examples/ipsec-secgw/ipsec-secgw.c

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