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 #include <rte_ip.h>
  45 #include <rte_ip_frag.h>
  46
  47 #include "ipsec.h"
  48 #include "parser.h"
  49
  50 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  51
  52 #define MAX_JUMBO_PKT_LEN  9600
  53
  54 #define MEMPOOL_CACHE_SIZE 256
  55
  56 #define NB_MBUF (32000)
  57
  58 #define CDEV_QUEUE_DESC 2048
  59 #define CDEV_MAP_ENTRIES 16384
  60 #define CDEV_MP_NB_OBJS 1024
  61 #define CDEV_MP_CACHE_SZ 64
  62 #define MAX_QUEUE_PAIRS 1
  63
  64 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  65
  66 #define NB_SOCKETS 4
  67
  68 /* Configure how many packets ahead to prefetch, when reading packets */
  69 #define PREFETCH_OFFSET 3
  70
  71 #define MAX_RX_QUEUE_PER_LCORE 16
  72
  73 #define MAX_LCORE_PARAMS 1024
  74
  75 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
  76
  77 /*
  78  * Configurable number of RX/TX ring descriptors
  79  */
  80 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  81 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  82 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  83 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  84
  85 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
  86 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  87         (((uint64_t)((a) & 0xff) << 56) | \
  88         ((uint64_t)((b) & 0xff) << 48) | \
  89         ((uint64_t)((c) & 0xff) << 40) | \
  90         ((uint64_t)((d) & 0xff) << 32) | \
  91         ((uint64_t)((e) & 0xff) << 24) | \
  92         ((uint64_t)((f) & 0xff) << 16) | \
  93         ((uint64_t)((g) & 0xff) << 8)  | \
  94         ((uint64_t)(h) & 0xff))
  95 #else
  96 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  97         (((uint64_t)((h) & 0xff) << 56) | \
  98         ((uint64_t)((g) & 0xff) << 48) | \
  99         ((uint64_t)((f) & 0xff) << 40) | \
 100         ((uint64_t)((e) & 0xff) << 32) | \
 101         ((uint64_t)((d) & 0xff) << 24) | \
 102         ((uint64_t)((c) & 0xff) << 16) | \
 103         ((uint64_t)((b) & 0xff) << 8) | \
 104         ((uint64_t)(a) & 0xff))
 105 #endif
 106 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 107
 108 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 109                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
 110                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
 111                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
 112                 0, 0)
 113
 114 #define FRAG_TBL_BUCKET_ENTRIES 4
 115 #define FRAG_TTL_MS             (10 * MS_PER_S)
 116
 117 #define MTU_TO_FRAMELEN(x)      ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
 118
 119 /* port/source ethernet addr and destination ethernet addr */
 120 struct ethaddr_info {
 121         uint64_t src, dst;
 122 };
 123
 124 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 125         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 126         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 127         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 128         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 129 };
 130
 131 #define CMD_LINE_OPT_CONFIG             "config"
 132 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 133 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 134 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
 135 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
 136 #define CMD_LINE_OPT_REASSEMBLE         "reassemble"
 137 #define CMD_LINE_OPT_MTU                "mtu"
 138
 139 enum {
 140         /* long options mapped to a short option */
 141
 142         /* first long only option value must be >= 256, so that we won't
 143          * conflict with short options
 144          */
 145         CMD_LINE_OPT_MIN_NUM = 256,
 146         CMD_LINE_OPT_CONFIG_NUM,
 147         CMD_LINE_OPT_SINGLE_SA_NUM,
 148         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 149         CMD_LINE_OPT_RX_OFFLOAD_NUM,
 150         CMD_LINE_OPT_TX_OFFLOAD_NUM,
 151         CMD_LINE_OPT_REASSEMBLE_NUM,
 152         CMD_LINE_OPT_MTU_NUM,
 153 };
 154
 155 static const struct option lgopts[] = {
 156         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 157         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 158         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 159         {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
 160         {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
 161         {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
 162         {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
 163         {NULL, 0, 0, 0}
 164 };
 165
 166 /* mask of enabled ports */
 167 static uint32_t enabled_port_mask;
 168 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 169 static uint32_t unprotected_port_mask;
 170 static int32_t promiscuous_on = 1;
 171 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 172 static uint32_t nb_lcores;
 173 static uint32_t single_sa;
 174 static uint32_t single_sa_idx;
 175
 176 /*
 177  * RX/TX HW offload capabilities to enable/use on ethernet ports.
 178  * By default all capabilities are enabled.
 179  */
 180 static uint64_t dev_rx_offload = UINT64_MAX;
 181 static uint64_t dev_tx_offload = UINT64_MAX;
 182
 183 /*
 184  * global values that determine multi-seg policy
 185  */
 186 static uint32_t frag_tbl_sz;
 187 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
 188 static uint32_t mtu_size = RTE_ETHER_MTU;
 189
 190 /* application wide librte_ipsec/SA parameters */
 191 struct app_sa_prm app_sa_prm = {.enable = 0};
 192 static const char *cfgfile;
 193
 194 struct lcore_rx_queue {
 195         uint16_t port_id;
 196         uint8_t queue_id;
 197 } __rte_cache_aligned;
 198
 199 struct lcore_params {
 200         uint16_t port_id;
 201         uint8_t queue_id;
 202         uint8_t lcore_id;
 203 } __rte_cache_aligned;
 204
 205 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 206
 207 static struct lcore_params *lcore_params;
 208 static uint16_t nb_lcore_params;
 209
 210 static struct rte_hash *cdev_map_in;
 211 static struct rte_hash *cdev_map_out;
 212
 213 struct buffer {
 214         uint16_t len;
 215         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 216 };
 217
 218 struct lcore_conf {
 219         uint16_t nb_rx_queue;
 220         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 221         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 222         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 223         struct ipsec_ctx inbound;
 224         struct ipsec_ctx outbound;
 225         struct rt_ctx *rt4_ctx;
 226         struct rt_ctx *rt6_ctx;
 227         struct {
 228                 struct rte_ip_frag_tbl *tbl;
 229                 struct rte_mempool *pool_dir;
 230                 struct rte_mempool *pool_indir;
 231                 struct rte_ip_frag_death_row dr;
 232         } frag;
 233 } __rte_cache_aligned;
 234
 235 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 236
 237 static struct rte_eth_conf port_conf = {
 238         .rxmode = {
 239                 .mq_mode        = ETH_MQ_RX_RSS,
 240                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 241                 .split_hdr_size = 0,
 242                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 243         },
 244         .rx_adv_conf = {
 245                 .rss_conf = {
 246                         .rss_key = NULL,
 247                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 248                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 249                 },
 250         },
 251         .txmode = {
 252                 .mq_mode = ETH_MQ_TX_NONE,
 253         },
 254 };
 255
 256 static struct socket_ctx socket_ctx[NB_SOCKETS];
 257
 258 /*
 259  * Determine is multi-segment support required:
 260  *  - either frame buffer size is smaller then mtu
 261  *  - or reassmeble support is requested
 262  */
 263 static int
 264 multi_seg_required(void)
 265 {
 266         return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
 267                 frame_buf_size || frag_tbl_sz != 0);
 268 }
 269
 270 static inline void
 271 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
 272         uint32_t l2_len)
 273 {
 274         uint32_t plen, trim;
 275
 276         plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
 277         if (plen < m->pkt_len) {
 278                 trim = m->pkt_len - plen;
 279                 rte_pktmbuf_trim(m, trim);
 280         }
 281 }
 282
 283 static inline void
 284 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
 285         uint32_t l2_len)
 286 {
 287         uint32_t plen, trim;
 288
 289         plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
 290         if (plen < m->pkt_len) {
 291                 trim = m->pkt_len - plen;
 292                 rte_pktmbuf_trim(m, trim);
 293         }
 294 }
 295
 296 static inline void
 297 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 298 {
 299         const struct rte_ether_hdr *eth;
 300         const struct rte_ipv4_hdr *iph4;
 301         const struct rte_ipv6_hdr *iph6;
 302
 303         eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
 304         if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 305
 306                 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
 307                         RTE_ETHER_HDR_LEN);
 308                 adjust_ipv4_pktlen(pkt, iph4, 0);
 309
 310                 if (iph4->next_proto_id == IPPROTO_ESP)
 311                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 312                 else {
 313                         t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
 314                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 315                 }
 316                 pkt->l2_len = 0;
 317                 pkt->l3_len = sizeof(*iph4);
 318         } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 319                 int next_proto;
 320                 size_t l3len, ext_len;
 321                 uint8_t *p;
 322
 323                 /* get protocol type */
 324                 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
 325                         RTE_ETHER_HDR_LEN);
 326                 adjust_ipv6_pktlen(pkt, iph6, 0);
 327
 328                 next_proto = iph6->proto;
 329
 330                 /* determine l3 header size up to ESP extension */
 331                 l3len = sizeof(struct ip6_hdr);
 332                 p = rte_pktmbuf_mtod(pkt, uint8_t *);
 333                 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
 334                         (next_proto = rte_ipv6_get_next_ext(p + l3len,
 335                                                 next_proto, &ext_len)) >= 0)
 336                         l3len += ext_len;
 337
 338                 /* drop packet when IPv6 header exceeds first segment length */
 339                 if (unlikely(l3len > pkt->data_len)) {
 340                         rte_pktmbuf_free(pkt);
 341                         return;
 342                 }
 343
 344                 if (next_proto == IPPROTO_ESP)
 345                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 346                 else {
 347                         t->ip6.data[t->ip6.num] = &iph6->proto;
 348                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 349                 }
 350                 pkt->l2_len = 0;
 351                 pkt->l3_len = l3len;
 352         } else {
 353                 /* Unknown/Unsupported type, drop the packet */
 354                 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
 355                         rte_be_to_cpu_16(eth->ether_type));
 356                 rte_pktmbuf_free(pkt);
 357                 return;
 358         }
 359
 360         /* Check if the packet has been processed inline. For inline protocol
 361          * processed packets, the metadata in the mbuf can be used to identify
 362          * the security processing done on the packet. The metadata will be
 363          * used to retrieve the application registered userdata associated
 364          * with the security session.
