examples/ipsec-secgw/ipsec-secgw.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <netinet/in.h>
  11 #include <netinet/ip.h>
  12 #include <netinet/ip6.h>
  13 #include <string.h>
  14 #include <sys/queue.h>
  15 #include <stdarg.h>
  16 #include <errno.h>
  17 #include <getopt.h>
  18
  19 #include <rte_common.h>
  20 #include <rte_byteorder.h>
  21 #include <rte_log.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_mempool.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_acl.h>
  38 #include <rte_lpm.h>
  39 #include <rte_lpm6.h>
  40 #include <rte_hash.h>
  41 #include <rte_jhash.h>
  42 #include <rte_cryptodev.h>
  43
  44 #include "ipsec.h"
  45 #include "parser.h"
  46
  47 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  48
  49 #define MAX_JUMBO_PKT_LEN  9600
  50
  51 #define MEMPOOL_CACHE_SIZE 256
  52
  53 #define NB_MBUF (32000)
  54
  55 #define CDEV_QUEUE_DESC 2048
  56 #define CDEV_MAP_ENTRIES 1024
  57 #define CDEV_MP_NB_OBJS 2048
  58 #define CDEV_MP_CACHE_SZ 64
  59 #define MAX_QUEUE_PAIRS 1
  60
  61 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  62
  63 #define NB_SOCKETS 4
  64
  65 /* Configure how many packets ahead to prefetch, when reading packets */
  66 #define PREFETCH_OFFSET 3
  67
  68 #define MAX_RX_QUEUE_PER_LCORE 16
  69
  70 #define MAX_LCORE_PARAMS 1024
  71
  72 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
  73
  74 /*
  75  * Configurable number of RX/TX ring descriptors
  76  */
  77 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  78 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  79 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  80 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  81
  82 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
  83 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  84         (((uint64_t)((a) & 0xff) << 56) | \
  85         ((uint64_t)((b) & 0xff) << 48) | \
  86         ((uint64_t)((c) & 0xff) << 40) | \
  87         ((uint64_t)((d) & 0xff) << 32) | \
  88         ((uint64_t)((e) & 0xff) << 24) | \
  89         ((uint64_t)((f) & 0xff) << 16) | \
  90         ((uint64_t)((g) & 0xff) << 8)  | \
  91         ((uint64_t)(h) & 0xff))
  92 #else
  93 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  94         (((uint64_t)((h) & 0xff) << 56) | \
  95         ((uint64_t)((g) & 0xff) << 48) | \
  96         ((uint64_t)((f) & 0xff) << 40) | \
  97         ((uint64_t)((e) & 0xff) << 32) | \
  98         ((uint64_t)((d) & 0xff) << 24) | \
  99         ((uint64_t)((c) & 0xff) << 16) | \
 100         ((uint64_t)((b) & 0xff) << 8) | \
 101         ((uint64_t)(a) & 0xff))
 102 #endif
 103 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 104
 105 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 106                 addr.addr_bytes[0], addr.addr_bytes[1], \
 107                 addr.addr_bytes[2], addr.addr_bytes[3], \
 108                 addr.addr_bytes[4], addr.addr_bytes[5], \
 109                 0, 0)
 110
 111 /* port/source ethernet addr and destination ethernet addr */
 112 struct ethaddr_info {
 113         uint64_t src, dst;
 114 };
 115
 116 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 117         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 118         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 119         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 120         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 121 };
 122
 123 #define CMD_LINE_OPT_CONFIG             "config"
 124 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 125 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 126
 127 enum {
 128         /* long options mapped to a short option */
 129
 130         /* first long only option value must be >= 256, so that we won't
 131          * conflict with short options
 132          */
 133         CMD_LINE_OPT_MIN_NUM = 256,
 134         CMD_LINE_OPT_CONFIG_NUM,
 135         CMD_LINE_OPT_SINGLE_SA_NUM,
 136         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 137 };
 138
 139 static const struct option lgopts[] = {
 140         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 141         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 142         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 143         {NULL, 0, 0, 0}
 144 };
 145
 146 /* mask of enabled ports */
 147 static uint32_t enabled_port_mask;
 148 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 149 static uint32_t unprotected_port_mask;
 150 static int32_t promiscuous_on = 1;
 151 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 152 static uint32_t nb_lcores;
 153 static uint32_t single_sa;
 154 static uint32_t single_sa_idx;
 155 static uint32_t frame_size;
 156
 157 struct lcore_rx_queue {
 158         uint16_t port_id;
 159         uint8_t queue_id;
 160 } __rte_cache_aligned;
 161
 162 struct lcore_params {
 163         uint16_t port_id;
 164         uint8_t queue_id;
 165         uint8_t lcore_id;
 166 } __rte_cache_aligned;
 167
 168 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 169
 170 static struct lcore_params *lcore_params;
 171 static uint16_t nb_lcore_params;
 172
 173 static struct rte_hash *cdev_map_in;
 174 static struct rte_hash *cdev_map_out;
 175
 176 struct buffer {
 177         uint16_t len;
 178         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 179 };
 180
 181 struct lcore_conf {
 182         uint16_t nb_rx_queue;
 183         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 184         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 185         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 186         struct ipsec_ctx inbound;
 187         struct ipsec_ctx outbound;
 188         struct rt_ctx *rt4_ctx;
 189         struct rt_ctx *rt6_ctx;
 190 } __rte_cache_aligned;
 191
 192 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 193
 194 static struct rte_eth_conf port_conf = {
 195         .rxmode = {
 196                 .mq_mode        = ETH_MQ_RX_RSS,
 197                 .max_rx_pkt_len = ETHER_MAX_LEN,
 198                 .split_hdr_size = 0,
 199                 .offloads = DEV_RX_OFFLOAD_CHECKSUM |
 200                             DEV_RX_OFFLOAD_CRC_STRIP,
 201                 .ignore_offload_bitfield = 1,
 202         },
 203         .rx_adv_conf = {
 204                 .rss_conf = {
 205                         .rss_key = NULL,
 206                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 207                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 208                 },
 209         },
 210         .txmode = {
 211                 .mq_mode = ETH_MQ_TX_NONE,
 212                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
 213                              DEV_TX_OFFLOAD_MULTI_SEGS),
 214         },
 215 };
 216
 217 static struct socket_ctx socket_ctx[NB_SOCKETS];
 218
 219 struct traffic_type {
 220         const uint8_t *data[MAX_PKT_BURST * 2];
 221         struct rte_mbuf *pkts[MAX_PKT_BURST * 2];
 222         uint32_t res[MAX_PKT_BURST * 2];
 223         uint32_t num;
 224 };
 225
 226 struct ipsec_traffic {
 227         struct traffic_type ipsec;
 228         struct traffic_type ip4;
 229         struct traffic_type ip6;
 230 };
 231
 232 static inline void
 233 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 234 {
 235         uint8_t *nlp;
 236         struct ether_hdr *eth;
 237
 238         eth = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 239         if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
 240                 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
 241                 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip, ip_p));
 242                 if (*nlp == IPPROTO_ESP)
 243                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 244                 else {
 245                         t->ip4.data[t->ip4.num] = nlp;
 246                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 247                 }
 248         } else if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
 249                 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
 250                 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip6_hdr, ip6_nxt));
 251                 if (*nlp == IPPROTO_ESP)
 252                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 253                 else {
 254                         t->ip6.data[t->ip6.num] = nlp;
 255                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 256                 }
 257         } else {
 258                 /* Unknown/Unsupported type, drop the packet */
 259                 RTE_LOG(ERR, IPSEC, "Unsupported packet type\n");
 260                 rte_pktmbuf_free(pkt);
 261         }
 262
 263         /* Check if the packet has been processed inline. For inline protocol
 264          * processed packets, the metadata in the mbuf can be used to identify
 265          * the security processing done on the packet. The metadata will be
 266          * used to retrieve the application registered userdata associated
 267          * with the security session.
