drivers/net/sfc/sfc_flow.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  *
   3  * Copyright (c) 2017-2018 Solarflare Communications Inc.
   4  * All rights reserved.
   5  *
   6  * This software was jointly developed between OKTET Labs (under contract
   7  * for Solarflare) and Solarflare Communications, Inc.
   8  */
   9
  10 #include <rte_byteorder.h>
  11 #include <rte_tailq.h>
  12 #include <rte_common.h>
  13 #include <rte_ethdev_driver.h>
  14 #include <rte_eth_ctrl.h>
  15 #include <rte_ether.h>
  16 #include <rte_flow.h>
  17 #include <rte_flow_driver.h>
  18
  19 #include "efx.h"
  20
  21 #include "sfc.h"
  22 #include "sfc_rx.h"
  23 #include "sfc_filter.h"
  24 #include "sfc_flow.h"
  25 #include "sfc_log.h"
  26
  27 /*
  28  * At now flow API is implemented in such a manner that each
  29  * flow rule is converted to one or more hardware filters.
  30  * All elements of flow rule (attributes, pattern items, actions)
  31  * correspond to one or more fields in the efx_filter_spec_s structure
  32  * that is responsible for the hardware filter.
  33  * If some required field is unset in the flow rule, then a handful
  34  * of filter copies will be created to cover all possible values
  35  * of such a field.
  36  */
  37
  38 enum sfc_flow_item_layers {
  39         SFC_FLOW_ITEM_ANY_LAYER,
  40         SFC_FLOW_ITEM_START_LAYER,
  41         SFC_FLOW_ITEM_L2,
  42         SFC_FLOW_ITEM_L3,
  43         SFC_FLOW_ITEM_L4,
  44 };
  45
  46 typedef int (sfc_flow_item_parse)(const struct rte_flow_item *item,
  47                                   efx_filter_spec_t *spec,
  48                                   struct rte_flow_error *error);
  49
  50 struct sfc_flow_item {
  51         enum rte_flow_item_type type;           /* Type of item */
  52         enum sfc_flow_item_layers layer;        /* Layer of item */
  53         enum sfc_flow_item_layers prev_layer;   /* Previous layer of item */
  54         sfc_flow_item_parse *parse;             /* Parsing function */
  55 };
  56
  57 static sfc_flow_item_parse sfc_flow_parse_void;
  58 static sfc_flow_item_parse sfc_flow_parse_eth;
  59 static sfc_flow_item_parse sfc_flow_parse_vlan;
  60 static sfc_flow_item_parse sfc_flow_parse_ipv4;
  61 static sfc_flow_item_parse sfc_flow_parse_ipv6;
  62 static sfc_flow_item_parse sfc_flow_parse_tcp;
  63 static sfc_flow_item_parse sfc_flow_parse_udp;
  64 static sfc_flow_item_parse sfc_flow_parse_vxlan;
  65 static sfc_flow_item_parse sfc_flow_parse_geneve;
  66 static sfc_flow_item_parse sfc_flow_parse_nvgre;
  67
  68 typedef int (sfc_flow_spec_set_vals)(struct sfc_flow_spec *spec,
  69                                      unsigned int filters_count_for_one_val,
  70                                      struct rte_flow_error *error);
  71
  72 typedef boolean_t (sfc_flow_spec_check)(efx_filter_match_flags_t match,
  73                                         efx_filter_spec_t *spec,
  74                                         struct sfc_filter *filter);
  75
  76 struct sfc_flow_copy_flag {
  77         /* EFX filter specification match flag */
  78         efx_filter_match_flags_t flag;
  79         /* Number of values of corresponding field */
  80         unsigned int vals_count;
  81         /* Function to set values in specifications */
  82         sfc_flow_spec_set_vals *set_vals;
  83         /*
  84          * Function to check that the specification is suitable
  85          * for adding this match flag
  86          */
  87         sfc_flow_spec_check *spec_check;
  88 };
  89
  90 static sfc_flow_spec_set_vals sfc_flow_set_unknown_dst_flags;
  91 static sfc_flow_spec_check sfc_flow_check_unknown_dst_flags;
  92 static sfc_flow_spec_set_vals sfc_flow_set_ethertypes;
  93 static sfc_flow_spec_set_vals sfc_flow_set_ifrm_unknown_dst_flags;
  94 static sfc_flow_spec_check sfc_flow_check_ifrm_unknown_dst_flags;
  95
  96 static boolean_t
  97 sfc_flow_is_zero(const uint8_t *buf, unsigned int size)
  98 {
  99         uint8_t sum = 0;
 100         unsigned int i;
 101
 102         for (i = 0; i < size; i++)
 103                 sum |= buf[i];
 104
 105         return (sum == 0) ? B_TRUE : B_FALSE;
 106 }
 107
 108 /*
 109  * Validate item and prepare structures spec and mask for parsing
 110  */
 111 static int
 112 sfc_flow_parse_init(const struct rte_flow_item *item,
 113                     const void **spec_ptr,
 114                     const void **mask_ptr,
 115                     const void *supp_mask,
 116                     const void *def_mask,
 117                     unsigned int size,
 118                     struct rte_flow_error *error)
 119 {
 120         const uint8_t *spec;
 121         const uint8_t *mask;
 122         const uint8_t *last;
 123         uint8_t match;
 124         uint8_t supp;
 125         unsigned int i;
 126
 127         if (item == NULL) {
 128                 rte_flow_error_set(error, EINVAL,
 129                                    RTE_FLOW_ERROR_TYPE_ITEM, NULL,
 130                                    "NULL item");
 131                 return -rte_errno;
 132         }
 133
 134         if ((item->last != NULL || item->mask != NULL) && item->spec == NULL) {
 135                 rte_flow_error_set(error, EINVAL,
 136                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
 137                                    "Mask or last is set without spec");
 138                 return -rte_errno;
 139         }
 140
 141         /*
 142          * If "mask" is not set, default mask is used,
 143          * but if default mask is NULL, "mask" should be set
 144          */
 145         if (item->mask == NULL) {
 146                 if (def_mask == NULL) {
 147                         rte_flow_error_set(error, EINVAL,
 148                                 RTE_FLOW_ERROR_TYPE_ITEM, NULL,
 149                                 "Mask should be specified");
 150                         return -rte_errno;
 151                 }
 152
 153                 mask = def_mask;
 154         } else {
 155                 mask = item->mask;
 156         }
 157
 158         spec = item->spec;
 159         last = item->last;
 160
 161         if (spec == NULL)
 162                 goto exit;
 163
 164         /*
 165          * If field values in "last" are either 0 or equal to the corresponding
 166          * values in "spec" then they are ignored
 167          */
 168         if (last != NULL &&
 169             !sfc_flow_is_zero(last, size) &&
 170             memcmp(last, spec, size) != 0) {
 171                 rte_flow_error_set(error, ENOTSUP,
 172                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
 173                                    "Ranging is not supported");
 174                 return -rte_errno;
 175         }
 176
 177         if (supp_mask == NULL) {
 178                 rte_flow_error_set(error, EINVAL,
 179                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
 180                         "Supported mask for item should be specified");
 181                 return -rte_errno;
 182         }
 183
 184         /* Check that mask and spec not asks for more match than supp_mask */
 185         for (i = 0; i < size; i++) {
 186                 match = spec[i] | mask[i];
 187                 supp = ((const uint8_t *)supp_mask)[i];
 188
 189                 if ((match | supp) != supp) {
 190                         rte_flow_error_set(error, ENOTSUP,
 191                                            RTE_FLOW_ERROR_TYPE_ITEM, item,
 192                                            "Item's field is not supported");
 193                         return -rte_errno;
 194                 }
 195         }
 196
 197 exit:
 198         *spec_ptr = spec;
 199         *mask_ptr = mask;
 200         return 0;
 201 }
 202
 203 /*
 204  * Protocol parsers.
 205  * Masking is not supported, so masks in items should be either
 206  * full or empty (zeroed) and set only for supported fields which
 207  * are specified in the supp_mask.
 208  */
 209
 210 static int
 211 sfc_flow_parse_void(__rte_unused const struct rte_flow_item *item,
 212                     __rte_unused efx_filter_spec_t *efx_spec,
 213                     __rte_unused struct rte_flow_error *error)
 214 {
 215         return 0;
 216 }
 217
 218 /**
 219  * Convert Ethernet item to EFX filter specification.
 220  *
 221  * @param item[in]
 222  *   Item specification. Outer frame specification may only comprise
 223  *   source/destination addresses and Ethertype field.
 224  *   Inner frame specification may contain destination address only.
 225  *   There is support for individual/group mask as well as for empty and full.
 226  *   If the mask is NULL, default mask will be used. Ranging is not supported.
 227  * @param efx_spec[in, out]
 228  *   EFX filter specification to update.
 229  * @param[out] error
 230  *   Perform verbose error reporting if not NULL.
 231  */
 232 static int
 233 sfc_flow_parse_eth(const struct rte_flow_item *item,
 234                    efx_filter_spec_t *efx_spec,
 235                    struct rte_flow_error *error)
 236 {
 237         int rc;
 238         const struct rte_flow_item_eth *spec = NULL;
 239         const struct rte_flow_item_eth *mask = NULL;
 240         const struct rte_flow_item_eth supp_mask = {
 241                 .dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
 242                 .src.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
 243                 .type = 0xffff,
 244         };
 245         const struct rte_flow_item_eth ifrm_supp_mask = {
 246                 .dst.addr_bytes = { 0xff, 0xff, 0xff, 0xff, 0xff, 0xff },
 247         };
 248         const uint8_t ig_mask[EFX_MAC_ADDR_LEN] = {
 249                 0x01, 0x00, 0x00, 0x00, 0x00, 0x00
 250         };
 251         const struct rte_flow_item_eth *supp_mask_p;
 252         const struct rte_flow_item_eth *def_mask_p;
 253         uint8_t *loc_mac = NULL;
 254         boolean_t is_ifrm = (efx_spec->efs_encap_type !=
 255                 EFX_TUNNEL_PROTOCOL_NONE);
 256
 257         if (is_ifrm) {
 258                 supp_mask_p = &ifrm_supp_mask;
 259                 def_mask_p = &ifrm_supp_mask;
 260                 loc_mac = efx_spec->efs_ifrm_loc_mac;
 261         } else {
 262                 supp_mask_p = &supp_mask;
 263                 def_mask_p = &rte_flow_item_eth_mask;
 264                 loc_mac = efx_spec->efs_loc_mac;
 265         }
 266
 267         rc = sfc_flow_parse_init(item,
 268                                  (const void **)&spec,
 269                                  (const void **)&mask,
 270                                  supp_mask_p, def_mask_p,
 271                                  sizeof(struct rte_flow_item_eth),
 272                                  error);
 273         if (rc != 0)
 274                 return rc;
 275
 276         /* If "spec" is not set, could be any Ethernet */
 277         if (spec == NULL)
 278                 return 0;
 279
 280         if (is_same_ether_addr(&mask->dst, &supp_mask.dst)) {
 281                 efx_spec->efs_match_flags |= is_ifrm ?
