On Mon, Mar 18, 2019 at 12:48 PM David Ahern <d...@cumulusnetworks.com> wrote:
>
> On 3/18/19 12:36 PM, Xin Long wrote:
> > diff --git a/net/ipv6/route.c b/net/ipv6/route.c
> > index 4ef4bbd..754777d 100644
> > --- a/net/ipv6/route.c
> > +++ b/net/ipv6/route.c
> > @@ -1040,13 +1040,17 @@ static struct rt6_info *ip6_create_rt_rcu(struct
> > fib6_info *rt)
> > struct rt6_info *nrt;
> >
> > if (!fib6_info_hold_safe(rt))
> > - return NULL;
> > + goto fallback;
> >
> > nrt = ip6_dst_alloc(dev_net(dev), dev, flags);
> > - if (nrt)
> > + if (nrt) {
> > ip6_rt_copy_init(nrt, rt);
> > - else
> > + } else {
> > fib6_info_release(rt);
> > +fallback:
> > + nrt = dev_net(dev)->ipv6.ip6_null_entry;
> > + dst_hold(&nrt->dst);
> > + }
> >
> > return nrt;
> > }
>
> It seems appropriate for ip6_create_rt_rcu to always return an rt6_info
> given its use and v6 design. Since ip6_dst_alloc can also fail (e.g.,
> exceed gc_thresh) the fallback case should apply to both.
Agree. We should take care of ip6_dst_alloc() as well.
