drivers/net/mlx5/mlx5_flow_tcf.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright 2018 6WIND S.A.
   3  * Copyright 2018 Mellanox Technologies, Ltd
   4  */
   5
   6 #include <assert.h>
   7 #include <errno.h>
   8 #include <libmnl/libmnl.h>
   9 #include <linux/gen_stats.h>
  10 #include <linux/if_ether.h>
  11 #include <linux/netlink.h>
  12 #include <linux/pkt_cls.h>
  13 #include <linux/pkt_sched.h>
  14 #include <linux/rtnetlink.h>
  15 #include <linux/tc_act/tc_gact.h>
  16 #include <linux/tc_act/tc_mirred.h>
  17 #include <netinet/in.h>
  18 #include <stdalign.h>
  19 #include <stdbool.h>
  20 #include <stddef.h>
  21 #include <stdint.h>
  22 #include <stdlib.h>
  23 #include <sys/socket.h>
  24
  25 #include <rte_byteorder.h>
  26 #include <rte_errno.h>
  27 #include <rte_ether.h>
  28 #include <rte_flow.h>
  29 #include <rte_malloc.h>
  30 #include <rte_common.h>
  31
  32 #include "mlx5.h"
  33 #include "mlx5_flow.h"
  34 #include "mlx5_autoconf.h"
  35
  36 #ifdef HAVE_TC_ACT_VLAN
  37
  38 #include <linux/tc_act/tc_vlan.h>
  39
  40 #else /* HAVE_TC_ACT_VLAN */
  41
  42 #define TCA_VLAN_ACT_POP 1
  43 #define TCA_VLAN_ACT_PUSH 2
  44 #define TCA_VLAN_ACT_MODIFY 3
  45 #define TCA_VLAN_PARMS 2
  46 #define TCA_VLAN_PUSH_VLAN_ID 3
  47 #define TCA_VLAN_PUSH_VLAN_PROTOCOL 4
  48 #define TCA_VLAN_PAD 5
  49 #define TCA_VLAN_PUSH_VLAN_PRIORITY 6
  50
  51 struct tc_vlan {
  52         tc_gen;
  53         int v_action;
  54 };
  55
  56 #endif /* HAVE_TC_ACT_VLAN */
  57
  58 #ifdef HAVE_TC_ACT_PEDIT
  59
  60 #include <linux/tc_act/tc_pedit.h>
  61
  62 #else /* HAVE_TC_ACT_VLAN */
  63
  64 enum {
  65         TCA_PEDIT_UNSPEC,
  66         TCA_PEDIT_TM,
  67         TCA_PEDIT_PARMS,
  68         TCA_PEDIT_PAD,
  69         TCA_PEDIT_PARMS_EX,
  70         TCA_PEDIT_KEYS_EX,
  71         TCA_PEDIT_KEY_EX,
  72         __TCA_PEDIT_MAX
  73 };
  74
  75 enum {
  76         TCA_PEDIT_KEY_EX_HTYPE = 1,
  77         TCA_PEDIT_KEY_EX_CMD = 2,
  78         __TCA_PEDIT_KEY_EX_MAX
  79 };
  80
  81 enum pedit_header_type {
  82         TCA_PEDIT_KEY_EX_HDR_TYPE_NETWORK = 0,
  83         TCA_PEDIT_KEY_EX_HDR_TYPE_ETH = 1,
  84         TCA_PEDIT_KEY_EX_HDR_TYPE_IP4 = 2,
  85         TCA_PEDIT_KEY_EX_HDR_TYPE_IP6 = 3,
  86         TCA_PEDIT_KEY_EX_HDR_TYPE_TCP = 4,
  87         TCA_PEDIT_KEY_EX_HDR_TYPE_UDP = 5,
  88         __PEDIT_HDR_TYPE_MAX,
  89 };
  90
  91 enum pedit_cmd {
  92         TCA_PEDIT_KEY_EX_CMD_SET = 0,
  93         TCA_PEDIT_KEY_EX_CMD_ADD = 1,
  94         __PEDIT_CMD_MAX,
  95 };
  96
  97 struct tc_pedit_key {
  98         __u32 mask; /* AND */
  99         __u32 val; /*XOR */
 100         __u32 off; /*offset */
 101         __u32 at;
 102         __u32 offmask;
 103         __u32 shift;
 104 };
 105
 106 __extension__
 107 struct tc_pedit_sel {
 108         tc_gen;
 109         unsigned char nkeys;
 110         unsigned char flags;
 111         struct tc_pedit_key keys[0];
 112 };
 113
 114 #endif /* HAVE_TC_ACT_VLAN */
 115
 116 #ifdef HAVE_TC_ACT_TUNNEL_KEY
 117
 118 #include <linux/tc_act/tc_tunnel_key.h>
 119
 120 #ifndef HAVE_TCA_TUNNEL_KEY_ENC_DST_PORT
 121 #define TCA_TUNNEL_KEY_ENC_DST_PORT 9
 122 #endif
 123
 124 #ifndef HAVE_TCA_TUNNEL_KEY_NO_CSUM
 125 #define TCA_TUNNEL_KEY_NO_CSUM 10
 126 #endif
 127
 128 #else /* HAVE_TC_ACT_TUNNEL_KEY */
 129
 130 #define TCA_ACT_TUNNEL_KEY 17
 131 #define TCA_TUNNEL_KEY_ACT_SET 1
 132 #define TCA_TUNNEL_KEY_ACT_RELEASE 2
 133 #define TCA_TUNNEL_KEY_PARMS 2
 134 #define TCA_TUNNEL_KEY_ENC_IPV4_SRC 3
 135 #define TCA_TUNNEL_KEY_ENC_IPV4_DST 4
 136 #define TCA_TUNNEL_KEY_ENC_IPV6_SRC 5
 137 #define TCA_TUNNEL_KEY_ENC_IPV6_DST 6
 138 #define TCA_TUNNEL_KEY_ENC_KEY_ID 7
 139 #define TCA_TUNNEL_KEY_ENC_DST_PORT 9
 140 #define TCA_TUNNEL_KEY_NO_CSUM 10
 141
 142 struct tc_tunnel_key {
 143         tc_gen;
 144         int t_action;
 145 };
 146
 147 #endif /* HAVE_TC_ACT_TUNNEL_KEY */
 148
 149 /* Normally found in linux/netlink.h. */
 150 #ifndef NETLINK_CAP_ACK
 151 #define NETLINK_CAP_ACK 10
 152 #endif
 153
 154 /* Normally found in linux/pkt_sched.h. */
 155 #ifndef TC_H_MIN_INGRESS
 156 #define TC_H_MIN_INGRESS 0xfff2u
 157 #endif
 158
 159 /* Normally found in linux/pkt_cls.h. */
 160 #ifndef TCA_CLS_FLAGS_SKIP_SW
 161 #define TCA_CLS_FLAGS_SKIP_SW (1 << 1)
 162 #endif
 163 #ifndef HAVE_TCA_CHAIN
 164 #define TCA_CHAIN 11
 165 #endif
 166 #ifndef HAVE_TCA_FLOWER_ACT
 167 #define TCA_FLOWER_ACT 3
 168 #endif
 169 #ifndef HAVE_TCA_FLOWER_FLAGS
 170 #define TCA_FLOWER_FLAGS 22
 171 #endif
 172 #ifndef HAVE_TCA_FLOWER_KEY_ETH_TYPE
 173 #define TCA_FLOWER_KEY_ETH_TYPE 8
 174 #endif
 175 #ifndef HAVE_TCA_FLOWER_KEY_ETH_DST
 176 #define TCA_FLOWER_KEY_ETH_DST 4
 177 #endif
 178 #ifndef HAVE_TCA_FLOWER_KEY_ETH_DST_MASK
 179 #define TCA_FLOWER_KEY_ETH_DST_MASK 5
 180 #endif
 181 #ifndef HAVE_TCA_FLOWER_KEY_ETH_SRC
 182 #define TCA_FLOWER_KEY_ETH_SRC 6
 183 #endif
 184 #ifndef HAVE_TCA_FLOWER_KEY_ETH_SRC_MASK
 185 #define TCA_FLOWER_KEY_ETH_SRC_MASK 7
 186 #endif
 187 #ifndef HAVE_TCA_FLOWER_KEY_IP_PROTO
 188 #define TCA_FLOWER_KEY_IP_PROTO 9
 189 #endif
 190 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_SRC
 191 #define TCA_FLOWER_KEY_IPV4_SRC 10
 192 #endif
 193 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_SRC_MASK
 194 #define TCA_FLOWER_KEY_IPV4_SRC_MASK 11
 195 #endif
 196 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_DST
 197 #define TCA_FLOWER_KEY_IPV4_DST 12
 198 #endif
 199 #ifndef HAVE_TCA_FLOWER_KEY_IPV4_DST_MASK
 200 #define TCA_FLOWER_KEY_IPV4_DST_MASK 13
 201 #endif
 202 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_SRC
 203 #define TCA_FLOWER_KEY_IPV6_SRC 14
 204 #endif
 205 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_SRC_MASK
 206 #define TCA_FLOWER_KEY_IPV6_SRC_MASK 15
 207 #endif
 208 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_DST
 209 #define TCA_FLOWER_KEY_IPV6_DST 16
 210 #endif
 211 #ifndef HAVE_TCA_FLOWER_KEY_IPV6_DST_MASK
 212 #define TCA_FLOWER_KEY_IPV6_DST_MASK 17
 213 #endif
 214 #ifndef HAVE_TCA_FLOWER_KEY_TCP_SRC
 215 #define TCA_FLOWER_KEY_TCP_SRC 18
 216 #endif
 217 #ifndef HAVE_TCA_FLOWER_KEY_TCP_SRC_MASK
 218 #define TCA_FLOWER_KEY_TCP_SRC_MASK 35
 219 #endif
 220 #ifndef HAVE_TCA_FLOWER_KEY_TCP_DST
 221 #define TCA_FLOWER_KEY_TCP_DST 19
 222 #endif
 223 #ifndef HAVE_TCA_FLOWER_KEY_TCP_DST_MASK
 224 #define TCA_FLOWER_KEY_TCP_DST_MASK 36
 225 #endif
 226 #ifndef HAVE_TCA_FLOWER_KEY_UDP_SRC
 227 #define TCA_FLOWER_KEY_UDP_SRC 20
 228 #endif
 229 #ifndef HAVE_TCA_FLOWER_KEY_UDP_SRC_MASK
 230 #define TCA_FLOWER_KEY_UDP_SRC_MASK 37
 231 #endif
 232 #ifndef HAVE_TCA_FLOWER_KEY_UDP_DST
 233 #define TCA_FLOWER_KEY_UDP_DST 21
 234 #endif
 235 #ifndef HAVE_TCA_FLOWER_KEY_UDP_DST_MASK
 236 #define TCA_FLOWER_KEY_UDP_DST_MASK 38
 237 #endif
 238 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_ID
 239 #define TCA_FLOWER_KEY_VLAN_ID 23
 240 #endif
 241 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_PRIO
 242 #define TCA_FLOWER_KEY_VLAN_PRIO 24
 243 #endif
 244 #ifndef HAVE_TCA_FLOWER_KEY_VLAN_ETH_TYPE
 245 #define TCA_FLOWER_KEY_VLAN_ETH_TYPE 25
 246 #endif
 247 #ifndef HAVE_TCA_FLOWER_KEY_ENC_KEY_ID
 248 #define TCA_FLOWER_KEY_ENC_KEY_ID 26
 249 #endif
 250 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_SRC
 251 #define TCA_FLOWER_KEY_ENC_IPV4_SRC 27
 252 #endif
 253 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK
 254 #define TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK 28
 255 #endif
 256 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_DST
 257 #define TCA_FLOWER_KEY_ENC_IPV4_DST 29
 258 #endif
 259 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV4_DST_MASK
 260 #define TCA_FLOWER_KEY_ENC_IPV4_DST_MASK 30
 261 #endif
 262 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_SRC
 263 #define TCA_FLOWER_KEY_ENC_IPV6_SRC 31
 264 #endif
 265 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK
 266 #define TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK 32
 267 #endif
 268 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_DST
 269 #define TCA_FLOWER_KEY_ENC_IPV6_DST 33
 270 #endif
 271 #ifndef HAVE_TCA_FLOWER_KEY_ENC_IPV6_DST_MASK
 272 #define TCA_FLOWER_KEY_ENC_IPV6_DST_MASK 34
 273 #endif
 274 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_SRC_PORT
 275 #define TCA_FLOWER_KEY_ENC_UDP_SRC_PORT 43
 276 #endif
 277 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK
 278 #define TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK 44
 279 #endif
 280 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_DST_PORT
 281 #define TCA_FLOWER_KEY_ENC_UDP_DST_PORT 45
 282 #endif
 283 #ifndef HAVE_TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK
 284 #define TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK 46
 285 #endif
 286 #ifndef HAVE_TCA_FLOWER_KEY_TCP_FLAGS
 287 #define TCA_FLOWER_KEY_TCP_FLAGS 71
 288 #endif
 289 #ifndef HAVE_TCA_FLOWER_KEY_TCP_FLAGS_MASK
 290 #define TCA_FLOWER_KEY_TCP_FLAGS_MASK 72
 291 #endif
 292 #ifndef HAVE_TC_ACT_GOTO_CHAIN
 293 #define TC_ACT_GOTO_CHAIN 0x20000000
 294 #endif
 295
 296 #ifndef IPV6_ADDR_LEN
 297 #define IPV6_ADDR_LEN 16
 298 #endif
 299
 300 #ifndef IPV4_ADDR_LEN
 301 #define IPV4_ADDR_LEN 4
 302 #endif
 303
 304 #ifndef TP_PORT_LEN
 305 #define TP_PORT_LEN 2 /* Transport Port (UDP/TCP) Length */
 306 #endif
 307
 308 #ifndef TTL_LEN
 309 #define TTL_LEN 1
 310 #endif
 311
 312 #ifndef TCA_ACT_MAX_PRIO
 313 #define TCA_ACT_MAX_PRIO 32
 314 #endif
 315
 316 /** UDP port range of VXLAN devices created by driver. */
 317 #define MLX5_VXLAN_PORT_MIN 30000
 318 #define MLX5_VXLAN_PORT_MAX 60000
 319 #define MLX5_VXLAN_DEVICE_PFX "vmlx_"
 320
 321 /** Tunnel action type, used for @p type in header structure. */
 322 enum flow_tcf_tunact_type {
 323         FLOW_TCF_TUNACT_VXLAN_DECAP,
 324         FLOW_TCF_TUNACT_VXLAN_ENCAP,
 325 };
 326
 327 /** Flags used for @p mask in tunnel action encap descriptors. */
 328 #define FLOW_TCF_ENCAP_ETH_SRC (1u << 0)
 329 #define FLOW_TCF_ENCAP_ETH_DST (1u << 1)
 330 #define FLOW_TCF_ENCAP_IPV4_SRC (1u << 2)
 331 #define FLOW_TCF_ENCAP_IPV4_DST (1u << 3)
 332 #define FLOW_TCF_ENCAP_IPV6_SRC (1u << 4)
 333 #define FLOW_TCF_ENCAP_IPV6_DST (1u << 5)
 334 #define FLOW_TCF_ENCAP_UDP_SRC (1u << 6)
 335 #define FLOW_TCF_ENCAP_UDP_DST (1u << 7)
 336 #define FLOW_TCF_ENCAP_VXLAN_VNI (1u << 8)
 337
 338 /**
 339  * Structure for holding netlink context.
 340  * Note the size of the message buffer which is MNL_SOCKET_BUFFER_SIZE.
 341  * Using this (8KB) buffer size ensures that netlink messages will never be
 342  * truncated.
 343  */
 344 struct mlx5_flow_tcf_context {
 345         struct mnl_socket *nl; /* NETLINK_ROUTE libmnl socket. */
 346         uint32_t seq; /* Message sequence number. */
 347         uint32_t buf_size; /* Message buffer size. */
 348         uint8_t *buf; /* Message buffer. */
 349 };
 350
 351 /**
 352  * Neigh rule structure. The neigh rule is applied via Netlink to
 353  * outer tunnel iface in order to provide destination MAC address
 354  * for the VXLAN encapsultion. The neigh rule is implicitly related
 355  * to the Flow itself and can be shared by multiple Flows.
 356  */
 357 struct tcf_neigh_rule {
 358         LIST_ENTRY(tcf_neigh_rule) next;
 359         uint32_t refcnt;
 360         struct ether_addr eth;
 361         uint16_t mask;
 362         union {
 363                 struct {
 364                         rte_be32_t dst;
 365                 } ipv4;
 366                 struct {
 367                         uint8_t dst[IPV6_ADDR_LEN];
 368                 } ipv6;
 369         };
 370 };
 371
 372 /**
 373  * Local rule structure. The local rule is applied via Netlink to
 374  * outer tunnel iface in order to provide local and peer IP addresses
 375  * of the VXLAN tunnel for encapsulation. The local rule is implicitly
 376  * related to the Flow itself and can be shared by multiple Flows.
 377  */
 378 struct tcf_local_rule {
 379         LIST_ENTRY(tcf_local_rule) next;
 380         uint32_t refcnt;
 381         uint16_t mask;
 382         union {
 383                 struct {
 384                         rte_be32_t dst;
 385                         rte_be32_t src;
 386                 } ipv4;
 387                 struct {
 388                         uint8_t dst[IPV6_ADDR_LEN];
 389                         uint8_t src[IPV6_ADDR_LEN];
 390                 } ipv6;
 391         };
 392 };
 393
 394 /** VXLAN virtual netdev. */
 395 struct tcf_vtep {
 396         LIST_ENTRY(tcf_vtep) next;
 397         LIST_HEAD(, tcf_neigh_rule) neigh;
 398         LIST_HEAD(, tcf_local_rule) local;
 399         uint32_t refcnt;
 400         unsigned int ifindex; /**< Own interface index. */
 401         unsigned int ifouter; /**< Index of device attached to. */
 402         uint16_t port;
 403         uint8_t created;
 404 };
 405
 406 /** Tunnel descriptor header, common for all tunnel types. */
 407 struct flow_tcf_tunnel_hdr {
 408         uint32_t type; /**< Tunnel action type. */
 409         struct tcf_vtep *vtep; /**< Virtual tunnel endpoint device. */
 410         unsigned int ifindex_org; /**< Original dst/src interface */
 411         unsigned int *ifindex_ptr; /**< Interface ptr in message. */
 412 };
 413
 414 struct flow_tcf_vxlan_decap {
 415         struct flow_tcf_tunnel_hdr hdr;
 416         uint16_t udp_port;
 417 };
 418
 419 struct flow_tcf_vxlan_encap {
 420         struct flow_tcf_tunnel_hdr hdr;
 421         uint32_t mask;
 422         struct {
 423                 struct ether_addr dst;
 424                 struct ether_addr src;
 425         } eth;
 426         union {
 427                 struct {
 428                         rte_be32_t dst;
 429                         rte_be32_t src;
 430                 } ipv4;
 431                 struct {
 432                         uint8_t dst[IPV6_ADDR_LEN];
 433                         uint8_t src[IPV6_ADDR_LEN];
 434                 } ipv6;
 435         };
 436 struct {
 437                 rte_be16_t src;
 438                 rte_be16_t dst;
 439         } udp;
 440         struct {
 441                 uint8_t vni[3];
 442         } vxlan;
 443 };
 444
 445 /** Structure used when extracting the values of a flow counters
 446  * from a netlink message.
 447  */
 448 struct flow_tcf_stats_basic {
 449         bool valid;
 450         struct gnet_stats_basic counters;
 451 };
 452
 453 /** Empty masks for known item types. */
 454 static const union {
 455         struct rte_flow_item_port_id port_id;
 456         struct rte_flow_item_eth eth;
 457         struct rte_flow_item_vlan vlan;
 458         struct rte_flow_item_ipv4 ipv4;
 459         struct rte_flow_item_ipv6 ipv6;
 460         struct rte_flow_item_tcp tcp;
 461         struct rte_flow_item_udp udp;
 462         struct rte_flow_item_vxlan vxlan;
 463 } flow_tcf_mask_empty;
 464
 465 /** Supported masks for known item types. */
 466 static const struct {
 467         struct rte_flow_item_port_id port_id;
 468         struct rte_flow_item_eth eth;
 469         struct rte_flow_item_vlan vlan;
 470         struct rte_flow_item_ipv4 ipv4;
 471         struct rte_flow_item_ipv6 ipv6;
 472         struct rte_flow_item_tcp tcp;
 473         struct rte_flow_item_udp udp;
 474         struct rte_flow_item_vxlan vxlan;
 475 } flow_tcf_mask_supported = {
 476         .port_id = {
 477                 .id = 0xffffffff,
 478         },
 479         .eth = {
 480                 .type = RTE_BE16(0xffff),
 481                 .dst.addr_bytes = "\xff\xff\xff\xff\xff\xff",
 482                 .src.addr_bytes = "\xff\xff\xff\xff\xff\xff",
 483         },
 484         .vlan = {
 485                 /* PCP and VID only, no DEI. */
 486                 .tci = RTE_BE16(0xefff),
 487                 .inner_type = RTE_BE16(0xffff),
 488         },
 489         .ipv4.hdr = {
 490                 .next_proto_id = 0xff,
 491                 .src_addr = RTE_BE32(0xffffffff),
 492                 .dst_addr = RTE_BE32(0xffffffff),
 493         },
 494         .ipv6.hdr = {
 495                 .proto = 0xff,
 496                 .src_addr =
 497                         "\xff\xff\xff\xff\xff\xff\xff\xff"
 498                         "\xff\xff\xff\xff\xff\xff\xff\xff",
 499                 .dst_addr =
 500                         "\xff\xff\xff\xff\xff\xff\xff\xff"
 501                         "\xff\xff\xff\xff\xff\xff\xff\xff",
 502         },
 503         .tcp.hdr = {
 504                 .src_port = RTE_BE16(0xffff),
 505                 .dst_port = RTE_BE16(0xffff),
 506                 .tcp_flags = 0xff,
 507         },
 508         .udp.hdr = {
 509                 .src_port = RTE_BE16(0xffff),
 510                 .dst_port = RTE_BE16(0xffff),
 511         },
 512         .vxlan = {
 513                .vni = "\xff\xff\xff",
 514         },
 515 };
 516
 517 #define SZ_NLATTR_HDR MNL_ALIGN(sizeof(struct nlattr))
 518 #define SZ_NLATTR_NEST SZ_NLATTR_HDR
 519 #define SZ_NLATTR_DATA_OF(len) MNL_ALIGN(SZ_NLATTR_HDR + (len))
 520 #define SZ_NLATTR_TYPE_OF(typ) SZ_NLATTR_DATA_OF(sizeof(typ))
 521 #define SZ_NLATTR_STRZ_OF(str) SZ_NLATTR_DATA_OF(strlen(str) + 1)
 522
 523 #define PTOI_TABLE_SZ_MAX(dev) (mlx5_dev_to_port_id((dev)->device, NULL, 0) + 2)
 524
 525 /** DPDK port to network interface index (ifindex) conversion. */
 526 struct flow_tcf_ptoi {
 527         uint16_t port_id; /**< DPDK port ID. */
 528         unsigned int ifindex; /**< Network interface index. */
 529 };
 530
 531 /* Due to a limitation on driver/FW. */
 532 #define MLX5_TCF_GROUP_ID_MAX 3
 533
 534 /*
 535  * Due to a limitation on driver/FW, priority ranges from 1 to 16 in kernel.
 536  * Priority in rte_flow attribute starts from 0 and is added by 1 in
 537  * translation. This is subject to be changed to determine the max priority
 538  * based on trial-and-error like Verbs driver once the restriction is lifted or
 539  * the range is extended.
 540  */
 541 #define MLX5_TCF_GROUP_PRIORITY_MAX 15
 542
 543 #define MLX5_TCF_FATE_ACTIONS \
 544         (MLX5_FLOW_ACTION_DROP | MLX5_FLOW_ACTION_PORT_ID | \
 545          MLX5_FLOW_ACTION_JUMP)
 546
 547 #define MLX5_TCF_VLAN_ACTIONS \
 548         (MLX5_FLOW_ACTION_OF_POP_VLAN | MLX5_FLOW_ACTION_OF_PUSH_VLAN | \
 549          MLX5_FLOW_ACTION_OF_SET_VLAN_VID | MLX5_FLOW_ACTION_OF_SET_VLAN_PCP)
 550
 551 #define MLX5_TCF_VXLAN_ACTIONS \
 552         (MLX5_FLOW_ACTION_VXLAN_ENCAP | MLX5_FLOW_ACTION_VXLAN_DECAP)
 553
 554 #define MLX5_TCF_PEDIT_ACTIONS \
 555         (MLX5_FLOW_ACTION_SET_IPV4_SRC | MLX5_FLOW_ACTION_SET_IPV4_DST | \
 556          MLX5_FLOW_ACTION_SET_IPV6_SRC | MLX5_FLOW_ACTION_SET_IPV6_DST | \
 557          MLX5_FLOW_ACTION_SET_TP_SRC | MLX5_FLOW_ACTION_SET_TP_DST | \
 558          MLX5_FLOW_ACTION_SET_TTL | MLX5_FLOW_ACTION_DEC_TTL | \
 559          MLX5_FLOW_ACTION_SET_MAC_SRC | MLX5_FLOW_ACTION_SET_MAC_DST)
 560
 561 #define MLX5_TCF_CONFIG_ACTIONS \
 562         (MLX5_FLOW_ACTION_PORT_ID | MLX5_FLOW_ACTION_JUMP | \
 563          MLX5_FLOW_ACTION_OF_PUSH_VLAN | MLX5_FLOW_ACTION_OF_SET_VLAN_VID | \
 564          MLX5_FLOW_ACTION_OF_SET_VLAN_PCP | \
 565          (MLX5_TCF_PEDIT_ACTIONS & ~MLX5_FLOW_ACTION_DEC_TTL))
 566
 567 #define MAX_PEDIT_KEYS 128
 568 #define SZ_PEDIT_KEY_VAL 4
 569
 570 #define NUM_OF_PEDIT_KEYS(sz) \
 571         (((sz) / SZ_PEDIT_KEY_VAL) + (((sz) % SZ_PEDIT_KEY_VAL) ? 1 : 0))
 572
 573 struct pedit_key_ex {
 574         enum pedit_header_type htype;
 575         enum pedit_cmd cmd;
 576 };
 577
 578 struct pedit_parser {
 579         struct tc_pedit_sel sel;
 580         struct tc_pedit_key keys[MAX_PEDIT_KEYS];
 581         struct pedit_key_ex keys_ex[MAX_PEDIT_KEYS];
 582 };
 583
 584 /**
 585  * Create space for using the implicitly created TC flow counter.
 586  *
 587  * @param[in] dev
 588  *   Pointer to the Ethernet device structure.
 589  *
 590  * @return
 591  *   A pointer to the counter data structure, NULL otherwise and
 592  *   rte_errno is set.
 593  */
 594 static struct mlx5_flow_counter *
 595 flow_tcf_counter_new(void)
 596 {
 597         struct mlx5_flow_counter *cnt;
 598
 599         /*
 600          * eswitch counter cannot be shared and its id is unknown.
 601          * currently returning all with id 0.
 602          * in the future maybe better to switch to unique numbers.
 603          */
 604         struct mlx5_flow_counter tmpl = {
 605                 .ref_cnt = 1,
 606         };
 607         cnt = rte_calloc(__func__, 1, sizeof(*cnt), 0);
 608         if (!cnt) {
 609                 rte_errno = ENOMEM;
 610                 return NULL;
 611         }
 612         *cnt = tmpl;
 613         /* Implicit counter, do not add to list. */
 614         return cnt;
 615 }
 616
 617 /**
 618  * Set pedit key of MAC address
 619  *
 620  * @param[in] actions
 621  *   pointer to action specification
 622  * @param[in,out] p_parser
 623  *   pointer to pedit_parser
 624  */
 625 static void
 626 flow_tcf_pedit_key_set_mac(const struct rte_flow_action *actions,
 627                            struct pedit_parser *p_parser)
 628 {
 629         int idx = p_parser->sel.nkeys;
 630         uint32_t off = actions->type == RTE_FLOW_ACTION_TYPE_SET_MAC_SRC ?
 631                                         offsetof(struct ether_hdr, s_addr) :
 632                                         offsetof(struct ether_hdr, d_addr);
 633         const struct rte_flow_action_set_mac *conf =
 634                 (const struct rte_flow_action_set_mac *)actions->conf;
 635
 636         p_parser->keys[idx].off = off;
 637         p_parser->keys[idx].mask = ~UINT32_MAX;
 638         p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_ETH;
 639         p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
 640         memcpy(&p_parser->keys[idx].val,
 641                 conf->mac_addr, SZ_PEDIT_KEY_VAL);
 642         idx++;
 643         p_parser->keys[idx].off = off + SZ_PEDIT_KEY_VAL;
 644         p_parser->keys[idx].mask = 0xFFFF0000;
 645         p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_ETH;
 646         p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
 647         memcpy(&p_parser->keys[idx].val,
 648                 conf->mac_addr + SZ_PEDIT_KEY_VAL,
 649                 ETHER_ADDR_LEN - SZ_PEDIT_KEY_VAL);
 650         p_parser->sel.nkeys = (++idx);
 651 }
 652
 653 /**
 654  * Set pedit key of decrease/set ttl
 655  *
 656  * @param[in] actions
 657  *   pointer to action specification
 658  * @param[in,out] p_parser
 659  *   pointer to pedit_parser
 660  * @param[in] item_flags
 661  *   flags of all items presented
 662  */
 663 static void
 664 flow_tcf_pedit_key_set_dec_ttl(const struct rte_flow_action *actions,
 665                                 struct pedit_parser *p_parser,
 666                                 uint64_t item_flags)
 667 {
 668         int idx = p_parser->sel.nkeys;
 669
 670         p_parser->keys[idx].mask = 0xFFFFFF00;
 671         if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4) {
 672                 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP4;
 673                 p_parser->keys[idx].off =
 674                         offsetof(struct ipv4_hdr, time_to_live);
 675         }
 676         if (item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6) {
 677                 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP6;
 678                 p_parser->keys[idx].off =
 679                         offsetof(struct ipv6_hdr, hop_limits);
 680         }
 681         if (actions->type == RTE_FLOW_ACTION_TYPE_DEC_TTL) {
 682                 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_ADD;
 683                 p_parser->keys[idx].val = 0x000000FF;
 684         } else {
 685                 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
 686                 p_parser->keys[idx].val =
 687                         (__u32)((const struct rte_flow_action_set_ttl *)
 688                          actions->conf)->ttl_value;
 689         }
 690         p_parser->sel.nkeys = (++idx);
 691 }
 692
 693 /**
 694  * Set pedit key of transport (TCP/UDP) port value
 695  *
 696  * @param[in] actions
 697  *   pointer to action specification
 698  * @param[in,out] p_parser
 699  *   pointer to pedit_parser
 700  * @param[in] item_flags
 701  *   flags of all items presented
 702  */
 703 static void
 704 flow_tcf_pedit_key_set_tp_port(const struct rte_flow_action *actions,
 705                                 struct pedit_parser *p_parser,
 706                                 uint64_t item_flags)
 707 {
 708         int idx = p_parser->sel.nkeys;
 709
 710         if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP)
 711                 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_UDP;
 712         if (item_flags & MLX5_FLOW_LAYER_OUTER_L4_TCP)
 713                 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_TCP;
 714         p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
 715         /* offset of src/dst port is same for TCP and UDP */
 716         p_parser->keys[idx].off =
 717                 actions->type == RTE_FLOW_ACTION_TYPE_SET_TP_SRC ?
