examples/ipsec-secgw/ipsec-secgw.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <netinet/in.h>
  11 #include <netinet/ip.h>
  12 #include <netinet/ip6.h>
  13 #include <string.h>
  14 #include <sys/queue.h>
  15 #include <stdarg.h>
  16 #include <errno.h>
  17 #include <getopt.h>
  18
  19 #include <rte_common.h>
  20 #include <rte_byteorder.h>
  21 #include <rte_log.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_mempool.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_acl.h>
  38 #include <rte_lpm.h>
  39 #include <rte_lpm6.h>
  40 #include <rte_hash.h>
  41 #include <rte_jhash.h>
  42 #include <rte_cryptodev.h>
  43 #include <rte_security.h>
  44
  45 #include "ipsec.h"
  46 #include "parser.h"
  47
  48 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  49
  50 #define MAX_JUMBO_PKT_LEN  9600
  51
  52 #define MEMPOOL_CACHE_SIZE 256
  53
  54 #define NB_MBUF (32000)
  55
  56 #define CDEV_QUEUE_DESC 2048
  57 #define CDEV_MAP_ENTRIES 16384
  58 #define CDEV_MP_NB_OBJS 1024
  59 #define CDEV_MP_CACHE_SZ 64
  60 #define MAX_QUEUE_PAIRS 1
  61
  62 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  63
  64 #define NB_SOCKETS 4
  65
  66 /* Configure how many packets ahead to prefetch, when reading packets */
  67 #define PREFETCH_OFFSET 3
  68
  69 #define MAX_RX_QUEUE_PER_LCORE 16
  70
  71 #define MAX_LCORE_PARAMS 1024
  72
  73 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
  74
  75 /*
  76  * Configurable number of RX/TX ring descriptors
  77  */
  78 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  79 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  80 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  81 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  82
  83 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
  84 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  85         (((uint64_t)((a) & 0xff) << 56) | \
  86         ((uint64_t)((b) & 0xff) << 48) | \
  87         ((uint64_t)((c) & 0xff) << 40) | \
  88         ((uint64_t)((d) & 0xff) << 32) | \
  89         ((uint64_t)((e) & 0xff) << 24) | \
  90         ((uint64_t)((f) & 0xff) << 16) | \
  91         ((uint64_t)((g) & 0xff) << 8)  | \
  92         ((uint64_t)(h) & 0xff))
  93 #else
  94 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  95         (((uint64_t)((h) & 0xff) << 56) | \
  96         ((uint64_t)((g) & 0xff) << 48) | \
  97         ((uint64_t)((f) & 0xff) << 40) | \
  98         ((uint64_t)((e) & 0xff) << 32) | \
  99         ((uint64_t)((d) & 0xff) << 24) | \
 100         ((uint64_t)((c) & 0xff) << 16) | \
 101         ((uint64_t)((b) & 0xff) << 8) | \
 102         ((uint64_t)(a) & 0xff))
 103 #endif
 104 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 105
 106 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 107                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
 108                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
 109                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
 110                 0, 0)
 111
 112 /* port/source ethernet addr and destination ethernet addr */
 113 struct ethaddr_info {
 114         uint64_t src, dst;
 115 };
 116
 117 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 118         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 119         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 120         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 121         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 122 };
 123
 124 #define CMD_LINE_OPT_CONFIG             "config"
 125 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 126 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 127 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
 128 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
 129
 130 enum {
 131         /* long options mapped to a short option */
 132
 133         /* first long only option value must be >= 256, so that we won't
 134          * conflict with short options
 135          */
 136         CMD_LINE_OPT_MIN_NUM = 256,
 137         CMD_LINE_OPT_CONFIG_NUM,
 138         CMD_LINE_OPT_SINGLE_SA_NUM,
 139         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 140         CMD_LINE_OPT_RX_OFFLOAD_NUM,
 141         CMD_LINE_OPT_TX_OFFLOAD_NUM,
 142 };
 143
 144 static const struct option lgopts[] = {
 145         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 146         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 147         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 148         {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
 149         {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
 150         {NULL, 0, 0, 0}
 151 };
 152
 153 /* mask of enabled ports */
 154 static uint32_t enabled_port_mask;
 155 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 156 static uint32_t unprotected_port_mask;
 157 static int32_t promiscuous_on = 1;
 158 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 159 static uint32_t nb_lcores;
 160 static uint32_t single_sa;
 161 static uint32_t single_sa_idx;
 162 static uint32_t frame_size;
 163
 164 /*
 165  * RX/TX HW offload capabilities to enable/use on ethernet ports.
 166  * By default all capabilities are enabled.
 167  */
 168 static uint64_t dev_rx_offload = UINT64_MAX;
 169 static uint64_t dev_tx_offload = UINT64_MAX;
 170
 171 struct lcore_rx_queue {
 172         uint16_t port_id;
 173         uint8_t queue_id;
 174 } __rte_cache_aligned;
 175
 176 struct lcore_params {
 177         uint16_t port_id;
 178         uint8_t queue_id;
 179         uint8_t lcore_id;
 180 } __rte_cache_aligned;
 181
 182 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 183
 184 static struct lcore_params *lcore_params;
 185 static uint16_t nb_lcore_params;
 186
 187 static struct rte_hash *cdev_map_in;
 188 static struct rte_hash *cdev_map_out;
 189
 190 struct buffer {
 191         uint16_t len;
 192         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 193 };
 194
 195 struct lcore_conf {
 196         uint16_t nb_rx_queue;
 197         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 198         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 199         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 200         struct ipsec_ctx inbound;
 201         struct ipsec_ctx outbound;
 202         struct rt_ctx *rt4_ctx;
 203         struct rt_ctx *rt6_ctx;
 204 } __rte_cache_aligned;
 205
 206 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 207
 208 static struct rte_eth_conf port_conf = {
 209         .rxmode = {
 210                 .mq_mode        = ETH_MQ_RX_RSS,
 211                 .max_rx_pkt_len = ETHER_MAX_LEN,
 212                 .split_hdr_size = 0,
 213                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 214         },
 215         .rx_adv_conf = {
 216                 .rss_conf = {
 217                         .rss_key = NULL,
 218                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 219                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 220                 },
 221         },
 222         .txmode = {
 223                 .mq_mode = ETH_MQ_TX_NONE,
 224         },
 225 };
 226
 227 static struct socket_ctx socket_ctx[NB_SOCKETS];
 228
 229 struct traffic_type {
 230         const uint8_t *data[MAX_PKT_BURST * 2];
 231         struct rte_mbuf *pkts[MAX_PKT_BURST * 2];
 232         uint32_t res[MAX_PKT_BURST * 2];
 233         uint32_t num;
 234 };
 235
 236 struct ipsec_traffic {
 237         struct traffic_type ipsec;
 238         struct traffic_type ip4;
 239         struct traffic_type ip6;
 240 };
 241
 242 static inline void
 243 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 244 {
 245         uint8_t *nlp;
 246         struct ether_hdr *eth;
 247
 248         eth = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 249         if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
 250                 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
 251                 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip, ip_p));
 252                 if (*nlp == IPPROTO_ESP)
 253                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 254                 else {
 255                         t->ip4.data[t->ip4.num] = nlp;
 256                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 257                 }
 258         } else if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
 259                 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
 260                 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip6_hdr, ip6_nxt));
 261                 if (*nlp == IPPROTO_ESP)
 262                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 263                 else {
 264                         t->ip6.data[t->ip6.num] = nlp;
 265                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 266                 }
 267         } else {
 268                 /* Unknown/Unsupported type, drop the packet */
 269                 RTE_LOG(ERR, IPSEC, "Unsupported packet type\n");
 270                 rte_pktmbuf_free(pkt);
 271         }
 272
 273         /* Check if the packet has been processed inline. For inline protocol
 274          * processed packets, the metadata in the mbuf can be used to identify
 275          * the security processing done on the packet. The metadata will be
 276          * used to retrieve the application registered userdata associated
 277          * with the security session.
