drivers/net/ethernet/mellanox/mlx5/core/en_tc.c

   1 /*
   2  * Copyright (c) 2016, Mellanox Technologies. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32
  33 #include <net/flow_dissector.h>
  34 #include <net/sch_generic.h>
  35 #include <net/pkt_cls.h>
  36 #include <net/tc_act/tc_gact.h>
  37 #include <net/tc_act/tc_skbedit.h>
  38 #include <linux/mlx5/fs.h>
  39 #include <linux/mlx5/device.h>
  40 #include <linux/rhashtable.h>
  41 #include <net/switchdev.h>
  42 #include <net/tc_act/tc_mirred.h>
  43 #include <net/tc_act/tc_vlan.h>
  44 #include <net/tc_act/tc_tunnel_key.h>
  45 #include <net/vxlan.h>
  46 #include "en.h"
  47 #include "en_tc.h"
  48 #include "eswitch.h"
  49 #include "vxlan.h"
  50
  51 enum {
  52         MLX5E_TC_FLOW_ESWITCH   = BIT(0),
  53 };
  54
  55 struct mlx5e_tc_flow {
  56         struct rhash_head       node;
  57         u64                     cookie;
  58         u8                      flags;
  59         struct mlx5_flow_handle *rule;
  60         struct list_head        encap; /* flows sharing the same encap */
  61         struct mlx5_esw_flow_attr *attr;
  62 };
  63
  64 enum {
  65         MLX5_HEADER_TYPE_VXLAN = 0x0,
  66         MLX5_HEADER_TYPE_NVGRE = 0x1,
  67 };
  68
  69 #define MLX5E_TC_TABLE_NUM_ENTRIES 1024
  70 #define MLX5E_TC_TABLE_NUM_GROUPS 4
  71
  72 static struct mlx5_flow_handle *
  73 mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv,
  74                       struct mlx5_flow_spec *spec,
  75                       u32 action, u32 flow_tag)
  76 {
  77         struct mlx5_core_dev *dev = priv->mdev;
  78         struct mlx5_flow_destination dest = { 0 };
  79         struct mlx5_flow_act flow_act = {
  80                 .action = action,
  81                 .flow_tag = flow_tag,
  82                 .encap_id = 0,
  83         };
  84         struct mlx5_fc *counter = NULL;
  85         struct mlx5_flow_handle *rule;
  86         bool table_created = false;
  87
  88         if (action & MLX5_FLOW_CONTEXT_ACTION_FWD_DEST) {
  89                 dest.type = MLX5_FLOW_DESTINATION_TYPE_FLOW_TABLE;
  90                 dest.ft = priv->fs.vlan.ft.t;
  91         } else if (action & MLX5_FLOW_CONTEXT_ACTION_COUNT) {
  92                 counter = mlx5_fc_create(dev, true);
  93                 if (IS_ERR(counter))
  94                         return ERR_CAST(counter);
  95
  96                 dest.type = MLX5_FLOW_DESTINATION_TYPE_COUNTER;
  97                 dest.counter = counter;
  98         }
  99
 100         if (IS_ERR_OR_NULL(priv->fs.tc.t)) {
 101                 priv->fs.tc.t =
 102                         mlx5_create_auto_grouped_flow_table(priv->fs.ns,
 103                                                             MLX5E_TC_PRIO,
 104                                                             MLX5E_TC_TABLE_NUM_ENTRIES,
 105                                                             MLX5E_TC_TABLE_NUM_GROUPS,
 106                                                             0, 0);
 107                 if (IS_ERR(priv->fs.tc.t)) {
 108                         netdev_err(priv->netdev,
 109                                    "Failed to create tc offload table\n");
 110                         rule = ERR_CAST(priv->fs.tc.t);
 111                         goto err_create_ft;
 112                 }
 113
 114                 table_created = true;
 115         }
 116
 117         spec->match_criteria_enable = MLX5_MATCH_OUTER_HEADERS;
 118         rule = mlx5_add_flow_rules(priv->fs.tc.t, spec, &flow_act, &dest, 1);
 119
 120         if (IS_ERR(rule))
 121                 goto err_add_rule;
 122
 123         return rule;
 124
 125 err_add_rule:
 126         if (table_created) {
 127                 mlx5_destroy_flow_table(priv->fs.tc.t);
 128                 priv->fs.tc.t = NULL;
 129         }
 130 err_create_ft:
 131         mlx5_fc_destroy(dev, counter);
 132
 133         return rule;
 134 }
 135
 136 static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv,
 137                                   struct mlx5e_tc_flow *flow)
 138 {
 139         struct mlx5_fc *counter = NULL;
 140
 141         if (!IS_ERR(flow->rule)) {
 142                 counter = mlx5_flow_rule_counter(flow->rule);
 143                 mlx5_del_flow_rules(flow->rule);
 144                 mlx5_fc_destroy(priv->mdev, counter);
 145         }
 146
 147         if (!mlx5e_tc_num_filters(priv) && (priv->fs.tc.t)) {
 148                 mlx5_destroy_flow_table(priv->fs.tc.t);
 149                 priv->fs.tc.t = NULL;
 150         }
 151 }
 152
 153 static struct mlx5_flow_handle *
 154 mlx5e_tc_add_fdb_flow(struct mlx5e_priv *priv,
 155                       struct mlx5_flow_spec *spec,
 156                       struct mlx5_esw_flow_attr *attr)
 157 {
 158         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 159         int err;
 160
 161         err = mlx5_eswitch_add_vlan_action(esw, attr);
 162         if (err)
 163                 return ERR_PTR(err);
 164
 165         return mlx5_eswitch_add_offloaded_rule(esw, spec, attr);
 166 }
 167
 168 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
 169                                struct mlx5e_tc_flow *flow);
 170
 171 static void mlx5e_tc_del_fdb_flow(struct mlx5e_priv *priv,
 172                                   struct mlx5e_tc_flow *flow)
 173 {
 174         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 175
 176         mlx5_eswitch_del_offloaded_rule(esw, flow->rule, flow->attr);
 177
 178         mlx5_eswitch_del_vlan_action(esw, flow->attr);
 179
 180         if (flow->attr->action & MLX5_FLOW_CONTEXT_ACTION_ENCAP)
 181                 mlx5e_detach_encap(priv, flow);
 182 }
 183
 184 static void mlx5e_detach_encap(struct mlx5e_priv *priv,
 185                                struct mlx5e_tc_flow *flow)
 186 {
 187         struct list_head *next = flow->encap.next;
 188
 189         list_del(&flow->encap);
 190         if (list_empty(next)) {
 191                 struct mlx5_encap_entry *e;
 192
 193                 e = list_entry(next, struct mlx5_encap_entry, flows);
 194                 if (e->n) {
 195                         mlx5_encap_dealloc(priv->mdev, e->encap_id);
 196                         neigh_release(e->n);
 197                 }
 198                 hlist_del_rcu(&e->encap_hlist);
 199                 kfree(e);
 200         }
 201 }
 202
 203 /* we get here also when setting rule to the FW failed, etc. It means that the
 204  * flow rule itself might not exist, but some offloading related to the actions
 205  * should be cleaned.