 718                 offsetof(struct tcp_hdr, src_port) :
 719                 offsetof(struct tcp_hdr, dst_port);
 720         p_parser->keys[idx].mask = 0xFFFF0000;
 721         p_parser->keys[idx].val =
 722                 (__u32)((const struct rte_flow_action_set_tp *)
 723                                 actions->conf)->port;
 724         p_parser->sel.nkeys = (++idx);
 725 }
 726
 727 /**
 728  * Set pedit key of ipv6 address
 729  *
 730  * @param[in] actions
 731  *   pointer to action specification
 732  * @param[in,out] p_parser
 733  *   pointer to pedit_parser
 734  */
 735 static void
 736 flow_tcf_pedit_key_set_ipv6_addr(const struct rte_flow_action *actions,
 737                                  struct pedit_parser *p_parser)
 738 {
 739         int idx = p_parser->sel.nkeys;
 740         int keys = NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
 741         int off_base =
 742                 actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC ?
 743                 offsetof(struct ipv6_hdr, src_addr) :
 744                 offsetof(struct ipv6_hdr, dst_addr);
 745         const struct rte_flow_action_set_ipv6 *conf =
 746                 (const struct rte_flow_action_set_ipv6 *)actions->conf;
 747
 748         for (int i = 0; i < keys; i++, idx++) {
 749                 p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP6;
 750                 p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
 751                 p_parser->keys[idx].off = off_base + i * SZ_PEDIT_KEY_VAL;
 752                 p_parser->keys[idx].mask = ~UINT32_MAX;
 753                 memcpy(&p_parser->keys[idx].val,
 754                         conf->ipv6_addr + i *  SZ_PEDIT_KEY_VAL,
 755                         SZ_PEDIT_KEY_VAL);
 756         }
 757         p_parser->sel.nkeys += keys;
 758 }
 759
 760 /**
 761  * Set pedit key of ipv4 address
 762  *
 763  * @param[in] actions
 764  *   pointer to action specification
 765  * @param[in,out] p_parser
 766  *   pointer to pedit_parser
 767  */
 768 static void
 769 flow_tcf_pedit_key_set_ipv4_addr(const struct rte_flow_action *actions,
 770                                  struct pedit_parser *p_parser)
 771 {
 772         int idx = p_parser->sel.nkeys;
 773
 774         p_parser->keys_ex[idx].htype = TCA_PEDIT_KEY_EX_HDR_TYPE_IP4;
 775         p_parser->keys_ex[idx].cmd = TCA_PEDIT_KEY_EX_CMD_SET;
 776         p_parser->keys[idx].off =
 777                 actions->type == RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC ?
 778                 offsetof(struct ipv4_hdr, src_addr) :
 779                 offsetof(struct ipv4_hdr, dst_addr);
 780         p_parser->keys[idx].mask = ~UINT32_MAX;
 781         p_parser->keys[idx].val =
 782                 ((const struct rte_flow_action_set_ipv4 *)
 783                  actions->conf)->ipv4_addr;
 784         p_parser->sel.nkeys = (++idx);
 785 }
 786
 787 /**
 788  * Create the pedit's na attribute in netlink message
 789  * on pre-allocate message buffer
 790  *
 791  * @param[in,out] nl
 792  *   pointer to pre-allocated netlink message buffer
 793  * @param[in,out] actions
 794  *   pointer to pointer of actions specification.
 795  * @param[in,out] action_flags
 796  *   pointer to actions flags
 797  * @param[in] item_flags
 798  *   flags of all item presented
 799  */
 800 static void
 801 flow_tcf_create_pedit_mnl_msg(struct nlmsghdr *nl,
 802                               const struct rte_flow_action **actions,
 803                               uint64_t item_flags)
 804 {
 805         struct pedit_parser p_parser;
 806         struct nlattr *na_act_options;
 807         struct nlattr *na_pedit_keys;
 808
 809         memset(&p_parser, 0, sizeof(p_parser));
 810         mnl_attr_put_strz(nl, TCA_ACT_KIND, "pedit");
 811         na_act_options = mnl_attr_nest_start(nl, TCA_ACT_OPTIONS);
 812         /* all modify header actions should be in one tc-pedit action */
 813         for (; (*actions)->type != RTE_FLOW_ACTION_TYPE_END; (*actions)++) {
 814                 switch ((*actions)->type) {
 815                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
 816                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
 817                         flow_tcf_pedit_key_set_ipv4_addr(*actions, &p_parser);
 818                         break;
 819                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
 820                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
 821                         flow_tcf_pedit_key_set_ipv6_addr(*actions, &p_parser);
 822                         break;
 823                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
 824                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
 825                         flow_tcf_pedit_key_set_tp_port(*actions,
 826                                                         &p_parser, item_flags);
 827                         break;
 828                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
 829                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
 830                         flow_tcf_pedit_key_set_dec_ttl(*actions,
 831                                                         &p_parser, item_flags);
 832                         break;
 833                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
 834                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
 835                         flow_tcf_pedit_key_set_mac(*actions, &p_parser);
 836                         break;
 837                 default:
 838                         goto pedit_mnl_msg_done;
 839                 }
 840         }
 841 pedit_mnl_msg_done:
 842         p_parser.sel.action = TC_ACT_PIPE;
 843         mnl_attr_put(nl, TCA_PEDIT_PARMS_EX,
 844                      sizeof(p_parser.sel) +
 845                      p_parser.sel.nkeys * sizeof(struct tc_pedit_key),
 846                      &p_parser);
 847         na_pedit_keys =
 848                 mnl_attr_nest_start(nl, TCA_PEDIT_KEYS_EX | NLA_F_NESTED);
 849         for (int i = 0; i < p_parser.sel.nkeys; i++) {
 850                 struct nlattr *na_pedit_key =
 851                         mnl_attr_nest_start(nl,
 852                                             TCA_PEDIT_KEY_EX | NLA_F_NESTED);
 853                 mnl_attr_put_u16(nl, TCA_PEDIT_KEY_EX_HTYPE,
 854                                  p_parser.keys_ex[i].htype);
 855                 mnl_attr_put_u16(nl, TCA_PEDIT_KEY_EX_CMD,
 856                                  p_parser.keys_ex[i].cmd);
 857                 mnl_attr_nest_end(nl, na_pedit_key);
 858         }
 859         mnl_attr_nest_end(nl, na_pedit_keys);
 860         mnl_attr_nest_end(nl, na_act_options);
 861         (*actions)--;
 862 }
 863
 864 /**
 865  * Calculate max memory size of one TC-pedit actions.
 866  * One TC-pedit action can contain set of keys each defining
 867  * a rewrite element (rte_flow action)
 868  *
 869  * @param[in,out] actions
 870  *   actions specification.
 871  * @param[in,out] action_flags
 872  *   actions flags
 873  * @param[in,out] size
 874  *   accumulated size
 875  * @return
 876  *   Max memory size of one TC-pedit action
 877  */
 878 static int
 879 flow_tcf_get_pedit_actions_size(const struct rte_flow_action **actions,
 880                                 uint64_t *action_flags)
 881 {
 882         int pedit_size = 0;
 883         int keys = 0;
 884         uint64_t flags = 0;
 885
 886         pedit_size += SZ_NLATTR_NEST + /* na_act_index. */
 887                       SZ_NLATTR_STRZ_OF("pedit") +
 888                       SZ_NLATTR_NEST; /* TCA_ACT_OPTIONS. */
 889         for (; (*actions)->type != RTE_FLOW_ACTION_TYPE_END; (*actions)++) {
 890                 switch ((*actions)->type) {
 891                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
 892                         keys += NUM_OF_PEDIT_KEYS(IPV4_ADDR_LEN);
 893                         flags |= MLX5_FLOW_ACTION_SET_IPV4_SRC;
 894                         break;
 895                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
 896                         keys += NUM_OF_PEDIT_KEYS(IPV4_ADDR_LEN);
 897                         flags |= MLX5_FLOW_ACTION_SET_IPV4_DST;
 898                         break;
 899                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
 900                         keys += NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
 901                         flags |= MLX5_FLOW_ACTION_SET_IPV6_SRC;
 902                         break;
 903                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
 904                         keys += NUM_OF_PEDIT_KEYS(IPV6_ADDR_LEN);
 905                         flags |= MLX5_FLOW_ACTION_SET_IPV6_DST;
 906                         break;
 907                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
 908                         /* TCP is as same as UDP */
 909                         keys += NUM_OF_PEDIT_KEYS(TP_PORT_LEN);
 910                         flags |= MLX5_FLOW_ACTION_SET_TP_SRC;
 911                         break;
 912                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
 913                         /* TCP is as same as UDP */
 914                         keys += NUM_OF_PEDIT_KEYS(TP_PORT_LEN);
 915                         flags |= MLX5_FLOW_ACTION_SET_TP_DST;
 916                         break;
 917                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
 918                         keys += NUM_OF_PEDIT_KEYS(TTL_LEN);
 919                         flags |= MLX5_FLOW_ACTION_SET_TTL;
 920                         break;
 921                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
 922                         keys += NUM_OF_PEDIT_KEYS(TTL_LEN);
 923                         flags |= MLX5_FLOW_ACTION_DEC_TTL;
 924                         break;
 925                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
 926                         keys += NUM_OF_PEDIT_KEYS(ETHER_ADDR_LEN);
 927                         flags |= MLX5_FLOW_ACTION_SET_MAC_SRC;
 928                         break;
 929                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
 930                         keys += NUM_OF_PEDIT_KEYS(ETHER_ADDR_LEN);
 931                         flags |= MLX5_FLOW_ACTION_SET_MAC_DST;
 932                         break;
 933                 default:
 934                         goto get_pedit_action_size_done;
 935                 }
 936         }
 937 get_pedit_action_size_done:
 938         /* TCA_PEDIT_PARAMS_EX */
 939         pedit_size +=
 940                 SZ_NLATTR_DATA_OF(sizeof(struct tc_pedit_sel) +
 941                                   keys * sizeof(struct tc_pedit_key));
 942         pedit_size += SZ_NLATTR_NEST; /* TCA_PEDIT_KEYS */
 943         pedit_size += keys *
 944                       /* TCA_PEDIT_KEY_EX + HTYPE + CMD */
 945                       (SZ_NLATTR_NEST + SZ_NLATTR_DATA_OF(2) +
 946                        SZ_NLATTR_DATA_OF(2));
 947         (*action_flags) |= flags;
 948         (*actions)--;
 949         return pedit_size;
 950 }
 951
 952 /**
 953  * Retrieve mask for pattern item.
 954  *
 955  * This function does basic sanity checks on a pattern item in order to
 956  * return the most appropriate mask for it.
 957  *
 958  * @param[in] item
 959  *   Item specification.
 960  * @param[in] mask_default
 961  *   Default mask for pattern item as specified by the flow API.
 962  * @param[in] mask_supported
 963  *   Mask fields supported by the implementation.
 964  * @param[in] mask_empty
 965  *   Empty mask to return when there is no specification.
 966  * @param[out] error
 967  *   Perform verbose error reporting if not NULL.
 968  *
 969  * @return
 970  *   Either @p item->mask or one of the mask parameters on success, NULL
 971  *   otherwise and rte_errno is set.
 972  */
 973 static const void *
 974 flow_tcf_item_mask(const struct rte_flow_item *item, const void *mask_default,
 975                    const void *mask_supported, const void *mask_empty,
 976                    size_t mask_size, struct rte_flow_error *error)
 977 {
 978         const uint8_t *mask;
 979         size_t i;
 980
 981         /* item->last and item->mask cannot exist without item->spec. */
 982         if (!item->spec && (item->mask || item->last)) {
 983                 rte_flow_error_set(error, EINVAL,
 984                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
 985                                    "\"mask\" or \"last\" field provided without"
 986                                    " a corresponding \"spec\"");
 987                 return NULL;
 988         }
 989         /* No spec, no mask, no problem. */
 990         if (!item->spec)
 991                 return mask_empty;
 992         mask = item->mask ? item->mask : mask_default;
 993         assert(mask);
 994         /*
 995          * Single-pass check to make sure that:
 996          * - Mask is supported, no bits are set outside mask_supported.
 997          * - Both item->spec and item->last are included in mask.
 998          */
 999         for (i = 0; i != mask_size; ++i) {
1000                 if (!mask[i])
1001                         continue;
1002                 if ((mask[i] | ((const uint8_t *)mask_supported)[i]) !=
1003                     ((const uint8_t *)mask_supported)[i]) {
1004                         rte_flow_error_set(error, ENOTSUP,
1005                                            RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1006                                            "unsupported field found"
1007                                            " in \"mask\"");
1008                         return NULL;
1009                 }
1010                 if (item->last &&
1011                     (((const uint8_t *)item->spec)[i] & mask[i]) !=
1012                     (((const uint8_t *)item->last)[i] & mask[i])) {
1013                         rte_flow_error_set(error, EINVAL,
1014                                            RTE_FLOW_ERROR_TYPE_ITEM_LAST,
1015                                            item->last,
1016                                            "range between \"spec\" and \"last\""
1017                                            " not comprised in \"mask\"");
1018                         return NULL;
1019                 }
1020         }
1021         return mask;
1022 }
1023
1024 /**
1025  * Build a conversion table between port ID and ifindex.
1026  *
1027  * @param[in] dev
1028  *   Pointer to Ethernet device.
1029  * @param[out] ptoi
1030  *   Pointer to ptoi table.
1031  * @param[in] len
1032  *   Size of ptoi table provided.
1033  *
1034  * @return
1035  *   Size of ptoi table filled.
1036  */
1037 static unsigned int
1038 flow_tcf_build_ptoi_table(struct rte_eth_dev *dev, struct flow_tcf_ptoi *ptoi,
1039                           unsigned int len)
1040 {
1041         unsigned int n = mlx5_dev_to_port_id(dev->device, NULL, 0);
1042         uint16_t port_id[n + 1];
1043         unsigned int i;
1044         unsigned int own = 0;
1045
1046         /* At least one port is needed when no switch domain is present. */
1047         if (!n) {
1048                 n = 1;
1049                 port_id[0] = dev->data->port_id;
1050         } else {
1051                 n = RTE_MIN(mlx5_dev_to_port_id(dev->device, port_id, n), n);
1052         }
1053         if (n > len)
1054                 return 0;
1055         for (i = 0; i != n; ++i) {
1056                 struct rte_eth_dev_info dev_info;
1057
1058                 rte_eth_dev_info_get(port_id[i], &dev_info);
1059                 if (port_id[i] == dev->data->port_id)
1060                         own = i;
1061                 ptoi[i].port_id = port_id[i];
1062                 ptoi[i].ifindex = dev_info.if_index;
1063         }
1064         /* Ensure first entry of ptoi[] is the current device. */
1065         if (own) {
1066                 ptoi[n] = ptoi[0];
1067                 ptoi[0] = ptoi[own];
1068                 ptoi[own] = ptoi[n];
1069         }
1070         /* An entry with zero ifindex terminates ptoi[]. */
1071         ptoi[n].port_id = 0;
1072         ptoi[n].ifindex = 0;
1073         return n;
1074 }
1075
1076 /**
1077  * Verify the @p attr will be correctly understood by the E-switch.
1078  *
1079  * @param[in] attr
1080  *   Pointer to flow attributes
1081  * @param[out] error
1082  *   Pointer to error structure.
1083  *
1084  * @return
1085  *   0 on success, a negative errno value otherwise and rte_errno is set.
1086  */
1087 static int
1088 flow_tcf_validate_attributes(const struct rte_flow_attr *attr,
1089                              struct rte_flow_error *error)
1090 {
1091         /*
1092          * Supported attributes: groups, some priorities and ingress only.
1093          * group is supported only if kernel supports chain. Don't care about
1094          * transfer as it is the caller's problem.
1095          */
1096         if (attr->group > MLX5_TCF_GROUP_ID_MAX)
1097                 return rte_flow_error_set(error, ENOTSUP,
1098                                           RTE_FLOW_ERROR_TYPE_ATTR_GROUP, attr,
1099                                           "group ID larger than "
1100                                           RTE_STR(MLX5_TCF_GROUP_ID_MAX)
1101                                           " isn't supported");
1102         else if (attr->priority > MLX5_TCF_GROUP_PRIORITY_MAX)
1103                 return rte_flow_error_set(error, ENOTSUP,
1104                                           RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY,
1105                                           attr,
1106                                           "priority more than "
1107                                           RTE_STR(MLX5_TCF_GROUP_PRIORITY_MAX)
1108                                           " is not supported");
1109         if (!attr->ingress)
1110                 return rte_flow_error_set(error, EINVAL,
1111                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1112                                           attr, "only ingress is supported");
1113         if (attr->egress)
1114                 return rte_flow_error_set(error, ENOTSUP,
1115                                           RTE_FLOW_ERROR_TYPE_ATTR_INGRESS,
1116                                           attr, "egress is not supported");
1117         return 0;
1118 }
1119
1120 /**
1121  * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_ETH item for E-Switch.
1122  * The routine checks the L2 fields to be used in encapsulation header.
1123  *
1124  * @param[in] item
1125  *   Pointer to the item structure.
1126  * @param[out] error
1127  *   Pointer to the error structure.
1128  *
1129  * @return
1130  *   0 on success, a negative errno value otherwise and rte_errno is set.
1131  **/
1132 static int
1133 flow_tcf_validate_vxlan_encap_eth(const struct rte_flow_item *item,
1134                                   struct rte_flow_error *error)
1135 {
1136         const struct rte_flow_item_eth *spec = item->spec;
1137         const struct rte_flow_item_eth *mask = item->mask;
1138
1139         if (!spec) {
1140                 /*
1141                  * Specification for L2 addresses can be empty
1142                  * because these ones are optional and not
1143                  * required directly by tc rule. Kernel tries
1144                  * to resolve these ones on its own
1145                  */
1146                 return 0;
1147         }
1148         if (!mask) {
1149                 /* If mask is not specified use the default one. */
1150                 mask = &rte_flow_item_eth_mask;
1151         }
1152         if (memcmp(&mask->dst,
1153                    &flow_tcf_mask_empty.eth.dst,
1154                    sizeof(flow_tcf_mask_empty.eth.dst))) {
1155                 if (memcmp(&mask->dst,
1156                            &rte_flow_item_eth_mask.dst,
1157                            sizeof(rte_flow_item_eth_mask.dst)))
1158                         return rte_flow_error_set
1159                                 (error, ENOTSUP,
1160                                  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1161                                  "no support for partial mask on"
1162                                  " \"eth.dst\" field");
1163         }
1164         if (memcmp(&mask->src,
1165                    &flow_tcf_mask_empty.eth.src,
1166                    sizeof(flow_tcf_mask_empty.eth.src))) {
1167                 if (memcmp(&mask->src,
1168                            &rte_flow_item_eth_mask.src,
1169                            sizeof(rte_flow_item_eth_mask.src)))
1170                         return rte_flow_error_set
1171                                 (error, ENOTSUP,
1172                                  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1173                                  "no support for partial mask on"
1174                                  " \"eth.src\" field");
1175         }
1176         if (mask->type != RTE_BE16(0x0000)) {
1177                 if (mask->type != RTE_BE16(0xffff))
1178                         return rte_flow_error_set
1179                                 (error, ENOTSUP,
1180                                  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1181                                  "no support for partial mask on"
1182                                  " \"eth.type\" field");
1183                 DRV_LOG(WARNING,
1184                         "outer ethernet type field"
1185                         " cannot be forced for vxlan"
1186                         " encapsulation, parameter ignored");
1187         }
1188         return 0;
1189 }
1190
1191 /**
1192  * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_IPV4 item for E-Switch.
1193  * The routine checks the IPv4 fields to be used in encapsulation header.
1194  *
1195  * @param[in] item
1196  *   Pointer to the item structure.
1197  * @param[out] error
1198  *   Pointer to the error structure.
1199  *
1200  * @return
1201  *   0 on success, a negative errno value otherwise and rte_errno is set.
1202  **/
1203 static int
1204 flow_tcf_validate_vxlan_encap_ipv4(const struct rte_flow_item *item,
1205                                    struct rte_flow_error *error)
1206 {
1207         const struct rte_flow_item_ipv4 *spec = item->spec;
1208         const struct rte_flow_item_ipv4 *mask = item->mask;
1209
1210         if (!spec) {
1211                 /*
1212                  * Specification for IP addresses cannot be empty
1213                  * because it is required by tunnel_key parameter.
1214                  */
1215                 return rte_flow_error_set(error, EINVAL,
1216                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1217                                           "NULL outer ipv4 address"
1218                                           " specification for vxlan"
1219                                           " encapsulation");
1220         }
1221         if (!mask)
1222                 mask = &rte_flow_item_ipv4_mask;
1223         if (mask->hdr.dst_addr != RTE_BE32(0x00000000)) {
1224                 if (mask->hdr.dst_addr != RTE_BE32(0xffffffff))
1225                         return rte_flow_error_set
1226                                 (error, ENOTSUP,
1227                                  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1228                                  "no support for partial mask on"
1229                                  " \"ipv4.hdr.dst_addr\" field"
1230                                  " for vxlan encapsulation");
1231                 /* More IPv4 address validations can be put here. */
1232         } else {
1233                 /*
1234                  * Kernel uses the destination IP address to determine
1235                  * the routing path and obtain the MAC destination
1236                  * address, so IP destination address must be
1237                  * specified in the tc rule.
1238                  */
1239                 return rte_flow_error_set(error, EINVAL,
1240                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1241                                           "outer ipv4 destination address"
1242                                           " must be specified for"
1243                                           " vxlan encapsulation");
1244         }
1245         if (mask->hdr.src_addr != RTE_BE32(0x00000000)) {
1246                 if (mask->hdr.src_addr != RTE_BE32(0xffffffff))
1247                         return rte_flow_error_set
1248                                 (error, ENOTSUP,
1249                                  RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1250                                  "no support for partial mask on"
1251                                  " \"ipv4.hdr.src_addr\" field"
1252                                  " for vxlan encapsulation");
1253                 /* More IPv4 address validations can be put here. */
1254         } else {
1255                 /*
1256                  * Kernel uses the source IP address to select the
1257                  * interface for egress encapsulated traffic, so
1258                  * it must be specified in the tc rule.
1259                  */
1260                 return rte_flow_error_set(error, EINVAL,
1261                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1262                                           "outer ipv4 source address"
1263                                           " must be specified for"
1264                                           " vxlan encapsulation");
1265         }
1266         return 0;
1267 }
1268
1269 /**
1270  * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_IPV6 item for E-Switch.
1271  * The routine checks the IPv6 fields to be used in encapsulation header.
1272  *
1273  * @param[in] item
1274  *   Pointer to the item structure.
1275  * @param[out] error
1276  *   Pointer to the error structure.
1277  *
1278  * @return
1279  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1280  **/
1281 static int
1282 flow_tcf_validate_vxlan_encap_ipv6(const struct rte_flow_item *item,
1283                                    struct rte_flow_error *error)
1284 {
1285         const struct rte_flow_item_ipv6 *spec = item->spec;
1286         const struct rte_flow_item_ipv6 *mask = item->mask;
1287
1288         if (!spec) {
1289                 /*
1290                  * Specification for IP addresses cannot be empty
1291                  * because it is required by tunnel_key parameter.
1292                  */
1293                 return rte_flow_error_set(error, EINVAL,
1294                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1295                                           "NULL outer ipv6 address"
1296                                           " specification for"
1297                                           " vxlan encapsulation");
1298         }
1299         if (!mask)
1300                 mask = &rte_flow_item_ipv6_mask;
1301         if (memcmp(&mask->hdr.dst_addr,
1302                    &flow_tcf_mask_empty.ipv6.hdr.dst_addr,
1303                    IPV6_ADDR_LEN)) {
1304                 if (memcmp(&mask->hdr.dst_addr,
1305                            &rte_flow_item_ipv6_mask.hdr.dst_addr,
1306                            IPV6_ADDR_LEN))
1307                         return rte_flow_error_set
1308                                         (error, ENOTSUP,
1309                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1310                                          "no support for partial mask on"
1311                                          " \"ipv6.hdr.dst_addr\" field"
1312                                          " for vxlan encapsulation");
1313                 /* More IPv6 address validations can be put here. */
1314         } else {
1315                 /*
1316                  * Kernel uses the destination IP address to determine
1317                  * the routing path and obtain the MAC destination
1318                  * address (heigh or gate), so IP destination address
1319                  * must be specified within the tc rule.
1320                  */
1321                 return rte_flow_error_set(error, EINVAL,
1322                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1323                                           "outer ipv6 destination address"
1324                                           " must be specified for"
1325                                           " vxlan encapsulation");
1326         }
1327         if (memcmp(&mask->hdr.src_addr,
1328                    &flow_tcf_mask_empty.ipv6.hdr.src_addr,
1329                    IPV6_ADDR_LEN)) {
1330                 if (memcmp(&mask->hdr.src_addr,
1331                            &rte_flow_item_ipv6_mask.hdr.src_addr,
1332                            IPV6_ADDR_LEN))
1333                         return rte_flow_error_set
1334                                         (error, ENOTSUP,
1335                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1336                                          "no support for partial mask on"
1337                                          " \"ipv6.hdr.src_addr\" field"
1338                                          " for vxlan encapsulation");
1339                 /* More L3 address validation can be put here. */
1340         } else {
1341                 /*
1342                  * Kernel uses the source IP address to select the
1343                  * interface for egress encapsulated traffic, so
1344                  * it must be specified in the tc rule.
1345                  */
1346                 return rte_flow_error_set(error, EINVAL,
1347                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1348                                           "outer L3 source address"
1349                                           " must be specified for"
1350                                           " vxlan encapsulation");
1351         }
1352         return 0;
1353 }
1354
1355 /**
1356  * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_UDP item for E-Switch.
1357  * The routine checks the UDP fields to be used in encapsulation header.
1358  *
1359  * @param[in] item
1360  *   Pointer to the item structure.
1361  * @param[out] error
1362  *   Pointer to the error structure.
1363  *
1364  * @return
1365  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1366  **/
1367 static int
1368 flow_tcf_validate_vxlan_encap_udp(const struct rte_flow_item *item,
1369                                   struct rte_flow_error *error)
1370 {
1371         const struct rte_flow_item_udp *spec = item->spec;
1372         const struct rte_flow_item_udp *mask = item->mask;
1373
1374         if (!spec) {
1375                 /*
1376                  * Specification for UDP ports cannot be empty
1377                  * because it is required by tunnel_key parameter.
1378                  */
1379                 return rte_flow_error_set(error, EINVAL,
1380                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1381                                           "NULL UDP port specification "
1382                                           " for vxlan encapsulation");
1383         }
1384         if (!mask)
1385                 mask = &rte_flow_item_udp_mask;
1386         if (mask->hdr.dst_port != RTE_BE16(0x0000)) {
1387                 if (mask->hdr.dst_port != RTE_BE16(0xffff))
1388                         return rte_flow_error_set
1389                                         (error, ENOTSUP,
1390                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1391                                          "no support for partial mask on"
1392                                          " \"udp.hdr.dst_port\" field"
1393                                          " for vxlan encapsulation");
1394                 if (!spec->hdr.dst_port)
1395                         return rte_flow_error_set
1396                                         (error, EINVAL,
1397                                          RTE_FLOW_ERROR_TYPE_ITEM, item,
1398                                          "outer UDP remote port cannot be"
1399                                          " 0 for vxlan encapsulation");
1400         } else {
1401                 return rte_flow_error_set(error, EINVAL,
1402                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1403                                           "outer UDP remote port"
1404                                           " must be specified for"
1405                                           " vxlan encapsulation");
1406         }
1407         if (mask->hdr.src_port != RTE_BE16(0x0000)) {
1408                 if (mask->hdr.src_port != RTE_BE16(0xffff))
1409                         return rte_flow_error_set
1410                                         (error, ENOTSUP,
1411                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1412                                          "no support for partial mask on"
1413                                          " \"udp.hdr.src_port\" field"
1414                                          " for vxlan encapsulation");
1415                 DRV_LOG(WARNING,
1416                         "outer UDP source port cannot be"
1417                         " forced for vxlan encapsulation,"
1418                         " parameter ignored");
1419         }
1420         return 0;
1421 }
1422
1423 /**
1424  * Validate VXLAN_ENCAP action RTE_FLOW_ITEM_TYPE_VXLAN item for E-Switch.
1425  * The routine checks the VNIP fields to be used in encapsulation header.
1426  *
1427  * @param[in] item
1428  *   Pointer to the item structure.
1429  * @param[out] error
1430  *   Pointer to the error structure.
1431  *
1432  * @return
1433  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1434  **/
1435 static int
1436 flow_tcf_validate_vxlan_encap_vni(const struct rte_flow_item *item,
1437                                   struct rte_flow_error *error)
1438 {
1439         const struct rte_flow_item_vxlan *spec = item->spec;
1440         const struct rte_flow_item_vxlan *mask = item->mask;
1441
1442         if (!spec) {
1443                 /* Outer VNI is required by tunnel_key parameter. */
1444                 return rte_flow_error_set(error, EINVAL,
1445                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1446                                           "NULL VNI specification"
1447                                           " for vxlan encapsulation");
1448         }
1449         if (!mask)
1450                 mask = &rte_flow_item_vxlan_mask;
1451         if (!mask->vni[0] && !mask->vni[1] && !mask->vni[2])
1452                 return rte_flow_error_set(error, EINVAL,
1453                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1454                                           "outer VNI must be specified "
1455                                           "for vxlan encapsulation");
1456         if (mask->vni[0] != 0xff ||
1457             mask->vni[1] != 0xff ||
1458             mask->vni[2] != 0xff)
1459                 return rte_flow_error_set(error, ENOTSUP,
1460                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1461                                           "no support for partial mask on"
1462                                           " \"vxlan.vni\" field");
1463
1464         if (!spec->vni[0] && !spec->vni[1] && !spec->vni[2])
1465                 return rte_flow_error_set(error, EINVAL,
1466                                           RTE_FLOW_ERROR_TYPE_ITEM, item,
1467                                           "vxlan vni cannot be 0");
1468         return 0;
1469 }
1470
1471 /**
1472  * Validate VXLAN_ENCAP action item list for E-Switch.
1473  * The routine checks items to be used in encapsulation header.
1474  *
1475  * @param[in] action
1476  *   Pointer to the VXLAN_ENCAP action structure.