 282                         EFX_FILTER_MATCH_IFRM_LOC_MAC :
 283                         EFX_FILTER_MATCH_LOC_MAC;
 284                 rte_memcpy(loc_mac, spec->dst.addr_bytes,
 285                            EFX_MAC_ADDR_LEN);
 286         } else if (memcmp(mask->dst.addr_bytes, ig_mask,
 287                           EFX_MAC_ADDR_LEN) == 0) {
 288                 if (is_unicast_ether_addr(&spec->dst))
 289                         efx_spec->efs_match_flags |= is_ifrm ?
 290                                 EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST :
 291                                 EFX_FILTER_MATCH_UNKNOWN_UCAST_DST;
 292                 else
 293                         efx_spec->efs_match_flags |= is_ifrm ?
 294                                 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST :
 295                                 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST;
 296         } else if (!is_zero_ether_addr(&mask->dst)) {
 297                 goto fail_bad_mask;
 298         }
 299
 300         /*
 301          * ifrm_supp_mask ensures that the source address and
 302          * ethertype masks are equal to zero in inner frame,
 303          * so these fields are filled in only for the outer frame
 304          */
 305         if (is_same_ether_addr(&mask->src, &supp_mask.src)) {
 306                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_MAC;
 307                 rte_memcpy(efx_spec->efs_rem_mac, spec->src.addr_bytes,
 308                            EFX_MAC_ADDR_LEN);
 309         } else if (!is_zero_ether_addr(&mask->src)) {
 310                 goto fail_bad_mask;
 311         }
 312
 313         /*
 314          * Ether type is in big-endian byte order in item and
 315          * in little-endian in efx_spec, so byte swap is used
 316          */
 317         if (mask->type == supp_mask.type) {
 318                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
 319                 efx_spec->efs_ether_type = rte_bswap16(spec->type);
 320         } else if (mask->type != 0) {
 321                 goto fail_bad_mask;
 322         }
 323
 324         return 0;
 325
 326 fail_bad_mask:
 327         rte_flow_error_set(error, EINVAL,
 328                            RTE_FLOW_ERROR_TYPE_ITEM, item,
 329                            "Bad mask in the ETH pattern item");
 330         return -rte_errno;
 331 }
 332
 333 /**
 334  * Convert VLAN item to EFX filter specification.
 335  *
 336  * @param item[in]
 337  *   Item specification. Only VID field is supported.
 338  *   The mask can not be NULL. Ranging is not supported.
 339  * @param efx_spec[in, out]
 340  *   EFX filter specification to update.
 341  * @param[out] error
 342  *   Perform verbose error reporting if not NULL.
 343  */
 344 static int
 345 sfc_flow_parse_vlan(const struct rte_flow_item *item,
 346                     efx_filter_spec_t *efx_spec,
 347                     struct rte_flow_error *error)
 348 {
 349         int rc;
 350         uint16_t vid;
 351         const struct rte_flow_item_vlan *spec = NULL;
 352         const struct rte_flow_item_vlan *mask = NULL;
 353         const struct rte_flow_item_vlan supp_mask = {
 354                 .tci = rte_cpu_to_be_16(ETH_VLAN_ID_MAX),
 355                 .inner_type = RTE_BE16(0xffff),
 356         };
 357
 358         rc = sfc_flow_parse_init(item,
 359                                  (const void **)&spec,
 360                                  (const void **)&mask,
 361                                  &supp_mask,
 362                                  NULL,
 363                                  sizeof(struct rte_flow_item_vlan),
 364                                  error);
 365         if (rc != 0)
 366                 return rc;
 367
 368         /*
 369          * VID is in big-endian byte order in item and
 370          * in little-endian in efx_spec, so byte swap is used.
 371          * If two VLAN items are included, the first matches
 372          * the outer tag and the next matches the inner tag.
 373          */
 374         if (mask->tci == supp_mask.tci) {
 375                 vid = rte_bswap16(spec->tci);
 376
 377                 if (!(efx_spec->efs_match_flags &
 378                       EFX_FILTER_MATCH_OUTER_VID)) {
 379                         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_OUTER_VID;
 380                         efx_spec->efs_outer_vid = vid;
 381                 } else if (!(efx_spec->efs_match_flags &
 382                              EFX_FILTER_MATCH_INNER_VID)) {
 383                         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_INNER_VID;
 384                         efx_spec->efs_inner_vid = vid;
 385                 } else {
 386                         rte_flow_error_set(error, EINVAL,
 387                                            RTE_FLOW_ERROR_TYPE_ITEM, item,
 388                                            "More than two VLAN items");
 389                         return -rte_errno;
 390                 }
 391         } else {
 392                 rte_flow_error_set(error, EINVAL,
 393                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
 394                                    "VLAN ID in TCI match is required");
 395                 return -rte_errno;
 396         }
 397
 398         if (efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE) {
 399                 rte_flow_error_set(error, EINVAL,
 400                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
 401                                    "VLAN TPID matching is not supported");
 402                 return -rte_errno;
 403         }
 404         if (mask->inner_type == supp_mask.inner_type) {
 405                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
 406                 efx_spec->efs_ether_type = rte_bswap16(spec->inner_type);
 407         } else if (mask->inner_type) {
 408                 rte_flow_error_set(error, EINVAL,
 409                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
 410                                    "Bad mask for VLAN inner_type");
 411                 return -rte_errno;
 412         }
 413
 414         return 0;
 415 }
 416
 417 /**
 418  * Convert IPv4 item to EFX filter specification.
 419  *
 420  * @param item[in]
 421  *   Item specification. Only source and destination addresses and
 422  *   protocol fields are supported. If the mask is NULL, default
 423  *   mask will be used. Ranging is not supported.
 424  * @param efx_spec[in, out]
 425  *   EFX filter specification to update.
 426  * @param[out] error
 427  *   Perform verbose error reporting if not NULL.
 428  */
 429 static int
 430 sfc_flow_parse_ipv4(const struct rte_flow_item *item,
 431                     efx_filter_spec_t *efx_spec,
 432                     struct rte_flow_error *error)
 433 {
 434         int rc;
 435         const struct rte_flow_item_ipv4 *spec = NULL;
 436         const struct rte_flow_item_ipv4 *mask = NULL;
 437         const uint16_t ether_type_ipv4 = rte_cpu_to_le_16(EFX_ETHER_TYPE_IPV4);
 438         const struct rte_flow_item_ipv4 supp_mask = {
 439                 .hdr = {
 440                         .src_addr = 0xffffffff,
 441                         .dst_addr = 0xffffffff,
 442                         .next_proto_id = 0xff,
 443                 }
 444         };
 445
 446         rc = sfc_flow_parse_init(item,
 447                                  (const void **)&spec,
 448                                  (const void **)&mask,
 449                                  &supp_mask,
 450                                  &rte_flow_item_ipv4_mask,
 451                                  sizeof(struct rte_flow_item_ipv4),
 452                                  error);
 453         if (rc != 0)
 454                 return rc;
 455
 456         /*
 457          * Filtering by IPv4 source and destination addresses requires
 458          * the appropriate ETHER_TYPE in hardware filters
 459          */
 460         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE)) {
 461                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
 462                 efx_spec->efs_ether_type = ether_type_ipv4;
 463         } else if (efx_spec->efs_ether_type != ether_type_ipv4) {
 464                 rte_flow_error_set(error, EINVAL,
 465                         RTE_FLOW_ERROR_TYPE_ITEM, item,
 466                         "Ethertype in pattern with IPV4 item should be appropriate");
 467                 return -rte_errno;
 468         }
 469
 470         if (spec == NULL)
 471                 return 0;
 472
 473         /*
 474          * IPv4 addresses are in big-endian byte order in item and in
 475          * efx_spec
 476          */
 477         if (mask->hdr.src_addr == supp_mask.hdr.src_addr) {
 478                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_HOST;
 479                 efx_spec->efs_rem_host.eo_u32[0] = spec->hdr.src_addr;
 480         } else if (mask->hdr.src_addr != 0) {
 481                 goto fail_bad_mask;
 482         }
 483
 484         if (mask->hdr.dst_addr == supp_mask.hdr.dst_addr) {
 485                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_HOST;
 486                 efx_spec->efs_loc_host.eo_u32[0] = spec->hdr.dst_addr;
 487         } else if (mask->hdr.dst_addr != 0) {
 488                 goto fail_bad_mask;
 489         }
 490
 491         if (mask->hdr.next_proto_id == supp_mask.hdr.next_proto_id) {
 492                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
 493                 efx_spec->efs_ip_proto = spec->hdr.next_proto_id;
 494         } else if (mask->hdr.next_proto_id != 0) {
 495                 goto fail_bad_mask;
 496         }
 497
 498         return 0;
 499
 500 fail_bad_mask:
 501         rte_flow_error_set(error, EINVAL,
 502                            RTE_FLOW_ERROR_TYPE_ITEM, item,
 503                            "Bad mask in the IPV4 pattern item");
 504         return -rte_errno;
 505 }
 506
 507 /**
 508  * Convert IPv6 item to EFX filter specification.
 509  *
 510  * @param item[in]
 511  *   Item specification. Only source and destination addresses and
 512  *   next header fields are supported. If the mask is NULL, default
 513  *   mask will be used. Ranging is not supported.
 514  * @param efx_spec[in, out]
 515  *   EFX filter specification to update.
 516  * @param[out] error
 517  *   Perform verbose error reporting if not NULL.
