Re: regression when connecting to ipv6 localhost

[Eric Dumazet <eric.dumazet@gmail.com>]

On Tue, 2012-10-16 at 18:07 +0200, Jan Hinnerk Stosch wrote:
> 2012/10/15 Jan Hinnerk Stosch <janhinnerk.stosch@gmail.com>:
> >
> > Yes. Attached..
> 
> Today I read about the existence of the "git bisect" function and if I
> have not done anything wrong the "bad commit" which lead to this
> regression is ddbe503203855939946430e39bae58de11b70b69.
> Perhaps this can help you.

Interesting, I was about to ask you to do a bisection ;)

So the hash value of ::1 was changed, and it has bad effect on your
machine...

Could it be hash collision and an old bug being uncovered ?

Could you trace inet6_addr_hash() a bit ?

diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
index d7c56f8..096c04c 100644
--- a/net/ipv6/addrconf.c
+++ b/net/ipv6/addrconf.c
@@ -580,7 +580,10 @@ ipv6_link_dev_addr(struct inet6_dev *idev, struct inet6_ifaddr *ifp)
 
 static u32 inet6_addr_hash(const struct in6_addr *addr)
 {
-	return hash_32(ipv6_addr_hash(addr), IN6_ADDR_HSIZE_SHIFT);
+	u32 res = hash_32(ipv6_addr_hash(addr), IN6_ADDR_HSIZE_SHIFT);
+
+	pr_err("inet6_addr_hash(%pi6) %u\n", addr, res);
+	return res;
 }
 
 /* On success it returns ifp with increased reference count */