC PATCHv2] batman-adv: Add missing hardif_free_ref in forw_packet_free
  2011-05-11 18:59   ` [B.A.T.M.A.N.] [RFC PATCHv2] " Sven Eckelmann
@ 2011-05-14 15:26     ` Marek Lindner
  0 siblings, 0 replies; 5+ messages in thread
From: Marek Lindner @ 2011-05-14 15:26 UTC (permalink / raw)
  To: The list for a Better Approach To Mobile Ad-hoc Networking

On Wednesday 11 May 2011 20:59:06 Sven Eckelmann wrote:
> add_bcast_packet_to_list increases the refcount for if_incoming but the
> reference count is never decreased. The reference count must be
> increased for all kinds of forwarded packets which have the primary
> interface stored and forw_packet_free must decrease them. Also
> purge_outstanding_packets has to invoke forw_packet_free when a work
> item was really cancelled.

Applied in revision d18e485.

Thanks,
Marek

^ permalink raw reply	[flat|nested] 5+ messages in thread

end of thread, other threads:[~2011-05-14 15:26 UTC | newest]

Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2011-05-11  0:17 [B.A.T.M.A.N.] bug - unregister_netdevice: usage_count = 1 Antonio Quartulli
2011-05-11 12:44 ` [B.A.T.M.A.N.] [RFC PATCH] batman-adv: Add missing hardif_free_ref in forw_packet_free Sven Eckelmann
2011-05-11 18:59   ` [B.A.T.M.A.N.] [RFC PATCHv2] " Sven Eckelmann
2011-05-14 15:26     ` Marek Lindner
2011-05-13 13:05   ` [B.A.T.M.A.N.] [RFC PATCH] " Antonio Quartulli

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.