 206  */
 207 static void mlx5e_tc_del_flow(struct mlx5e_priv *priv,
 208                               struct mlx5e_tc_flow *flow)
 209 {
 210         if (flow->flags & MLX5E_TC_FLOW_ESWITCH)
 211                 mlx5e_tc_del_fdb_flow(priv, flow);
 212         else
 213                 mlx5e_tc_del_nic_flow(priv, flow);
 214 }
 215
 216 static void parse_vxlan_attr(struct mlx5_flow_spec *spec,
 217                              struct tc_cls_flower_offload *f)
 218 {
 219         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 220                                        outer_headers);
 221         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 222                                        outer_headers);
 223         void *misc_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 224                                     misc_parameters);
 225         void *misc_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 226                                     misc_parameters);
 227
 228         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ip_protocol);
 229         MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol, IPPROTO_UDP);
 230
 231         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID)) {
 232                 struct flow_dissector_key_keyid *key =
 233                         skb_flow_dissector_target(f->dissector,
 234                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
 235                                                   f->key);
 236                 struct flow_dissector_key_keyid *mask =
 237                         skb_flow_dissector_target(f->dissector,
 238                                                   FLOW_DISSECTOR_KEY_ENC_KEYID,
 239                                                   f->mask);
 240                 MLX5_SET(fte_match_set_misc, misc_c, vxlan_vni,
 241                          be32_to_cpu(mask->keyid));
 242                 MLX5_SET(fte_match_set_misc, misc_v, vxlan_vni,
 243                          be32_to_cpu(key->keyid));
 244         }
 245 }
 246
 247 static int parse_tunnel_attr(struct mlx5e_priv *priv,
 248                              struct mlx5_flow_spec *spec,
 249                              struct tc_cls_flower_offload *f)
 250 {
 251         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 252                                        outer_headers);
 253         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 254                                        outer_headers);
 255
 256         struct flow_dissector_key_control *enc_control =
 257                 skb_flow_dissector_target(f->dissector,
 258                                           FLOW_DISSECTOR_KEY_ENC_CONTROL,
 259                                           f->key);
 260
 261         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) {
 262                 struct flow_dissector_key_ports *key =
 263                         skb_flow_dissector_target(f->dissector,
 264                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
 265                                                   f->key);
 266                 struct flow_dissector_key_ports *mask =
 267                         skb_flow_dissector_target(f->dissector,
 268                                                   FLOW_DISSECTOR_KEY_ENC_PORTS,
 269                                                   f->mask);
 270                 struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 271                 struct net_device *up_dev = mlx5_eswitch_get_uplink_netdev(esw);
 272                 struct mlx5e_priv *up_priv = netdev_priv(up_dev);
 273
 274                 /* Full udp dst port must be given */
 275                 if (memchr_inv(&mask->dst, 0xff, sizeof(mask->dst)))
 276                         goto vxlan_match_offload_err;
 277
 278                 if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->dst)) &&
 279                     MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap))
 280                         parse_vxlan_attr(spec, f);
 281                 else {
 282                         netdev_warn(priv->netdev,
 283                                     "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->dst));
 284                         return -EOPNOTSUPP;
 285                 }
 286
 287                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 288                          udp_dport, ntohs(mask->dst));
 289                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 290                          udp_dport, ntohs(key->dst));
 291
 292                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 293                          udp_sport, ntohs(mask->src));
 294                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 295                          udp_sport, ntohs(key->src));
 296         } else { /* udp dst port must be given */
 297 vxlan_match_offload_err:
 298                 netdev_warn(priv->netdev,
 299                             "IP tunnel decap offload supported only for vxlan, must set UDP dport\n");
 300                 return -EOPNOTSUPP;
 301         }
 302
 303         if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
 304                 struct flow_dissector_key_ipv4_addrs *key =
 305                         skb_flow_dissector_target(f->dissector,
 306                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
 307                                                   f->key);
 308                 struct flow_dissector_key_ipv4_addrs *mask =
 309                         skb_flow_dissector_target(f->dissector,
 310                                                   FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS,
 311                                                   f->mask);
 312                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 313                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
 314                          ntohl(mask->src));
 315                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 316                          src_ipv4_src_ipv6.ipv4_layout.ipv4,
 317                          ntohl(key->src));
 318
 319                 MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 320                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
 321                          ntohl(mask->dst));
 322                 MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 323                          dst_ipv4_dst_ipv6.ipv4_layout.ipv4,
 324                          ntohl(key->dst));
 325
 326                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
 327                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IP);
 328         } else if (enc_control->addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
 329                 struct flow_dissector_key_ipv6_addrs *key =
 330                         skb_flow_dissector_target(f->dissector,
 331                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
 332                                                   f->key);
 333                 struct flow_dissector_key_ipv6_addrs *mask =
 334                         skb_flow_dissector_target(f->dissector,
 335                                                   FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS,
 336                                                   f->mask);
 337
 338                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 339                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 340                        &mask->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 341                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 342                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 343                        &key->src, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 344
 345                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 346                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 347                        &mask->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 348                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 349                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 350                        &key->dst, MLX5_FLD_SZ_BYTES(ipv6_layout, ipv6));
 351
 352                 MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, ethertype);
 353                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype, ETH_P_IPV6);
 354         }
 355
 356         /* Enforce DMAC when offloading incoming tunneled flows.
 357          * Flow counters require a match on the DMAC.
