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 #include "sad.h"
  50
  51 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  52
  53 #define MAX_JUMBO_PKT_LEN  9600
  54
  55 #define MEMPOOL_CACHE_SIZE 256
  56
  57 #define NB_MBUF (32000)
  58
  59 #define CDEV_QUEUE_DESC 2048
  60 #define CDEV_MAP_ENTRIES 16384
  61 #define CDEV_MP_NB_OBJS 1024
  62 #define CDEV_MP_CACHE_SZ 64
  63 #define MAX_QUEUE_PAIRS 1
  64
  65 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  66
  67 #define NB_SOCKETS 4
  68
  69 /* Configure how many packets ahead to prefetch, when reading packets */
  70 #define PREFETCH_OFFSET 3
  71
  72 #define MAX_RX_QUEUE_PER_LCORE 16
  73
  74 #define MAX_LCORE_PARAMS 1024
  75
  76 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
  77
  78 /*
  79  * Configurable number of RX/TX ring descriptors
  80  */
  81 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  82 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  83 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  84 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  85
  86 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
  87 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  88         (((uint64_t)((a) & 0xff) << 56) | \
  89         ((uint64_t)((b) & 0xff) << 48) | \
  90         ((uint64_t)((c) & 0xff) << 40) | \
  91         ((uint64_t)((d) & 0xff) << 32) | \
  92         ((uint64_t)((e) & 0xff) << 24) | \
  93         ((uint64_t)((f) & 0xff) << 16) | \
  94         ((uint64_t)((g) & 0xff) << 8)  | \
  95         ((uint64_t)(h) & 0xff))
  96 #else
  97 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  98         (((uint64_t)((h) & 0xff) << 56) | \
  99         ((uint64_t)((g) & 0xff) << 48) | \
 100         ((uint64_t)((f) & 0xff) << 40) | \
 101         ((uint64_t)((e) & 0xff) << 32) | \
 102         ((uint64_t)((d) & 0xff) << 24) | \
 103         ((uint64_t)((c) & 0xff) << 16) | \
 104         ((uint64_t)((b) & 0xff) << 8) | \
 105         ((uint64_t)(a) & 0xff))
 106 #endif
 107 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 108
 109 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 110                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
 111                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
 112                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
 113                 0, 0)
 114
 115 #define FRAG_TBL_BUCKET_ENTRIES 4
 116 #define MAX_FRAG_TTL_NS         (10LL * NS_PER_S)
 117
 118 #define MTU_TO_FRAMELEN(x)      ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
 119
 120 /* port/source ethernet addr and destination ethernet addr */
 121 struct ethaddr_info {
 122         uint64_t src, dst;
 123 };
 124
 125 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 126         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 127         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 128         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 129         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 130 };
 131
 132 #define CMD_LINE_OPT_CONFIG             "config"
 133 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 134 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 135 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
 136 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
 137 #define CMD_LINE_OPT_REASSEMBLE         "reassemble"
 138 #define CMD_LINE_OPT_MTU                "mtu"
 139 #define CMD_LINE_OPT_FRAG_TTL           "frag-ttl"
 140
 141 enum {
 142         /* long options mapped to a short option */
 143
 144         /* first long only option value must be >= 256, so that we won't
 145          * conflict with short options
 146          */
 147         CMD_LINE_OPT_MIN_NUM = 256,
 148         CMD_LINE_OPT_CONFIG_NUM,
 149         CMD_LINE_OPT_SINGLE_SA_NUM,
 150         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 151         CMD_LINE_OPT_RX_OFFLOAD_NUM,
 152         CMD_LINE_OPT_TX_OFFLOAD_NUM,
 153         CMD_LINE_OPT_REASSEMBLE_NUM,
 154         CMD_LINE_OPT_MTU_NUM,
 155         CMD_LINE_OPT_FRAG_TTL_NUM,
 156 };
 157
 158 static const struct option lgopts[] = {
 159         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 160         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 161         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 162         {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
 163         {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
 164         {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
 165         {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
 166         {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
 167         {NULL, 0, 0, 0}
 168 };
 169
 170 /* mask of enabled ports */
 171 static uint32_t enabled_port_mask;
 172 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 173 static uint32_t unprotected_port_mask;
 174 static int32_t promiscuous_on = 1;
 175 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 176 static uint32_t nb_lcores;
 177 static uint32_t single_sa;
 178 static uint32_t single_sa_idx;
 179
 180 /*
 181  * RX/TX HW offload capabilities to enable/use on ethernet ports.
 182  * By default all capabilities are enabled.
