diff mbox

net: Fix resetting network_header in neigh_resolve_output and neigh_connected_output

Message ID 15592436817.c86fc05b10946.6198596471268005484@abahmed.com
State Changes Requested, archived
Delegated to: David Miller
Headers show

Commit Message

Abdelrhman Ahmed June 27, 2016, 2:28 p.m. UTC
neigh_resolve_output and neigh_connected_output resets the skb to 
network_header because of the retry loop and this reset will pull down the
data pointer to the network header in the first iteration then hardware header
will be added, but it will overwrite any data which is inserted between network
header and hardware header (for example by netfilter hooks) only for the first
packet(s) before using cached hardware header as neigh_hh_output (which is
called for using cached hardware header) does not reset to the network header.

The fix is to reset with reference to skb's data before loop instead of network
header.

Fixes: e1f165032c8b ("net: Fix skb_under_panic oops in neigh_resolve_output")
Signed-off-by: Abdelrhman Ahmed <ab@abahmed.com>
---
 net/core/neighbour.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)

Comments

David Miller July 1, 2016, 7:57 p.m. UTC | #1
From: Abdelrhman Ahmed <ab@abahmed.com>
Date: Mon, 27 Jun 2016 16:28:59 +0200

> @@ -1293,15 +1293,19 @@ int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
>  	int rc = 0;
>  
>  	if (!neigh_event_send(neigh, skb)) {
> -		int err;
> +		int err, offset;
>  		struct net_device *dev = neigh->dev;
> +		unsigned char *data;
>  		unsigned int seq;
>  
>  		if (dev->header_ops->cache && !neigh->hh.hh_len)
>  			neigh_hh_init(neigh);
>  
> +		data = skb->data;
> +
>  		do {
> -			__skb_pull(skb, skb_network_offset(skb));
> +			offset = data - skb->data;
> +			__skb_pull(skb, offset);

This is definitely not right, using skb->data for this.  It may work
for the cases you have tested but it is not generally correct.

You must use the skb network header.

You are just trying to avoid doing the pull more than once if we loop
right?  Then simply use a boolean to track that.
diff mbox

Patch

diff --git a/net/core/neighbour.c b/net/core/neighbour.c
index 29dd8cc..7aac242 100644
--- a/net/core/neighbour.c
+++ b/net/core/neighbour.c
@@ -1293,15 +1293,19 @@  int neigh_resolve_output(struct neighbour *neigh, struct sk_buff *skb)
 	int rc = 0;
 
 	if (!neigh_event_send(neigh, skb)) {
-		int err;
+		int err, offset;
 		struct net_device *dev = neigh->dev;
+		unsigned char *data;
 		unsigned int seq;
 
 		if (dev->header_ops->cache && !neigh->hh.hh_len)
 			neigh_hh_init(neigh);
 
+		data = skb->data;
+
 		do {
-			__skb_pull(skb, skb_network_offset(skb));
+			offset = data - skb->data;
+			__skb_pull(skb, offset);
 			seq = read_seqbegin(&neigh->ha_lock);
 			err = dev_hard_header(skb, dev, ntohs(skb->protocol),
 					      neigh->ha, NULL, skb->len);
@@ -1326,11 +1330,15 @@  EXPORT_SYMBOL(neigh_resolve_output);
 int neigh_connected_output(struct neighbour *neigh, struct sk_buff *skb)
 {
 	struct net_device *dev = neigh->dev;
+	unsigned char *data;
 	unsigned int seq;
-	int err;
+	int err, offset;
+
+	data = skb->data;
 
 	do {
-		__skb_pull(skb, skb_network_offset(skb));
+		offset = data - skb->data;
+		__skb_pull(skb, offset);
 		seq = read_seqbegin(&neigh->ha_lock);
 		err = dev_hard_header(skb, dev, ntohs(skb->protocol),
 				      neigh->ha, NULL, skb->len);