 358          */
 359         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_47_16);
 360         MLX5_SET_TO_ONES(fte_match_set_lyr_2_4, headers_c, dmac_15_0);
 361         ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 362                                      dmac_47_16), priv->netdev->dev_addr);
 363
 364         /* let software handle IP fragments */
 365         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
 366         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag, 0);
 367
 368         return 0;
 369 }
 370
 371 static int __parse_cls_flower(struct mlx5e_priv *priv,
 372                               struct mlx5_flow_spec *spec,
 373                               struct tc_cls_flower_offload *f,
 374                               u8 *min_inline)
 375 {
 376         void *headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 377                                        outer_headers);
 378         void *headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 379                                        outer_headers);
 380         u16 addr_type = 0;
 381         u8 ip_proto = 0;
 382
 383         *min_inline = MLX5_INLINE_MODE_L2;
 384
 385         if (f->dissector->used_keys &
 386             ~(BIT(FLOW_DISSECTOR_KEY_CONTROL) |
 387               BIT(FLOW_DISSECTOR_KEY_BASIC) |
 388               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
 389               BIT(FLOW_DISSECTOR_KEY_VLAN) |
 390               BIT(FLOW_DISSECTOR_KEY_IPV4_ADDRS) |
 391               BIT(FLOW_DISSECTOR_KEY_IPV6_ADDRS) |
 392               BIT(FLOW_DISSECTOR_KEY_PORTS) |
 393               BIT(FLOW_DISSECTOR_KEY_ENC_KEYID) |
 394               BIT(FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) |
 395               BIT(FLOW_DISSECTOR_KEY_ENC_IPV6_ADDRS) |
 396               BIT(FLOW_DISSECTOR_KEY_ENC_PORTS) |
 397               BIT(FLOW_DISSECTOR_KEY_ENC_CONTROL))) {
 398                 netdev_warn(priv->netdev, "Unsupported key used: 0x%x\n",
 399                             f->dissector->used_keys);
 400                 return -EOPNOTSUPP;
 401         }
 402
 403         if ((dissector_uses_key(f->dissector,
 404                                 FLOW_DISSECTOR_KEY_ENC_IPV4_ADDRS) ||
 405              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_KEYID) ||
 406              dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_PORTS)) &&
 407             dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ENC_CONTROL)) {
 408                 struct flow_dissector_key_control *key =
 409                         skb_flow_dissector_target(f->dissector,
 410                                                   FLOW_DISSECTOR_KEY_ENC_CONTROL,
 411                                                   f->key);
 412                 switch (key->addr_type) {
 413                 case FLOW_DISSECTOR_KEY_IPV4_ADDRS:
 414                 case FLOW_DISSECTOR_KEY_IPV6_ADDRS:
 415                         if (parse_tunnel_attr(priv, spec, f))
 416                                 return -EOPNOTSUPP;
 417                         break;
 418                 default:
 419                         return -EOPNOTSUPP;
 420                 }
 421
 422                 /* In decap flow, header pointers should point to the inner
 423                  * headers, outer header were already set by parse_tunnel_attr
 424                  */
 425                 headers_c = MLX5_ADDR_OF(fte_match_param, spec->match_criteria,
 426                                          inner_headers);
 427                 headers_v = MLX5_ADDR_OF(fte_match_param, spec->match_value,
 428                                          inner_headers);
 429         }
 430
 431         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_CONTROL)) {
 432                 struct flow_dissector_key_control *key =
 433                         skb_flow_dissector_target(f->dissector,
 434                                                   FLOW_DISSECTOR_KEY_CONTROL,
 435                                                   f->key);
 436
 437                 struct flow_dissector_key_control *mask =
 438                         skb_flow_dissector_target(f->dissector,
 439                                                   FLOW_DISSECTOR_KEY_CONTROL,
 440                                                   f->mask);
 441                 addr_type = key->addr_type;
 442
 443                 if (mask->flags & FLOW_DIS_IS_FRAGMENT) {
 444                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, frag, 1);
 445                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, frag,
 446                                  key->flags & FLOW_DIS_IS_FRAGMENT);
 447
 448                         /* the HW doesn't need L3 inline to match on frag=no */
 449                         if (key->flags & FLOW_DIS_IS_FRAGMENT)
 450                                 *min_inline = MLX5_INLINE_MODE_IP;
 451                 }
 452         }
 453
 454         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
 455                 struct flow_dissector_key_basic *key =
 456                         skb_flow_dissector_target(f->dissector,
 457                                                   FLOW_DISSECTOR_KEY_BASIC,
 458                                                   f->key);
 459                 struct flow_dissector_key_basic *mask =
 460                         skb_flow_dissector_target(f->dissector,
 461                                                   FLOW_DISSECTOR_KEY_BASIC,
 462                                                   f->mask);
 463                 ip_proto = key->ip_proto;
 464
 465                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ethertype,
 466                          ntohs(mask->n_proto));
 467                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ethertype,
 468                          ntohs(key->n_proto));
 469
 470                 MLX5_SET(fte_match_set_lyr_2_4, headers_c, ip_protocol,
 471                          mask->ip_proto);
 472                 MLX5_SET(fte_match_set_lyr_2_4, headers_v, ip_protocol,
 473                          key->ip_proto);
 474
 475                 if (mask->ip_proto)
 476                         *min_inline = MLX5_INLINE_MODE_IP;
 477         }
 478
 479         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
 480                 struct flow_dissector_key_eth_addrs *key =
 481                         skb_flow_dissector_target(f->dissector,
 482                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
 483                                                   f->key);
 484                 struct flow_dissector_key_eth_addrs *mask =
 485                         skb_flow_dissector_target(f->dissector,
 486                                                   FLOW_DISSECTOR_KEY_ETH_ADDRS,
 487                                                   f->mask);
 488