 268          */
 269
 270         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 271                 struct ipsec_sa *sa;
 272                 struct ipsec_mbuf_metadata *priv;
 273                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 274                                                 rte_eth_dev_get_sec_ctx(
 275                                                 pkt->port);
 276
 277                 /* Retrieve the userdata registered. Here, the userdata
 278                  * registered is the SA pointer.
 279                  */
 280
 281                 sa = (struct ipsec_sa *)
 282                                 rte_security_get_userdata(ctx, pkt->udata64);
 283
 284                 if (sa == NULL) {
 285                         /* userdata could not be retrieved */
 286                         return;
 287                 }
 288
 289                 /* Save SA as priv member in mbuf. This will be used in the
 290                  * IPsec selector(SP-SA) check.
 291                  */
 292
 293                 priv = get_priv(pkt);
 294                 priv->sa = sa;
 295         }
 296 }
 297
 298 static inline void
 299 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 300                 uint16_t nb_pkts)
 301 {
 302         int32_t i;
 303
 304         t->ipsec.num = 0;
 305         t->ip4.num = 0;
 306         t->ip6.num = 0;
 307
 308         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 309                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 310                                         void *));
 311                 prepare_one_packet(pkts[i], t);
 312         }
 313         /* Process left packets */
 314         for (; i < nb_pkts; i++)
 315                 prepare_one_packet(pkts[i], t);
 316 }
 317
 318 static inline void
 319 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port)
 320 {
 321         struct ip *ip;
 322         struct ether_hdr *ethhdr;
 323
 324         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 325
 326         ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
 327
 328         if (ip->ip_v == IPVERSION) {
 329                 pkt->ol_flags |= PKT_TX_IP_CKSUM | PKT_TX_IPV4;
 330                 pkt->l3_len = sizeof(struct ip);
 331                 pkt->l2_len = ETHER_HDR_LEN;
 332
 333                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
 334         } else {
 335                 pkt->ol_flags |= PKT_TX_IPV6;
 336                 pkt->l3_len = sizeof(struct ip6_hdr);
 337                 pkt->l2_len = ETHER_HDR_LEN;
 338
 339                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
 340         }
 341
 342         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 343                         sizeof(struct ether_addr));
 344         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 345                         sizeof(struct ether_addr));
 346 }
 347
 348 static inline void
 349 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port)
 350 {
 351         int32_t i;
 352         const int32_t prefetch_offset = 2;
 353
 354         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 355                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 356                 prepare_tx_pkt(pkts[i], port);
 357         }
 358         /* Process left packets */
 359         for (; i < nb_pkts; i++)
 360                 prepare_tx_pkt(pkts[i], port);
 361 }
 362
 363 /* Send burst of packets on an output interface */
 364 static inline int32_t
 365 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 366 {
 367         struct rte_mbuf **m_table;
 368         int32_t ret;
 369         uint16_t queueid;
 370
 371         queueid = qconf->tx_queue_id[port];
 372         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 373
 374         prepare_tx_burst(m_table, n, port);
 375
 376         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 377         if (unlikely(ret < n)) {
 378                 do {
 379                         rte_pktmbuf_free(m_table[ret]);
 380                 } while (++ret < n);
 381         }
 382
 383         return 0;
 384 }
 385
 386 /* Enqueue a single packet, and send burst if queue is filled */
 387 static inline int32_t
 388 send_single_packet(struct rte_mbuf *m, uint16_t port)
 389 {
 390         uint32_t lcore_id;
 391         uint16_t len;
 392         struct lcore_conf *qconf;
 393
 394         lcore_id = rte_lcore_id();
 395
 396         qconf = &lcore_conf[lcore_id];
 397         len = qconf->tx_mbufs[port].len;
 398         qconf->tx_mbufs[port].m_table[len] = m;
 399         len++;
 400
 401         /* enough pkts to be sent */
 402         if (unlikely(len == MAX_PKT_BURST)) {
 403                 send_burst(qconf, MAX_PKT_BURST, port);
 404                 len = 0;
 405         }
 406
 407         qconf->tx_mbufs[port].len = len;
 408         return 0;
 409 }
 410
 411 static inline void
 412 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 413                 uint16_t lim)
 414 {
 415         struct rte_mbuf *m;
 416         uint32_t i, j, res, sa_idx;
 417
 418         if (ip->num == 0 || sp == NULL)
 419                 return;
 420
 421         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 422                         ip->num, DEFAULT_MAX_CATEGORIES);
 423
 424         j = 0;
 425         for (i = 0; i < ip->num; i++) {
 426                 m = ip->pkts[i];
 427                 res = ip->res[i];
 428                 if (res & BYPASS) {
 429                         ip->pkts[j++] = m;
 430                         continue;
 431                 }
 432                 if (res & DISCARD) {
 433                         rte_pktmbuf_free(m);
 434                         continue;
 435                 }
 436
 437                 /* Only check SPI match for processed IPSec packets */
 438                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 439                         rte_pktmbuf_free(m);
 440                         continue;
 441                 }
 442
 443                 sa_idx = ip->res[i] & PROTECT_MASK;
 444                 if (sa_idx >= IPSEC_SA_MAX_ENTRIES ||
 445                                 !inbound_sa_check(sa, m, sa_idx)) {
 446                         rte_pktmbuf_free(m);
 447                         continue;
 448                 }
 449                 ip->pkts[j++] = m;
 450         }
 451         ip->num = j;
 452 }
 453
 454 static inline void
 455 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 456                 struct ipsec_traffic *traffic)
 457 {
 458         struct rte_mbuf *m;
 459         uint16_t idx, nb_pkts_in, i, n_ip4, n_ip6;
 460
 461         nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 462                         traffic->ipsec.num, MAX_PKT_BURST);
 463
 464         n_ip4 = traffic->ip4.num;
 465         n_ip6 = traffic->ip6.num;
 466
 467         /* SP/ACL Inbound check ipsec and ip4 */
 468         for (i = 0; i < nb_pkts_in; i++) {
 469                 m = traffic->ipsec.pkts[i];
 470                 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 471                 if (ip->ip_v == IPVERSION) {
 472                         idx = traffic->ip4.num++;
 473                         traffic->ip4.pkts[idx] = m;
 474                         traffic->ip4.data[idx] = rte_pktmbuf_mtod_offset(m,
 475                                         uint8_t *, offsetof(struct ip, ip_p));
 476                 } else if (ip->ip_v == IP6_VERSION) {
 477                         idx = traffic->ip6.num++;
 478                         traffic->ip6.pkts[idx] = m;
 479                         traffic->ip6.data[idx] = rte_pktmbuf_mtod_offset(m,
 480                                         uint8_t *,
 481                                         offsetof(struct ip6_hdr, ip6_nxt));
 482                 } else
 483                         rte_pktmbuf_free(m);
 484         }
 485
 486         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 487                         n_ip4);
 488
 489         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 490                         n_ip6);
 491 }
 492
 493 static inline void
 494 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 495                 struct traffic_type *ipsec)
 496 {
 497         struct rte_mbuf *m;
 498         uint32_t i, j, sa_idx;
 499
 500         if (ip->num == 0 || sp == NULL)
 501                 return;
 502
