mlxsw: spectrum_router: Support GRE tunnels
authorPetr Machata <petrm@mellanox.com>
Sat, 2 Sep 2017 21:49:29 +0000 (23:49 +0200)
committerDavid S. Miller <davem@davemloft.net>
Mon, 4 Sep 2017 03:23:26 +0000 (20:23 -0700)
This patch introduces callbacks and tunnel type to offload GRE 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_ipip.c
drivers/net/ethernet/mellanox/mlxsw/spectrum_ipip.h
drivers/net/ethernet/mellanox/mlxsw/spectrum_router.c
drivers/net/ethernet/mellanox/mlxsw/spectrum_router.h

index 20b0b969d41db1ccc09d66da26fe0465d90d304d..702fe945227c213d7e5f4d75815fb600825db5e3 100644 (file)
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <net/ip_tunnels.h>
+
 #include "spectrum_ipip.h"
 
+static bool
+mlxsw_sp_ipip_netdev_has_ikey(const struct net_device *ol_dev)
+{
+       struct ip_tunnel *tun = netdev_priv(ol_dev);
+
+       return !!(tun->parms.i_flags & TUNNEL_KEY);
+}
+
+static bool
+mlxsw_sp_ipip_netdev_has_okey(const struct net_device *ol_dev)
+{
+       struct ip_tunnel *tun = netdev_priv(ol_dev);
+
+       return !!(tun->parms.o_flags & TUNNEL_KEY);
+}
+
+static u32 mlxsw_sp_ipip_netdev_ikey(const struct net_device *ol_dev)
+{
+       struct ip_tunnel *tun = netdev_priv(ol_dev);
+
+       return mlxsw_sp_ipip_netdev_has_ikey(ol_dev) ?
+               be32_to_cpu(tun->parms.i_key) : 0;
+}
+
+static u32 mlxsw_sp_ipip_netdev_okey(const struct net_device *ol_dev)
+{
+       struct ip_tunnel *tun = netdev_priv(ol_dev);
+
+       return mlxsw_sp_ipip_netdev_has_okey(ol_dev) ?
+               be32_to_cpu(tun->parms.o_key) : 0;
+}
+
+static int
+mlxsw_sp_ipip_nexthop_update_gre4(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
+                                 struct mlxsw_sp_ipip_entry *ipip_entry)
+{
+       u16 rif_index = mlxsw_sp_ipip_lb_rif_index(ipip_entry->ol_lb);
+       __be32 daddr4 = mlxsw_sp_ipip_netdev_daddr4(ipip_entry->ol_dev);
+       char ratr_pl[MLXSW_REG_RATR_LEN];
+
+       mlxsw_reg_ratr_pack(ratr_pl, MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY,
+                           true, MLXSW_REG_RATR_TYPE_IPIP,
+                           adj_index, rif_index);
+       mlxsw_reg_ratr_ipip4_entry_pack(ratr_pl, be32_to_cpu(daddr4));
+
+       return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
+}
+
+static int
+mlxsw_sp_ipip_fib_entry_op_gre4_rtdp(struct mlxsw_sp *mlxsw_sp,
+                                    u32 tunnel_index,
+                                    struct mlxsw_sp_ipip_entry *ipip_entry)
+{
+       bool has_ikey = mlxsw_sp_ipip_netdev_has_ikey(ipip_entry->ol_dev);
+       u16 rif_index = mlxsw_sp_ipip_lb_rif_index(ipip_entry->ol_lb);
+       u32 ikey = mlxsw_sp_ipip_netdev_ikey(ipip_entry->ol_dev);
+       char rtdp_pl[MLXSW_REG_RTDP_LEN];
+       unsigned int type_check;
+       u32 daddr4;
+
+       mlxsw_reg_rtdp_pack(rtdp_pl, MLXSW_REG_RTDP_TYPE_IPIP, tunnel_index);
+
+       type_check = has_ikey ?
+               MLXSW_REG_RTDP_IPIP_TYPE_CHECK_ALLOW_GRE_KEY :
+               MLXSW_REG_RTDP_IPIP_TYPE_CHECK_ALLOW_GRE;
+
+       /* Linux demuxes tunnels based on packet SIP (which must match tunnel
+        * remote IP). Thus configure decap so that it filters out packets that
+        * are not IPv4 or have the wrong SIP. IPIP_DECAP_ERROR trap is
+        * generated for packets that fail this criterion. Linux then handles
+        * such packets in slow path and generates ICMP destination unreachable.
+        */
+       daddr4 = be32_to_cpu(mlxsw_sp_ipip_netdev_daddr4(ipip_entry->ol_dev));
+       mlxsw_reg_rtdp_ipip4_pack(rtdp_pl, rif_index,
+                                 MLXSW_REG_RTDP_IPIP_SIP_CHECK_FILTER_IPV4,
+                                 type_check, has_ikey, daddr4, ikey);
+
+       return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rtdp), rtdp_pl);
+}
+
+static int
+mlxsw_sp_ipip_fib_entry_op_gre4_ralue(struct mlxsw_sp *mlxsw_sp,
+                                     u32 dip, u8 prefix_len, u16 ul_vr_id,
+                                     enum mlxsw_reg_ralue_op op,
+                                     u32 tunnel_index)
+{
+       char ralue_pl[MLXSW_REG_RALUE_LEN];
+