 489                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 490                                              dmac_47_16),
 491                                 mask->dst);
 492                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 493                                              dmac_47_16),
 494                                 key->dst);
 495
 496                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 497                                              smac_47_16),
 498                                 mask->src);
 499                 ether_addr_copy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 500                                              smac_47_16),
 501                                 key->src);
 502         }
 503
 504         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
 505                 struct flow_dissector_key_vlan *key =
 506                         skb_flow_dissector_target(f->dissector,
 507                                                   FLOW_DISSECTOR_KEY_VLAN,
 508                                                   f->key);
 509                 struct flow_dissector_key_vlan *mask =
 510                         skb_flow_dissector_target(f->dissector,
 511                                                   FLOW_DISSECTOR_KEY_VLAN,
 512                                                   f->mask);
 513                 if (mask->vlan_id || mask->vlan_priority) {
 514                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, cvlan_tag, 1);
 515                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, cvlan_tag, 1);
 516
 517                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_vid, mask->vlan_id);
 518                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_vid, key->vlan_id);
 519
 520                         MLX5_SET(fte_match_set_lyr_2_4, headers_c, first_prio, mask->vlan_priority);
 521                         MLX5_SET(fte_match_set_lyr_2_4, headers_v, first_prio, key->vlan_priority);
 522                 }
 523         }
 524
 525         if (addr_type == FLOW_DISSECTOR_KEY_IPV4_ADDRS) {
 526                 struct flow_dissector_key_ipv4_addrs *key =
 527                         skb_flow_dissector_target(f->dissector,
 528                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 529                                                   f->key);
 530                 struct flow_dissector_key_ipv4_addrs *mask =
 531                         skb_flow_dissector_target(f->dissector,
 532                                                   FLOW_DISSECTOR_KEY_IPV4_ADDRS,
 533                                                   f->mask);
 534
 535                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 536                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
 537                        &mask->src, sizeof(mask->src));
 538                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 539                                     src_ipv4_src_ipv6.ipv4_layout.ipv4),
 540                        &key->src, sizeof(key->src));
 541                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 542                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 543                        &mask->dst, sizeof(mask->dst));
 544                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 545                                     dst_ipv4_dst_ipv6.ipv4_layout.ipv4),
 546                        &key->dst, sizeof(key->dst));
 547
 548                 if (mask->src || mask->dst)
 549                         *min_inline = MLX5_INLINE_MODE_IP;
 550         }
 551
 552         if (addr_type == FLOW_DISSECTOR_KEY_IPV6_ADDRS) {
 553                 struct flow_dissector_key_ipv6_addrs *key =
 554                         skb_flow_dissector_target(f->dissector,
 555                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 556                                                   f->key);
 557                 struct flow_dissector_key_ipv6_addrs *mask =
 558                         skb_flow_dissector_target(f->dissector,
 559                                                   FLOW_DISSECTOR_KEY_IPV6_ADDRS,
 560                                                   f->mask);
 561
 562                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 563                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 564                        &mask->src, sizeof(mask->src));
 565                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 566                                     src_ipv4_src_ipv6.ipv6_layout.ipv6),
 567                        &key->src, sizeof(key->src));
 568
 569                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_c,
 570                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 571                        &mask->dst, sizeof(mask->dst));
 572                 memcpy(MLX5_ADDR_OF(fte_match_set_lyr_2_4, headers_v,
 573                                     dst_ipv4_dst_ipv6.ipv6_layout.ipv6),
 574                        &key->dst, sizeof(key->dst));
 575
 576                 if (ipv6_addr_type(&mask->src) != IPV6_ADDR_ANY ||
 577                     ipv6_addr_type(&mask->dst) != IPV6_ADDR_ANY)
 578                         *min_inline = MLX5_INLINE_MODE_IP;
 579         }
 580
 581         if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_PORTS)) {
 582                 struct flow_dissector_key_ports *key =
 583                         skb_flow_dissector_target(f->dissector,
 584                                                   FLOW_DISSECTOR_KEY_PORTS,
 585                                                   f->key);
 586                 struct flow_dissector_key_ports *mask =
 587                         skb_flow_dissector_target(f->dissector,
 588                                                   FLOW_DISSECTOR_KEY_PORTS,
 589                                                   f->mask);
 590                 switch (ip_proto) {
 591                 case IPPROTO_TCP:
 592                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 593                                  tcp_sport, ntohs(mask->src));
 594                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 595                                  tcp_sport, ntohs(key->src));
 596
 597                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 598                                  tcp_dport, ntohs(mask->dst));
 599                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 600                                  tcp_dport, ntohs(key->dst));
 601                         break;
 602
 603                 case IPPROTO_UDP:
 604                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 605                                  udp_sport, ntohs(mask->src));
 606                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 607                                  udp_sport, ntohs(key->src));