 365          */
 366
 367         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 368                 struct ipsec_sa *sa;
 369                 struct ipsec_mbuf_metadata *priv;
 370                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 371                                                 rte_eth_dev_get_sec_ctx(
 372                                                 pkt->port);
 373
 374                 /* Retrieve the userdata registered. Here, the userdata
 375                  * registered is the SA pointer.
 376                  */
 377
 378                 sa = (struct ipsec_sa *)
 379                                 rte_security_get_userdata(ctx, pkt->udata64);
 380
 381                 if (sa == NULL) {
 382                         /* userdata could not be retrieved */
 383                         return;
 384                 }
 385
 386                 /* Save SA as priv member in mbuf. This will be used in the
 387                  * IPsec selector(SP-SA) check.
 388                  */
 389
 390                 priv = get_priv(pkt);
 391                 priv->sa = sa;
 392         }
 393 }
 394
 395 static inline void
 396 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 397                 uint16_t nb_pkts)
 398 {
 399         int32_t i;
 400
 401         t->ipsec.num = 0;
 402         t->ip4.num = 0;
 403         t->ip6.num = 0;
 404
 405         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 406                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 407                                         void *));
 408                 prepare_one_packet(pkts[i], t);
 409         }
 410         /* Process left packets */
 411         for (; i < nb_pkts; i++)
 412                 prepare_one_packet(pkts[i], t);
 413 }
 414
 415 static inline void
 416 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 417                 const struct lcore_conf *qconf)
 418 {
 419         struct ip *ip;
 420         struct rte_ether_hdr *ethhdr;
 421
 422         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 423
 424         ethhdr = (struct rte_ether_hdr *)
 425                 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
 426
 427         if (ip->ip_v == IPVERSION) {
 428                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 429                 pkt->l3_len = sizeof(struct ip);
 430                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 431
 432                 ip->ip_sum = 0;
 433
 434                 /* calculate IPv4 cksum in SW */
 435                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 436                         ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
 437
 438                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
 439         } else {
 440                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 441                 pkt->l3_len = sizeof(struct ip6_hdr);
 442                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 443
 444                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
 445         }
 446
 447         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 448                         sizeof(struct rte_ether_addr));
 449         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 450                         sizeof(struct rte_ether_addr));
 451 }
 452
 453 static inline void
 454 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 455                 const struct lcore_conf *qconf)
 456 {
 457         int32_t i;
 458         const int32_t prefetch_offset = 2;
 459
 460         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 461                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 462                 prepare_tx_pkt(pkts[i], port, qconf);
 463         }
 464         /* Process left packets */
 465         for (; i < nb_pkts; i++)
 466                 prepare_tx_pkt(pkts[i], port, qconf);
 467 }
 468
 469 /* Send burst of packets on an output interface */
 470 static inline int32_t
 471 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 472 {
 473         struct rte_mbuf **m_table;
 474         int32_t ret;
 475         uint16_t queueid;
 476
 477         queueid = qconf->tx_queue_id[port];
 478         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 479
 480         prepare_tx_burst(m_table, n, port, qconf);
 481
 482         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 483         if (unlikely(ret < n)) {
 484                 do {
 485                         rte_pktmbuf_free(m_table[ret]);
 486                 } while (++ret < n);
 487         }
 488
 489         return 0;
 490 }
 491
 492 /*
 493  * Helper function to fragment and queue for TX one packet.
 494  */
 495 static inline uint32_t
 496 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
 497         uint16_t port, uint8_t proto)
 498 {
 499         struct buffer *tbl;
 500         uint32_t len, n;
 501         int32_t rc;
 502
 503         tbl =  qconf->tx_mbufs + port;
 504         len = tbl->len;
 505
 506         /* free space for new fragments */
 507         if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >=  RTE_DIM(tbl->m_table)) {
 508                 send_burst(qconf, len, port);
 509                 len = 0;
 510         }
 511
 512         n = RTE_DIM(tbl->m_table) - len;
 513
 514         if (proto == IPPROTO_IP)
 515                 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
 516                         n, mtu_size, qconf->frag.pool_dir,
 517                         qconf->frag.pool_indir);
 518         else
 519                 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
 520                         n, mtu_size, qconf->frag.pool_dir,
 521                         qconf->frag.pool_indir);
 522
 523         if (rc >= 0)
 524                 len += rc;
 525         else
 526                 RTE_LOG(ERR, IPSEC,
 527                         "%s: failed to fragment packet with size %u, "
 528                         "error code: %d\n",
 529                         __func__, m->pkt_len, rte_errno);
 530
 531         rte_pktmbuf_free(m);
 532         return len;
 533 }
 534
 535 /* Enqueue a single packet, and send burst if queue is filled */
 536 static inline int32_t
 537 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
 538 {
 539         uint32_t lcore_id;
 540         uint16_t len;
 541         struct lcore_conf *qconf;
 542
 543         lcore_id = rte_lcore_id();
 544
 545         qconf = &lcore_conf[lcore_id];
 546         len = qconf->tx_mbufs[port].len;
 547
 548         if (m->pkt_len <= mtu_size) {
 549                 qconf->tx_mbufs[port].m_table[len] = m;
 550                 len++;
 551
 552         /* need to fragment the packet */
 553         } else if (frag_tbl_sz > 0)
 554                 len = send_fragment_packet(qconf, m, port, proto);
 555         else
 556                 rte_pktmbuf_free(m);
 557
 558         /* enough pkts to be sent */
 559         if (unlikely(len == MAX_PKT_BURST)) {
 560                 send_burst(qconf, MAX_PKT_BURST, port);
 561                 len = 0;
 562         }
 563
 564         qconf->tx_mbufs[port].len = len;
 565         return 0;
 566 }
 567
 568 static inline void
 569 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 570                 uint16_t lim)
 571 {
 572         struct rte_mbuf *m;
 573         uint32_t i, j, res, sa_idx;
 574
 575         if (ip->num == 0 || sp == NULL)
 576                 return;
 577
 578         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 579                         ip->num, DEFAULT_MAX_CATEGORIES);
 580
 581         j = 0;
 582         for (i = 0; i < ip->num; i++) {
 583                 m = ip->pkts[i];
 584                 res = ip->res[i];
 585                 if (res == BYPASS) {
 586                         ip->pkts[j++] = m;
 587                         continue;
 588                 }
 589                 if (res == DISCARD) {
 590                         rte_pktmbuf_free(m);
 591                         continue;
 592                 }
 593
 594                 /* Only check SPI match for processed IPSec packets */
 595                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 596                         rte_pktmbuf_free(m);
 597                         continue;
 598                 }
 599
 600                 sa_idx = SPI2IDX(res);
 601                 if (!inbound_sa_check(sa, m, sa_idx)) {
 602                         rte_pktmbuf_free(m);
 603                         continue;
 604                 }
 605                 ip->pkts[j++] = m;
 606         }
 607         ip->num = j;
 608 }
 609
 610 static void
 611 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 612 {
 613         uint32_t i, n4, n6;
 614         struct ip *ip;
 615         struct rte_mbuf *m;
 616
 617         n4 = trf->ip4.num;
 618         n6 = trf->ip6.num;
 619
 620         for (i = 0; i < num; i++) {
 621
 622                 m = mb[i];
 623                 ip = rte_pktmbuf_mtod(m, struct ip *);
 624
 625                 if (ip->ip_v == IPVERSION) {
 626                         trf->ip4.pkts[n4] = m;
 627                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 628                                         uint8_t *, offsetof(struct ip, ip_p));
 629                         n4++;
 630                 } else if (ip->ip_v == IP6_VERSION) {
 631                         trf->ip6.pkts[n6] = m;
 632                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 633                                         uint8_t *,
 634                                         offsetof(struct ip6_hdr, ip6_nxt));
 635                         n6++;
 636                 } else
 637                         rte_pktmbuf_free(m);
 638         }
 639
 640         trf->ip4.num = n4;
 641         trf->ip6.num = n6;
 642 }
 643
 644
 645 static inline void
 646 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 647                 struct ipsec_traffic *traffic)
 648 {
 649         uint16_t nb_pkts_in, n_ip4, n_ip6;
 650
 651         n_ip4 = traffic->ip4.num;
 652         n_ip6 = traffic->ip6.num;
 653
 654         if (app_sa_prm.enable == 0) {
 655                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 656                                 traffic->ipsec.num, MAX_PKT_BURST);
 657                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 658         } else {
 659                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 660                         traffic->ipsec.saptr, traffic->ipsec.num);
 661                 ipsec_process(ipsec_ctx, traffic);
 662         }
 663
 664         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 665                         n_ip4);
 666
 667         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 668                         n_ip6);
 669 }
 670
 671 static inline void
 672 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 673                 struct traffic_type *ipsec)
 674 {
 675         struct rte_mbuf *m;
 676         uint32_t i, j, sa_idx;
 677
 678         if (ip->num == 0 || sp == NULL)
 679                 return;
 680
 681         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 682                         ip->num, DEFAULT_MAX_CATEGORIES);
 683
 684         j = 0;
 685         for (i = 0; i < ip->num; i++) {
 686                 m = ip->pkts[i];
 687                 sa_idx = SPI2IDX(ip->res[i]);
 688                 if (ip->res[i] == DISCARD)
 689                         rte_pktmbuf_free(m);
 690                 else if (ip->res[i] == BYPASS)
 691                         ip->pkts[j++] = m;
 692                 else {
 693                         ipsec->res[ipsec->num] = sa_idx;
 694                         ipsec->pkts[ipsec->num++] = m;
 695                 }
 696         }
 697         ip->num = j;
 698 }
 699
 700 static inline void
 701 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 702                 struct ipsec_traffic *traffic)
 703 {
 704         struct rte_mbuf *m;
 705         uint16_t idx, nb_pkts_out, i;
 706