 183  */
 184 static uint64_t dev_rx_offload = UINT64_MAX;
 185 static uint64_t dev_tx_offload = UINT64_MAX;
 186
 187 /*
 188  * global values that determine multi-seg policy
 189  */
 190 static uint32_t frag_tbl_sz;
 191 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
 192 static uint32_t mtu_size = RTE_ETHER_MTU;
 193 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
 194
 195 /* application wide librte_ipsec/SA parameters */
 196 struct app_sa_prm app_sa_prm = {
 197                         .enable = 0,
 198                         .cache_sz = SA_CACHE_SZ
 199                 };
 200 static const char *cfgfile;
 201
 202 struct lcore_rx_queue {
 203         uint16_t port_id;
 204         uint8_t queue_id;
 205 } __rte_cache_aligned;
 206
 207 struct lcore_params {
 208         uint16_t port_id;
 209         uint8_t queue_id;
 210         uint8_t lcore_id;
 211 } __rte_cache_aligned;
 212
 213 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 214
 215 static struct lcore_params *lcore_params;
 216 static uint16_t nb_lcore_params;
 217
 218 static struct rte_hash *cdev_map_in;
 219 static struct rte_hash *cdev_map_out;
 220
 221 struct buffer {
 222         uint16_t len;
 223         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 224 };
 225
 226 struct lcore_conf {
 227         uint16_t nb_rx_queue;
 228         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 229         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 230         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 231         struct ipsec_ctx inbound;
 232         struct ipsec_ctx outbound;
 233         struct rt_ctx *rt4_ctx;
 234         struct rt_ctx *rt6_ctx;
 235         struct {
 236                 struct rte_ip_frag_tbl *tbl;
 237                 struct rte_mempool *pool_dir;
 238                 struct rte_mempool *pool_indir;
 239                 struct rte_ip_frag_death_row dr;
 240         } frag;
 241 } __rte_cache_aligned;
 242
 243 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 244
 245 static struct rte_eth_conf port_conf = {
 246         .rxmode = {
 247                 .mq_mode        = ETH_MQ_RX_RSS,
 248                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 249                 .split_hdr_size = 0,
 250                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 251         },
 252         .rx_adv_conf = {
 253                 .rss_conf = {
 254                         .rss_key = NULL,
 255                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 256                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 257                 },
 258         },
 259         .txmode = {
 260                 .mq_mode = ETH_MQ_TX_NONE,
 261         },
 262 };
 263
 264 static struct socket_ctx socket_ctx[NB_SOCKETS];
 265
 266 /*
 267  * Determine is multi-segment support required:
 268  *  - either frame buffer size is smaller then mtu
 269  *  - or reassmeble support is requested
 270  */
 271 static int
 272 multi_seg_required(void)
 273 {
 274         return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
 275                 frame_buf_size || frag_tbl_sz != 0);
 276 }
 277
 278 static inline void
 279 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
 280         uint32_t l2_len)
 281 {
 282         uint32_t plen, trim;
 283
 284         plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
 285         if (plen < m->pkt_len) {
 286                 trim = m->pkt_len - plen;
 287                 rte_pktmbuf_trim(m, trim);
 288         }
 289 }
 290
 291 static inline void
 292 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
 293         uint32_t l2_len)
 294 {
 295         uint32_t plen, trim;
 296
 297         plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
 298         if (plen < m->pkt_len) {
 299                 trim = m->pkt_len - plen;
 300                 rte_pktmbuf_trim(m, trim);
 301         }
 302 }
 303
 304 static inline void
 305 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 306 {
 307         const struct rte_ether_hdr *eth;
 308         const struct rte_ipv4_hdr *iph4;
 309         const struct rte_ipv6_hdr *iph6;
 310
 311         eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
 312         if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 313
 314                 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
 315                         RTE_ETHER_HDR_LEN);
 316                 adjust_ipv4_pktlen(pkt, iph4, 0);
 317
 318                 if (iph4->next_proto_id == IPPROTO_ESP)
 319                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 320                 else {
 321                         t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
 322                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 323                 }
 324                 pkt->l2_len = 0;
 325                 pkt->l3_len = sizeof(*iph4);
 326                 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
 327         } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 328                 int next_proto;
 329                 size_t l3len, ext_len;
 330                 uint8_t *p;
 331
 332                 /* get protocol type */
 333                 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
 334                         RTE_ETHER_HDR_LEN);
 335                 adjust_ipv6_pktlen(pkt, iph6, 0);
 336
 337                 next_proto = iph6->proto;
 338
 339                 /* determine l3 header size up to ESP extension */
 340                 l3len = sizeof(struct ip6_hdr);
 341                 p = rte_pktmbuf_mtod(pkt, uint8_t *);
 342                 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
 343                         (next_proto = rte_ipv6_get_next_ext(p + l3len,
 344                                                 next_proto, &ext_len)) >= 0)
 345                         l3len += ext_len;
 346
 347                 /* drop packet when IPv6 header exceeds first segment length */
 348                 if (unlikely(l3len > pkt->data_len)) {
 349                         rte_pktmbuf_free(pkt);
 350                         return;
 351                 }
 352
 353                 if (next_proto == IPPROTO_ESP)
 354                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 355                 else {
 356                         t->ip6.data[t->ip6.num] = &iph6->proto;
 357                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 358                 }
 359                 pkt->l2_len = 0;
 360                 pkt->l3_len = l3len;
 361                 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
 362         } else {
 363                 /* Unknown/Unsupported type, drop the packet */
 364                 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
 365                         rte_be_to_cpu_16(eth->ether_type));
 366                 rte_pktmbuf_free(pkt);
 367                 return;
 368         }
 369
 370         /* Check if the packet has been processed inline. For inline protocol
 371          * processed packets, the metadata in the mbuf can be used to identify
 372          * the security processing done on the packet. The metadata will be
 373          * used to retrieve the application registered userdata associated
 374          * with the security session.