 608
 609                         MLX5_SET(fte_match_set_lyr_2_4, headers_c,
 610                                  udp_dport, ntohs(mask->dst));
 611                         MLX5_SET(fte_match_set_lyr_2_4, headers_v,
 612                                  udp_dport, ntohs(key->dst));
 613                         break;
 614                 default:
 615                         netdev_err(priv->netdev,
 616                                    "Only UDP and TCP transport are supported\n");
 617                         return -EINVAL;
 618                 }
 619
 620                 if (mask->src || mask->dst)
 621                         *min_inline = MLX5_INLINE_MODE_TCP_UDP;
 622         }
 623
 624         return 0;
 625 }
 626
 627 static int parse_cls_flower(struct mlx5e_priv *priv,
 628                             struct mlx5e_tc_flow *flow,
 629                             struct mlx5_flow_spec *spec,
 630                             struct tc_cls_flower_offload *f)
 631 {
 632         struct mlx5_core_dev *dev = priv->mdev;
 633         struct mlx5_eswitch *esw = dev->priv.eswitch;
 634         struct mlx5_eswitch_rep *rep = priv->ppriv;
 635         u8 min_inline;
 636         int err;
 637
 638         err = __parse_cls_flower(priv, spec, f, &min_inline);
 639
 640         if (!err && (flow->flags & MLX5E_TC_FLOW_ESWITCH) &&
 641             rep->vport != FDB_UPLINK_VPORT) {
 642                 if (min_inline > esw->offloads.inline_mode) {
 643                         netdev_warn(priv->netdev,
 644                                     "Flow is not offloaded due to min inline setting, required %d actual %d\n",
 645                                     min_inline, esw->offloads.inline_mode);
 646                         return -EOPNOTSUPP;
 647                 }
 648         }
 649
 650         return err;
 651 }
 652
 653 static int parse_tc_nic_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
 654                                 u32 *action, u32 *flow_tag)
 655 {
 656         const struct tc_action *a;
 657         LIST_HEAD(actions);
 658
 659         if (tc_no_actions(exts))
 660                 return -EINVAL;
 661
 662         *flow_tag = MLX5_FS_DEFAULT_FLOW_TAG;
 663         *action = 0;
 664
 665         tcf_exts_to_list(exts, &actions);
 666         list_for_each_entry(a, &actions, list) {
 667                 /* Only support a single action per rule */
 668                 if (*action)
 669                         return -EINVAL;
 670
 671                 if (is_tcf_gact_shot(a)) {
 672                         *action |= MLX5_FLOW_CONTEXT_ACTION_DROP;
 673                         if (MLX5_CAP_FLOWTABLE(priv->mdev,
 674                                                flow_table_properties_nic_receive.flow_counter))
 675                                 *action |= MLX5_FLOW_CONTEXT_ACTION_COUNT;
 676                         continue;
 677                 }
 678
 679                 if (is_tcf_skbedit_mark(a)) {
 680                         u32 mark = tcf_skbedit_mark(a);
 681
 682                         if (mark & ~MLX5E_TC_FLOW_ID_MASK) {
 683                                 netdev_warn(priv->netdev, "Bad flow mark - only 16 bit is supported: 0x%x\n",
 684                                             mark);
 685                                 return -EINVAL;
 686                         }
 687
 688                         *flow_tag = mark;
 689                         *action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST;
 690                         continue;
 691                 }
 692
 693                 return -EINVAL;
 694         }
 695
 696         return 0;
 697 }
 698
 699 static inline int cmp_encap_info(struct ip_tunnel_key *a,
 700                                  struct ip_tunnel_key *b)
 701 {
 702         return memcmp(a, b, sizeof(*a));
 703 }
 704
 705 static inline int hash_encap_info(struct ip_tunnel_key *key)
 706 {
 707         return jhash(key, sizeof(*key), 0);
 708 }
 709
 710 static int mlx5e_route_lookup_ipv4(struct mlx5e_priv *priv,
 711                                    struct net_device *mirred_dev,
 712                                    struct net_device **out_dev,
 713                                    struct flowi4 *fl4,
 714                                    struct neighbour **out_n,
 715                                    int *out_ttl)
 716 {
 717         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 718         struct rtable *rt;
 719         struct neighbour *n = NULL;
 720
 721 #if IS_ENABLED(CONFIG_INET)
 722         int ret;
 723
 724         rt = ip_route_output_key(dev_net(mirred_dev), fl4);
 725         ret = PTR_ERR_OR_ZERO(rt);
 726         if (ret)
 727                 return ret;
 728 #else
 729         return -EOPNOTSUPP;
 730 #endif
 731         /* if the egress device isn't on the same HW e-switch, we use the uplink */
 732         if (!switchdev_port_same_parent_id(priv->netdev, rt->dst.dev))
 733                 *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
 734         else
 735                 *out_dev = rt->dst.dev;
 736
 737         *out_ttl = ip4_dst_hoplimit(&rt->dst);
 738         n = dst_neigh_lookup(&rt->dst, &fl4->daddr);
 739         ip_rt_put(rt);
 740         if (!n)
 741                 return -ENOMEM;
 742
 743         *out_n = n;
 744         return 0;
 745 }
 746
 747 static int mlx5e_route_lookup_ipv6(struct mlx5e_priv *priv,
 748                                    struct net_device *mirred_dev,
 749                                    struct net_device **out_dev,
 750                                    struct flowi6 *fl6,
 751                                    struct neighbour **out_n,
 752                                    int *out_ttl)
 753 {
 754         struct neighbour *n = NULL;
 755         struct dst_entry *dst;
 756
 757 #if IS_ENABLED(CONFIG_INET) && IS_ENABLED(CONFIG_IPV6)
 758         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
 759         int ret;
 760
 761         dst = ip6_route_output(dev_net(mirred_dev), NULL, fl6);
 762         ret = dst->error;
 763         if (ret) {
 764                 dst_release(dst);
 765                 return ret;
 766         }
 767
 768         *out_ttl = ip6_dst_hoplimit(dst);
 769
 770         /* if the egress device isn't on the same HW e-switch, we use the uplink */
 771         if (!switchdev_port_same_parent_id(priv->netdev, dst->dev))
 772                 *out_dev = mlx5_eswitch_get_uplink_netdev(esw);
 773         else
 774                 *out_dev = dst->dev;
 775 #else
 776         return -EOPNOTSUPP;
 777 #endif
 778
 779         n = dst_neigh_lookup(dst, &fl6->daddr);
 780         dst_release(dst);
 781         if (!n)
 782                 return -ENOMEM;
 783
 784         *out_n = n;
 785         return 0;
 786 }
 787
 788 static int gen_vxlan_header_ipv4(struct net_device *out_dev,