 707         /* Drop any IPsec traffic from protected ports */
 708         for (i = 0; i < traffic->ipsec.num; i++)
 709                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 710
 711         traffic->ipsec.num = 0;
 712
 713         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 714
 715         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 716
 717         if (app_sa_prm.enable == 0) {
 718
 719                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 720                                 traffic->ipsec.res, traffic->ipsec.num,
 721                                 MAX_PKT_BURST);
 722
 723                 for (i = 0; i < nb_pkts_out; i++) {
 724                         m = traffic->ipsec.pkts[i];
 725                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 726                         if (ip->ip_v == IPVERSION) {
 727                                 idx = traffic->ip4.num++;
 728                                 traffic->ip4.pkts[idx] = m;
 729                         } else {
 730                                 idx = traffic->ip6.num++;
 731                                 traffic->ip6.pkts[idx] = m;
 732                         }
 733                 }
 734         } else {
 735                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 736                         traffic->ipsec.saptr, traffic->ipsec.num);
 737                 ipsec_process(ipsec_ctx, traffic);
 738         }
 739 }
 740
 741 static inline void
 742 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 743                 struct ipsec_traffic *traffic)
 744 {
 745         struct rte_mbuf *m;
 746         uint32_t nb_pkts_in, i, idx;
 747
 748         /* Drop any IPv4 traffic from unprotected ports */
 749         for (i = 0; i < traffic->ip4.num; i++)
 750                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 751
 752         traffic->ip4.num = 0;
 753
 754         /* Drop any IPv6 traffic from unprotected ports */
 755         for (i = 0; i < traffic->ip6.num; i++)
 756                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 757
 758         traffic->ip6.num = 0;
 759
 760         if (app_sa_prm.enable == 0) {
 761
 762                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 763                                 traffic->ipsec.num, MAX_PKT_BURST);
 764
 765                 for (i = 0; i < nb_pkts_in; i++) {
 766                         m = traffic->ipsec.pkts[i];
 767                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 768                         if (ip->ip_v == IPVERSION) {
 769                                 idx = traffic->ip4.num++;
 770                                 traffic->ip4.pkts[idx] = m;
 771                         } else {
 772                                 idx = traffic->ip6.num++;
 773                                 traffic->ip6.pkts[idx] = m;
 774                         }
 775                 }
 776         } else {
 777                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 778                         traffic->ipsec.saptr, traffic->ipsec.num);
 779                 ipsec_process(ipsec_ctx, traffic);
 780         }
 781 }
 782
 783 static inline void
 784 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 785                 struct ipsec_traffic *traffic)
 786 {
 787         struct rte_mbuf *m;
 788         uint32_t nb_pkts_out, i, n;
 789         struct ip *ip;
 790
 791         /* Drop any IPsec traffic from protected ports */
 792         for (i = 0; i < traffic->ipsec.num; i++)
 793                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 794
 795         n = 0;
 796
 797         for (i = 0; i < traffic->ip4.num; i++) {
 798                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 799                 traffic->ipsec.res[n++] = single_sa_idx;
 800         }
 801
 802         for (i = 0; i < traffic->ip6.num; i++) {
 803                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 804                 traffic->ipsec.res[n++] = single_sa_idx;
 805         }
 806
 807         traffic->ip4.num = 0;
 808         traffic->ip6.num = 0;
 809         traffic->ipsec.num = n;
 810
 811         if (app_sa_prm.enable == 0) {
 812
 813                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 814                                 traffic->ipsec.res, traffic->ipsec.num,
 815                                 MAX_PKT_BURST);
 816
 817                 /* They all sue the same SA (ip4 or ip6 tunnel) */
 818                 m = traffic->ipsec.pkts[0];
 819                 ip = rte_pktmbuf_mtod(m, struct ip *);
 820                 if (ip->ip_v == IPVERSION) {
 821                         traffic->ip4.num = nb_pkts_out;
 822                         for (i = 0; i < nb_pkts_out; i++)
 823                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 824                 } else {
 825                         traffic->ip6.num = nb_pkts_out;
 826                         for (i = 0; i < nb_pkts_out; i++)
 827                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 828                 }
 829         } else {
 830                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 831                         traffic->ipsec.saptr, traffic->ipsec.num);
 832                 ipsec_process(ipsec_ctx, traffic);
 833         }
 834 }
 835
 836 static inline int32_t
 837 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 838 {
 839         struct ipsec_mbuf_metadata *priv;
 840         struct ipsec_sa *sa;
 841
 842         priv = get_priv(pkt);
 843
 844         sa = priv->sa;
 845         if (unlikely(sa == NULL)) {
 846                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 847                 goto fail;
 848         }
 849
 850         if (is_ipv6)
 851                 return sa->portid;
 852
 853         /* else */
 854         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 855
 856 fail:
 857         if (is_ipv6)
 858                 return -1;
 859
 860         /* else */
 861         return 0;
 862 }
 863
 864 static inline void
 865 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 866 {
 867         uint32_t hop[MAX_PKT_BURST * 2];
 868         uint32_t dst_ip[MAX_PKT_BURST * 2];
 869         int32_t pkt_hop = 0;
 870         uint16_t i, offset;
 871         uint16_t lpm_pkts = 0;
 872
 873         if (nb_pkts == 0)
 874                 return;
 875
 876         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 877          * have port ID in the SA
 878          */
 879
 880         for (i = 0; i < nb_pkts; i++) {
 881                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 882                         /* Security offload not enabled. So an LPM lookup is
 883                          * required to get the hop
 884                          */
 885                         offset = offsetof(struct ip, ip_dst);
 886                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 887                                         uint32_t *, offset);
 888                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 889                         lpm_pkts++;
 890                 }
 891         }
 892
 893         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 894
 895         lpm_pkts = 0;
 896
 897         for (i = 0; i < nb_pkts; i++) {
 898                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 899                         /* Read hop from the SA */
 900                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 901                 } else {
 902                         /* Need to use hop returned by lookup */
 903                         pkt_hop = hop[lpm_pkts++];
 904                 }
 905
 906                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 907                         rte_pktmbuf_free(pkts[i]);
 908                         continue;
 909                 }
 910                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
 911         }
 912 }
 913
 914 static inline void
 915 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 916 {
 917         int32_t hop[MAX_PKT_BURST * 2];
 918         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 919         uint8_t *ip6_dst;
 920         int32_t pkt_hop = 0;
 921         uint16_t i, offset;
 922         uint16_t lpm_pkts = 0;
 923
 924         if (nb_pkts == 0)
 925                 return;
 926
 927         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 928          * have port ID in the SA
 929          */
 930
 931         for (i = 0; i < nb_pkts; i++) {
 932                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 933                         /* Security offload not enabled. So an LPM lookup is
 934                          * required to get the hop
 935                          */
 936                         offset = offsetof(struct ip6_hdr, ip6_dst);
 937                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 938                                         offset);
 939                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 940                         lpm_pkts++;
 941                 }
 942         }
 943
 944         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 945                         lpm_pkts);
 946
 947         lpm_pkts = 0;
 948
 949         for (i = 0; i < nb_pkts; i++) {
 950                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 951                         /* Read hop from the SA */
 952                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 953                 } else {
 954                         /* Need to use hop returned by lookup */
 955                         pkt_hop = hop[lpm_pkts++];
 956                 }
 957
 958                 if (pkt_hop == -1) {
 959                         rte_pktmbuf_free(pkts[i]);
 960                         continue;
 961                 }
 962                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
 963         }
 964 }
 965
 966 static inline void
 967 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 968                 uint8_t nb_pkts, uint16_t portid)
 969 {
 970         struct ipsec_traffic traffic;
 971
 972         prepare_traffic(pkts, &traffic, nb_pkts);
 973
 974         if (unlikely(single_sa)) {
 975                 if (UNPROTECTED_PORT(portid))
 976                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 977                 else
 978                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 979         } else {
 980                 if (UNPROTECTED_PORT(portid))
 981                         process_pkts_inbound(&qconf->inbound, &traffic);
 982                 else
 983                         process_pkts_outbound(&qconf->outbound, &traffic);
 984         }
 985
 986         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 987         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 988 }
 989
 990 static inline void
 991 drain_tx_buffers(struct lcore_conf *qconf)
 992 {
 993         struct buffer *buf;
 994         uint32_t portid;
 995
 996         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 997                 buf = &qconf->tx_mbufs[portid];
 998                 if (buf->len == 0)
 999                         continue;
1000                 send_burst(qconf, buf->len, portid);
1001                 buf->len = 0;
1002         }
1003 }
1004
1005 static inline void
1006 drain_crypto_buffers(struct lcore_conf *qconf)
1007 {
1008         uint32_t i;
1009         struct ipsec_ctx *ctx;
1010
1011         /* drain inbound buffers*/
1012         ctx = &qconf->inbound;
1013         for (i = 0; i != ctx->nb_qps; i++) {
1014                 if (ctx->tbl[i].len != 0)
1015                         enqueue_cop_burst(ctx->tbl  + i);
1016         }
1017
1018         /* drain outbound buffers*/
1019         ctx = &qconf->outbound;
1020         for (i = 0; i != ctx->nb_qps; i++) {
1021                 if (ctx->tbl[i].len != 0)
1022                         enqueue_cop_burst(ctx->tbl  + i);
1023         }
1024 }
1025
1026 static void