 278          */
 279
 280         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 281                 struct ipsec_sa *sa;
 282                 struct ipsec_mbuf_metadata *priv;
 283                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 284                                                 rte_eth_dev_get_sec_ctx(
 285                                                 pkt->port);
 286
 287                 /* Retrieve the userdata registered. Here, the userdata
 288                  * registered is the SA pointer.
 289                  */
 290
 291                 sa = (struct ipsec_sa *)
 292                                 rte_security_get_userdata(ctx, pkt->udata64);
 293
 294                 if (sa == NULL) {
 295                         /* userdata could not be retrieved */
 296                         return;
 297                 }
 298
 299                 /* Save SA as priv member in mbuf. This will be used in the
 300                  * IPsec selector(SP-SA) check.
 301                  */
 302
 303                 priv = get_priv(pkt);
 304                 priv->sa = sa;
 305         }
 306 }
 307
 308 static inline void
 309 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 310                 uint16_t nb_pkts)
 311 {
 312         int32_t i;
 313
 314         t->ipsec.num = 0;
 315         t->ip4.num = 0;
 316         t->ip6.num = 0;
 317
 318         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 319                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 320                                         void *));
 321                 prepare_one_packet(pkts[i], t);
 322         }
 323         /* Process left packets */
 324         for (; i < nb_pkts; i++)
 325                 prepare_one_packet(pkts[i], t);
 326 }
 327
 328 static inline void
 329 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 330                 const struct lcore_conf *qconf)
 331 {
 332         struct ip *ip;
 333         struct ether_hdr *ethhdr;
 334
 335         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 336
 337         ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
 338
 339         if (ip->ip_v == IPVERSION) {
 340                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 341                 pkt->l3_len = sizeof(struct ip);
 342                 pkt->l2_len = ETHER_HDR_LEN;
 343
 344                 ip->ip_sum = 0;
 345
 346                 /* calculate IPv4 cksum in SW */
 347                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 348                         ip->ip_sum = rte_ipv4_cksum((struct ipv4_hdr *)ip);
 349
 350                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
 351         } else {
 352                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 353                 pkt->l3_len = sizeof(struct ip6_hdr);
 354                 pkt->l2_len = ETHER_HDR_LEN;
 355
 356                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
 357         }
 358
 359         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 360                         sizeof(struct ether_addr));
 361         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 362                         sizeof(struct ether_addr));
 363 }
 364
 365 static inline void
 366 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 367                 const struct lcore_conf *qconf)
 368 {
 369         int32_t i;
 370         const int32_t prefetch_offset = 2;
 371
 372         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 373                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 374                 prepare_tx_pkt(pkts[i], port, qconf);
 375         }
 376         /* Process left packets */
 377         for (; i < nb_pkts; i++)
 378                 prepare_tx_pkt(pkts[i], port, qconf);
 379 }
 380
 381 /* Send burst of packets on an output interface */
 382 static inline int32_t
 383 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 384 {
 385         struct rte_mbuf **m_table;
 386         int32_t ret;
 387         uint16_t queueid;
 388
 389         queueid = qconf->tx_queue_id[port];
 390         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 391
 392         prepare_tx_burst(m_table, n, port, qconf);
 393
 394         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 395         if (unlikely(ret < n)) {
 396                 do {
 397                         rte_pktmbuf_free(m_table[ret]);
 398                 } while (++ret < n);
 399         }
 400
 401         return 0;
 402 }
 403
 404 /* Enqueue a single packet, and send burst if queue is filled */
 405 static inline int32_t
 406 send_single_packet(struct rte_mbuf *m, uint16_t port)
 407 {
 408         uint32_t lcore_id;
 409         uint16_t len;
 410         struct lcore_conf *qconf;
 411
 412         lcore_id = rte_lcore_id();
 413
 414         qconf = &lcore_conf[lcore_id];
 415         len = qconf->tx_mbufs[port].len;
 416         qconf->tx_mbufs[port].m_table[len] = m;
 417         len++;
 418
 419         /* enough pkts to be sent */
 420         if (unlikely(len == MAX_PKT_BURST)) {
 421                 send_burst(qconf, MAX_PKT_BURST, port);
 422                 len = 0;
 423         }
 424
 425         qconf->tx_mbufs[port].len = len;
 426         return 0;
 427 }
 428
 429 static inline void
 430 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 431                 uint16_t lim)
 432 {
 433         struct rte_mbuf *m;
 434         uint32_t i, j, res, sa_idx;
 435
 436         if (ip->num == 0 || sp == NULL)
 437                 return;
 438
 439         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 440                         ip->num, DEFAULT_MAX_CATEGORIES);
 441
 442         j = 0;
 443         for (i = 0; i < ip->num; i++) {
 444                 m = ip->pkts[i];
 445                 res = ip->res[i];
 446                 if (res & BYPASS) {
 447                         ip->pkts[j++] = m;
 448                         continue;
 449                 }
 450                 if (res & DISCARD) {
 451                         rte_pktmbuf_free(m);
 452                         continue;
 453                 }
 454
 455                 /* Only check SPI match for processed IPSec packets */
 456                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 457                         rte_pktmbuf_free(m);
 458                         continue;
 459                 }
 460
 461                 sa_idx = ip->res[i] & PROTECT_MASK;
 462                 if (sa_idx >= IPSEC_SA_MAX_ENTRIES ||
 463                                 !inbound_sa_check(sa, m, sa_idx)) {
 464                         rte_pktmbuf_free(m);
 465                         continue;
 466                 }
 467                 ip->pkts[j++] = m;
 468         }
 469         ip->num = j;
 470 }
 471
 472 static void
 473 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 474 {
 475         uint32_t i, n4, n6;
 476         struct ip *ip;
 477         struct rte_mbuf *m;
 478
 479         n4 = trf->ip4.num;
 480         n6 = trf->ip6.num;
 481
 482         for (i = 0; i < num; i++) {
 483
 484                 m = mb[i];
 485                 ip = rte_pktmbuf_mtod(m, struct ip *);
 486
 487                 if (ip->ip_v == IPVERSION) {
 488                         trf->ip4.pkts[n4] = m;
 489                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 490                                         uint8_t *, offsetof(struct ip, ip_p));
 491                         n4++;
 492                 } else if (ip->ip_v == IP6_VERSION) {
 493                         trf->ip6.pkts[n6] = m;
 494                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 495                                         uint8_t *,
 496                                         offsetof(struct ip6_hdr, ip6_nxt));
 497                         n6++;
 498                 } else
 499                         rte_pktmbuf_free(m);
 500         }
 501
 502         trf->ip4.num = n4;
 503         trf->ip6.num = n6;
 504 }
 505
 506
 507 static inline void
 508 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 509                 struct ipsec_traffic *traffic)
 510 {
 511         uint16_t nb_pkts_in, n_ip4, n_ip6;
 512
 513         n_ip4 = traffic->ip4.num;
 514         n_ip6 = traffic->ip6.num;
 515
 516         nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 517                         traffic->ipsec.num, MAX_PKT_BURST);
 518
 519         split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 520
 521         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 522                         n_ip4);
 523
 524         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 525                         n_ip6);
 526 }
 527
 528 static inline void
 529 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 530                 struct traffic_type *ipsec)
 531 {
 532         struct rte_mbuf *m;
 533         uint32_t i, j, sa_idx;
 534
 535         if (ip->num == 0 || sp == NULL)
 536                 return;
 537
 538         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 539                         ip->num, DEFAULT_MAX_CATEGORIES);
 540
 541         j = 0;
 542         for (i = 0; i < ip->num; i++) {
 543                 m = ip->pkts[i];
 544                 sa_idx = ip->res[i] & PROTECT_MASK;
 545                 if (ip->res[i] & DISCARD)
 546                         rte_pktmbuf_free(m);
 547                 else if (ip->res[i] & BYPASS)
 548                         ip->pkts[j++] = m;
 549                 else if (sa_idx < IPSEC_SA_MAX_ENTRIES) {
 550                         ipsec->res[ipsec->num] = sa_idx;
 551                         ipsec->pkts[ipsec->num++] = m;
 552                 } else /* invalid SA idx */
 553                         rte_pktmbuf_free(m);
 554         }
 555         ip->num = j;
 556 }
 557
 558 static inline void
 559 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 560                 struct ipsec_traffic *traffic)
 561 {
 562         struct rte_mbuf *m;
 563         uint16_t idx, nb_pkts_out, i;
 564
 565         /* Drop any IPsec traffic from protected ports */
 566         for (i = 0; i < traffic->ipsec.num; i++)
 567                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 568
 569         traffic->ipsec.num = 0;
 570
 571         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 572