1477  * @param[out] error
1478  *   Pointer to the error structure.
1479  *
1480  * @return
1481  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1482  **/
1483 static int
1484 flow_tcf_validate_vxlan_encap(const struct rte_flow_action *action,
1485                               struct rte_flow_error *error)
1486 {
1487         const struct rte_flow_item *items;
1488         int ret;
1489         uint32_t item_flags = 0;
1490
1491         if (!action->conf)
1492                 return rte_flow_error_set(error, EINVAL,
1493                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1494                                           "Missing vxlan tunnel"
1495                                           " action configuration");
1496         items = ((const struct rte_flow_action_vxlan_encap *)
1497                                         action->conf)->definition;
1498         if (!items)
1499                 return rte_flow_error_set(error, EINVAL,
1500                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1501                                           "Missing vxlan tunnel"
1502                                           " encapsulation parameters");
1503         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1504                 switch (items->type) {
1505                 case RTE_FLOW_ITEM_TYPE_VOID:
1506                         break;
1507                 case RTE_FLOW_ITEM_TYPE_ETH:
1508                         ret = mlx5_flow_validate_item_eth(items, item_flags,
1509                                                           error);
1510                         if (ret < 0)
1511                                 return ret;
1512                         ret = flow_tcf_validate_vxlan_encap_eth(items, error);
1513                         if (ret < 0)
1514                                 return ret;
1515                         item_flags |= MLX5_FLOW_LAYER_OUTER_L2;
1516                         break;
1517                 break;
1518                 case RTE_FLOW_ITEM_TYPE_IPV4:
1519                         ret = mlx5_flow_validate_item_ipv4(items, item_flags,
1520                                                            error);
1521                         if (ret < 0)
1522                                 return ret;
1523                         ret = flow_tcf_validate_vxlan_encap_ipv4(items, error);
1524                         if (ret < 0)
1525                                 return ret;
1526                         item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
1527                         break;
1528                 case RTE_FLOW_ITEM_TYPE_IPV6:
1529                         ret = mlx5_flow_validate_item_ipv6(items, item_flags,
1530                                                            error);
1531                         if (ret < 0)
1532                                 return ret;
1533                         ret = flow_tcf_validate_vxlan_encap_ipv6(items, error);
1534                         if (ret < 0)
1535                                 return ret;
1536                         item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
1537                         break;
1538                 case RTE_FLOW_ITEM_TYPE_UDP:
1539                         ret = mlx5_flow_validate_item_udp(items, item_flags,
1540                                                            0xFF, error);
1541                         if (ret < 0)
1542                                 return ret;
1543                         ret = flow_tcf_validate_vxlan_encap_udp(items, error);
1544                         if (ret < 0)
1545                                 return ret;
1546                         item_flags |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
1547                         break;
1548                 case RTE_FLOW_ITEM_TYPE_VXLAN:
1549                         ret = mlx5_flow_validate_item_vxlan(items,
1550                                                             item_flags, error);
1551                         if (ret < 0)
1552                                 return ret;
1553                         ret = flow_tcf_validate_vxlan_encap_vni(items, error);
1554                         if (ret < 0)
1555                                 return ret;
1556                         item_flags |= MLX5_FLOW_LAYER_VXLAN;
1557                         break;
1558                 default:
1559                         return rte_flow_error_set
1560                                         (error, ENOTSUP,
1561                                          RTE_FLOW_ERROR_TYPE_ITEM, items,
1562                                          "vxlan encap item not supported");
1563                 }
1564         }
1565         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3))
1566                 return rte_flow_error_set(error, EINVAL,
1567                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1568                                           "no outer IP layer found"
1569                                           " for vxlan encapsulation");
1570         if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
1571                 return rte_flow_error_set(error, EINVAL,
1572                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1573                                           "no outer UDP layer found"
1574                                           " for vxlan encapsulation");
1575         if (!(item_flags & MLX5_FLOW_LAYER_VXLAN))
1576                 return rte_flow_error_set(error, EINVAL,
1577                                           RTE_FLOW_ERROR_TYPE_ACTION, action,
1578                                           "no VXLAN VNI found"
1579                                           " for vxlan encapsulation");
1580         return 0;
1581 }
1582
1583 /**
1584  * Validate RTE_FLOW_ITEM_TYPE_IPV4 item if VXLAN_DECAP action
1585  * is present in actions list.
1586  *
1587  * @param[in] ipv4
1588  *   Outer IPv4 address item (if any, NULL otherwise).
1589  * @param[out] error
1590  *   Pointer to the error structure.
1591  *
1592  * @return
1593  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1594  **/
1595 static int
1596 flow_tcf_validate_vxlan_decap_ipv4(const struct rte_flow_item *ipv4,
1597                                    struct rte_flow_error *error)
1598 {
1599         const struct rte_flow_item_ipv4 *spec = ipv4->spec;
1600         const struct rte_flow_item_ipv4 *mask = ipv4->mask;
1601
1602         if (!spec) {
1603                 /*
1604                  * Specification for IP addresses cannot be empty
1605                  * because it is required as decap parameter.
1606                  */
1607                 return rte_flow_error_set(error, EINVAL,
1608                                           RTE_FLOW_ERROR_TYPE_ITEM, ipv4,
1609                                           "NULL outer ipv4 address"
1610                                           " specification for vxlan"
1611                                           " for vxlan decapsulation");
1612         }
1613         if (!mask)
1614                 mask = &rte_flow_item_ipv4_mask;
1615         if (mask->hdr.dst_addr != RTE_BE32(0x00000000)) {
1616                 if (mask->hdr.dst_addr != RTE_BE32(0xffffffff))
1617                         return rte_flow_error_set
1618                                         (error, ENOTSUP,
1619                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1620                                          "no support for partial mask on"
1621                                          " \"ipv4.hdr.dst_addr\" field");
1622                 /* More IP address validations can be put here. */
1623         } else {
1624                 /*
1625                  * Kernel uses the destination IP address
1626                  * to determine the ingress network interface
1627                  * for traffic being decapsulated.
1628                  */
1629                 return rte_flow_error_set(error, EINVAL,
1630                                           RTE_FLOW_ERROR_TYPE_ITEM, ipv4,
1631                                           "outer ipv4 destination address"
1632                                           " must be specified for"
1633                                           " vxlan decapsulation");
1634         }
1635         /* Source IP address is optional for decap. */
1636         if (mask->hdr.src_addr != RTE_BE32(0x00000000) &&
1637             mask->hdr.src_addr != RTE_BE32(0xffffffff))
1638                 return rte_flow_error_set(error, ENOTSUP,
1639                                           RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1640                                           "no support for partial mask on"
1641                                           " \"ipv4.hdr.src_addr\" field");
1642         return 0;
1643 }
1644
1645 /**
1646  * Validate RTE_FLOW_ITEM_TYPE_IPV6 item if VXLAN_DECAP action
1647  * is present in actions list.
1648  *
1649  * @param[in] ipv6
1650  *   Outer IPv6 address item (if any, NULL otherwise).
1651  * @param[out] error
1652  *   Pointer to the error structure.
1653  *
1654  * @return
1655  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1656  **/
1657 static int
1658 flow_tcf_validate_vxlan_decap_ipv6(const struct rte_flow_item *ipv6,
1659                                    struct rte_flow_error *error)
1660 {
1661         const struct rte_flow_item_ipv6 *spec = ipv6->spec;
1662         const struct rte_flow_item_ipv6 *mask = ipv6->mask;
1663
1664         if (!spec) {
1665                 /*
1666                  * Specification for IP addresses cannot be empty
1667                  * because it is required as decap parameter.
1668                  */
1669                 return rte_flow_error_set(error, EINVAL,
1670                                           RTE_FLOW_ERROR_TYPE_ITEM, ipv6,
1671                                           "NULL outer ipv6 address"
1672                                           " specification for vxlan"
1673                                           " decapsulation");
1674         }
1675         if (!mask)
1676                 mask = &rte_flow_item_ipv6_mask;
1677         if (memcmp(&mask->hdr.dst_addr,
1678                    &flow_tcf_mask_empty.ipv6.hdr.dst_addr,
1679                    IPV6_ADDR_LEN)) {
1680                 if (memcmp(&mask->hdr.dst_addr,
1681                         &rte_flow_item_ipv6_mask.hdr.dst_addr,
1682                         IPV6_ADDR_LEN))
1683                         return rte_flow_error_set
1684                                         (error, ENOTSUP,
1685                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1686                                          "no support for partial mask on"
1687                                          " \"ipv6.hdr.dst_addr\" field");
1688                 /* More IP address validations can be put here. */
1689         } else {
1690                 /*
1691                  * Kernel uses the destination IP address
1692                  * to determine the ingress network interface
1693                  * for traffic being decapsulated.
1694                  */
1695                 return rte_flow_error_set(error, EINVAL,
1696                                           RTE_FLOW_ERROR_TYPE_ITEM, ipv6,
1697                                           "outer ipv6 destination address must be "
1698                                           "specified for vxlan decapsulation");
1699         }
1700         /* Source IP address is optional for decap. */
1701         if (memcmp(&mask->hdr.src_addr,
1702                    &flow_tcf_mask_empty.ipv6.hdr.src_addr,
1703                    IPV6_ADDR_LEN)) {
1704                 if (memcmp(&mask->hdr.src_addr,
1705                            &rte_flow_item_ipv6_mask.hdr.src_addr,
1706                            IPV6_ADDR_LEN))
1707                         return rte_flow_error_set
1708                                         (error, ENOTSUP,
1709                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1710                                          "no support for partial mask on"
1711                                          " \"ipv6.hdr.src_addr\" field");
1712         }
1713         return 0;
1714 }
1715
1716 /**
1717  * Validate RTE_FLOW_ITEM_TYPE_UDP item if VXLAN_DECAP action
1718  * is present in actions list.
1719  *
1720  * @param[in] udp
1721  *   Outer UDP layer item (if any, NULL otherwise).
1722  * @param[out] error
1723  *   Pointer to the error structure.
1724  *
1725  * @return
1726  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1727  **/
1728 static int
1729 flow_tcf_validate_vxlan_decap_udp(const struct rte_flow_item *udp,
1730                                   struct rte_flow_error *error)
1731 {
1732         const struct rte_flow_item_udp *spec = udp->spec;
1733         const struct rte_flow_item_udp *mask = udp->mask;
1734
1735         if (!spec)
1736                 /*
1737                  * Specification for UDP ports cannot be empty
1738                  * because it is required as decap parameter.
1739                  */
1740                 return rte_flow_error_set(error, EINVAL,
1741                                           RTE_FLOW_ERROR_TYPE_ITEM, udp,
1742                                           "NULL UDP port specification"
1743                                           " for VXLAN decapsulation");
1744         if (!mask)
1745                 mask = &rte_flow_item_udp_mask;
1746         if (mask->hdr.dst_port != RTE_BE16(0x0000)) {
1747                 if (mask->hdr.dst_port != RTE_BE16(0xffff))
1748                         return rte_flow_error_set
1749                                         (error, ENOTSUP,
1750                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1751                                          "no support for partial mask on"
1752                                          " \"udp.hdr.dst_port\" field");
1753                 if (!spec->hdr.dst_port)
1754                         return rte_flow_error_set
1755                                         (error, EINVAL,
1756                                          RTE_FLOW_ERROR_TYPE_ITEM, udp,
1757                                          "zero decap local UDP port");
1758         } else {
1759                 return rte_flow_error_set(error, EINVAL,
1760                                           RTE_FLOW_ERROR_TYPE_ITEM, udp,
1761                                           "outer UDP destination port must be "
1762                                           "specified for vxlan decapsulation");
1763         }
1764         if (mask->hdr.src_port != RTE_BE16(0x0000)) {
1765                 if (mask->hdr.src_port != RTE_BE16(0xffff))
1766                         return rte_flow_error_set
1767                                         (error, ENOTSUP,
1768                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK, mask,
1769                                          "no support for partial mask on"
1770                                          " \"udp.hdr.src_port\" field");
1771                 DRV_LOG(WARNING,
1772                         "outer UDP local port cannot be "
1773                         "forced for VXLAN encapsulation, "
1774                         "parameter ignored");
1775         }
1776         return 0;
1777 }
1778
1779 /**
1780  * Validate flow for E-Switch.
1781  *
1782  * @param[in] priv
1783  *   Pointer to the priv structure.
1784  * @param[in] attr
1785  *   Pointer to the flow attributes.
1786  * @param[in] items
1787  *   Pointer to the list of items.
1788  * @param[in] actions
1789  *   Pointer to the list of actions.
1790  * @param[out] error
1791  *   Pointer to the error structure.
1792  *
1793  * @return
1794  *   0 on success, a negative errno value otherwise and rte_ernno is set.
1795  */
1796 static int
1797 flow_tcf_validate(struct rte_eth_dev *dev,
1798                   const struct rte_flow_attr *attr,
1799                   const struct rte_flow_item items[],
1800                   const struct rte_flow_action actions[],
1801                   struct rte_flow_error *error)
1802 {
1803         union {
1804                 const struct rte_flow_item_port_id *port_id;
1805                 const struct rte_flow_item_eth *eth;
1806                 const struct rte_flow_item_vlan *vlan;
1807                 const struct rte_flow_item_ipv4 *ipv4;
1808                 const struct rte_flow_item_ipv6 *ipv6;
1809                 const struct rte_flow_item_tcp *tcp;
1810                 const struct rte_flow_item_udp *udp;
1811                 const struct rte_flow_item_vxlan *vxlan;
1812         } spec, mask;
1813         union {
1814                 const struct rte_flow_action_port_id *port_id;
1815                 const struct rte_flow_action_jump *jump;
1816                 const struct rte_flow_action_of_push_vlan *of_push_vlan;
1817                 const struct rte_flow_action_of_set_vlan_vid *
1818                         of_set_vlan_vid;
1819                 const struct rte_flow_action_of_set_vlan_pcp *
1820                         of_set_vlan_pcp;
1821                 const struct rte_flow_action_vxlan_encap *vxlan_encap;
1822                 const struct rte_flow_action_set_ipv4 *set_ipv4;
1823                 const struct rte_flow_action_set_ipv6 *set_ipv6;
1824         } conf;
1825         uint64_t item_flags = 0;
1826         uint64_t action_flags = 0;
1827         uint8_t next_protocol = -1;
1828         unsigned int tcm_ifindex = 0;
1829         uint8_t pedit_validated = 0;
1830         struct flow_tcf_ptoi ptoi[PTOI_TABLE_SZ_MAX(dev)];
1831         struct rte_eth_dev *port_id_dev = NULL;
1832         bool in_port_id_set;
1833         int ret;
1834
1835         claim_nonzero(flow_tcf_build_ptoi_table(dev, ptoi,
1836                                                 PTOI_TABLE_SZ_MAX(dev)));
1837         ret = flow_tcf_validate_attributes(attr, error);
1838         if (ret < 0)
1839                 return ret;
1840         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1841                 unsigned int i;
1842                 uint64_t current_action_flag = 0;
1843
1844                 switch (actions->type) {
1845                 case RTE_FLOW_ACTION_TYPE_VOID:
1846                         break;
1847                 case RTE_FLOW_ACTION_TYPE_PORT_ID:
1848                         current_action_flag = MLX5_FLOW_ACTION_PORT_ID;
1849                         if (!actions->conf)
1850                                 break;
1851                         conf.port_id = actions->conf;
1852                         if (conf.port_id->original)
1853                                 i = 0;
1854                         else
1855                                 for (i = 0; ptoi[i].ifindex; ++i)
1856                                         if (ptoi[i].port_id == conf.port_id->id)
1857                                                 break;
1858                         if (!ptoi[i].ifindex)
1859                                 return rte_flow_error_set
1860                                         (error, ENODEV,
1861                                          RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1862                                          conf.port_id,
1863                                          "missing data to convert port ID to"
1864                                          " ifindex");
1865                         port_id_dev = &rte_eth_devices[conf.port_id->id];
1866                         break;
1867                 case RTE_FLOW_ACTION_TYPE_JUMP:
1868                         current_action_flag = MLX5_FLOW_ACTION_JUMP;
1869                         if (!actions->conf)
1870                                 break;
1871                         conf.jump = actions->conf;
1872                         if (attr->group >= conf.jump->group)
1873                                 return rte_flow_error_set
1874                                         (error, ENOTSUP,
1875                                          RTE_FLOW_ERROR_TYPE_ACTION,
1876                                          actions,
1877                                          "can jump only to a group forward");
1878                         break;
1879                 case RTE_FLOW_ACTION_TYPE_DROP:
1880                         current_action_flag = MLX5_FLOW_ACTION_DROP;
1881                         break;
1882                 case RTE_FLOW_ACTION_TYPE_COUNT:
1883                         break;
1884                 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
1885                         current_action_flag = MLX5_FLOW_ACTION_OF_POP_VLAN;
1886                         break;
1887                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
1888                         current_action_flag = MLX5_FLOW_ACTION_OF_PUSH_VLAN;
1889                         break;
1890                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
1891                         if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
1892                                 return rte_flow_error_set
1893                                         (error, ENOTSUP,
1894                                          RTE_FLOW_ERROR_TYPE_ACTION, actions,
1895                                          "vlan modify is not supported,"
1896                                          " set action must follow push action");
1897                         current_action_flag = MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
1898                         break;
1899                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
1900                         if (!(action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN))
1901                                 return rte_flow_error_set
1902                                         (error, ENOTSUP,
1903                                          RTE_FLOW_ERROR_TYPE_ACTION, actions,
1904                                          "vlan modify is not supported,"
1905                                          " set action must follow push action");
1906                         current_action_flag = MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
1907                         break;
1908                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
1909                         current_action_flag = MLX5_FLOW_ACTION_VXLAN_DECAP;
1910                         break;
1911                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
1912                         ret = flow_tcf_validate_vxlan_encap(actions, error);
1913                         if (ret < 0)
1914                                 return ret;
1915                         current_action_flag = MLX5_FLOW_ACTION_VXLAN_ENCAP;
1916                         break;
1917                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
1918                         current_action_flag = MLX5_FLOW_ACTION_SET_IPV4_SRC;
1919                         break;
1920                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
1921                         current_action_flag = MLX5_FLOW_ACTION_SET_IPV4_DST;
1922                         break;
1923                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
1924                         current_action_flag = MLX5_FLOW_ACTION_SET_IPV6_SRC;
1925                         break;
1926                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
1927                         current_action_flag = MLX5_FLOW_ACTION_SET_IPV6_DST;
1928                         break;
1929                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
1930                         current_action_flag = MLX5_FLOW_ACTION_SET_TP_SRC;
1931                         break;
1932                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
1933                         current_action_flag = MLX5_FLOW_ACTION_SET_TP_DST;
1934                         break;
1935                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
1936                         current_action_flag = MLX5_FLOW_ACTION_SET_TTL;
1937                         break;
1938                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
1939                         current_action_flag = MLX5_FLOW_ACTION_DEC_TTL;
1940                         break;
1941                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
1942                         current_action_flag = MLX5_FLOW_ACTION_SET_MAC_SRC;
1943                         break;
1944                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
1945                         current_action_flag = MLX5_FLOW_ACTION_SET_MAC_DST;
1946                         break;
1947                 default:
1948                         return rte_flow_error_set(error, ENOTSUP,
1949                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1950                                                   actions,
1951                                                   "action not supported");
1952                 }
1953                 if (current_action_flag & MLX5_TCF_CONFIG_ACTIONS) {
1954                         if (!actions->conf)
1955                                 return rte_flow_error_set
1956                                         (error, EINVAL,
1957                                          RTE_FLOW_ERROR_TYPE_ACTION_CONF,
1958                                          actions,
1959                                          "action configuration not set");
1960                 }
1961                 if ((current_action_flag & MLX5_TCF_PEDIT_ACTIONS) &&
1962                     pedit_validated)
1963                         return rte_flow_error_set(error, ENOTSUP,
1964                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1965                                                   actions,
1966                                                   "set actions should be "
1967                                                   "listed successively");
1968                 if ((current_action_flag & ~MLX5_TCF_PEDIT_ACTIONS) &&
1969                     (action_flags & MLX5_TCF_PEDIT_ACTIONS))
1970                         pedit_validated = 1;
1971                 if ((current_action_flag & MLX5_TCF_FATE_ACTIONS) &&
1972                     (action_flags & MLX5_TCF_FATE_ACTIONS))
1973                         return rte_flow_error_set(error, EINVAL,
1974                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1975                                                   actions,
1976                                                   "can't have multiple fate"
1977                                                   " actions");
1978                 if ((current_action_flag & MLX5_TCF_VXLAN_ACTIONS) &&
1979                     (action_flags & MLX5_TCF_VXLAN_ACTIONS))
1980                         return rte_flow_error_set(error, EINVAL,
1981                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1982                                                   actions,
1983                                                   "can't have multiple vxlan"
1984                                                   " actions");
1985                 if ((current_action_flag & MLX5_TCF_VXLAN_ACTIONS) &&
1986                     (action_flags & MLX5_TCF_VLAN_ACTIONS))
1987                         return rte_flow_error_set(error, ENOTSUP,
1988                                                   RTE_FLOW_ERROR_TYPE_ACTION,
1989                                                   actions,
1990                                                   "can't have vxlan and vlan"
1991                                                   " actions in the same rule");
1992                 action_flags |= current_action_flag;
1993         }
1994         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
1995                 unsigned int i;
1996
1997                 if ((item_flags & MLX5_FLOW_LAYER_TUNNEL) &&
1998                     items->type != RTE_FLOW_ITEM_TYPE_ETH)
1999                         return rte_flow_error_set(error, ENOTSUP,
2000                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2001                                                   items,
2002                                                   "only L2 inner item"
2003                                                   " is supported");
2004                 switch (items->type) {
2005                 case RTE_FLOW_ITEM_TYPE_VOID:
2006                         break;
2007                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
2008                         mask.port_id = flow_tcf_item_mask
2009                                 (items, &rte_flow_item_port_id_mask,
2010                                  &flow_tcf_mask_supported.port_id,
2011                                  &flow_tcf_mask_empty.port_id,
2012                                  sizeof(flow_tcf_mask_supported.port_id),
2013                                  error);
2014                         if (!mask.port_id)
2015                                 return -rte_errno;
2016                         if (mask.port_id == &flow_tcf_mask_empty.port_id) {
2017                                 in_port_id_set = 1;
2018                                 break;
2019                         }
2020                         spec.port_id = items->spec;
2021                         if (mask.port_id->id && mask.port_id->id != 0xffffffff)
2022                                 return rte_flow_error_set
2023                                         (error, ENOTSUP,
2024                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2025                                          mask.port_id,
2026                                          "no support for partial mask on"
2027                                          " \"id\" field");
2028                         if (!mask.port_id->id)
2029                                 i = 0;
2030                         else
2031                                 for (i = 0; ptoi[i].ifindex; ++i)
2032                                         if (ptoi[i].port_id == spec.port_id->id)
2033                                                 break;
2034                         if (!ptoi[i].ifindex)
2035                                 return rte_flow_error_set
2036                                         (error, ENODEV,
2037                                          RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2038                                          spec.port_id,
2039                                          "missing data to convert port ID to"
2040                                          " ifindex");
2041                         if (in_port_id_set && ptoi[i].ifindex != tcm_ifindex)
2042                                 return rte_flow_error_set
2043                                         (error, ENOTSUP,
2044                                          RTE_FLOW_ERROR_TYPE_ITEM_SPEC,
2045                                          spec.port_id,
2046                                          "cannot match traffic for"
2047                                          " several port IDs through"
2048                                          " a single flow rule");
2049                         tcm_ifindex = ptoi[i].ifindex;
2050                         in_port_id_set = 1;
2051                         break;
2052                 case RTE_FLOW_ITEM_TYPE_ETH:
2053                         ret = mlx5_flow_validate_item_eth(items, item_flags,
2054                                                           error);
2055                         if (ret < 0)
2056                                 return ret;
2057                         item_flags |= (item_flags & MLX5_FLOW_LAYER_TUNNEL) ?
2058                                         MLX5_FLOW_LAYER_INNER_L2 :
2059                                         MLX5_FLOW_LAYER_OUTER_L2;
2060                         /* TODO:
2061                          * Redundant check due to different supported mask.
2062                          * Same for the rest of items.
2063                          */
2064                         mask.eth = flow_tcf_item_mask
2065                                 (items, &rte_flow_item_eth_mask,
2066                                  &flow_tcf_mask_supported.eth,
2067                                  &flow_tcf_mask_empty.eth,
2068                                  sizeof(flow_tcf_mask_supported.eth),
2069                                  error);
2070                         if (!mask.eth)
2071                                 return -rte_errno;
2072                         if (mask.eth->type && mask.eth->type !=
2073                             RTE_BE16(0xffff))
2074                                 return rte_flow_error_set
2075                                         (error, ENOTSUP,
2076                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2077                                          mask.eth,
2078                                          "no support for partial mask on"
2079                                          " \"type\" field");
2080                         break;
2081                 case RTE_FLOW_ITEM_TYPE_VLAN:
2082                         ret = mlx5_flow_validate_item_vlan(items, item_flags,
2083                                                            error);
2084                         if (ret < 0)
2085                                 return ret;
2086                         item_flags |= MLX5_FLOW_LAYER_OUTER_VLAN;
2087                         mask.vlan = flow_tcf_item_mask
2088                                 (items, &rte_flow_item_vlan_mask,
2089                                  &flow_tcf_mask_supported.vlan,
2090                                  &flow_tcf_mask_empty.vlan,
2091                                  sizeof(flow_tcf_mask_supported.vlan),
2092                                  error);
2093                         if (!mask.vlan)
2094                                 return -rte_errno;
2095                         if ((mask.vlan->tci & RTE_BE16(0xe000) &&
2096                              (mask.vlan->tci & RTE_BE16(0xe000)) !=
2097                               RTE_BE16(0xe000)) ||
2098                             (mask.vlan->tci & RTE_BE16(0x0fff) &&
2099                              (mask.vlan->tci & RTE_BE16(0x0fff)) !=
2100                               RTE_BE16(0x0fff)) ||
2101                             (mask.vlan->inner_type &&
2102                              mask.vlan->inner_type != RTE_BE16(0xffff)))
2103                                 return rte_flow_error_set
2104                                         (error, ENOTSUP,
2105                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2106                                          mask.vlan,
2107                                          "no support for partial masks on"
2108                                          " \"tci\" (PCP and VID parts) and"
2109                                          " \"inner_type\" fields");
2110                         break;
2111                 case RTE_FLOW_ITEM_TYPE_IPV4:
2112                         ret = mlx5_flow_validate_item_ipv4(items, item_flags,
2113                                                            error);
2114                         if (ret < 0)
2115                                 return ret;
2116                         item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
2117                         mask.ipv4 = flow_tcf_item_mask
2118                                 (items, &rte_flow_item_ipv4_mask,
2119                                  &flow_tcf_mask_supported.ipv4,
2120                                  &flow_tcf_mask_empty.ipv4,
2121                                  sizeof(flow_tcf_mask_supported.ipv4),
2122                                  error);
2123                         if (!mask.ipv4)
2124                                 return -rte_errno;
2125                         if (mask.ipv4->hdr.next_proto_id &&
2126                             mask.ipv4->hdr.next_proto_id != 0xff)
2127                                 return rte_flow_error_set
2128                                         (error, ENOTSUP,
2129                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2130                                          mask.ipv4,
2131                                          "no support for partial mask on"
2132                                          " \"hdr.next_proto_id\" field");
2133                         else if (mask.ipv4->hdr.next_proto_id)
2134                                 next_protocol =
2135                                         ((const struct rte_flow_item_ipv4 *)
2136                                          (items->spec))->hdr.next_proto_id;
2137                         if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2138                                 ret = flow_tcf_validate_vxlan_decap_ipv4
2139                                                                 (items, error);
2140                                 if (ret < 0)
2141                                         return ret;
2142                         }
2143                         break;
2144                 case RTE_FLOW_ITEM_TYPE_IPV6:
2145                         ret = mlx5_flow_validate_item_ipv6(items, item_flags,
2146                                                            error);
2147                         if (ret < 0)
2148                                 return ret;
2149                         item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
2150                         mask.ipv6 = flow_tcf_item_mask
2151                                 (items, &rte_flow_item_ipv6_mask,
2152                                  &flow_tcf_mask_supported.ipv6,
2153                                  &flow_tcf_mask_empty.ipv6,
2154                                  sizeof(flow_tcf_mask_supported.ipv6),
2155                                  error);
2156                         if (!mask.ipv6)
2157                                 return -rte_errno;
2158                         if (mask.ipv6->hdr.proto &&
2159                             mask.ipv6->hdr.proto != 0xff)
2160                                 return rte_flow_error_set
2161                                         (error, ENOTSUP,
2162                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2163                                          mask.ipv6,
2164                                          "no support for partial mask on"
2165                                          " \"hdr.proto\" field");
2166                         else if (mask.ipv6->hdr.proto)
2167                                 next_protocol =
2168                                         ((const struct rte_flow_item_ipv6 *)
2169                                          (items->spec))->hdr.proto;
2170                         if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2171                                 ret = flow_tcf_validate_vxlan_decap_ipv6
2172                                                                 (items, error);
2173                                 if (ret < 0)
2174                                         return ret;
2175                         }
2176                         break;
2177                 case RTE_FLOW_ITEM_TYPE_UDP:
2178                         ret = mlx5_flow_validate_item_udp(items, item_flags,
2179                                                           next_protocol, error);
2180                         if (ret < 0)
2181                                 return ret;
2182                         item_flags |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
2183                         mask.udp = flow_tcf_item_mask
2184                                 (items, &rte_flow_item_udp_mask,
2185                                  &flow_tcf_mask_supported.udp,
2186                                  &flow_tcf_mask_empty.udp,
2187                                  sizeof(flow_tcf_mask_supported.udp),
2188                                  error);
2189                         if (!mask.udp)
2190                                 return -rte_errno;
2191                         if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2192                                 ret = flow_tcf_validate_vxlan_decap_udp
2193                                                                 (items, error);
2194                                 if (ret < 0)
2195                                         return ret;
2196                         }
2197                         break;
2198                 case RTE_FLOW_ITEM_TYPE_TCP:
2199                         ret = mlx5_flow_validate_item_tcp
2200                                              (items, item_flags,
2201                                               next_protocol,
2202                                               &flow_tcf_mask_supported.tcp,
2203                                               error);
2204                         if (ret < 0)
2205                                 return ret;
2206                         item_flags |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
2207                         mask.tcp = flow_tcf_item_mask
2208                                 (items, &rte_flow_item_tcp_mask,
2209                                  &flow_tcf_mask_supported.tcp,
2210                                  &flow_tcf_mask_empty.tcp,
2211                                  sizeof(flow_tcf_mask_supported.tcp),
2212                                  error);
2213                         if (!mask.tcp)
2214                                 return -rte_errno;
2215                         break;
2216                 case RTE_FLOW_ITEM_TYPE_VXLAN:
2217                         if (!(action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP))
2218                                 return rte_flow_error_set
2219                                         (error, ENOTSUP,
2220                                          RTE_FLOW_ERROR_TYPE_ITEM,
2221                                          items,
2222                                          "vni pattern should be followed by"
2223                                          " vxlan decapsulation action");
2224                         ret = mlx5_flow_validate_item_vxlan(items,
2225                                                             item_flags, error);
2226                         if (ret < 0)
2227                                 return ret;
2228                         item_flags |= MLX5_FLOW_LAYER_VXLAN;
2229                         mask.vxlan = flow_tcf_item_mask
2230                                 (items, &rte_flow_item_vxlan_mask,
2231                                  &flow_tcf_mask_supported.vxlan,
2232                                  &flow_tcf_mask_empty.vxlan,
2233                                  sizeof(flow_tcf_mask_supported.vxlan), error);
2234                         if (!mask.vxlan)
2235                                 return -rte_errno;
2236                         if (mask.vxlan->vni[0] != 0xff ||
2237                             mask.vxlan->vni[1] != 0xff ||
2238                             mask.vxlan->vni[2] != 0xff)
2239                                 return rte_flow_error_set
2240                                         (error, ENOTSUP,
2241                                          RTE_FLOW_ERROR_TYPE_ITEM_MASK,
2242                                          mask.vxlan,
2243                                          "no support for partial or "
2244                                          "empty mask on \"vxlan.vni\" field");
2245                         break;
2246                 default:
2247                         return rte_flow_error_set(error, ENOTSUP,
2248                                                   RTE_FLOW_ERROR_TYPE_ITEM,
2249                                                   items, "item not supported");
2250                 }
2251         }
2252         if ((action_flags & MLX5_TCF_PEDIT_ACTIONS) &&
2253             (action_flags & MLX5_FLOW_ACTION_DROP))
2254                 return rte_flow_error_set(error, ENOTSUP,
2255                                           RTE_FLOW_ERROR_TYPE_ACTION,
2256                                           actions,
2257                                           "set action is not compatible with "
2258                                           "drop action");
2259         if ((action_flags & MLX5_TCF_PEDIT_ACTIONS) &&
2260             !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2261                 return rte_flow_error_set(error, ENOTSUP,
2262                                           RTE_FLOW_ERROR_TYPE_ACTION,
2263                                           actions,
2264                                           "set action must be followed by "
2265                                           "port_id action");
2266         if (action_flags &
2267            (MLX5_FLOW_ACTION_SET_IPV4_SRC | MLX5_FLOW_ACTION_SET_IPV4_DST)) {
2268                 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV4))
2269                         return rte_flow_error_set(error, EINVAL,
2270                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2271                                                   actions,
2272                                                   "no ipv4 item found in"
2273                                                   " pattern");
2274         }
2275         if (action_flags &
2276            (MLX5_FLOW_ACTION_SET_IPV6_SRC | MLX5_FLOW_ACTION_SET_IPV6_DST)) {
2277                 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L3_IPV6))
2278                         return rte_flow_error_set(error, EINVAL,
2279                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2280                                                   actions,
2281                                                   "no ipv6 item found in"
2282                                                   " pattern");
2283         }
2284         if (action_flags &
2285            (MLX5_FLOW_ACTION_SET_TP_SRC | MLX5_FLOW_ACTION_SET_TP_DST)) {
2286                 if (!(item_flags &
2287                      (MLX5_FLOW_LAYER_OUTER_L4_UDP |
2288                       MLX5_FLOW_LAYER_OUTER_L4_TCP)))
2289                         return rte_flow_error_set(error, EINVAL,
2290                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2291                                                   actions,
2292                                                   "no TCP/UDP item found in"
2293                                                   " pattern");
2294         }
2295         /*
2296          * FW syndrome (0xA9C090):
2297          *     set_flow_table_entry: push vlan action fte in fdb can ONLY be
2298          *     forward to the uplink.