1027 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1028                 struct ipsec_ctx *ctx)
1029 {
1030         uint32_t n;
1031         struct ipsec_traffic trf;
1032
1033         if (app_sa_prm.enable == 0) {
1034
1035                 /* dequeue packets from crypto-queue */
1036                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1037                         RTE_DIM(trf.ipsec.pkts));
1038
1039                 trf.ip4.num = 0;
1040                 trf.ip6.num = 0;
1041
1042                 /* split traffic by ipv4-ipv6 */
1043                 split46_traffic(&trf, trf.ipsec.pkts, n);
1044         } else
1045                 ipsec_cqp_process(ctx, &trf);
1046
1047         /* process ipv4 packets */
1048         if (trf.ip4.num != 0) {
1049                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1050                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1051         }
1052
1053         /* process ipv6 packets */
1054         if (trf.ip6.num != 0) {
1055                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1056                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1057         }
1058 }
1059
1060 static void
1061 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1062                 struct ipsec_ctx *ctx)
1063 {
1064         uint32_t n;
1065         struct ipsec_traffic trf;
1066
1067         if (app_sa_prm.enable == 0) {
1068
1069                 /* dequeue packets from crypto-queue */
1070                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1071                         RTE_DIM(trf.ipsec.pkts));
1072
1073                 trf.ip4.num = 0;
1074                 trf.ip6.num = 0;
1075
1076                 /* split traffic by ipv4-ipv6 */
1077                 split46_traffic(&trf, trf.ipsec.pkts, n);
1078         } else
1079                 ipsec_cqp_process(ctx, &trf);
1080
1081         /* process ipv4 packets */
1082         if (trf.ip4.num != 0)
1083                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1084
1085         /* process ipv6 packets */
1086         if (trf.ip6.num != 0)
1087                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1088 }
1089
1090 /* main processing loop */
1091 static int32_t
1092 main_loop(__attribute__((unused)) void *dummy)
1093 {
1094         struct rte_mbuf *pkts[MAX_PKT_BURST];
1095         uint32_t lcore_id;
1096         uint64_t prev_tsc, diff_tsc, cur_tsc;
1097         int32_t i, nb_rx;
1098         uint16_t portid;
1099         uint8_t queueid;
1100         struct lcore_conf *qconf;
1101         int32_t socket_id;
1102         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1103                         / US_PER_S * BURST_TX_DRAIN_US;
1104         struct lcore_rx_queue *rxql;
1105
1106         prev_tsc = 0;
1107         lcore_id = rte_lcore_id();
1108         qconf = &lcore_conf[lcore_id];
1109         rxql = qconf->rx_queue_list;
1110         socket_id = rte_lcore_to_socket_id(lcore_id);
1111
1112         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1113         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1114         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1115         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1116         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1117         qconf->inbound.cdev_map = cdev_map_in;
1118         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1119         qconf->inbound.session_priv_pool =
1120                         socket_ctx[socket_id].session_priv_pool;
1121         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1122         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1123         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1124         qconf->outbound.cdev_map = cdev_map_out;
1125         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1126         qconf->outbound.session_priv_pool =
1127                         socket_ctx[socket_id].session_priv_pool;
1128         qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1129         qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1130
1131         if (qconf->nb_rx_queue == 0) {
1132                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1133                         lcore_id);
1134                 return 0;
1135         }
1136
1137         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1138
1139         for (i = 0; i < qconf->nb_rx_queue; i++) {
1140                 portid = rxql[i].port_id;
1141                 queueid = rxql[i].queue_id;
1142                 RTE_LOG(INFO, IPSEC,
1143                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1144                         lcore_id, portid, queueid);
1145         }
1146
1147         while (1) {
1148                 cur_tsc = rte_rdtsc();
1149
1150                 /* TX queue buffer drain */
1151                 diff_tsc = cur_tsc - prev_tsc;
1152
1153                 if (unlikely(diff_tsc > drain_tsc)) {
1154                         drain_tx_buffers(qconf);
1155                         drain_crypto_buffers(qconf);
1156                         prev_tsc = cur_tsc;
1157                 }
1158
1159                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1160
1161                         /* Read packets from RX queues */
1162                         portid = rxql[i].port_id;
1163                         queueid = rxql[i].queue_id;
1164                         nb_rx = rte_eth_rx_burst(portid, queueid,
1165                                         pkts, MAX_PKT_BURST);
1166
1167                         if (nb_rx > 0)
1168                                 process_pkts(qconf, pkts, nb_rx, portid);
1169
1170                         /* dequeue and process completed crypto-ops */
1171                         if (UNPROTECTED_PORT(portid))
1172                                 drain_inbound_crypto_queues(qconf,
1173                                         &qconf->inbound);
1174                         else
1175                                 drain_outbound_crypto_queues(qconf,
1176                                         &qconf->outbound);
1177                 }
1178         }
1179 }
1180
1181 static int32_t
1182 check_params(void)
1183 {
1184         uint8_t lcore;
1185         uint16_t portid;
1186         uint16_t i;
1187         int32_t socket_id;
1188
1189         if (lcore_params == NULL) {
1190                 printf("Error: No port/queue/core mappings\n");
1191                 return -1;
1192         }
1193
1194         for (i = 0; i < nb_lcore_params; ++i) {
1195                 lcore = lcore_params[i].lcore_id;
1196                 if (!rte_lcore_is_enabled(lcore)) {
1197                         printf("error: lcore %hhu is not enabled in "
1198                                 "lcore mask\n", lcore);
1199                         return -1;
1200                 }
1201                 socket_id = rte_lcore_to_socket_id(lcore);
1202                 if (socket_id != 0 && numa_on == 0) {
1203                         printf("warning: lcore %hhu is on socket %d "
1204                                 "with numa off\n",
1205                                 lcore, socket_id);
1206                 }
1207                 portid = lcore_params[i].port_id;
1208                 if ((enabled_port_mask & (1 << portid)) == 0) {
1209                         printf("port %u is not enabled in port mask\n", portid);
1210                         return -1;
1211                 }
1212                 if (!rte_eth_dev_is_valid_port(portid)) {
1213                         printf("port %u is not present on the board\n", portid);
1214                         return -1;
1215                 }
1216         }
1217         return 0;
1218 }
1219
1220 static uint8_t
1221 get_port_nb_rx_queues(const uint16_t port)
1222 {
1223         int32_t queue = -1;
1224         uint16_t i;
1225
1226         for (i = 0; i < nb_lcore_params; ++i) {
1227                 if (lcore_params[i].port_id == port &&
1228                                 lcore_params[i].queue_id > queue)
1229                         queue = lcore_params[i].queue_id;
1230         }
1231         return (uint8_t)(++queue);
1232 }
1233
1234 static int32_t
1235 init_lcore_rx_queues(void)
1236 {
1237         uint16_t i, nb_rx_queue;
1238         uint8_t lcore;
1239
1240         for (i = 0; i < nb_lcore_params; ++i) {
1241                 lcore = lcore_params[i].lcore_id;
1242                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1243                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1244                         printf("error: too many queues (%u) for lcore: %u\n",
1245                                         nb_rx_queue + 1, lcore);
1246                         return -1;
1247                 }
1248                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1249                         lcore_params[i].port_id;
1250                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1251                         lcore_params[i].queue_id;
1252                 lcore_conf[lcore].nb_rx_queue++;
1253         }
1254         return 0;
1255 }
1256
1257 /* display usage */
1258 static void
1259 print_usage(const char *prgname)
1260 {
1261         fprintf(stderr, "%s [EAL options] --"
1262                 " -p PORTMASK"
1263                 " [-P]"
1264                 " [-u PORTMASK]"
1265                 " [-j FRAMESIZE]"
1266                 " [-l]"
1267                 " [-w REPLAY_WINDOW_SIZE]"
1268                 " [-e]"
1269                 " [-a]"
1270                 " -f CONFIG_FILE"
1271                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1272                 " [--single-sa SAIDX]"
1273                 " [--cryptodev_mask MASK]"
1274                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1275                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1276                 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1277                 " [--" CMD_LINE_OPT_MTU " MTU]"
1278                 "\n\n"
1279                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1280                 "  -P : Enable promiscuous mode\n"
1281                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1282                 "  -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1283                 "     value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1284                 "  -l enables code-path that uses librte_ipsec\n"
1285                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1286                 "     size for each SA\n"
1287                 "  -e enables ESN\n"
1288                 "  -a enables SA SQN atomic behaviour\n"
1289                 "  -f CONFIG_FILE: Configuration file\n"
1290                 "  --config (port,queue,lcore): Rx queue configuration\n"
1291                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1292                 "                     bypassing the SP\n"
1293                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1294                 "                         devices to configure\n"
1295                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1296                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1297                 "                         (DEV_RX_OFFLOAD_*)\n"
1298                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1299                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1300                 "                         (DEV_TX_OFFLOAD_*)\n"
1301                 "  --" CMD_LINE_OPT_REASSEMBLE " NUM"
1302                 ": max number of entries in reassemble(fragment) table\n"
1303                 "    (zero (default value) disables reassembly)\n"
1304                 "  --" CMD_LINE_OPT_MTU " MTU"
1305                 ": MTU value on all ports (default value: 1500)\n"
1306                 "    outgoing packets with bigger size will be fragmented\n"
1307                 "    incoming packets with bigger size will be discarded\n"
1308                 "\n",
1309                 prgname);
1310 }
1311
1312 static int
