Re: possible recursive locking in ATM layer

Re: possible recursive locking in ATM layer

[Arjan van de Ven]

From: Arjan van de Ven <arjan@linux.intel.com>

> Linux version 2.6.17-git22 (duncan@baldrick) (gcc version 4.0.3 (Ubuntu 4.0.3-1ubuntu5)) #20 PREEMPT Tue Jul 4 10:35:04 CEST 2006
> 
> [ 2381.598609] =============================================
> [ 2381.619314] [ INFO: possible recursive locking detected ]
> [ 2381.635497] ---------------------------------------------
> [ 2381.651706] atmarpd/2696 is trying to acquire lock:
> [ 2381.666354]  (&skb_queue_lock_key){-+..}, at: [<c028c540>] skb_migrate+0x24/0x6c
> [ 2381.688848]


ok this is a real potential deadlock in a way, it takes two locks of 2
skbuffs without doing any kind of lock ordering; I think the following
patch should fix it. Just sort the lock taking order by address of the
skb.. it's not pretty but it's the best this can do in a minimally
invasive way.

I still agree with the comment that this code shouldn't live in the atm
layer...

Signed-off-by: Arjan van de Ven <arjan@linux.intel.com>

---
 net/atm/ipcommon.c |   13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

Index: linux-2.6.17-mm6/net/atm/ipcommon.c
===================================================================
--- linux-2.6.17-mm6.orig/net/atm/ipcommon.c
+++ linux-2.6.17-mm6/net/atm/ipcommon.c
@@ -25,8 +25,8 @@
 /*
  * skb_migrate appends the list at "from" to "to", emptying "from" in the
  * process. skb_migrate is atomic with respect to all other skb operations on
- * "from" and "to". Note that it locks both lists at the same time, so beware
- * of potential deadlocks.
+ * "from" and "to". Note that it locks both lists at the same time, so to deal
+ * with the lock ordering, the locks are taken in address order.
  *
  * This function should live in skbuff.c or skbuff.h.
  */
@@ -39,8 +39,13 @@ void skb_migrate(struct sk_buff_head *fr
 	struct sk_buff *skb_to = (struct sk_buff *) to;
 	struct sk_buff *prev;
 
-	spin_lock_irqsave(&from->lock,flags);
-	spin_lock(&to->lock);
+	if (from<to) {
+		spin_lock_irqsave(&from->lock,flags);
+		spin_lock_nested(&to->lock, SINGLE_DEPTH_NESTING);
+	} else {
+		spin_lock_irqsave(&to->lock, flags);
+		spin_lock_nested(&from->lock, SINGLE_DEPTH_NESTING);
+	}
 	prev = from->prev;
 	from->next->prev = to->prev;
 	prev->next = skb_to;

Re: possible recursive locking in ATM layer

[Duncan Sands]

> ok this is a real potential deadlock in a way, it takes two locks of 2
> skbuffs without doing any kind of lock ordering; I think the following
> patch should fix it. Just sort the lock taking order by address of the
> skb.. it's not pretty but it's the best this can do in a minimally
> invasive way.

The patch is effective.  Thanks for making it!

Best wishes,

Duncan.

Re: possible recursive locking in ATM layer

[Avi Kivity]

Arjan van de Ven wrote:
>
> From: Arjan van de Ven <arjan@linux.intel.com>
>
> > Linux version 2.6.17-git22 (duncan@baldrick) (gcc version 4.0.3 
> (Ubuntu 4.0.3-1ubuntu5)) #20 PREEMPT Tue Jul 4 10:35:04 CEST 2006
>
> >
> > [ 2381.598609] =============================================
> > [ 2381.619314] [ INFO: possible recursive locking detected ]
> > [ 2381.635497] ---------------------------------------------
> > [ 2381.651706] atmarpd/2696 is trying to acquire lock:
> > [ 2381.666354]  (&skb_queue_lock_key){-+..}, at: [<c028c540>] 
> skb_migrate+0x24/0x6c
> > [ 2381.688848]
>
>
> ok this is a real potential deadlock in a way, it takes two locks of 2
> skbuffs without doing any kind of lock ordering; I think the following
> patch should fix it. Just sort the lock taking order by address of the
> skb.. it's not pretty but it's the best this can do in a minimally
> invasive way.
>

Isn't it a deadlock only if skb_migrate(a, b) and skb_migrate(b, a) can 
be called concurrently?

-- 
error compiling committee.c: too many arguments to function

Re: possible recursive locking in ATM layer

[Arjan van de Ven]

On Wed, 2006-07-05 at 16:33 +0300, Avi Kivity wrote:
> Arjan van de Ven wrote:
> >
> > From: Arjan van de Ven <arjan@linux.intel.com>
> >
> > > Linux version 2.6.17-git22 (duncan@baldrick) (gcc version 4.0.3 
> > (Ubuntu 4.0.3-1ubuntu5)) #20 PREEMPT Tue Jul 4 10:35:04 CEST 2006
> >
> > >
> > > [ 2381.598609] =============================================
> > > [ 2381.619314] [ INFO: possible recursive locking detected ]
> > > [ 2381.635497] ---------------------------------------------
> > > [ 2381.651706] atmarpd/2696 is trying to acquire lock:
> > > [ 2381.666354]  (&skb_queue_lock_key){-+..}, at: [<c028c540>] 
> > skb_migrate+0x24/0x6c
> > > [ 2381.688848]
> >
> >
> > ok this is a real potential deadlock in a way, it takes two locks of 2
> > skbuffs without doing any kind of lock ordering; I think the following
> > patch should fix it. Just sort the lock taking order by address of the
> > skb.. it's not pretty but it's the best this can do in a minimally
> > invasive way.
> >
> 
> Isn't it a deadlock only if skb_migrate(a, b) and skb_migrate(b, a) can 
> be called concurrently?

yes, well, and if there are no  other double-takers...
>