2299          */
2300         if ((action_flags & MLX5_FLOW_ACTION_OF_PUSH_VLAN) &&
2301             (action_flags & MLX5_FLOW_ACTION_PORT_ID) &&
2302             ((struct priv *)port_id_dev->data->dev_private)->representor)
2303                 return rte_flow_error_set(error, ENOTSUP,
2304                                           RTE_FLOW_ERROR_TYPE_ACTION, actions,
2305                                           "vlan push can only be applied"
2306                                           " when forwarding to uplink port");
2307         /*
2308          * FW syndrome (0x294609):
2309          *     set_flow_table_entry: modify/pop/push actions in fdb flow table
2310          *     are supported only while forwarding to vport.
2311          */
2312         if ((action_flags & MLX5_TCF_VLAN_ACTIONS) &&
2313             !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2314                 return rte_flow_error_set(error, ENOTSUP,
2315                                           RTE_FLOW_ERROR_TYPE_ACTION, actions,
2316                                           "vlan actions are supported"
2317                                           " only with port_id action");
2318         if ((action_flags & MLX5_TCF_VXLAN_ACTIONS) &&
2319             !(action_flags & MLX5_FLOW_ACTION_PORT_ID))
2320                 return rte_flow_error_set(error, ENOTSUP,
2321                                           RTE_FLOW_ERROR_TYPE_ACTION, NULL,
2322                                           "vxlan actions are supported"
2323                                           " only with port_id action");
2324         if (!(action_flags & MLX5_TCF_FATE_ACTIONS))
2325                 return rte_flow_error_set(error, EINVAL,
2326                                           RTE_FLOW_ERROR_TYPE_ACTION, actions,
2327                                           "no fate action is found");
2328         if (action_flags &
2329            (MLX5_FLOW_ACTION_SET_TTL | MLX5_FLOW_ACTION_DEC_TTL)) {
2330                 if (!(item_flags &
2331                      (MLX5_FLOW_LAYER_OUTER_L3_IPV4 |
2332                       MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
2333                         return rte_flow_error_set(error, EINVAL,
2334                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2335                                                   actions,
2336                                                   "no IP found in pattern");
2337         }
2338         if (action_flags &
2339             (MLX5_FLOW_ACTION_SET_MAC_SRC | MLX5_FLOW_ACTION_SET_MAC_DST)) {
2340                 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L2))
2341                         return rte_flow_error_set(error, ENOTSUP,
2342                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2343                                                   actions,
2344                                                   "no ethernet found in"
2345                                                   " pattern");
2346         }
2347         if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2348                 if (!(item_flags &
2349                      (MLX5_FLOW_LAYER_OUTER_L3_IPV4 |
2350                       MLX5_FLOW_LAYER_OUTER_L3_IPV6)))
2351                         return rte_flow_error_set(error, EINVAL,
2352                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2353                                                   NULL,
2354                                                   "no outer IP pattern found"
2355                                                   " for vxlan decap action");
2356                 if (!(item_flags & MLX5_FLOW_LAYER_OUTER_L4_UDP))
2357                         return rte_flow_error_set(error, EINVAL,
2358                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2359                                                   NULL,
2360                                                   "no outer UDP pattern found"
2361                                                   " for vxlan decap action");
2362                 if (!(item_flags & MLX5_FLOW_LAYER_VXLAN))
2363                         return rte_flow_error_set(error, EINVAL,
2364                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2365                                                   NULL,
2366                                                   "no VNI pattern found"
2367                                                   " for vxlan decap action");
2368         }
2369         return 0;
2370 }
2371
2372 /**
2373  * Calculate maximum size of memory for flow items of Linux TC flower.
2374  *
2375  * @param[in] attr
2376  *   Pointer to the flow attributes.
2377  * @param[in] items
2378  *   Pointer to the list of items.
2379  *
2380  * @return
2381  *   Maximum size of memory for items.
2382  */
2383 static int
2384 flow_tcf_get_items_size(const struct rte_flow_attr *attr,
2385                         const struct rte_flow_item items[])
2386 {
2387         int size = 0;
2388
2389         size += SZ_NLATTR_STRZ_OF("flower") +
2390                 SZ_NLATTR_NEST + /* TCA_OPTIONS. */
2391                 SZ_NLATTR_TYPE_OF(uint32_t); /* TCA_CLS_FLAGS_SKIP_SW. */
2392         if (attr->group > 0)
2393                 size += SZ_NLATTR_TYPE_OF(uint32_t); /* TCA_CHAIN. */
2394         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2395                 switch (items->type) {
2396                 case RTE_FLOW_ITEM_TYPE_VOID:
2397                         break;
2398                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
2399                         break;
2400                 case RTE_FLOW_ITEM_TYPE_ETH:
2401                         size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2402                                 SZ_NLATTR_DATA_OF(ETHER_ADDR_LEN) * 4;
2403                                 /* dst/src MAC addr and mask. */
2404                         break;
2405                 case RTE_FLOW_ITEM_TYPE_VLAN:
2406                         size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2407                                 SZ_NLATTR_TYPE_OF(uint16_t) +
2408                                 /* VLAN Ether type. */
2409                                 SZ_NLATTR_TYPE_OF(uint8_t) + /* VLAN prio. */
2410                                 SZ_NLATTR_TYPE_OF(uint16_t); /* VLAN ID. */
2411                         break;
2412                 case RTE_FLOW_ITEM_TYPE_IPV4:
2413                         size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2414                                 SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2415                                 SZ_NLATTR_TYPE_OF(uint32_t) * 4;
2416                                 /* dst/src IP addr and mask. */
2417                         break;
2418                 case RTE_FLOW_ITEM_TYPE_IPV6:
2419                         size += SZ_NLATTR_TYPE_OF(uint16_t) + /* Ether type. */
2420                                 SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2421                                 SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN) * 4;
2422                                 /* dst/src IP addr and mask. */
2423                         break;
2424                 case RTE_FLOW_ITEM_TYPE_UDP:
2425                         size += SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2426                                 SZ_NLATTR_TYPE_OF(uint16_t) * 4;
2427                                 /* dst/src port and mask. */
2428                         break;
2429                 case RTE_FLOW_ITEM_TYPE_TCP:
2430                         size += SZ_NLATTR_TYPE_OF(uint8_t) + /* IP proto. */
2431                                 SZ_NLATTR_TYPE_OF(uint16_t) * 4;
2432                                 /* dst/src port and mask. */
2433                         break;
2434                 case RTE_FLOW_ITEM_TYPE_VXLAN:
2435                         size += SZ_NLATTR_TYPE_OF(uint32_t);
2436                         break;
2437                 default:
2438                         DRV_LOG(WARNING,
2439                                 "unsupported item %p type %d,"
2440                                 " items must be validated before flow creation",
2441                                 (const void *)items, items->type);
2442                         break;
2443                 }
2444         }
2445         return size;
2446 }
2447
2448 /**
2449  * Calculate size of memory to store the VXLAN encapsultion
2450  * related items in the Netlink message buffer. Items list
2451  * is specified by RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action.
2452  * The item list should be validated.
2453  *
2454  * @param[in] action
2455  *   RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action object.
2456  *   List of pattern items to scan data from.
2457  *
2458  * @return
2459  *   The size the part of Netlink message buffer to store the
2460  *   VXLAN encapsulation item attributes.
2461  */
2462 static int
2463 flow_tcf_vxlan_encap_size(const struct rte_flow_action *action)
2464 {
2465         const struct rte_flow_item *items;
2466         int size = 0;
2467
2468         assert(action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP);
2469         assert(action->conf);
2470
2471         items = ((const struct rte_flow_action_vxlan_encap *)
2472                                         action->conf)->definition;
2473         assert(items);
2474         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2475                 switch (items->type) {
2476                 case RTE_FLOW_ITEM_TYPE_VOID:
2477                         break;
2478                 case RTE_FLOW_ITEM_TYPE_ETH:
2479                         /* This item does not require message buffer. */
2480                         break;
2481                 case RTE_FLOW_ITEM_TYPE_IPV4:
2482                         size += SZ_NLATTR_DATA_OF(IPV4_ADDR_LEN) * 2;
2483                         break;
2484                 case RTE_FLOW_ITEM_TYPE_IPV6:
2485                         size += SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN) * 2;
2486                         break;
2487                 case RTE_FLOW_ITEM_TYPE_UDP: {
2488                         const struct rte_flow_item_udp *udp = items->mask;
2489
2490                         size += SZ_NLATTR_TYPE_OF(uint16_t);
2491                         if (!udp || udp->hdr.src_port != RTE_BE16(0x0000))
2492                                 size += SZ_NLATTR_TYPE_OF(uint16_t);
2493                         break;
2494                 }
2495                 case RTE_FLOW_ITEM_TYPE_VXLAN:
2496                         size += SZ_NLATTR_TYPE_OF(uint32_t);
2497                         break;
2498                 default:
2499                         assert(false);
2500                         DRV_LOG(WARNING,
2501                                 "unsupported item %p type %d,"
2502                                 " items must be validated"
2503                                 " before flow creation",
2504                                 (const void *)items, items->type);
2505                         return 0;
2506                 }
2507         }
2508         return size;
2509 }
2510
2511 /**
2512  * Calculate maximum size of memory for flow actions of Linux TC flower and
2513  * extract specified actions.
2514  *
2515  * @param[in] actions
2516  *   Pointer to the list of actions.
2517  * @param[out] action_flags
2518  *   Pointer to the detected actions.
2519  *
2520  * @return
2521  *   Maximum size of memory for actions.
2522  */
2523 static int
2524 flow_tcf_get_actions_and_size(const struct rte_flow_action actions[],
2525                               uint64_t *action_flags)
2526 {
2527         int size = 0;
2528         uint64_t flags = 0;
2529
2530         size += SZ_NLATTR_NEST; /* TCA_FLOWER_ACT. */
2531         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
2532                 switch (actions->type) {
2533                 case RTE_FLOW_ACTION_TYPE_VOID:
2534                         break;
2535                 case RTE_FLOW_ACTION_TYPE_PORT_ID:
2536                         size += SZ_NLATTR_NEST + /* na_act_index. */
2537                                 SZ_NLATTR_STRZ_OF("mirred") +
2538                                 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2539                                 SZ_NLATTR_TYPE_OF(struct tc_mirred);
2540                         flags |= MLX5_FLOW_ACTION_PORT_ID;
2541                         break;
2542                 case RTE_FLOW_ACTION_TYPE_JUMP:
2543                         size += SZ_NLATTR_NEST + /* na_act_index. */
2544                                 SZ_NLATTR_STRZ_OF("gact") +
2545                                 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2546                                 SZ_NLATTR_TYPE_OF(struct tc_gact);
2547                         flags |= MLX5_FLOW_ACTION_JUMP;
2548                         break;
2549                 case RTE_FLOW_ACTION_TYPE_DROP:
2550                         size += SZ_NLATTR_NEST + /* na_act_index. */
2551                                 SZ_NLATTR_STRZ_OF("gact") +
2552                                 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2553                                 SZ_NLATTR_TYPE_OF(struct tc_gact);
2554                         flags |= MLX5_FLOW_ACTION_DROP;
2555                         break;
2556                 case RTE_FLOW_ACTION_TYPE_COUNT:
2557                         break;
2558                 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
2559                         flags |= MLX5_FLOW_ACTION_OF_POP_VLAN;
2560                         goto action_of_vlan;
2561                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
2562                         flags |= MLX5_FLOW_ACTION_OF_PUSH_VLAN;
2563                         goto action_of_vlan;
2564                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
2565                         flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_VID;
2566                         goto action_of_vlan;
2567                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
2568                         flags |= MLX5_FLOW_ACTION_OF_SET_VLAN_PCP;
2569                         goto action_of_vlan;
2570 action_of_vlan:
2571                         size += SZ_NLATTR_NEST + /* na_act_index. */
2572                                 SZ_NLATTR_STRZ_OF("vlan") +
2573                                 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2574                                 SZ_NLATTR_TYPE_OF(struct tc_vlan) +
2575                                 SZ_NLATTR_TYPE_OF(uint16_t) +
2576                                 /* VLAN protocol. */
2577                                 SZ_NLATTR_TYPE_OF(uint16_t) + /* VLAN ID. */
2578                                 SZ_NLATTR_TYPE_OF(uint8_t); /* VLAN prio. */
2579                         break;
2580                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
2581                         size += SZ_NLATTR_NEST + /* na_act_index. */
2582                                 SZ_NLATTR_STRZ_OF("tunnel_key") +
2583                                 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2584                                 SZ_NLATTR_TYPE_OF(uint8_t);
2585                         size += SZ_NLATTR_TYPE_OF(struct tc_tunnel_key);
2586                         size += flow_tcf_vxlan_encap_size(actions) +
2587                                 RTE_ALIGN_CEIL /* preceding encap params. */
2588                                 (sizeof(struct flow_tcf_vxlan_encap),
2589                                 MNL_ALIGNTO);
2590                         flags |= MLX5_FLOW_ACTION_VXLAN_ENCAP;
2591                         break;
2592                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
2593                         size += SZ_NLATTR_NEST + /* na_act_index. */
2594                                 SZ_NLATTR_STRZ_OF("tunnel_key") +
2595                                 SZ_NLATTR_NEST + /* TCA_ACT_OPTIONS. */
2596                                 SZ_NLATTR_TYPE_OF(uint8_t);
2597                         size += SZ_NLATTR_TYPE_OF(struct tc_tunnel_key);
2598                         size += RTE_ALIGN_CEIL /* preceding decap params. */
2599                                 (sizeof(struct flow_tcf_vxlan_decap),
2600                                 MNL_ALIGNTO);
2601                         flags |= MLX5_FLOW_ACTION_VXLAN_DECAP;
2602                         break;
2603                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
2604                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
2605                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
2606                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
2607                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
2608                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
2609                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
2610                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
2611                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
2612                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
2613                         size += flow_tcf_get_pedit_actions_size(&actions,
2614                                                                 &flags);
2615                         break;
2616                 default:
2617                         DRV_LOG(WARNING,
2618                                 "unsupported action %p type %d,"
2619                                 " items must be validated before flow creation",
2620                                 (const void *)actions, actions->type);
2621                         break;
2622                 }
2623         }
2624         *action_flags = flags;
2625         return size;
2626 }
2627
2628 /**
2629  * Brand rtnetlink buffer with unique handle.
2630  *
2631  * This handle should be unique for a given network interface to avoid
2632  * collisions.
2633  *
2634  * @param nlh
2635  *   Pointer to Netlink message.
2636  * @param handle
2637  *   Unique 32-bit handle to use.
2638  */
2639 static void
2640 flow_tcf_nl_brand(struct nlmsghdr *nlh, uint32_t handle)
2641 {
2642         struct tcmsg *tcm = mnl_nlmsg_get_payload(nlh);
2643
2644         tcm->tcm_handle = handle;
2645         DRV_LOG(DEBUG, "Netlink msg %p is branded with handle %x",
2646                 (void *)nlh, handle);
2647 }
2648
2649 /**
2650  * Prepare a flow object for Linux TC flower. It calculates the maximum size of
2651  * memory required, allocates the memory, initializes Netlink message headers
2652  * and set unique TC message handle.
2653  *
2654  * @param[in] attr
2655  *   Pointer to the flow attributes.
2656  * @param[in] items
2657  *   Pointer to the list of items.
2658  * @param[in] actions
2659  *   Pointer to the list of actions.
2660  * @param[out] error
2661  *   Pointer to the error structure.
2662  *
2663  * @return
2664  *   Pointer to mlx5_flow object on success,
2665  *   otherwise NULL and rte_ernno is set.
2666  */
2667 static struct mlx5_flow *
2668 flow_tcf_prepare(const struct rte_flow_attr *attr,
2669                  const struct rte_flow_item items[],
2670                  const struct rte_flow_action actions[],
2671                  struct rte_flow_error *error)
2672 {
2673         size_t size = RTE_ALIGN_CEIL
2674                         (sizeof(struct mlx5_flow),
2675                          alignof(struct flow_tcf_tunnel_hdr)) +
2676                       MNL_ALIGN(sizeof(struct nlmsghdr)) +
2677                       MNL_ALIGN(sizeof(struct tcmsg));
2678         struct mlx5_flow *dev_flow;
2679         uint64_t action_flags = 0;
2680         struct nlmsghdr *nlh;
2681         struct tcmsg *tcm;
2682         uint8_t *sp, *tun = NULL;
2683
2684         size += flow_tcf_get_items_size(attr, items);
2685         size += flow_tcf_get_actions_and_size(actions, &action_flags);
2686         dev_flow = rte_zmalloc(__func__, size, MNL_ALIGNTO);
2687         if (!dev_flow) {
2688                 rte_flow_error_set(error, ENOMEM,
2689                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2690                                    "not enough memory to create E-Switch flow");
2691                 return NULL;
2692         }
2693         sp = (uint8_t *)(dev_flow + 1);
2694         if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP) {
2695                 sp = RTE_PTR_ALIGN
2696                         (sp, alignof(struct flow_tcf_tunnel_hdr));
2697                 tun = sp;
2698                 sp += RTE_ALIGN_CEIL
2699                         (sizeof(struct flow_tcf_vxlan_encap),
2700                         MNL_ALIGNTO);
2701 #ifndef NDEBUG
2702                 size -= RTE_ALIGN_CEIL
2703                         (sizeof(struct flow_tcf_vxlan_encap),
2704                         MNL_ALIGNTO);
2705 #endif
2706         } else if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP) {
2707                 sp = RTE_PTR_ALIGN
2708                         (sp, alignof(struct flow_tcf_tunnel_hdr));
2709                 tun = sp;
2710                 sp += RTE_ALIGN_CEIL
2711                         (sizeof(struct flow_tcf_vxlan_decap),
2712                         MNL_ALIGNTO);
2713 #ifndef NDEBUG
2714                 size -= RTE_ALIGN_CEIL
2715                         (sizeof(struct flow_tcf_vxlan_decap),
2716                         MNL_ALIGNTO);
2717 #endif
2718         } else {
2719                 sp = RTE_PTR_ALIGN(sp, MNL_ALIGNTO);
2720         }
2721         nlh = mnl_nlmsg_put_header(sp);
2722         tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
2723         *dev_flow = (struct mlx5_flow){
2724                 .tcf = (struct mlx5_flow_tcf){
2725 #ifndef NDEBUG
2726                         .nlsize = size - RTE_ALIGN_CEIL
2727                                 (sizeof(struct mlx5_flow),
2728                                  alignof(struct flow_tcf_tunnel_hdr)),
2729 #endif
2730                         .tunnel = (struct flow_tcf_tunnel_hdr *)tun,
2731                         .nlh = nlh,
2732                         .tcm = tcm,
2733                 },
2734         };
2735         if (action_flags & MLX5_FLOW_ACTION_VXLAN_DECAP)
2736                 dev_flow->tcf.tunnel->type = FLOW_TCF_TUNACT_VXLAN_DECAP;
2737         else if (action_flags & MLX5_FLOW_ACTION_VXLAN_ENCAP)
2738                 dev_flow->tcf.tunnel->type = FLOW_TCF_TUNACT_VXLAN_ENCAP;
2739         /*
2740          * Generate a reasonably unique handle based on the address of the
2741          * target buffer.
2742          *
2743          * This is straightforward on 32-bit systems where the flow pointer can
2744          * be used directly. Otherwise, its least significant part is taken
2745          * after shifting it by the previous power of two of the pointed buffer
2746          * size.
2747          */
2748         if (sizeof(dev_flow) <= 4)
2749                 flow_tcf_nl_brand(nlh, (uintptr_t)dev_flow);
2750         else
2751                 flow_tcf_nl_brand(nlh, (uintptr_t)dev_flow >>
2752                                        rte_log2_u32(rte_align32prevpow2(size)));
2753         return dev_flow;
2754 }
2755
2756 /**
2757  * Make adjustments for supporting count actions.
2758  *
2759  * @param[in] dev
2760  *   Pointer to the Ethernet device structure.
2761  * @param[in] dev_flow
2762  *   Pointer to mlx5_flow.
2763  * @param[out] error
2764  *   Pointer to error structure.
2765  *
2766  * @return
2767  *   0 On success else a negative errno value is returned and rte_errno is set.
2768  */
2769 static int
2770 flow_tcf_translate_action_count(struct rte_eth_dev *dev __rte_unused,
2771                                   struct mlx5_flow *dev_flow,
2772                                   struct rte_flow_error *error)
2773 {
2774         struct rte_flow *flow = dev_flow->flow;
2775
2776         if (!flow->counter) {
2777                 flow->counter = flow_tcf_counter_new();
2778                 if (!flow->counter)
2779                         return rte_flow_error_set(error, rte_errno,
2780                                                   RTE_FLOW_ERROR_TYPE_ACTION,
2781                                                   NULL,
2782                                                   "cannot get counter"
2783                                                   " context.");
2784         }
2785         return 0;
2786 }
2787
2788 /**
2789  * Convert VXLAN VNI to 32-bit integer.
2790  *
2791  * @param[in] vni
2792  *   VXLAN VNI in 24-bit wire format.
2793  *
2794  * @return
2795  *   VXLAN VNI as a 32-bit integer value in network endian.
2796  */
2797 static inline rte_be32_t
2798 vxlan_vni_as_be32(const uint8_t vni[3])
2799 {
2800         union {
2801                 uint8_t vni[4];
2802                 rte_be32_t dword;
2803         } ret = {
2804                 .vni = { 0, vni[0], vni[1], vni[2] },
2805         };
2806         return ret.dword;
2807 }
2808
2809 /**
2810  * Helper function to process RTE_FLOW_ITEM_TYPE_ETH entry in configuration
2811  * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the MAC address fields
2812  * in the encapsulation parameters structure. The item must be prevalidated,
2813  * no any validation checks performed by function.
2814  *
2815  * @param[in] spec
2816  *   RTE_FLOW_ITEM_TYPE_ETH entry specification.
2817  * @param[in] mask
2818  *   RTE_FLOW_ITEM_TYPE_ETH entry mask.
2819  * @param[out] encap
2820  *   Structure to fill the gathered MAC address data.
2821  */
2822 static void
2823 flow_tcf_parse_vxlan_encap_eth(const struct rte_flow_item_eth *spec,
2824                                const struct rte_flow_item_eth *mask,
2825                                struct flow_tcf_vxlan_encap *encap)
2826 {
2827         /* Item must be validated before. No redundant checks. */
2828         assert(spec);
2829         if (!mask || !memcmp(&mask->dst,
2830                              &rte_flow_item_eth_mask.dst,
2831                              sizeof(rte_flow_item_eth_mask.dst))) {
2832                 /*
2833                  * Ethernet addresses are not supported by
2834                  * tc as tunnel_key parameters. Destination
2835                  * address is needed to form encap packet
2836                  * header and retrieved by kernel from
2837                  * implicit sources (ARP table, etc),
2838                  * address masks are not supported at all.
2839                  */
2840                 encap->eth.dst = spec->dst;
2841                 encap->mask |= FLOW_TCF_ENCAP_ETH_DST;
2842         }
2843         if (!mask || !memcmp(&mask->src,
2844                              &rte_flow_item_eth_mask.src,
2845                              sizeof(rte_flow_item_eth_mask.src))) {
2846                 /*
2847                  * Ethernet addresses are not supported by
2848                  * tc as tunnel_key parameters. Source ethernet
2849                  * address is ignored anyway.
2850                  */
2851                 encap->eth.src = spec->src;
2852                 encap->mask |= FLOW_TCF_ENCAP_ETH_SRC;
2853         }
2854 }
2855
2856 /**
2857  * Helper function to process RTE_FLOW_ITEM_TYPE_IPV4 entry in configuration
2858  * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the IPV4 address fields
2859  * in the encapsulation parameters structure. The item must be prevalidated,
2860  * no any validation checks performed by function.
2861  *
2862  * @param[in] spec
2863  *   RTE_FLOW_ITEM_TYPE_IPV4 entry specification.
2864  * @param[out] encap
2865  *   Structure to fill the gathered IPV4 address data.
2866  */
2867 static void
2868 flow_tcf_parse_vxlan_encap_ipv4(const struct rte_flow_item_ipv4 *spec,
2869                                 struct flow_tcf_vxlan_encap *encap)
2870 {
2871         /* Item must be validated before. No redundant checks. */
2872         assert(spec);
2873         encap->ipv4.dst = spec->hdr.dst_addr;
2874         encap->ipv4.src = spec->hdr.src_addr;
2875         encap->mask |= FLOW_TCF_ENCAP_IPV4_SRC |
2876                        FLOW_TCF_ENCAP_IPV4_DST;
2877 }
2878
2879 /**
2880  * Helper function to process RTE_FLOW_ITEM_TYPE_IPV6 entry in configuration
2881  * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the IPV6 address fields
2882  * in the encapsulation parameters structure. The item must be prevalidated,
2883  * no any validation checks performed by function.
2884  *
2885  * @param[in] spec
2886  *   RTE_FLOW_ITEM_TYPE_IPV6 entry specification.
2887  * @param[out] encap
2888  *   Structure to fill the gathered IPV6 address data.
2889  */
2890 static void
2891 flow_tcf_parse_vxlan_encap_ipv6(const struct rte_flow_item_ipv6 *spec,
2892                                 struct flow_tcf_vxlan_encap *encap)
2893 {
2894         /* Item must be validated before. No redundant checks. */
2895         assert(spec);
2896         memcpy(encap->ipv6.dst, spec->hdr.dst_addr, IPV6_ADDR_LEN);
2897         memcpy(encap->ipv6.src, spec->hdr.src_addr, IPV6_ADDR_LEN);
2898         encap->mask |= FLOW_TCF_ENCAP_IPV6_SRC |
2899                        FLOW_TCF_ENCAP_IPV6_DST;
2900 }
2901
2902 /**
2903  * Helper function to process RTE_FLOW_ITEM_TYPE_UDP entry in configuration
2904  * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the UDP port fields
2905  * in the encapsulation parameters structure. The item must be prevalidated,
2906  * no any validation checks performed by function.
2907  *
2908  * @param[in] spec
2909  *   RTE_FLOW_ITEM_TYPE_UDP entry specification.
2910  * @param[in] mask
2911  *   RTE_FLOW_ITEM_TYPE_UDP entry mask.
2912  * @param[out] encap
2913  *   Structure to fill the gathered UDP port data.
2914  */
2915 static void
2916 flow_tcf_parse_vxlan_encap_udp(const struct rte_flow_item_udp *spec,
2917                                const struct rte_flow_item_udp *mask,
2918                                struct flow_tcf_vxlan_encap *encap)
2919 {
2920         assert(spec);
2921         encap->udp.dst = spec->hdr.dst_port;
2922         encap->mask |= FLOW_TCF_ENCAP_UDP_DST;
2923         if (!mask || mask->hdr.src_port != RTE_BE16(0x0000)) {
2924                 encap->udp.src = spec->hdr.src_port;
2925                 encap->mask |= FLOW_TCF_ENCAP_IPV4_SRC;
2926         }
2927 }
2928
2929 /**
2930  * Helper function to process RTE_FLOW_ITEM_TYPE_VXLAN entry in configuration
2931  * of action RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. Fills the VNI fields
2932  * in the encapsulation parameters structure. The item must be prevalidated,
2933  * no any validation checks performed by function.