 573         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 574
 575         nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 576                         traffic->ipsec.res, traffic->ipsec.num,
 577                         MAX_PKT_BURST);
 578
 579         for (i = 0; i < nb_pkts_out; i++) {
 580                 m = traffic->ipsec.pkts[i];
 581                 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 582                 if (ip->ip_v == IPVERSION) {
 583                         idx = traffic->ip4.num++;
 584                         traffic->ip4.pkts[idx] = m;
 585                 } else {
 586                         idx = traffic->ip6.num++;
 587                         traffic->ip6.pkts[idx] = m;
 588                 }
 589         }
 590 }
 591
 592 static inline void
 593 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 594                 struct ipsec_traffic *traffic)
 595 {
 596         struct rte_mbuf *m;
 597         uint32_t nb_pkts_in, i, idx;
 598
 599         /* Drop any IPv4 traffic from unprotected ports */
 600         for (i = 0; i < traffic->ip4.num; i++)
 601                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 602
 603         traffic->ip4.num = 0;
 604
 605         /* Drop any IPv6 traffic from unprotected ports */
 606         for (i = 0; i < traffic->ip6.num; i++)
 607                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 608
 609         traffic->ip6.num = 0;
 610
 611         nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 612                         traffic->ipsec.num, MAX_PKT_BURST);
 613
 614         for (i = 0; i < nb_pkts_in; i++) {
 615                 m = traffic->ipsec.pkts[i];
 616                 struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 617                 if (ip->ip_v == IPVERSION) {
 618                         idx = traffic->ip4.num++;
 619                         traffic->ip4.pkts[idx] = m;
 620                 } else {
 621                         idx = traffic->ip6.num++;
 622                         traffic->ip6.pkts[idx] = m;
 623                 }
 624         }
 625 }
 626
 627 static inline void
 628 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 629                 struct ipsec_traffic *traffic)
 630 {
 631         struct rte_mbuf *m;
 632         uint32_t nb_pkts_out, i, n;
 633         struct ip *ip;
 634
 635         /* Drop any IPsec traffic from protected ports */
 636         for (i = 0; i < traffic->ipsec.num; i++)
 637                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 638
 639         n = 0;
 640
 641         for (i = 0; i < traffic->ip4.num; i++) {
 642                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 643                 traffic->ipsec.res[n++] = single_sa_idx;
 644         }
 645
 646         for (i = 0; i < traffic->ip6.num; i++) {
 647                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 648                 traffic->ipsec.res[n++] = single_sa_idx;
 649         }
 650
 651         traffic->ip4.num = 0;
 652         traffic->ip6.num = 0;
 653         traffic->ipsec.num = n;
 654
 655         nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 656                         traffic->ipsec.res, traffic->ipsec.num,
 657                         MAX_PKT_BURST);
 658
 659         /* They all sue the same SA (ip4 or ip6 tunnel) */
 660         m = traffic->ipsec.pkts[i];
 661         ip = rte_pktmbuf_mtod(m, struct ip *);
 662         if (ip->ip_v == IPVERSION) {
 663                 traffic->ip4.num = nb_pkts_out;
 664                 for (i = 0; i < nb_pkts_out; i++)
 665                         traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 666         } else {
 667                 traffic->ip6.num = nb_pkts_out;
 668                 for (i = 0; i < nb_pkts_out; i++)
 669                         traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 670         }
 671 }
 672
 673 static inline int32_t
 674 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 675 {
 676         struct ipsec_mbuf_metadata *priv;
 677         struct ipsec_sa *sa;
 678
 679         priv = get_priv(pkt);
 680
 681         sa = priv->sa;
 682         if (unlikely(sa == NULL)) {
 683                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 684                 goto fail;
 685         }
 686
 687         if (is_ipv6)
 688                 return sa->portid;
 689
 690         /* else */
 691         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 692
 693 fail:
 694         if (is_ipv6)
 695                 return -1;
 696
 697         /* else */
 698         return 0;
 699 }
 700
 701 static inline void
 702 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 703 {
 704         uint32_t hop[MAX_PKT_BURST * 2];
 705         uint32_t dst_ip[MAX_PKT_BURST * 2];
 706         int32_t pkt_hop = 0;
 707         uint16_t i, offset;
 708         uint16_t lpm_pkts = 0;
 709
 710         if (nb_pkts == 0)
 711                 return;
 712
 713         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 714          * have port ID in the SA
 715          */
 716
 717         for (i = 0; i < nb_pkts; i++) {
 718                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 719                         /* Security offload not enabled. So an LPM lookup is
 720                          * required to get the hop
 721                          */
 722                         offset = offsetof(struct ip, ip_dst);
 723                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 724                                         uint32_t *, offset);
 725                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 726                         lpm_pkts++;
 727                 }
 728         }
 729
 730         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 731
 732         lpm_pkts = 0;
 733
 734         for (i = 0; i < nb_pkts; i++) {
 735                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 736                         /* Read hop from the SA */
 737                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 738                 } else {
 739                         /* Need to use hop returned by lookup */
 740                         pkt_hop = hop[lpm_pkts++];
 741                 }
 742
 743                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 744                         rte_pktmbuf_free(pkts[i]);
 745                         continue;
 746                 }
 747                 send_single_packet(pkts[i], pkt_hop & 0xff);
 748         }
 749 }
 750
 751 static inline void
 752 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 753 {
 754         int32_t hop[MAX_PKT_BURST * 2];
 755         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 756         uint8_t *ip6_dst;
 757         int32_t pkt_hop = 0;
 758         uint16_t i, offset;
 759         uint16_t lpm_pkts = 0;
 760
 761         if (nb_pkts == 0)
 762                 return;
 763
 764         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 765          * have port ID in the SA
 766          */
 767
 768         for (i = 0; i < nb_pkts; i++) {
 769                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 770                         /* Security offload not enabled. So an LPM lookup is
 771                          * required to get the hop
 772                          */
 773                         offset = offsetof(struct ip6_hdr, ip6_dst);
 774                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 775                                         offset);
 776                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 777                         lpm_pkts++;
 778                 }
 779         }
 780
 781         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 782                         lpm_pkts);
 783
 784         lpm_pkts = 0;
 785
 786         for (i = 0; i < nb_pkts; i++) {
 787                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 788                         /* Read hop from the SA */
 789                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 790                 } else {
 791                         /* Need to use hop returned by lookup */
 792                         pkt_hop = hop[lpm_pkts++];
 793                 }
 794
 795                 if (pkt_hop == -1) {
 796                         rte_pktmbuf_free(pkts[i]);
 797                         continue;
 798                 }
 799                 send_single_packet(pkts[i], pkt_hop & 0xff);
 800         }
 801 }
 802
 803 static inline void
 804 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 805                 uint8_t nb_pkts, uint16_t portid)
 806 {
 807         struct ipsec_traffic traffic;
 808
 809         prepare_traffic(pkts, &traffic, nb_pkts);
 810
 811         if (unlikely(single_sa)) {
 812                 if (UNPROTECTED_PORT(portid))
 813                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 814                 else
 815                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 816         } else {
 817                 if (UNPROTECTED_PORT(portid))
 818                         process_pkts_inbound(&qconf->inbound, &traffic);
 819                 else
 820                         process_pkts_outbound(&qconf->outbound, &traffic);
 821         }
 822
 823         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 824         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 825 }
 826
 827 static inline void
 828 drain_tx_buffers(struct lcore_conf *qconf)
 829 {
 830         struct buffer *buf;
 831         uint32_t portid;
 832
 833         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 834                 buf = &qconf->tx_mbufs[portid];
 835                 if (buf->len == 0)
 836                         continue;
 837                 send_burst(qconf, buf->len, portid);
 838                 buf->len = 0;
 839         }
 840 }
 841
 842 static inline void
 843 drain_crypto_buffers(struct lcore_conf *qconf)
 844 {
 845         uint32_t i;
 846         struct ipsec_ctx *ctx;
 847
 848         /* drain inbound buffers*/
 849         ctx = &qconf->inbound;
 850         for (i = 0; i != ctx->nb_qps; i++) {
 851                 if (ctx->tbl[i].len != 0)
 852                         enqueue_cop_burst(ctx->tbl  + i);