 518  */
 519 static int
 520 sfc_flow_parse_ipv6(const struct rte_flow_item *item,
 521                     efx_filter_spec_t *efx_spec,
 522                     struct rte_flow_error *error)
 523 {
 524         int rc;
 525         const struct rte_flow_item_ipv6 *spec = NULL;
 526         const struct rte_flow_item_ipv6 *mask = NULL;
 527         const uint16_t ether_type_ipv6 = rte_cpu_to_le_16(EFX_ETHER_TYPE_IPV6);
 528         const struct rte_flow_item_ipv6 supp_mask = {
 529                 .hdr = {
 530                         .src_addr = { 0xff, 0xff, 0xff, 0xff,
 531                                       0xff, 0xff, 0xff, 0xff,
 532                                       0xff, 0xff, 0xff, 0xff,
 533                                       0xff, 0xff, 0xff, 0xff },
 534                         .dst_addr = { 0xff, 0xff, 0xff, 0xff,
 535                                       0xff, 0xff, 0xff, 0xff,
 536                                       0xff, 0xff, 0xff, 0xff,
 537                                       0xff, 0xff, 0xff, 0xff },
 538                         .proto = 0xff,
 539                 }
 540         };
 541
 542         rc = sfc_flow_parse_init(item,
 543                                  (const void **)&spec,
 544                                  (const void **)&mask,
 545                                  &supp_mask,
 546                                  &rte_flow_item_ipv6_mask,
 547                                  sizeof(struct rte_flow_item_ipv6),
 548                                  error);
 549         if (rc != 0)
 550                 return rc;
 551
 552         /*
 553          * Filtering by IPv6 source and destination addresses requires
 554          * the appropriate ETHER_TYPE in hardware filters
 555          */
 556         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE)) {
 557                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ETHER_TYPE;
 558                 efx_spec->efs_ether_type = ether_type_ipv6;
 559         } else if (efx_spec->efs_ether_type != ether_type_ipv6) {
 560                 rte_flow_error_set(error, EINVAL,
 561                         RTE_FLOW_ERROR_TYPE_ITEM, item,
 562                         "Ethertype in pattern with IPV6 item should be appropriate");
 563                 return -rte_errno;
 564         }
 565
 566         if (spec == NULL)
 567                 return 0;
 568
 569         /*
 570          * IPv6 addresses are in big-endian byte order in item and in
 571          * efx_spec
 572          */
 573         if (memcmp(mask->hdr.src_addr, supp_mask.hdr.src_addr,
 574                    sizeof(mask->hdr.src_addr)) == 0) {
 575                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_HOST;
 576
 577                 RTE_BUILD_BUG_ON(sizeof(efx_spec->efs_rem_host) !=
 578                                  sizeof(spec->hdr.src_addr));
 579                 rte_memcpy(&efx_spec->efs_rem_host, spec->hdr.src_addr,
 580                            sizeof(efx_spec->efs_rem_host));
 581         } else if (!sfc_flow_is_zero(mask->hdr.src_addr,
 582                                      sizeof(mask->hdr.src_addr))) {
 583                 goto fail_bad_mask;
 584         }
 585
 586         if (memcmp(mask->hdr.dst_addr, supp_mask.hdr.dst_addr,
 587                    sizeof(mask->hdr.dst_addr)) == 0) {
 588                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_HOST;
 589
 590                 RTE_BUILD_BUG_ON(sizeof(efx_spec->efs_loc_host) !=
 591                                  sizeof(spec->hdr.dst_addr));
 592                 rte_memcpy(&efx_spec->efs_loc_host, spec->hdr.dst_addr,
 593                            sizeof(efx_spec->efs_loc_host));
 594         } else if (!sfc_flow_is_zero(mask->hdr.dst_addr,
 595                                      sizeof(mask->hdr.dst_addr))) {
 596                 goto fail_bad_mask;
 597         }
 598
 599         if (mask->hdr.proto == supp_mask.hdr.proto) {
 600                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
 601                 efx_spec->efs_ip_proto = spec->hdr.proto;
 602         } else if (mask->hdr.proto != 0) {
 603                 goto fail_bad_mask;
 604         }
 605
 606         return 0;
 607
 608 fail_bad_mask:
 609         rte_flow_error_set(error, EINVAL,
 610                            RTE_FLOW_ERROR_TYPE_ITEM, item,
 611                            "Bad mask in the IPV6 pattern item");
 612         return -rte_errno;
 613 }
 614
 615 /**
 616  * Convert TCP item to EFX filter specification.
 617  *
 618  * @param item[in]
 619  *   Item specification. Only source and destination ports fields
 620  *   are supported. If the mask is NULL, default mask will be used.
 621  *   Ranging is not supported.
 622  * @param efx_spec[in, out]
 623  *   EFX filter specification to update.
 624  * @param[out] error
 625  *   Perform verbose error reporting if not NULL.
 626  */
 627 static int
 628 sfc_flow_parse_tcp(const struct rte_flow_item *item,
 629                    efx_filter_spec_t *efx_spec,
 630                    struct rte_flow_error *error)
 631 {
 632         int rc;
 633         const struct rte_flow_item_tcp *spec = NULL;
 634         const struct rte_flow_item_tcp *mask = NULL;
 635         const struct rte_flow_item_tcp supp_mask = {
 636                 .hdr = {
 637                         .src_port = 0xffff,
 638                         .dst_port = 0xffff,
 639                 }
 640         };
 641
 642         rc = sfc_flow_parse_init(item,
 643                                  (const void **)&spec,
 644                                  (const void **)&mask,
 645                                  &supp_mask,
 646                                  &rte_flow_item_tcp_mask,
 647                                  sizeof(struct rte_flow_item_tcp),
 648                                  error);
 649         if (rc != 0)
 650                 return rc;
 651
 652         /*
 653          * Filtering by TCP source and destination ports requires
 654          * the appropriate IP_PROTO in hardware filters
 655          */
 656         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
 657                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
 658                 efx_spec->efs_ip_proto = EFX_IPPROTO_TCP;
 659         } else if (efx_spec->efs_ip_proto != EFX_IPPROTO_TCP) {
 660                 rte_flow_error_set(error, EINVAL,
 661                         RTE_FLOW_ERROR_TYPE_ITEM, item,
 662                         "IP proto in pattern with TCP item should be appropriate");
 663                 return -rte_errno;
 664         }
 665
 666         if (spec == NULL)
 667                 return 0;
 668
 669         /*
 670          * Source and destination ports are in big-endian byte order in item and
 671          * in little-endian in efx_spec, so byte swap is used
 672          */
 673         if (mask->hdr.src_port == supp_mask.hdr.src_port) {
 674                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_PORT;
 675                 efx_spec->efs_rem_port = rte_bswap16(spec->hdr.src_port);
 676         } else if (mask->hdr.src_port != 0) {
 677                 goto fail_bad_mask;
 678         }
 679
 680         if (mask->hdr.dst_port == supp_mask.hdr.dst_port) {
 681                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_PORT;
 682                 efx_spec->efs_loc_port = rte_bswap16(spec->hdr.dst_port);
 683         } else if (mask->hdr.dst_port != 0) {
 684                 goto fail_bad_mask;
 685         }
 686
 687         return 0;
 688
 689 fail_bad_mask:
 690         rte_flow_error_set(error, EINVAL,
 691                            RTE_FLOW_ERROR_TYPE_ITEM, item,
 692                            "Bad mask in the TCP pattern item");
 693         return -rte_errno;
 694 }
 695
 696 /**
 697  * Convert UDP item to EFX filter specification.
 698  *
 699  * @param item[in]
 700  *   Item specification. Only source and destination ports fields
 701  *   are supported. If the mask is NULL, default mask will be used.
 702  *   Ranging is not supported.
 703  * @param efx_spec[in, out]
 704  *   EFX filter specification to update.
 705  * @param[out] error
 706  *   Perform verbose error reporting if not NULL.
 707  */
 708 static int
 709 sfc_flow_parse_udp(const struct rte_flow_item *item,
 710                    efx_filter_spec_t *efx_spec,
 711                    struct rte_flow_error *error)
 712 {
 713         int rc;
 714         const struct rte_flow_item_udp *spec = NULL;
 715         const struct rte_flow_item_udp *mask = NULL;
 716         const struct rte_flow_item_udp supp_mask = {
 717                 .hdr = {
 718                         .src_port = 0xffff,
 719                         .dst_port = 0xffff,
 720                 }
 721         };
 722
 723         rc = sfc_flow_parse_init(item,
 724                                  (const void **)&spec,
 725                                  (const void **)&mask,
 726                                  &supp_mask,
 727                                  &rte_flow_item_udp_mask,
 728                                  sizeof(struct rte_flow_item_udp),
 729                                  error);
 730         if (rc != 0)
 731                 return rc;
 732
 733         /*
 734          * Filtering by UDP source and destination ports requires
 735          * the appropriate IP_PROTO in hardware filters
 736          */
 737         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
 738                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
 739                 efx_spec->efs_ip_proto = EFX_IPPROTO_UDP;
 740         } else if (efx_spec->efs_ip_proto != EFX_IPPROTO_UDP) {
 741                 rte_flow_error_set(error, EINVAL,
 742                         RTE_FLOW_ERROR_TYPE_ITEM, item,
 743                         "IP proto in pattern with UDP item should be appropriate");
 744                 return -rte_errno;
 745         }
 746
 747         if (spec == NULL)
 748                 return 0;
 749
 750         /*
 751          * Source and destination ports are in big-endian byte order in item and
 752          * in little-endian in efx_spec, so byte swap is used
 753          */
 754         if (mask->hdr.src_port == supp_mask.hdr.src_port) {
 755                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_REM_PORT;
 756                 efx_spec->efs_rem_port = rte_bswap16(spec->hdr.src_port);
 757         } else if (mask->hdr.src_port != 0) {
 758                 goto fail_bad_mask;
 759         }
 760
 761         if (mask->hdr.dst_port == supp_mask.hdr.dst_port) {
 762                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_LOC_PORT;
 763                 efx_spec->efs_loc_port = rte_bswap16(spec->hdr.dst_port);
 764         } else if (mask->hdr.dst_port != 0) {
 765                 goto fail_bad_mask;
 766         }
 767
 768         return 0;
 769
 770 fail_bad_mask:
 771         rte_flow_error_set(error, EINVAL,
 772                            RTE_FLOW_ERROR_TYPE_ITEM, item,
 773                            "Bad mask in the UDP pattern item");
 774         return -rte_errno;
 775 }
 776
 777 /*
 778  * Filters for encapsulated packets match based on the EtherType and IP
 779  * protocol in the outer frame.
 780  */
 781 static int
 782 sfc_flow_set_match_flags_for_encap_pkts(const struct rte_flow_item *item,
 783                                         efx_filter_spec_t *efx_spec,
 784                                         uint8_t ip_proto,
 785                                         struct rte_flow_error *error)
 786 {
 787         if (!(efx_spec->efs_match_flags & EFX_FILTER_MATCH_IP_PROTO)) {
 788                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_IP_PROTO;
 789                 efx_spec->efs_ip_proto = ip_proto;
 790         } else if (efx_spec->efs_ip_proto != ip_proto) {
 791                 switch (ip_proto) {
 792                 case EFX_IPPROTO_UDP:
 793                         rte_flow_error_set(error, EINVAL,
 794                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
 795                                 "Outer IP header protocol must be UDP "
 796                                 "in VxLAN/GENEVE pattern");
 797                         return -rte_errno;
 798
 799                 case EFX_IPPROTO_GRE:
 800                         rte_flow_error_set(error, EINVAL,
 801                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
 802                                 "Outer IP header protocol must be GRE "
 803                                 "in NVGRE pattern");
 804                         return -rte_errno;
 805
 806                 default:
 807                         rte_flow_error_set(error, EINVAL,
 808                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
 809                                 "Only VxLAN/GENEVE/NVGRE tunneling patterns "
 810                                 "are supported");
 811                         return -rte_errno;
 812                 }
 813         }
 814
 815         if (efx_spec->efs_match_flags & EFX_FILTER_MATCH_ETHER_TYPE &&
 816             efx_spec->efs_ether_type != EFX_ETHER_TYPE_IPV4 &&
 817             efx_spec->efs_ether_type != EFX_ETHER_TYPE_IPV6) {
 818                 rte_flow_error_set(error, EINVAL,
 819                         RTE_FLOW_ERROR_TYPE_ITEM, item,
 820                         "Outer frame EtherType in pattern with tunneling "
 821                         "must be IPv4 or IPv6");
 822                 return -rte_errno;
 823         }
 824
 825         return 0;
 826 }
 827
 828 static int
 829 sfc_flow_set_efx_spec_vni_or_vsid(efx_filter_spec_t *efx_spec,
 830                                   const uint8_t *vni_or_vsid_val,
 831                                   const uint8_t *vni_or_vsid_mask,
 832                                   const struct rte_flow_item *item,
 833                                   struct rte_flow_error *error)
 834 {
 835         const uint8_t vni_or_vsid_full_mask[EFX_VNI_OR_VSID_LEN] = {
 836                 0xff, 0xff, 0xff
 837         };
 838
 839         if (memcmp(vni_or_vsid_mask, vni_or_vsid_full_mask,
 840                    EFX_VNI_OR_VSID_LEN) == 0) {
 841                 efx_spec->efs_match_flags |= EFX_FILTER_MATCH_VNI_OR_VSID;
 842                 rte_memcpy(efx_spec->efs_vni_or_vsid, vni_or_vsid_val,
 843                            EFX_VNI_OR_VSID_LEN);
 844         } else if (!sfc_flow_is_zero(vni_or_vsid_mask, EFX_VNI_OR_VSID_LEN)) {
 845                 rte_flow_error_set(error, EINVAL,
 846                                    RTE_FLOW_ERROR_TYPE_ITEM, item,
 847                                    "Unsupported VNI/VSID mask");
 848                 return -rte_errno;
 849         }
 850
 851         return 0;
 852 }
 853
 854 /**
 855  * Convert VXLAN item to EFX filter specification.