2934  *
2935  * @param[in] spec
2936  *   RTE_FLOW_ITEM_TYPE_VXLAN entry specification.
2937  * @param[out] encap
2938  *   Structure to fill the gathered VNI address data.
2939  */
2940 static void
2941 flow_tcf_parse_vxlan_encap_vni(const struct rte_flow_item_vxlan *spec,
2942                                struct flow_tcf_vxlan_encap *encap)
2943 {
2944         /* Item must be validated before. Do not redundant checks. */
2945         assert(spec);
2946         memcpy(encap->vxlan.vni, spec->vni, sizeof(encap->vxlan.vni));
2947         encap->mask |= FLOW_TCF_ENCAP_VXLAN_VNI;
2948 }
2949
2950 /**
2951  * Populate consolidated encapsulation object from list of pattern items.
2952  *
2953  * Helper function to process configuration of action such as
2954  * RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP. The item list should be
2955  * validated, there is no way to return an meaningful error.
2956  *
2957  * @param[in] action
2958  *   RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP action object.
2959  *   List of pattern items to gather data from.
2960  * @param[out] src
2961  *   Structure to fill gathered data.
2962  */
2963 static void
2964 flow_tcf_vxlan_encap_parse(const struct rte_flow_action *action,
2965                            struct flow_tcf_vxlan_encap *encap)
2966 {
2967         union {
2968                 const struct rte_flow_item_eth *eth;
2969                 const struct rte_flow_item_ipv4 *ipv4;
2970                 const struct rte_flow_item_ipv6 *ipv6;
2971                 const struct rte_flow_item_udp *udp;
2972                 const struct rte_flow_item_vxlan *vxlan;
2973         } spec, mask;
2974         const struct rte_flow_item *items;
2975
2976         assert(action->type == RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP);
2977         assert(action->conf);
2978
2979         items = ((const struct rte_flow_action_vxlan_encap *)
2980                                         action->conf)->definition;
2981         assert(items);
2982         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
2983                 switch (items->type) {
2984                 case RTE_FLOW_ITEM_TYPE_VOID:
2985                         break;
2986                 case RTE_FLOW_ITEM_TYPE_ETH:
2987                         mask.eth = items->mask;
2988                         spec.eth = items->spec;
2989                         flow_tcf_parse_vxlan_encap_eth(spec.eth, mask.eth,
2990                                                        encap);
2991                         break;
2992                 case RTE_FLOW_ITEM_TYPE_IPV4:
2993                         spec.ipv4 = items->spec;
2994                         flow_tcf_parse_vxlan_encap_ipv4(spec.ipv4, encap);
2995                         break;
2996                 case RTE_FLOW_ITEM_TYPE_IPV6:
2997                         spec.ipv6 = items->spec;
2998                         flow_tcf_parse_vxlan_encap_ipv6(spec.ipv6, encap);
2999                         break;
3000                 case RTE_FLOW_ITEM_TYPE_UDP:
3001                         mask.udp = items->mask;
3002                         spec.udp = items->spec;
3003                         flow_tcf_parse_vxlan_encap_udp(spec.udp, mask.udp,
3004                                                        encap);
3005                         break;
3006                 case RTE_FLOW_ITEM_TYPE_VXLAN:
3007                         spec.vxlan = items->spec;
3008                         flow_tcf_parse_vxlan_encap_vni(spec.vxlan, encap);
3009                         break;
3010                 default:
3011                         assert(false);
3012                         DRV_LOG(WARNING,
3013                                 "unsupported item %p type %d,"
3014                                 " items must be validated"
3015                                 " before flow creation",
3016                                 (const void *)items, items->type);
3017                         encap->mask = 0;
3018                         return;
3019                 }
3020         }
3021 }
3022
3023 /**
3024  * Translate flow for Linux TC flower and construct Netlink message.
3025  *
3026  * @param[in] priv
3027  *   Pointer to the priv structure.
3028  * @param[in, out] flow
3029  *   Pointer to the sub flow.
3030  * @param[in] attr
3031  *   Pointer to the flow attributes.
3032  * @param[in] items
3033  *   Pointer to the list of items.
3034  * @param[in] actions
3035  *   Pointer to the list of actions.
3036  * @param[out] error
3037  *   Pointer to the error structure.
3038  *
3039  * @return
3040  *   0 on success, a negative errno value otherwise and rte_ernno is set.
3041  */
3042 static int
3043 flow_tcf_translate(struct rte_eth_dev *dev, struct mlx5_flow *dev_flow,
3044                    const struct rte_flow_attr *attr,
3045                    const struct rte_flow_item items[],
3046                    const struct rte_flow_action actions[],
3047                    struct rte_flow_error *error)
3048 {
3049         union {
3050                 const struct rte_flow_item_port_id *port_id;
3051                 const struct rte_flow_item_eth *eth;
3052                 const struct rte_flow_item_vlan *vlan;
3053                 const struct rte_flow_item_ipv4 *ipv4;
3054                 const struct rte_flow_item_ipv6 *ipv6;
3055                 const struct rte_flow_item_tcp *tcp;
3056                 const struct rte_flow_item_udp *udp;
3057                 const struct rte_flow_item_vxlan *vxlan;
3058         } spec, mask;
3059         union {
3060                 const struct rte_flow_action_port_id *port_id;
3061                 const struct rte_flow_action_jump *jump;
3062                 const struct rte_flow_action_of_push_vlan *of_push_vlan;
3063                 const struct rte_flow_action_of_set_vlan_vid *
3064                         of_set_vlan_vid;
3065                 const struct rte_flow_action_of_set_vlan_pcp *
3066                         of_set_vlan_pcp;
3067         } conf;
3068         union {
3069                 struct flow_tcf_tunnel_hdr *hdr;
3070                 struct flow_tcf_vxlan_decap *vxlan;
3071         } decap = {
3072                 .hdr = NULL,
3073         };
3074         union {
3075                 struct flow_tcf_tunnel_hdr *hdr;
3076                 struct flow_tcf_vxlan_encap *vxlan;
3077         } encap = {
3078                 .hdr = NULL,
3079         };
3080         struct flow_tcf_ptoi ptoi[PTOI_TABLE_SZ_MAX(dev)];
3081         struct nlmsghdr *nlh = dev_flow->tcf.nlh;
3082         struct tcmsg *tcm = dev_flow->tcf.tcm;
3083         uint32_t na_act_index_cur;
3084         bool eth_type_set = 0;
3085         bool vlan_present = 0;
3086         bool vlan_eth_type_set = 0;
3087         bool ip_proto_set = 0;
3088         struct nlattr *na_flower;
3089         struct nlattr *na_flower_act;
3090         struct nlattr *na_vlan_id = NULL;
3091         struct nlattr *na_vlan_priority = NULL;
3092         uint64_t item_flags = 0;
3093         int ret;
3094
3095         claim_nonzero(flow_tcf_build_ptoi_table(dev, ptoi,
3096                                                 PTOI_TABLE_SZ_MAX(dev)));
3097         if (dev_flow->tcf.tunnel) {
3098                 switch (dev_flow->tcf.tunnel->type) {
3099                 case FLOW_TCF_TUNACT_VXLAN_DECAP:
3100                         decap.vxlan = dev_flow->tcf.vxlan_decap;
3101                         break;
3102                 case FLOW_TCF_TUNACT_VXLAN_ENCAP:
3103                         encap.vxlan = dev_flow->tcf.vxlan_encap;
3104                         break;
3105                 /* New tunnel actions can be added here. */
3106                 default:
3107                         assert(false);
3108                         break;
3109                 }
3110         }
3111         nlh = dev_flow->tcf.nlh;
3112         tcm = dev_flow->tcf.tcm;
3113         /* Prepare API must have been called beforehand. */
3114         assert(nlh != NULL && tcm != NULL);
3115         tcm->tcm_family = AF_UNSPEC;
3116         tcm->tcm_ifindex = ptoi[0].ifindex;
3117         tcm->tcm_parent = TC_H_MAKE(TC_H_INGRESS, TC_H_MIN_INGRESS);
3118         /*
3119          * Priority cannot be zero to prevent the kernel from picking one
3120          * automatically.
3121          */
3122         tcm->tcm_info = TC_H_MAKE((attr->priority + 1) << 16,
3123                                   RTE_BE16(ETH_P_ALL));
3124         if (attr->group > 0)
3125                 mnl_attr_put_u32(nlh, TCA_CHAIN, attr->group);
3126         mnl_attr_put_strz(nlh, TCA_KIND, "flower");
3127         na_flower = mnl_attr_nest_start(nlh, TCA_OPTIONS);
3128         for (; items->type != RTE_FLOW_ITEM_TYPE_END; items++) {
3129                 unsigned int i;
3130
3131                 switch (items->type) {
3132                 case RTE_FLOW_ITEM_TYPE_VOID:
3133                         break;
3134                 case RTE_FLOW_ITEM_TYPE_PORT_ID:
3135                         mask.port_id = flow_tcf_item_mask
3136                                 (items, &rte_flow_item_port_id_mask,
3137                                  &flow_tcf_mask_supported.port_id,
3138                                  &flow_tcf_mask_empty.port_id,
3139                                  sizeof(flow_tcf_mask_supported.port_id),
3140                                  error);
3141                         assert(mask.port_id);
3142                         if (mask.port_id == &flow_tcf_mask_empty.port_id)
3143                                 break;
3144                         spec.port_id = items->spec;
3145                         if (!mask.port_id->id)
3146                                 i = 0;
3147                         else
3148                                 for (i = 0; ptoi[i].ifindex; ++i)
3149                                         if (ptoi[i].port_id == spec.port_id->id)
3150                                                 break;
3151                         assert(ptoi[i].ifindex);
3152                         tcm->tcm_ifindex = ptoi[i].ifindex;
3153                         break;
3154                 case RTE_FLOW_ITEM_TYPE_ETH:
3155                         item_flags |= (item_flags & MLX5_FLOW_LAYER_VXLAN) ?
3156                                       MLX5_FLOW_LAYER_INNER_L2 :
3157                                       MLX5_FLOW_LAYER_OUTER_L2;
3158                         mask.eth = flow_tcf_item_mask
3159                                 (items, &rte_flow_item_eth_mask,
3160                                  &flow_tcf_mask_supported.eth,
3161                                  &flow_tcf_mask_empty.eth,
3162                                  sizeof(flow_tcf_mask_supported.eth),
3163                                  error);
3164                         assert(mask.eth);
3165                         if (mask.eth == &flow_tcf_mask_empty.eth)
3166                                 break;
3167                         spec.eth = items->spec;
3168                         if (decap.vxlan &&
3169                             !(item_flags & MLX5_FLOW_LAYER_VXLAN)) {
3170                                 DRV_LOG(WARNING,
3171                                         "outer L2 addresses cannot be forced"
3172                                         " for vxlan decapsulation, parameter"
3173                                         " ignored");
3174                                 break;
3175                         }
3176                         if (mask.eth->type) {
3177                                 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_ETH_TYPE,
3178                                                  spec.eth->type);
3179                                 eth_type_set = 1;
3180                         }
3181                         if (!is_zero_ether_addr(&mask.eth->dst)) {
3182                                 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_DST,
3183                                              ETHER_ADDR_LEN,
3184                                              spec.eth->dst.addr_bytes);
3185                                 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_DST_MASK,
3186                                              ETHER_ADDR_LEN,
3187                                              mask.eth->dst.addr_bytes);
3188                         }
3189                         if (!is_zero_ether_addr(&mask.eth->src)) {
3190                                 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_SRC,
3191                                              ETHER_ADDR_LEN,
3192                                              spec.eth->src.addr_bytes);
3193                                 mnl_attr_put(nlh, TCA_FLOWER_KEY_ETH_SRC_MASK,
3194                                              ETHER_ADDR_LEN,
3195                                              mask.eth->src.addr_bytes);
3196                         }
3197                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3198                         break;
3199                 case RTE_FLOW_ITEM_TYPE_VLAN:
3200                         assert(!encap.hdr);
3201                         assert(!decap.hdr);
3202                         item_flags |= MLX5_FLOW_LAYER_OUTER_VLAN;
3203                         mask.vlan = flow_tcf_item_mask
3204                                 (items, &rte_flow_item_vlan_mask,
3205                                  &flow_tcf_mask_supported.vlan,
3206                                  &flow_tcf_mask_empty.vlan,
3207                                  sizeof(flow_tcf_mask_supported.vlan),
3208                                  error);
3209                         assert(mask.vlan);
3210                         if (!eth_type_set)
3211                                 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_ETH_TYPE,
3212                                                  RTE_BE16(ETH_P_8021Q));
3213                         eth_type_set = 1;
3214                         vlan_present = 1;
3215                         if (mask.vlan == &flow_tcf_mask_empty.vlan)
3216                                 break;
3217                         spec.vlan = items->spec;
3218                         if (mask.vlan->inner_type) {
3219                                 mnl_attr_put_u16(nlh,
3220                                                  TCA_FLOWER_KEY_VLAN_ETH_TYPE,
3221                                                  spec.vlan->inner_type);
3222                                 vlan_eth_type_set = 1;
3223                         }
3224                         if (mask.vlan->tci & RTE_BE16(0xe000))
3225                                 mnl_attr_put_u8(nlh, TCA_FLOWER_KEY_VLAN_PRIO,
3226                                                 (rte_be_to_cpu_16
3227                                                  (spec.vlan->tci) >> 13) & 0x7);
3228                         if (mask.vlan->tci & RTE_BE16(0x0fff))
3229                                 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_VLAN_ID,
3230                                                  rte_be_to_cpu_16
3231                                                  (spec.vlan->tci &
3232                                                   RTE_BE16(0x0fff)));
3233                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3234                         break;
3235                 case RTE_FLOW_ITEM_TYPE_IPV4:
3236                         item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV4;
3237                         mask.ipv4 = flow_tcf_item_mask
3238                                 (items, &rte_flow_item_ipv4_mask,
3239                                  &flow_tcf_mask_supported.ipv4,
3240                                  &flow_tcf_mask_empty.ipv4,
3241                                  sizeof(flow_tcf_mask_supported.ipv4),
3242                                  error);
3243                         assert(mask.ipv4);
3244                         spec.ipv4 = items->spec;
3245                         if (!decap.vxlan) {
3246                                 if (!eth_type_set && !vlan_eth_type_set)
3247                                         mnl_attr_put_u16
3248                                                 (nlh,
3249                                                  vlan_present ?
3250                                                  TCA_FLOWER_KEY_VLAN_ETH_TYPE :
3251                                                  TCA_FLOWER_KEY_ETH_TYPE,
3252                                                  RTE_BE16(ETH_P_IP));
3253                                 eth_type_set = 1;
3254                                 vlan_eth_type_set = 1;
3255                                 if (mask.ipv4 == &flow_tcf_mask_empty.ipv4)
3256                                         break;
3257                                 if (mask.ipv4->hdr.next_proto_id) {
3258                                         mnl_attr_put_u8
3259                                                 (nlh, TCA_FLOWER_KEY_IP_PROTO,
3260                                                  spec.ipv4->hdr.next_proto_id);
3261                                         ip_proto_set = 1;
3262                                 }
3263                         } else {
3264                                 assert(mask.ipv4 != &flow_tcf_mask_empty.ipv4);
3265                         }
3266                         if (mask.ipv4->hdr.src_addr) {
3267                                 mnl_attr_put_u32
3268                                         (nlh, decap.vxlan ?
3269                                          TCA_FLOWER_KEY_ENC_IPV4_SRC :
3270                                          TCA_FLOWER_KEY_IPV4_SRC,
3271                                          spec.ipv4->hdr.src_addr);
3272                                 mnl_attr_put_u32
3273                                         (nlh, decap.vxlan ?
3274                                          TCA_FLOWER_KEY_ENC_IPV4_SRC_MASK :
3275                                          TCA_FLOWER_KEY_IPV4_SRC_MASK,
3276                                          mask.ipv4->hdr.src_addr);
3277                         }
3278                         if (mask.ipv4->hdr.dst_addr) {
3279                                 mnl_attr_put_u32
3280                                         (nlh, decap.vxlan ?
3281                                          TCA_FLOWER_KEY_ENC_IPV4_DST :
3282                                          TCA_FLOWER_KEY_IPV4_DST,
3283                                          spec.ipv4->hdr.dst_addr);
3284                                 mnl_attr_put_u32
3285                                         (nlh, decap.vxlan ?
3286                                          TCA_FLOWER_KEY_ENC_IPV4_DST_MASK :
3287                                          TCA_FLOWER_KEY_IPV4_DST_MASK,
3288                                          mask.ipv4->hdr.dst_addr);
3289                         }
3290                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3291                         break;
3292                 case RTE_FLOW_ITEM_TYPE_IPV6:
3293                         item_flags |= MLX5_FLOW_LAYER_OUTER_L3_IPV6;
3294                         mask.ipv6 = flow_tcf_item_mask
3295                                 (items, &rte_flow_item_ipv6_mask,
3296                                  &flow_tcf_mask_supported.ipv6,
3297                                  &flow_tcf_mask_empty.ipv6,
3298                                  sizeof(flow_tcf_mask_supported.ipv6),
3299                                  error);
3300                         assert(mask.ipv6);
3301                         spec.ipv6 = items->spec;
3302                         if (!decap.vxlan) {
3303                                 if (!eth_type_set || !vlan_eth_type_set) {
3304                                         mnl_attr_put_u16
3305                                                 (nlh,
3306                                                  vlan_present ?
3307                                                  TCA_FLOWER_KEY_VLAN_ETH_TYPE :
3308                                                  TCA_FLOWER_KEY_ETH_TYPE,
3309                                                  RTE_BE16(ETH_P_IPV6));
3310                                 }
3311                                 eth_type_set = 1;
3312                                 vlan_eth_type_set = 1;
3313                                 if (mask.ipv6 == &flow_tcf_mask_empty.ipv6)
3314                                         break;
3315                                 if (mask.ipv6->hdr.proto) {
3316                                         mnl_attr_put_u8
3317                                                 (nlh, TCA_FLOWER_KEY_IP_PROTO,
3318                                                  spec.ipv6->hdr.proto);
3319                                         ip_proto_set = 1;
3320                                 }
3321                         } else {
3322                                 assert(mask.ipv6 != &flow_tcf_mask_empty.ipv6);
3323                         }
3324                         if (!IN6_IS_ADDR_UNSPECIFIED(mask.ipv6->hdr.src_addr)) {
3325                                 mnl_attr_put(nlh, decap.vxlan ?
3326                                              TCA_FLOWER_KEY_ENC_IPV6_SRC :
3327                                              TCA_FLOWER_KEY_IPV6_SRC,
3328                                              IPV6_ADDR_LEN,
3329                                              spec.ipv6->hdr.src_addr);
3330                                 mnl_attr_put(nlh, decap.vxlan ?
3331                                              TCA_FLOWER_KEY_ENC_IPV6_SRC_MASK :
3332                                              TCA_FLOWER_KEY_IPV6_SRC_MASK,
3333                                              IPV6_ADDR_LEN,
3334                                              mask.ipv6->hdr.src_addr);
3335                         }
3336                         if (!IN6_IS_ADDR_UNSPECIFIED(mask.ipv6->hdr.dst_addr)) {
3337                                 mnl_attr_put(nlh, decap.vxlan ?
3338                                              TCA_FLOWER_KEY_ENC_IPV6_DST :
3339                                              TCA_FLOWER_KEY_IPV6_DST,
3340                                              IPV6_ADDR_LEN,
3341                                              spec.ipv6->hdr.dst_addr);
3342                                 mnl_attr_put(nlh, decap.vxlan ?
3343                                              TCA_FLOWER_KEY_ENC_IPV6_DST_MASK :
3344                                              TCA_FLOWER_KEY_IPV6_DST_MASK,
3345                                              IPV6_ADDR_LEN,
3346                                              mask.ipv6->hdr.dst_addr);
3347                         }
3348                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3349                         break;
3350                 case RTE_FLOW_ITEM_TYPE_UDP:
3351                         item_flags |= MLX5_FLOW_LAYER_OUTER_L4_UDP;
3352                         mask.udp = flow_tcf_item_mask
3353                                 (items, &rte_flow_item_udp_mask,
3354                                  &flow_tcf_mask_supported.udp,
3355                                  &flow_tcf_mask_empty.udp,
3356                                  sizeof(flow_tcf_mask_supported.udp),
3357                                  error);
3358                         assert(mask.udp);
3359                         spec.udp = items->spec;
3360                         if (!decap.vxlan) {
3361                                 if (!ip_proto_set)
3362                                         mnl_attr_put_u8
3363                                                 (nlh, TCA_FLOWER_KEY_IP_PROTO,
3364                                                 IPPROTO_UDP);
3365                                 if (mask.udp == &flow_tcf_mask_empty.udp)
3366                                         break;
3367                         } else {
3368                                 assert(mask.udp != &flow_tcf_mask_empty.udp);
3369                                 decap.vxlan->udp_port =
3370                                         rte_be_to_cpu_16
3371                                                 (spec.udp->hdr.dst_port);
3372                         }
3373                         if (mask.udp->hdr.src_port) {
3374                                 mnl_attr_put_u16
3375                                         (nlh, decap.vxlan ?
3376                                          TCA_FLOWER_KEY_ENC_UDP_SRC_PORT :
3377                                          TCA_FLOWER_KEY_UDP_SRC,
3378                                          spec.udp->hdr.src_port);
3379                                 mnl_attr_put_u16
3380                                         (nlh, decap.vxlan ?
3381                                          TCA_FLOWER_KEY_ENC_UDP_SRC_PORT_MASK :
3382                                          TCA_FLOWER_KEY_UDP_SRC_MASK,
3383                                          mask.udp->hdr.src_port);
3384                         }
3385                         if (mask.udp->hdr.dst_port) {
3386                                 mnl_attr_put_u16
3387                                         (nlh, decap.vxlan ?
3388                                          TCA_FLOWER_KEY_ENC_UDP_DST_PORT :
3389                                          TCA_FLOWER_KEY_UDP_DST,
3390                                          spec.udp->hdr.dst_port);
3391                                 mnl_attr_put_u16
3392                                         (nlh, decap.vxlan ?
3393                                          TCA_FLOWER_KEY_ENC_UDP_DST_PORT_MASK :
3394                                          TCA_FLOWER_KEY_UDP_DST_MASK,
3395                                          mask.udp->hdr.dst_port);
3396                         }
3397                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3398                         break;
3399                 case RTE_FLOW_ITEM_TYPE_TCP:
3400                         item_flags |= MLX5_FLOW_LAYER_OUTER_L4_TCP;
3401                         mask.tcp = flow_tcf_item_mask
3402                                 (items, &rte_flow_item_tcp_mask,
3403                                  &flow_tcf_mask_supported.tcp,
3404                                  &flow_tcf_mask_empty.tcp,
3405                                  sizeof(flow_tcf_mask_supported.tcp),
3406                                  error);
3407                         assert(mask.tcp);
3408                         if (!ip_proto_set)
3409                                 mnl_attr_put_u8(nlh, TCA_FLOWER_KEY_IP_PROTO,
3410                                                 IPPROTO_TCP);
3411                         if (mask.tcp == &flow_tcf_mask_empty.tcp)
3412                                 break;
3413                         spec.tcp = items->spec;
3414                         if (mask.tcp->hdr.src_port) {
3415                                 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_TCP_SRC,
3416                                                  spec.tcp->hdr.src_port);
3417                                 mnl_attr_put_u16(nlh,
3418                                                  TCA_FLOWER_KEY_TCP_SRC_MASK,
3419                                                  mask.tcp->hdr.src_port);
3420                         }
3421                         if (mask.tcp->hdr.dst_port) {
3422                                 mnl_attr_put_u16(nlh, TCA_FLOWER_KEY_TCP_DST,
3423                                                  spec.tcp->hdr.dst_port);
3424                                 mnl_attr_put_u16(nlh,
3425                                                  TCA_FLOWER_KEY_TCP_DST_MASK,
3426                                                  mask.tcp->hdr.dst_port);
3427                         }
3428                         if (mask.tcp->hdr.tcp_flags) {
3429                                 mnl_attr_put_u16
3430                                         (nlh,
3431                                          TCA_FLOWER_KEY_TCP_FLAGS,
3432                                          rte_cpu_to_be_16
3433                                                 (spec.tcp->hdr.tcp_flags));
3434                                 mnl_attr_put_u16
3435                                         (nlh,
3436                                          TCA_FLOWER_KEY_TCP_FLAGS_MASK,
3437                                          rte_cpu_to_be_16
3438                                                 (mask.tcp->hdr.tcp_flags));
3439                         }
3440                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3441                         break;
3442                 case RTE_FLOW_ITEM_TYPE_VXLAN:
3443                         assert(decap.vxlan);
3444                         item_flags |= MLX5_FLOW_LAYER_VXLAN;
3445                         spec.vxlan = items->spec;
3446                         mnl_attr_put_u32(nlh,
3447                                          TCA_FLOWER_KEY_ENC_KEY_ID,
3448                                          vxlan_vni_as_be32(spec.vxlan->vni));
3449                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3450                         break;
3451                 default:
3452                         return rte_flow_error_set(error, ENOTSUP,
3453                                                   RTE_FLOW_ERROR_TYPE_ITEM,
3454                                                   NULL, "item not supported");
3455                 }
3456         }
3457         na_flower_act = mnl_attr_nest_start(nlh, TCA_FLOWER_ACT);
3458         na_act_index_cur = 1;
3459         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
3460                 struct nlattr *na_act_index;
3461                 struct nlattr *na_act;
3462                 unsigned int vlan_act;
3463                 unsigned int i;
3464
3465                 switch (actions->type) {
3466                 case RTE_FLOW_ACTION_TYPE_VOID:
3467                         break;
3468                 case RTE_FLOW_ACTION_TYPE_PORT_ID:
3469                         conf.port_id = actions->conf;
3470                         if (conf.port_id->original)
3471                                 i = 0;
3472                         else
3473                                 for (i = 0; ptoi[i].ifindex; ++i)
3474                                         if (ptoi[i].port_id == conf.port_id->id)
3475                                                 break;
3476                         assert(ptoi[i].ifindex);
3477                         na_act_index =
3478                                 mnl_attr_nest_start(nlh, na_act_index_cur++);
3479                         assert(na_act_index);
3480                         mnl_attr_put_strz(nlh, TCA_ACT_KIND, "mirred");
3481                         na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3482                         assert(na_act);
3483                         if (encap.hdr) {
3484                                 assert(dev_flow->tcf.tunnel);
3485                                 dev_flow->tcf.tunnel->ifindex_ptr =
3486                                         &((struct tc_mirred *)
3487                                         mnl_attr_get_payload
3488                                         (mnl_nlmsg_get_payload_tail
3489                                                 (nlh)))->ifindex;
3490                         }
3491                         mnl_attr_put(nlh, TCA_MIRRED_PARMS,
3492                                      sizeof(struct tc_mirred),
3493                                      &(struct tc_mirred){
3494                                         .action = TC_ACT_STOLEN,
3495                                         .eaction = TCA_EGRESS_REDIR,
3496                                         .ifindex = ptoi[i].ifindex,
3497                                      });
3498                         mnl_attr_nest_end(nlh, na_act);
3499                         mnl_attr_nest_end(nlh, na_act_index);
3500                         break;
3501                 case RTE_FLOW_ACTION_TYPE_JUMP:
3502                         conf.jump = actions->conf;
3503                         na_act_index =
3504                                 mnl_attr_nest_start(nlh, na_act_index_cur++);
3505                         assert(na_act_index);
3506                         mnl_attr_put_strz(nlh, TCA_ACT_KIND, "gact");
3507                         na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3508                         assert(na_act);
3509                         mnl_attr_put(nlh, TCA_GACT_PARMS,
3510                                      sizeof(struct tc_gact),
3511                                      &(struct tc_gact){
3512                                         .action = TC_ACT_GOTO_CHAIN |
3513                                                   conf.jump->group,
3514                                      });
3515                         mnl_attr_nest_end(nlh, na_act);
3516                         mnl_attr_nest_end(nlh, na_act_index);
3517                         break;
3518                 case RTE_FLOW_ACTION_TYPE_DROP:
3519                         na_act_index =
3520                                 mnl_attr_nest_start(nlh, na_act_index_cur++);
3521                         assert(na_act_index);
3522                         mnl_attr_put_strz(nlh, TCA_ACT_KIND, "gact");
3523                         na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3524                         assert(na_act);
3525                         mnl_attr_put(nlh, TCA_GACT_PARMS,
3526                                      sizeof(struct tc_gact),
3527                                      &(struct tc_gact){
3528                                         .action = TC_ACT_SHOT,
3529                                      });
3530                         mnl_attr_nest_end(nlh, na_act);
3531                         mnl_attr_nest_end(nlh, na_act_index);
3532                         break;
3533                 case RTE_FLOW_ACTION_TYPE_COUNT:
3534                         /*
3535                          * Driver adds the count action implicitly for
3536                          * each rule it creates.