 503         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 504                         ip->num, DEFAULT_MAX_CATEGORIES);
 505
 506         j = 0;
 507         for (i = 0; i < ip->num; i++) {
 508                 m = ip->pkts[i];
 509                 sa_idx = ip->res[i] & PROTECT_MASK;
 510                 if (ip->res[i] & DISCARD)
 511                         rte_pktmbuf_free(m);
 512                 else if (sa_idx < IPSEC_SA_MAX_ENTRIES) {
 513                         ipsec->res[ipsec->num] = sa_idx;
 514                         ipsec->pkts[ipsec->num++] = m;
 515                 } else /* BYPASS */
 516                         ip->pkts[j++] = m;
 517         }
 518         ip->num = j;
 519 }
 520
 521 static inline void
 522 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 523                 struct ipsec_traffic *traffic)
 524 {
 525         struct rte_mbuf *m;
 526         uint16_t idx, nb_pkts_out, i;
 527
 528         /* Drop any IPsec traffic from protected ports */
 529         for (i = 0; i < traffic->ipsec.num; i++)
 530                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 531
 532         traffic->ipsec.num = 0;
 533
 534         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 535
 536         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 537
 538         nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 539                         traffic->ipsec.res, traffic->ipsec.num,
 540                         MAX_PKT_BURST);
 541
 542         for (i = 0; i < nb_pkts_out; i++) {
 543                 m = traffic->ipsec.pkts[i];
 544                 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 545                 if (ip->ip_v == IPVERSION) {
 546                         idx = traffic->ip4.num++;
 547                         traffic->ip4.pkts[idx] = m;
 548                 } else {
 549                         idx = traffic->ip6.num++;
 550                         traffic->ip6.pkts[idx] = m;
 551                 }
 552         }
 553 }
 554
 555 static inline void
 556 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 557                 struct ipsec_traffic *traffic)
 558 {
 559         struct rte_mbuf *m;
 560         uint32_t nb_pkts_in, i, idx;
 561
 562         /* Drop any IPv4 traffic from unprotected ports */
 563         for (i = 0; i < traffic->ip4.num; i++)
 564                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 565
 566         traffic->ip4.num = 0;
 567
 568         /* Drop any IPv6 traffic from unprotected ports */
 569         for (i = 0; i < traffic->ip6.num; i++)
 570                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 571
 572         traffic->ip6.num = 0;
 573
 574         nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 575                         traffic->ipsec.num, MAX_PKT_BURST);
 576
 577         for (i = 0; i < nb_pkts_in; i++) {
 578                 m = traffic->ipsec.pkts[i];
 579                 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 580                 if (ip->ip_v == IPVERSION) {
 581                         idx = traffic->ip4.num++;
 582                         traffic->ip4.pkts[idx] = m;
 583                 } else {
 584                         idx = traffic->ip6.num++;
 585                         traffic->ip6.pkts[idx] = m;
 586                 }
 587         }
 588 }
 589
 590 static inline void
 591 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 592                 struct ipsec_traffic *traffic)
 593 {
 594         struct rte_mbuf *m;
 595         uint32_t nb_pkts_out, i;
 596         struct ip *ip;
 597
 598         /* Drop any IPsec traffic from protected ports */
 599         for (i = 0; i < traffic->ipsec.num; i++)
 600                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 601
 602         traffic->ipsec.num = 0;
 603
 604         for (i = 0; i < traffic->ip4.num; i++)
 605                 traffic->ip4.res[i] = single_sa_idx;
 606
 607         for (i = 0; i < traffic->ip6.num; i++)
 608                 traffic->ip6.res[i] = single_sa_idx;
 609
 610         nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ip4.pkts,
 611                         traffic->ip4.res, traffic->ip4.num,
 612                         MAX_PKT_BURST);
 613
 614         /* They all sue the same SA (ip4 or ip6 tunnel) */
 615         m = traffic->ipsec.pkts[i];
 616         ip = rte_pktmbuf_mtod(m, struct ip *);
 617         if (ip->ip_v == IPVERSION)
 618                 traffic->ip4.num = nb_pkts_out;
 619         else
 620                 traffic->ip6.num = nb_pkts_out;
 621 }
 622
 623 static inline int32_t
 624 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 625 {
 626         struct ipsec_mbuf_metadata *priv;
 627         struct ipsec_sa *sa;
 628
 629         priv = get_priv(pkt);
 630
 631         sa = priv->sa;
 632         if (unlikely(sa == NULL)) {
 633                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 634                 goto fail;
 635         }
 636
 637         if (is_ipv6)
 638                 return sa->portid;
 639
 640         /* else */
 641         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 642
 643 fail:
 644         if (is_ipv6)
 645                 return -1;
 646
 647         /* else */
 648         return 0;
 649 }
 650
 651 static inline void
 652 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 653 {
 654         uint32_t hop[MAX_PKT_BURST * 2];
 655         uint32_t dst_ip[MAX_PKT_BURST * 2];
 656         int32_t pkt_hop = 0;
 657         uint16_t i, offset;
 658         uint16_t lpm_pkts = 0;
 659
 660         if (nb_pkts == 0)
 661                 return;
 662
 663         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 664          * have port ID in the SA
 665          */
 666
 667         for (i = 0; i < nb_pkts; i++) {
 668                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 669                         /* Security offload not enabled. So an LPM lookup is
 670                          * required to get the hop
 671                          */
 672                         offset = offsetof(struct ip, ip_dst);
 673                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 674                                         uint32_t *, offset);
 675                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 676                         lpm_pkts++;
 677                 }
 678         }
 679
 680         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 681
 682         lpm_pkts = 0;
 683
 684         for (i = 0; i < nb_pkts; i++) {
 685                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 686                         /* Read hop from the SA */
 687                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 688                 } else {
 689                         /* Need to use hop returned by lookup */
 690                         pkt_hop = hop[lpm_pkts++];
 691                 }
 692
 693                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 694                         rte_pktmbuf_free(pkts[i]);
 695                         continue;
 696                 }
 697                 send_single_packet(pkts[i], pkt_hop & 0xff);
 698         }
 699 }
 700
 701 static inline void
 702 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 703 {
 704         int32_t hop[MAX_PKT_BURST * 2];
 705         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 706         uint8_t *ip6_dst;
 707         int32_t pkt_hop = 0;
 708         uint16_t i, offset;
 709         uint16_t lpm_pkts = 0;
 710
 711         if (nb_pkts == 0)
 712                 return;
 713
 714         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 715          * have port ID in the SA
 716          */
 717
 718         for (i = 0; i < nb_pkts; i++) {
 719                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 720                         /* Security offload not enabled. So an LPM lookup is
 721                          * required to get the hop
 722                          */
 723                         offset = offsetof(struct ip6_hdr, ip6_dst);
 724                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 725                                         offset);
 726                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 727                         lpm_pkts++;
 728                 }
 729         }
 730