 856  *
 857  * @param item[in]
 858  *   Item specification. Only VXLAN network identifier field is supported.
 859  *   If the mask is NULL, default mask will be used.
 860  *   Ranging is not supported.
 861  * @param efx_spec[in, out]
 862  *   EFX filter specification to update.
 863  * @param[out] error
 864  *   Perform verbose error reporting if not NULL.
 865  */
 866 static int
 867 sfc_flow_parse_vxlan(const struct rte_flow_item *item,
 868                      efx_filter_spec_t *efx_spec,
 869                      struct rte_flow_error *error)
 870 {
 871         int rc;
 872         const struct rte_flow_item_vxlan *spec = NULL;
 873         const struct rte_flow_item_vxlan *mask = NULL;
 874         const struct rte_flow_item_vxlan supp_mask = {
 875                 .vni = { 0xff, 0xff, 0xff }
 876         };
 877
 878         rc = sfc_flow_parse_init(item,
 879                                  (const void **)&spec,
 880                                  (const void **)&mask,
 881                                  &supp_mask,
 882                                  &rte_flow_item_vxlan_mask,
 883                                  sizeof(struct rte_flow_item_vxlan),
 884                                  error);
 885         if (rc != 0)
 886                 return rc;
 887
 888         rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
 889                                                      EFX_IPPROTO_UDP, error);
 890         if (rc != 0)
 891                 return rc;
 892
 893         efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_VXLAN;
 894         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
 895
 896         if (spec == NULL)
 897                 return 0;
 898
 899         rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->vni,
 900                                                mask->vni, item, error);
 901
 902         return rc;
 903 }
 904
 905 /**
 906  * Convert GENEVE item to EFX filter specification.
 907  *
 908  * @param item[in]
 909  *   Item specification. Only Virtual Network Identifier and protocol type
 910  *   fields are supported. But protocol type can be only Ethernet (0x6558).
 911  *   If the mask is NULL, default mask will be used.
 912  *   Ranging is not supported.
 913  * @param efx_spec[in, out]
 914  *   EFX filter specification to update.
 915  * @param[out] error
 916  *   Perform verbose error reporting if not NULL.
 917  */
 918 static int
 919 sfc_flow_parse_geneve(const struct rte_flow_item *item,
 920                       efx_filter_spec_t *efx_spec,
 921                       struct rte_flow_error *error)
 922 {
 923         int rc;
 924         const struct rte_flow_item_geneve *spec = NULL;
 925         const struct rte_flow_item_geneve *mask = NULL;
 926         const struct rte_flow_item_geneve supp_mask = {
 927                 .protocol = RTE_BE16(0xffff),
 928                 .vni = { 0xff, 0xff, 0xff }
 929         };
 930
 931         rc = sfc_flow_parse_init(item,
 932                                  (const void **)&spec,
 933                                  (const void **)&mask,
 934                                  &supp_mask,
 935                                  &rte_flow_item_geneve_mask,
 936                                  sizeof(struct rte_flow_item_geneve),
 937                                  error);
 938         if (rc != 0)
 939                 return rc;
 940
 941         rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
 942                                                      EFX_IPPROTO_UDP, error);
 943         if (rc != 0)
 944                 return rc;
 945
 946         efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_GENEVE;
 947         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
 948
 949         if (spec == NULL)
 950                 return 0;
 951
 952         if (mask->protocol == supp_mask.protocol) {
 953                 if (spec->protocol != rte_cpu_to_be_16(ETHER_TYPE_TEB)) {
 954                         rte_flow_error_set(error, EINVAL,
 955                                 RTE_FLOW_ERROR_TYPE_ITEM, item,
 956                                 "GENEVE encap. protocol must be Ethernet "
 957                                 "(0x6558) in the GENEVE pattern item");
 958                         return -rte_errno;
 959                 }
 960         } else if (mask->protocol != 0) {
 961                 rte_flow_error_set(error, EINVAL,
 962                         RTE_FLOW_ERROR_TYPE_ITEM, item,
 963                         "Unsupported mask for GENEVE encap. protocol");
 964                 return -rte_errno;
 965         }
 966
 967         rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->vni,
 968                                                mask->vni, item, error);
 969
 970         return rc;
 971 }
 972
 973 /**
 974  * Convert NVGRE item to EFX filter specification.
 975  *
 976  * @param item[in]
 977  *   Item specification. Only virtual subnet ID field is supported.
 978  *   If the mask is NULL, default mask will be used.
 979  *   Ranging is not supported.
 980  * @param efx_spec[in, out]
 981  *   EFX filter specification to update.
 982  * @param[out] error
 983  *   Perform verbose error reporting if not NULL.
 984  */
 985 static int
 986 sfc_flow_parse_nvgre(const struct rte_flow_item *item,
 987                      efx_filter_spec_t *efx_spec,
 988                      struct rte_flow_error *error)
 989 {
 990         int rc;
 991         const struct rte_flow_item_nvgre *spec = NULL;
 992         const struct rte_flow_item_nvgre *mask = NULL;
 993         const struct rte_flow_item_nvgre supp_mask = {
 994                 .tni = { 0xff, 0xff, 0xff }
 995         };
 996
 997         rc = sfc_flow_parse_init(item,
 998                                  (const void **)&spec,
 999                                  (const void **)&mask,
1000                                  &supp_mask,
1001                                  &rte_flow_item_nvgre_mask,
1002                                  sizeof(struct rte_flow_item_nvgre),
1003                                  error);
1004         if (rc != 0)
1005                 return rc;
1006
1007         rc = sfc_flow_set_match_flags_for_encap_pkts(item, efx_spec,
1008                                                      EFX_IPPROTO_GRE, error);
1009         if (rc != 0)
1010                 return rc;
1011
1012         efx_spec->efs_encap_type = EFX_TUNNEL_PROTOCOL_NVGRE;
1013         efx_spec->efs_match_flags |= EFX_FILTER_MATCH_ENCAP_TYPE;
1014
1015         if (spec == NULL)
1016                 return 0;
1017
1018         rc = sfc_flow_set_efx_spec_vni_or_vsid(efx_spec, spec->tni,
1019                                                mask->tni, item, error);
1020
1021         return rc;
1022 }
1023
1024 static const struct sfc_flow_item sfc_flow_items[] = {
1025         {
1026                 .type = RTE_FLOW_ITEM_TYPE_VOID,
1027                 .prev_layer = SFC_FLOW_ITEM_ANY_LAYER,
1028                 .layer = SFC_FLOW_ITEM_ANY_LAYER,
1029                 .parse = sfc_flow_parse_void,
1030         },
1031         {
1032                 .type = RTE_FLOW_ITEM_TYPE_ETH,
1033                 .prev_layer = SFC_FLOW_ITEM_START_LAYER,
1034                 .layer = SFC_FLOW_ITEM_L2,
1035                 .parse = sfc_flow_parse_eth,
1036         },
1037         {
1038                 .type = RTE_FLOW_ITEM_TYPE_VLAN,
1039                 .prev_layer = SFC_FLOW_ITEM_L2,
1040                 .layer = SFC_FLOW_ITEM_L2,
1041                 .parse = sfc_flow_parse_vlan,
1042         },
1043         {
1044                 .type = RTE_FLOW_ITEM_TYPE_IPV4,
1045                 .prev_layer = SFC_FLOW_ITEM_L2,
1046                 .layer = SFC_FLOW_ITEM_L3,
1047                 .parse = sfc_flow_parse_ipv4,
1048         },
1049         {
1050                 .type = RTE_FLOW_ITEM_TYPE_IPV6,
1051                 .prev_layer = SFC_FLOW_ITEM_L2,
1052                 .layer = SFC_FLOW_ITEM_L3,
1053                 .parse = sfc_flow_parse_ipv6,
1054         },
1055         {
1056                 .type = RTE_FLOW_ITEM_TYPE_TCP,
1057                 .prev_layer = SFC_FLOW_ITEM_L3,
1058                 .layer = SFC_FLOW_ITEM_L4,
1059                 .parse = sfc_flow_parse_tcp,
1060         },
1061         {
1062                 .type = RTE_FLOW_ITEM_TYPE_UDP,
1063                 .prev_layer = SFC_FLOW_ITEM_L3,
1064                 .layer = SFC_FLOW_ITEM_L4,
1065                 .parse = sfc_flow_parse_udp,
1066         },
1067         {
1068                 .type = RTE_FLOW_ITEM_TYPE_VXLAN,
1069                 .prev_layer = SFC_FLOW_ITEM_L4,
1070                 .layer = SFC_FLOW_ITEM_START_LAYER,
1071                 .parse = sfc_flow_parse_vxlan,
1072         },
1073         {
1074                 .type = RTE_FLOW_ITEM_TYPE_GENEVE,
1075                 .prev_layer = SFC_FLOW_ITEM_L4,
1076                 .layer = SFC_FLOW_ITEM_START_LAYER,
1077                 .parse = sfc_flow_parse_geneve,
1078         },
1079         {
1080                 .type = RTE_FLOW_ITEM_TYPE_NVGRE,
1081                 .prev_layer = SFC_FLOW_ITEM_L3,
1082                 .layer = SFC_FLOW_ITEM_START_LAYER,
1083                 .parse = sfc_flow_parse_nvgre,
1084         },
1085 };
1086
1087 /*
1088  * Protocol-independent flow API support
1089  */
1090 static int
1091 sfc_flow_parse_attr(const struct rte_flow_attr *attr,
1092                     struct rte_flow *flow,
1093                     struct rte_flow_error *error)
1094 {
1095         if (attr == NULL) {
1096                 rte_flow_error_set(error, EINVAL,
1097                                    RTE_FLOW_ERROR_TYPE_ATTR, NULL,
1098                                    "NULL attribute");
1099                 return -rte_errno;
1100         }
1101         if (attr->group != 0) {
1102                 rte_flow_error_set(error, ENOTSUP,
1103                                    RTE_FLOW_ERROR_TYPE_ATTR_GROUP, attr,
1104                                    "Groups are not supported");
1105                 return -rte_errno;
1106         }
1107         if (attr->priority != 0) {
1108                 rte_flow_error_set(error, ENOTSUP,
1109                                    RTE_FLOW_ERROR_TYPE_ATTR_PRIORITY, attr,
1110                                    "Priorities are not supported");
1111                 return -rte_errno;
1112         }
1113         if (attr->egress != 0) {
1114                 rte_flow_error_set(error, ENOTSUP,
1115                                    RTE_FLOW_ERROR_TYPE_ATTR_EGRESS, attr,
1116                                    "Egress is not supported");
1117                 return -rte_errno;
1118         }
1119         if (attr->ingress == 0) {
1120                 rte_flow_error_set(error, ENOTSUP,
1121                                    RTE_FLOW_ERROR_TYPE_ATTR_INGRESS, attr,
1122                                    "Only ingress is supported");
1123                 return -rte_errno;
1124         }
1125
1126         flow->spec.template.efs_flags |= EFX_FILTER_FLAG_RX;
1127         flow->spec.template.efs_rss_context = EFX_RSS_CONTEXT_DEFAULT;
1128
1129         return 0;
1130 }
1131
1132 /* Get item from array sfc_flow_items */
1133 static const struct sfc_flow_item *
1134 sfc_flow_get_item(enum rte_flow_item_type type)
1135 {
1136         unsigned int i;
1137
1138         for (i = 0; i < RTE_DIM(sfc_flow_items); i++)
1139                 if (sfc_flow_items[i].type == type)
1140                         return &sfc_flow_items[i];
1141
1142         return NULL;
1143 }
1144
1145 static int
1146 sfc_flow_parse_pattern(const struct rte_flow_item pattern[],
1147                        struct rte_flow *flow,
1148                        struct rte_flow_error *error)
1149 {
1150         int rc;
1151         unsigned int prev_layer = SFC_FLOW_ITEM_ANY_LAYER;
1152         boolean_t is_ifrm = B_FALSE;
1153         const struct sfc_flow_item *item;
1154
1155         if (pattern == NULL) {
1156                 rte_flow_error_set(error, EINVAL,
1157                                    RTE_FLOW_ERROR_TYPE_ITEM_NUM, NULL,
1158                                    "NULL pattern");
1159                 return -rte_errno;
1160         }
1161
1162         for (; pattern->type != RTE_FLOW_ITEM_TYPE_END; pattern++) {
1163                 item = sfc_flow_get_item(pattern->type);
1164                 if (item == NULL) {
1165                         rte_flow_error_set(error, ENOTSUP,
1166                                            RTE_FLOW_ERROR_TYPE_ITEM, pattern,
1167                                            "Unsupported pattern item");
1168                         return -rte_errno;
1169                 }
1170
1171                 /*
1172                  * Omitting one or several protocol layers at the beginning
1173                  * of pattern is supported
1174                  */
1175                 if (item->prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
1176                     prev_layer != SFC_FLOW_ITEM_ANY_LAYER &&
1177                     item->prev_layer != prev_layer) {
1178                         rte_flow_error_set(error, ENOTSUP,
1179                                            RTE_FLOW_ERROR_TYPE_ITEM, pattern,
1180                                            "Unexpected sequence of pattern items");
1181                         return -rte_errno;
1182                 }
1183
1184                 /*
1185                  * Allow only VOID and ETH pattern items in the inner frame.