 375          */
 376
 377         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 378                 struct ipsec_sa *sa;
 379                 struct ipsec_mbuf_metadata *priv;
 380                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 381                                                 rte_eth_dev_get_sec_ctx(
 382                                                 pkt->port);
 383
 384                 /* Retrieve the userdata registered. Here, the userdata
 385                  * registered is the SA pointer.
 386                  */
 387
 388                 sa = (struct ipsec_sa *)
 389                                 rte_security_get_userdata(ctx, pkt->udata64);
 390
 391                 if (sa == NULL) {
 392                         /* userdata could not be retrieved */
 393                         return;
 394                 }
 395
 396                 /* Save SA as priv member in mbuf. This will be used in the
 397                  * IPsec selector(SP-SA) check.
 398                  */
 399
 400                 priv = get_priv(pkt);
 401                 priv->sa = sa;
 402         }
 403 }
 404
 405 static inline void
 406 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 407                 uint16_t nb_pkts)
 408 {
 409         int32_t i;
 410
 411         t->ipsec.num = 0;
 412         t->ip4.num = 0;
 413         t->ip6.num = 0;
 414
 415         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 416                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 417                                         void *));
 418                 prepare_one_packet(pkts[i], t);
 419         }
 420         /* Process left packets */
 421         for (; i < nb_pkts; i++)
 422                 prepare_one_packet(pkts[i], t);
 423 }
 424
 425 static inline void
 426 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 427                 const struct lcore_conf *qconf)
 428 {
 429         struct ip *ip;
 430         struct rte_ether_hdr *ethhdr;
 431
 432         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 433
 434         ethhdr = (struct rte_ether_hdr *)
 435                 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
 436
 437         if (ip->ip_v == IPVERSION) {
 438                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 439                 pkt->l3_len = sizeof(struct ip);
 440                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 441
 442                 ip->ip_sum = 0;
 443
 444                 /* calculate IPv4 cksum in SW */
 445                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 446                         ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
 447
 448                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
 449         } else {
 450                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 451                 pkt->l3_len = sizeof(struct ip6_hdr);
 452                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 453
 454                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
 455         }
 456
 457         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 458                         sizeof(struct rte_ether_addr));
 459         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 460                         sizeof(struct rte_ether_addr));
 461 }
 462
 463 static inline void
 464 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 465                 const struct lcore_conf *qconf)
 466 {
 467         int32_t i;
 468         const int32_t prefetch_offset = 2;
 469
 470         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 471                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 472                 prepare_tx_pkt(pkts[i], port, qconf);
 473         }
 474         /* Process left packets */
 475         for (; i < nb_pkts; i++)
 476                 prepare_tx_pkt(pkts[i], port, qconf);
 477 }
 478
 479 /* Send burst of packets on an output interface */
 480 static inline int32_t
 481 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 482 {
 483         struct rte_mbuf **m_table;
 484         int32_t ret;
 485         uint16_t queueid;
 486
 487         queueid = qconf->tx_queue_id[port];
 488         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 489
 490         prepare_tx_burst(m_table, n, port, qconf);
 491
 492         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 493         if (unlikely(ret < n)) {
 494                 do {
 495                         rte_pktmbuf_free(m_table[ret]);
 496                 } while (++ret < n);
 497         }
 498
 499         return 0;
 500 }
 501
 502 /*
 503  * Helper function to fragment and queue for TX one packet.
 504  */
 505 static inline uint32_t
 506 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
 507         uint16_t port, uint8_t proto)
 508 {
 509         struct buffer *tbl;
 510         uint32_t len, n;
 511         int32_t rc;
 512
 513         tbl =  qconf->tx_mbufs + port;
 514         len = tbl->len;
 515
 516         /* free space for new fragments */
 517         if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >=  RTE_DIM(tbl->m_table)) {
 518                 send_burst(qconf, len, port);
 519                 len = 0;
 520         }
 521
 522         n = RTE_DIM(tbl->m_table) - len;
 523
 524         if (proto == IPPROTO_IP)
 525                 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
 526                         n, mtu_size, qconf->frag.pool_dir,
 527                         qconf->frag.pool_indir);
 528         else
 529                 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
 530                         n, mtu_size, qconf->frag.pool_dir,
 531                         qconf->frag.pool_indir);
 532
 533         if (rc >= 0)
 534                 len += rc;
 535         else
 536                 RTE_LOG(ERR, IPSEC,
 537                         "%s: failed to fragment packet with size %u, "
 538                         "error code: %d\n",
 539                         __func__, m->pkt_len, rte_errno);
 540
 541         rte_pktmbuf_free(m);
 542         return len;
 543 }
 544
 545 /* Enqueue a single packet, and send burst if queue is filled */
 546 static inline int32_t
 547 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
 548 {
 549         uint32_t lcore_id;
 550         uint16_t len;
 551         struct lcore_conf *qconf;
 552
 553         lcore_id = rte_lcore_id();
 554
 555         qconf = &lcore_conf[lcore_id];
