[PATCH 3/3] Fix use of skb after netif_rx

[PATCH 3/3] Fix use of skb after netif_rx

[Julia Lawall]

From: Julia Lawall <julia@diku.dk>

Recently, Wang Chen submitted a patch
(d30f53aeb31d453a5230f526bea592af07944564) to move a call to netif_rx(skb)
after a subsequent reference to skb, because netif_rx may call kfree_skb on
its argument.  netif_rx_ni calls netif_rx, so the same problem occurs in
the files below.

I have left the updating of dev->last_rx after the calls to netif_rx_ni
because it seems time dependent, but moved the other field updates before.

This was found using the following semantic match.
(http://www.emn.fr/x-info/coccinelle/)

// <smpl>
@@
expression skb, e,e1;
@@

(
 netif_rx(skb);
|
 netif_rx_ni(skb);
)
  ... when != skb = e
(
  skb = e1
|
* skb
)
// </smpl>

Signed-off-by: Julia Lawall <julia@diku.dk>
---

diff a/drivers/s390/net/ctcmain.c b/drivers/s390/net/ctcmain.c
--- a/drivers/s390/net/ctcmain.c	2007-12-05 09:21:56.000000000 +0100
+++ b/drivers/s390/net/ctcmain.c	2007-12-05 19:03:40.000000000 +0100
@@ -478,14 +478,14 @@ ctc_unpack_skb(struct channel *ch, struc
 		skb->dev = pskb->dev;
 		skb->protocol = pskb->protocol;
 		pskb->ip_summed = CHECKSUM_UNNECESSARY;
-		netif_rx_ni(skb);
 		/**
-		 * Successful rx; reset logflags
+		 * reset logflags
 		 */
 		ch->logflags = 0;
-		dev->last_rx = jiffies;
 		privptr->stats.rx_packets++;
 		privptr->stats.rx_bytes += skb->len;
+		netif_rx_ni(skb);
+		dev->last_rx = jiffies;
 		if (len > 0) {
 			skb_pull(pskb, header->length);
 			if (skb_tailroom(pskb) < LL_HEADER_LENGTH) {
diff a/drivers/s390/net/netiucv.c b/drivers/s390/net/netiucv.c
--- a/drivers/s390/net/netiucv.c	2007-10-22 11:25:20.000000000 +0200
+++ b/drivers/s390/net/netiucv.c	2007-12-05 19:03:10.000000000 +0100
@@ -639,14 +639,14 @@ static void netiucv_unpack_skb(struct iu
 		skb->dev = pskb->dev;
 		skb->protocol = pskb->protocol;
 		pskb->ip_summed = CHECKSUM_UNNECESSARY;
+		privptr->stats.rx_packets++;
+		privptr->stats.rx_bytes += skb->len;
 		/*
 		 * Since receiving is always initiated from a tasklet (in iucv.c),
 		 * we must use netif_rx_ni() instead of netif_rx()
 		 */
 		netif_rx_ni(skb);
 		dev->last_rx = jiffies;
-		privptr->stats.rx_packets++;
-		privptr->stats.rx_bytes += skb->len;
 		skb_pull(pskb, header->next);
 		skb_put(pskb, NETIUCV_HDRLEN);
 	}

Re: [PATCH 3/3] Fix use of skb after netif_rx

[Wang Chen]

Julia Lawall said the following on 2007-12-10 4:05:
> From: Julia Lawall <julia@diku.dk>
> // <smpl>
> @@
> expression skb, e,e1;
> @@
> 
> (
>  netif_rx(skb);
> |
>  netif_rx_ni(skb);
> )
>   ... when != skb = e
> (
>   skb = e1
> |
> * skb
> )
> // </smpl>
> 
> diff a/drivers/s390/net/ctcmain.c b/drivers/s390/net/ctcmain.c
> diff a/drivers/s390/net/netiucv.c b/drivers/s390/net/netiucv.c

Julia, seems that your semantic patch misses following place.

drivers/s390/net/qeth_main.c:2733
...
#endif
			rxrc = netif_rx(skb);
		card->dev->last_rx = jiffies;
		card->stats.rx_packets++;
		card->stats.rx_bytes += skb->len;
...

--
WCN

Re: [PATCH 3/3] Fix use of skb after netif_rx

[Julia Lawall]

> > // </smpl>
> > 
> > diff a/drivers/s390/net/ctcmain.c b/drivers/s390/net/ctcmain.c
> > diff a/drivers/s390/net/netiucv.c b/drivers/s390/net/netiucv.c
> 
> Julia, seems that your semantic patch misses following place.
> 
> drivers/s390/net/qeth_main.c:2733
> ...
> #endif
> 			rxrc = netif_rx(skb);
> 		card->dev->last_rx = jiffies;
> 		card->stats.rx_packets++;
> 		card->stats.rx_bytes += skb->len;
> ...

Actually, I found this one as well, but I wasn't sure what to do with it.  
This one is a bit more complicated because the line with the call to 
netif_rx is in an else branch if the #ifdef above is taken.  So I wasn't 
sure what would be the best way to solve the problem in this case.

Perhaps the solution would be just to save the value of the len field 
in a local variable in this case, as you proposed in your original patch.

julia

Re: [PATCH 3/3] Fix use of skb after netif_rx

[Wang Chen]

Julia Lawall said the following on 2007-12-10 15:18:
>> Julia, seems that your semantic patch misses following place.
>>
>> drivers/s390/net/qeth_main.c:2733
>> ...
>> #endif
>> 			rxrc = netif_rx(skb);
>> 		card->dev->last_rx = jiffies;
>> 		card->stats.rx_packets++;
>> 		card->stats.rx_bytes += skb->len;
>> ...
> 
> Actually, I found this one as well, but I wasn't sure what to do with it.  
> This one is a bit more complicated because the line with the call to 
> netif_rx is in an else branch if the #ifdef above is taken.  So I wasn't 
> sure what would be the best way to solve the problem in this case.
> 
> Perhaps the solution would be just to save the value of the len field 
> in a local variable in this case, as you proposed in your original patch.
> 

I agree.

BTW, please send driver patch to Jeff Garzik <jgarzik@pobox.com> and
cc to netdev@vger.kernel.org.

--
WCN

Re: [PATCH 3/3] Fix use of skb after netif_rx

[David Miller]

From: Julia Lawall <julia@diku.dk>
Date: Sun, 9 Dec 2007 21:05:30 +0100 (CET)

> From: Julia Lawall <julia@diku.dk>
> 
> Recently, Wang Chen submitted a patch
> (d30f53aeb31d453a5230f526bea592af07944564) to move a call to netif_rx(skb)
> after a subsequent reference to skb, because netif_rx may call kfree_skb on
> its argument.  netif_rx_ni calls netif_rx, so the same problem occurs in
> the files below.
> 
> I have left the updating of dev->last_rx after the calls to netif_rx_ni
> because it seems time dependent, but moved the other field updates before.
> 
> This was found using the following semantic match.
> (http://www.emn.fr/x-info/coccinelle/)
 ...
> Signed-off-by: Julia Lawall <julia@diku.dk>

Applied.