1186                  * Also check that there is only one tunneling protocol.
1187                  */
1188                 switch (item->type) {
1189                 case RTE_FLOW_ITEM_TYPE_VOID:
1190                 case RTE_FLOW_ITEM_TYPE_ETH:
1191                         break;
1192
1193                 case RTE_FLOW_ITEM_TYPE_VXLAN:
1194                 case RTE_FLOW_ITEM_TYPE_GENEVE:
1195                 case RTE_FLOW_ITEM_TYPE_NVGRE:
1196                         if (is_ifrm) {
1197                                 rte_flow_error_set(error, EINVAL,
1198                                         RTE_FLOW_ERROR_TYPE_ITEM,
1199                                         pattern,
1200                                         "More than one tunneling protocol");
1201                                 return -rte_errno;
1202                         }
1203                         is_ifrm = B_TRUE;
1204                         break;
1205
1206                 default:
1207                         if (is_ifrm) {
1208                                 rte_flow_error_set(error, EINVAL,
1209                                         RTE_FLOW_ERROR_TYPE_ITEM,
1210                                         pattern,
1211                                         "There is an unsupported pattern item "
1212                                         "in the inner frame");
1213                                 return -rte_errno;
1214                         }
1215                         break;
1216                 }
1217
1218                 rc = item->parse(pattern, &flow->spec.template, error);
1219                 if (rc != 0)
1220                         return rc;
1221
1222                 if (item->layer != SFC_FLOW_ITEM_ANY_LAYER)
1223                         prev_layer = item->layer;
1224         }
1225
1226         return 0;
1227 }
1228
1229 static int
1230 sfc_flow_parse_queue(struct sfc_adapter *sa,
1231                      const struct rte_flow_action_queue *queue,
1232                      struct rte_flow *flow)
1233 {
1234         struct sfc_rxq *rxq;
1235
1236         if (queue->index >= sa->rxq_count)
1237                 return -EINVAL;
1238
1239         rxq = sa->rxq_info[queue->index].rxq;
1240         flow->spec.template.efs_dmaq_id = (uint16_t)rxq->hw_index;
1241
1242         return 0;
1243 }
1244
1245 #if EFSYS_OPT_RX_SCALE
1246 static int
1247 sfc_flow_parse_rss(struct sfc_adapter *sa,
1248                    const struct rte_flow_action_rss *rss,
1249                    struct rte_flow *flow)
1250 {
1251         unsigned int rxq_sw_index;
1252         struct sfc_rxq *rxq;
1253         unsigned int rxq_hw_index_min;
1254         unsigned int rxq_hw_index_max;
1255         const uint8_t *rss_key;
1256         struct sfc_flow_rss *sfc_rss_conf = &flow->rss_conf;
1257         unsigned int i;
1258
1259         if (rss->queue_num == 0)
1260                 return -EINVAL;
1261
1262         rxq_sw_index = sa->rxq_count - 1;
1263         rxq = sa->rxq_info[rxq_sw_index].rxq;
1264         rxq_hw_index_min = rxq->hw_index;
1265         rxq_hw_index_max = 0;
1266
1267         for (i = 0; i < rss->queue_num; ++i) {
1268                 rxq_sw_index = rss->queue[i];
1269
1270                 if (rxq_sw_index >= sa->rxq_count)
1271                         return -EINVAL;
1272
1273                 rxq = sa->rxq_info[rxq_sw_index].rxq;
1274
1275                 if (rxq->hw_index < rxq_hw_index_min)
1276                         rxq_hw_index_min = rxq->hw_index;
1277
1278                 if (rxq->hw_index > rxq_hw_index_max)
1279                         rxq_hw_index_max = rxq->hw_index;
1280         }
1281
1282         switch (rss->func) {
1283         case RTE_ETH_HASH_FUNCTION_DEFAULT:
1284         case RTE_ETH_HASH_FUNCTION_TOEPLITZ:
1285                 break;
1286         default:
1287                 return -EINVAL;
1288         }
1289
1290         if (rss->level)
1291                 return -EINVAL;
1292
1293         if ((rss->types & ~SFC_RSS_OFFLOADS) != 0)
1294                 return -EINVAL;
1295
1296         if (rss->key_len) {
1297                 if (rss->key_len != sizeof(sa->rss_key))
1298                         return -EINVAL;
1299
1300                 rss_key = rss->key;
1301         } else {
1302                 rss_key = sa->rss_key;
1303         }
1304
1305         flow->rss = B_TRUE;
1306
1307         sfc_rss_conf->rxq_hw_index_min = rxq_hw_index_min;
1308         sfc_rss_conf->rxq_hw_index_max = rxq_hw_index_max;
1309         sfc_rss_conf->rss_hash_types = sfc_rte_to_efx_hash_type(rss->types);
1310         rte_memcpy(sfc_rss_conf->rss_key, rss_key, sizeof(sa->rss_key));
1311
1312         for (i = 0; i < RTE_DIM(sfc_rss_conf->rss_tbl); ++i) {
1313                 unsigned int rxq_sw_index = rss->queue[i % rss->queue_num];
1314                 struct sfc_rxq *rxq = sa->rxq_info[rxq_sw_index].rxq;
1315
1316                 sfc_rss_conf->rss_tbl[i] = rxq->hw_index - rxq_hw_index_min;
1317         }
1318
1319         return 0;
1320 }
1321 #endif /* EFSYS_OPT_RX_SCALE */
1322
1323 static int
1324 sfc_flow_spec_flush(struct sfc_adapter *sa, struct sfc_flow_spec *spec,
1325                     unsigned int filters_count)
1326 {
1327         unsigned int i;
1328         int ret = 0;
1329
1330         for (i = 0; i < filters_count; i++) {
1331                 int rc;
1332
1333                 rc = efx_filter_remove(sa->nic, &spec->filters[i]);
1334                 if (ret == 0 && rc != 0) {
1335                         sfc_err(sa, "failed to remove filter specification "
1336                                 "(rc = %d)", rc);
1337                         ret = rc;
1338                 }
1339         }
1340
1341         return ret;
1342 }
1343
1344 static int
1345 sfc_flow_spec_insert(struct sfc_adapter *sa, struct sfc_flow_spec *spec)
1346 {
1347         unsigned int i;
1348         int rc = 0;
1349
1350         for (i = 0; i < spec->count; i++) {
1351                 rc = efx_filter_insert(sa->nic, &spec->filters[i]);
1352                 if (rc != 0) {
1353                         sfc_flow_spec_flush(sa, spec, i);
1354                         break;
1355                 }
1356         }
1357
1358         return rc;
1359 }
1360
1361 static int
1362 sfc_flow_spec_remove(struct sfc_adapter *sa, struct sfc_flow_spec *spec)
1363 {
1364         return sfc_flow_spec_flush(sa, spec, spec->count);
1365 }
1366
1367 static int
1368 sfc_flow_filter_insert(struct sfc_adapter *sa,
1369                        struct rte_flow *flow)
1370 {
1371 #if EFSYS_OPT_RX_SCALE
1372         struct sfc_flow_rss *rss = &flow->rss_conf;
1373         uint32_t efs_rss_context = EFX_RSS_CONTEXT_DEFAULT;
1374         unsigned int i;
1375         int rc = 0;
1376
1377         if (flow->rss) {
1378                 unsigned int rss_spread = MIN(rss->rxq_hw_index_max -
1379                                               rss->rxq_hw_index_min + 1,
1380                                               EFX_MAXRSS);
1381
1382                 rc = efx_rx_scale_context_alloc(sa->nic,
1383                                                 EFX_RX_SCALE_EXCLUSIVE,
1384                                                 rss_spread,
1385                                                 &efs_rss_context);
1386                 if (rc != 0)
1387                         goto fail_scale_context_alloc;
1388
1389                 rc = efx_rx_scale_mode_set(sa->nic, efs_rss_context,
1390                                            EFX_RX_HASHALG_TOEPLITZ,
1391                                            rss->rss_hash_types, B_TRUE);
1392                 if (rc != 0)
1393                         goto fail_scale_mode_set;
1394
1395                 rc = efx_rx_scale_key_set(sa->nic, efs_rss_context,
1396                                           rss->rss_key,
1397                                           sizeof(sa->rss_key));
1398                 if (rc != 0)
1399                         goto fail_scale_key_set;
1400
1401                 /*
1402                  * At this point, fully elaborated filter specifications
1403                  * have been produced from the template. To make sure that
1404                  * RSS behaviour is consistent between them, set the same
1405                  * RSS context value everywhere.