 556         len = qconf->tx_mbufs[port].len;
 557
 558         if (m->pkt_len <= mtu_size) {
 559                 qconf->tx_mbufs[port].m_table[len] = m;
 560                 len++;
 561
 562         /* need to fragment the packet */
 563         } else if (frag_tbl_sz > 0)
 564                 len = send_fragment_packet(qconf, m, port, proto);
 565         else
 566                 rte_pktmbuf_free(m);
 567
 568         /* enough pkts to be sent */
 569         if (unlikely(len == MAX_PKT_BURST)) {
 570                 send_burst(qconf, MAX_PKT_BURST, port);
 571                 len = 0;
 572         }
 573
 574         qconf->tx_mbufs[port].len = len;
 575         return 0;
 576 }
 577
 578 static inline void
 579 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 580                 uint16_t lim)
 581 {
 582         struct rte_mbuf *m;
 583         uint32_t i, j, res, sa_idx;
 584
 585         if (ip->num == 0 || sp == NULL)
 586                 return;
 587
 588         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 589                         ip->num, DEFAULT_MAX_CATEGORIES);
 590
 591         j = 0;
 592         for (i = 0; i < ip->num; i++) {
 593                 m = ip->pkts[i];
 594                 res = ip->res[i];
 595                 if (res == BYPASS) {
 596                         ip->pkts[j++] = m;
 597                         continue;
 598                 }
 599                 if (res == DISCARD) {
 600                         rte_pktmbuf_free(m);
 601                         continue;
 602                 }
 603
 604                 /* Only check SPI match for processed IPSec packets */
 605                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 606                         rte_pktmbuf_free(m);
 607                         continue;
 608                 }
 609
 610                 sa_idx = res - 1;
 611                 if (!inbound_sa_check(sa, m, sa_idx)) {
 612                         rte_pktmbuf_free(m);
 613                         continue;
 614                 }
 615                 ip->pkts[j++] = m;
 616         }
 617         ip->num = j;
 618 }
 619
 620 static void
 621 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 622 {
 623         uint32_t i, n4, n6;
 624         struct ip *ip;
 625         struct rte_mbuf *m;
 626
 627         n4 = trf->ip4.num;
 628         n6 = trf->ip6.num;
 629
 630         for (i = 0; i < num; i++) {
 631
 632                 m = mb[i];
 633                 ip = rte_pktmbuf_mtod(m, struct ip *);
 634
 635                 if (ip->ip_v == IPVERSION) {
 636                         trf->ip4.pkts[n4] = m;
 637                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 638                                         uint8_t *, offsetof(struct ip, ip_p));
 639                         n4++;
 640                 } else if (ip->ip_v == IP6_VERSION) {
 641                         trf->ip6.pkts[n6] = m;
 642                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 643                                         uint8_t *,
 644                                         offsetof(struct ip6_hdr, ip6_nxt));
 645                         n6++;
 646                 } else
 647                         rte_pktmbuf_free(m);
 648         }
 649
 650         trf->ip4.num = n4;
 651         trf->ip6.num = n6;
 652 }
 653
 654
 655 static inline void
 656 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 657                 struct ipsec_traffic *traffic)
 658 {
 659         uint16_t nb_pkts_in, n_ip4, n_ip6;
 660
 661         n_ip4 = traffic->ip4.num;
 662         n_ip6 = traffic->ip6.num;
 663
 664         if (app_sa_prm.enable == 0) {
 665                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 666                                 traffic->ipsec.num, MAX_PKT_BURST);
 667                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 668         } else {
 669                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 670                         traffic->ipsec.saptr, traffic->ipsec.num);
 671                 ipsec_process(ipsec_ctx, traffic);
 672         }
 673
 674         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 675                         n_ip4);
 676
 677         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 678                         n_ip6);
 679 }
 680
 681 static inline void
 682 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 683                 struct traffic_type *ipsec)
 684 {
 685         struct rte_mbuf *m;
 686         uint32_t i, j, sa_idx;
 687
 688         if (ip->num == 0 || sp == NULL)
 689                 return;
 690
 691         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 692                         ip->num, DEFAULT_MAX_CATEGORIES);
 693
 694         j = 0;
 695         for (i = 0; i < ip->num; i++) {
 696                 m = ip->pkts[i];
 697                 sa_idx = ip->res[i] - 1;
 698                 if (ip->res[i] == DISCARD)
 699                         rte_pktmbuf_free(m);
 700                 else if (ip->res[i] == BYPASS)
 701                         ip->pkts[j++] = m;
 702                 else {
 703                         ipsec->res[ipsec->num] = sa_idx;
 704                         ipsec->pkts[ipsec->num++] = m;
 705                 }
 706         }
 707         ip->num = j;
 708 }
 709
 710 static inline void
 711 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 712                 struct ipsec_traffic *traffic)
 713 {
 714         struct rte_mbuf *m;
 715         uint16_t idx, nb_pkts_out, i;
 716
 717         /* Drop any IPsec traffic from protected ports */
 718         for (i = 0; i < traffic->ipsec.num; i++)
 719                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 720
 721         traffic->ipsec.num = 0;
 722
 723         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 724
 725         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 726
 727         if (app_sa_prm.enable == 0) {
 728
 729                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 730                                 traffic->ipsec.res, traffic->ipsec.num,
 731                                 MAX_PKT_BURST);
 732
 733                 for (i = 0; i < nb_pkts_out; i++) {
 734                         m = traffic->ipsec.pkts[i];
 735                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 736                         if (ip->ip_v == IPVERSION) {
 737                                 idx = traffic->ip4.num++;
 738                                 traffic->ip4.pkts[idx] = m;