 853         }
 854
 855         /* drain outbound buffers*/
 856         ctx = &qconf->outbound;
 857         for (i = 0; i != ctx->nb_qps; i++) {
 858                 if (ctx->tbl[i].len != 0)
 859                         enqueue_cop_burst(ctx->tbl  + i);
 860         }
 861 }
 862
 863 static void
 864 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
 865                 struct ipsec_ctx *ctx)
 866 {
 867         uint32_t n;
 868         struct ipsec_traffic trf;
 869
 870         /* dequeue packets from crypto-queue */
 871         n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
 872                         RTE_DIM(trf.ipsec.pkts));
 873         if (n == 0)
 874                 return;
 875
 876         trf.ip4.num = 0;
 877         trf.ip6.num = 0;
 878
 879         /* split traffic by ipv4-ipv6 */
 880         split46_traffic(&trf, trf.ipsec.pkts, n);
 881
 882         /* process ipv4 packets */
 883         inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
 884         route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
 885
 886         /* process ipv6 packets */
 887         inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
 888         route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
 889 }
 890
 891 static void
 892 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
 893                 struct ipsec_ctx *ctx)
 894 {
 895         uint32_t n;
 896         struct ipsec_traffic trf;
 897
 898         /* dequeue packets from crypto-queue */
 899         n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
 900                         RTE_DIM(trf.ipsec.pkts));
 901         if (n == 0)
 902                 return;
 903
 904         trf.ip4.num = 0;
 905         trf.ip6.num = 0;
 906
 907         /* split traffic by ipv4-ipv6 */
 908         split46_traffic(&trf, trf.ipsec.pkts, n);
 909
 910         /* process ipv4 packets */
 911         route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
 912
 913         /* process ipv6 packets */
 914         route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
 915 }
 916
 917 /* main processing loop */
 918 static int32_t
 919 main_loop(__attribute__((unused)) void *dummy)
 920 {
 921         struct rte_mbuf *pkts[MAX_PKT_BURST];
 922         uint32_t lcore_id;
 923         uint64_t prev_tsc, diff_tsc, cur_tsc;
 924         int32_t i, nb_rx;
 925         uint16_t portid;
 926         uint8_t queueid;
 927         struct lcore_conf *qconf;
 928         int32_t socket_id;
 929         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
 930                         / US_PER_S * BURST_TX_DRAIN_US;
 931         struct lcore_rx_queue *rxql;
 932
 933         prev_tsc = 0;
 934         lcore_id = rte_lcore_id();
 935         qconf = &lcore_conf[lcore_id];
 936         rxql = qconf->rx_queue_list;
 937         socket_id = rte_lcore_to_socket_id(lcore_id);
 938
 939         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
 940         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
 941         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
 942         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
 943         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
 944         qconf->inbound.cdev_map = cdev_map_in;
 945         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
 946         qconf->inbound.session_priv_pool =
 947                         socket_ctx[socket_id].session_priv_pool;
 948         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
 949         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
 950         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
 951         qconf->outbound.cdev_map = cdev_map_out;
 952         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
 953         qconf->outbound.session_priv_pool =
 954                         socket_ctx[socket_id].session_priv_pool;
 955
 956         if (qconf->nb_rx_queue == 0) {
 957                 RTE_LOG(INFO, IPSEC, "lcore %u has nothing to do\n", lcore_id);
 958                 return 0;
 959         }
 960
 961         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
 962
 963         for (i = 0; i < qconf->nb_rx_queue; i++) {
 964                 portid = rxql[i].port_id;
 965                 queueid = rxql[i].queue_id;
 966                 RTE_LOG(INFO, IPSEC,
 967                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 968                         lcore_id, portid, queueid);
 969         }
 970
 971         while (1) {
 972                 cur_tsc = rte_rdtsc();
 973
 974                 /* TX queue buffer drain */
 975                 diff_tsc = cur_tsc - prev_tsc;
 976
 977                 if (unlikely(diff_tsc > drain_tsc)) {
 978                         drain_tx_buffers(qconf);
 979                         drain_crypto_buffers(qconf);
 980                         prev_tsc = cur_tsc;
 981                 }
 982
 983                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
 984
 985                         /* Read packets from RX queues */
 986                         portid = rxql[i].port_id;
 987                         queueid = rxql[i].queue_id;
 988                         nb_rx = rte_eth_rx_burst(portid, queueid,
 989                                         pkts, MAX_PKT_BURST);
 990
 991                         if (nb_rx > 0)
 992                                 process_pkts(qconf, pkts, nb_rx, portid);
 993
 994                         /* dequeue and process completed crypto-ops */
 995                         if (UNPROTECTED_PORT(portid))
 996                                 drain_inbound_crypto_queues(qconf,
 997                                         &qconf->inbound);
 998                         else
 999                                 drain_outbound_crypto_queues(qconf,
1000                                         &qconf->outbound);
1001                 }
1002         }
1003 }
1004
1005 static int32_t
1006 check_params(void)
1007 {
1008         uint8_t lcore;
1009         uint16_t portid;
1010         uint16_t i;
1011         int32_t socket_id;
1012
1013         if (lcore_params == NULL) {
1014                 printf("Error: No port/queue/core mappings\n");
1015                 return -1;
1016         }
1017
1018         for (i = 0; i < nb_lcore_params; ++i) {
1019                 lcore = lcore_params[i].lcore_id;
1020                 if (!rte_lcore_is_enabled(lcore)) {
1021                         printf("error: lcore %hhu is not enabled in "
1022                                 "lcore mask\n", lcore);
1023                         return -1;
1024                 }
1025                 socket_id = rte_lcore_to_socket_id(lcore);
1026                 if (socket_id != 0 && numa_on == 0) {
1027                         printf("warning: lcore %hhu is on socket %d "
1028                                 "with numa off\n",
1029                                 lcore, socket_id);
1030                 }
1031                 portid = lcore_params[i].port_id;
1032                 if ((enabled_port_mask & (1 << portid)) == 0) {
1033                         printf("port %u is not enabled in port mask\n", portid);
1034                         return -1;
1035                 }
1036                 if (!rte_eth_dev_is_valid_port(portid)) {
1037                         printf("port %u is not present on the board\n", portid);
1038                         return -1;
1039                 }
1040         }
1041         return 0;
1042 }
1043
1044 static uint8_t
1045 get_port_nb_rx_queues(const uint16_t port)
1046 {
1047         int32_t queue = -1;
1048         uint16_t i;
1049
1050         for (i = 0; i < nb_lcore_params; ++i) {
1051                 if (lcore_params[i].port_id == port &&
1052                                 lcore_params[i].queue_id > queue)
1053                         queue = lcore_params[i].queue_id;
1054         }
1055         return (uint8_t)(++queue);
1056 }
1057
1058 static int32_t
1059 init_lcore_rx_queues(void)
1060 {
1061         uint16_t i, nb_rx_queue;
1062         uint8_t lcore;
1063
1064         for (i = 0; i < nb_lcore_params; ++i) {
1065                 lcore = lcore_params[i].lcore_id;
1066                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1067                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1068                         printf("error: too many queues (%u) for lcore: %u\n",
1069                                         nb_rx_queue + 1, lcore);
1070                         return -1;
1071                 }
1072                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1073                         lcore_params[i].port_id;
1074                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1075                         lcore_params[i].queue_id;
1076                 lcore_conf[lcore].nb_rx_queue++;
1077         }
1078         return 0;
1079 }
1080
1081 /* display usage */
1082 static void
1083 print_usage(const char *prgname)
1084 {
1085         fprintf(stderr, "%s [EAL options] --"
1086                 " -p PORTMASK"
1087                 " [-P]"
1088                 " [-u PORTMASK]"
1089                 " [-j FRAMESIZE]"
1090                 " -f CONFIG_FILE"
1091                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1092                 " [--single-sa SAIDX]"
1093                 " [--cryptodev_mask MASK]"
1094                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1095                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1096                 "\n\n"
1097                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1098                 "  -P : Enable promiscuous mode\n"
1099                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1100                 "  -j FRAMESIZE: Enable jumbo frame with 'FRAMESIZE' as maximum\n"
1101                 "                packet size\n"
1102                 "  -f CONFIG_FILE: Configuration file\n"
1103                 "  --config (port,queue,lcore): Rx queue configuration\n"
1104                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1105                 "                     bypassing the SP\n"
1106                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1107                 "                         devices to configure\n"