1406                  */
1407                 for (i = 0; i < flow->spec.count; i++) {
1408                         efx_filter_spec_t *spec = &flow->spec.filters[i];
1409
1410                         spec->efs_rss_context = efs_rss_context;
1411                         spec->efs_dmaq_id = rss->rxq_hw_index_min;
1412                         spec->efs_flags |= EFX_FILTER_FLAG_RX_RSS;
1413                 }
1414         }
1415
1416         rc = sfc_flow_spec_insert(sa, &flow->spec);
1417         if (rc != 0)
1418                 goto fail_filter_insert;
1419
1420         if (flow->rss) {
1421                 /*
1422                  * Scale table is set after filter insertion because
1423                  * the table entries are relative to the base RxQ ID
1424                  * and the latter is submitted to the HW by means of
1425                  * inserting a filter, so by the time of the request
1426                  * the HW knows all the information needed to verify
1427                  * the table entries, and the operation will succeed
1428                  */
1429                 rc = efx_rx_scale_tbl_set(sa->nic, efs_rss_context,
1430                                           rss->rss_tbl, RTE_DIM(rss->rss_tbl));
1431                 if (rc != 0)
1432                         goto fail_scale_tbl_set;
1433         }
1434
1435         return 0;
1436
1437 fail_scale_tbl_set:
1438         sfc_flow_spec_remove(sa, &flow->spec);
1439
1440 fail_filter_insert:
1441 fail_scale_key_set:
1442 fail_scale_mode_set:
1443         if (efs_rss_context != EFX_RSS_CONTEXT_DEFAULT)
1444                 efx_rx_scale_context_free(sa->nic, efs_rss_context);
1445
1446 fail_scale_context_alloc:
1447         return rc;
1448 #else /* !EFSYS_OPT_RX_SCALE */
1449         return sfc_flow_spec_insert(sa, &flow->spec);
1450 #endif /* EFSYS_OPT_RX_SCALE */
1451 }
1452
1453 static int
1454 sfc_flow_filter_remove(struct sfc_adapter *sa,
1455                        struct rte_flow *flow)
1456 {
1457         int rc = 0;
1458
1459         rc = sfc_flow_spec_remove(sa, &flow->spec);
1460         if (rc != 0)
1461                 return rc;
1462
1463 #if EFSYS_OPT_RX_SCALE
1464         if (flow->rss) {
1465                 /*
1466                  * All specifications for a given flow rule have the same RSS
1467                  * context, so that RSS context value is taken from the first
1468                  * filter specification
1469                  */
1470                 efx_filter_spec_t *spec = &flow->spec.filters[0];
1471
1472                 rc = efx_rx_scale_context_free(sa->nic, spec->efs_rss_context);
1473         }
1474 #endif /* EFSYS_OPT_RX_SCALE */
1475
1476         return rc;
1477 }
1478
1479 static int
1480 sfc_flow_parse_actions(struct sfc_adapter *sa,
1481                        const struct rte_flow_action actions[],
1482                        struct rte_flow *flow,
1483                        struct rte_flow_error *error)
1484 {
1485         int rc;
1486         boolean_t is_specified = B_FALSE;
1487
1488         if (actions == NULL) {
1489                 rte_flow_error_set(error, EINVAL,
1490                                    RTE_FLOW_ERROR_TYPE_ACTION_NUM, NULL,
1491                                    "NULL actions");
1492                 return -rte_errno;
1493         }
1494
1495         for (; actions->type != RTE_FLOW_ACTION_TYPE_END; actions++) {
1496                 /* This one may appear anywhere multiple times. */
1497                 if (actions->type == RTE_FLOW_ACTION_TYPE_VOID)
1498                         continue;
1499                 /* Fate-deciding actions may appear exactly once. */
1500                 if (is_specified) {
1501                         rte_flow_error_set
1502                                 (error, ENOTSUP, RTE_FLOW_ERROR_TYPE_ACTION,
1503                                  actions,
1504                                  "Cannot combine several fate-deciding actions,"
1505                                  "choose between QUEUE, RSS or DROP");
1506                         return -rte_errno;
1507                 }
1508                 switch (actions->type) {
1509                 case RTE_FLOW_ACTION_TYPE_QUEUE:
1510                         rc = sfc_flow_parse_queue(sa, actions->conf, flow);
1511                         if (rc != 0) {
1512                                 rte_flow_error_set(error, EINVAL,
1513                                         RTE_FLOW_ERROR_TYPE_ACTION, actions,
1514                                         "Bad QUEUE action");
1515                                 return -rte_errno;
1516                         }
1517
1518                         is_specified = B_TRUE;
1519                         break;
1520
1521 #if EFSYS_OPT_RX_SCALE
1522                 case RTE_FLOW_ACTION_TYPE_RSS:
1523                         rc = sfc_flow_parse_rss(sa, actions->conf, flow);
1524                         if (rc != 0) {
1525                                 rte_flow_error_set(error, rc,
1526                                         RTE_FLOW_ERROR_TYPE_ACTION, actions,
1527                                         "Bad RSS action");
1528                                 return -rte_errno;
1529                         }
1530
1531                         is_specified = B_TRUE;
1532                         break;
1533 #endif /* EFSYS_OPT_RX_SCALE */
1534
1535                 case RTE_FLOW_ACTION_TYPE_DROP:
1536                         flow->spec.template.efs_dmaq_id =
1537                                 EFX_FILTER_SPEC_RX_DMAQ_ID_DROP;
1538
1539                         is_specified = B_TRUE;
1540                         break;
1541
1542                 default:
1543                         rte_flow_error_set(error, ENOTSUP,
1544                                            RTE_FLOW_ERROR_TYPE_ACTION, actions,
1545                                            "Action is not supported");
1546                         return -rte_errno;
1547                 }
1548         }
1549
1550         /* When fate is unknown, drop traffic. */
1551         if (!is_specified) {
1552                 flow->spec.template.efs_dmaq_id =
1553                         EFX_FILTER_SPEC_RX_DMAQ_ID_DROP;
1554         }
1555
1556         return 0;
1557 }
1558
1559 /**
1560  * Set the EFX_FILTER_MATCH_UNKNOWN_UCAST_DST
1561  * and EFX_FILTER_MATCH_UNKNOWN_MCAST_DST match flags in the same
1562  * specifications after copying.
1563  *
1564  * @param spec[in, out]
1565  *   SFC flow specification to update.
1566  * @param filters_count_for_one_val[in]
1567  *   How many specifications should have the same match flag, what is the
1568  *   number of specifications before copying.
1569  * @param error[out]
1570  *   Perform verbose error reporting if not NULL.
1571  */
1572 static int
1573 sfc_flow_set_unknown_dst_flags(struct sfc_flow_spec *spec,
1574                                unsigned int filters_count_for_one_val,
1575                                struct rte_flow_error *error)
1576 {
1577         unsigned int i;
1578         static const efx_filter_match_flags_t vals[] = {
1579                 EFX_FILTER_MATCH_UNKNOWN_UCAST_DST,
1580                 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST
1581         };
1582
1583         if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1584                 rte_flow_error_set(error, EINVAL,
1585                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1586                         "Number of specifications is incorrect while copying "
1587                         "by unknown destination flags");
1588                 return -rte_errno;
1589         }
1590
1591         for (i = 0; i < spec->count; i++) {
1592                 /* The check above ensures that divisor can't be zero here */
1593                 spec->filters[i].efs_match_flags |=
1594                         vals[i / filters_count_for_one_val];
1595         }
1596
1597         return 0;
1598 }
1599
1600 /**
1601  * Check that the following conditions are met:
1602  * - the list of supported filters has a filter
1603  *   with EFX_FILTER_MATCH_UNKNOWN_MCAST_DST flag instead of
1604  *   EFX_FILTER_MATCH_UNKNOWN_UCAST_DST, since this filter will also
1605  *   be inserted.
1606  *
1607  * @param match[in]
1608  *   The match flags of filter.
1609  * @param spec[in]
1610  *   Specification to be supplemented.
1611  * @param filter[in]
1612  *   SFC filter with list of supported filters.
1613  */
1614 static boolean_t
1615 sfc_flow_check_unknown_dst_flags(efx_filter_match_flags_t match,
1616                                  __rte_unused efx_filter_spec_t *spec,
1617                                  struct sfc_filter *filter)
1618 {
1619         unsigned int i;
1620         efx_filter_match_flags_t match_mcast_dst;
1621
1622         match_mcast_dst =
1623                 (match & ~EFX_FILTER_MATCH_UNKNOWN_UCAST_DST) |
1624                 EFX_FILTER_MATCH_UNKNOWN_MCAST_DST;
1625         for (i = 0; i < filter->supported_match_num; i++) {
1626                 if (match_mcast_dst == filter->supported_match[i])
1627                         return B_TRUE;
1628         }
1629
1630         return B_FALSE;
1631 }
1632
1633 /**
1634  * Set the EFX_FILTER_MATCH_ETHER_TYPE match flag and EFX_ETHER_TYPE_IPV4 and
1635  * EFX_ETHER_TYPE_IPV6 values of the corresponding field in the same
1636  * specifications after copying.
1637  *
1638  * @param spec[in, out]
1639  *   SFC flow specification to update.
1640  * @param filters_count_for_one_val[in]
1641  *   How many specifications should have the same EtherType value, what is the
1642  *   number of specifications before copying.
1643  * @param error[out]
1644  *   Perform verbose error reporting if not NULL.
1645  */
1646 static int
1647 sfc_flow_set_ethertypes(struct sfc_flow_spec *spec,
1648                         unsigned int filters_count_for_one_val,
1649                         struct rte_flow_error *error)
1650 {
1651         unsigned int i;
1652         static const uint16_t vals[] = {
1653                 EFX_ETHER_TYPE_IPV4, EFX_ETHER_TYPE_IPV6
1654         };
1655
1656         if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1657                 rte_flow_error_set(error, EINVAL,
1658                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1659                         "Number of specifications is incorrect "
1660                         "while copying by Ethertype");
1661                 return -rte_errno;
1662         }
1663
1664         for (i = 0; i < spec->count; i++) {
1665                 spec->filters[i].efs_match_flags |=
1666                         EFX_FILTER_MATCH_ETHER_TYPE;
1667
1668                 /*
1669                  * The check above ensures that
1670                  * filters_count_for_one_val is not 0
1671                  */
1672                 spec->filters[i].efs_ether_type =
1673                         vals[i / filters_count_for_one_val];
1674         }
1675
1676         return 0;
1677 }
1678
1679 /**
1680  * Set the EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST and
1681  * EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST match flags in the same
1682  * specifications after copying.
1683  *
1684  * @param spec[in, out]
1685  *   SFC flow specification to update.
1686  * @param filters_count_for_one_val[in]
1687  *   How many specifications should have the same match flag, what is the
1688  *   number of specifications before copying.
1689  * @param error[out]
1690  *   Perform verbose error reporting if not NULL.