3537                          */
3538                         ret = flow_tcf_translate_action_count(dev,
3539                                                               dev_flow, error);
3540                         if (ret < 0)
3541                                 return ret;
3542                         break;
3543                 case RTE_FLOW_ACTION_TYPE_OF_POP_VLAN:
3544                         conf.of_push_vlan = NULL;
3545                         vlan_act = TCA_VLAN_ACT_POP;
3546                         goto action_of_vlan;
3547                 case RTE_FLOW_ACTION_TYPE_OF_PUSH_VLAN:
3548                         conf.of_push_vlan = actions->conf;
3549                         vlan_act = TCA_VLAN_ACT_PUSH;
3550                         goto action_of_vlan;
3551                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID:
3552                         conf.of_set_vlan_vid = actions->conf;
3553                         if (na_vlan_id)
3554                                 goto override_na_vlan_id;
3555                         vlan_act = TCA_VLAN_ACT_MODIFY;
3556                         goto action_of_vlan;
3557                 case RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP:
3558                         conf.of_set_vlan_pcp = actions->conf;
3559                         if (na_vlan_priority)
3560                                 goto override_na_vlan_priority;
3561                         vlan_act = TCA_VLAN_ACT_MODIFY;
3562                         goto action_of_vlan;
3563 action_of_vlan:
3564                         na_act_index =
3565                                 mnl_attr_nest_start(nlh, na_act_index_cur++);
3566                         assert(na_act_index);
3567                         mnl_attr_put_strz(nlh, TCA_ACT_KIND, "vlan");
3568                         na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3569                         assert(na_act);
3570                         mnl_attr_put(nlh, TCA_VLAN_PARMS,
3571                                      sizeof(struct tc_vlan),
3572                                      &(struct tc_vlan){
3573                                         .action = TC_ACT_PIPE,
3574                                         .v_action = vlan_act,
3575                                      });
3576                         if (vlan_act == TCA_VLAN_ACT_POP) {
3577                                 mnl_attr_nest_end(nlh, na_act);
3578                                 mnl_attr_nest_end(nlh, na_act_index);
3579                                 break;
3580                         }
3581                         if (vlan_act == TCA_VLAN_ACT_PUSH)
3582                                 mnl_attr_put_u16(nlh,
3583                                                  TCA_VLAN_PUSH_VLAN_PROTOCOL,
3584                                                  conf.of_push_vlan->ethertype);
3585                         na_vlan_id = mnl_nlmsg_get_payload_tail(nlh);
3586                         mnl_attr_put_u16(nlh, TCA_VLAN_PAD, 0);
3587                         na_vlan_priority = mnl_nlmsg_get_payload_tail(nlh);
3588                         mnl_attr_put_u8(nlh, TCA_VLAN_PAD, 0);
3589                         mnl_attr_nest_end(nlh, na_act);
3590                         mnl_attr_nest_end(nlh, na_act_index);
3591                         if (actions->type ==
3592                             RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_VID) {
3593 override_na_vlan_id:
3594                                 na_vlan_id->nla_type = TCA_VLAN_PUSH_VLAN_ID;
3595                                 *(uint16_t *)mnl_attr_get_payload(na_vlan_id) =
3596                                         rte_be_to_cpu_16
3597                                         (conf.of_set_vlan_vid->vlan_vid);
3598                         } else if (actions->type ==
3599                                    RTE_FLOW_ACTION_TYPE_OF_SET_VLAN_PCP) {
3600 override_na_vlan_priority:
3601                                 na_vlan_priority->nla_type =
3602                                         TCA_VLAN_PUSH_VLAN_PRIORITY;
3603                                 *(uint8_t *)mnl_attr_get_payload
3604                                         (na_vlan_priority) =
3605                                         conf.of_set_vlan_pcp->vlan_pcp;
3606                         }
3607                         break;
3608                 case RTE_FLOW_ACTION_TYPE_VXLAN_DECAP:
3609                         assert(decap.vxlan);
3610                         assert(dev_flow->tcf.tunnel);
3611                         dev_flow->tcf.tunnel->ifindex_ptr =
3612                                 (unsigned int *)&tcm->tcm_ifindex;
3613                         na_act_index =
3614                                 mnl_attr_nest_start(nlh, na_act_index_cur++);
3615                         assert(na_act_index);
3616                         mnl_attr_put_strz(nlh, TCA_ACT_KIND, "tunnel_key");
3617                         na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3618                         assert(na_act);
3619                         mnl_attr_put(nlh, TCA_TUNNEL_KEY_PARMS,
3620                                 sizeof(struct tc_tunnel_key),
3621                                 &(struct tc_tunnel_key){
3622                                         .action = TC_ACT_PIPE,
3623                                         .t_action = TCA_TUNNEL_KEY_ACT_RELEASE,
3624                                         });
3625                         mnl_attr_nest_end(nlh, na_act);
3626                         mnl_attr_nest_end(nlh, na_act_index);
3627                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3628                         break;
3629                 case RTE_FLOW_ACTION_TYPE_VXLAN_ENCAP:
3630                         assert(encap.vxlan);
3631                         flow_tcf_vxlan_encap_parse(actions, encap.vxlan);
3632                         na_act_index =
3633                                 mnl_attr_nest_start(nlh, na_act_index_cur++);
3634                         assert(na_act_index);
3635                         mnl_attr_put_strz(nlh, TCA_ACT_KIND, "tunnel_key");
3636                         na_act = mnl_attr_nest_start(nlh, TCA_ACT_OPTIONS);
3637                         assert(na_act);
3638                         mnl_attr_put(nlh, TCA_TUNNEL_KEY_PARMS,
3639                                 sizeof(struct tc_tunnel_key),
3640                                 &(struct tc_tunnel_key){
3641                                         .action = TC_ACT_PIPE,
3642                                         .t_action = TCA_TUNNEL_KEY_ACT_SET,
3643                                         });
3644                         if (encap.vxlan->mask & FLOW_TCF_ENCAP_UDP_DST)
3645                                 mnl_attr_put_u16(nlh,
3646                                          TCA_TUNNEL_KEY_ENC_DST_PORT,
3647                                          encap.vxlan->udp.dst);
3648                         if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV4_SRC)
3649                                 mnl_attr_put_u32(nlh,
3650                                          TCA_TUNNEL_KEY_ENC_IPV4_SRC,
3651                                          encap.vxlan->ipv4.src);
3652                         if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV4_DST)
3653                                 mnl_attr_put_u32(nlh,
3654                                          TCA_TUNNEL_KEY_ENC_IPV4_DST,
3655                                          encap.vxlan->ipv4.dst);
3656                         if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV6_SRC)
3657                                 mnl_attr_put(nlh,
3658                                          TCA_TUNNEL_KEY_ENC_IPV6_SRC,
3659                                          sizeof(encap.vxlan->ipv6.src),
3660                                          &encap.vxlan->ipv6.src);
3661                         if (encap.vxlan->mask & FLOW_TCF_ENCAP_IPV6_DST)
3662                                 mnl_attr_put(nlh,
3663                                          TCA_TUNNEL_KEY_ENC_IPV6_DST,
3664                                          sizeof(encap.vxlan->ipv6.dst),
3665                                          &encap.vxlan->ipv6.dst);
3666                         if (encap.vxlan->mask & FLOW_TCF_ENCAP_VXLAN_VNI)
3667                                 mnl_attr_put_u32(nlh,
3668                                          TCA_TUNNEL_KEY_ENC_KEY_ID,
3669                                          vxlan_vni_as_be32
3670                                                 (encap.vxlan->vxlan.vni));
3671                         mnl_attr_put_u8(nlh, TCA_TUNNEL_KEY_NO_CSUM, 0);
3672                         mnl_attr_nest_end(nlh, na_act);
3673                         mnl_attr_nest_end(nlh, na_act_index);
3674                         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3675                         break;
3676                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_SRC:
3677                 case RTE_FLOW_ACTION_TYPE_SET_IPV4_DST:
3678                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_SRC:
3679                 case RTE_FLOW_ACTION_TYPE_SET_IPV6_DST:
3680                 case RTE_FLOW_ACTION_TYPE_SET_TP_SRC:
3681                 case RTE_FLOW_ACTION_TYPE_SET_TP_DST:
3682                 case RTE_FLOW_ACTION_TYPE_SET_TTL:
3683                 case RTE_FLOW_ACTION_TYPE_DEC_TTL:
3684                 case RTE_FLOW_ACTION_TYPE_SET_MAC_SRC:
3685                 case RTE_FLOW_ACTION_TYPE_SET_MAC_DST:
3686                         na_act_index =
3687                                 mnl_attr_nest_start(nlh, na_act_index_cur++);
3688                         flow_tcf_create_pedit_mnl_msg(nlh,
3689                                                       &actions, item_flags);
3690                         mnl_attr_nest_end(nlh, na_act_index);
3691                         break;
3692                 default:
3693                         return rte_flow_error_set(error, ENOTSUP,
3694                                                   RTE_FLOW_ERROR_TYPE_ACTION,
3695                                                   actions,
3696                                                   "action not supported");
3697                 }
3698         }
3699         assert(na_flower);
3700         assert(na_flower_act);
3701         mnl_attr_nest_end(nlh, na_flower_act);
3702         mnl_attr_put_u32(nlh, TCA_FLOWER_FLAGS, decap.vxlan ?
3703                                                 0 : TCA_CLS_FLAGS_SKIP_SW);
3704         mnl_attr_nest_end(nlh, na_flower);
3705         if (dev_flow->tcf.tunnel && dev_flow->tcf.tunnel->ifindex_ptr)
3706                 dev_flow->tcf.tunnel->ifindex_org =
3707                         *dev_flow->tcf.tunnel->ifindex_ptr;
3708         assert(dev_flow->tcf.nlsize >= nlh->nlmsg_len);
3709         return 0;
3710 }
3711
3712 /**
3713  * Send Netlink message with acknowledgment.
3714  *
3715  * @param tcf
3716  *   Flow context to use.
3717  * @param nlh
3718  *   Message to send. This function always raises the NLM_F_ACK flag before
3719  *   sending.
3720  * @param[in] cb
3721  *   Callback handler for received message.
3722  * @param[in] arg
3723  *   Context pointer for callback handler.
3724  *
3725  * @return
3726  *   0 on success, a negative errno value otherwise and rte_errno is set.
3727  */
3728 static int
3729 flow_tcf_nl_ack(struct mlx5_flow_tcf_context *tcf,
3730                 struct nlmsghdr *nlh,
3731                 mnl_cb_t cb, void *arg)
3732 {
3733         unsigned int portid = mnl_socket_get_portid(tcf->nl);
3734         uint32_t seq = tcf->seq++;
3735         int err, ret;
3736
3737         assert(tcf->nl);
3738         assert(tcf->buf);
3739         if (!seq)
3740                 /* seq 0 is reserved for kernel event-driven notifications. */
3741                 seq = tcf->seq++;
3742         nlh->nlmsg_seq = seq;
3743         nlh->nlmsg_flags |= NLM_F_ACK;
3744         ret = mnl_socket_sendto(tcf->nl, nlh, nlh->nlmsg_len);
3745         err = (ret <= 0) ? errno : 0;
3746         nlh = (struct nlmsghdr *)(tcf->buf);
3747         /*
3748          * The following loop postpones non-fatal errors until multipart
3749          * messages are complete.
3750          */
3751         if (ret > 0)
3752                 while (true) {
3753                         ret = mnl_socket_recvfrom(tcf->nl, tcf->buf,
3754                                                   tcf->buf_size);
3755                         if (ret < 0) {
3756                                 err = errno;
3757                                 if (err != ENOSPC)
3758                                         break;
3759                         }
3760                         if (!err) {
3761                                 ret = mnl_cb_run(nlh, ret, seq, portid,
3762                                                  cb, arg);
3763                                 if (ret < 0) {
3764                                         err = errno;
3765                                         break;
3766                                 }
3767                         }
3768                         /* Will receive till end of multipart message */
3769                         if (!(nlh->nlmsg_flags & NLM_F_MULTI) ||
3770                               nlh->nlmsg_type == NLMSG_DONE)
3771                                 break;
3772                 }
3773         if (!err)
3774                 return 0;
3775         rte_errno = err;
3776         return -err;
3777 }
3778
3779 #define MNL_BUF_EXTRA_SPACE 16
3780 #define MNL_REQUEST_SIZE_MIN 256
3781 #define MNL_REQUEST_SIZE_MAX 2048
3782 #define MNL_REQUEST_SIZE RTE_MIN(RTE_MAX(sysconf(_SC_PAGESIZE), \
3783                                  MNL_REQUEST_SIZE_MIN), MNL_REQUEST_SIZE_MAX)
3784
3785 /* Data structures used by flow_tcf_xxx_cb() routines. */
3786 struct tcf_nlcb_buf {
3787         LIST_ENTRY(tcf_nlcb_buf) next;
3788         uint32_t size;
3789         alignas(struct nlmsghdr)
3790         uint8_t msg[]; /**< Netlink message data. */
3791 };
3792
3793 struct tcf_nlcb_context {
3794         unsigned int ifindex; /**< Base interface index. */
3795         uint32_t bufsize;
3796         LIST_HEAD(, tcf_nlcb_buf) nlbuf;
3797 };
3798
3799 /**
3800  * Allocate space for netlink command in buffer list
3801  *
3802  * @param[in, out] ctx
3803  *   Pointer to callback context with command buffers list.
3804  * @param[in] size
3805  *   Required size of data buffer to be allocated.
3806  *
3807  * @return
3808  *   Pointer to allocated memory, aligned as message header.
3809  *   NULL if some error occurred.
3810  */
3811 static struct nlmsghdr *
3812 flow_tcf_alloc_nlcmd(struct tcf_nlcb_context *ctx, uint32_t size)
3813 {
3814         struct tcf_nlcb_buf *buf;
3815         struct nlmsghdr *nlh;
3816
3817         size = NLMSG_ALIGN(size);
3818         buf = LIST_FIRST(&ctx->nlbuf);
3819         if (buf && (buf->size + size) <= ctx->bufsize) {
3820                 nlh = (struct nlmsghdr *)&buf->msg[buf->size];
3821                 buf->size += size;
3822                 return nlh;
3823         }
3824         if (size > ctx->bufsize) {
3825                 DRV_LOG(WARNING, "netlink: too long command buffer requested");
3826                 return NULL;
3827         }
3828         buf = rte_malloc(__func__,
3829                         ctx->bufsize + sizeof(struct tcf_nlcb_buf),
3830                         alignof(struct tcf_nlcb_buf));
3831         if (!buf) {
3832                 DRV_LOG(WARNING, "netlink: no memory for command buffer");
3833                 return NULL;
3834         }
3835         LIST_INSERT_HEAD(&ctx->nlbuf, buf, next);
3836         buf->size = size;
3837         nlh = (struct nlmsghdr *)&buf->msg[0];
3838         return nlh;
3839 }
3840
3841 /**
3842  * Send the buffers with prepared netlink commands. Scans the list and
3843  * sends all found buffers. Buffers are sent and freed anyway in order
3844  * to prevent memory leakage if some every message in received packet.
3845  *
3846  * @param[in] tcf
3847  *   Context object initialized by mlx5_flow_tcf_context_create().
3848  * @param[in, out] ctx
3849  *   Pointer to callback context with command buffers list.
3850  *
3851  * @return
3852  *   Zero value on success, negative errno value otherwise
3853  *   and rte_errno is set.
3854  */
3855 static int
3856 flow_tcf_send_nlcmd(struct mlx5_flow_tcf_context *tcf,
3857                     struct tcf_nlcb_context *ctx)
3858 {
3859         struct tcf_nlcb_buf *bc = LIST_FIRST(&ctx->nlbuf);
3860         int ret = 0;
3861
3862         while (bc) {
3863                 struct tcf_nlcb_buf *bn = LIST_NEXT(bc, next);
3864                 struct nlmsghdr *nlh;
3865                 uint32_t msg = 0;
3866                 int rc;
3867
3868                 while (msg < bc->size) {
3869                         /*
3870                          * Send Netlink commands from buffer in one by one
3871                          * fashion. If we send multiple rule deletion commands
3872                          * in one Netlink message and some error occurs it may
3873                          * cause multiple ACK error messages and break sequence
3874                          * numbers of Netlink communication, because we expect
3875                          * the only one ACK reply.
3876                          */
3877                         assert((bc->size - msg) >= sizeof(struct nlmsghdr));
3878                         nlh = (struct nlmsghdr *)&bc->msg[msg];
3879                         assert((bc->size - msg) >= nlh->nlmsg_len);
3880                         msg += nlh->nlmsg_len;
3881                         rc = flow_tcf_nl_ack(tcf, nlh, NULL, NULL);
3882                         if (rc) {
3883                                 DRV_LOG(WARNING,
3884                                         "netlink: cleanup error %d", rc);
3885                                 if (!ret)
3886                                         ret = rc;
3887                         }
3888                 }
3889                 rte_free(bc);
3890                 bc = bn;
3891         }
3892         LIST_INIT(&ctx->nlbuf);
3893         return ret;
3894 }
3895
3896 /**
3897  * Collect local IP address rules with scope link attribute  on specified
3898  * network device. This is callback routine called by libmnl mnl_cb_run()
3899  * in loop for every message in received packet.
3900  *
3901  * @param[in] nlh
3902  *   Pointer to reply header.
3903  * @param[in, out] arg
3904  *   Opaque data pointer for this callback.
3905  *
3906  * @return
3907  *   A positive, nonzero value on success, negative errno value otherwise
3908  *   and rte_errno is set.
3909  */
3910 static int
3911 flow_tcf_collect_local_cb(const struct nlmsghdr *nlh, void *arg)
3912 {
3913         struct tcf_nlcb_context *ctx = arg;
3914         struct nlmsghdr *cmd;
3915         struct ifaddrmsg *ifa;
3916         struct nlattr *na;
3917         struct nlattr *na_local = NULL;
3918         struct nlattr *na_peer = NULL;
3919         unsigned char family;
3920         uint32_t size;
3921
3922         if (nlh->nlmsg_type != RTM_NEWADDR) {
3923                 rte_errno = EINVAL;
3924                 return -rte_errno;
3925         }
3926         ifa = mnl_nlmsg_get_payload(nlh);
3927         family = ifa->ifa_family;
3928         if (ifa->ifa_index != ctx->ifindex ||
3929             ifa->ifa_scope != RT_SCOPE_LINK ||
3930             !(ifa->ifa_flags & IFA_F_PERMANENT) ||
3931             (family != AF_INET && family != AF_INET6))
3932                 return 1;
3933         mnl_attr_for_each(na, nlh, sizeof(*ifa)) {
3934                 switch (mnl_attr_get_type(na)) {
3935                 case IFA_LOCAL:
3936                         na_local = na;
3937                         break;
3938                 case IFA_ADDRESS:
3939                         na_peer = na;
3940                         break;
3941                 }
3942                 if (na_local && na_peer)
3943                         break;
3944         }
3945         if (!na_local || !na_peer)
3946                 return 1;
3947         /* Local rule found with scope link, permanent and assigned peer. */
3948         size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
3949                MNL_ALIGN(sizeof(struct ifaddrmsg)) +
3950                (family == AF_INET6 ? 2 * SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN)
3951                                    : 2 * SZ_NLATTR_TYPE_OF(uint32_t));
3952         cmd = flow_tcf_alloc_nlcmd(ctx, size);
3953         if (!cmd) {
3954                 rte_errno = ENOMEM;
3955                 return -rte_errno;
3956         }
3957         cmd = mnl_nlmsg_put_header(cmd);
3958         cmd->nlmsg_type = RTM_DELADDR;
3959         cmd->nlmsg_flags = NLM_F_REQUEST;
3960         ifa = mnl_nlmsg_put_extra_header(cmd, sizeof(*ifa));
3961         ifa->ifa_flags = IFA_F_PERMANENT;
3962         ifa->ifa_scope = RT_SCOPE_LINK;
3963         ifa->ifa_index = ctx->ifindex;
3964         if (family == AF_INET) {
3965                 ifa->ifa_family = AF_INET;
3966                 ifa->ifa_prefixlen = 32;
3967                 mnl_attr_put_u32(cmd, IFA_LOCAL, mnl_attr_get_u32(na_local));
3968                 mnl_attr_put_u32(cmd, IFA_ADDRESS, mnl_attr_get_u32(na_peer));
3969         } else {
3970                 ifa->ifa_family = AF_INET6;
3971                 ifa->ifa_prefixlen = 128;
3972                 mnl_attr_put(cmd, IFA_LOCAL, IPV6_ADDR_LEN,
3973                         mnl_attr_get_payload(na_local));
3974                 mnl_attr_put(cmd, IFA_ADDRESS, IPV6_ADDR_LEN,
3975                         mnl_attr_get_payload(na_peer));
3976         }
3977         assert(size == cmd->nlmsg_len);
3978         return 1;
3979 }
3980
3981 /**
3982  * Cleanup the local IP addresses on outer interface.
3983  *
3984  * @param[in] tcf
3985  *   Context object initialized by mlx5_flow_tcf_context_create().
3986  * @param[in] ifindex
3987  *   Network inferface index to perform cleanup.
3988  */
3989 static void
3990 flow_tcf_encap_local_cleanup(struct mlx5_flow_tcf_context *tcf,
3991                             unsigned int ifindex)
3992 {
3993         struct nlmsghdr *nlh;
3994         struct ifaddrmsg *ifa;
3995         struct tcf_nlcb_context ctx = {
3996                 .ifindex = ifindex,
3997                 .bufsize = MNL_REQUEST_SIZE,
3998                 .nlbuf = LIST_HEAD_INITIALIZER(),
3999         };
4000         int ret;
4001
4002         assert(ifindex);
4003         /*
4004          * Seek and destroy leftovers of local IP addresses with
4005          * matching properties "scope link".
4006          */
4007         nlh = mnl_nlmsg_put_header(tcf->buf);
4008         nlh->nlmsg_type = RTM_GETADDR;
4009         nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4010         ifa = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifa));
4011         ifa->ifa_family = AF_UNSPEC;
4012         ifa->ifa_index = ifindex;
4013         ifa->ifa_scope = RT_SCOPE_LINK;
4014         ret = flow_tcf_nl_ack(tcf, nlh, flow_tcf_collect_local_cb, &ctx);
4015         if (ret)
4016                 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4017         ret = flow_tcf_send_nlcmd(tcf, &ctx);
4018         if (ret)
4019                 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4020 }
4021
4022 /**
4023  * Collect neigh permament rules on specified network device.
4024  * This is callback routine called by libmnl mnl_cb_run() in loop for
4025  * every message in received packet.
4026  *
4027  * @param[in] nlh
4028  *   Pointer to reply header.
4029  * @param[in, out] arg
4030  *   Opaque data pointer for this callback.
4031  *
4032  * @return
4033  *   A positive, nonzero value on success, negative errno value otherwise
4034  *   and rte_errno is set.
4035  */
4036 static int
4037 flow_tcf_collect_neigh_cb(const struct nlmsghdr *nlh, void *arg)
4038 {
4039         struct tcf_nlcb_context *ctx = arg;
4040         struct nlmsghdr *cmd;
4041         struct ndmsg *ndm;
4042         struct nlattr *na;
4043         struct nlattr *na_ip = NULL;
4044         struct nlattr *na_mac = NULL;
4045         unsigned char family;
4046         uint32_t size;
4047
4048         if (nlh->nlmsg_type != RTM_NEWNEIGH) {
4049                 rte_errno = EINVAL;
4050                 return -rte_errno;
4051         }
4052         ndm = mnl_nlmsg_get_payload(nlh);
4053         family = ndm->ndm_family;
4054         if (ndm->ndm_ifindex != (int)ctx->ifindex ||
4055            !(ndm->ndm_state & NUD_PERMANENT) ||
4056            (family != AF_INET && family != AF_INET6))
4057                 return 1;
4058         mnl_attr_for_each(na, nlh, sizeof(*ndm)) {
4059                 switch (mnl_attr_get_type(na)) {
4060                 case NDA_DST:
4061                         na_ip = na;
4062                         break;
4063                 case NDA_LLADDR:
4064                         na_mac = na;
4065                         break;
4066                 }
4067                 if (na_mac && na_ip)
4068                         break;
4069         }
4070         if (!na_mac || !na_ip)
4071                 return 1;
4072         /* Neigh rule with permenent attribute found. */
4073         size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
4074                MNL_ALIGN(sizeof(struct ndmsg)) +
4075                SZ_NLATTR_DATA_OF(ETHER_ADDR_LEN) +
4076                (family == AF_INET6 ? SZ_NLATTR_DATA_OF(IPV6_ADDR_LEN)
4077                                    : SZ_NLATTR_TYPE_OF(uint32_t));
4078         cmd = flow_tcf_alloc_nlcmd(ctx, size);
4079         if (!cmd) {
4080                 rte_errno = ENOMEM;
4081                 return -rte_errno;
4082         }
4083         cmd = mnl_nlmsg_put_header(cmd);
4084         cmd->nlmsg_type = RTM_DELNEIGH;
4085         cmd->nlmsg_flags = NLM_F_REQUEST;
4086         ndm = mnl_nlmsg_put_extra_header(cmd, sizeof(*ndm));
4087         ndm->ndm_ifindex = ctx->ifindex;
4088         ndm->ndm_state = NUD_PERMANENT;
4089         ndm->ndm_flags = 0;
4090         ndm->ndm_type = 0;
4091         if (family == AF_INET) {
4092                 ndm->ndm_family = AF_INET;
4093                 mnl_attr_put_u32(cmd, NDA_DST, mnl_attr_get_u32(na_ip));
4094         } else {
4095                 ndm->ndm_family = AF_INET6;
4096                 mnl_attr_put(cmd, NDA_DST, IPV6_ADDR_LEN,
4097                              mnl_attr_get_payload(na_ip));
4098         }
4099         mnl_attr_put(cmd, NDA_LLADDR, ETHER_ADDR_LEN,
4100                      mnl_attr_get_payload(na_mac));
4101         assert(size == cmd->nlmsg_len);
4102         return 1;
4103 }
4104
4105 /**
4106  * Cleanup the neigh rules on outer interface.
4107  *
4108  * @param[in] tcf
4109  *   Context object initialized by mlx5_flow_tcf_context_create().
4110  * @param[in] ifindex
4111  *   Network inferface index to perform cleanup.
4112  */
4113 static void
4114 flow_tcf_encap_neigh_cleanup(struct mlx5_flow_tcf_context *tcf,
4115                             unsigned int ifindex)
4116 {
4117         struct nlmsghdr *nlh;
4118         struct ndmsg *ndm;
4119         struct tcf_nlcb_context ctx = {
4120                 .ifindex = ifindex,
4121                 .bufsize = MNL_REQUEST_SIZE,
4122                 .nlbuf = LIST_HEAD_INITIALIZER(),
4123         };
4124         int ret;
4125
4126         assert(ifindex);
4127         /* Seek and destroy leftovers of neigh rules. */
4128         nlh = mnl_nlmsg_put_header(tcf->buf);
4129         nlh->nlmsg_type = RTM_GETNEIGH;
4130         nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4131         ndm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ndm));
4132         ndm->ndm_family = AF_UNSPEC;
4133         ndm->ndm_ifindex = ifindex;
4134         ndm->ndm_state = NUD_PERMANENT;
4135         ret = flow_tcf_nl_ack(tcf, nlh, flow_tcf_collect_neigh_cb, &ctx);
4136         if (ret)
4137                 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4138         ret = flow_tcf_send_nlcmd(tcf, &ctx);
4139         if (ret)
4140                 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4141 }
4142
4143 /**
4144  * Collect indices of VXLAN encap/decap interfaces associated with device.
4145  * This is callback routine called by libmnl mnl_cb_run() in loop for
4146  * every message in received packet.
4147  *
4148  * @param[in] nlh
4149  *   Pointer to reply header.
4150  * @param[in, out] arg
4151  *   Opaque data pointer for this callback.
4152  *
4153  * @return
4154  *   A positive, nonzero value on success, negative errno value otherwise
4155  *   and rte_errno is set.
4156  */
4157 static int
4158 flow_tcf_collect_vxlan_cb(const struct nlmsghdr *nlh, void *arg)
4159 {
4160         struct tcf_nlcb_context *ctx = arg;
4161         struct nlmsghdr *cmd;
4162         struct ifinfomsg *ifm;
4163         struct nlattr *na;
4164         struct nlattr *na_info = NULL;
4165         struct nlattr *na_vxlan = NULL;
4166         bool found = false;
4167         unsigned int vxindex;
4168         uint32_t size;
4169
4170         if (nlh->nlmsg_type != RTM_NEWLINK) {
4171                 rte_errno = EINVAL;
4172                 return -rte_errno;
4173         }
4174         ifm = mnl_nlmsg_get_payload(nlh);
4175         if (!ifm->ifi_index) {
4176                 rte_errno = EINVAL;
4177                 return -rte_errno;
4178         }
4179         mnl_attr_for_each(na, nlh, sizeof(*ifm))
4180                 if (mnl_attr_get_type(na) == IFLA_LINKINFO) {
4181                         na_info = na;
4182                         break;
4183                 }
4184         if (!na_info)
4185                 return 1;
4186         mnl_attr_for_each_nested(na, na_info) {
4187                 switch (mnl_attr_get_type(na)) {
4188                 case IFLA_INFO_KIND:
4189                         if (!strncmp("vxlan", mnl_attr_get_str(na),
4190                                      mnl_attr_get_len(na)))
4191                                 found = true;
4192                         break;
4193                 case IFLA_INFO_DATA:
4194                         na_vxlan = na;
4195                         break;
4196                 }
4197                 if (found && na_vxlan)
4198                         break;
4199         }
4200         if (!found || !na_vxlan)
4201                 return 1;
4202         found = false;
4203         mnl_attr_for_each_nested(na, na_vxlan) {
4204                 if (mnl_attr_get_type(na) == IFLA_VXLAN_LINK &&
4205                     mnl_attr_get_u32(na) == ctx->ifindex) {
4206                         found = true;
4207                         break;
4208                 }
4209         }
4210         if (!found)
4211                 return 1;
4212         /* Attached VXLAN device found, store the command to delete. */
4213         vxindex = ifm->ifi_index;
4214         size = MNL_ALIGN(sizeof(struct nlmsghdr)) +
4215                MNL_ALIGN(sizeof(struct ifinfomsg));
4216         cmd = flow_tcf_alloc_nlcmd(ctx, size);
4217         if (!cmd) {
4218                 rte_errno = ENOMEM;
4219                 return -rte_errno;
4220         }
4221         cmd = mnl_nlmsg_put_header(cmd);
4222         cmd->nlmsg_type = RTM_DELLINK;
4223         cmd->nlmsg_flags = NLM_F_REQUEST;
4224         ifm = mnl_nlmsg_put_extra_header(cmd, sizeof(*ifm));
4225         ifm->ifi_family = AF_UNSPEC;
4226         ifm->ifi_index = vxindex;
4227         assert(size == cmd->nlmsg_len);
4228         return 1;
4229 }
4230
4231 /**
4232  * Cleanup the outer interface. Removes all found vxlan devices
4233  * attached to specified index, flushes the meigh and local IP
4234  * datavase.
4235  *
4236  * @param[in] tcf
4237  *   Context object initialized by mlx5_flow_tcf_context_create().
4238  * @param[in] ifindex
4239  *   Network inferface index to perform cleanup.
4240  */
4241 static void
4242 flow_tcf_encap_iface_cleanup(struct mlx5_flow_tcf_context *tcf,
4243                             unsigned int ifindex)
4244 {
4245         struct nlmsghdr *nlh;
4246         struct ifinfomsg *ifm;
4247         struct tcf_nlcb_context ctx = {
4248                 .ifindex = ifindex,
4249                 .bufsize = MNL_REQUEST_SIZE,
4250                 .nlbuf = LIST_HEAD_INITIALIZER(),
4251         };
4252         int ret;
4253
4254         assert(ifindex);
4255         /*
4256          * Seek and destroy leftover VXLAN encap/decap interfaces with
4257          * matching properties.
4258          */
4259         nlh = mnl_nlmsg_put_header(tcf->buf);
4260         nlh->nlmsg_type = RTM_GETLINK;
4261         nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_DUMP;
4262         ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4263         ifm->ifi_family = AF_UNSPEC;
4264         ret = flow_tcf_nl_ack(tcf, nlh, flow_tcf_collect_vxlan_cb, &ctx);
4265         if (ret)
4266                 DRV_LOG(WARNING, "netlink: query device list error %d", ret);
4267         ret = flow_tcf_send_nlcmd(tcf, &ctx);
4268         if (ret)
4269                 DRV_LOG(WARNING, "netlink: device delete error %d", ret);
4270 }
4271
4272 /**
4273  * Emit Netlink message to add/remove local address to the outer device.
4274  * The address being added is visible within the link only (scope link).
4275  *
4276  * Note that an implicit route is maintained by the kernel due to the
4277  * presence of a peer address (IFA_ADDRESS).
4278  *
4279  * These rules are used for encapsultion only and allow to assign
4280  * the outer tunnel source IP address.