1313 parse_mask(const char *str, uint64_t *val)
1314 {
1315         char *end;
1316         unsigned long t;
1317
1318         errno = 0;
1319         t = strtoul(str, &end, 0);
1320         if (errno != 0 || end[0] != 0)
1321                 return -EINVAL;
1322
1323         *val = t;
1324         return 0;
1325 }
1326
1327 static int32_t
1328 parse_portmask(const char *portmask)
1329 {
1330         char *end = NULL;
1331         unsigned long pm;
1332
1333         /* parse hexadecimal string */
1334         pm = strtoul(portmask, &end, 16);
1335         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1336                 return -1;
1337
1338         if ((pm == 0) && errno)
1339                 return -1;
1340
1341         return pm;
1342 }
1343
1344 static int32_t
1345 parse_decimal(const char *str)
1346 {
1347         char *end = NULL;
1348         unsigned long num;
1349
1350         num = strtoul(str, &end, 10);
1351         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
1352                 return -1;
1353
1354         return num;
1355 }
1356
1357 static int32_t
1358 parse_config(const char *q_arg)
1359 {
1360         char s[256];
1361         const char *p, *p0 = q_arg;
1362         char *end;
1363         enum fieldnames {
1364                 FLD_PORT = 0,
1365                 FLD_QUEUE,
1366                 FLD_LCORE,
1367                 _NUM_FLD
1368         };
1369         unsigned long int_fld[_NUM_FLD];
1370         char *str_fld[_NUM_FLD];
1371         int32_t i;
1372         uint32_t size;
1373
1374         nb_lcore_params = 0;
1375
1376         while ((p = strchr(p0, '(')) != NULL) {
1377                 ++p;
1378                 p0 = strchr(p, ')');
1379                 if (p0 == NULL)
1380                         return -1;
1381
1382                 size = p0 - p;
1383                 if (size >= sizeof(s))
1384                         return -1;
1385
1386                 snprintf(s, sizeof(s), "%.*s", size, p);
1387                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1388                                 _NUM_FLD)
1389                         return -1;
1390                 for (i = 0; i < _NUM_FLD; i++) {
1391                         errno = 0;
1392                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1393                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1394                                 return -1;
1395                 }
1396                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1397                         printf("exceeded max number of lcore params: %hu\n",
1398                                 nb_lcore_params);
1399                         return -1;
1400                 }
1401                 lcore_params_array[nb_lcore_params].port_id =
1402                         (uint8_t)int_fld[FLD_PORT];
1403                 lcore_params_array[nb_lcore_params].queue_id =
1404                         (uint8_t)int_fld[FLD_QUEUE];
1405                 lcore_params_array[nb_lcore_params].lcore_id =
1406                         (uint8_t)int_fld[FLD_LCORE];
1407                 ++nb_lcore_params;
1408         }
1409         lcore_params = lcore_params_array;
1410         return 0;
1411 }
1412
1413 static void
1414 print_app_sa_prm(const struct app_sa_prm *prm)
1415 {
1416         printf("librte_ipsec usage: %s\n",
1417                 (prm->enable == 0) ? "disabled" : "enabled");
1418
1419         if (prm->enable == 0)
1420                 return;
1421
1422         printf("replay window size: %u\n", prm->window_size);
1423         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1424         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1425 }
1426
1427 static int32_t
1428 parse_args(int32_t argc, char **argv)
1429 {
1430         int32_t opt, ret;
1431         char **argvopt;
1432         int32_t option_index;
1433         char *prgname = argv[0];
1434         int32_t f_present = 0;
1435
1436         argvopt = argv;
1437
1438         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:",
1439                                 lgopts, &option_index)) != EOF) {
1440
1441                 switch (opt) {
1442                 case 'p':
1443                         enabled_port_mask = parse_portmask(optarg);
1444                         if (enabled_port_mask == 0) {
1445                                 printf("invalid portmask\n");
1446                                 print_usage(prgname);
1447                                 return -1;
1448                         }
1449                         break;
1450                 case 'P':
1451                         printf("Promiscuous mode selected\n");
1452                         promiscuous_on = 1;
1453                         break;
1454                 case 'u':
1455                         unprotected_port_mask = parse_portmask(optarg);
1456                         if (unprotected_port_mask == 0) {
1457                                 printf("invalid unprotected portmask\n");
1458                                 print_usage(prgname);
1459                                 return -1;
1460                         }
1461                         break;
1462                 case 'f':
1463                         if (f_present == 1) {
1464                                 printf("\"-f\" option present more than "
1465                                         "once!\n");
1466                                 print_usage(prgname);
1467                                 return -1;
1468                         }
1469                         cfgfile = optarg;
1470                         f_present = 1;
1471                         break;
1472                 case 'j':
1473                         ret = parse_decimal(optarg);
1474                         if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1475                                         ret > UINT16_MAX) {
1476                                 printf("Invalid frame buffer size value: %s\n",
1477                                         optarg);
1478                                 print_usage(prgname);
1479                                 return -1;
1480                         }
1481                         frame_buf_size = ret;
1482                         printf("Custom frame buffer size %u\n", frame_buf_size);
1483                         break;
1484                 case 'l':
1485                         app_sa_prm.enable = 1;
1486                         break;
1487                 case 'w':
1488                         app_sa_prm.enable = 1;
1489                         app_sa_prm.window_size = parse_decimal(optarg);
1490                         break;
1491                 case 'e':
1492                         app_sa_prm.enable = 1;
1493                         app_sa_prm.enable_esn = 1;
1494                         break;
1495                 case 'a':
1496                         app_sa_prm.enable = 1;
1497                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1498                         break;
1499                 case CMD_LINE_OPT_CONFIG_NUM:
1500                         ret = parse_config(optarg);
1501                         if (ret) {
1502                                 printf("Invalid config\n");
1503                                 print_usage(prgname);
1504                                 return -1;
1505                         }
1506                         break;
1507                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1508                         ret = parse_decimal(optarg);
1509                         if (ret == -1) {
1510                                 printf("Invalid argument[sa_idx]\n");
1511                                 print_usage(prgname);
1512                                 return -1;
1513                         }
1514
1515                         /* else */
1516                         single_sa = 1;
1517                         single_sa_idx = ret;
1518                         printf("Configured with single SA index %u\n",
1519                                         single_sa_idx);
1520                         break;
1521                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1522                         ret = parse_portmask(optarg);
1523                         if (ret == -1) {
1524                                 printf("Invalid argument[portmask]\n");
1525                                 print_usage(prgname);
1526                                 return -1;
1527                         }
1528
1529                         /* else */
1530                         enabled_cryptodev_mask = ret;
1531                         break;
1532                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1533                         ret = parse_mask(optarg, &dev_rx_offload);
1534                         if (ret != 0) {
1535                                 printf("Invalid argument for \'%s\': %s\n",
1536                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1537                                 print_usage(prgname);
1538                                 return -1;
1539                         }
1540                         break;
1541                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1542                         ret = parse_mask(optarg, &dev_tx_offload);
1543                         if (ret != 0) {
1544                                 printf("Invalid argument for \'%s\': %s\n",
1545                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1546                                 print_usage(prgname);
1547                                 return -1;
1548                         }
1549                         break;
1550                 case CMD_LINE_OPT_REASSEMBLE_NUM:
1551                         ret = parse_decimal(optarg);
1552                         if (ret < 0) {
1553                                 printf("Invalid argument for \'%s\': %s\n",
1554                                         CMD_LINE_OPT_REASSEMBLE, optarg);
1555                                 print_usage(prgname);
1556                                 return -1;
1557                         }
1558                         frag_tbl_sz = ret;
1559                         break;
1560                 case CMD_LINE_OPT_MTU_NUM:
1561                         ret = parse_decimal(optarg);
1562                         if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1563                                 printf("Invalid argument for \'%s\': %s\n",
1564                                         CMD_LINE_OPT_MTU, optarg);
1565                                 print_usage(prgname);
1566                                 return -1;
1567                         }
1568                         mtu_size = ret;
1569                         break;
1570                 default:
1571                         print_usage(prgname);
1572                         return -1;
1573                 }
1574         }
1575
1576         if (f_present == 0) {
1577                 printf("Mandatory option \"-f\" not present\n");
1578                 return -1;
1579         }
1580
1581         /* check do we need to enable multi-seg support */
1582         if (multi_seg_required()) {
1583                 /* legacy mode doesn't support multi-seg */
1584                 app_sa_prm.enable = 1;
1585                 printf("frame buf size: %u, mtu: %u, "
1586                         "number of reassemble entries: %u\n"
1587                         "multi-segment support is required\n",
1588                         frame_buf_size, mtu_size, frag_tbl_sz);
1589         }
1590
1591         print_app_sa_prm(&app_sa_prm);
1592
1593         if (optind >= 0)
1594                 argv[optind-1] = prgname;
1595
1596         ret = optind-1;
1597         optind = 1; /* reset getopt lib */
1598         return ret;
1599 }
1600
1601 static void
1602 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1603 {
1604         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1605         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1606         printf("%s%s", name, buf);
1607 }
1608
1609 /*
1610  * Update destination ethaddr for the port.
