[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1463086553.23934.165.camel@edumazet-glaptop3.roam.corp.google.com>
Date: Thu, 12 May 2016 13:55:53 -0700
From: Eric Dumazet <eric.dumazet@...il.com>
To: Alexander Duyck <alexander.duyck@...il.com>
Cc: Alexander Duyck <aduyck@...antis.com>,
Netdev <netdev@...r.kernel.org>,
David Miller <davem@...emloft.net>,
Thomas Graf <tgraf@...g.ch>, Tom Herbert <tom@...bertland.com>,
Jiri Benc <jbenc@...hat.com>
Subject: Re: [net-next PATCH v2] udp: Resolve NULL pointer dereference over
flow-based vxlan device
On Thu, 2016-05-12 at 13:27 -0700, Alexander Duyck wrote:
> I'm assuming this was using skb_dst(skb)->dev in order to allow for
> use of this function by other callers since the original function
> __udp4_lib_lookup_skb was using that. If we change this then it
> reduces the likelihood of the code being reusable by other callers.
> In such a case I would probably want to go through and also rename the
> functions to make sure they are tagged as being GRO specific.
>
__udp4_lib_lookup_skb() should use skb->dev as well,
or get the net pointer from its caller.
It is called from __udp4_lib_rcv() and this one simply does :
struct net *net = dev_net(skb->dev);
So for consistency (and small performance gain) I would suggest :
diff --git a/net/ipv4/udp.c b/net/ipv4/udp.c
index f67f52ba4809..d9006f2d28eb 100644
--- a/net/ipv4/udp.c
+++ b/net/ipv4/udp.c
@@ -604,7 +604,7 @@ static inline struct sock *__udp4_lib_lookup_skb(struct sk_buff *skb,
{
const struct iphdr *iph = ip_hdr(skb);
- return __udp4_lib_lookup(dev_net(skb_dst(skb)->dev), iph->saddr, sport,
+ return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
iph->daddr, dport, inet_iif(skb),
udptable, skb);
}
Powered by blists - more mailing lists