1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
2 /* Copyright (c) 2017-2018 Mellanox Technologies. All rights reserved */
4 #include <net/ip_tunnels.h>
5 #include <net/ip6_tunnel.h>
6 #include <net/inet_ecn.h>
8 #include "spectrum_ipip.h"
12 mlxsw_sp_ipip_netdev_parms4(const struct net_device *ol_dev)
14 struct ip_tunnel *tun = netdev_priv(ol_dev);
20 mlxsw_sp_ipip_netdev_parms6(const struct net_device *ol_dev)
22 struct ip6_tnl *tun = netdev_priv(ol_dev);
27 static bool mlxsw_sp_ipip_parms4_has_ikey(struct ip_tunnel_parm parms)
29 return !!(parms.i_flags & TUNNEL_KEY);
32 static bool mlxsw_sp_ipip_parms4_has_okey(struct ip_tunnel_parm parms)
34 return !!(parms.o_flags & TUNNEL_KEY);
37 static u32 mlxsw_sp_ipip_parms4_ikey(struct ip_tunnel_parm parms)
39 return mlxsw_sp_ipip_parms4_has_ikey(parms) ?
40 be32_to_cpu(parms.i_key) : 0;
43 static u32 mlxsw_sp_ipip_parms4_okey(struct ip_tunnel_parm parms)
45 return mlxsw_sp_ipip_parms4_has_okey(parms) ?
46 be32_to_cpu(parms.o_key) : 0;
49 static union mlxsw_sp_l3addr
50 mlxsw_sp_ipip_parms4_saddr(struct ip_tunnel_parm parms)
52 return (union mlxsw_sp_l3addr) { .addr4 = parms.iph.saddr };
55 static union mlxsw_sp_l3addr
56 mlxsw_sp_ipip_parms6_saddr(struct __ip6_tnl_parm parms)
58 return (union mlxsw_sp_l3addr) { .addr6 = parms.laddr };
61 static union mlxsw_sp_l3addr
62 mlxsw_sp_ipip_parms4_daddr(struct ip_tunnel_parm parms)
64 return (union mlxsw_sp_l3addr) { .addr4 = parms.iph.daddr };
67 static union mlxsw_sp_l3addr
68 mlxsw_sp_ipip_parms6_daddr(struct __ip6_tnl_parm parms)
70 return (union mlxsw_sp_l3addr) { .addr6 = parms.raddr };
74 mlxsw_sp_ipip_netdev_saddr(enum mlxsw_sp_l3proto proto,
75 const struct net_device *ol_dev)
77 struct ip_tunnel_parm parms4;
78 struct __ip6_tnl_parm parms6;
81 case MLXSW_SP_L3_PROTO_IPV4:
82 parms4 = mlxsw_sp_ipip_netdev_parms4(ol_dev);
83 return mlxsw_sp_ipip_parms4_saddr(parms4);
84 case MLXSW_SP_L3_PROTO_IPV6:
85 parms6 = mlxsw_sp_ipip_netdev_parms6(ol_dev);
86 return mlxsw_sp_ipip_parms6_saddr(parms6);
90 return (union mlxsw_sp_l3addr) {0};
93 static __be32 mlxsw_sp_ipip_netdev_daddr4(const struct net_device *ol_dev)
96 struct ip_tunnel_parm parms4 = mlxsw_sp_ipip_netdev_parms4(ol_dev);
98 return mlxsw_sp_ipip_parms4_daddr(parms4).addr4;
101 static union mlxsw_sp_l3addr
102 mlxsw_sp_ipip_netdev_daddr(enum mlxsw_sp_l3proto proto,
103 const struct net_device *ol_dev)
105 struct ip_tunnel_parm parms4;
106 struct __ip6_tnl_parm parms6;
109 case MLXSW_SP_L3_PROTO_IPV4:
110 parms4 = mlxsw_sp_ipip_netdev_parms4(ol_dev);
111 return mlxsw_sp_ipip_parms4_daddr(parms4);
112 case MLXSW_SP_L3_PROTO_IPV6:
113 parms6 = mlxsw_sp_ipip_netdev_parms6(ol_dev);
114 return mlxsw_sp_ipip_parms6_daddr(parms6);
118 return (union mlxsw_sp_l3addr) {0};
