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