4281  *
4282  * @param[in] tcf
4283  *   Libmnl socket context object.
4284  * @param[in] encap
4285  *   Encapsulation properties (source address and its peer).
4286  * @param[in] ifindex
4287  *   Network interface to apply rule.
4288  * @param[in] enable
4289  *   Toggle between add and remove.
4290  * @param[out] error
4291  *   Perform verbose error reporting if not NULL.
4292  *
4293  * @return
4294  *   0 on success, a negative errno value otherwise and rte_errno is set.
4295  */
4296 static int
4297 flow_tcf_rule_local(struct mlx5_flow_tcf_context *tcf,
4298                     const struct flow_tcf_vxlan_encap *encap,
4299                     unsigned int ifindex,
4300                     bool enable,
4301                     struct rte_flow_error *error)
4302 {
4303         struct nlmsghdr *nlh;
4304         struct ifaddrmsg *ifa;
4305         alignas(struct nlmsghdr)
4306         uint8_t buf[mnl_nlmsg_size(sizeof(*ifa) + 128)];
4307
4308         nlh = mnl_nlmsg_put_header(buf);
4309         nlh->nlmsg_type = enable ? RTM_NEWADDR : RTM_DELADDR;
4310         nlh->nlmsg_flags =
4311                 NLM_F_REQUEST | (enable ? NLM_F_CREATE | NLM_F_REPLACE : 0);
4312         nlh->nlmsg_seq = 0;
4313         ifa = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifa));
4314         ifa->ifa_flags = IFA_F_PERMANENT;
4315         ifa->ifa_scope = RT_SCOPE_LINK;
4316         ifa->ifa_index = ifindex;
4317         if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4318                 ifa->ifa_family = AF_INET;
4319                 ifa->ifa_prefixlen = 32;
4320                 mnl_attr_put_u32(nlh, IFA_LOCAL, encap->ipv4.src);
4321                 if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST)
4322                         mnl_attr_put_u32(nlh, IFA_ADDRESS,
4323                                               encap->ipv4.dst);
4324         } else {
4325                 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4326                 ifa->ifa_family = AF_INET6;
4327                 ifa->ifa_prefixlen = 128;
4328                 mnl_attr_put(nlh, IFA_LOCAL,
4329                                   sizeof(encap->ipv6.src),
4330                                   &encap->ipv6.src);
4331                 if (encap->mask & FLOW_TCF_ENCAP_IPV6_DST)
4332                         mnl_attr_put(nlh, IFA_ADDRESS,
4333                                           sizeof(encap->ipv6.dst),
4334                                           &encap->ipv6.dst);
4335         }
4336         if (!flow_tcf_nl_ack(tcf, nlh, NULL, NULL))
4337                 return 0;
4338         return rte_flow_error_set(error, rte_errno,
4339                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4340                                   "netlink: cannot complete IFA request"
4341                                   " (ip addr add)");
4342 }
4343
4344 /**
4345  * Emit Netlink message to add/remove neighbor.
4346  *
4347  * @param[in] tcf
4348  *   Libmnl socket context object.
4349  * @param[in] encap
4350  *   Encapsulation properties (destination address).
4351  * @param[in] ifindex
4352  *   Network interface.
4353  * @param[in] enable
4354  *   Toggle between add and remove.
4355  * @param[out] error
4356  *   Perform verbose error reporting if not NULL.
4357  *
4358  * @return
4359  *   0 on success, a negative errno value otherwise and rte_errno is set.
4360  */
4361 static int
4362 flow_tcf_rule_neigh(struct mlx5_flow_tcf_context *tcf,
4363                      const struct flow_tcf_vxlan_encap *encap,
4364                      unsigned int ifindex,
4365                      bool enable,
4366                      struct rte_flow_error *error)
4367 {
4368         struct nlmsghdr *nlh;
4369         struct ndmsg *ndm;
4370         alignas(struct nlmsghdr)
4371         uint8_t buf[mnl_nlmsg_size(sizeof(*ndm) + 128)];
4372
4373         nlh = mnl_nlmsg_put_header(buf);
4374         nlh->nlmsg_type = enable ? RTM_NEWNEIGH : RTM_DELNEIGH;
4375         nlh->nlmsg_flags =
4376                 NLM_F_REQUEST | (enable ? NLM_F_CREATE | NLM_F_REPLACE : 0);
4377         nlh->nlmsg_seq = 0;
4378         ndm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ndm));
4379         ndm->ndm_ifindex = ifindex;
4380         ndm->ndm_state = NUD_PERMANENT;
4381         ndm->ndm_flags = 0;
4382         ndm->ndm_type = 0;
4383         if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4384                 ndm->ndm_family = AF_INET;
4385                 mnl_attr_put_u32(nlh, NDA_DST, encap->ipv4.dst);
4386         } else {
4387                 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4388                 ndm->ndm_family = AF_INET6;
4389                 mnl_attr_put(nlh, NDA_DST, sizeof(encap->ipv6.dst),
4390                                                  &encap->ipv6.dst);
4391         }
4392         if (encap->mask & FLOW_TCF_ENCAP_ETH_SRC && enable)
4393                 DRV_LOG(WARNING,
4394                         "outer ethernet source address cannot be "
4395                         "forced for VXLAN encapsulation");
4396         if (encap->mask & FLOW_TCF_ENCAP_ETH_DST)
4397                 mnl_attr_put(nlh, NDA_LLADDR, sizeof(encap->eth.dst),
4398                                                     &encap->eth.dst);
4399         if (!flow_tcf_nl_ack(tcf, nlh, NULL, NULL))
4400                 return 0;
4401         return rte_flow_error_set(error, rte_errno,
4402                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4403                                   "netlink: cannot complete ND request"
4404                                   " (ip neigh)");
4405 }
4406
4407 /**
4408  * Manage the local IP addresses and their peers IP addresses on the
4409  * outer interface for encapsulation purposes. The kernel searches the
4410  * appropriate device for tunnel egress traffic using the outer source
4411  * IP, this IP should be assigned to the outer network device, otherwise
4412  * kernel rejects the rule.
4413  *
4414  * Adds or removes the addresses using the Netlink command like this:
4415  *   ip addr add <src_ip> peer <dst_ip> scope link dev <ifouter>
4416  *
4417  * The addresses are local to the netdev ("scope link"), this reduces
4418  * the risk of conflicts. Note that an implicit route is maintained by
4419  * the kernel due to the presence of a peer address (IFA_ADDRESS).
4420  *
4421  * @param[in] tcf
4422  *   Libmnl socket context object.
4423  * @param[in] vtep
4424  *   VTEP object, contains rule database and ifouter index.
4425  * @param[in] dev_flow
4426  *   Flow object, contains the tunnel parameters (for encap only).
4427  * @param[in] enable
4428  *   Toggle between add and remove.
4429  * @param[out] error
4430  *   Perform verbose error reporting if not NULL.
4431  *
4432  * @return
4433  *   0 on success, a negative errno value otherwise and rte_errno is set.
4434  */
4435 static int
4436 flow_tcf_encap_local(struct mlx5_flow_tcf_context *tcf,
4437                      struct tcf_vtep *vtep,
4438                      struct mlx5_flow *dev_flow,
4439                      bool enable,
4440                      struct rte_flow_error *error)
4441 {
4442         const struct flow_tcf_vxlan_encap *encap = dev_flow->tcf.vxlan_encap;
4443         struct tcf_local_rule *rule;
4444         bool found = false;
4445         int ret;
4446
4447         assert(encap);
4448         assert(encap->hdr.type == FLOW_TCF_TUNACT_VXLAN_ENCAP);
4449         if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4450                 assert(encap->mask & FLOW_TCF_ENCAP_IPV4_DST);
4451                 LIST_FOREACH(rule, &vtep->local, next) {
4452                         if (rule->mask & FLOW_TCF_ENCAP_IPV4_SRC &&
4453                             encap->ipv4.src == rule->ipv4.src &&
4454                             encap->ipv4.dst == rule->ipv4.dst) {
4455                                 found = true;
4456                                 break;
4457                         }
4458                 }
4459         } else {
4460                 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4461                 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4462                 LIST_FOREACH(rule, &vtep->local, next) {
4463                         if (rule->mask & FLOW_TCF_ENCAP_IPV6_SRC &&
4464                             !memcmp(&encap->ipv6.src, &rule->ipv6.src,
4465                                             sizeof(encap->ipv6.src)) &&
4466                             !memcmp(&encap->ipv6.dst, &rule->ipv6.dst,
4467                                             sizeof(encap->ipv6.dst))) {
4468                                 found = true;
4469                                 break;
4470                         }
4471                 }
4472         }
4473         if (found) {
4474                 if (enable) {
4475                         rule->refcnt++;
4476                         return 0;
4477                 }
4478                 if (!rule->refcnt || !--rule->refcnt) {
4479                         LIST_REMOVE(rule, next);
4480                         return flow_tcf_rule_local(tcf, encap,
4481                                         vtep->ifouter, false, error);
4482                 }
4483                 return 0;
4484         }
4485         if (!enable) {
4486                 DRV_LOG(WARNING, "disabling not existing local rule");
4487                 rte_flow_error_set(error, ENOENT,
4488                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4489                                    "disabling not existing local rule");
4490                 return -ENOENT;
4491         }
4492         rule = rte_zmalloc(__func__, sizeof(struct tcf_local_rule),
4493                                 alignof(struct tcf_local_rule));
4494         if (!rule) {
4495                 rte_flow_error_set(error, ENOMEM,
4496                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4497                                    "unable to allocate memory for local rule");
4498                 return -rte_errno;
4499         }
4500         *rule = (struct tcf_local_rule){.refcnt = 0,
4501                                         .mask = 0,
4502                                         };
4503         if (encap->mask & FLOW_TCF_ENCAP_IPV4_SRC) {
4504                 rule->mask = FLOW_TCF_ENCAP_IPV4_SRC
4505                            | FLOW_TCF_ENCAP_IPV4_DST;
4506                 rule->ipv4.src = encap->ipv4.src;
4507                 rule->ipv4.dst = encap->ipv4.dst;
4508         } else {
4509                 rule->mask = FLOW_TCF_ENCAP_IPV6_SRC
4510                            | FLOW_TCF_ENCAP_IPV6_DST;
4511                 memcpy(&rule->ipv6.src, &encap->ipv6.src, IPV6_ADDR_LEN);
4512                 memcpy(&rule->ipv6.dst, &encap->ipv6.dst, IPV6_ADDR_LEN);
4513         }
4514         ret = flow_tcf_rule_local(tcf, encap, vtep->ifouter, true, error);
4515         if (ret) {
4516                 rte_free(rule);
4517                 return ret;
4518         }
4519         rule->refcnt++;
4520         LIST_INSERT_HEAD(&vtep->local, rule, next);
4521         return 0;
4522 }
4523
4524 /**
4525  * Manage the destination MAC/IP addresses neigh database, kernel uses
4526  * this one to determine the destination MAC address within encapsulation
4527  * header. Adds or removes the entries using the Netlink command like this:
4528  *   ip neigh add dev <ifouter> lladdr <dst_mac> to <dst_ip> nud permanent
4529  *
4530  * @param[in] tcf
4531  *   Libmnl socket context object.
4532  * @param[in] vtep
4533  *   VTEP object, contains rule database and ifouter index.
4534  * @param[in] dev_flow
4535  *   Flow object, contains the tunnel parameters (for encap only).
4536  * @param[in] enable
4537  *   Toggle between add and remove.
4538  * @param[out] error
4539  *   Perform verbose error reporting if not NULL.
4540  *
4541  * @return
4542  *   0 on success, a negative errno value otherwise and rte_errno is set.
4543  */
4544 static int
4545 flow_tcf_encap_neigh(struct mlx5_flow_tcf_context *tcf,
4546                      struct tcf_vtep *vtep,
4547                      struct mlx5_flow *dev_flow,
4548                      bool enable,
4549                      struct rte_flow_error *error)
4550 {
4551         const struct flow_tcf_vxlan_encap *encap = dev_flow->tcf.vxlan_encap;
4552         struct tcf_neigh_rule *rule;
4553         bool found = false;
4554         int ret;
4555
4556         assert(encap);
4557         assert(encap->hdr.type == FLOW_TCF_TUNACT_VXLAN_ENCAP);
4558         if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4559                 assert(encap->mask & FLOW_TCF_ENCAP_IPV4_SRC);
4560                 LIST_FOREACH(rule, &vtep->neigh, next) {
4561                         if (rule->mask & FLOW_TCF_ENCAP_IPV4_DST &&
4562                             encap->ipv4.dst == rule->ipv4.dst) {
4563                                 found = true;
4564                                 break;
4565                         }
4566                 }
4567         } else {
4568                 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_SRC);
4569                 assert(encap->mask & FLOW_TCF_ENCAP_IPV6_DST);
4570                 LIST_FOREACH(rule, &vtep->neigh, next) {
4571                         if (rule->mask & FLOW_TCF_ENCAP_IPV6_DST &&
4572                             !memcmp(&encap->ipv6.dst, &rule->ipv6.dst,
4573                                                 sizeof(encap->ipv6.dst))) {
4574                                 found = true;
4575                                 break;
4576                         }
4577                 }
4578         }
4579         if (found) {
4580                 if (memcmp(&encap->eth.dst, &rule->eth,
4581                            sizeof(encap->eth.dst))) {
4582                         DRV_LOG(WARNING, "Destination MAC differs"
4583                                          " in neigh rule");
4584                         rte_flow_error_set(error, EEXIST,
4585                                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
4586                                            NULL, "Different MAC address"
4587                                            " neigh rule for the same"
4588                                            " destination IP");
4589                                         return -EEXIST;
4590                 }
4591                 if (enable) {
4592                         rule->refcnt++;
4593                         return 0;
4594                 }
4595                 if (!rule->refcnt || !--rule->refcnt) {
4596                         LIST_REMOVE(rule, next);
4597                         return flow_tcf_rule_neigh(tcf, encap,
4598                                                    vtep->ifouter,
4599                                                    false, error);
4600                 }
4601                 return 0;
4602         }
4603         if (!enable) {
4604                 DRV_LOG(WARNING, "Disabling not existing neigh rule");
4605                 rte_flow_error_set(error, ENOENT,
4606                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4607                                    "unable to allocate memory for neigh rule");
4608                 return -ENOENT;
4609         }
4610         rule = rte_zmalloc(__func__, sizeof(struct tcf_neigh_rule),
4611                                 alignof(struct tcf_neigh_rule));
4612         if (!rule) {
4613                 rte_flow_error_set(error, ENOMEM,
4614                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4615                                    "unable to allocate memory for neigh rule");
4616                 return -rte_errno;
4617         }
4618         *rule = (struct tcf_neigh_rule){.refcnt = 0,
4619                                         .mask = 0,
4620                                         };
4621         if (encap->mask & FLOW_TCF_ENCAP_IPV4_DST) {
4622                 rule->mask = FLOW_TCF_ENCAP_IPV4_DST;
4623                 rule->ipv4.dst = encap->ipv4.dst;
4624         } else {
4625                 rule->mask = FLOW_TCF_ENCAP_IPV6_DST;
4626                 memcpy(&rule->ipv6.dst, &encap->ipv6.dst, IPV6_ADDR_LEN);
4627         }
4628         memcpy(&rule->eth, &encap->eth.dst, sizeof(rule->eth));
4629         ret = flow_tcf_rule_neigh(tcf, encap, vtep->ifouter, true, error);
4630         if (ret) {
4631                 rte_free(rule);
4632                 return ret;
4633         }
4634         rule->refcnt++;
4635         LIST_INSERT_HEAD(&vtep->neigh, rule, next);
4636         return 0;
4637 }
4638
4639 /* VTEP device list is shared between PMD port instances. */
4640 static LIST_HEAD(, tcf_vtep) vtep_list_vxlan = LIST_HEAD_INITIALIZER();
4641 static pthread_mutex_t vtep_list_mutex = PTHREAD_MUTEX_INITIALIZER;
4642
4643 /**
4644  * Deletes VTEP network device.
4645  *
4646  * @param[in] tcf
4647  *   Context object initialized by mlx5_flow_tcf_context_create().
4648  * @param[in] vtep
4649  *   Object represinting the network device to delete. Memory
4650  *   allocated for this object is freed by routine.
4651  */
4652 static void
4653 flow_tcf_vtep_delete(struct mlx5_flow_tcf_context *tcf,
4654                      struct tcf_vtep *vtep)
4655 {
4656         struct nlmsghdr *nlh;
4657         struct ifinfomsg *ifm;
4658         alignas(struct nlmsghdr)
4659         uint8_t buf[mnl_nlmsg_size(MNL_ALIGN(sizeof(*ifm))) +
4660                     MNL_BUF_EXTRA_SPACE];
4661         int ret;
4662
4663         assert(!vtep->refcnt);
4664         /* Delete only ifaces those we actually created. */
4665         if (vtep->created && vtep->ifindex) {
4666                 DRV_LOG(INFO, "VTEP delete (%d)", vtep->ifindex);
4667                 nlh = mnl_nlmsg_put_header(buf);
4668                 nlh->nlmsg_type = RTM_DELLINK;
4669                 nlh->nlmsg_flags = NLM_F_REQUEST;
4670                 ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4671                 ifm->ifi_family = AF_UNSPEC;
4672                 ifm->ifi_index = vtep->ifindex;
4673                 assert(sizeof(buf) >= nlh->nlmsg_len);
4674                 ret = flow_tcf_nl_ack(tcf, nlh, NULL, NULL);
4675                 if (ret)
4676                         DRV_LOG(WARNING, "netlink: error deleting vxlan"
4677                                          " encap/decap ifindex %u",
4678                                          ifm->ifi_index);
4679         }
4680         rte_free(vtep);
4681 }
4682
4683 /**
4684  * Creates VTEP network device.
4685  *
4686  * @param[in] tcf
4687  *   Context object initialized by mlx5_flow_tcf_context_create().
4688  * @param[in] ifouter
4689  *   Outer interface to attach new-created VXLAN device
4690  *   If zero the VXLAN device will not be attached to any device.
4691  *   These VTEPs are used for decapsulation and can be precreated
4692  *   and shared between processes.
4693  * @param[in] port
4694  *   UDP port of created VTEP device.
4695  * @param[out] error
4696  *   Perform verbose error reporting if not NULL.
4697  *
4698  * @return
4699  * Pointer to created device structure on success,
4700  * NULL otherwise and rte_errno is set.
4701  */
4702 #ifdef HAVE_IFLA_VXLAN_COLLECT_METADATA
4703 static struct tcf_vtep*
4704 flow_tcf_vtep_create(struct mlx5_flow_tcf_context *tcf,
4705                      unsigned int ifouter,
4706                      uint16_t port, struct rte_flow_error *error)
4707 {
4708         struct tcf_vtep *vtep;
4709         struct nlmsghdr *nlh;
4710         struct ifinfomsg *ifm;
4711         char name[sizeof(MLX5_VXLAN_DEVICE_PFX) + 24];
4712         alignas(struct nlmsghdr)
4713         uint8_t buf[mnl_nlmsg_size(sizeof(*ifm)) +
4714                     SZ_NLATTR_DATA_OF(sizeof(name)) +
4715                     SZ_NLATTR_NEST * 2 +
4716                     SZ_NLATTR_STRZ_OF("vxlan") +
4717                     SZ_NLATTR_DATA_OF(sizeof(uint32_t)) +
4718                     SZ_NLATTR_DATA_OF(sizeof(uint16_t)) +
4719                     SZ_NLATTR_DATA_OF(sizeof(uint8_t)) * 3 +
4720                     MNL_BUF_EXTRA_SPACE];
4721         struct nlattr *na_info;
4722         struct nlattr *na_vxlan;
4723         rte_be16_t vxlan_port = rte_cpu_to_be_16(port);
4724         int ret;
4725
4726         vtep = rte_zmalloc(__func__, sizeof(*vtep), alignof(struct tcf_vtep));
4727         if (!vtep) {
4728                 rte_flow_error_set(error, ENOMEM,
4729                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4730                                    "unable to allocate memory for VTEP");
4731                 return NULL;
4732         }
4733         *vtep = (struct tcf_vtep){
4734                         .port = port,
4735                         .local = LIST_HEAD_INITIALIZER(),
4736                         .neigh = LIST_HEAD_INITIALIZER(),
4737         };
4738         memset(buf, 0, sizeof(buf));
4739         nlh = mnl_nlmsg_put_header(buf);
4740         nlh->nlmsg_type = RTM_NEWLINK;
4741         nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE  | NLM_F_EXCL;
4742         ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4743         ifm->ifi_family = AF_UNSPEC;
4744         ifm->ifi_type = 0;
4745         ifm->ifi_index = 0;
4746         ifm->ifi_flags = IFF_UP;
4747         ifm->ifi_change = 0xffffffff;
4748         snprintf(name, sizeof(name), "%s%u", MLX5_VXLAN_DEVICE_PFX, port);
4749         mnl_attr_put_strz(nlh, IFLA_IFNAME, name);
4750         na_info = mnl_attr_nest_start(nlh, IFLA_LINKINFO);
4751         assert(na_info);
4752         mnl_attr_put_strz(nlh, IFLA_INFO_KIND, "vxlan");
4753         na_vxlan = mnl_attr_nest_start(nlh, IFLA_INFO_DATA);
4754         if (ifouter)
4755                 mnl_attr_put_u32(nlh, IFLA_VXLAN_LINK, ifouter);
4756         assert(na_vxlan);
4757         mnl_attr_put_u8(nlh, IFLA_VXLAN_COLLECT_METADATA, 1);
4758         mnl_attr_put_u8(nlh, IFLA_VXLAN_UDP_ZERO_CSUM6_RX, 1);
4759         mnl_attr_put_u8(nlh, IFLA_VXLAN_LEARNING, 0);
4760         mnl_attr_put_u16(nlh, IFLA_VXLAN_PORT, vxlan_port);
4761         mnl_attr_nest_end(nlh, na_vxlan);
4762         mnl_attr_nest_end(nlh, na_info);
4763         assert(sizeof(buf) >= nlh->nlmsg_len);
4764         ret = flow_tcf_nl_ack(tcf, nlh, NULL, NULL);
4765         if (ret) {
4766                 DRV_LOG(WARNING,
4767                         "netlink: VTEP %s create failure (%d)",
4768                         name, rte_errno);
4769                 if (rte_errno != EEXIST || ifouter)
4770                         /*
4771                          * Some unhandled error occurred or device is
4772                          * for encapsulation and cannot be shared.
4773                          */
4774                         goto error;
4775         } else {
4776                 /*
4777                  * Mark device we actually created.
4778                  * We should explicitly delete
4779                  * when we do not need it anymore.
4780                  */
4781                 vtep->created = 1;
4782         }
4783         /* Try to get ifindex of created of pre-existing device. */
4784         ret = if_nametoindex(name);
4785         if (!ret) {
4786                 DRV_LOG(WARNING,
4787                         "VTEP %s failed to get index (%d)", name, errno);
4788                 rte_flow_error_set
4789                         (error, -errno,
4790                          RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4791                          "netlink: failed to retrieve VTEP ifindex");
4792                 goto error;
4793         }
4794         vtep->ifindex = ret;
4795         vtep->ifouter = ifouter;
4796         memset(buf, 0, sizeof(buf));
4797         nlh = mnl_nlmsg_put_header(buf);
4798         nlh->nlmsg_type = RTM_NEWLINK;
4799         nlh->nlmsg_flags = NLM_F_REQUEST;
4800         ifm = mnl_nlmsg_put_extra_header(nlh, sizeof(*ifm));
4801         ifm->ifi_family = AF_UNSPEC;
4802         ifm->ifi_type = 0;
4803         ifm->ifi_index = vtep->ifindex;
4804         ifm->ifi_flags = IFF_UP;
4805         ifm->ifi_change = IFF_UP;
4806         ret = flow_tcf_nl_ack(tcf, nlh, NULL, NULL);
4807         if (ret) {
4808                 rte_flow_error_set(error, -errno,
4809                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4810                                    "netlink: failed to set VTEP link up");
4811                 DRV_LOG(WARNING, "netlink: VTEP %s set link up failure (%d)",
4812                         name, rte_errno);
4813                 goto clean;
4814         }
4815         ret = mlx5_flow_tcf_init(tcf, vtep->ifindex, error);
4816         if (ret) {
4817                 DRV_LOG(WARNING, "VTEP %s init failure (%d)", name, rte_errno);
4818                 goto clean;
4819         }
4820         DRV_LOG(INFO, "VTEP create (%d, %d)", vtep->port, vtep->ifindex);
4821         vtep->refcnt = 1;
4822         return vtep;
4823 clean:
4824         flow_tcf_vtep_delete(tcf, vtep);
4825         return NULL;
4826 error:
4827         rte_free(vtep);
4828         return NULL;
4829 }
4830 #else
4831 static struct tcf_vtep*
4832 flow_tcf_vtep_create(struct mlx5_flow_tcf_context *tcf __rte_unused,
4833                      unsigned int ifouter __rte_unused,
4834                      uint16_t port __rte_unused,
4835                      struct rte_flow_error *error)
4836 {
4837         rte_flow_error_set(error, ENOTSUP,
4838                            RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4839                            "netlink: failed to create VTEP, "
4840                            "vxlan metadata are not supported by kernel");
4841         return NULL;
4842 }
4843 #endif /* HAVE_IFLA_VXLAN_COLLECT_METADATA */
4844
4845 /**
4846  * Acquire target interface index for VXLAN tunneling decapsulation.
4847  * In order to share the UDP port within the other interfaces the
4848  * VXLAN device created as not attached to any interface (if created).
4849  *
4850  * @param[in] tcf
4851  *   Context object initialized by mlx5_flow_tcf_context_create().
4852  * @param[in] dev_flow
4853  *   Flow tcf object with tunnel structure pointer set.
4854  * @param[out] error
4855  *   Perform verbose error reporting if not NULL.
4856  * @return
4857  *   Interface descriptor pointer on success,
4858  *   NULL otherwise and rte_errno is set.
4859  */
4860 static struct tcf_vtep*
4861 flow_tcf_decap_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
4862                             struct mlx5_flow *dev_flow,
4863                             struct rte_flow_error *error)
4864 {
4865         struct tcf_vtep *vtep;
4866         uint16_t port = dev_flow->tcf.vxlan_decap->udp_port;
4867
4868         LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
4869                 if (vtep->port == port)
4870                         break;
4871         }
4872         if (vtep && vtep->ifouter) {
4873                 rte_flow_error_set(error, -errno,
4874                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
4875                                    "Failed to create decap VTEP with specified"
4876                                    " UDP port, atatched device exists");
4877                 return NULL;
4878         }
4879         if (vtep) {
4880                 /* Device exists, just increment the reference counter. */
4881                 vtep->refcnt++;
4882                 assert(vtep->ifindex);
4883                 return vtep;
4884         }
4885         /* No decapsulation device exists, try to create the new one. */
4886         vtep = flow_tcf_vtep_create(tcf, 0, port, error);
4887         if (vtep)
4888                 LIST_INSERT_HEAD(&vtep_list_vxlan, vtep, next);
4889         return vtep;
4890 }
4891
4892 /**
4893  * Aqcuire target interface index for VXLAN tunneling encapsulation.
4894  *
4895  * @param[in] tcf
4896  *   Context object initialized by mlx5_flow_tcf_context_create().
4897  * @param[in] ifouter
4898  *   Network interface index to attach VXLAN encap device to.
4899  * @param[in] dev_flow
4900  *   Flow tcf object with tunnel structure pointer set.
4901  * @param[out] error
4902  *   Perform verbose error reporting if not NULL.
4903  * @return
4904  *   Interface descriptor pointer on success,
4905  *   NULL otherwise and rte_errno is set.
4906  */
4907 static struct tcf_vtep*
4908 flow_tcf_encap_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
4909                             unsigned int ifouter,
4910                             struct mlx5_flow *dev_flow __rte_unused,
4911                             struct rte_flow_error *error)
4912 {
4913         static uint16_t encap_port = MLX5_VXLAN_PORT_MIN - 1;
4914         struct tcf_vtep *vtep;
4915         int ret;
4916
4917         assert(ifouter);
4918         /* Look whether the attached VTEP for encap is created. */
4919         LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
4920                 if (vtep->ifouter == ifouter)
4921                         break;
4922         }
4923         if (vtep) {
4924                 /* VTEP already exists, just increment the reference. */
4925                 vtep->refcnt++;
4926         } else {
4927                 uint16_t pcnt;
4928
4929                 /* Not found, we should create the new attached VTEP. */
4930                 flow_tcf_encap_iface_cleanup(tcf, ifouter);
4931                 flow_tcf_encap_local_cleanup(tcf, ifouter);
4932                 flow_tcf_encap_neigh_cleanup(tcf, ifouter);
4933                 for (pcnt = 0; pcnt <= (MLX5_VXLAN_PORT_MAX
4934                                      - MLX5_VXLAN_PORT_MIN); pcnt++) {
4935                         encap_port++;
4936                         /* Wraparound the UDP port index. */
4937                         if (encap_port < MLX5_VXLAN_PORT_MIN ||
4938                             encap_port > MLX5_VXLAN_PORT_MAX)
4939                                 encap_port = MLX5_VXLAN_PORT_MIN;
4940                         /* Check whether UDP port is in already in use. */
4941                         LIST_FOREACH(vtep, &vtep_list_vxlan, next) {
4942                                 if (vtep->port == encap_port)
4943                                         break;
4944                         }
4945                         if (vtep) {
4946                                 /* Port is in use, try the next one. */
4947                                 vtep = NULL;
4948                                 continue;
4949                         }
4950                         vtep = flow_tcf_vtep_create(tcf, ifouter,
4951                                                     encap_port, error);
4952                         if (vtep) {
4953                                 LIST_INSERT_HEAD(&vtep_list_vxlan, vtep, next);
4954                                 break;
4955                         }
4956                         if (rte_errno != EEXIST)
4957                                 break;
4958                 }
4959                 if (!vtep)
4960                         return NULL;
4961         }
4962         assert(vtep->ifouter == ifouter);
4963         assert(vtep->ifindex);
4964         /* Create local ipaddr with peer to specify the outer IPs. */
4965         ret = flow_tcf_encap_local(tcf, vtep, dev_flow, true, error);
4966         if (!ret) {
4967                 /* Create neigh rule to specify outer destination MAC. */
4968                 ret = flow_tcf_encap_neigh(tcf, vtep, dev_flow, true, error);
4969                 if (ret)
4970                         flow_tcf_encap_local(tcf, vtep,
4971                                              dev_flow, false, error);
4972         }
4973         if (ret) {
4974                 if (--vtep->refcnt == 0)
4975                         flow_tcf_vtep_delete(tcf, vtep);
4976                 return NULL;
4977         }
4978         return vtep;
4979 }
4980
4981 /**
4982  * Acquires target interface index for tunneling of any type.
4983  * Creates the new VTEP if needed.