1611  */
1612 int
1613 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1614 {
1615         if (port >= RTE_DIM(ethaddr_tbl))
1616                 return -EINVAL;
1617
1618         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1619         return 0;
1620 }
1621
1622 /* Check the link status of all ports in up to 9s, and print them finally */
1623 static void
1624 check_all_ports_link_status(uint32_t port_mask)
1625 {
1626 #define CHECK_INTERVAL 100 /* 100ms */
1627 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1628         uint16_t portid;
1629         uint8_t count, all_ports_up, print_flag = 0;
1630         struct rte_eth_link link;
1631         int ret;
1632
1633         printf("\nChecking link status");
1634         fflush(stdout);
1635         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1636                 all_ports_up = 1;
1637                 RTE_ETH_FOREACH_DEV(portid) {
1638                         if ((port_mask & (1 << portid)) == 0)
1639                                 continue;
1640                         memset(&link, 0, sizeof(link));
1641                         ret = rte_eth_link_get_nowait(portid, &link);
1642                         if (ret < 0) {
1643                                 all_ports_up = 0;
1644                                 if (print_flag == 1)
1645                                         printf("Port %u link get failed: %s\n",
1646                                                 portid, rte_strerror(-ret));
1647                                 continue;
1648                         }
1649                         /* print link status if flag set */
1650                         if (print_flag == 1) {
1651                                 if (link.link_status)
1652                                         printf(
1653                                         "Port%d Link Up - speed %u Mbps -%s\n",
1654                                                 portid, link.link_speed,
1655                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1656                                         ("full-duplex") : ("half-duplex\n"));
1657                                 else
1658                                         printf("Port %d Link Down\n", portid);
1659                                 continue;
1660                         }
1661                         /* clear all_ports_up flag if any link down */
1662                         if (link.link_status == ETH_LINK_DOWN) {
1663                                 all_ports_up = 0;
1664                                 break;
1665                         }
1666                 }
1667                 /* after finally printing all link status, get out */
1668                 if (print_flag == 1)
1669                         break;
1670
1671                 if (all_ports_up == 0) {
1672                         printf(".");
1673                         fflush(stdout);
1674                         rte_delay_ms(CHECK_INTERVAL);
1675                 }
1676
1677                 /* set the print_flag if all ports up or timeout */
1678                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1679                         print_flag = 1;
1680                         printf("done\n");
1681                 }
1682         }
1683 }
1684
1685 static int32_t
1686 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1687                 uint16_t qp, struct lcore_params *params,
1688                 struct ipsec_ctx *ipsec_ctx,
1689                 const struct rte_cryptodev_capabilities *cipher,
1690                 const struct rte_cryptodev_capabilities *auth,
1691                 const struct rte_cryptodev_capabilities *aead)
1692 {
1693         int32_t ret = 0;
1694         unsigned long i;
1695         struct cdev_key key = { 0 };
1696
1697         key.lcore_id = params->lcore_id;
1698         if (cipher)
1699                 key.cipher_algo = cipher->sym.cipher.algo;
1700         if (auth)
1701                 key.auth_algo = auth->sym.auth.algo;
1702         if (aead)
1703                 key.aead_algo = aead->sym.aead.algo;
1704
1705         ret = rte_hash_lookup(map, &key);
1706         if (ret != -ENOENT)
1707                 return 0;
1708
1709         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1710                 if (ipsec_ctx->tbl[i].id == cdev_id)
1711                         break;
1712
1713         if (i == ipsec_ctx->nb_qps) {
1714                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1715                         printf("Maximum number of crypto devices assigned to "
1716                                 "a core, increase MAX_QP_PER_LCORE value\n");
1717                         return 0;
1718                 }
1719                 ipsec_ctx->tbl[i].id = cdev_id;
1720                 ipsec_ctx->tbl[i].qp = qp;
1721                 ipsec_ctx->nb_qps++;
1722                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1723                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1724                                 cdev_id, qp, i);
1725         }
1726
1727         ret = rte_hash_add_key_data(map, &key, (void *)i);
1728         if (ret < 0) {
1729                 printf("Faled to insert cdev mapping for (lcore %u, "
1730                                 "cdev %u, qp %u), errno %d\n",
1731                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1732                                 ipsec_ctx->tbl[i].qp, ret);
1733                 return 0;
1734         }
1735
1736         return 1;
1737 }
1738
1739 static int32_t
1740 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1741                 uint16_t qp, struct lcore_params *params)
1742 {
1743         int32_t ret = 0;
1744         const struct rte_cryptodev_capabilities *i, *j;
1745         struct rte_hash *map;
1746         struct lcore_conf *qconf;
1747         struct ipsec_ctx *ipsec_ctx;
1748         const char *str;
1749
1750         qconf = &lcore_conf[params->lcore_id];
1751
1752         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1753                 map = cdev_map_out;
1754                 ipsec_ctx = &qconf->outbound;
1755                 str = "Outbound";
1756         } else {
1757                 map = cdev_map_in;
1758                 ipsec_ctx = &qconf->inbound;
1759                 str = "Inbound";
1760         }
1761
1762         /* Required cryptodevs with operation chainning */
1763         if (!(dev_info->feature_flags &
1764                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1765                 return ret;
1766
1767         for (i = dev_info->capabilities;
1768                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1769                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1770                         continue;
1771
1772                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1773                         ret |= add_mapping(map, str, cdev_id, qp, params,
1774                                         ipsec_ctx, NULL, NULL, i);
1775                         continue;
1776                 }
1777
1778                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1779                         continue;
1780
1781                 for (j = dev_info->capabilities;
1782                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1783                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1784                                 continue;
1785
1786                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1787                                 continue;
1788
1789                         ret |= add_mapping(map, str, cdev_id, qp, params,
1790                                                 ipsec_ctx, i, j, NULL);
1791                 }
1792         }
1793
1794         return ret;
1795 }
1796
1797 /* Check if the device is enabled by cryptodev_mask */
1798 static int
1799 check_cryptodev_mask(uint8_t cdev_id)
1800 {
1801         if (enabled_cryptodev_mask & (1 << cdev_id))
1802                 return 0;
1803
1804         return -1;
1805 }
1806
1807 static int32_t
1808 cryptodevs_init(void)
1809 {
1810         struct rte_cryptodev_config dev_conf;
1811         struct rte_cryptodev_qp_conf qp_conf;
1812         uint16_t idx, max_nb_qps, qp, i;
1813         int16_t cdev_id;
1814         struct rte_hash_parameters params = { 0 };
1815
1816         const uint64_t mseg_flag = multi_seg_required() ?
1817                                 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
1818
1819         params.entries = CDEV_MAP_ENTRIES;
1820         params.key_len = sizeof(struct cdev_key);
1821         params.hash_func = rte_jhash;
1822         params.hash_func_init_val = 0;
1823         params.socket_id = rte_socket_id();
1824
1825         params.name = "cdev_map_in";
1826         cdev_map_in = rte_hash_create(&params);
1827         if (cdev_map_in == NULL)
1828                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1829                                 rte_errno);
1830
1831         params.name = "cdev_map_out";
1832         cdev_map_out = rte_hash_create(&params);
1833         if (cdev_map_out == NULL)
1834                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1835                                 rte_errno);
1836
1837         printf("lcore/cryptodev/qp mappings:\n");
1838
1839         idx = 0;
1840         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1841                 struct rte_cryptodev_info cdev_info;
1842
1843                 if (check_cryptodev_mask((uint8_t)cdev_id))
1844                         continue;
1845
1846                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1847
1848                 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
1849                         rte_exit(EXIT_FAILURE,
1850                                 "Device %hd does not support \'%s\' feature\n",
1851                                 cdev_id,
1852                                 rte_cryptodev_get_feature_name(mseg_flag));
1853
1854                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1855                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1856                 else
1857                         max_nb_qps = nb_lcore_params;
1858
1859                 qp = 0;
1860                 i = 0;
1861                 while (qp < max_nb_qps && i < nb_lcore_params) {
1862                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1863                                                 &lcore_params[idx]))
1864                                 qp++;
1865                         idx++;
1866                         idx = idx % nb_lcore_params;
1867                         i++;
1868                 }
1869
1870                 if (qp == 0)
1871                         continue;
1872
1873                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1874                 dev_conf.nb_queue_pairs = qp;
1875                 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
1876
1877                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1878                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1879                         rte_exit(EXIT_FAILURE,
1880                                 "Device does not support at least %u "
1881                                 "sessions", CDEV_MP_NB_OBJS);
1882
1883                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1884                         rte_panic("Failed to initialize cryptodev %u\n",
1885                                         cdev_id);
1886
1887                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1888                 qp_conf.mp_session =
1889                         socket_ctx[dev_conf.socket_id].session_pool;
1890                 qp_conf.mp_session_private =
1891                         socket_ctx[dev_conf.socket_id].session_priv_pool;
1892                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1893                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1894                                         &qp_conf, dev_conf.socket_id))
1895                                 rte_panic("Failed to setup queue %u for "
1896                                                 "cdev_id %u\n", 0, cdev_id);
1897
1898                 if (rte_cryptodev_start(cdev_id))
1899                         rte_panic("Failed to start cryptodev %u\n",
1900                                         cdev_id);
1901         }
1902
1903         printf("\n");
1904
1905         return 0;
1906 }
1907
1908 static void
1909 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
1910 {
1911         uint32_t frame_size;
1912         struct rte_eth_dev_info dev_info;
1913         struct rte_eth_txconf *txconf;
1914         uint16_t nb_tx_queue, nb_rx_queue;
1915         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1916         int32_t ret, socket_id;
1917         struct lcore_conf *qconf;
