Re: BUG in net/l2tp/l2tp_core.c

["François Cachereul" <f.cachereul@alphalink.fr>]

On 10/09/2013 01:05 PM, Eric Dumazet wrote:
> On Wed, 2013-10-09 at 12:11 +0200, François Cachereul wrote:
>> Hi,
>>
>> I got the following BUG when using l2tp modules with smp kernel.
>> I noticed that l2tp_xmit_skb uses bh_lock_sock/bh_unlok_sock which
>> doesn't seem to be correct because it does a lot of stuff and probably
>> sleep before releasing the lock. I try replacing
>> bh_lock_sock/bh_unlock_sock with lock_sock/release_sock and the BUG
>> doesn't happened anymore. Is it correct ?
>>
>> Regards
>> François
>>
> 
> At first glance, you might read commit
> 6af88da14ee284aaad6e4326da09a89191ab6165
> ("l2tp: Fix locking in l2tp_core.c")
> 
> l2tp_eth_dev_xmit() is called from BH context, so you cannot use
> lock_sock()
> 
> Try the following patch instead :
> 
> diff --git a/net/l2tp/l2tp_ppp.c b/net/l2tp/l2tp_ppp.c
> index f0a7ada..ffda81e 100644
> --- a/net/l2tp/l2tp_ppp.c
> +++ b/net/l2tp/l2tp_ppp.c
> @@ -353,7 +353,9 @@ static int pppol2tp_sendmsg(struct kiocb *iocb, struct socket *sock, struct msgh
>  		goto error_put_sess_tun;
>  	}
>  
> +	local_bh_disable();
>  	l2tp_xmit_skb(session, skb, session->hdr_len);
> +	local_bh_enable();
>  
>  	sock_put(ps->tunnel_sock);
>  	sock_put(sk);
> @@ -422,7 +424,9 @@ static int pppol2tp_xmit(struct ppp_channel *chan, struct sk_buff *skb)
>  	skb->data[0] = ppph[0];
>  	skb->data[1] = ppph[1];
>  
> +	local_bh_disable();
>  	l2tp_xmit_skb(session, skb, session->hdr_len);
> +	local_bh_enable();
>  
>  	sock_put(sk_tun);
>  	sock_put(sk);
> 
> 

That works. Thanks

François