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