 789                                  char buf[],
 790                                  unsigned char h_dest[ETH_ALEN],
 791                                  int ttl,
 792                                  __be32 daddr,
 793                                  __be32 saddr,
 794                                  __be16 udp_dst_port,
 795                                  __be32 vx_vni)
 796 {
 797         int encap_size = VXLAN_HLEN + sizeof(struct iphdr) + ETH_HLEN;
 798         struct ethhdr *eth = (struct ethhdr *)buf;
 799         struct iphdr  *ip = (struct iphdr *)((char *)eth + sizeof(struct ethhdr));
 800         struct udphdr *udp = (struct udphdr *)((char *)ip + sizeof(struct iphdr));
 801         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
 802
 803         memset(buf, 0, encap_size);
 804
 805         ether_addr_copy(eth->h_dest, h_dest);
 806         ether_addr_copy(eth->h_source, out_dev->dev_addr);
 807         eth->h_proto = htons(ETH_P_IP);
 808
 809         ip->daddr = daddr;
 810         ip->saddr = saddr;
 811
 812         ip->ttl = ttl;
 813         ip->protocol = IPPROTO_UDP;
 814         ip->version = 0x4;
 815         ip->ihl = 0x5;
 816
 817         udp->dest = udp_dst_port;
 818         vxh->vx_flags = VXLAN_HF_VNI;
 819         vxh->vx_vni = vxlan_vni_field(vx_vni);
 820
 821         return encap_size;
 822 }
 823
 824 static int gen_vxlan_header_ipv6(struct net_device *out_dev,
 825                                  char buf[],
 826                                  unsigned char h_dest[ETH_ALEN],
 827                                  int ttl,
 828                                  struct in6_addr *daddr,
 829                                  struct in6_addr *saddr,
 830                                  __be16 udp_dst_port,
 831                                  __be32 vx_vni)
 832 {
 833         int encap_size = VXLAN_HLEN + sizeof(struct ipv6hdr) + ETH_HLEN;
 834         struct ethhdr *eth = (struct ethhdr *)buf;
 835         struct ipv6hdr *ip6h = (struct ipv6hdr *)((char *)eth + sizeof(struct ethhdr));
 836         struct udphdr *udp = (struct udphdr *)((char *)ip6h + sizeof(struct ipv6hdr));
 837         struct vxlanhdr *vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr));
 838
 839         memset(buf, 0, encap_size);
 840
 841         ether_addr_copy(eth->h_dest, h_dest);
 842         ether_addr_copy(eth->h_source, out_dev->dev_addr);
 843         eth->h_proto = htons(ETH_P_IPV6);
 844
 845         ip6_flow_hdr(ip6h, 0, 0);
 846         /* the HW fills up ipv6 payload len */
 847         ip6h->nexthdr     = IPPROTO_UDP;
 848         ip6h->hop_limit   = ttl;
 849         ip6h->daddr       = *daddr;
 850         ip6h->saddr       = *saddr;
 851
 852         udp->dest = udp_dst_port;
 853         vxh->vx_flags = VXLAN_HF_VNI;
 854         vxh->vx_vni = vxlan_vni_field(vx_vni);
 855
 856         return encap_size;
 857 }
 858
 859 static int mlx5e_create_encap_header_ipv4(struct mlx5e_priv *priv,
 860                                           struct net_device *mirred_dev,
 861                                           struct mlx5_encap_entry *e,
 862                                           struct net_device **out_dev)
 863 {
 864         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
 865         struct ip_tunnel_key *tun_key = &e->tun_info.key;
 866         int encap_size, ttl, err;
 867         struct neighbour *n = NULL;
 868         struct flowi4 fl4 = {};
 869         char *encap_header;
 870
 871         encap_header = kzalloc(max_encap_size, GFP_KERNEL);
 872         if (!encap_header)
 873                 return -ENOMEM;
 874
 875         switch (e->tunnel_type) {
 876         case MLX5_HEADER_TYPE_VXLAN:
 877                 fl4.flowi4_proto = IPPROTO_UDP;
 878                 fl4.fl4_dport = tun_key->tp_dst;
 879                 break;
 880         default:
 881                 err = -EOPNOTSUPP;
 882                 goto out;
 883         }
 884         fl4.flowi4_tos = tun_key->tos;
 885         fl4.daddr = tun_key->u.ipv4.dst;
 886         fl4.saddr = tun_key->u.ipv4.src;
 887
 888         err = mlx5e_route_lookup_ipv4(priv, mirred_dev, out_dev,
 889                                       &fl4, &n, &ttl);
 890         if (err)
 891                 goto out;
 892
 893         if (!(n->nud_state & NUD_VALID)) {
 894                 pr_warn("%s: can't offload, neighbour to %pI4 invalid\n", __func__, &fl4.daddr);
 895                 err = -EOPNOTSUPP;
 896                 goto out;
 897         }
 898
 899         e->n = n;
 900         e->out_dev = *out_dev;
 901
 902         neigh_ha_snapshot(e->h_dest, n, *out_dev);
 903
 904         switch (e->tunnel_type) {
 905         case MLX5_HEADER_TYPE_VXLAN:
 906                 encap_size = gen_vxlan_header_ipv4(*out_dev, encap_header,
 907                                                    e->h_dest, ttl,
 908                                                    fl4.daddr,
 909                                                    fl4.saddr, tun_key->tp_dst,
 910                                                    tunnel_id_to_key32(tun_key->tun_id));
 911                 break;
 912         default:
 913                 err = -EOPNOTSUPP;
 914                 goto out;
 915         }
 916
 917         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
 918                                encap_size, encap_header, &e->encap_id);
 919 out:
 920         if (err && n)
 921                 neigh_release(n);
 922         kfree(encap_header);
 923         return err;
 924 }
 925
 926 static int mlx5e_create_encap_header_ipv6(struct mlx5e_priv *priv,
 927                                           struct net_device *mirred_dev,
 928                                           struct mlx5_encap_entry *e,
 929                                           struct net_device **out_dev)
 930
 931 {
 932         int max_encap_size = MLX5_CAP_ESW(priv->mdev, max_encap_header_size);
 933         struct ip_tunnel_key *tun_key = &e->tun_info.key;
 934         int encap_size, err, ttl = 0;
 935         struct neighbour *n = NULL;
 936         struct flowi6 fl6 = {};
 937         char *encap_header;
 938
 939         encap_header = kzalloc(max_encap_size, GFP_KERNEL);
 940         if (!encap_header)
 941                 return -ENOMEM;
 942
 943         switch (e->tunnel_type) {
 944         case MLX5_HEADER_TYPE_VXLAN:
 945                 fl6.flowi6_proto = IPPROTO_UDP;
 946                 fl6.fl6_dport = tun_key->tp_dst;
 947                 break;
 948         default:
 949                 err = -EOPNOTSUPP;
 950                 goto out;
 951         }
 952
 953         fl6.flowlabel = ip6_make_flowinfo(RT_TOS(tun_key->tos), tun_key->label);
 954         fl6.daddr = tun_key->u.ipv6.dst;
 955         fl6.saddr = tun_key->u.ipv6.src;
 956
 957         err = mlx5e_route_lookup_ipv6(priv, mirred_dev, out_dev,