1108                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1109                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1110                 "                         (DEV_RX_OFFLOAD_*)\n"
1111                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1112                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1113                 "                         (DEV_TX_OFFLOAD_*)\n"
1114                 "\n",
1115                 prgname);
1116 }
1117
1118 static int
1119 parse_mask(const char *str, uint64_t *val)
1120 {
1121         char *end;
1122         unsigned long t;
1123
1124         errno = 0;
1125         t = strtoul(str, &end, 0);
1126         if (errno != 0 || end[0] != 0)
1127                 return -EINVAL;
1128
1129         *val = t;
1130         return 0;
1131 }
1132
1133 static int32_t
1134 parse_portmask(const char *portmask)
1135 {
1136         char *end = NULL;
1137         unsigned long pm;
1138
1139         /* parse hexadecimal string */
1140         pm = strtoul(portmask, &end, 16);
1141         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1142                 return -1;
1143
1144         if ((pm == 0) && errno)
1145                 return -1;
1146
1147         return pm;
1148 }
1149
1150 static int32_t
1151 parse_decimal(const char *str)
1152 {
1153         char *end = NULL;
1154         unsigned long num;
1155
1156         num = strtoul(str, &end, 10);
1157         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
1158                 return -1;
1159
1160         return num;
1161 }
1162
1163 static int32_t
1164 parse_config(const char *q_arg)
1165 {
1166         char s[256];
1167         const char *p, *p0 = q_arg;
1168         char *end;
1169         enum fieldnames {
1170                 FLD_PORT = 0,
1171                 FLD_QUEUE,
1172                 FLD_LCORE,
1173                 _NUM_FLD
1174         };
1175         unsigned long int_fld[_NUM_FLD];
1176         char *str_fld[_NUM_FLD];
1177         int32_t i;
1178         uint32_t size;
1179
1180         nb_lcore_params = 0;
1181
1182         while ((p = strchr(p0, '(')) != NULL) {
1183                 ++p;
1184                 p0 = strchr(p, ')');
1185                 if (p0 == NULL)
1186                         return -1;
1187
1188                 size = p0 - p;
1189                 if (size >= sizeof(s))
1190                         return -1;
1191
1192                 snprintf(s, sizeof(s), "%.*s", size, p);
1193                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1194                                 _NUM_FLD)
1195                         return -1;
1196                 for (i = 0; i < _NUM_FLD; i++) {
1197                         errno = 0;
1198                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1199                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1200                                 return -1;
1201                 }
1202                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1203                         printf("exceeded max number of lcore params: %hu\n",
1204                                 nb_lcore_params);
1205                         return -1;
1206                 }
1207                 lcore_params_array[nb_lcore_params].port_id =
1208                         (uint8_t)int_fld[FLD_PORT];
1209                 lcore_params_array[nb_lcore_params].queue_id =
1210                         (uint8_t)int_fld[FLD_QUEUE];
1211                 lcore_params_array[nb_lcore_params].lcore_id =
1212                         (uint8_t)int_fld[FLD_LCORE];
1213                 ++nb_lcore_params;
1214         }
1215         lcore_params = lcore_params_array;
1216         return 0;
1217 }
1218
1219 static int32_t
1220 parse_args(int32_t argc, char **argv)
1221 {
1222         int32_t opt, ret;
1223         char **argvopt;
1224         int32_t option_index;
1225         char *prgname = argv[0];
1226         int32_t f_present = 0;
1227
1228         argvopt = argv;
1229
1230         while ((opt = getopt_long(argc, argvopt, "p:Pu:f:j:",
1231                                 lgopts, &option_index)) != EOF) {
1232
1233                 switch (opt) {
1234                 case 'p':
1235                         enabled_port_mask = parse_portmask(optarg);
1236                         if (enabled_port_mask == 0) {
1237                                 printf("invalid portmask\n");
1238                                 print_usage(prgname);
1239                                 return -1;
1240                         }
1241                         break;
1242                 case 'P':
1243                         printf("Promiscuous mode selected\n");
1244                         promiscuous_on = 1;
1245                         break;
1246                 case 'u':
1247                         unprotected_port_mask = parse_portmask(optarg);
1248                         if (unprotected_port_mask == 0) {
1249                                 printf("invalid unprotected portmask\n");
1250                                 print_usage(prgname);
1251                                 return -1;
1252                         }
1253                         break;
1254                 case 'f':
1255                         if (f_present == 1) {
1256                                 printf("\"-f\" option present more than "
1257                                         "once!\n");
1258                                 print_usage(prgname);
1259                                 return -1;
1260                         }
1261                         if (parse_cfg_file(optarg) < 0) {
1262                                 printf("parsing file \"%s\" failed\n",
1263                                         optarg);
1264                                 print_usage(prgname);
1265                                 return -1;
1266                         }
1267                         f_present = 1;
1268                         break;
1269                 case 'j':
1270                         {
1271                                 int32_t size = parse_decimal(optarg);
1272                                 if (size <= 1518) {
1273                                         printf("Invalid jumbo frame size\n");
1274                                         if (size < 0) {
1275                                                 print_usage(prgname);
1276                                                 return -1;
1277                                         }
1278                                         printf("Using default value 9000\n");
1279                                         frame_size = 9000;
1280                                 } else {
1281                                         frame_size = size;
1282                                 }
1283                         }
1284                         printf("Enabled jumbo frames size %u\n", frame_size);
1285                         break;
1286                 case CMD_LINE_OPT_CONFIG_NUM:
1287                         ret = parse_config(optarg);
1288                         if (ret) {
1289                                 printf("Invalid config\n");
1290                                 print_usage(prgname);
1291                                 return -1;
1292                         }
1293                         break;
1294                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1295                         ret = parse_decimal(optarg);
1296                         if (ret == -1) {
1297                                 printf("Invalid argument[sa_idx]\n");
1298                                 print_usage(prgname);
1299                                 return -1;
1300                         }
1301
1302                         /* else */
1303                         single_sa = 1;
1304                         single_sa_idx = ret;
1305                         printf("Configured with single SA index %u\n",
1306                                         single_sa_idx);
1307                         break;
1308                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1309                         ret = parse_portmask(optarg);
1310                         if (ret == -1) {
1311                                 printf("Invalid argument[portmask]\n");
1312                                 print_usage(prgname);
1313                                 return -1;
1314                         }
1315
1316                         /* else */
1317                         enabled_cryptodev_mask = ret;
1318                         break;
1319                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1320                         ret = parse_mask(optarg, &dev_rx_offload);
1321                         if (ret != 0) {
1322                                 printf("Invalid argument for \'%s\': %s\n",
1323                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1324                                 print_usage(prgname);
1325                                 return -1;
1326                         }
1327                         break;
1328                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1329                         ret = parse_mask(optarg, &dev_tx_offload);
1330                         if (ret != 0) {
1331                                 printf("Invalid argument for \'%s\': %s\n",
1332                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1333                                 print_usage(prgname);
1334                                 return -1;
1335                         }
1336                         break;
1337                 default:
1338                         print_usage(prgname);
1339                         return -1;
1340                 }
1341         }
1342
1343         if (f_present == 0) {
1344                 printf("Mandatory option \"-f\" not present\n");
1345                 return -1;
1346         }
1347
1348         if (optind >= 0)
1349                 argv[optind-1] = prgname;
1350
1351         ret = optind-1;
1352         optind = 1; /* reset getopt lib */
1353         return ret;
1354 }
1355
1356 static void
1357 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1358 {
1359         char buf[ETHER_ADDR_FMT_SIZE];
1360         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1361         printf("%s%s", name, buf);
1362 }
1363
1364 /*
1365  * Update destination ethaddr for the port.