121 bool mlxsw_sp_l3addr_is_zero(union mlxsw_sp_l3addr addr)
123 union mlxsw_sp_l3addr naddr = {0};
125 return !memcmp(&addr, &naddr, sizeof(naddr));
129 mlxsw_sp_ipip_nexthop_update_gre4(struct mlxsw_sp *mlxsw_sp, u32 adj_index,
130 struct mlxsw_sp_ipip_entry *ipip_entry,
131 bool force, char *ratr_pl)
133 u16 rif_index = mlxsw_sp_ipip_lb_rif_index(ipip_entry->ol_lb);
134 __be32 daddr4 = mlxsw_sp_ipip_netdev_daddr4(ipip_entry->ol_dev);
135 enum mlxsw_reg_ratr_op op;
137 op = force ? MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY :
138 MLXSW_REG_RATR_OP_WRITE_WRITE_ENTRY_ON_ACTIVITY;
139 mlxsw_reg_ratr_pack(ratr_pl, op, true, MLXSW_REG_RATR_TYPE_IPIP,
140 adj_index, rif_index);
141 mlxsw_reg_ratr_ipip4_entry_pack(ratr_pl, be32_to_cpu(daddr4));
143 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(ratr), ratr_pl);
147 mlxsw_sp_ipip_decap_config_gre4(struct mlxsw_sp *mlxsw_sp,
148 struct mlxsw_sp_ipip_entry *ipip_entry,
151 u16 rif_index = mlxsw_sp_ipip_lb_rif_index(ipip_entry->ol_lb);
152 u16 ul_rif_id = mlxsw_sp_ipip_lb_ul_rif_id(ipip_entry->ol_lb);
153 char rtdp_pl[MLXSW_REG_RTDP_LEN];
154 struct ip_tunnel_parm parms;
155 unsigned int type_check;
160 parms = mlxsw_sp_ipip_netdev_parms4(ipip_entry->ol_dev);
161 has_ikey = mlxsw_sp_ipip_parms4_has_ikey(parms);
162 ikey = mlxsw_sp_ipip_parms4_ikey(parms);
164 mlxsw_reg_rtdp_pack(rtdp_pl, MLXSW_REG_RTDP_TYPE_IPIP, tunnel_index);
165 mlxsw_reg_rtdp_egress_router_interface_set(rtdp_pl, ul_rif_id);
167 type_check = has_ikey ?
168 MLXSW_REG_RTDP_IPIP_TYPE_CHECK_ALLOW_GRE_KEY :
169 MLXSW_REG_RTDP_IPIP_TYPE_CHECK_ALLOW_GRE;
171 /* Linux demuxes tunnels based on packet SIP (which must match tunnel
172 * remote IP). Thus configure decap so that it filters out packets that
173 * are not IPv4 or have the wrong SIP. IPIP_DECAP_ERROR trap is
174 * generated for packets that fail this criterion. Linux then handles
175 * such packets in slow path and generates ICMP destination unreachable.
177 daddr4 = be32_to_cpu(mlxsw_sp_ipip_netdev_daddr4(ipip_entry->ol_dev));
178 mlxsw_reg_rtdp_ipip4_pack(rtdp_pl, rif_index,
179 MLXSW_REG_RTDP_IPIP_SIP_CHECK_FILTER_IPV4,
180 type_check, has_ikey, daddr4, ikey);
182 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(rtdp), rtdp_pl);
185 static bool mlxsw_sp_ipip_tunnel_complete(enum mlxsw_sp_l3proto proto,
186 const struct net_device *ol_dev)
188 union mlxsw_sp_l3addr saddr = mlxsw_sp_ipip_netdev_saddr(proto, ol_dev);
189 union mlxsw_sp_l3addr daddr = mlxsw_sp_ipip_netdev_daddr(proto, ol_dev);
191 /* Tunnels with unset local or remote address are valid in Linux and
192 * used for lightweight tunnels (LWT) and Non-Broadcast Multi-Access
193 * (NBMA) tunnels. In principle these can be offloaded, but the driver
194 * currently doesn't support this. So punt.