1691  */
1692 static int
1693 sfc_flow_set_ifrm_unknown_dst_flags(struct sfc_flow_spec *spec,
1694                                     unsigned int filters_count_for_one_val,
1695                                     struct rte_flow_error *error)
1696 {
1697         unsigned int i;
1698         static const efx_filter_match_flags_t vals[] = {
1699                 EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST,
1700                 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST
1701         };
1702
1703         if (filters_count_for_one_val * RTE_DIM(vals) != spec->count) {
1704                 rte_flow_error_set(error, EINVAL,
1705                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1706                         "Number of specifications is incorrect while copying "
1707                         "by inner frame unknown destination flags");
1708                 return -rte_errno;
1709         }
1710
1711         for (i = 0; i < spec->count; i++) {
1712                 /* The check above ensures that divisor can't be zero here */
1713                 spec->filters[i].efs_match_flags |=
1714                         vals[i / filters_count_for_one_val];
1715         }
1716
1717         return 0;
1718 }
1719
1720 /**
1721  * Check that the following conditions are met:
1722  * - the specification corresponds to a filter for encapsulated traffic
1723  * - the list of supported filters has a filter
1724  *   with EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST flag instead of
1725  *   EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST, since this filter will also
1726  *   be inserted.
1727  *
1728  * @param match[in]
1729  *   The match flags of filter.
1730  * @param spec[in]
1731  *   Specification to be supplemented.
1732  * @param filter[in]
1733  *   SFC filter with list of supported filters.
1734  */
1735 static boolean_t
1736 sfc_flow_check_ifrm_unknown_dst_flags(efx_filter_match_flags_t match,
1737                                       efx_filter_spec_t *spec,
1738                                       struct sfc_filter *filter)
1739 {
1740         unsigned int i;
1741         efx_tunnel_protocol_t encap_type = spec->efs_encap_type;
1742         efx_filter_match_flags_t match_mcast_dst;
1743
1744         if (encap_type == EFX_TUNNEL_PROTOCOL_NONE)
1745                 return B_FALSE;
1746
1747         match_mcast_dst =
1748                 (match & ~EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST) |
1749                 EFX_FILTER_MATCH_IFRM_UNKNOWN_MCAST_DST;
1750         for (i = 0; i < filter->supported_match_num; i++) {
1751                 if (match_mcast_dst == filter->supported_match[i])
1752                         return B_TRUE;
1753         }
1754
1755         return B_FALSE;
1756 }
1757
1758 /*
1759  * Match flags that can be automatically added to filters.
1760  * Selecting the last minimum when searching for the copy flag ensures that the
1761  * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST flag has a higher priority than
1762  * EFX_FILTER_MATCH_ETHER_TYPE. This is because the filter
1763  * EFX_FILTER_MATCH_UNKNOWN_UCAST_DST is at the end of the list of supported
1764  * filters.
1765  */
1766 static const struct sfc_flow_copy_flag sfc_flow_copy_flags[] = {
1767         {
1768                 .flag = EFX_FILTER_MATCH_UNKNOWN_UCAST_DST,
1769                 .vals_count = 2,
1770                 .set_vals = sfc_flow_set_unknown_dst_flags,
1771                 .spec_check = sfc_flow_check_unknown_dst_flags,
1772         },
1773         {
1774                 .flag = EFX_FILTER_MATCH_ETHER_TYPE,
1775                 .vals_count = 2,
1776                 .set_vals = sfc_flow_set_ethertypes,
1777                 .spec_check = NULL,
1778         },
1779         {
1780                 .flag = EFX_FILTER_MATCH_IFRM_UNKNOWN_UCAST_DST,
1781                 .vals_count = 2,
1782                 .set_vals = sfc_flow_set_ifrm_unknown_dst_flags,
1783                 .spec_check = sfc_flow_check_ifrm_unknown_dst_flags,
1784         },
1785 };
1786
1787 /* Get item from array sfc_flow_copy_flags */
1788 static const struct sfc_flow_copy_flag *
1789 sfc_flow_get_copy_flag(efx_filter_match_flags_t flag)
1790 {
1791         unsigned int i;
1792
1793         for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1794                 if (sfc_flow_copy_flags[i].flag == flag)
1795                         return &sfc_flow_copy_flags[i];
1796         }
1797
1798         return NULL;
1799 }
1800
1801 /**
1802  * Make copies of the specifications, set match flag and values
1803  * of the field that corresponds to it.
1804  *
1805  * @param spec[in, out]
1806  *   SFC flow specification to update.
1807  * @param flag[in]
1808  *   The match flag to add.
1809  * @param error[out]
1810  *   Perform verbose error reporting if not NULL.
1811  */
1812 static int
1813 sfc_flow_spec_add_match_flag(struct sfc_flow_spec *spec,
1814                              efx_filter_match_flags_t flag,
1815                              struct rte_flow_error *error)
1816 {
1817         unsigned int i;
1818         unsigned int new_filters_count;
1819         unsigned int filters_count_for_one_val;
1820         const struct sfc_flow_copy_flag *copy_flag;
1821         int rc;
1822
1823         copy_flag = sfc_flow_get_copy_flag(flag);
1824         if (copy_flag == NULL) {
1825                 rte_flow_error_set(error, ENOTSUP,
1826                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1827                                    "Unsupported spec field for copying");
1828                 return -rte_errno;
1829         }
1830
1831         new_filters_count = spec->count * copy_flag->vals_count;
1832         if (new_filters_count > SF_FLOW_SPEC_NB_FILTERS_MAX) {
1833                 rte_flow_error_set(error, EINVAL,
1834                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1835                         "Too much EFX specifications in the flow rule");
1836                 return -rte_errno;
1837         }
1838
1839         /* Copy filters specifications */
1840         for (i = spec->count; i < new_filters_count; i++)
1841                 spec->filters[i] = spec->filters[i - spec->count];
1842
1843         filters_count_for_one_val = spec->count;
1844         spec->count = new_filters_count;
1845
1846         rc = copy_flag->set_vals(spec, filters_count_for_one_val, error);
1847         if (rc != 0)
1848                 return rc;
1849
1850         return 0;
1851 }
1852
1853 /**
1854  * Check that the given set of match flags missing in the original filter spec
1855  * could be covered by adding spec copies which specify the corresponding
1856  * flags and packet field values to match.
1857  *
1858  * @param miss_flags[in]
1859  *   Flags that are missing until the supported filter.
1860  * @param spec[in]
1861  *   Specification to be supplemented.
1862  * @param filter[in]
1863  *   SFC filter.
1864  *
1865  * @return
1866  *   Number of specifications after copy or 0, if the flags can not be added.
1867  */
1868 static unsigned int
1869 sfc_flow_check_missing_flags(efx_filter_match_flags_t miss_flags,
1870                              efx_filter_spec_t *spec,
1871                              struct sfc_filter *filter)
1872 {
1873         unsigned int i;
1874         efx_filter_match_flags_t copy_flags = 0;
1875         efx_filter_match_flags_t flag;
1876         efx_filter_match_flags_t match = spec->efs_match_flags | miss_flags;
1877         sfc_flow_spec_check *check;
1878         unsigned int multiplier = 1;
1879
1880         for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1881                 flag = sfc_flow_copy_flags[i].flag;
1882                 check = sfc_flow_copy_flags[i].spec_check;
1883                 if ((flag & miss_flags) == flag) {
1884                         if (check != NULL && (!check(match, spec, filter)))
1885                                 continue;
1886
1887                         copy_flags |= flag;
1888                         multiplier *= sfc_flow_copy_flags[i].vals_count;
1889                 }
1890         }
1891
1892         if (copy_flags == miss_flags)
1893                 return multiplier;
1894
1895         return 0;
1896 }
1897
1898 /**
1899  * Attempt to supplement the specification template to the minimally
1900  * supported set of match flags. To do this, it is necessary to copy
1901  * the specifications, filling them with the values of fields that
1902  * correspond to the missing flags.
1903  * The necessary and sufficient filter is built from the fewest number
1904  * of copies which could be made to cover the minimally required set
1905  * of flags.
1906  *
1907  * @param sa[in]
1908  *   SFC adapter.
1909  * @param spec[in, out]
1910  *   SFC flow specification to update.
1911  * @param error[out]
1912  *   Perform verbose error reporting if not NULL.
1913  */
1914 static int
1915 sfc_flow_spec_filters_complete(struct sfc_adapter *sa,
1916                                struct sfc_flow_spec *spec,
1917                                struct rte_flow_error *error)
1918 {
1919         struct sfc_filter *filter = &sa->filter;
1920         efx_filter_match_flags_t miss_flags;
1921         efx_filter_match_flags_t min_miss_flags = 0;
1922         efx_filter_match_flags_t match;
1923         unsigned int min_multiplier = UINT_MAX;
1924         unsigned int multiplier;
1925         unsigned int i;
1926         int rc;
1927
1928         match = spec->template.efs_match_flags;
1929         for (i = 0; i < filter->supported_match_num; i++) {
1930                 if ((match & filter->supported_match[i]) == match) {
1931                         miss_flags = filter->supported_match[i] & (~match);
1932                         multiplier = sfc_flow_check_missing_flags(miss_flags,
1933                                 &spec->template, filter);
1934                         if (multiplier > 0) {
1935                                 if (multiplier <= min_multiplier) {
1936                                         min_multiplier = multiplier;
1937                                         min_miss_flags = miss_flags;
1938                                 }
1939                         }
1940                 }
1941         }
1942
1943         if (min_multiplier == UINT_MAX) {
1944                 rte_flow_error_set(error, ENOTSUP,
1945                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
1946                                    "Flow rule pattern is not supported");
1947                 return -rte_errno;
1948         }
1949
1950         for (i = 0; i < RTE_DIM(sfc_flow_copy_flags); i++) {
1951                 efx_filter_match_flags_t flag = sfc_flow_copy_flags[i].flag;
1952
1953                 if ((flag & min_miss_flags) == flag) {
1954                         rc = sfc_flow_spec_add_match_flag(spec, flag, error);
1955                         if (rc != 0)
1956                                 return rc;
1957                 }
1958         }
1959
1960         return 0;
1961 }
1962
1963 /**
1964  * Check that set of match flags is referred to by a filter. Filter is
1965  * described by match flags with the ability to add OUTER_VID and INNER_VID
1966  * flags.
1967  *
1968  * @param match_flags[in]
1969  *   Set of match flags.
1970  * @param flags_pattern[in]
1971  *   Pattern of filter match flags.
1972  */
1973 static boolean_t
1974 sfc_flow_is_match_with_vids(efx_filter_match_flags_t match_flags,
1975                             efx_filter_match_flags_t flags_pattern)
1976 {
1977         if ((match_flags & flags_pattern) != flags_pattern)
1978                 return B_FALSE;
1979
1980         switch (match_flags & ~flags_pattern) {
1981         case 0:
1982         case EFX_FILTER_MATCH_OUTER_VID:
1983         case EFX_FILTER_MATCH_OUTER_VID | EFX_FILTER_MATCH_INNER_VID:
1984                 return B_TRUE;
1985         default:
1986                 return B_FALSE;
1987         }
1988 }
1989
1990 /**
1991  * Check whether the spec maps to a hardware filter which is known to be
1992  * ineffective despite being valid.
1993  *
1994  * @param spec[in]
1995  *   SFC flow specification.