+       mlxsw_reg_ralue_pack4(ralue_pl, MLXSW_REG_RALXX_PROTOCOL_IPV4, op,
+                             ul_vr_id, prefix_len, dip);
+       mlxsw_reg_ralue_act_ip2me_tun_pack(ralue_pl, tunnel_index);
+       return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ralue), ralue_pl);
+}
+
+static int mlxsw_sp_ipip_fib_entry_op_gre4(struct mlxsw_sp *mlxsw_sp,
+                                       struct mlxsw_sp_ipip_entry *ipip_entry,
+                                       enum mlxsw_reg_ralue_op op,
+                                       u32 tunnel_index)
+{
+       u16 ul_vr_id = mlxsw_sp_ipip_lb_ul_vr_id(ipip_entry->ol_lb);
+       __be32 dip;
+       int err;
+
+       err = mlxsw_sp_ipip_fib_entry_op_gre4_rtdp(mlxsw_sp, tunnel_index,
+                                                  ipip_entry);
+       if (err)
+               return err;
+
+       dip = mlxsw_sp_ipip_netdev_saddr(MLXSW_SP_L3_PROTO_IPV4,
+                                        ipip_entry->ol_dev).addr4;
+       return mlxsw_sp_ipip_fib_entry_op_gre4_ralue(mlxsw_sp, be32_to_cpu(dip),
+                                                    32, ul_vr_id, op,
+                                                    tunnel_index);
+}
+
+static bool mlxsw_sp_ipip_tunnel_complete(enum mlxsw_sp_l3proto proto,
+                                         const struct net_device *ol_dev)
+{
+       union mlxsw_sp_l3addr saddr = mlxsw_sp_ipip_netdev_saddr(proto, ol_dev);
+       union mlxsw_sp_l3addr daddr = mlxsw_sp_ipip_netdev_daddr(proto, ol_dev);
+       union mlxsw_sp_l3addr naddr = {0};
+
+       /* Tunnels with unset local or remote address are valid in Linux and
+        * used for lightweight tunnels (LWT) and Non-Broadcast Multi-Access
+        * (NBMA) tunnels. In principle these can be offloaded, but the driver
+        * currently doesn't support this. So punt.
+        */
+       return memcmp(&saddr, &naddr, sizeof(naddr)) &&
+              memcmp(&daddr, &naddr, sizeof(naddr));
+}
+
+static bool mlxsw_sp_ipip_can_offload_gre4(const struct mlxsw_sp *mlxsw_sp,
+                                          const struct net_device *ol_dev,
+                                          enum mlxsw_sp_l3proto ol_proto)
+{
+       struct ip_tunnel *tunnel = netdev_priv(ol_dev);
+       __be16 okflags = TUNNEL_KEY; /* We can't offload any other features. */
+       bool inherit_ttl = tunnel->parms.iph.ttl == 0;
+       bool inherit_tos = tunnel->parms.iph.tos & 0x1;
+
+       return (tunnel->parms.i_flags & ~okflags) == 0 &&
+              (tunnel->parms.o_flags & ~okflags) == 0 &&
+              inherit_ttl && inherit_tos &&
+              mlxsw_sp_ipip_tunnel_complete(MLXSW_SP_L3_PROTO_IPV4, ol_dev);
+}
+
+static struct mlxsw_sp_rif_ipip_lb_config
+mlxsw_sp_ipip_ol_loopback_config_gre4(struct mlxsw_sp *mlxsw_sp,
+                                     const struct net_device *ol_dev)
+{
+       enum mlxsw_reg_ritr_loopback_ipip_type lb_ipipt;
+
+       lb_ipipt = mlxsw_sp_ipip_netdev_has_okey(ol_dev) ?
+               MLXSW_REG_RITR_LOOPBACK_IPIP_TYPE_IP_IN_GRE_KEY_IN_IP :
+               MLXSW_REG_RITR_LOOPBACK_IPIP_TYPE_IP_IN_GRE_IN_IP;
+       return (struct mlxsw_sp_rif_ipip_lb_config){
+               .lb_ipipt = lb_ipipt,
+               .okey = mlxsw_sp_ipip_netdev_okey(ol_dev),
+               .ul_protocol = MLXSW_SP_L3_PROTO_IPV4,
+               .saddr = mlxsw_sp_ipip_netdev_saddr(MLXSW_SP_L3_PROTO_IPV4,
+                                                   ol_dev),
+       };
+}
+
+static const struct mlxsw_sp_ipip_ops mlxsw_sp_ipip_gre4_ops = {
+       .dev_type = ARPHRD_IPGRE,
+       .ul_proto = MLXSW_SP_L3_PROTO_IPV4,
+       .nexthop_update = mlxsw_sp_ipip_nexthop_update_gre4,
+       .fib_entry_op = mlxsw_sp_ipip_fib_entry_op_gre4,
+       .can_offload = mlxsw_sp_ipip_can_offload_gre4,
+       .ol_loopback_config = mlxsw_sp_ipip_ol_loopback_config_gre4,
+};
+
 const struct mlxsw_sp_ipip_ops *mlxsw_sp_ipip_ops_arr[] = {
+       [MLXSW_SP_IPIP_TYPE_GRE4] = &mlxsw_sp_ipip_gre4_ops,
 };
index cd986bde856aa6de7bda8efc9155c19605fa86b7..1c2db831d83b1c4f05a0d469e08a89e2ccf86c8f 100644 (file)
@@ -39,6 +39,7 @@
 #include <net/ip_fib.h>
 