196 return !mlxsw_sp_l3addr_is_zero(saddr) &&
197 !mlxsw_sp_l3addr_is_zero(daddr);
200 static bool mlxsw_sp_ipip_can_offload_gre4(const struct mlxsw_sp *mlxsw_sp,
201 const struct net_device *ol_dev)
203 struct ip_tunnel *tunnel = netdev_priv(ol_dev);
204 __be16 okflags = TUNNEL_KEY; /* We can't offload any other features. */
205 bool inherit_ttl = tunnel->parms.iph.ttl == 0;
206 bool inherit_tos = tunnel->parms.iph.tos & 0x1;
208 return (tunnel->parms.i_flags & ~okflags) == 0 &&
209 (tunnel->parms.o_flags & ~okflags) == 0 &&
210 inherit_ttl && inherit_tos &&
211 mlxsw_sp_ipip_tunnel_complete(MLXSW_SP_L3_PROTO_IPV4, ol_dev);
214 static struct mlxsw_sp_rif_ipip_lb_config
215 mlxsw_sp_ipip_ol_loopback_config_gre4(struct mlxsw_sp *mlxsw_sp,
216 const struct net_device *ol_dev)
218 struct ip_tunnel_parm parms = mlxsw_sp_ipip_netdev_parms4(ol_dev);
219 enum mlxsw_reg_ritr_loopback_ipip_type lb_ipipt;
221 lb_ipipt = mlxsw_sp_ipip_parms4_has_okey(parms) ?
222 MLXSW_REG_RITR_LOOPBACK_IPIP_TYPE_IP_IN_GRE_KEY_IN_IP :
223 MLXSW_REG_RITR_LOOPBACK_IPIP_TYPE_IP_IN_GRE_IN_IP;
224 return (struct mlxsw_sp_rif_ipip_lb_config){
225 .lb_ipipt = lb_ipipt,
226 .okey = mlxsw_sp_ipip_parms4_okey(parms),
227 .ul_protocol = MLXSW_SP_L3_PROTO_IPV4,
228 .saddr = mlxsw_sp_ipip_netdev_saddr(MLXSW_SP_L3_PROTO_IPV4,
234 mlxsw_sp_ipip_ol_netdev_change_gre4(struct mlxsw_sp *mlxsw_sp,
235 struct mlxsw_sp_ipip_entry *ipip_entry,
236 struct netlink_ext_ack *extack)
238 union mlxsw_sp_l3addr old_saddr, new_saddr;
239 union mlxsw_sp_l3addr old_daddr, new_daddr;
240 struct ip_tunnel_parm new_parms;
241 bool update_tunnel = false;
242 bool update_decap = false;
243 bool update_nhs = false;
246 new_parms = mlxsw_sp_ipip_netdev_parms4(ipip_entry->ol_dev);
248 new_saddr = mlxsw_sp_ipip_parms4_saddr(new_parms);
249 old_saddr = mlxsw_sp_ipip_parms4_saddr(ipip_entry->parms4);
250 new_daddr = mlxsw_sp_ipip_parms4_daddr(new_parms);
251 old_daddr = mlxsw_sp_ipip_parms4_daddr(ipip_entry->parms4);
253 if (!mlxsw_sp_l3addr_eq(&new_saddr, &old_saddr)) {
254 u16 ul_tb_id = mlxsw_sp_ipip_dev_ul_tb_id(ipip_entry->ol_dev);
256 /* Since the local address has changed, if there is another
257 * tunnel with a matching saddr, both need to be demoted.