 731         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 732                         lpm_pkts);
 733
 734         lpm_pkts = 0;
 735
 736         for (i = 0; i < nb_pkts; i++) {
 737                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 738                         /* Read hop from the SA */
 739                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 740                 } else {
 741                         /* Need to use hop returned by lookup */
 742                         pkt_hop = hop[lpm_pkts++];
 743                 }
 744
 745                 if (pkt_hop == -1) {
 746                         rte_pktmbuf_free(pkts[i]);
 747                         continue;
 748                 }
 749                 send_single_packet(pkts[i], pkt_hop & 0xff);
 750         }
 751 }
 752
 753 static inline void
 754 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 755                 uint8_t nb_pkts, uint16_t portid)
 756 {
 757         struct ipsec_traffic traffic;
 758
 759         prepare_traffic(pkts, &traffic, nb_pkts);
 760
 761         if (unlikely(single_sa)) {
 762                 if (UNPROTECTED_PORT(portid))
 763                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 764                 else
 765                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 766         } else {
 767                 if (UNPROTECTED_PORT(portid))
 768                         process_pkts_inbound(&qconf->inbound, &traffic);
 769                 else
 770                         process_pkts_outbound(&qconf->outbound, &traffic);
 771         }
 772
 773         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 774         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 775 }
 776
 777 static inline void
 778 drain_buffers(struct lcore_conf *qconf)
 779 {
 780         struct buffer *buf;
 781         uint32_t portid;
 782
 783         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 784                 buf = &qconf->tx_mbufs[portid];
 785                 if (buf->len == 0)
 786                         continue;
 787                 send_burst(qconf, buf->len, portid);
 788                 buf->len = 0;
 789         }
 790 }
 791
 792 /* main processing loop */
 793 static int32_t
 794 main_loop(__attribute__((unused)) void *dummy)
 795 {
 796         struct rte_mbuf *pkts[MAX_PKT_BURST];
 797         uint32_t lcore_id;
 798         uint64_t prev_tsc, diff_tsc, cur_tsc;
 799         int32_t i, nb_rx;
 800         uint16_t portid;
 801         uint8_t queueid;
 802         struct lcore_conf *qconf;
 803         int32_t socket_id;
 804         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
 805                         / US_PER_S * BURST_TX_DRAIN_US;
 806         struct lcore_rx_queue *rxql;
 807
 808         prev_tsc = 0;
 809         lcore_id = rte_lcore_id();
 810         qconf = &lcore_conf[lcore_id];
 811         rxql = qconf->rx_queue_list;
 812         socket_id = rte_lcore_to_socket_id(lcore_id);
 813
 814         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
 815         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
 816         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
 817         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
 818         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
 819         qconf->inbound.cdev_map = cdev_map_in;
 820         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
 821         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
 822         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
 823         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
 824         qconf->outbound.cdev_map = cdev_map_out;
 825         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
 826
 827         if (qconf->nb_rx_queue == 0) {
 828                 RTE_LOG(INFO, IPSEC, "lcore %u has nothing to do\n", lcore_id);
 829                 return 0;
 830         }
 831
 832         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
 833
 834         for (i = 0; i < qconf->nb_rx_queue; i++) {
 835                 portid = rxql[i].port_id;
 836                 queueid = rxql[i].queue_id;
 837                 RTE_LOG(INFO, IPSEC,
 838                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 839                         lcore_id, portid, queueid);
 840         }
 841
 842         while (1) {
 843                 cur_tsc = rte_rdtsc();
 844
 845                 /* TX queue buffer drain */
 846                 diff_tsc = cur_tsc - prev_tsc;
 847
 848                 if (unlikely(diff_tsc > drain_tsc)) {
 849                         drain_buffers(qconf);
 850                         prev_tsc = cur_tsc;
 851                 }
 852
 853                 /* Read packet from RX queues */
 854                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
 855                         portid = rxql[i].port_id;
 856                         queueid = rxql[i].queue_id;
 857                         nb_rx = rte_eth_rx_burst(portid, queueid,
 858                                         pkts, MAX_PKT_BURST);
 859
 860                         if (nb_rx > 0)
 861                                 process_pkts(qconf, pkts, nb_rx, portid);
 862                 }
 863         }
 864 }
 865
 866 static int32_t
 867 check_params(void)
 868 {
 869         uint8_t lcore;
 870         uint16_t portid;
 871         uint16_t i;
 872         int32_t socket_id;
 873
 874         if (lcore_params == NULL) {
 875                 printf("Error: No port/queue/core mappings\n");
 876                 return -1;
 877         }
 878
 879         for (i = 0; i < nb_lcore_params; ++i) {
 880                 lcore = lcore_params[i].lcore_id;
 881                 if (!rte_lcore_is_enabled(lcore)) {
 882                         printf("error: lcore %hhu is not enabled in "
 883                                 "lcore mask\n", lcore);
 884                         return -1;
 885                 }
 886                 socket_id = rte_lcore_to_socket_id(lcore);
 887                 if (socket_id != 0 && numa_on == 0) {
 888                         printf("warning: lcore %hhu is on socket %d "
 889                                 "with numa off\n",
 890                                 lcore, socket_id);
 891                 }
 892                 portid = lcore_params[i].port_id;
 893                 if ((enabled_port_mask & (1 << portid)) == 0) {
 894                         printf("port %u is not enabled in port mask\n", portid);
 895                         return -1;
 896                 }
 897                 if (!rte_eth_dev_is_valid_port(portid)) {
 898                         printf("port %u is not present on the board\n", portid);
 899                         return -1;
 900                 }
 901         }
 902         return 0;
 903 }
 904
 905 static uint8_t
 906 get_port_nb_rx_queues(const uint16_t port)
 907 {
 908         int32_t queue = -1;
 909         uint16_t i;
 910
 911         for (i = 0; i < nb_lcore_params; ++i) {
 912                 if (lcore_params[i].port_id == port &&
 913                                 lcore_params[i].queue_id > queue)
 914                         queue = lcore_params[i].queue_id;
 915         }
 916         return (uint8_t)(++queue);
 917 }
 918
 919 static int32_t
 920 init_lcore_rx_queues(void)
 921 {
 922         uint16_t i, nb_rx_queue;
 923         uint8_t lcore;
 924
 925         for (i = 0; i < nb_lcore_params; ++i) {
 926                 lcore = lcore_params[i].lcore_id;
 927                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
 928                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
 929                         printf("error: too many queues (%u) for lcore: %u\n",
 930                                         nb_rx_queue + 1, lcore);
 931                         return -1;
 932                 }
 933                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
 934                         lcore_params[i].port_id;
 935                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
 936                         lcore_params[i].queue_id;
 937                 lcore_conf[lcore].nb_rx_queue++;
 938         }
 939         return 0;
 940 }
 941
 942 /* display usage */
 943 static void
 944 print_usage(const char *prgname)
 945 {
 946         printf("%s [EAL options] -- -p PORTMASK -P -u PORTMASK"
 947                 "  --"CMD_LINE_OPT_CONFIG" (port,queue,lcore)[,(port,queue,lcore]"
