From: Eric Dumazet <edumazet@google.com>

On Wed, 2013-06-26 at 02:29 -0700, Eric Dumazet wrote:

On Wed, 2013-06-26 at 02:22 -0700, Eric Dumazet wrote:

quoted

ip6_rt_put

Oh well...

It sounds we mix ipv4/ipv6 dst on an IPV6 socket.

So an IPV4 actor thinks he got a "struct rtable" pointer and messes
critical fields which overlay struct rt6_info components, for example
rt6i_node

rt6_inode  shares the same storage with : 

rt_type/rt_is_input/rt_uses_gateway/rt_iif

In my case, rt_iif is 0

CC Steffen Klassert, because of commit
8141ed9fcedb278f4a3a78680591bef1e55f75fb
("ipv4: Add a socket release callback for datagram sockets")

ip6_sk_dst_lookup_flow()/ip6_sk_dst_check() look buggy, they assume the
socket dst is an IPv6 one.

But its not true if ip4_datagram_connect() was called before.

Following patch fixes the problem for me :

[PATCH] ipv6: ip6_sk_dst_check() must not assume ipv6 dst

It's possible to use AF_INET6 sockets and to connect to an IPv4
destination. After this, socket dst cache is a pointer to a rtable,
not rt6_info.

ip6_sk_dst_check() should check the socket dst cache is IPv6, or else
various corruptions/crashes can happen.

Dave Jones can reproduce immediate crash with 
trinity -q -l off -n -c sendmsg -c connect

With help from Hannes Frederic Sowa

Reported-by: Dave Jones <redacted>
Reported-by: Hannes Frederic Sowa <redacted>
Signed-off-by: Eric Dumazet <edumazet@google.com>
---
 net/ipv6/ip6_output.c |    8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index 95703ba..d5d20cd 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c

@@ -821,11 +821,17 @@ static struct dst_entry *ip6_sk_dst_check(struct sock *sk,
 					  const struct flowi6 *fl6)
 {
 	struct ipv6_pinfo *np = inet6_sk(sk);
-	struct rt6_info *rt = (struct rt6_info *)dst;
+	struct rt6_info *rt;
 
 	if (!dst)
 		goto out;
 
+	if (dst->ops->family != AF_INET6) {
+		dst_release(dst);
+		return NULL;
+	}
+
+	rt = (struct rt6_info *)dst;
 	/* Yes, checking route validity in not connected
 	 * case is not very simple. Take into account,
 	 * that we do not support routing by source, TOS,