4984  *
4985  * @param[in] tcf
4986  *   Context object initialized by mlx5_flow_tcf_context_create().
4987  * @param[in] ifouter
4988  *   Network interface index to attach VXLAN encap device to.
4989  * @param[in] dev_flow
4990  *   Flow tcf object with tunnel structure pointer set.
4991  * @param[out] error
4992  *   Perform verbose error reporting if not NULL.
4993  * @return
4994  *   Interface descriptor pointer on success,
4995  *   NULL otherwise and rte_errno is set.
4996  */
4997 static struct tcf_vtep*
4998 flow_tcf_vtep_acquire(struct mlx5_flow_tcf_context *tcf,
4999                       unsigned int ifouter,
5000                       struct mlx5_flow *dev_flow,
5001                       struct rte_flow_error *error)
5002 {
5003         struct tcf_vtep *vtep = NULL;
5004
5005         assert(dev_flow->tcf.tunnel);
5006         pthread_mutex_lock(&vtep_list_mutex);
5007         switch (dev_flow->tcf.tunnel->type) {
5008         case FLOW_TCF_TUNACT_VXLAN_ENCAP:
5009                 vtep = flow_tcf_encap_vtep_acquire(tcf, ifouter,
5010                                                   dev_flow, error);
5011                 break;
5012         case FLOW_TCF_TUNACT_VXLAN_DECAP:
5013                 vtep = flow_tcf_decap_vtep_acquire(tcf, dev_flow, error);
5014                 break;
5015         default:
5016                 rte_flow_error_set(error, ENOTSUP,
5017                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5018                                    "unsupported tunnel type");
5019                 break;
5020         }
5021         pthread_mutex_unlock(&vtep_list_mutex);
5022         return vtep;
5023 }
5024
5025 /**
5026  * Release tunneling interface by ifindex. Decrements reference
5027  * counter and actually removes the device if counter is zero.
5028  *
5029  * @param[in] tcf
5030  *   Context object initialized by mlx5_flow_tcf_context_create().
5031  * @param[in] vtep
5032  *   VTEP device descriptor structure.
5033  * @param[in] dev_flow
5034  *   Flow tcf object with tunnel structure pointer set.
5035  */
5036 static void
5037 flow_tcf_vtep_release(struct mlx5_flow_tcf_context *tcf,
5038                       struct tcf_vtep *vtep,
5039                       struct mlx5_flow *dev_flow)
5040 {
5041         assert(dev_flow->tcf.tunnel);
5042         pthread_mutex_lock(&vtep_list_mutex);
5043         switch (dev_flow->tcf.tunnel->type) {
5044         case FLOW_TCF_TUNACT_VXLAN_DECAP:
5045                 break;
5046         case FLOW_TCF_TUNACT_VXLAN_ENCAP:
5047                 /* Remove the encap ancillary rules first. */
5048                 flow_tcf_encap_neigh(tcf, vtep, dev_flow, false, NULL);
5049                 flow_tcf_encap_local(tcf, vtep, dev_flow, false, NULL);
5050                 break;
5051         default:
5052                 assert(false);
5053                 DRV_LOG(WARNING, "Unsupported tunnel type");
5054                 break;
5055         }
5056         assert(vtep->refcnt);
5057         if (--vtep->refcnt == 0) {
5058                 LIST_REMOVE(vtep, next);
5059                 flow_tcf_vtep_delete(tcf, vtep);
5060         }
5061         pthread_mutex_unlock(&vtep_list_mutex);
5062 }
5063
5064
5065 /**
5066  * Apply flow to E-Switch by sending Netlink message.
5067  *
5068  * @param[in] dev
5069  *   Pointer to Ethernet device.
5070  * @param[in, out] flow
5071  *   Pointer to the sub flow.
5072  * @param[out] error
5073  *   Pointer to the error structure.
5074  *
5075  * @return
5076  *   0 on success, a negative errno value otherwise and rte_ernno is set.
5077  */
5078 static int
5079 flow_tcf_apply(struct rte_eth_dev *dev, struct rte_flow *flow,
5080                struct rte_flow_error *error)
5081 {
5082         struct priv *priv = dev->data->dev_private;
5083         struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5084         struct mlx5_flow *dev_flow;
5085         struct nlmsghdr *nlh;
5086
5087         dev_flow = LIST_FIRST(&flow->dev_flows);
5088         /* E-Switch flow can't be expanded. */
5089         assert(!LIST_NEXT(dev_flow, next));
5090         if (dev_flow->tcf.applied)
5091                 return 0;
5092         nlh = dev_flow->tcf.nlh;
5093         nlh->nlmsg_type = RTM_NEWTFILTER;
5094         nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
5095         if (dev_flow->tcf.tunnel) {
5096                 /*
5097                  * Replace the interface index, target for
5098                  * encapsulation, source for decapsulation.
5099                  */
5100                 assert(!dev_flow->tcf.tunnel->vtep);
5101                 assert(dev_flow->tcf.tunnel->ifindex_ptr);
5102                 /* Acquire actual VTEP device when rule is being applied. */
5103                 dev_flow->tcf.tunnel->vtep =
5104                         flow_tcf_vtep_acquire(ctx,
5105                                         dev_flow->tcf.tunnel->ifindex_org,
5106                                         dev_flow, error);
5107                 if (!dev_flow->tcf.tunnel->vtep)
5108                         return -rte_errno;
5109                 DRV_LOG(INFO, "Replace ifindex: %d->%d",
5110                                 dev_flow->tcf.tunnel->vtep->ifindex,
5111                                 dev_flow->tcf.tunnel->ifindex_org);
5112                 *dev_flow->tcf.tunnel->ifindex_ptr =
5113                         dev_flow->tcf.tunnel->vtep->ifindex;
5114         }
5115         if (!flow_tcf_nl_ack(ctx, nlh, NULL, NULL)) {
5116                 dev_flow->tcf.applied = 1;
5117                 return 0;
5118         }
5119         if (dev_flow->tcf.tunnel) {
5120                 /* Rollback the VTEP configuration if rule apply failed. */
5121                 assert(dev_flow->tcf.tunnel->vtep);
5122                 flow_tcf_vtep_release(ctx, dev_flow->tcf.tunnel->vtep,
5123                                       dev_flow);
5124                 dev_flow->tcf.tunnel->vtep = NULL;
5125         }
5126         return rte_flow_error_set(error, rte_errno,
5127                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5128                                   "netlink: failed to create TC flow rule");
5129 }
5130
5131 /**
5132  * Remove flow from E-Switch by sending Netlink message.
5133  *
5134  * @param[in] dev
5135  *   Pointer to Ethernet device.
5136  * @param[in, out] flow
5137  *   Pointer to the sub flow.
5138  */
5139 static void
5140 flow_tcf_remove(struct rte_eth_dev *dev, struct rte_flow *flow)
5141 {
5142         struct priv *priv = dev->data->dev_private;
5143         struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5144         struct mlx5_flow *dev_flow;
5145         struct nlmsghdr *nlh;
5146
5147         if (!flow)
5148                 return;
5149         dev_flow = LIST_FIRST(&flow->dev_flows);
5150         if (!dev_flow)
5151                 return;
5152         /* E-Switch flow can't be expanded. */
5153         assert(!LIST_NEXT(dev_flow, next));
5154         if (dev_flow->tcf.applied) {
5155                 nlh = dev_flow->tcf.nlh;
5156                 nlh->nlmsg_type = RTM_DELTFILTER;
5157                 nlh->nlmsg_flags = NLM_F_REQUEST;
5158                 flow_tcf_nl_ack(ctx, nlh, NULL, NULL);
5159                 if (dev_flow->tcf.tunnel) {
5160                         assert(dev_flow->tcf.tunnel->vtep);
5161                         flow_tcf_vtep_release(ctx,
5162                                 dev_flow->tcf.tunnel->vtep,
5163                                 dev_flow);
5164                         dev_flow->tcf.tunnel->vtep = NULL;
5165                 }
5166                 dev_flow->tcf.applied = 0;
5167         }
5168 }
5169
5170 /**
5171  * Remove flow from E-Switch and release resources of the device flow.
5172  *
5173  * @param[in] dev
5174  *   Pointer to Ethernet device.
5175  * @param[in, out] flow
5176  *   Pointer to the sub flow.
5177  */
5178 static void
5179 flow_tcf_destroy(struct rte_eth_dev *dev, struct rte_flow *flow)
5180 {
5181         struct mlx5_flow *dev_flow;
5182
5183         if (!flow)
5184                 return;
5185         flow_tcf_remove(dev, flow);
5186         if (flow->counter) {
5187                 if (--flow->counter->ref_cnt == 0) {
5188                         rte_free(flow->counter);
5189                         flow->counter = NULL;
5190                 }
5191         }
5192         dev_flow = LIST_FIRST(&flow->dev_flows);
5193         if (!dev_flow)
5194                 return;
5195         /* E-Switch flow can't be expanded. */
5196         assert(!LIST_NEXT(dev_flow, next));
5197         LIST_REMOVE(dev_flow, next);
5198         rte_free(dev_flow);
5199 }
5200
5201 /**
5202  * Helper routine for figuring the space size required for a parse buffer.
5203  *
5204  * @param array
5205  *   array of values to use.
5206  * @param idx
5207  *   Current location in array.
5208  * @param value
5209  *   Value to compare with.
5210  *
5211  * @return
5212  *   The maximum between the given value and the array value on index.
5213  */
5214 static uint16_t
5215 flow_tcf_arr_val_max(uint16_t array[], int idx, uint16_t value)
5216 {
5217         return idx < 0 ? (value) : RTE_MAX((array)[idx], value);
5218 }
5219
5220 /**
5221  * Parse rtnetlink message attributes filling the attribute table with the info
5222  * retrieved.
5223  *
5224  * @param tb
5225  *   Attribute table to be filled.
5226  * @param[out] max
5227  *   Maxinum entry in the attribute table.
5228  * @param rte
5229  *   The attributes section in the message to be parsed.
5230  * @param len
5231  *   The length of the attributes section in the message.
5232  */
5233 static void
5234 flow_tcf_nl_parse_rtattr(struct rtattr *tb[], int max,
5235                          struct rtattr *rta, int len)
5236 {
5237         unsigned short type;
5238         memset(tb, 0, sizeof(struct rtattr *) * (max + 1));
5239         while (RTA_OK(rta, len)) {
5240                 type = rta->rta_type;
5241                 if (type <= max && !tb[type])
5242                         tb[type] = rta;
5243                 rta = RTA_NEXT(rta, len);
5244         }
5245 }
5246
5247 /**
5248  * Extract flow counters from flower action.
5249  *
5250  * @param rta
5251  *   flower action stats properties in the Netlink message received.
5252  * @param rta_type
5253  *   The backward sequence of rta_types, as written in the attribute table,
5254  *   we need to traverse in order to get to the requested object.
5255  * @param idx
5256  *   Current location in rta_type table.
5257  * @param[out] data
5258  *   data holding the count statistics of the rte_flow retrieved from
5259  *   the message.
5260  *
5261  * @return
5262  *   0 if data was found and retrieved, -1 otherwise.
5263  */
5264 static int
5265 flow_tcf_nl_action_stats_parse_and_get(struct rtattr *rta,
5266                                        uint16_t rta_type[], int idx,
5267                                        struct gnet_stats_basic *data)
5268 {
5269         int tca_stats_max = flow_tcf_arr_val_max(rta_type, idx,
5270                                                  TCA_STATS_BASIC);
5271         struct rtattr *tbs[tca_stats_max + 1];
5272
5273         if (rta == NULL || idx < 0)
5274                 return -1;
5275         flow_tcf_nl_parse_rtattr(tbs, tca_stats_max,
5276                                  RTA_DATA(rta), RTA_PAYLOAD(rta));
5277         switch (rta_type[idx]) {
5278         case TCA_STATS_BASIC:
5279                 if (tbs[TCA_STATS_BASIC]) {
5280                         memcpy(data, RTA_DATA(tbs[TCA_STATS_BASIC]),
5281                                RTE_MIN(RTA_PAYLOAD(tbs[TCA_STATS_BASIC]),
5282                                sizeof(*data)));
5283                         return 0;
5284                 }
5285                 break;
5286         default:
5287                 break;
5288         }
5289         return -1;
5290 }
5291
5292 /**
5293  * Parse flower single action retrieving the requested action attribute,
5294  * if found.
5295  *
5296  * @param arg
5297  *   flower action properties in the Netlink message received.
5298  * @param rta_type
5299  *   The backward sequence of rta_types, as written in the attribute table,
5300  *   we need to traverse in order to get to the requested object.
5301  * @param idx
5302  *   Current location in rta_type table.
5303  * @param[out] data
5304  *   Count statistics retrieved from the message query.
5305  *
5306  * @return
5307  *   0 if data was found and retrieved, -1 otherwise.
5308  */
5309 static int
5310 flow_tcf_nl_parse_one_action_and_get(struct rtattr *arg,
5311                                      uint16_t rta_type[], int idx, void *data)
5312 {
5313         int tca_act_max = flow_tcf_arr_val_max(rta_type, idx, TCA_ACT_STATS);
5314         struct rtattr *tb[tca_act_max + 1];
5315
5316         if (arg == NULL || idx < 0)
5317                 return -1;
5318         flow_tcf_nl_parse_rtattr(tb, tca_act_max,
5319                                  RTA_DATA(arg), RTA_PAYLOAD(arg));
5320         if (tb[TCA_ACT_KIND] == NULL)
5321                 return -1;
5322         switch (rta_type[idx]) {
5323         case TCA_ACT_STATS:
5324                 if (tb[TCA_ACT_STATS])
5325                         return flow_tcf_nl_action_stats_parse_and_get
5326                                         (tb[TCA_ACT_STATS],
5327                                          rta_type, --idx,
5328                                          (struct gnet_stats_basic *)data);
5329                 break;
5330         default:
5331                 break;
5332         }
5333         return -1;
5334 }
5335
5336 /**
5337  * Parse flower action section in the message retrieving the requested
5338  * attribute from the first action that provides it.
5339  *
5340  * @param opt
5341  *   flower section in the Netlink message received.
5342  * @param rta_type
5343  *   The backward sequence of rta_types, as written in the attribute table,
5344  *   we need to traverse in order to get to the requested object.
5345  * @param idx
5346  *   Current location in rta_type table.
5347  * @param[out] data
5348  *   data retrieved from the message query.
5349  *
5350  * @return
5351  *   0 if data was found and retrieved, -1 otherwise.
5352  */
5353 static int
5354 flow_tcf_nl_action_parse_and_get(struct rtattr *arg,
5355                                  uint16_t rta_type[], int idx, void *data)
5356 {
5357         struct rtattr *tb[TCA_ACT_MAX_PRIO + 1];
5358         int i;
5359
5360         if (arg == NULL || idx < 0)
5361                 return -1;
5362         flow_tcf_nl_parse_rtattr(tb, TCA_ACT_MAX_PRIO,
5363                                  RTA_DATA(arg), RTA_PAYLOAD(arg));
5364         switch (rta_type[idx]) {
5365         /*
5366          * flow counters are stored in the actions defined by the flow
5367          * and not in the flow itself, therefore we need to traverse the
5368          * flower chain of actions in search for them.
5369          *
5370          * Note that the index is not decremented here.
5371          */
5372         case TCA_ACT_STATS:
5373                 for (i = 0; i <= TCA_ACT_MAX_PRIO; i++) {
5374                         if (tb[i] &&
5375                         !flow_tcf_nl_parse_one_action_and_get(tb[i],
5376                                                               rta_type,
5377                                                               idx, data))
5378                                 return 0;
5379                 }
5380                 break;
5381         default:
5382                 break;
5383         }
5384         return -1;
5385 }
5386
5387 /**
5388  * Parse flower classifier options in the message, retrieving the requested
5389  * attribute if found.
5390  *
5391  * @param opt
5392  *   flower section in the Netlink message received.
5393  * @param rta_type
5394  *   The backward sequence of rta_types, as written in the attribute table,
5395  *   we need to traverse in order to get to the requested object.
5396  * @param idx
5397  *   Current location in rta_type table.
5398  * @param[out] data
5399  *   data retrieved from the message query.
5400  *
5401  * @return
5402  *   0 if data was found and retrieved, -1 otherwise.
5403  */
5404 static int
5405 flow_tcf_nl_opts_parse_and_get(struct rtattr *opt,
5406                                uint16_t rta_type[], int idx, void *data)
5407 {
5408         int tca_flower_max = flow_tcf_arr_val_max(rta_type, idx,
5409                                                   TCA_FLOWER_ACT);
5410         struct rtattr *tb[tca_flower_max + 1];
5411
5412         if (!opt || idx < 0)
5413                 return -1;
5414         flow_tcf_nl_parse_rtattr(tb, tca_flower_max,
5415                                  RTA_DATA(opt), RTA_PAYLOAD(opt));
5416         switch (rta_type[idx]) {
5417         case TCA_FLOWER_ACT:
5418                 if (tb[TCA_FLOWER_ACT])
5419                         return flow_tcf_nl_action_parse_and_get
5420                                                         (tb[TCA_FLOWER_ACT],
5421                                                          rta_type, --idx, data);
5422                 break;
5423         default:
5424                 break;
5425         }
5426         return -1;
5427 }
5428
5429 /**
5430  * Parse Netlink reply on filter query, retrieving the flow counters.
5431  *
5432  * @param nlh
5433  *   Message received from Netlink.
5434  * @param rta_type
5435  *   The backward sequence of rta_types, as written in the attribute table,
5436  *   we need to traverse in order to get to the requested object.
5437  * @param idx
5438  *   Current location in rta_type table.
5439  * @param[out] data
5440  *   data retrieved from the message query.
5441  *
5442  * @return
5443  *   0 if data was found and retrieved, -1 otherwise.
5444  */
5445 static int
5446 flow_tcf_nl_filter_parse_and_get(struct nlmsghdr *cnlh,
5447                                  uint16_t rta_type[], int idx, void *data)
5448 {
5449         struct nlmsghdr *nlh = cnlh;
5450         struct tcmsg *t = NLMSG_DATA(nlh);
5451         int len = nlh->nlmsg_len;
5452         int tca_max = flow_tcf_arr_val_max(rta_type, idx, TCA_OPTIONS);
5453         struct rtattr *tb[tca_max + 1];
5454
5455         if (idx < 0)
5456                 return -1;
5457         if (nlh->nlmsg_type != RTM_NEWTFILTER &&
5458             nlh->nlmsg_type != RTM_GETTFILTER &&
5459             nlh->nlmsg_type != RTM_DELTFILTER)
5460                 return -1;
5461         len -= NLMSG_LENGTH(sizeof(*t));
5462         if (len < 0)
5463                 return -1;
5464         flow_tcf_nl_parse_rtattr(tb, tca_max, TCA_RTA(t), len);
5465         /* Not a TC flower flow - bail out */
5466         if (!tb[TCA_KIND] ||
5467             strcmp(RTA_DATA(tb[TCA_KIND]), "flower"))
5468                 return -1;
5469         switch (rta_type[idx]) {
5470         case TCA_OPTIONS:
5471                 if (tb[TCA_OPTIONS])
5472                         return flow_tcf_nl_opts_parse_and_get(tb[TCA_OPTIONS],
5473                                                               rta_type,
5474                                                               --idx, data);
5475                 break;
5476         default:
5477                 break;
5478         }
5479         return -1;
5480 }
5481
5482 /**
5483  * A callback to parse Netlink reply on TC flower query.
5484  *
5485  * @param nlh
5486  *   Message received from Netlink.
5487  * @param[out] data
5488  *   Pointer to data area to be filled by the parsing routine.
5489  *   assumed to be a pinter to struct flow_tcf_stats_basic.
5490  *
5491  * @return
5492  *   MNL_CB_OK value.
5493  */
5494 static int
5495 flow_tcf_nl_message_get_stats_basic(const struct nlmsghdr *nlh, void *data)
5496 {
5497         /*
5498          * The backward sequence of rta_types to pass in order to get
5499          *  to the counters.
5500          */
5501         uint16_t rta_type[] = { TCA_STATS_BASIC, TCA_ACT_STATS,
5502                                 TCA_FLOWER_ACT, TCA_OPTIONS };
5503         struct flow_tcf_stats_basic *sb_data = data;
5504         union {
5505                 const struct nlmsghdr *c;
5506                 struct nlmsghdr *nc;
5507         } tnlh = { .c = nlh };
5508
5509         if (!flow_tcf_nl_filter_parse_and_get(tnlh.nc, rta_type,
5510                                               RTE_DIM(rta_type) - 1,
5511                                               (void *)&sb_data->counters))
5512                 sb_data->valid = true;
5513         return MNL_CB_OK;
5514 }
5515
5516 /**
5517  * Query a TC flower rule for its statistics via netlink.
5518  *
5519  * @param[in] dev
5520  *   Pointer to Ethernet device.
5521  * @param[in] flow
5522  *   Pointer to the sub flow.
5523  * @param[out] data
5524  *   data retrieved by the query.
5525  * @param[out] error
5526  *   Perform verbose error reporting if not NULL.
5527  *
5528  * @return
5529  *   0 on success, a negative errno value otherwise and rte_errno is set.
5530  */
5531 static int
5532 flow_tcf_query_count(struct rte_eth_dev *dev,
5533                           struct rte_flow *flow,
5534                           void *data,
5535                           struct rte_flow_error *error)
5536 {
5537         struct flow_tcf_stats_basic sb_data = { 0 };
5538         struct rte_flow_query_count *qc = data;
5539         struct priv *priv = dev->data->dev_private;
5540         struct mlx5_flow_tcf_context *ctx = priv->tcf_context;
5541         struct mnl_socket *nl = ctx->nl;
5542         struct mlx5_flow *dev_flow;
5543         struct nlmsghdr *nlh;
5544         uint32_t seq = priv->tcf_context->seq++;
5545         ssize_t ret;
5546         assert(qc);
5547
5548         dev_flow = LIST_FIRST(&flow->dev_flows);
5549         /* E-Switch flow can't be expanded. */
5550         assert(!LIST_NEXT(dev_flow, next));
5551         if (!dev_flow->flow->counter)
5552                 goto notsup_exit;
5553         nlh = dev_flow->tcf.nlh;
5554         nlh->nlmsg_type = RTM_GETTFILTER;
5555         nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_ECHO;
5556         nlh->nlmsg_seq = seq;
5557         if (mnl_socket_sendto(nl, nlh, nlh->nlmsg_len) == -1)
5558                 goto error_exit;
5559         do {
5560                 ret = mnl_socket_recvfrom(nl, ctx->buf, ctx->buf_size);
5561                 if (ret <= 0)
5562                         break;
5563                 ret = mnl_cb_run(ctx->buf, ret, seq,
5564                                  mnl_socket_get_portid(nl),
5565                                  flow_tcf_nl_message_get_stats_basic,
5566                                  (void *)&sb_data);
5567         } while (ret > 0);
5568         /* Return the delta from last reset. */
5569         if (sb_data.valid) {
5570                 /* Return the delta from last reset. */
5571                 qc->hits_set = 1;
5572                 qc->bytes_set = 1;
5573                 qc->hits = sb_data.counters.packets - flow->counter->hits;
5574                 qc->bytes = sb_data.counters.bytes - flow->counter->bytes;
5575                 if (qc->reset) {
5576                         flow->counter->hits = sb_data.counters.packets;
5577                         flow->counter->bytes = sb_data.counters.bytes;
5578                 }
5579                 return 0;
5580         }
5581         return rte_flow_error_set(error, EINVAL,
5582                                   RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5583                                   NULL,
5584                                   "flow does not have counter");
5585 error_exit:
5586         return rte_flow_error_set
5587                         (error, errno, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5588                          NULL, "netlink: failed to read flow rule counters");
5589 notsup_exit:
5590         return rte_flow_error_set
5591                         (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
5592                          NULL, "counters are not available.");
5593 }
5594
5595 /**
5596  * Query a flow.
5597  *
5598  * @see rte_flow_query()
5599  * @see rte_flow_ops
5600  */
5601 static int
5602 flow_tcf_query(struct rte_eth_dev *dev,
5603                struct rte_flow *flow,
5604                const struct rte_flow_action *actions,
5605                void *data,
5606                struct rte_flow_error *error)
5607 {
5608         int ret = -EINVAL;
5609
5610         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
5611                 switch (actions->type) {
5612                 case RTE_FLOW_ACTION_TYPE_VOID:
5613                         break;
5614                 case RTE_FLOW_ACTION_TYPE_COUNT:
5615                         ret = flow_tcf_query_count(dev, flow, data, error);
5616                         break;
5617                 default:
5618                         return rte_flow_error_set(error, ENOTSUP,
5619                                                   RTE_FLOW_ERROR_TYPE_ACTION,
5620                                                   actions,
5621                                                   "action not supported");
5622                 }
5623         }
5624         return ret;
5625 }
5626
5627 const struct mlx5_flow_driver_ops mlx5_flow_tcf_drv_ops = {
5628         .validate = flow_tcf_validate,
5629         .prepare = flow_tcf_prepare,
5630         .translate = flow_tcf_translate,
5631         .apply = flow_tcf_apply,
5632         .remove = flow_tcf_remove,
5633         .destroy = flow_tcf_destroy,
5634         .query = flow_tcf_query,
5635 };
5636
5637 /**
5638  * Create and configure a libmnl socket for Netlink flow rules.
5639  *
5640  * @return
5641  *   A valid libmnl socket object pointer on success, NULL otherwise and
5642  *   rte_errno is set.
5643  */
5644 static struct mnl_socket *
5645 flow_tcf_mnl_socket_create(void)
5646 {
5647         struct mnl_socket *nl = mnl_socket_open(NETLINK_ROUTE);
5648
5649         if (nl) {
5650                 mnl_socket_setsockopt(nl, NETLINK_CAP_ACK, &(int){ 1 },
5651                                       sizeof(int));
5652                 if (!mnl_socket_bind(nl, 0, MNL_SOCKET_AUTOPID))
5653                         return nl;
5654         }
5655         rte_errno = errno;
5656         if (nl)
5657                 mnl_socket_close(nl);
5658         return NULL;
5659 }
5660
5661 /**
5662  * Destroy a libmnl socket.
5663  *
5664  * @param nl
5665  *   Libmnl socket of the @p NETLINK_ROUTE kind.
5666  */
5667 static void
5668 flow_tcf_mnl_socket_destroy(struct mnl_socket *nl)
5669 {
5670         if (nl)
5671                 mnl_socket_close(nl);
5672 }
5673
5674 /**
5675  * Initialize ingress qdisc of a given network interface.
5676  *
5677  * @param ctx
5678  *   Pointer to tc-flower context to use.
5679  * @param ifindex
5680  *   Index of network interface to initialize.
5681  * @param[out] error
5682  *   Perform verbose error reporting if not NULL.
5683  *
5684  * @return
5685  *   0 on success, a negative errno value otherwise and rte_errno is set.
5686  */
5687 int
5688 mlx5_flow_tcf_init(struct mlx5_flow_tcf_context *ctx,
5689                    unsigned int ifindex, struct rte_flow_error *error)
5690 {
5691         struct nlmsghdr *nlh;
5692         struct tcmsg *tcm;
5693         alignas(struct nlmsghdr)
5694         uint8_t buf[mnl_nlmsg_size(sizeof(*tcm)) +
5695                     SZ_NLATTR_STRZ_OF("ingress") +
5696                     MNL_BUF_EXTRA_SPACE];
5697
5698         /* Destroy existing ingress qdisc and everything attached to it. */
5699         nlh = mnl_nlmsg_put_header(buf);
5700         nlh->nlmsg_type = RTM_DELQDISC;
5701         nlh->nlmsg_flags = NLM_F_REQUEST;
5702         tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
5703         tcm->tcm_family = AF_UNSPEC;
5704         tcm->tcm_ifindex = ifindex;
5705         tcm->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
5706         tcm->tcm_parent = TC_H_INGRESS;
5707         assert(sizeof(buf) >= nlh->nlmsg_len);
5708         /* Ignore errors when qdisc is already absent. */
5709         if (flow_tcf_nl_ack(ctx, nlh, NULL, NULL) &&
5710             rte_errno != EINVAL && rte_errno != ENOENT)
5711                 return rte_flow_error_set(error, rte_errno,
5712                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5713                                           "netlink: failed to remove ingress"
5714                                           " qdisc");
5715         /* Create fresh ingress qdisc. */
5716         nlh = mnl_nlmsg_put_header(buf);
5717         nlh->nlmsg_type = RTM_NEWQDISC;
5718         nlh->nlmsg_flags = NLM_F_REQUEST | NLM_F_CREATE | NLM_F_EXCL;
5719         tcm = mnl_nlmsg_put_extra_header(nlh, sizeof(*tcm));
5720         tcm->tcm_family = AF_UNSPEC;
5721         tcm->tcm_ifindex = ifindex;
5722         tcm->tcm_handle = TC_H_MAKE(TC_H_INGRESS, 0);
5723         tcm->tcm_parent = TC_H_INGRESS;
5724         mnl_attr_put_strz_check(nlh, sizeof(buf), TCA_KIND, "ingress");
5725         assert(sizeof(buf) >= nlh->nlmsg_len);
5726         if (flow_tcf_nl_ack(ctx, nlh, NULL, NULL))
5727                 return rte_flow_error_set(error, rte_errno,
5728                                           RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
5729                                           "netlink: failed to create ingress"
5730                                           " qdisc");
5731         return 0;
5732 }
5733
5734 /**
5735  * Create libmnl context for Netlink flow rules.
5736  *
5737  * @return
5738  *   A valid libmnl socket object pointer on success, NULL otherwise and
5739  *   rte_errno is set.
5740  */
5741 struct mlx5_flow_tcf_context *
5742 mlx5_flow_tcf_context_create(void)
5743 {
5744         struct mlx5_flow_tcf_context *ctx = rte_zmalloc(__func__,
5745                                                         sizeof(*ctx),
5746                                                         sizeof(uint32_t));
5747         if (!ctx)
5748                 goto error;
5749         ctx->nl = flow_tcf_mnl_socket_create();
5750         if (!ctx->nl)
5751                 goto error;
5752         ctx->buf_size = MNL_SOCKET_BUFFER_SIZE;
5753         ctx->buf = rte_zmalloc(__func__,
5754                                ctx->buf_size, sizeof(uint32_t));
5755         if (!ctx->buf)
5756                 goto error;
5757         ctx->seq = random();
5758         return ctx;
5759 error:
5760         mlx5_flow_tcf_context_destroy(ctx);
5761         return NULL;
5762 }
5763
5764 /**
5765  * Destroy a libmnl context.
5766  *
5767  * @param ctx
5768  *   Libmnl socket of the @p NETLINK_ROUTE kind.
5769  */
5770 void
5771 mlx5_flow_tcf_context_destroy(struct mlx5_flow_tcf_context *ctx)
5772 {
5773         if (!ctx)
5774                 return;
5775         flow_tcf_mnl_socket_destroy(ctx->nl);
5776         rte_free(ctx->buf);
5777         rte_free(ctx);
5778 }