 948                 " --single-sa SAIDX -f CONFIG_FILE\n"
 949                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 950                 "  -P : enable promiscuous mode\n"
 951                 "  -u PORTMASK: hexadecimal bitmask of unprotected ports\n"
 952                 "  -j FRAMESIZE: jumbo frame maximum size\n"
 953                 "  --"CMD_LINE_OPT_CONFIG": (port,queue,lcore): "
 954                 "rx queues configuration\n"
 955                 "  --single-sa SAIDX: use single SA index for outbound, "
 956                 "bypassing the SP\n"
 957                 "  --cryptodev_mask MASK: hexadecimal bitmask of the "
 958                 "crypto devices to configure\n"
 959                 "  -f CONFIG_FILE: Configuration file path\n",
 960                 prgname);
 961 }
 962
 963 static int32_t
 964 parse_portmask(const char *portmask)
 965 {
 966         char *end = NULL;
 967         unsigned long pm;
 968
 969         /* parse hexadecimal string */
 970         pm = strtoul(portmask, &end, 16);
 971         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 972                 return -1;
 973
 974         if ((pm == 0) && errno)
 975                 return -1;
 976
 977         return pm;
 978 }
 979
 980 static int32_t
 981 parse_decimal(const char *str)
 982 {
 983         char *end = NULL;
 984         unsigned long num;
 985
 986         num = strtoul(str, &end, 10);
 987         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
 988                 return -1;
 989
 990         return num;
 991 }
 992
 993 static int32_t
 994 parse_config(const char *q_arg)
 995 {
 996         char s[256];
 997         const char *p, *p0 = q_arg;
 998         char *end;
 999         enum fieldnames {
1000                 FLD_PORT = 0,
1001                 FLD_QUEUE,
1002                 FLD_LCORE,
1003                 _NUM_FLD
1004         };
1005         unsigned long int_fld[_NUM_FLD];
1006         char *str_fld[_NUM_FLD];
1007         int32_t i;
1008         uint32_t size;
1009
1010         nb_lcore_params = 0;
1011
1012         while ((p = strchr(p0, '(')) != NULL) {
1013                 ++p;
1014                 p0 = strchr(p, ')');
1015                 if (p0 == NULL)
1016                         return -1;
1017
1018                 size = p0 - p;
1019                 if (size >= sizeof(s))
1020                         return -1;
1021
1022                 snprintf(s, sizeof(s), "%.*s", size, p);
1023                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1024                                 _NUM_FLD)
1025                         return -1;
1026                 for (i = 0; i < _NUM_FLD; i++) {
1027                         errno = 0;
1028                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1029                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1030                                 return -1;
1031                 }
1032                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1033                         printf("exceeded max number of lcore params: %hu\n",
1034                                 nb_lcore_params);
1035                         return -1;
1036                 }
1037                 lcore_params_array[nb_lcore_params].port_id =
1038                         (uint8_t)int_fld[FLD_PORT];
1039                 lcore_params_array[nb_lcore_params].queue_id =
1040                         (uint8_t)int_fld[FLD_QUEUE];
1041                 lcore_params_array[nb_lcore_params].lcore_id =
1042                         (uint8_t)int_fld[FLD_LCORE];
1043                 ++nb_lcore_params;
1044         }
1045         lcore_params = lcore_params_array;
1046         return 0;
1047 }
1048
1049 static int32_t
1050 parse_args(int32_t argc, char **argv)
1051 {
1052         int32_t opt, ret;
1053         char **argvopt;
1054         int32_t option_index;
1055         char *prgname = argv[0];
1056         int32_t f_present = 0;
1057
1058         argvopt = argv;
1059
1060         while ((opt = getopt_long(argc, argvopt, "p:Pu:f:j:",
1061                                 lgopts, &option_index)) != EOF) {
1062
1063                 switch (opt) {
1064                 case 'p':
1065                         enabled_port_mask = parse_portmask(optarg);
1066                         if (enabled_port_mask == 0) {
1067                                 printf("invalid portmask\n");
1068                                 print_usage(prgname);
1069                                 return -1;
1070                         }
1071                         break;
1072                 case 'P':
1073                         printf("Promiscuous mode selected\n");
1074                         promiscuous_on = 1;
1075                         break;
1076                 case 'u':
1077                         unprotected_port_mask = parse_portmask(optarg);
1078                         if (unprotected_port_mask == 0) {
1079                                 printf("invalid unprotected portmask\n");
1080                                 print_usage(prgname);
1081                                 return -1;
1082                         }
1083                         break;
1084                 case 'f':
1085                         if (f_present == 1) {
1086                                 printf("\"-f\" option present more than "
1087                                         "once!\n");
1088                                 print_usage(prgname);
1089                                 return -1;
1090                         }
1091                         if (parse_cfg_file(optarg) < 0) {
1092                                 printf("parsing file \"%s\" failed\n",
1093                                         optarg);
1094                                 print_usage(prgname);
1095                                 return -1;
1096                         }
1097                         f_present = 1;
1098                         break;
1099                 case 'j':
1100                         {
1101                                 int32_t size = parse_decimal(optarg);
1102                                 if (size <= 1518) {
1103                                         printf("Invalid jumbo frame size\n");
1104                                         if (size < 0) {
1105                                                 print_usage(prgname);
1106                                                 return -1;
1107                                         }
1108                                         printf("Using default value 9000\n");
1109                                         frame_size = 9000;
1110                                 } else {
1111                                         frame_size = size;
1112                                 }
1113                         }
1114                         printf("Enabled jumbo frames size %u\n", frame_size);
1115                         break;
1116                 case CMD_LINE_OPT_CONFIG_NUM:
1117                         ret = parse_config(optarg);
1118                         if (ret) {
1119                                 printf("Invalid config\n");
1120                                 print_usage(prgname);
1121                                 return -1;
1122                         }
1123                         break;
1124                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1125                         ret = parse_decimal(optarg);
1126                         if (ret == -1) {
1127                                 printf("Invalid argument[sa_idx]\n");
1128                                 print_usage(prgname);
1129                                 return -1;
1130                         }
1131
1132                         /* else */
1133                         single_sa = 1;
1134                         single_sa_idx = ret;
1135                         printf("Configured with single SA index %u\n",
1136                                         single_sa_idx);
1137                         break;
1138                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1139                         ret = parse_portmask(optarg);
1140                         if (ret == -1) {
1141                                 printf("Invalid argument[portmask]\n");
1142                                 print_usage(prgname);
1143                                 return -1;
1144                         }
1145
1146                         /* else */
