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