1366  */
1367 int
1368 add_dst_ethaddr(uint16_t port, const struct ether_addr *addr)
1369 {
1370         if (port > RTE_DIM(ethaddr_tbl))
1371                 return -EINVAL;
1372
1373         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1374         return 0;
1375 }
1376
1377 /* Check the link status of all ports in up to 9s, and print them finally */
1378 static void
1379 check_all_ports_link_status(uint32_t port_mask)
1380 {
1381 #define CHECK_INTERVAL 100 /* 100ms */
1382 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1383         uint16_t portid;
1384         uint8_t count, all_ports_up, print_flag = 0;
1385         struct rte_eth_link link;
1386
1387         printf("\nChecking link status");
1388         fflush(stdout);
1389         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1390                 all_ports_up = 1;
1391                 RTE_ETH_FOREACH_DEV(portid) {
1392                         if ((port_mask & (1 << portid)) == 0)
1393                                 continue;
1394                         memset(&link, 0, sizeof(link));
1395                         rte_eth_link_get_nowait(portid, &link);
1396                         /* print link status if flag set */
1397                         if (print_flag == 1) {
1398                                 if (link.link_status)
1399                                         printf(
1400                                         "Port%d Link Up - speed %u Mbps -%s\n",
1401                                                 portid, link.link_speed,
1402                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1403                                         ("full-duplex") : ("half-duplex\n"));
1404                                 else
1405                                         printf("Port %d Link Down\n", portid);
1406                                 continue;
1407                         }
1408                         /* clear all_ports_up flag if any link down */
1409                         if (link.link_status == ETH_LINK_DOWN) {
1410                                 all_ports_up = 0;
1411                                 break;
1412                         }
1413                 }
1414                 /* after finally printing all link status, get out */
1415                 if (print_flag == 1)
1416                         break;
1417
1418                 if (all_ports_up == 0) {
1419                         printf(".");
1420                         fflush(stdout);
1421                         rte_delay_ms(CHECK_INTERVAL);
1422                 }
1423
1424                 /* set the print_flag if all ports up or timeout */
1425                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1426                         print_flag = 1;
1427                         printf("done\n");
1428                 }
1429         }
1430 }
1431
1432 static int32_t
1433 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1434                 uint16_t qp, struct lcore_params *params,
1435                 struct ipsec_ctx *ipsec_ctx,
1436                 const struct rte_cryptodev_capabilities *cipher,
1437                 const struct rte_cryptodev_capabilities *auth,
1438                 const struct rte_cryptodev_capabilities *aead)
1439 {
1440         int32_t ret = 0;
1441         unsigned long i;
1442         struct cdev_key key = { 0 };
1443
1444         key.lcore_id = params->lcore_id;
1445         if (cipher)
1446                 key.cipher_algo = cipher->sym.cipher.algo;
1447         if (auth)
1448                 key.auth_algo = auth->sym.auth.algo;
1449         if (aead)
1450                 key.aead_algo = aead->sym.aead.algo;
1451
1452         ret = rte_hash_lookup(map, &key);
1453         if (ret != -ENOENT)
1454                 return 0;
1455
1456         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1457                 if (ipsec_ctx->tbl[i].id == cdev_id)
1458                         break;
1459
1460         if (i == ipsec_ctx->nb_qps) {
1461                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1462                         printf("Maximum number of crypto devices assigned to "
1463                                 "a core, increase MAX_QP_PER_LCORE value\n");
1464                         return 0;
1465                 }
1466                 ipsec_ctx->tbl[i].id = cdev_id;
1467                 ipsec_ctx->tbl[i].qp = qp;
1468                 ipsec_ctx->nb_qps++;
1469                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1470                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1471                                 cdev_id, qp, i);
1472         }
1473
1474         ret = rte_hash_add_key_data(map, &key, (void *)i);
1475         if (ret < 0) {
1476                 printf("Faled to insert cdev mapping for (lcore %u, "
1477                                 "cdev %u, qp %u), errno %d\n",
1478                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1479                                 ipsec_ctx->tbl[i].qp, ret);
1480                 return 0;
1481         }
1482
1483         return 1;
1484 }
1485
1486 static int32_t
1487 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1488                 uint16_t qp, struct lcore_params *params)
1489 {
1490         int32_t ret = 0;
1491         const struct rte_cryptodev_capabilities *i, *j;
1492         struct rte_hash *map;
1493         struct lcore_conf *qconf;
1494         struct ipsec_ctx *ipsec_ctx;
1495         const char *str;
1496
1497         qconf = &lcore_conf[params->lcore_id];
1498
1499         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1500                 map = cdev_map_out;
1501                 ipsec_ctx = &qconf->outbound;
1502                 str = "Outbound";
1503         } else {
1504                 map = cdev_map_in;
1505                 ipsec_ctx = &qconf->inbound;
1506                 str = "Inbound";
1507         }
1508
1509         /* Required cryptodevs with operation chainning */
1510         if (!(dev_info->feature_flags &
1511                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1512                 return ret;
1513
1514         for (i = dev_info->capabilities;
1515                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1516                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1517                         continue;
1518
1519                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1520                         ret |= add_mapping(map, str, cdev_id, qp, params,
1521                                         ipsec_ctx, NULL, NULL, i);
1522                         continue;
1523                 }
1524
1525                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1526                         continue;
1527
1528                 for (j = dev_info->capabilities;
1529                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1530                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1531                                 continue;
1532
1533                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1534                                 continue;
1535
1536                         ret |= add_mapping(map, str, cdev_id, qp, params,
1537                                                 ipsec_ctx, i, j, NULL);
1538                 }
1539         }
1540
1541         return ret;
1542 }
1543
1544 /* Check if the device is enabled by cryptodev_mask */
1545 static int
1546 check_cryptodev_mask(uint8_t cdev_id)
1547 {
1548         if (enabled_cryptodev_mask & (1 << cdev_id))
1549                 return 0;
1550
1551         return -1;
1552 }
1553
1554 static int32_t
1555 cryptodevs_init(void)
1556 {
1557         struct rte_cryptodev_config dev_conf;
1558         struct rte_cryptodev_qp_conf qp_conf;
1559         uint16_t idx, max_nb_qps, qp, i;
1560         int16_t cdev_id, port_id;
1561         struct rte_hash_parameters params = { 0 };
1562
1563         params.entries = CDEV_MAP_ENTRIES;
1564         params.key_len = sizeof(struct cdev_key);
1565         params.hash_func = rte_jhash;
1566         params.hash_func_init_val = 0;
1567         params.socket_id = rte_socket_id();
1568
1569         params.name = "cdev_map_in";
1570         cdev_map_in = rte_hash_create(&params);
1571         if (cdev_map_in == NULL)
1572                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1573                                 rte_errno);
1574
1575         params.name = "cdev_map_out";
1576         cdev_map_out = rte_hash_create(&params);
1577         if (cdev_map_out == NULL)
1578                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1579                                 rte_errno);
1580
1581         printf("lcore/cryptodev/qp mappings:\n");
1582
1583         uint32_t max_sess_sz = 0, sess_sz;
1584         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1585                 void *sec_ctx;
1586
1587                 /* Get crypto priv session size */
1588                 sess_sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
1589                 if (sess_sz > max_sess_sz)
1590                         max_sess_sz = sess_sz;
1591
1592                 /*
1593                  * If crypto device is security capable, need to check the
1594                  * size of security session as well.