1918         struct rte_ether_addr ethaddr;
1919         struct rte_eth_conf local_port_conf = port_conf;
1920
1921         ret = rte_eth_dev_info_get(portid, &dev_info);
1922         if (ret != 0)
1923                 rte_exit(EXIT_FAILURE,
1924                         "Error during getting device (port %u) info: %s\n",
1925                         portid, strerror(-ret));
1926
1927         /* limit allowed HW offloafs, as user requested */
1928         dev_info.rx_offload_capa &= dev_rx_offload;
1929         dev_info.tx_offload_capa &= dev_tx_offload;
1930
1931         printf("Configuring device port %u:\n", portid);
1932
1933         ret = rte_eth_macaddr_get(portid, &ethaddr);
1934         if (ret != 0)
1935                 rte_exit(EXIT_FAILURE,
1936                         "Error getting MAC address (port %u): %s\n",
1937                         portid, rte_strerror(-ret));
1938
1939         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
1940         print_ethaddr("Address: ", &ethaddr);
1941         printf("\n");
1942
1943         nb_rx_queue = get_port_nb_rx_queues(portid);
1944         nb_tx_queue = nb_lcores;
1945
1946         if (nb_rx_queue > dev_info.max_rx_queues)
1947                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1948                                 "(max rx queue is %u)\n",
1949                                 nb_rx_queue, dev_info.max_rx_queues);
1950
1951         if (nb_tx_queue > dev_info.max_tx_queues)
1952                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1953                                 "(max tx queue is %u)\n",
1954                                 nb_tx_queue, dev_info.max_tx_queues);
1955
1956         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1957                         nb_rx_queue, nb_tx_queue);
1958
1959         frame_size = MTU_TO_FRAMELEN(mtu_size);
1960         if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
1961                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1962         local_port_conf.rxmode.max_rx_pkt_len = frame_size;
1963
1964         if (multi_seg_required()) {
1965                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
1966                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
1967         }
1968
1969         local_port_conf.rxmode.offloads |= req_rx_offloads;
1970         local_port_conf.txmode.offloads |= req_tx_offloads;
1971
1972         /* Check that all required capabilities are supported */
1973         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
1974                         local_port_conf.rxmode.offloads)
1975                 rte_exit(EXIT_FAILURE,
1976                         "Error: port %u required RX offloads: 0x%" PRIx64
1977                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
1978                         portid, local_port_conf.rxmode.offloads,
1979                         dev_info.rx_offload_capa);
1980
1981         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
1982                         local_port_conf.txmode.offloads)
1983                 rte_exit(EXIT_FAILURE,
1984                         "Error: port %u required TX offloads: 0x%" PRIx64
1985                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
1986                         portid, local_port_conf.txmode.offloads,
1987                         dev_info.tx_offload_capa);
1988
1989         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1990                 local_port_conf.txmode.offloads |=
1991                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1992
1993         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
1994                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
1995
1996         printf("port %u configurng rx_offloads=0x%" PRIx64
1997                 ", tx_offloads=0x%" PRIx64 "\n",
1998                 portid, local_port_conf.rxmode.offloads,
1999                 local_port_conf.txmode.offloads);
2000
2001         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2002                 dev_info.flow_type_rss_offloads;
2003         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2004                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
2005                 printf("Port %u modified RSS hash function based on hardware support,"
2006                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2007                         portid,
2008                         port_conf.rx_adv_conf.rss_conf.rss_hf,
2009                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2010         }
2011
2012         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2013                         &local_port_conf);
2014         if (ret < 0)
2015                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2016                                 "err=%d, port=%d\n", ret, portid);
2017
2018         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2019         if (ret < 0)
2020                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2021                                 "err=%d, port=%d\n", ret, portid);
2022
2023         /* init one TX queue per lcore */
2024         tx_queueid = 0;
2025         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2026                 if (rte_lcore_is_enabled(lcore_id) == 0)
2027                         continue;
2028
2029                 if (numa_on)
2030                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2031                 else
2032                         socket_id = 0;
2033
2034                 /* init TX queue */
2035                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2036
2037                 txconf = &dev_info.default_txconf;
2038                 txconf->offloads = local_port_conf.txmode.offloads;
2039
2040                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2041                                 socket_id, txconf);
2042                 if (ret < 0)
2043                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2044                                         "err=%d, port=%d\n", ret, portid);
2045
2046                 qconf = &lcore_conf[lcore_id];
2047                 qconf->tx_queue_id[portid] = tx_queueid;
2048
2049                 /* Pre-populate pkt offloads based on capabilities */
2050                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2051                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2052                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2053                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2054
2055                 tx_queueid++;
2056
2057                 /* init RX queues */
2058                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2059                         struct rte_eth_rxconf rxq_conf;
2060
2061                         if (portid != qconf->rx_queue_list[queue].port_id)
2062                                 continue;
2063
2064                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
2065
2066                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2067                                         socket_id);
2068
2069                         rxq_conf = dev_info.default_rxconf;
2070                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
2071                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2072                                         nb_rxd, socket_id, &rxq_conf,
2073                                         socket_ctx[socket_id].mbuf_pool);
2074                         if (ret < 0)
2075                                 rte_exit(EXIT_FAILURE,
2076                                         "rte_eth_rx_queue_setup: err=%d, "
2077                                         "port=%d\n", ret, portid);
2078                 }
2079         }
2080         printf("\n");
2081 }
2082
2083 static size_t
2084 max_session_size(void)
2085 {
2086         size_t max_sz, sz;
2087         void *sec_ctx;
2088         int16_t cdev_id, port_id, n;
2089
2090         max_sz = 0;
2091         n =  rte_cryptodev_count();
2092         for (cdev_id = 0; cdev_id != n; cdev_id++) {
2093                 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2094                 if (sz > max_sz)
2095                         max_sz = sz;
2096                 /*
2097                  * If crypto device is security capable, need to check the
2098                  * size of security session as well.
2099                  */
2100
2101                 /* Get security context of the crypto device */
2102                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2103                 if (sec_ctx == NULL)
2104                         continue;
2105
2106                 /* Get size of security session */
2107                 sz = rte_security_session_get_size(sec_ctx);
2108                 if (sz > max_sz)
2109                         max_sz = sz;
2110         }
2111
2112         RTE_ETH_FOREACH_DEV(port_id) {
2113                 if ((enabled_port_mask & (1 << port_id)) == 0)
2114                         continue;
2115
2116                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2117                 if (sec_ctx == NULL)
2118                         continue;
2119
2120                 sz = rte_security_session_get_size(sec_ctx);
2121                 if (sz > max_sz)
2122                         max_sz = sz;
2123         }
2124
2125         return max_sz;
2126 }
2127
2128 static void
2129 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2130 {
2131         char mp_name[RTE_MEMPOOL_NAMESIZE];
2132         struct rte_mempool *sess_mp;
2133
2134         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2135                         "sess_mp_%u", socket_id);
2136         sess_mp = rte_cryptodev_sym_session_pool_create(
2137                         mp_name, CDEV_MP_NB_OBJS,
2138                         sess_sz, CDEV_MP_CACHE_SZ, 0,
2139                         socket_id);
2140         ctx->session_pool = sess_mp;
2141
2142         if (ctx->session_pool == NULL)
2143                 rte_exit(EXIT_FAILURE,
2144                         "Cannot init session pool on socket %d\n", socket_id);
2145         else
2146                 printf("Allocated session pool on socket %d\n", socket_id);
2147 }
2148
2149 static void
2150 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2151         size_t sess_sz)
2152 {
2153         char mp_name[RTE_MEMPOOL_NAMESIZE];
2154         struct rte_mempool *sess_mp;
2155
2156         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2157                         "sess_mp_priv_%u", socket_id);
2158         sess_mp = rte_mempool_create(mp_name,
2159                         CDEV_MP_NB_OBJS,
2160                         sess_sz,
2161                         CDEV_MP_CACHE_SZ,
2162                         0, NULL, NULL, NULL,
2163                         NULL, socket_id,
2164                         0);
2165         ctx->session_priv_pool = sess_mp;
2166
2167         if (ctx->session_priv_pool == NULL)
2168                 rte_exit(EXIT_FAILURE,
2169                         "Cannot init session priv pool on socket %d\n",
2170                         socket_id);
2171         else
2172                 printf("Allocated session priv pool on socket %d\n",
2173                         socket_id);
2174 }
2175
2176 static void
2177 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2178 {
2179         char s[64];
2180         int32_t ms;
2181
2182         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2183         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2184                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2185                         frame_buf_size, socket_id);
2186
2187         /*
2188          * if multi-segment support is enabled, then create a pool
2189          * for indirect mbufs.
2190          */
2191         ms = multi_seg_required();
2192         if (ms != 0) {
2193                 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2194                 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2195                         MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2196         }
2197
2198         if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2199                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2200                                 socket_id);
2201         else
2202                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2203 }
2204
2205 static inline int
2206 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2207 {
2208         struct ipsec_sa *sa;
2209
2210         /* For inline protocol processing, the metadata in the event will
2211          * uniquely identify the security session which raised the event.
2212          * Application would then need the userdata it had registered with the
2213          * security session to process the event.