259 if (mlxsw_sp_ipip_demote_tunnel_by_saddr(mlxsw_sp,
260 MLXSW_SP_L3_PROTO_IPV4,
263 mlxsw_sp_ipip_entry_demote_tunnel(mlxsw_sp, ipip_entry);
267 update_tunnel = true;
268 } else if ((mlxsw_sp_ipip_parms4_okey(ipip_entry->parms4) !=
269 mlxsw_sp_ipip_parms4_okey(new_parms)) ||
270 ipip_entry->parms4.link != new_parms.link) {
271 update_tunnel = true;
272 } else if (!mlxsw_sp_l3addr_eq(&new_daddr, &old_daddr)) {
274 } else if (mlxsw_sp_ipip_parms4_ikey(ipip_entry->parms4) !=
275 mlxsw_sp_ipip_parms4_ikey(new_parms)) {
280 err = __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
284 err = __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
287 else if (update_decap)
288 err = __mlxsw_sp_ipip_entry_update_tunnel(mlxsw_sp, ipip_entry,
292 ipip_entry->parms4 = new_parms;
296 static const struct mlxsw_sp_ipip_ops mlxsw_sp_ipip_gre4_ops = {
297 .dev_type = ARPHRD_IPGRE,
298 .ul_proto = MLXSW_SP_L3_PROTO_IPV4,
299 .nexthop_update = mlxsw_sp_ipip_nexthop_update_gre4,
300 .decap_config = mlxsw_sp_ipip_decap_config_gre4,
301 .can_offload = mlxsw_sp_ipip_can_offload_gre4,
302 .ol_loopback_config = mlxsw_sp_ipip_ol_loopback_config_gre4,
303 .ol_netdev_change = mlxsw_sp_ipip_ol_netdev_change_gre4,
306 const struct mlxsw_sp_ipip_ops *mlxsw_sp_ipip_ops_arr[] = {
307 [MLXSW_SP_IPIP_TYPE_GRE4] = &mlxsw_sp_ipip_gre4_ops,
310 static int mlxsw_sp_ipip_ecn_encap_init_one(struct mlxsw_sp *mlxsw_sp,
311 u8 inner_ecn, u8 outer_ecn)
313 char tieem_pl[MLXSW_REG_TIEEM_LEN];
315 mlxsw_reg_tieem_pack(tieem_pl, inner_ecn, outer_ecn);
316 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tieem), tieem_pl);
319 int mlxsw_sp_ipip_ecn_encap_init(struct mlxsw_sp *mlxsw_sp)
323 /* Iterate over inner ECN values */
324 for (i = INET_ECN_NOT_ECT; i <= INET_ECN_CE; i++) {
325 u8 outer_ecn = INET_ECN_encapsulate(0, i);
328 err = mlxsw_sp_ipip_ecn_encap_init_one(mlxsw_sp, i, outer_ecn);
336 static int mlxsw_sp_ipip_ecn_decap_init_one(struct mlxsw_sp *mlxsw_sp,
337 u8 inner_ecn, u8 outer_ecn)
339 char tidem_pl[MLXSW_REG_TIDEM_LEN];
343 new_inner_ecn = mlxsw_sp_tunnel_ecn_decap(outer_ecn, inner_ecn,
345 mlxsw_reg_tidem_pack(tidem_pl, outer_ecn, inner_ecn, new_inner_ecn,
346 trap_en, trap_en ? MLXSW_TRAP_ID_DECAP_ECN0 : 0);
347 return mlxsw_reg_write(mlxsw_sp->core, MLXSW_REG(tidem), tidem_pl);
350 int mlxsw_sp_ipip_ecn_decap_init(struct mlxsw_sp *mlxsw_sp)
354 /* Iterate over inner ECN values */
355 for (i = INET_ECN_NOT_ECT; i <= INET_ECN_CE; i++) {
356 /* Iterate over outer ECN values */
357 for (j = INET_ECN_NOT_ECT; j <= INET_ECN_CE; j++) {
358 err = mlxsw_sp_ipip_ecn_decap_init_one(mlxsw_sp, i, j);