1147                         enabled_cryptodev_mask = ret;
1148                         break;
1149                 default:
1150                         print_usage(prgname);
1151                         return -1;
1152                 }
1153         }
1154
1155         if (f_present == 0) {
1156                 printf("Mandatory option \"-f\" not present\n");
1157                 return -1;
1158         }
1159
1160         if (optind >= 0)
1161                 argv[optind-1] = prgname;
1162
1163         ret = optind-1;
1164         optind = 1; /* reset getopt lib */
1165         return ret;
1166 }
1167
1168 static void
1169 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1170 {
1171         char buf[ETHER_ADDR_FMT_SIZE];
1172         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1173         printf("%s%s", name, buf);
1174 }
1175
1176 /* Check the link status of all ports in up to 9s, and print them finally */
1177 static void
1178 check_all_ports_link_status(uint32_t port_mask)
1179 {
1180 #define CHECK_INTERVAL 100 /* 100ms */
1181 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1182         uint16_t portid;
1183         uint8_t count, all_ports_up, print_flag = 0;
1184         struct rte_eth_link link;
1185
1186         printf("\nChecking link status");
1187         fflush(stdout);
1188         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1189                 all_ports_up = 1;
1190                 RTE_ETH_FOREACH_DEV(portid) {
1191                         if ((port_mask & (1 << portid)) == 0)
1192                                 continue;
1193                         memset(&link, 0, sizeof(link));
1194                         rte_eth_link_get_nowait(portid, &link);
1195                         /* print link status if flag set */
1196                         if (print_flag == 1) {
1197                                 if (link.link_status)
1198                                         printf(
1199                                         "Port%d Link Up - speed %u Mbps -%s\n",
1200                                                 portid, link.link_speed,
1201                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1202                                         ("full-duplex") : ("half-duplex\n"));
1203                                 else
1204                                         printf("Port %d Link Down\n", portid);
1205                                 continue;
1206                         }
1207                         /* clear all_ports_up flag if any link down */
1208                         if (link.link_status == ETH_LINK_DOWN) {
1209                                 all_ports_up = 0;
1210                                 break;
1211                         }
1212                 }
1213                 /* after finally printing all link status, get out */
1214                 if (print_flag == 1)
1215                         break;
1216
1217                 if (all_ports_up == 0) {
1218                         printf(".");
1219                         fflush(stdout);
1220                         rte_delay_ms(CHECK_INTERVAL);
1221                 }
1222
1223                 /* set the print_flag if all ports up or timeout */
1224                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1225                         print_flag = 1;
1226                         printf("done\n");
1227                 }
1228         }
1229 }
1230
1231 static int32_t
1232 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1233                 uint16_t qp, struct lcore_params *params,
1234                 struct ipsec_ctx *ipsec_ctx,
1235                 const struct rte_cryptodev_capabilities *cipher,
1236                 const struct rte_cryptodev_capabilities *auth,
1237                 const struct rte_cryptodev_capabilities *aead)
1238 {
1239         int32_t ret = 0;
1240         unsigned long i;
1241         struct cdev_key key = { 0 };
1242
1243         key.lcore_id = params->lcore_id;
1244         if (cipher)
1245                 key.cipher_algo = cipher->sym.cipher.algo;
1246         if (auth)
1247                 key.auth_algo = auth->sym.auth.algo;
1248         if (aead)
1249                 key.aead_algo = aead->sym.aead.algo;
1250
1251         ret = rte_hash_lookup(map, &key);
1252         if (ret != -ENOENT)
1253                 return 0;
1254
1255         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1256                 if (ipsec_ctx->tbl[i].id == cdev_id)
1257                         break;
1258
1259         if (i == ipsec_ctx->nb_qps) {
1260                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1261                         printf("Maximum number of crypto devices assigned to "
1262                                 "a core, increase MAX_QP_PER_LCORE value\n");
1263                         return 0;
1264                 }
1265                 ipsec_ctx->tbl[i].id = cdev_id;
1266                 ipsec_ctx->tbl[i].qp = qp;
1267                 ipsec_ctx->nb_qps++;
1268                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1269                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1270                                 cdev_id, qp, i);
1271         }
1272
1273         ret = rte_hash_add_key_data(map, &key, (void *)i);
1274         if (ret < 0) {
1275                 printf("Faled to insert cdev mapping for (lcore %u, "
1276                                 "cdev %u, qp %u), errno %d\n",
1277                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1278                                 ipsec_ctx->tbl[i].qp, ret);
1279                 return 0;
1280         }
1281
1282         return 1;
1283 }
1284
1285 static int32_t
1286 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1287                 uint16_t qp, struct lcore_params *params)
1288 {
1289         int32_t ret = 0;
1290         const struct rte_cryptodev_capabilities *i, *j;
1291         struct rte_hash *map;
1292         struct lcore_conf *qconf;
1293         struct ipsec_ctx *ipsec_ctx;
1294         const char *str;
1295
1296         qconf = &lcore_conf[params->lcore_id];
1297
1298         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1299                 map = cdev_map_out;
1300                 ipsec_ctx = &qconf->outbound;
1301                 str = "Outbound";
1302         } else {
1303                 map = cdev_map_in;
1304                 ipsec_ctx = &qconf->inbound;
1305                 str = "Inbound";
1306         }
1307
1308         /* Required cryptodevs with operation chainning */
1309         if (!(dev_info->feature_flags &
1310                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1311                 return ret;
1312
1313         for (i = dev_info->capabilities;
1314                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1315                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1316                         continue;
1317
1318                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1319                         ret |= add_mapping(map, str, cdev_id, qp, params,
1320                                         ipsec_ctx, NULL, NULL, i);
1321                         continue;
1322                 }
1323
1324                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1325                         continue;
1326
1327                 for (j = dev_info->capabilities;
1328                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1329                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1330                                 continue;
1331
1332                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1333                                 continue;
1334
1335                         ret |= add_mapping(map, str, cdev_id, qp, params,
1336                                                 ipsec_ctx, i, j, NULL);
1337                 }
1338         }
1339
1340         return ret;
1341 }
1342
1343 /* Check if the device is enabled by cryptodev_mask */
1344 static int
1345 check_cryptodev_mask(uint8_t cdev_id)
1346 {
1347         if (enabled_cryptodev_mask & (1 << cdev_id))
1348                 return 0;
1349
1350         return -1;
1351 }
1352
1353 static int32_t
1354 cryptodevs_init(void)
1355 {
1356         struct rte_cryptodev_config dev_conf;
1357         struct rte_cryptodev_qp_conf qp_conf;
1358         uint16_t idx, max_nb_qps, qp, i;