 958                                       &fl6, &n, &ttl);
 959         if (err)
 960                 goto out;
 961
 962         if (!(n->nud_state & NUD_VALID)) {
 963                 pr_warn("%s: can't offload, neighbour to %pI6 invalid\n", __func__, &fl6.daddr);
 964                 err = -EOPNOTSUPP;
 965                 goto out;
 966         }
 967
 968         e->n = n;
 969         e->out_dev = *out_dev;
 970
 971         neigh_ha_snapshot(e->h_dest, n, *out_dev);
 972
 973         switch (e->tunnel_type) {
 974         case MLX5_HEADER_TYPE_VXLAN:
 975                 encap_size = gen_vxlan_header_ipv6(*out_dev, encap_header,
 976                                                    e->h_dest, ttl,
 977                                                    &fl6.daddr,
 978                                                    &fl6.saddr, tun_key->tp_dst,
 979                                                    tunnel_id_to_key32(tun_key->tun_id));
 980                 break;
 981         default:
 982                 err = -EOPNOTSUPP;
 983                 goto out;
 984         }
 985
 986         err = mlx5_encap_alloc(priv->mdev, e->tunnel_type,
 987                                encap_size, encap_header, &e->encap_id);
 988 out:
 989         if (err && n)
 990                 neigh_release(n);
 991         kfree(encap_header);
 992         return err;
 993 }
 994
 995 static int mlx5e_attach_encap(struct mlx5e_priv *priv,
 996                               struct ip_tunnel_info *tun_info,
 997                               struct net_device *mirred_dev,
 998                               struct mlx5_esw_flow_attr *attr)
 999 {
1000         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1001         struct net_device *up_dev = mlx5_eswitch_get_uplink_netdev(esw);
1002         struct mlx5e_priv *up_priv = netdev_priv(up_dev);
1003         unsigned short family = ip_tunnel_info_af(tun_info);
1004         struct ip_tunnel_key *key = &tun_info->key;
1005         struct mlx5_encap_entry *e;
1006         struct net_device *out_dev;
1007         int tunnel_type, err = -EOPNOTSUPP;
1008         uintptr_t hash_key;
1009         bool found = false;
1010
1011         /* udp dst port must be set */
1012         if (!memchr_inv(&key->tp_dst, 0, sizeof(key->tp_dst)))
1013                 goto vxlan_encap_offload_err;
1014
1015         /* setting udp src port isn't supported */
1016         if (memchr_inv(&key->tp_src, 0, sizeof(key->tp_src))) {
1017 vxlan_encap_offload_err:
1018                 netdev_warn(priv->netdev,
1019                             "must set udp dst port and not set udp src port\n");
1020                 return -EOPNOTSUPP;
1021         }
1022
1023         if (mlx5e_vxlan_lookup_port(up_priv, be16_to_cpu(key->tp_dst)) &&
1024             MLX5_CAP_ESW(priv->mdev, vxlan_encap_decap)) {
1025                 tunnel_type = MLX5_HEADER_TYPE_VXLAN;
1026         } else {
1027                 netdev_warn(priv->netdev,
1028                             "%d isn't an offloaded vxlan udp dport\n", be16_to_cpu(key->tp_dst));
1029                 return -EOPNOTSUPP;
1030         }
1031
1032         hash_key = hash_encap_info(key);
1033
1034         hash_for_each_possible_rcu(esw->offloads.encap_tbl, e,
1035                                    encap_hlist, hash_key) {
1036                 if (!cmp_encap_info(&e->tun_info.key, key)) {
1037                         found = true;
1038                         break;
1039                 }
1040         }
1041
1042         if (found) {
1043                 attr->encap = e;
1044                 return 0;
1045         }
1046
1047         e = kzalloc(sizeof(*e), GFP_KERNEL);
1048         if (!e)
1049                 return -ENOMEM;
1050
1051         e->tun_info = *tun_info;
1052         e->tunnel_type = tunnel_type;
1053         INIT_LIST_HEAD(&e->flows);
1054
1055         if (family == AF_INET)
1056                 err = mlx5e_create_encap_header_ipv4(priv, mirred_dev, e, &out_dev);
1057         else if (family == AF_INET6)
1058                 err = mlx5e_create_encap_header_ipv6(priv, mirred_dev, e, &out_dev);
1059
1060         if (err)
1061                 goto out_err;
1062
1063         attr->encap = e;
1064         hash_add_rcu(esw->offloads.encap_tbl, &e->encap_hlist, hash_key);
1065
1066         return err;
1067
1068 out_err:
1069         kfree(e);
1070         return err;
1071 }
1072
1073 static int parse_tc_fdb_actions(struct mlx5e_priv *priv, struct tcf_exts *exts,
1074                                 struct mlx5e_tc_flow *flow)
1075 {
1076         struct mlx5_esw_flow_attr *attr = flow->attr;
1077         struct ip_tunnel_info *info = NULL;
1078         const struct tc_action *a;
1079         LIST_HEAD(actions);
1080         bool encap = false;
1081         int err;
1082
1083         if (tc_no_actions(exts))
1084                 return -EINVAL;
1085
1086         memset(attr, 0, sizeof(*attr));
1087         attr->in_rep = priv->ppriv;
1088
1089         tcf_exts_to_list(exts, &actions);
1090         list_for_each_entry(a, &actions, list) {
1091                 if (is_tcf_gact_shot(a)) {
1092                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DROP |
1093                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1094                         continue;
1095                 }
1096
1097                 if (is_tcf_mirred_egress_redirect(a)) {
1098                         int ifindex = tcf_mirred_ifindex(a);
1099                         struct net_device *out_dev;
1100                         struct mlx5e_priv *out_priv;
1101
1102                         out_dev = __dev_get_by_index(dev_net(priv->netdev), ifindex);
1103
1104                         if (switchdev_port_same_parent_id(priv->netdev,
1105                                                           out_dev)) {
1106                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1107                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1108                                 out_priv = netdev_priv(out_dev);
1109                                 attr->out_rep = out_priv->ppriv;
1110                         } else if (encap) {
1111                                 err = mlx5e_attach_encap(priv, info,
1112                                                          out_dev, attr);
1113                                 if (err)
1114                                         return err;
1115                                 list_add(&flow->encap, &attr->encap->flows);
1116                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_ENCAP |
1117                                         MLX5_FLOW_CONTEXT_ACTION_FWD_DEST |
1118                                         MLX5_FLOW_CONTEXT_ACTION_COUNT;
1119                                 out_priv = netdev_priv(attr->encap->out_dev);
1120                                 attr->out_rep = out_priv->ppriv;
1121                         } else {
