mlxsw: spectrum_router: Handle encap to demoted tunnels
authorPetr Machata <petrm@mellanox.com>
Tue, 28 Nov 2017 12:17:13 +0000 (13:17 +0100)
committerDavid S. Miller <davem@davemloft.net>
Tue, 28 Nov 2017 14:55:47 +0000 (09:55 -0500)
Some tunnels that are offloadable on their own can nonetheless be
demoted to slow path if their local address is in conflict with that of
another tunnel. When a route is formed for such a tunnel,
mlxsw_sp_nexthop_ipip_init() fails to find the corresponding IPIP entry,
and that triggers a FIB abort.

Resolve the problem by not assuming that a tunnel for which
mlxsw_sp_ipip_ops.can_offload() holds also automatically has an IPIP
entry.

Fixes: af641713e97d ("mlxsw: spectrum_router: Onload conflicting tunnels")
Signed-off-by: Petr Machata <petrm@mellanox.com>
Reviewed-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c

index a23e452f7261b9e67dc90920e25c4b9fa88e7192..17e2153a8dc9c67cc4b5a9ec45bfb171b35984c1 100644 (file)
@@ -3361,22 +3361,19 @@ static bool mlxsw_sp_ipip_netdev_ul_up(struct net_device *ol_dev)
        return ul_dev ? (ul_dev->flags & IFF_UP) : true;
 }
 
-static int mlxsw_sp_nexthop_ipip_init(struct mlxsw_sp *mlxsw_sp,
-                                     struct mlxsw_sp_nexthop *nh,
-                                     struct net_device *ol_dev)
+static void mlxsw_sp_nexthop_ipip_init(struct mlxsw_sp *mlxsw_sp,
+                                      struct mlxsw_sp_nexthop *nh,
+                                      struct mlxsw_sp_ipip_entry *ipip_entry)
 {
        bool removing;
 
        if (!nh->nh_grp->gateway || nh->ipip_entry)
-               return 0;
-
-       nh->ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, ol_dev);
-       if (!nh->ipip_entry)
-               return -ENOENT;
+               return;
 
-       removing = !mlxsw_sp_ipip_netdev_ul_up(ol_dev);
+       nh->ipip_entry = ipip_entry;
+       removing = !mlxsw_sp_ipip_netdev_ul_up(ipip_entry->ol_dev);
        __mlxsw_sp_nexthop_neigh_update(nh, removing);
-       return 0;
+       mlxsw_sp_nexthop_rif_init(nh, &ipip_entry->ol_lb->common);
 }
 
 static void mlxsw_sp_nexthop_ipip_fini(struct mlxsw_sp *mlxsw_sp,
@@ -3421,21 +3418,21 @@ static int mlxsw_sp_nexthop4_type_init(struct mlxsw_sp *mlxsw_sp,
                                       struct mlxsw_sp_nexthop *nh,
                                       struct fib_nh *fib_nh)
 {
-       struct mlxsw_sp_router *router = mlxsw_sp->router;
+       const struct mlxsw_sp_ipip_ops *ipip_ops;
        struct net_device *dev = fib_nh->nh_dev;
-       enum mlxsw_sp_ipip_type ipipt;
+       struct mlxsw_sp_ipip_entry *ipip_entry;
        struct mlxsw_sp_rif *rif;
        int err;
 
-       if (mlxsw_sp_nexthop4_ipip_type(mlxsw_sp, fib_nh, &ipipt) &&
-           router->ipip_ops_arr[ipipt]->can_offload(mlxsw_sp, dev,
-                                                    MLXSW_SP_L3_PROTO_IPV4)) {
-               nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
-               err = mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, dev);
-               if (err)
-                       return err;
-               mlxsw_sp_nexthop_rif_init(nh, &nh->ipip_entry->ol_lb->common);
-               return 0;
+       ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
+       if (ipip_entry) {
+               ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
+               if (ipip_ops->can_offload(mlxsw_sp, dev,
+                                         MLXSW_SP_L3_PROTO_IPV4)) {
+                       nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
+                       mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
+                       return 0;
+               }
        }
 
        nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;
@@ -4712,21 +4709,21 @@ static int mlxsw_sp_nexthop6_type_init(struct mlxsw_sp *mlxsw_sp,
                                       struct mlxsw_sp_nexthop *nh,
                                       const struct rt6_info *rt)
 {
-       struct mlxsw_sp_router *router = mlxsw_sp->router;
+       const struct mlxsw_sp_ipip_ops *ipip_ops;
+       struct mlxsw_sp_ipip_entry *ipip_entry;
        struct net_device *dev = rt->dst.dev;
-       enum mlxsw_sp_ipip_type ipipt;
        struct mlxsw_sp_rif *rif;
        int err;
 
-       if (mlxsw_sp_nexthop6_ipip_type(mlxsw_sp, rt, &ipipt) &&
-           router->ipip_ops_arr[ipipt]->can_offload(mlxsw_sp, dev,
-                                                    MLXSW_SP_L3_PROTO_IPV6)) {
-               nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
-               err = mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, dev);
-               if (err)
-                       return err;
-               mlxsw_sp_nexthop_rif_init(nh, &nh->ipip_entry->ol_lb->common);
-               return 0;
+       ipip_entry = mlxsw_sp_ipip_entry_find_by_ol_dev(mlxsw_sp, dev);
+       if (ipip_entry) {
+               ipip_ops = mlxsw_sp->router->ipip_ops_arr[ipip_entry->ipipt];
+               if (ipip_ops->can_offload(mlxsw_sp, dev,
+                                         MLXSW_SP_L3_PROTO_IPV6)) {
+                       nh->type = MLXSW_SP_NEXTHOP_TYPE_IPIP;
+                       mlxsw_sp_nexthop_ipip_init(mlxsw_sp, nh, ipip_entry);
+                       return 0;
+               }
        }
 
        nh->type = MLXSW_SP_NEXTHOP_TYPE_ETH;