 enum mlxsw_sp_ipip_type {
+       MLXSW_SP_IPIP_TYPE_GRE4,
        MLXSW_SP_IPIP_TYPE_MAX,
 };
 
index 6068eea8152fe676dfb8ab0bf0c1deedceaa4274..f0fb898533fbdf64c56594d7984ddc9ff874a6e9 100644 (file)
@@ -1020,6 +1020,32 @@ mlxsw_sp_ipip_netdev_saddr(enum mlxsw_sp_l3proto proto,
        };
 }
 
+__be32 mlxsw_sp_ipip_netdev_daddr4(const struct net_device *ol_dev)
+{
+       struct ip_tunnel *tun = netdev_priv(ol_dev);
+
+       return tun->parms.iph.daddr;
+}
+
+union mlxsw_sp_l3addr
+mlxsw_sp_ipip_netdev_daddr(enum mlxsw_sp_l3proto proto,
+                          const struct net_device *ol_dev)
+{
+       switch (proto) {
+       case MLXSW_SP_L3_PROTO_IPV4:
+               return (union mlxsw_sp_l3addr) {
+                       .addr4 = mlxsw_sp_ipip_netdev_daddr4(ol_dev),
+               };
+       case MLXSW_SP_L3_PROTO_IPV6:
+               break;
+       };
+
+       WARN_ON(1);
+       return (union mlxsw_sp_l3addr) {
+               .addr4 = 0,
+       };
+}
+
 static bool mlxsw_sp_l3addr_eq(const union mlxsw_sp_l3addr *addr1,
                               const union mlxsw_sp_l3addr *addr2)
 {
index 96324769351ae20753120aab635a0a5706443648..345fcc4f38e9a33f08a890a73af1a19b1728eb9a 100644 (file)
@@ -103,5 +103,9 @@ bool mlxsw_sp_neigh_ipv6_ignore(struct mlxsw_sp_neigh_entry *neigh_entry);
 union mlxsw_sp_l3addr
 mlxsw_sp_ipip_netdev_saddr(enum mlxsw_sp_l3proto proto,
                           const struct net_device *ol_dev);
+union mlxsw_sp_l3addr
+mlxsw_sp_ipip_netdev_daddr(enum mlxsw_sp_l3proto proto,
+                          const struct net_device *ol_dev);
+__be32 mlxsw_sp_ipip_netdev_daddr4(const struct net_device *ol_dev);
 
 #endif /* _MLXSW_ROUTER_H_*/