Re: [PATCH] net: ieee802154: fix net_device reference release too early

[Stefan Schmidt]

Hello.

On Thu, 2017-05-18 at 15:14, Stefan Schmidt wrote:
> Hello.
> 
> On Thu, 2017-05-18 at 15:50, linzhang wrote:
> > This patch fixes the kernel oops when release net_device reference in 
> > advance. In function raw_sendmsg(i think the dgram_sendmsg has the same 
> > problem), there is a race condition between dev_put and dev_queue_xmit
> > when the device is gong that maybe lead to dev_queue_ximt to see
> > an illegal net_device pointer.
> > 
> 
> You have a test case to reproduce this oops? I fear I have not seen
> one.

If you have a test case handy adding it to the commit would be handy. If you do
not have one around we can do without.

> > So i think that dev_put should be behind of the dev_queue_xmit.
> > 
> > Also, explicit set skb->sk is needless, sock_alloc_send_skb is
> > already set it.
> 
> You could have put this fixup in a different patch.

I actually would request you to split this into two patches. One for the
removal of the sk setting and one for the race condition fix.

> > Signed-off-by: linzhang <xiaolou4617@xxxxxxxxx>
> 
> This looks more like a username instead of a real name. If you have Lin
> Zhang as you English real name that would be better here. :)

This would be also appreciated.

> > ---
> >  net/ieee802154/socket.c | 10 ++++------
> >  1 file changed, 4 insertions(+), 6 deletions(-)
> > 
> > diff --git a/net/ieee802154/socket.c b/net/ieee802154/socket.c
> > index eedba76..a60658c 100644
> > --- a/net/ieee802154/socket.c
> > +++ b/net/ieee802154/socket.c
> > @@ -301,15 +301,14 @@ static int raw_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
> >  		goto out_skb;
> >  
> >  	skb->dev = dev;
> > -	skb->sk  = sk;
> >  	skb->protocol = htons(ETH_P_IEEE802154);
> >  
> > -	dev_put(dev);
> > -
> >  	err = dev_queue_xmit(skb);
> >  	if (err > 0)
> >  		err = net_xmit_errno(err);
> >  
> > +	dev_put(dev);
> > +
> >  	return err ?: size;
> >  
> >  out_skb:
> > @@ -690,15 +689,14 @@ static int dgram_sendmsg(struct sock *sk, struct msghdr *msg, size_t size)
> >  		goto out_skb;
> >  
> >  	skb->dev = dev;
> > -	skb->sk  = sk;
> >  	skb->protocol = htons(ETH_P_IEEE802154);
> >  
> > -	dev_put(dev);
> > -
> >  	err = dev_queue_xmit(skb);
> >  	if (err > 0)
> >  		err = net_xmit_errno(err);
> >  
> > +	dev_put(dev);
> > +
> >  	return err ?: size;
> 
> Going to give this a test ride here now.

I gave it a ride in my testbed and I encountered no problems. While I have never
seen the race and oops myself doing the dev_put before the xmit can surely lead to
such a race and the fix is valid.

Once you have done the changes requested above and re-submit your two patches you can
add my

Acked-by: Stefan Schmidt <stefan@xxxxxxxxxxxxxxx>

to both of them.

regards
Stefan Schmidt