 739                         } else {
 740                                 idx = traffic->ip6.num++;
 741                                 traffic->ip6.pkts[idx] = m;
 742                         }
 743                 }
 744         } else {
 745                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 746                         traffic->ipsec.saptr, traffic->ipsec.num);
 747                 ipsec_process(ipsec_ctx, traffic);
 748         }
 749 }
 750
 751 static inline void
 752 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 753                 struct ipsec_traffic *traffic)
 754 {
 755         struct rte_mbuf *m;
 756         uint32_t nb_pkts_in, i, idx;
 757
 758         /* Drop any IPv4 traffic from unprotected ports */
 759         for (i = 0; i < traffic->ip4.num; i++)
 760                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 761
 762         traffic->ip4.num = 0;
 763
 764         /* Drop any IPv6 traffic from unprotected ports */
 765         for (i = 0; i < traffic->ip6.num; i++)
 766                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 767
 768         traffic->ip6.num = 0;
 769
 770         if (app_sa_prm.enable == 0) {
 771
 772                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 773                                 traffic->ipsec.num, MAX_PKT_BURST);
 774
 775                 for (i = 0; i < nb_pkts_in; i++) {
 776                         m = traffic->ipsec.pkts[i];
 777                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 778                         if (ip->ip_v == IPVERSION) {
 779                                 idx = traffic->ip4.num++;
 780                                 traffic->ip4.pkts[idx] = m;
 781                         } else {
 782                                 idx = traffic->ip6.num++;
 783                                 traffic->ip6.pkts[idx] = m;
 784                         }
 785                 }
 786         } else {
 787                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 788                         traffic->ipsec.saptr, traffic->ipsec.num);
 789                 ipsec_process(ipsec_ctx, traffic);
 790         }
 791 }
 792
 793 static inline void
 794 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 795                 struct ipsec_traffic *traffic)
 796 {
 797         struct rte_mbuf *m;
 798         uint32_t nb_pkts_out, i, n;
 799         struct ip *ip;
 800
 801         /* Drop any IPsec traffic from protected ports */
 802         for (i = 0; i < traffic->ipsec.num; i++)
 803                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 804
 805         n = 0;
 806
 807         for (i = 0; i < traffic->ip4.num; i++) {
 808                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 809                 traffic->ipsec.res[n++] = single_sa_idx;
 810         }
 811
 812         for (i = 0; i < traffic->ip6.num; i++) {
 813                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 814                 traffic->ipsec.res[n++] = single_sa_idx;
 815         }
 816
 817         traffic->ip4.num = 0;
 818         traffic->ip6.num = 0;
 819         traffic->ipsec.num = n;
 820
 821         if (app_sa_prm.enable == 0) {
 822
 823                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 824                                 traffic->ipsec.res, traffic->ipsec.num,
 825                                 MAX_PKT_BURST);
 826
 827                 /* They all sue the same SA (ip4 or ip6 tunnel) */
 828                 m = traffic->ipsec.pkts[0];
 829                 ip = rte_pktmbuf_mtod(m, struct ip *);
 830                 if (ip->ip_v == IPVERSION) {
 831                         traffic->ip4.num = nb_pkts_out;
 832                         for (i = 0; i < nb_pkts_out; i++)
 833                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 834                 } else {
 835                         traffic->ip6.num = nb_pkts_out;
 836                         for (i = 0; i < nb_pkts_out; i++)
 837                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 838                 }
 839         } else {
 840                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 841                         traffic->ipsec.saptr, traffic->ipsec.num);
 842                 ipsec_process(ipsec_ctx, traffic);
 843         }
 844 }
 845
 846 static inline int32_t
 847 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 848 {
 849         struct ipsec_mbuf_metadata *priv;
 850         struct ipsec_sa *sa;
 851
 852         priv = get_priv(pkt);
 853
 854         sa = priv->sa;
 855         if (unlikely(sa == NULL)) {
 856                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 857                 goto fail;
 858         }
 859
 860         if (is_ipv6)
 861                 return sa->portid;
 862
 863         /* else */
 864         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 865
 866 fail:
 867         if (is_ipv6)
 868                 return -1;
 869
 870         /* else */
 871         return 0;
 872 }
 873
 874 static inline void
 875 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 876 {
 877         uint32_t hop[MAX_PKT_BURST * 2];
 878         uint32_t dst_ip[MAX_PKT_BURST * 2];
 879         int32_t pkt_hop = 0;
 880         uint16_t i, offset;
 881         uint16_t lpm_pkts = 0;
 882
 883         if (nb_pkts == 0)
 884                 return;
 885
 886         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 887          * have port ID in the SA
 888          */
 889
 890         for (i = 0; i < nb_pkts; i++) {
 891                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 892                         /* Security offload not enabled. So an LPM lookup is
 893                          * required to get the hop
 894                          */
 895                         offset = offsetof(struct ip, ip_dst);
 896                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 897                                         uint32_t *, offset);
 898                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 899                         lpm_pkts++;
 900                 }
 901         }
 902
 903         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 904
 905         lpm_pkts = 0;
 906
 907         for (i = 0; i < nb_pkts; i++) {
 908                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 909                         /* Read hop from the SA */
