Skip to contentSkip to conversation

__netif_receive_skb_core: pass skb by reference

PATCH Superseded Boris Sukholitko 3 participants 5 messages May 18, 2020 – May 19, 2020

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 __netif_receive_skb_core: pass skb by reference

__netif_receive_skb_core may change the skb pointer passed into it (e.g. in rx_handler). The original skb may be freed as a result of this operation.

The callers of __netif_receive_skb_core may further process original skb by using pt_prev pointer returned by __netif_receive_skb_core thus leading to unpleasant effects.

The solution is to pass skb by reference into __netif_receive_skb_core.

---
 net/core/dev.c | 14 +++++++++-----
 1 file changed, 9 insertions(+), 5 deletions(-)
net/core/dev.c +9 -5
index 6d327b7aa813..0c2bbf479f19 100644
@@ -4988,7 +4988,7 @@ static inline int nf_ingress(struct sk_buff *skb, struct packet_type **pt_prev,
 	return 0;
 }
 
-static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
+static int __netif_receive_skb_core(struct sk_buff **pskb, bool pfmemalloc,
 				    struct packet_type **ppt_prev)
 {
 	struct packet_type *ptype, *pt_prev;
@@ -4997,6 +4997,7 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 	bool deliver_exact = false;
 	int ret = NET_RX_DROP;
 	__be16 type;
+	struct sk_buff *skb = *pskb;
 
 	net_timestamp_check(!netdev_tstamp_prequeue, skb);
 
@@ -5023,8 +5024,10 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 		ret2 = do_xdp_generic(rcu_dereference(skb->dev->xdp_prog), skb);
 		preempt_enable();
 
-		if (ret2 != XDP_PASS)
-			return NET_RX_DROP;
+		if (ret2 != XDP_PASS) {
+			ret = NET_RX_DROP;
+			goto out;
+		}
 		skb_reset_mac_len(skb);
 	}
 
@@ -5174,6 +5177,7 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 	}
 
 out:
+	*pskb = skb;
 	return ret;
 }
 
@@ -5183,7 +5187,7 @@ static int __netif_receive_skb_one_core(struct sk_buff *skb, bool pfmemalloc)
 	struct packet_type *pt_prev = NULL;
 	int ret;
 
-	ret = __netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
+	ret = __netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
 	if (pt_prev)
 		ret = INDIRECT_CALL_INET(pt_prev->func, ipv6_rcv, ip_rcv, skb,
 					 skb->dev, pt_prev, orig_dev);
@@ -5261,7 +5265,7 @@ static void __netif_receive_skb_list_core(struct list_head *head, bool pfmemallo
 		struct packet_type *pt_prev = NULL;
 
 		skb_list_del_init(skb);
-		__netif_receive_skb_core(skb, pfmemalloc, &pt_prev);
+		__netif_receive_skb_core(&skb, pfmemalloc, &pt_prev);
 		if (!pt_prev)
 			continue;
 		if (pt_curr != pt_prev || od_curr != orig_dev) {
Signature · 1 line
2.19.2
#1 lore raw
Quoted 11 lines · Boris Sukholitko

__netif_receive_skb_core may change the skb pointer passed into it (e.g. in rx_handler). The original skb may be freed as a result of this operation.

The callers of __netif_receive_skb_core may further process original skb by using pt_prev pointer returned by __netif_receive_skb_core thus leading to unpleasant effects.

The solution is to pass skb by reference into __netif_receive_skb_core.

Please provide a Fixes: tag for such a scary patch.

Thanks.

#3 lore raw
In reply to Boris Sukholitko
Quoted 11 lines · Boris Sukholitko

__netif_receive_skb_core may change the skb pointer passed into it (e.g. in rx_handler). The original skb may be freed as a result of this operation.

The callers of __netif_receive_skb_core may further process original skb by using pt_prev pointer returned by __netif_receive_skb_core thus leading to unpleasant effects.

The solution is to pass skb by reference into __netif_receive_skb_core.

I think this patch is correct, but there's a couple of things I'd like to see before adding my Ack.
Firstly, please add a Fixes: tag; I expect the relevant commit will be 88eb1944e18c ("net: core: propagate SKB lists through packet_type lookup") but I'm not 100% sure so do check that yourself.
Secondly:

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 5182

Quoted 7 lines
net/core/dev.c +1 -0
@@ -5174,6 +5177,7 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 	}
 
 out:
+	*pskb = skb;
 	return ret;
 }

Could we have some sort of WARN_ONs (maybe under #ifdef DEBUG) to check that we never have a NULL skb with a non-NULL pt_prev?  Or at least a comment at the top of the function stating this part of its contract with callers?  I've gone through and convinced myself that it never happens currently, but that depends on some fairly subtle details.

-ed

#4 lore raw
Quoted 3 lines · Edward Cree

Firstly, please add a Fixes: tag; I expect the relevant commit will be 88eb1944e18c ("net: core: propagate SKB lists through packet_type lookup") but I'm not 100% sure so do check that yourself.

You are right, this is the right commit. Fixes tag added.

Quoted 4 lines

Secondly:

Quoted 3 lines
(quoted diff) +0 -0
@@ -5174,6 +5177,7 @@ static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc,
 	}
 
Quoted 9 lines

net/core/dev.c static int __netif_receive_skb_core(struct sk_buff *skb, bool pfmemalloc, · line 5182

Quoted 4 lines
net/core/dev.c +1 -0
 out:
+	*pskb = skb;
 	return ret;
 }

Could we have some sort of WARN_ONs (maybe under #ifdef DEBUG) to check that we never have a NULL skb with a non-NULL pt_prev?  Or at least a comment at the top of the function stating this part of its contract with callers?  I've gone through and convinced myself that it never happens currently, but that depends on some fairly subtle details.

I've added the comment at this hunk. The *ppt_prev assignment happens several lines above and skb is being used right next to it.

Please see v2 of the patch with this changes.

Thanks,
Boris.