2214          */
2215
2216         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2217
2218         if (sa == NULL) {
2219                 /* userdata could not be retrieved */
2220                 return -1;
2221         }
2222
2223         /* Sequence number over flow. SA need to be re-established */
2224         RTE_SET_USED(sa);
2225         return 0;
2226 }
2227
2228 static int
2229 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2230                  void *param, void *ret_param)
2231 {
2232         uint64_t md;
2233         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2234         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2235                                         rte_eth_dev_get_sec_ctx(port_id);
2236
2237         RTE_SET_USED(param);
2238
2239         if (type != RTE_ETH_EVENT_IPSEC)
2240                 return -1;
2241
2242         event_desc = ret_param;
2243         if (event_desc == NULL) {
2244                 printf("Event descriptor not set\n");
2245                 return -1;
2246         }
2247
2248         md = event_desc->metadata;
2249
2250         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2251                 return inline_ipsec_event_esn_overflow(ctx, md);
2252         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2253                 printf("Invalid IPsec event reported\n");
2254                 return -1;
2255         }
2256
2257         return -1;
2258 }
2259
2260 static uint16_t
2261 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2262         struct rte_mbuf *pkt[], uint16_t nb_pkts,
2263         __rte_unused uint16_t max_pkts, void *user_param)
2264 {
2265         uint64_t tm;
2266         uint32_t i, k;
2267         struct lcore_conf *lc;
2268         struct rte_mbuf *mb;
2269         struct rte_ether_hdr *eth;
2270
2271         lc = user_param;
2272         k = 0;
2273         tm = 0;
2274
2275         for (i = 0; i != nb_pkts; i++) {
2276
2277                 mb = pkt[i];
2278                 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2279                 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2280
2281                         struct rte_ipv4_hdr *iph;
2282
2283                         iph = (struct rte_ipv4_hdr *)(eth + 1);
2284                         if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2285
2286                                 mb->l2_len = sizeof(*eth);
2287                                 mb->l3_len = sizeof(*iph);
2288                                 tm = (tm != 0) ? tm : rte_rdtsc();
2289                                 mb = rte_ipv4_frag_reassemble_packet(
2290                                         lc->frag.tbl, &lc->frag.dr,
2291                                         mb, tm, iph);
2292
2293                                 if (mb != NULL) {
2294                                         /* fix ip cksum after reassemble. */
2295                                         iph = rte_pktmbuf_mtod_offset(mb,
2296                                                 struct rte_ipv4_hdr *,
2297                                                 mb->l2_len);
2298                                         iph->hdr_checksum = 0;
2299                                         iph->hdr_checksum = rte_ipv4_cksum(iph);
2300                                 }
2301                         }
2302                 } else if (eth->ether_type ==
2303                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2304
2305                         struct rte_ipv6_hdr *iph;
2306                         struct ipv6_extension_fragment *fh;
2307
2308                         iph = (struct rte_ipv6_hdr *)(eth + 1);
2309                         fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2310                         if (fh != NULL) {
2311                                 mb->l2_len = sizeof(*eth);
2312                                 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2313                                         sizeof(*fh);
2314                                 tm = (tm != 0) ? tm : rte_rdtsc();
2315                                 mb = rte_ipv6_frag_reassemble_packet(
2316                                         lc->frag.tbl, &lc->frag.dr,
2317                                         mb, tm, iph, fh);
2318                                 if (mb != NULL)
2319                                         /* fix l3_len after reassemble. */
2320                                         mb->l3_len = mb->l3_len - sizeof(*fh);
2321                         }
2322                 }
2323
2324                 pkt[k] = mb;
2325                 k += (mb != NULL);
2326         }
2327
2328         /* some fragments were encountered, drain death row */
2329         if (tm != 0)
2330                 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2331
2332         return k;
2333 }
2334
2335
2336 static int
2337 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2338 {
2339         int32_t sid;
2340         uint32_t i;
2341         uint64_t frag_cycles;
2342         const struct lcore_rx_queue *rxq;
2343         const struct rte_eth_rxtx_callback *cb;
2344
2345         /* create fragment table */
2346         sid = rte_lcore_to_socket_id(cid);
2347         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) /
2348                 MS_PER_S * FRAG_TTL_MS;
2349
2350         lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2351                 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2352         if (lc->frag.tbl == NULL) {
2353                 printf("%s(%u): failed to create fragment table of size: %u, "
2354                         "error code: %d\n",
2355                         __func__, cid, frag_tbl_sz, rte_errno);
2356                 return -ENOMEM;
2357         }
2358
2359         /* setup reassemble RX callbacks for all queues */
2360         for (i = 0; i != lc->nb_rx_queue; i++) {
2361
2362                 rxq = lc->rx_queue_list + i;
2363                 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2364                         rx_callback, lc);
2365                 if (cb == NULL) {
2366                         printf("%s(%u): failed to install RX callback for "
2367                                 "portid=%u, queueid=%u, error code: %d\n",
2368                                 __func__, cid,
2369                                 rxq->port_id, rxq->queue_id, rte_errno);
2370                         return -ENOMEM;
2371                 }
2372         }
2373
2374         return 0;
2375 }
2376
2377 static int
2378 reassemble_init(void)
2379 {
2380         int32_t rc;
2381         uint32_t i, lc;
2382
2383         rc = 0;
2384         for (i = 0; i != nb_lcore_params; i++) {
2385                 lc = lcore_params[i].lcore_id;
2386                 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2387                 if (rc != 0)
2388                         break;
2389         }
2390
2391         return rc;
2392 }
2393
2394 int32_t
2395 main(int32_t argc, char **argv)
2396 {
2397         int32_t ret;
2398         uint32_t lcore_id;
2399         uint32_t i;
2400         uint8_t socket_id;
2401         uint16_t portid;
2402         uint64_t req_rx_offloads, req_tx_offloads;
2403         size_t sess_sz;
2404
2405         /* init EAL */
2406         ret = rte_eal_init(argc, argv);
2407         if (ret < 0)
2408                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2409         argc -= ret;
2410         argv += ret;
2411
2412         /* parse application arguments (after the EAL ones) */
2413         ret = parse_args(argc, argv);
2414         if (ret < 0)
2415                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2416
2417         /* parse configuration file */
2418         if (parse_cfg_file(cfgfile) < 0) {
2419                 printf("parsing file \"%s\" failed\n",
2420                         optarg);
2421                 print_usage(argv[0]);
2422                 return -1;
2423         }
2424
2425         if ((unprotected_port_mask & enabled_port_mask) !=
2426                         unprotected_port_mask)
2427                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2428                                 unprotected_port_mask);
2429
2430         if (check_params() < 0)
2431                 rte_exit(EXIT_FAILURE, "check_params failed\n");
2432
2433         ret = init_lcore_rx_queues();
2434         if (ret < 0)
2435                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2436
2437         nb_lcores = rte_lcore_count();
2438
2439         sess_sz = max_session_size();
2440
2441         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2442                 if (rte_lcore_is_enabled(lcore_id) == 0)
2443                         continue;
2444
2445                 if (numa_on)
2446                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2447                 else
2448                         socket_id = 0;
2449
2450                 /* mbuf_pool is initialised by the pool_init() function*/
2451                 if (socket_ctx[socket_id].mbuf_pool)
2452                         continue;
2453
2454                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2455                 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2456                 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2457                         sess_sz);
2458         }
2459
2460         RTE_ETH_FOREACH_DEV(portid) {
2461                 if ((enabled_port_mask & (1 << portid)) == 0)
2462                         continue;
2463
2464                 sa_check_offloads(portid, &req_rx_offloads, &req_tx_offloads);
2465                 port_init(portid, req_rx_offloads, req_tx_offloads);
2466         }
2467
2468         cryptodevs_init();
2469
2470         /* start ports */
2471         RTE_ETH_FOREACH_DEV(portid) {
2472                 if ((enabled_port_mask & (1 << portid)) == 0)
2473                         continue;
2474
2475                 /*
2476                  * Start device
2477                  * note: device must be started before a flow rule
2478                  * can be installed.
2479                  */
2480                 ret = rte_eth_dev_start(portid);
2481                 if (ret < 0)
2482                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2483                                         "err=%d, port=%d\n", ret, portid);
2484                 /*
2485                  * If enabled, put device in promiscuous mode.
2486                  * This allows IO forwarding mode to forward packets
2487                  * to itself through 2 cross-connected  ports of the
2488                  * target machine.
2489                  */
2490                 if (promiscuous_on) {
2491                         ret = rte_eth_promiscuous_enable(portid);
2492                         if (ret != 0)
2493                                 rte_exit(EXIT_FAILURE,
2494                                         "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2495                                         rte_strerror(-ret), portid);
2496                 }
2497
2498                 rte_eth_dev_callback_register(portid,
2499                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2500         }
2501
2502         /* fragment reassemble is enabled */
2503         if (frag_tbl_sz != 0) {
2504                 ret = reassemble_init();
2505                 if (ret != 0)
2506                         rte_exit(EXIT_FAILURE, "failed at reassemble init");
2507         }
2508
2509         /* Replicate each context per socket */
2510         for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2511                 socket_id = rte_socket_id_by_idx(i);
2512                 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2513                         (socket_ctx[socket_id].sa_in == NULL) &&
2514                         (socket_ctx[socket_id].sa_out == NULL)) {
2515                         sa_init(&socket_ctx[socket_id], socket_id);
2516                         sp4_init(&socket_ctx[socket_id], socket_id);
2517                         sp6_init(&socket_ctx[socket_id], socket_id);
2518                         rt_init(&socket_ctx[socket_id], socket_id);
2519                 }
2520         }
2521
2522         check_all_ports_link_status(enabled_port_mask);
2523
2524         /* launch per-lcore init on every lcore */
2525         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2526         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2527                 if (rte_eal_wait_lcore(lcore_id) < 0)
2528                         return -1;
2529         }
2530
2531         return 0;
2532 }