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