Re: [Bug #14378] Problems with net/core/skbuff.c

[Eric Dumazet <eric.dumazet@gmail.com>]

Massimo Cetra a écrit :
> David Miller ha scritto:
>> From: "Rafael J. Wysocki" <rjw@sisk.pl>
>> Date: Mon, 12 Oct 2009 00:22:04 +0200 (CEST)
>>
>>  
>>> Bug-Entry    : http://bugzilla.kernel.org/show_bug.cgi?id=14378
>>> Subject        : Problems with net/core/skbuff.c
>>> Submitter    : Massimo Cetra <mcetra@navynet.it>
>>> Date        : 2009-10-08 14:51 (4 days old)
>>> References    : http://marc.info/?l=linux-kernel&m=125501488220358&w=4
>>>     
>>
>> I don't know what to do about this one.
>>
>> The user indicates that they have the vserver patches applied,
>> so maybe there is some interaction with that stuff.
>>   
> Actually i found another oops which is very similar to the previous one.
> Here, vserver is not involved, and the problem starts at drbd which
> lives in kernel space (the other oops started at ocfs2).
> 
> Both ocfs2 and drbd make heavy use of network I/O so i guess the problem
> is something in the network layer.
> 
> Anything i can do to help to debugging and solving this issue ?
> 
> Thanks
> Max
> 

Problem is kfree_skb() is called from irq context, wich is not allowed.

static void skb_release_head_state(struct sk_buff *skb)
{
...
if (skb->destructor) {
	WARN_ON(in_irq());
	skb->destructor();
}
...
}

virtio_net start_xmit() function calls free_old_xmit_skbs() and 
free_old_xmit_skbs() ultimately calls kfree_skb()

Quick fix would be to use dev_kfree_skb_any() instead,
because netpoll can definitly calls start_xmit()
with irq disabled.

Could you please following patch ?

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 8d00976..54bf091 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -454,7 +454,7 @@ static unsigned int free_old_xmit_skbs(struct virtnet_info *vi)
 		vi->dev->stats.tx_bytes += skb->len;
 		vi->dev->stats.tx_packets++;
 		tot_sgs += skb_vnet_hdr(skb)->num_sg;
-		kfree_skb(skb);
+		dev_kfree_skb_any(skb);
 	}
 	return tot_sgs;
 }