1595                  */
1596
1597                 /* Get security context of the crypto device */
1598                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
1599                 if (sec_ctx == NULL)
1600                         continue;
1601
1602                 /* Get size of security session */
1603                 sess_sz = rte_security_session_get_size(sec_ctx);
1604                 if (sess_sz > max_sess_sz)
1605                         max_sess_sz = sess_sz;
1606         }
1607         RTE_ETH_FOREACH_DEV(port_id) {
1608                 void *sec_ctx;
1609
1610                 if ((enabled_port_mask & (1 << port_id)) == 0)
1611                         continue;
1612
1613                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
1614                 if (sec_ctx == NULL)
1615                         continue;
1616
1617                 sess_sz = rte_security_session_get_size(sec_ctx);
1618                 if (sess_sz > max_sess_sz)
1619                         max_sess_sz = sess_sz;
1620         }
1621
1622         idx = 0;
1623         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1624                 struct rte_cryptodev_info cdev_info;
1625
1626                 if (check_cryptodev_mask((uint8_t)cdev_id))
1627                         continue;
1628
1629                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1630
1631                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1632                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1633                 else
1634                         max_nb_qps = nb_lcore_params;
1635
1636                 qp = 0;
1637                 i = 0;
1638                 while (qp < max_nb_qps && i < nb_lcore_params) {
1639                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1640                                                 &lcore_params[idx]))
1641                                 qp++;
1642                         idx++;
1643                         idx = idx % nb_lcore_params;
1644                         i++;
1645                 }
1646
1647                 if (qp == 0)
1648                         continue;
1649
1650                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1651                 dev_conf.nb_queue_pairs = qp;
1652
1653                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1654                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1655                         rte_exit(EXIT_FAILURE,
1656                                 "Device does not support at least %u "
1657                                 "sessions", CDEV_MP_NB_OBJS);
1658
1659                 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1660                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1661                         struct rte_mempool *sess_mp;
1662
1663                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1664                                         "sess_mp_%u", dev_conf.socket_id);
1665                         sess_mp = rte_cryptodev_sym_session_pool_create(
1666                                         mp_name, CDEV_MP_NB_OBJS,
1667                                         0, CDEV_MP_CACHE_SZ, 0,
1668                                         dev_conf.socket_id);
1669                         socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1670                 }
1671
1672                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool) {
1673                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1674                         struct rte_mempool *sess_mp;
1675
1676                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1677                                         "sess_mp_priv_%u", dev_conf.socket_id);
1678                         sess_mp = rte_mempool_create(mp_name,
1679                                         CDEV_MP_NB_OBJS,
1680                                         max_sess_sz,
1681                                         CDEV_MP_CACHE_SZ,
1682                                         0, NULL, NULL, NULL,
1683                                         NULL, dev_conf.socket_id,
1684                                         0);
1685                         socket_ctx[dev_conf.socket_id].session_priv_pool =
1686                                         sess_mp;
1687                 }
1688
1689                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool ||
1690                                 !socket_ctx[dev_conf.socket_id].session_pool)
1691                         rte_exit(EXIT_FAILURE,
1692                                 "Cannot create session pool on socket %d\n",
1693                                 dev_conf.socket_id);
1694                 else
1695                         printf("Allocated session pool on socket %d\n",
1696                                         dev_conf.socket_id);
1697
1698                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1699                         rte_panic("Failed to initialize cryptodev %u\n",
1700                                         cdev_id);
1701
1702                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1703                 qp_conf.mp_session =
1704                         socket_ctx[dev_conf.socket_id].session_pool;
1705                 qp_conf.mp_session_private =
1706                         socket_ctx[dev_conf.socket_id].session_priv_pool;
1707                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1708                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1709                                         &qp_conf, dev_conf.socket_id))
1710                                 rte_panic("Failed to setup queue %u for "
1711                                                 "cdev_id %u\n", 0, cdev_id);
1712
1713                 if (rte_cryptodev_start(cdev_id))
1714                         rte_panic("Failed to start cryptodev %u\n",
1715                                         cdev_id);
1716         }
1717
1718         /* create session pools for eth devices that implement security */
1719         RTE_ETH_FOREACH_DEV(port_id) {
1720                 if ((enabled_port_mask & (1 << port_id)) &&
1721                                 rte_eth_dev_get_sec_ctx(port_id)) {
1722                         int socket_id = rte_eth_dev_socket_id(port_id);
1723
1724                         if (!socket_ctx[socket_id].session_pool) {
1725                                 char mp_name[RTE_MEMPOOL_NAMESIZE];
1726                                 struct rte_mempool *sess_mp;
1727
1728                                 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1729                                                 "sess_mp_%u", socket_id);
1730                                 sess_mp = rte_mempool_create(mp_name,
1731                                                 (CDEV_MP_NB_OBJS * 2),
1732                                                 max_sess_sz,
1733                                                 CDEV_MP_CACHE_SZ,
1734                                                 0, NULL, NULL, NULL,
1735                                                 NULL, socket_id,
1736                                                 0);
1737                                 if (sess_mp == NULL)
1738                                         rte_exit(EXIT_FAILURE,
1739                                                 "Cannot create session pool "
1740                                                 "on socket %d\n", socket_id);
1741                                 else
1742                                         printf("Allocated session pool "
1743                                                 "on socket %d\n", socket_id);
1744                                 socket_ctx[socket_id].session_pool = sess_mp;
1745                         }
1746                 }
1747         }
1748
1749
1750         printf("\n");
1751
1752         return 0;
1753 }
1754
1755 static void
1756 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
1757 {
1758         struct rte_eth_dev_info dev_info;
1759         struct rte_eth_txconf *txconf;
1760         uint16_t nb_tx_queue, nb_rx_queue;
1761         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1762         int32_t ret, socket_id;
1763         struct lcore_conf *qconf;
1764         struct ether_addr ethaddr;
1765         struct rte_eth_conf local_port_conf = port_conf;
1766
1767         rte_eth_dev_info_get(portid, &dev_info);
1768
1769         /* limit allowed HW offloafs, as user requested */
1770         dev_info.rx_offload_capa &= dev_rx_offload;
1771         dev_info.tx_offload_capa &= dev_tx_offload;
1772
1773         printf("Configuring device port %u:\n", portid);
1774
1775         rte_eth_macaddr_get(portid, &ethaddr);
1776         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
1777         print_ethaddr("Address: ", &ethaddr);
1778         printf("\n");
1779
1780         nb_rx_queue = get_port_nb_rx_queues(portid);
1781         nb_tx_queue = nb_lcores;
1782
1783         if (nb_rx_queue > dev_info.max_rx_queues)
1784                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1785                                 "(max rx queue is %u)\n",
1786                                 nb_rx_queue, dev_info.max_rx_queues);
1787
1788         if (nb_tx_queue > dev_info.max_tx_queues)
1789                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1790                                 "(max tx queue is %u)\n",
1791                                 nb_tx_queue, dev_info.max_tx_queues);
1792
1793         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1794                         nb_rx_queue, nb_tx_queue);
1795
1796         if (frame_size) {
1797                 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
1798                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1799         }
1800
1801         local_port_conf.rxmode.offloads |= req_rx_offloads;
1802         local_port_conf.txmode.offloads |= req_tx_offloads;
1803
1804         /* Check that all required capabilities are supported */
1805         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
1806                         local_port_conf.rxmode.offloads)
1807                 rte_exit(EXIT_FAILURE,
1808                         "Error: port %u required RX offloads: 0x%" PRIx64
1809                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
1810                         portid, local_port_conf.rxmode.offloads,
1811                         dev_info.rx_offload_capa);
1812
1813         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
1814                         local_port_conf.txmode.offloads)
1815                 rte_exit(EXIT_FAILURE,
1816                         "Error: port %u required TX offloads: 0x%" PRIx64
1817                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
1818                         portid, local_port_conf.txmode.offloads,
1819                         dev_info.tx_offload_capa);
1820
