On 8/26/20 10:48 AM, Ido Schimmel wrote:
> From: Ido Schimmel <ido...@nvidia.com>
>
> Each nexthop group contains an indication if it has IPv4 nexthops
> ('has_v4'). Its purpose is to prevent IPv6 routes from using groups with
> IPv4 nexthops.
>
> However, the indication is not updated when a nexthop is removed. This
> results in the kernel wrongly rejecting IPv6 routes from pointing to
> groups that only contain IPv6 nexthops. Example:
>
> # ip nexthop replace id 1 via 192.0.2.2 dev dummy10
> # ip nexthop replace id 2 via 2001:db8:1::2 dev dummy10
> # ip nexthop replace id 10 group 1/2
> # ip nexthop del id 1
> # ip route replace 2001:db8:10::/64 nhid 10
> Error: IPv6 routes can not use an IPv4 nexthop.
>
> Solve this by updating the indication according to the new set of
> member nexthops.
>
> Signed-off-by: Ido Schimmel <ido...@nvidia.com>
> ---
> net/ipv4/nexthop.c | 8 +++++++-
> 1 file changed, 7 insertions(+), 1 deletion(-)
>
> diff --git a/net/ipv4/nexthop.c b/net/ipv4/nexthop.c
> index 1b736e3e1baa..5199a2815df6 100644
> --- a/net/ipv4/nexthop.c
> +++ b/net/ipv4/nexthop.c
> @@ -797,7 +797,7 @@ static void remove_nh_grp_entry(struct net *net, struct
> nh_grp_entry *nhge,
> return;
> }
>
> - newg->has_v4 = nhg->has_v4;
> + newg->has_v4 = false;
> newg->mpath = nhg->mpath;
> newg->fdb_nh = nhg->fdb_nh;
> newg->num_nh = nhg->num_nh;
> @@ -806,12 +806,18 @@ static void remove_nh_grp_entry(struct net *net, struct
> nh_grp_entry *nhge,
> nhges = nhg->nh_entries;
> new_nhges = newg->nh_entries;
> for (i = 0, j = 0; i < nhg->num_nh; ++i) {
> + struct nh_info *nhi;
> +
> /* current nexthop getting removed */
> if (nhg->nh_entries[i].nh == nh) {
> newg->num_nh--;
> continue;
> }
>
> + nhi = rtnl_dereference(nhges[i].nh->nh_info);
> + if (nhi->family == AF_INET)
> + newg->has_v4 = true;
> +
> list_del(&nhges[i].nh_list);
> new_nhges[j].nh_parent = nhges[i].nh_parent;
> new_nhges[j].nh = nhges[i].nh;
>
Reviewed-by: David Ahern <dsah...@gmail.com>
