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