Re: ipv6 addrconfg warn_on hit: WARN_ON(ifp->idev->valid_ll_addr_cnt < 0);

[Hannes Frederic Sowa <hannes@stressinduktion.org>]

On Thu, Jan 16, 2014 at 06:27:06PM +0100, Jiri Pirko wrote:
> Thu, Jan 16, 2014 at 04:18:12PM CET, hannes@stressinduktion.org wrote:
> >On Thu, Jan 16, 2014 at 03:38:17PM +0100, Hannes Frederic Sowa wrote:
> >> On Thu, Jan 16, 2014 at 03:07:01PM +0100, Jiri Pirko wrote:
> >> > Thu, Jan 16, 2014 at 02:53:23PM CET, jiri@resnulli.us wrote:
> >> > >Hi Hannes.
> >> > >
> >> > >WARN_ON(ifp->idev->valid_ll_addr_cnt < 0);
> >> > >
> >> > >We did hit once this warning during the tests. The person who hit this
> >> > >says that it was during the setup of many macvlan devices.
> >> > >
> >> > >I examined the code but I'm not sure how this could happen. Looks like a
> >> > >race condition between addrconf_dad_completed() and addrconf_ifdown(). 
> >> > >Not sure how to easily resolve that though.
> >> 
> >> That seems to be the case. Actually we don't need to count precisiely
> >> here, we just must precisiely identify the situation where the first
> >> LL address comes into operational state. Maybe we can implement this
> >> somehow differently. I'll play with the code soon.
> >
> >Maybe something like this and then throw out the whole ll counting stuff:
> >
> >diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c
> >index 6913a82..105105a 100644
> >--- a/net/ipv6/addrconf.c
> >+++ b/net/ipv6/addrconf.c
> >@@ -3233,6 +3233,19 @@ out:
> > 	in6_ifa_put(ifp);
> > }
> > 
> >+/* idev must be at least read locked */
> >+static bool ipv6_lonely_lladdr(struct inet6_dev *idev, struct inet6_ifaddr *ifp)
> >+{
> >+	bool ret = true;
> >+	struct inet6_ifaddr *ifpiter;
> >+
> >+	list_for_each_entry(ifpiter, &idev->addr_list, if_list) {
> >+		if (ifp != ifpiter && ifpiter->scope == IFA_LINK)
> >+			ret = false;
> >+	}
> >+	return ret;
> >+}
> >+
> > static void addrconf_dad_completed(struct inet6_ifaddr *ifp)
> > {
> > 	struct net_device *dev = ifp->idev->dev;
> >@@ -3253,8 +3266,7 @@ static void addrconf_dad_completed(struct inet6_ifaddr *ifp)
> > 
> > 	read_lock_bh(&ifp->idev->lock);
> > 	spin_lock(&ifp->lock);
> >-	send_mld = ipv6_addr_type(&ifp->addr) & IPV6_ADDR_LINKLOCAL &&
> >-		   ifp->idev->valid_ll_addr_cnt == 1;
> >+	send_mld = ifp->scope == IFA_LINK && ipv6_lonely_lladdr(ifp->idev, ifp);
> > 	send_rs = send_mld &&
> > 		  ipv6_accept_ra(ifp->idev) &&
> > 		  ifp->idev->cnf.rtr_solicits > 0 &&
> 
> Sounds sane to me. Would you care to submit this please?
> Do not forget to remove all the valid_ll_addr_cnt stuff :)

I'll do. ;) Currently testing the changes.

Thanks,

  Hannes