1359         int16_t cdev_id, port_id;
1360         struct rte_hash_parameters params = { 0 };
1361
1362         params.entries = CDEV_MAP_ENTRIES;
1363         params.key_len = sizeof(struct cdev_key);
1364         params.hash_func = rte_jhash;
1365         params.hash_func_init_val = 0;
1366         params.socket_id = rte_socket_id();
1367
1368         params.name = "cdev_map_in";
1369         cdev_map_in = rte_hash_create(&params);
1370         if (cdev_map_in == NULL)
1371                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1372                                 rte_errno);
1373
1374         params.name = "cdev_map_out";
1375         cdev_map_out = rte_hash_create(&params);
1376         if (cdev_map_out == NULL)
1377                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1378                                 rte_errno);
1379
1380         printf("lcore/cryptodev/qp mappings:\n");
1381
1382         uint32_t max_sess_sz = 0, sess_sz;
1383         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1384                 sess_sz = rte_cryptodev_get_private_session_size(cdev_id);
1385                 if (sess_sz > max_sess_sz)
1386                         max_sess_sz = sess_sz;
1387         }
1388         RTE_ETH_FOREACH_DEV(port_id) {
1389                 void *sec_ctx;
1390
1391                 if ((enabled_port_mask & (1 << port_id)) == 0)
1392                         continue;
1393
1394                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
1395                 if (sec_ctx == NULL)
1396                         continue;
1397
1398                 sess_sz = rte_security_session_get_size(sec_ctx);
1399                 if (sess_sz > max_sess_sz)
1400                         max_sess_sz = sess_sz;
1401         }
1402
1403         idx = 0;
1404         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1405                 struct rte_cryptodev_info cdev_info;
1406
1407                 if (check_cryptodev_mask((uint8_t)cdev_id))
1408                         continue;
1409
1410                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1411
1412                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1413                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1414                 else
1415                         max_nb_qps = nb_lcore_params;
1416
1417                 qp = 0;
1418                 i = 0;
1419                 while (qp < max_nb_qps && i < nb_lcore_params) {
1420                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1421                                                 &lcore_params[idx]))
1422                                 qp++;
1423                         idx++;
1424                         idx = idx % nb_lcore_params;
1425                         i++;
1426                 }
1427
1428                 if (qp == 0)
1429                         continue;
1430
1431                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1432                 dev_conf.nb_queue_pairs = qp;
1433
1434                 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1435                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1436                         struct rte_mempool *sess_mp;
1437
1438                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1439                                         "sess_mp_%u", dev_conf.socket_id);
1440                         sess_mp = rte_mempool_create(mp_name,
1441                                         CDEV_MP_NB_OBJS,
1442                                         max_sess_sz,
1443                                         CDEV_MP_CACHE_SZ,
1444                                         0, NULL, NULL, NULL,
1445                                         NULL, dev_conf.socket_id,
1446                                         0);
1447                         if (sess_mp == NULL)
1448                                 rte_exit(EXIT_FAILURE,
1449                                         "Cannot create session pool on socket %d\n",
1450                                         dev_conf.socket_id);
1451                         else
1452                                 printf("Allocated session pool on socket %d\n",
1453                                         dev_conf.socket_id);
1454                         socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1455                 }
1456
1457                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1458                         rte_panic("Failed to initialize cryptodev %u\n",
1459                                         cdev_id);
1460
1461                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1462                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1463                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1464                                         &qp_conf, dev_conf.socket_id,
1465                                         socket_ctx[dev_conf.socket_id].session_pool))
1466                                 rte_panic("Failed to setup queue %u for "
1467                                                 "cdev_id %u\n", 0, cdev_id);
1468
1469                 if (rte_cryptodev_start(cdev_id))
1470                         rte_panic("Failed to start cryptodev %u\n",
1471                                         cdev_id);
1472         }
1473
1474         /* create session pools for eth devices that implement security */
1475         RTE_ETH_FOREACH_DEV(port_id) {
1476                 if ((enabled_port_mask & (1 << port_id)) &&
1477                                 rte_eth_dev_get_sec_ctx(port_id)) {
1478                         int socket_id = rte_eth_dev_socket_id(port_id);
1479
1480                         if (!socket_ctx[socket_id].session_pool) {
1481                                 char mp_name[RTE_MEMPOOL_NAMESIZE];
1482                                 struct rte_mempool *sess_mp;
1483
1484                                 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1485                                                 "sess_mp_%u", socket_id);
1486                                 sess_mp = rte_mempool_create(mp_name,
1487                                                 CDEV_MP_NB_OBJS,
1488                                                 max_sess_sz,
1489                                                 CDEV_MP_CACHE_SZ,
1490                                                 0, NULL, NULL, NULL,
1491                                                 NULL, socket_id,
1492                                                 0);
1493                                 if (sess_mp == NULL)
1494                                         rte_exit(EXIT_FAILURE,
1495                                                 "Cannot create session pool "
1496                                                 "on socket %d\n", socket_id);
1497                                 else
1498                                         printf("Allocated session pool "
1499                                                 "on socket %d\n", socket_id);
1500                                 socket_ctx[socket_id].session_pool = sess_mp;
1501                         }
1502                 }
1503         }
1504
1505
1506         printf("\n");
1507
1508         return 0;
1509 }
1510
1511 static void
1512 port_init(uint16_t portid)
1513 {
1514         struct rte_eth_dev_info dev_info;
1515         struct rte_eth_txconf *txconf;
1516         uint16_t nb_tx_queue, nb_rx_queue;
1517         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1518         int32_t ret, socket_id;
1519         struct lcore_conf *qconf;
1520         struct ether_addr ethaddr;
1521         struct rte_eth_conf local_port_conf = port_conf;
1522
1523         rte_eth_dev_info_get(portid, &dev_info);
1524
1525         printf("Configuring device port %u:\n", portid);
1526
1527         rte_eth_macaddr_get(portid, &ethaddr);
1528         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(ethaddr);
1529         print_ethaddr("Address: ", &ethaddr);
1530         printf("\n");
1531
1532         nb_rx_queue = get_port_nb_rx_queues(portid);
1533         nb_tx_queue = nb_lcores;
1534
1535         if (nb_rx_queue > dev_info.max_rx_queues)
1536                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1537                                 "(max rx queue is %u)\n",
1538                                 nb_rx_queue, dev_info.max_rx_queues);
1539
1540         if (nb_tx_queue > dev_info.max_tx_queues)