 910                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 911                 } else {
 912                         /* Need to use hop returned by lookup */
 913                         pkt_hop = hop[lpm_pkts++];
 914                 }
 915
 916                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 917                         rte_pktmbuf_free(pkts[i]);
 918                         continue;
 919                 }
 920                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
 921         }
 922 }
 923
 924 static inline void
 925 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 926 {
 927         int32_t hop[MAX_PKT_BURST * 2];
 928         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 929         uint8_t *ip6_dst;
 930         int32_t pkt_hop = 0;
 931         uint16_t i, offset;
 932         uint16_t lpm_pkts = 0;
 933
 934         if (nb_pkts == 0)
 935                 return;
 936
 937         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 938          * have port ID in the SA
 939          */
 940
 941         for (i = 0; i < nb_pkts; i++) {
 942                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 943                         /* Security offload not enabled. So an LPM lookup is
 944                          * required to get the hop
 945                          */
 946                         offset = offsetof(struct ip6_hdr, ip6_dst);
 947                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 948                                         offset);
 949                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 950                         lpm_pkts++;
 951                 }
 952         }
 953
 954         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 955                         lpm_pkts);
 956
 957         lpm_pkts = 0;
 958
 959         for (i = 0; i < nb_pkts; i++) {
 960                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 961                         /* Read hop from the SA */
 962                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 963                 } else {
 964                         /* Need to use hop returned by lookup */
 965                         pkt_hop = hop[lpm_pkts++];
 966                 }
 967
 968                 if (pkt_hop == -1) {
 969                         rte_pktmbuf_free(pkts[i]);
 970                         continue;
 971                 }
 972                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
 973         }
 974 }
 975
 976 static inline void
 977 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 978                 uint8_t nb_pkts, uint16_t portid)
 979 {
 980         struct ipsec_traffic traffic;
 981
 982         prepare_traffic(pkts, &traffic, nb_pkts);
 983
 984         if (unlikely(single_sa)) {
 985                 if (UNPROTECTED_PORT(portid))
 986                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 987                 else
 988                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 989         } else {
 990                 if (UNPROTECTED_PORT(portid))
 991                         process_pkts_inbound(&qconf->inbound, &traffic);
 992                 else
 993                         process_pkts_outbound(&qconf->outbound, &traffic);
 994         }
 995
 996         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 997         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 998 }
 999
1000 static inline void
1001 drain_tx_buffers(struct lcore_conf *qconf)
1002 {
1003         struct buffer *buf;
1004         uint32_t portid;
1005
1006         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1007                 buf = &qconf->tx_mbufs[portid];
1008                 if (buf->len == 0)
1009                         continue;
1010                 send_burst(qconf, buf->len, portid);
1011                 buf->len = 0;
1012         }
1013 }
1014
1015 static inline void
1016 drain_crypto_buffers(struct lcore_conf *qconf)
1017 {
1018         uint32_t i;
1019         struct ipsec_ctx *ctx;
1020
1021         /* drain inbound buffers*/
1022         ctx = &qconf->inbound;
1023         for (i = 0; i != ctx->nb_qps; i++) {
1024                 if (ctx->tbl[i].len != 0)
1025                         enqueue_cop_burst(ctx->tbl  + i);
1026         }
1027
1028         /* drain outbound buffers*/
1029         ctx = &qconf->outbound;
1030         for (i = 0; i != ctx->nb_qps; i++) {
1031                 if (ctx->tbl[i].len != 0)
1032                         enqueue_cop_burst(ctx->tbl  + i);
1033         }
1034 }
1035
1036 static void
1037 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1038                 struct ipsec_ctx *ctx)
1039 {
1040         uint32_t n;
1041         struct ipsec_traffic trf;
1042
1043         if (app_sa_prm.enable == 0) {
1044
1045                 /* dequeue packets from crypto-queue */
1046                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1047                         RTE_DIM(trf.ipsec.pkts));
1048
1049                 trf.ip4.num = 0;
1050                 trf.ip6.num = 0;
1051
1052                 /* split traffic by ipv4-ipv6 */
1053                 split46_traffic(&trf, trf.ipsec.pkts, n);
1054         } else
1055                 ipsec_cqp_process(ctx, &trf);
1056
1057         /* process ipv4 packets */
1058         if (trf.ip4.num != 0) {
1059                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1060                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1061         }
1062
1063         /* process ipv6 packets */
1064         if (trf.ip6.num != 0) {
1065                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1066                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1067         }
1068 }
1069
1070 static void
1071 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1072                 struct ipsec_ctx *ctx)
1073 {
1074         uint32_t n;
1075         struct ipsec_traffic trf;
1076
1077         if (app_sa_prm.enable == 0) {
1078
1079                 /* dequeue packets from crypto-queue */
1080                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1081                         RTE_DIM(trf.ipsec.pkts));
1082
1083                 trf.ip4.num = 0;
1084                 trf.ip6.num = 0;
1085
1086                 /* split traffic by ipv4-ipv6 */
1087                 split46_traffic(&trf, trf.ipsec.pkts, n);
1088         } else
1089                 ipsec_cqp_process(ctx, &trf);
1090
1091         /* process ipv4 packets */