1821         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1822                 local_port_conf.txmode.offloads |=
1823                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1824
1825         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
1826                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
1827
1828         printf("port %u configurng rx_offloads=0x%" PRIx64
1829                 ", tx_offloads=0x%" PRIx64 "\n",
1830                 portid, local_port_conf.rxmode.offloads,
1831                 local_port_conf.txmode.offloads);
1832
1833         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1834                 dev_info.flow_type_rss_offloads;
1835         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1836                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
1837                 printf("Port %u modified RSS hash function based on hardware support,"
1838                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1839                         portid,
1840                         port_conf.rx_adv_conf.rss_conf.rss_hf,
1841                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1842         }
1843
1844         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1845                         &local_port_conf);
1846         if (ret < 0)
1847                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1848                                 "err=%d, port=%d\n", ret, portid);
1849
1850         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
1851         if (ret < 0)
1852                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
1853                                 "err=%d, port=%d\n", ret, portid);
1854
1855         /* init one TX queue per lcore */
1856         tx_queueid = 0;
1857         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1858                 if (rte_lcore_is_enabled(lcore_id) == 0)
1859                         continue;
1860
1861                 if (numa_on)
1862                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1863                 else
1864                         socket_id = 0;
1865
1866                 /* init TX queue */
1867                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1868
1869                 txconf = &dev_info.default_txconf;
1870                 txconf->offloads = local_port_conf.txmode.offloads;
1871
1872                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1873                                 socket_id, txconf);
1874                 if (ret < 0)
1875                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1876                                         "err=%d, port=%d\n", ret, portid);
1877
1878                 qconf = &lcore_conf[lcore_id];
1879                 qconf->tx_queue_id[portid] = tx_queueid;
1880
1881                 /* Pre-populate pkt offloads based on capabilities */
1882                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
1883                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
1884                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
1885                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
1886
1887                 tx_queueid++;
1888
1889                 /* init RX queues */
1890                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1891                         struct rte_eth_rxconf rxq_conf;
1892
1893                         if (portid != qconf->rx_queue_list[queue].port_id)
1894                                 continue;
1895
1896                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
1897
1898                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1899                                         socket_id);
1900
1901                         rxq_conf = dev_info.default_rxconf;
1902                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
1903                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1904                                         nb_rxd, socket_id, &rxq_conf,
1905                                         socket_ctx[socket_id].mbuf_pool);
1906                         if (ret < 0)
1907                                 rte_exit(EXIT_FAILURE,
1908                                         "rte_eth_rx_queue_setup: err=%d, "
1909                                         "port=%d\n", ret, portid);
1910                 }
1911         }
1912         printf("\n");
1913 }
1914
1915 static void
1916 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1917 {
1918         char s[64];
1919         uint32_t buff_size = frame_size ? (frame_size + RTE_PKTMBUF_HEADROOM) :
1920                         RTE_MBUF_DEFAULT_BUF_SIZE;
1921
1922
1923         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1924         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1925                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1926                         buff_size,
1927                         socket_id);
1928         if (ctx->mbuf_pool == NULL)
1929                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
1930                                 socket_id);
1931         else
1932                 printf("Allocated mbuf pool on socket %d\n", socket_id);
1933 }
1934
1935 static inline int
1936 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
1937 {
1938         struct ipsec_sa *sa;
1939
1940         /* For inline protocol processing, the metadata in the event will
1941          * uniquely identify the security session which raised the event.
1942          * Application would then need the userdata it had registered with the
1943          * security session to process the event.
1944          */
1945
1946         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
1947
1948         if (sa == NULL) {
1949                 /* userdata could not be retrieved */
1950                 return -1;
1951         }
1952
1953         /* Sequence number over flow. SA need to be re-established */
1954         RTE_SET_USED(sa);
1955         return 0;
1956 }
1957
1958 static int
1959 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
1960                  void *param, void *ret_param)
1961 {
1962         uint64_t md;
1963         struct rte_eth_event_ipsec_desc *event_desc = NULL;
1964         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
1965                                         rte_eth_dev_get_sec_ctx(port_id);
1966
1967         RTE_SET_USED(param);
1968
1969         if (type != RTE_ETH_EVENT_IPSEC)
1970                 return -1;
1971
1972         event_desc = ret_param;
1973         if (event_desc == NULL) {
1974                 printf("Event descriptor not set\n");
1975                 return -1;
1976         }
1977
1978         md = event_desc->metadata;
1979
1980         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
1981                 return inline_ipsec_event_esn_overflow(ctx, md);
1982         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
1983                 printf("Invalid IPsec event reported\n");
1984                 return -1;
1985         }
1986
1987         return -1;
1988 }
1989
1990 int32_t
1991 main(int32_t argc, char **argv)
1992 {
1993         int32_t ret;
1994         uint32_t lcore_id;
1995         uint8_t socket_id;
1996         uint16_t portid;
1997         uint64_t req_rx_offloads, req_tx_offloads;
1998
1999         /* init EAL */
2000         ret = rte_eal_init(argc, argv);
2001         if (ret < 0)
2002                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2003         argc -= ret;
2004         argv += ret;
2005
2006         /* parse application arguments (after the EAL ones) */
2007         ret = parse_args(argc, argv);
2008         if (ret < 0)
2009                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2010
2011         if ((unprotected_port_mask & enabled_port_mask) !=
2012                         unprotected_port_mask)
2013                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2014                                 unprotected_port_mask);
2015
2016         if (check_params() < 0)
2017                 rte_exit(EXIT_FAILURE, "check_params failed\n");
2018
2019         ret = init_lcore_rx_queues();
2020         if (ret < 0)
2021                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2022
2023         nb_lcores = rte_lcore_count();
2024
2025         /* Replicate each context per socket */
2026         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2027                 if (rte_lcore_is_enabled(lcore_id) == 0)
2028                         continue;
2029
2030                 if (numa_on)
2031                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2032                 else
2033                         socket_id = 0;
2034
2035                 if (socket_ctx[socket_id].mbuf_pool)
2036                         continue;
2037
2038                 sa_init(&socket_ctx[socket_id], socket_id);
2039
2040                 sp4_init(&socket_ctx[socket_id], socket_id);
2041
2042                 sp6_init(&socket_ctx[socket_id], socket_id);
2043
2044                 rt_init(&socket_ctx[socket_id], socket_id);
2045
2046                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2047         }
2048
2049         RTE_ETH_FOREACH_DEV(portid) {
2050                 if ((enabled_port_mask & (1 << portid)) == 0)
2051                         continue;
2052
2053                 sa_check_offloads(portid, &req_rx_offloads, &req_tx_offloads);
2054                 port_init(portid, req_rx_offloads, req_tx_offloads);
2055         }
2056
2057         cryptodevs_init();
2058
2059         /* start ports */
2060         RTE_ETH_FOREACH_DEV(portid) {
2061                 if ((enabled_port_mask & (1 << portid)) == 0)
2062                         continue;
2063
2064                 /* Start device */
2065                 ret = rte_eth_dev_start(portid);
2066                 if (ret < 0)
2067                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2068                                         "err=%d, port=%d\n", ret, portid);
2069                 /*
2070                  * If enabled, put device in promiscuous mode.
2071                  * This allows IO forwarding mode to forward packets
2072                  * to itself through 2 cross-connected  ports of the
2073                  * target machine.
2074                  */
2075                 if (promiscuous_on)
2076                         rte_eth_promiscuous_enable(portid);
2077
2078                 rte_eth_dev_callback_register(portid,
2079                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2080         }
2081
2082         check_all_ports_link_status(enabled_port_mask);
2083
2084         /* launch per-lcore init on every lcore */
2085         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2086         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2087                 if (rte_eal_wait_lcore(lcore_id) < 0)
2088                         return -1;
2089         }
2090
2091         return 0;
2092 }