1541                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1542                                 "(max tx queue is %u)\n",
1543                                 nb_tx_queue, dev_info.max_tx_queues);
1544
1545         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1546                         nb_rx_queue, nb_tx_queue);
1547
1548         if (frame_size) {
1549                 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
1550                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1551         }
1552
1553         if (dev_info.rx_offload_capa & DEV_RX_OFFLOAD_SECURITY)
1554                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SECURITY;
1555         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_SECURITY)
1556                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_SECURITY;
1557         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1558                 local_port_conf.txmode.offloads |=
1559                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1560         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1561                         &local_port_conf);
1562         if (ret < 0)
1563                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1564                                 "err=%d, port=%d\n", ret, portid);
1565
1566         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
1567         if (ret < 0)
1568                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
1569                                 "err=%d, port=%d\n", ret, portid);
1570
1571         /* init one TX queue per lcore */
1572         tx_queueid = 0;
1573         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1574                 if (rte_lcore_is_enabled(lcore_id) == 0)
1575                         continue;
1576
1577                 if (numa_on)
1578                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1579                 else
1580                         socket_id = 0;
1581
1582                 /* init TX queue */
1583                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1584
1585                 txconf = &dev_info.default_txconf;
1586                 txconf->txq_flags = ETH_TXQ_FLAGS_IGNORE;
1587                 txconf->offloads = local_port_conf.txmode.offloads;
1588
1589                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1590                                 socket_id, txconf);
1591                 if (ret < 0)
1592                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1593                                         "err=%d, port=%d\n", ret, portid);
1594
1595                 qconf = &lcore_conf[lcore_id];
1596                 qconf->tx_queue_id[portid] = tx_queueid;
1597                 tx_queueid++;
1598
1599                 /* init RX queues */
1600                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1601                         struct rte_eth_rxconf rxq_conf;
1602
1603                         if (portid != qconf->rx_queue_list[queue].port_id)
1604                                 continue;
1605
1606                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
1607
1608                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1609                                         socket_id);
1610
1611                         rxq_conf = dev_info.default_rxconf;
1612                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
1613                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1614                                         nb_rxd, socket_id, &rxq_conf,
1615                                         socket_ctx[socket_id].mbuf_pool);
1616                         if (ret < 0)
1617                                 rte_exit(EXIT_FAILURE,
1618                                         "rte_eth_rx_queue_setup: err=%d, "
1619                                         "port=%d\n", ret, portid);
1620                 }
1621         }
1622         printf("\n");
1623 }
1624
1625 static void
1626 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1627 {
1628         char s[64];
1629         uint32_t buff_size = frame_size ? (frame_size + RTE_PKTMBUF_HEADROOM) :
1630                         RTE_MBUF_DEFAULT_BUF_SIZE;
1631
1632
1633         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1634         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1635                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1636                         buff_size,
1637                         socket_id);
1638         if (ctx->mbuf_pool == NULL)
1639                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
1640                                 socket_id);
1641         else
1642                 printf("Allocated mbuf pool on socket %d\n", socket_id);
1643 }
1644
1645 int32_t
1646 main(int32_t argc, char **argv)
1647 {
1648         int32_t ret;
1649         uint32_t lcore_id;
1650         uint8_t socket_id;
1651         uint16_t portid;
1652
1653         /* init EAL */
1654         ret = rte_eal_init(argc, argv);
1655         if (ret < 0)
1656                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1657         argc -= ret;
1658         argv += ret;
1659
1660         /* parse application arguments (after the EAL ones) */
1661         ret = parse_args(argc, argv);
1662         if (ret < 0)
1663                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
1664
1665         if ((unprotected_port_mask & enabled_port_mask) !=
1666                         unprotected_port_mask)
1667                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
1668                                 unprotected_port_mask);
1669
1670         if (check_params() < 0)
1671                 rte_exit(EXIT_FAILURE, "check_params failed\n");
1672
1673         ret = init_lcore_rx_queues();
1674         if (ret < 0)
1675                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
1676
1677         nb_lcores = rte_lcore_count();
1678
1679         /* Replicate each context per socket */
1680         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1681                 if (rte_lcore_is_enabled(lcore_id) == 0)
1682                         continue;
1683
1684                 if (numa_on)
1685                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1686                 else
1687                         socket_id = 0;
1688
1689                 if (socket_ctx[socket_id].mbuf_pool)
1690                         continue;
1691
1692                 sa_init(&socket_ctx[socket_id], socket_id);
1693
1694                 sp4_init(&socket_ctx[socket_id], socket_id);
1695
1696                 sp6_init(&socket_ctx[socket_id], socket_id);
1697
1698                 rt_init(&socket_ctx[socket_id], socket_id);
1699
1700                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
1701         }
1702
1703         RTE_ETH_FOREACH_DEV(portid) {
1704                 if ((enabled_port_mask & (1 << portid)) == 0)
1705                         continue;
1706
1707                 port_init(portid);
1708         }
1709
1710         cryptodevs_init();
1711
1712         /* start ports */
1713         RTE_ETH_FOREACH_DEV(portid) {
1714                 if ((enabled_port_mask & (1 << portid)) == 0)
1715                         continue;
1716
1717                 /* Start device */
1718                 ret = rte_eth_dev_start(portid);
1719                 if (ret < 0)
1720                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
1721                                         "err=%d, port=%d\n", ret, portid);
1722                 /*
1723                  * If enabled, put device in promiscuous mode.
1724                  * This allows IO forwarding mode to forward packets
1725                  * to itself through 2 cross-connected  ports of the
1726                  * target machine.
1727                  */
1728                 if (promiscuous_on)
1729                         rte_eth_promiscuous_enable(portid);
1730         }
1731
1732         check_all_ports_link_status(enabled_port_mask);
1733
1734         /* launch per-lcore init on every lcore */
1735         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1736         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1737                 if (rte_eal_wait_lcore(lcore_id) < 0)
1738                         return -1;
1739         }
1740
1741         return 0;
1742 }