1092         if (trf.ip4.num != 0)
1093                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1094
1095         /* process ipv6 packets */
1096         if (trf.ip6.num != 0)
1097                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1098 }
1099
1100 /* main processing loop */
1101 static int32_t
1102 main_loop(__attribute__((unused)) void *dummy)
1103 {
1104         struct rte_mbuf *pkts[MAX_PKT_BURST];
1105         uint32_t lcore_id;
1106         uint64_t prev_tsc, diff_tsc, cur_tsc;
1107         int32_t i, nb_rx;
1108         uint16_t portid;
1109         uint8_t queueid;
1110         struct lcore_conf *qconf;
1111         int32_t rc, socket_id;
1112         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1113                         / US_PER_S * BURST_TX_DRAIN_US;
1114         struct lcore_rx_queue *rxql;
1115
1116         prev_tsc = 0;
1117         lcore_id = rte_lcore_id();
1118         qconf = &lcore_conf[lcore_id];
1119         rxql = qconf->rx_queue_list;
1120         socket_id = rte_lcore_to_socket_id(lcore_id);
1121
1122         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1123         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1124         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1125         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1126         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1127         qconf->inbound.cdev_map = cdev_map_in;
1128         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1129         qconf->inbound.session_priv_pool =
1130                         socket_ctx[socket_id].session_priv_pool;
1131         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1132         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1133         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1134         qconf->outbound.cdev_map = cdev_map_out;
1135         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1136         qconf->outbound.session_priv_pool =
1137                         socket_ctx[socket_id].session_priv_pool;
1138         qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1139         qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1140
1141         rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1142         if (rc != 0) {
1143                 RTE_LOG(ERR, IPSEC,
1144                         "SAD cache init on lcore %u, failed with code: %d\n",
1145                         lcore_id, rc);
1146                 return rc;
1147         }
1148
1149         if (qconf->nb_rx_queue == 0) {
1150                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1151                         lcore_id);
1152                 return 0;
1153         }
1154
1155         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1156
1157         for (i = 0; i < qconf->nb_rx_queue; i++) {
1158                 portid = rxql[i].port_id;
1159                 queueid = rxql[i].queue_id;
1160                 RTE_LOG(INFO, IPSEC,
1161                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1162                         lcore_id, portid, queueid);
1163         }
1164
1165         while (1) {
1166                 cur_tsc = rte_rdtsc();
1167
1168                 /* TX queue buffer drain */
1169                 diff_tsc = cur_tsc - prev_tsc;
1170
1171                 if (unlikely(diff_tsc > drain_tsc)) {
1172                         drain_tx_buffers(qconf);
1173                         drain_crypto_buffers(qconf);
1174                         prev_tsc = cur_tsc;
1175                 }
1176
1177                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1178
1179                         /* Read packets from RX queues */
1180                         portid = rxql[i].port_id;
1181                         queueid = rxql[i].queue_id;
1182                         nb_rx = rte_eth_rx_burst(portid, queueid,
1183                                         pkts, MAX_PKT_BURST);
1184
1185                         if (nb_rx > 0)
1186                                 process_pkts(qconf, pkts, nb_rx, portid);
1187
1188                         /* dequeue and process completed crypto-ops */
1189                         if (UNPROTECTED_PORT(portid))
1190                                 drain_inbound_crypto_queues(qconf,
1191                                         &qconf->inbound);
1192                         else
1193                                 drain_outbound_crypto_queues(qconf,
1194                                         &qconf->outbound);
1195                 }
1196         }
1197 }
1198
1199 static int32_t
1200 check_params(void)
1201 {
1202         uint8_t lcore;
1203         uint16_t portid;
1204         uint16_t i;
1205         int32_t socket_id;
1206
1207         if (lcore_params == NULL) {
1208                 printf("Error: No port/queue/core mappings\n");
1209                 return -1;
1210         }
1211
1212         for (i = 0; i < nb_lcore_params; ++i) {
1213                 lcore = lcore_params[i].lcore_id;
1214                 if (!rte_lcore_is_enabled(lcore)) {
1215                         printf("error: lcore %hhu is not enabled in "
1216                                 "lcore mask\n", lcore);
1217                         return -1;
1218                 }
1219                 socket_id = rte_lcore_to_socket_id(lcore);
1220                 if (socket_id != 0 && numa_on == 0) {
1221                         printf("warning: lcore %hhu is on socket %d "
1222                                 "with numa off\n",
1223                                 lcore, socket_id);
1224                 }
1225                 portid = lcore_params[i].port_id;
1226                 if ((enabled_port_mask & (1 << portid)) == 0) {
1227                         printf("port %u is not enabled in port mask\n", portid);
1228                         return -1;
1229                 }
1230                 if (!rte_eth_dev_is_valid_port(portid)) {
1231                         printf("port %u is not present on the board\n", portid);
1232                         return -1;
1233                 }
1234         }
1235         return 0;
1236 }
1237
1238 static uint8_t
1239 get_port_nb_rx_queues(const uint16_t port)
1240 {
1241         int32_t queue = -1;
1242         uint16_t i;
1243
1244         for (i = 0; i < nb_lcore_params; ++i) {
1245                 if (lcore_params[i].port_id == port &&
1246                                 lcore_params[i].queue_id > queue)
1247                         queue = lcore_params[i].queue_id;