1996  */
1997 static boolean_t
1998 sfc_flow_is_match_flags_exception(struct sfc_flow_spec *spec)
1999 {
2000         unsigned int i;
2001         uint16_t ether_type;
2002         uint8_t ip_proto;
2003         efx_filter_match_flags_t match_flags;
2004
2005         for (i = 0; i < spec->count; i++) {
2006                 match_flags = spec->filters[i].efs_match_flags;
2007
2008                 if (sfc_flow_is_match_with_vids(match_flags,
2009                                                 EFX_FILTER_MATCH_ETHER_TYPE) ||
2010                     sfc_flow_is_match_with_vids(match_flags,
2011                                                 EFX_FILTER_MATCH_ETHER_TYPE |
2012                                                 EFX_FILTER_MATCH_LOC_MAC)) {
2013                         ether_type = spec->filters[i].efs_ether_type;
2014                         if (ether_type == EFX_ETHER_TYPE_IPV4 ||
2015                             ether_type == EFX_ETHER_TYPE_IPV6)
2016                                 return B_TRUE;
2017                 } else if (sfc_flow_is_match_with_vids(match_flags,
2018                                 EFX_FILTER_MATCH_ETHER_TYPE |
2019                                 EFX_FILTER_MATCH_IP_PROTO) ||
2020                            sfc_flow_is_match_with_vids(match_flags,
2021                                 EFX_FILTER_MATCH_ETHER_TYPE |
2022                                 EFX_FILTER_MATCH_IP_PROTO |
2023                                 EFX_FILTER_MATCH_LOC_MAC)) {
2024                         ip_proto = spec->filters[i].efs_ip_proto;
2025                         if (ip_proto == EFX_IPPROTO_TCP ||
2026                             ip_proto == EFX_IPPROTO_UDP)
2027                                 return B_TRUE;
2028                 }
2029         }
2030
2031         return B_FALSE;
2032 }
2033
2034 static int
2035 sfc_flow_validate_match_flags(struct sfc_adapter *sa,
2036                               struct rte_flow *flow,
2037                               struct rte_flow_error *error)
2038 {
2039         efx_filter_spec_t *spec_tmpl = &flow->spec.template;
2040         efx_filter_match_flags_t match_flags = spec_tmpl->efs_match_flags;
2041         int rc;
2042
2043         /* Initialize the first filter spec with template */
2044         flow->spec.filters[0] = *spec_tmpl;
2045         flow->spec.count = 1;
2046
2047         if (!sfc_filter_is_match_supported(sa, match_flags)) {
2048                 rc = sfc_flow_spec_filters_complete(sa, &flow->spec, error);
2049                 if (rc != 0)
2050                         return rc;
2051         }
2052
2053         if (sfc_flow_is_match_flags_exception(&flow->spec)) {
2054                 rte_flow_error_set(error, ENOTSUP,
2055                         RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2056                         "The flow rule pattern is unsupported");
2057                 return -rte_errno;
2058         }
2059
2060         return 0;
2061 }
2062
2063 static int
2064 sfc_flow_parse(struct rte_eth_dev *dev,
2065                const struct rte_flow_attr *attr,
2066                const struct rte_flow_item pattern[],
2067                const struct rte_flow_action actions[],
2068                struct rte_flow *flow,
2069                struct rte_flow_error *error)
2070 {
2071         struct sfc_adapter *sa = dev->data->dev_private;
2072         int rc;
2073
2074         rc = sfc_flow_parse_attr(attr, flow, error);
2075         if (rc != 0)
2076                 goto fail_bad_value;
2077
2078         rc = sfc_flow_parse_pattern(pattern, flow, error);
2079         if (rc != 0)
2080                 goto fail_bad_value;
2081
2082         rc = sfc_flow_parse_actions(sa, actions, flow, error);
2083         if (rc != 0)
2084                 goto fail_bad_value;
2085
2086         rc = sfc_flow_validate_match_flags(sa, flow, error);
2087         if (rc != 0)
2088                 goto fail_bad_value;
2089
2090         return 0;
2091
2092 fail_bad_value:
2093         return rc;
2094 }
2095
2096 static int
2097 sfc_flow_validate(struct rte_eth_dev *dev,
2098                   const struct rte_flow_attr *attr,
2099                   const struct rte_flow_item pattern[],
2100                   const struct rte_flow_action actions[],
2101                   struct rte_flow_error *error)
2102 {
2103         struct rte_flow flow;
2104
2105         memset(&flow, 0, sizeof(flow));
2106
2107         return sfc_flow_parse(dev, attr, pattern, actions, &flow, error);
2108 }
2109
2110 static struct rte_flow *
2111 sfc_flow_create(struct rte_eth_dev *dev,
2112                 const struct rte_flow_attr *attr,
2113                 const struct rte_flow_item pattern[],
2114                 const struct rte_flow_action actions[],
2115                 struct rte_flow_error *error)
2116 {
2117         struct sfc_adapter *sa = dev->data->dev_private;
2118         struct rte_flow *flow = NULL;
2119         int rc;
2120
2121         flow = rte_zmalloc("sfc_rte_flow", sizeof(*flow), 0);
2122         if (flow == NULL) {
2123                 rte_flow_error_set(error, ENOMEM,
2124                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2125                                    "Failed to allocate memory");
2126                 goto fail_no_mem;
2127         }
2128
2129         rc = sfc_flow_parse(dev, attr, pattern, actions, flow, error);
2130         if (rc != 0)
2131                 goto fail_bad_value;
2132
2133         TAILQ_INSERT_TAIL(&sa->filter.flow_list, flow, entries);
2134
2135         sfc_adapter_lock(sa);
2136
2137         if (sa->state == SFC_ADAPTER_STARTED) {
2138                 rc = sfc_flow_filter_insert(sa, flow);
2139                 if (rc != 0) {
2140                         rte_flow_error_set(error, rc,
2141                                 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2142                                 "Failed to insert filter");
2143                         goto fail_filter_insert;
2144                 }
2145         }
2146
2147         sfc_adapter_unlock(sa);
2148
2149         return flow;
2150
2151 fail_filter_insert:
2152         TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2153
2154 fail_bad_value:
2155         rte_free(flow);
2156         sfc_adapter_unlock(sa);
2157
2158 fail_no_mem:
2159         return NULL;
2160 }
2161
2162 static int
2163 sfc_flow_remove(struct sfc_adapter *sa,
2164                 struct rte_flow *flow,
2165                 struct rte_flow_error *error)
2166 {
2167         int rc = 0;
2168
2169         SFC_ASSERT(sfc_adapter_is_locked(sa));
2170
2171         if (sa->state == SFC_ADAPTER_STARTED) {
2172                 rc = sfc_flow_filter_remove(sa, flow);
2173                 if (rc != 0)
2174                         rte_flow_error_set(error, rc,
2175                                 RTE_FLOW_ERROR_TYPE_UNSPECIFIED, NULL,
2176                                 "Failed to destroy flow rule");
2177         }
2178
2179         TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2180         rte_free(flow);
2181
2182         return rc;
2183 }
2184
2185 static int
2186 sfc_flow_destroy(struct rte_eth_dev *dev,
2187                  struct rte_flow *flow,
2188                  struct rte_flow_error *error)
2189 {
2190         struct sfc_adapter *sa = dev->data->dev_private;
2191         struct rte_flow *flow_ptr;
2192         int rc = EINVAL;
2193
2194         sfc_adapter_lock(sa);
2195
2196         TAILQ_FOREACH(flow_ptr, &sa->filter.flow_list, entries) {
2197                 if (flow_ptr == flow)
2198                         rc = 0;
2199         }
2200         if (rc != 0) {
2201                 rte_flow_error_set(error, rc,
2202                                    RTE_FLOW_ERROR_TYPE_HANDLE, NULL,
2203                                    "Failed to find flow rule to destroy");
2204                 goto fail_bad_value;
2205         }
2206
2207         rc = sfc_flow_remove(sa, flow, error);
2208
2209 fail_bad_value:
2210         sfc_adapter_unlock(sa);
2211
2212         return -rc;
2213 }
2214
2215 static int
2216 sfc_flow_flush(struct rte_eth_dev *dev,
2217                struct rte_flow_error *error)
2218 {
2219         struct sfc_adapter *sa = dev->data->dev_private;
2220         struct rte_flow *flow;
2221         int rc = 0;
2222         int ret = 0;
2223
2224         sfc_adapter_lock(sa);
2225
2226         while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
2227                 rc = sfc_flow_remove(sa, flow, error);
2228                 if (rc != 0)
2229                         ret = rc;
2230         }
2231
2232         sfc_adapter_unlock(sa);
2233
2234         return -ret;
2235 }
2236
2237 static int
2238 sfc_flow_isolate(struct rte_eth_dev *dev, int enable,
2239                  struct rte_flow_error *error)
2240 {
2241         struct sfc_adapter *sa = dev->data->dev_private;
2242         struct sfc_port *port = &sa->port;
2243         int ret = 0;
2244
2245         sfc_adapter_lock(sa);
2246         if (sa->state != SFC_ADAPTER_INITIALIZED) {
2247                 rte_flow_error_set(error, EBUSY,
2248                                    RTE_FLOW_ERROR_TYPE_UNSPECIFIED,
2249                                    NULL, "please close the port first");
2250                 ret = -rte_errno;
2251         } else {
2252                 port->isolated = (enable) ? B_TRUE : B_FALSE;
2253         }
2254         sfc_adapter_unlock(sa);
2255
2256         return ret;
2257 }
2258
2259 const struct rte_flow_ops sfc_flow_ops = {
2260         .validate = sfc_flow_validate,
2261         .create = sfc_flow_create,
2262         .destroy = sfc_flow_destroy,
2263         .flush = sfc_flow_flush,
2264         .query = NULL,
2265         .isolate = sfc_flow_isolate,
2266 };
2267
2268 void
2269 sfc_flow_init(struct sfc_adapter *sa)
2270 {
2271         SFC_ASSERT(sfc_adapter_is_locked(sa));
2272
2273         TAILQ_INIT(&sa->filter.flow_list);
2274 }
2275
2276 void
2277 sfc_flow_fini(struct sfc_adapter *sa)
2278 {
2279         struct rte_flow *flow;
2280
2281         SFC_ASSERT(sfc_adapter_is_locked(sa));
2282
2283         while ((flow = TAILQ_FIRST(&sa->filter.flow_list)) != NULL) {
2284                 TAILQ_REMOVE(&sa->filter.flow_list, flow, entries);
2285                 rte_free(flow);
2286         }
2287 }
2288
2289 void
2290 sfc_flow_stop(struct sfc_adapter *sa)
2291 {
2292         struct rte_flow *flow;
2293
2294         SFC_ASSERT(sfc_adapter_is_locked(sa));
2295
2296         TAILQ_FOREACH(flow, &sa->filter.flow_list, entries)
2297                 sfc_flow_filter_remove(sa, flow);
2298 }
2299
2300 int
2301 sfc_flow_start(struct sfc_adapter *sa)
2302 {
2303         struct rte_flow *flow;
2304         int rc = 0;
2305
2306         sfc_log_init(sa, "entry");
2307
2308         SFC_ASSERT(sfc_adapter_is_locked(sa));
2309
2310         TAILQ_FOREACH(flow, &sa->filter.flow_list, entries) {
2311                 rc = sfc_flow_filter_insert(sa, flow);
2312                 if (rc != 0)
2313                         goto fail_bad_flow;
2314         }
2315
2316         sfc_log_init(sa, "done");
2317
2318 fail_bad_flow:
2319         return rc;
2320 }