1122                                 pr_err("devices %s %s not on same switch HW, can't offload forwarding\n",
1123                                        priv->netdev->name, out_dev->name);
1124                                 return -EINVAL;
1125                         }
1126                         continue;
1127                 }
1128
1129                 if (is_tcf_tunnel_set(a)) {
1130                         info = tcf_tunnel_info(a);
1131                         if (info)
1132                                 encap = true;
1133                         else
1134                                 return -EOPNOTSUPP;
1135                         continue;
1136                 }
1137
1138                 if (is_tcf_vlan(a)) {
1139                         if (tcf_vlan_action(a) == TCA_VLAN_ACT_POP) {
1140                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_POP;
1141                         } else if (tcf_vlan_action(a) == TCA_VLAN_ACT_PUSH) {
1142                                 if (tcf_vlan_push_proto(a) != htons(ETH_P_8021Q))
1143                                         return -EOPNOTSUPP;
1144
1145                                 attr->action |= MLX5_FLOW_CONTEXT_ACTION_VLAN_PUSH;
1146                                 attr->vlan = tcf_vlan_push_vid(a);
1147                         } else { /* action is TCA_VLAN_ACT_MODIFY */
1148                                 return -EOPNOTSUPP;
1149                         }
1150                         continue;
1151                 }
1152
1153                 if (is_tcf_tunnel_release(a)) {
1154                         attr->action |= MLX5_FLOW_CONTEXT_ACTION_DECAP;
1155                         continue;
1156                 }
1157
1158                 return -EINVAL;
1159         }
1160         return 0;
1161 }
1162
1163 int mlx5e_configure_flower(struct mlx5e_priv *priv, __be16 protocol,
1164                            struct tc_cls_flower_offload *f)
1165 {
1166         struct mlx5e_tc_table *tc = &priv->fs.tc;
1167         int err, attr_size = 0;
1168         u32 flow_tag, action;
1169         struct mlx5e_tc_flow *flow;
1170         struct mlx5_flow_spec *spec;
1171         struct mlx5_eswitch *esw = priv->mdev->priv.eswitch;
1172         u8 flow_flags = 0;
1173
1174         if (esw && esw->mode == SRIOV_OFFLOADS) {
1175                 flow_flags = MLX5E_TC_FLOW_ESWITCH;
1176                 attr_size  = sizeof(struct mlx5_esw_flow_attr);
1177         }
1178
1179         flow = kzalloc(sizeof(*flow) + attr_size, GFP_KERNEL);
1180         spec = mlx5_vzalloc(sizeof(*spec));
1181         if (!spec || !flow) {
1182                 err = -ENOMEM;
1183                 goto err_free;
1184         }
1185
1186         flow->cookie = f->cookie;
1187         flow->flags = flow_flags;
1188
1189         err = parse_cls_flower(priv, flow, spec, f);
1190         if (err < 0)
1191                 goto err_free;
1192
1193         if (flow->flags & MLX5E_TC_FLOW_ESWITCH) {
1194                 flow->attr  = (struct mlx5_esw_flow_attr *)(flow + 1);
1195                 err = parse_tc_fdb_actions(priv, f->exts, flow);
1196                 if (err < 0)
1197                         goto err_free;
1198                 flow->rule = mlx5e_tc_add_fdb_flow(priv, spec, flow->attr);
1199         } else {
1200                 err = parse_tc_nic_actions(priv, f->exts, &action, &flow_tag);
1201                 if (err < 0)
1202                         goto err_free;
1203                 flow->rule = mlx5e_tc_add_nic_flow(priv, spec, action, flow_tag);
1204         }
1205
1206         if (IS_ERR(flow->rule)) {
1207                 err = PTR_ERR(flow->rule);
1208                 goto err_del_rule;
1209         }
1210
1211         err = rhashtable_insert_fast(&tc->ht, &flow->node,
1212                                      tc->ht_params);
1213         if (err)
1214                 goto err_del_rule;
1215
1216         goto out;
1217
1218 err_del_rule:
1219         mlx5e_tc_del_flow(priv, flow);
1220
1221 err_free:
1222         kfree(flow);
1223 out:
1224         kvfree(spec);
1225         return err;
1226 }
1227
1228 int mlx5e_delete_flower(struct mlx5e_priv *priv,
1229                         struct tc_cls_flower_offload *f)
1230 {
1231         struct mlx5e_tc_flow *flow;
1232         struct mlx5e_tc_table *tc = &priv->fs.tc;
1233
1234         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
1235                                       tc->ht_params);
1236         if (!flow)
1237                 return -EINVAL;
1238
1239         rhashtable_remove_fast(&tc->ht, &flow->node, tc->ht_params);
1240
1241         mlx5e_tc_del_flow(priv, flow);
1242
1243
1244         kfree(flow);
1245
1246         return 0;
1247 }
1248
1249 int mlx5e_stats_flower(struct mlx5e_priv *priv,
1250                        struct tc_cls_flower_offload *f)
1251 {
1252         struct mlx5e_tc_table *tc = &priv->fs.tc;
1253         struct mlx5e_tc_flow *flow;
1254         struct tc_action *a;
1255         struct mlx5_fc *counter;
1256         LIST_HEAD(actions);
1257         u64 bytes;
1258         u64 packets;
1259         u64 lastuse;
1260
1261         flow = rhashtable_lookup_fast(&tc->ht, &f->cookie,
1262                                       tc->ht_params);
1263         if (!flow)
1264                 return -EINVAL;
1265
1266         counter = mlx5_flow_rule_counter(flow->rule);
1267         if (!counter)
1268                 return 0;
1269
1270         mlx5_fc_query_cached(counter, &bytes, &packets, &lastuse);
1271
1272         preempt_disable();
1273
1274         tcf_exts_to_list(f->exts, &actions);
1275         list_for_each_entry(a, &actions, list)
1276                 tcf_action_stats_update(a, bytes, packets, lastuse);
1277
1278         preempt_enable();
1279
1280         return 0;
1281 }
1282
1283 static const struct rhashtable_params mlx5e_tc_flow_ht_params = {
1284         .head_offset = offsetof(struct mlx5e_tc_flow, node),
1285         .key_offset = offsetof(struct mlx5e_tc_flow, cookie),
1286         .key_len = sizeof(((struct mlx5e_tc_flow *)0)->cookie),
1287         .automatic_shrinking = true,
1288 };
1289
1290 int mlx5e_tc_init(struct mlx5e_priv *priv)
1291 {
1292         struct mlx5e_tc_table *tc = &priv->fs.tc;
1293
1294         tc->ht_params = mlx5e_tc_flow_ht_params;
1295         return rhashtable_init(&tc->ht, &tc->ht_params);
1296 }
1297
1298 static void _mlx5e_tc_del_flow(void *ptr, void *arg)
1299 {
1300         struct mlx5e_tc_flow *flow = ptr;
1301         struct mlx5e_priv *priv = arg;
1302
1303         mlx5e_tc_del_flow(priv, flow);
1304         kfree(flow);
1305 }
1306
1307 void mlx5e_tc_cleanup(struct mlx5e_priv *priv)
1308 {
1309         struct mlx5e_tc_table *tc = &priv->fs.tc;
1310
1311         rhashtable_free_and_destroy(&tc->ht, _mlx5e_tc_del_flow, priv);
1312
1313         if (!IS_ERR_OR_NULL(tc->t)) {
1314                 mlx5_destroy_flow_table(tc->t);
1315                 tc->t = NULL;
1316         }
1317 }