1248         }
1249         return (uint8_t)(++queue);
1250 }
1251
1252 static int32_t
1253 init_lcore_rx_queues(void)
1254 {
1255         uint16_t i, nb_rx_queue;
1256         uint8_t lcore;
1257
1258         for (i = 0; i < nb_lcore_params; ++i) {
1259                 lcore = lcore_params[i].lcore_id;
1260                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1261                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1262                         printf("error: too many queues (%u) for lcore: %u\n",
1263                                         nb_rx_queue + 1, lcore);
1264                         return -1;
1265                 }
1266                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1267                         lcore_params[i].port_id;
1268                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1269                         lcore_params[i].queue_id;
1270                 lcore_conf[lcore].nb_rx_queue++;
1271         }
1272         return 0;
1273 }
1274
1275 /* display usage */
1276 static void
1277 print_usage(const char *prgname)
1278 {
1279         fprintf(stderr, "%s [EAL options] --"
1280                 " -p PORTMASK"
1281                 " [-P]"
1282                 " [-u PORTMASK]"
1283                 " [-j FRAMESIZE]"
1284                 " [-l]"
1285                 " [-w REPLAY_WINDOW_SIZE]"
1286                 " [-e]"
1287                 " [-a]"
1288                 " [-c]"
1289                 " -f CONFIG_FILE"
1290                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1291                 " [--single-sa SAIDX]"
1292                 " [--cryptodev_mask MASK]"
1293                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1294                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1295                 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1296                 " [--" CMD_LINE_OPT_MTU " MTU]"
1297                 "\n\n"
1298                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1299                 "  -P : Enable promiscuous mode\n"
1300                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1301                 "  -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1302                 "     value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1303                 "  -l enables code-path that uses librte_ipsec\n"
1304                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1305                 "     size for each SA\n"
1306                 "  -e enables ESN\n"
1307                 "  -a enables SA SQN atomic behaviour\n"
1308                 "  -c specifies inbound SAD cache size,\n"
1309                 "     zero value disables the cache (default value: 128)\n"
1310                 "  -f CONFIG_FILE: Configuration file\n"
1311                 "  --config (port,queue,lcore): Rx queue configuration\n"
1312                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1313                 "                     bypassing the SP\n"
1314                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1315                 "                         devices to configure\n"
1316                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1317                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1318                 "                         (DEV_RX_OFFLOAD_*)\n"
1319                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1320                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1321                 "                         (DEV_TX_OFFLOAD_*)\n"
1322                 "  --" CMD_LINE_OPT_REASSEMBLE " NUM"
1323                 ": max number of entries in reassemble(fragment) table\n"
1324                 "    (zero (default value) disables reassembly)\n"
1325                 "  --" CMD_LINE_OPT_MTU " MTU"
1326                 ": MTU value on all ports (default value: 1500)\n"
1327                 "    outgoing packets with bigger size will be fragmented\n"
1328                 "    incoming packets with bigger size will be discarded\n"
1329                 "  --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1330                 ": fragments lifetime in nanoseconds, default\n"
1331                 "    and maximum value is 10.000.000.000 ns (10 s)\n"
1332                 "\n",
1333                 prgname);
1334 }
1335
1336 static int
1337 parse_mask(const char *str, uint64_t *val)
1338 {
1339         char *end;
1340         unsigned long t;
1341
1342         errno = 0;
1343         t = strtoul(str, &end, 0);
1344         if (errno != 0 || end[0] != 0)
1345                 return -EINVAL;
1346
1347         *val = t;
1348         return 0;
1349 }
1350
1351 static int32_t
1352 parse_portmask(const char *portmask)
1353 {
1354         char *end = NULL;
1355         unsigned long pm;
1356
1357         /* parse hexadecimal string */
1358         pm = strtoul(portmask, &end, 16);
1359         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1360                 return -1;
1361
1362         if ((pm == 0) && errno)
1363                 return -1;
1364
1365         return pm;
1366 }
1367
1368 static int64_t
1369 parse_decimal(const char *str)
1370 {
1371         char *end = NULL;
1372         uint64_t num;
1373
1374         num = strtoull(str, &end, 10);
1375         if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1376                 || num > INT64_MAX)
1377                 return -1;
1378
1379         return num;
1380 }
1381
1382 static int32_t
1383 parse_config(const char *q_arg)
1384 {
1385         char s[256];
1386         const char *p, *p0 = q_arg;
1387         char *end;
1388         enum fieldnames {
1389                 FLD_PORT = 0,
1390                 FLD_QUEUE,
1391                 FLD_LCORE,
1392                 _NUM_FLD
1393         };
1394         unsigned long int_fld[_NUM_FLD];
1395         char *str_fld[_NUM_FLD];
1396         int32_t i;
1397         uint32_t size;
1398
1399         nb_lcore_params = 0;
1400
1401         while ((p = strchr(p0, '(')) != NULL) {
1402                 ++p;
1403                 p0 = strchr(p, ')');
1404                 if (p0 == NULL)
1405                         return -1;
1406
1407                 size = p0 - p;
1408                 if (size >= sizeof(s))
1409                         return -1;
1410
1411                 snprintf(s, sizeof(s), "%.*s", size, p);
1412                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1413                                 _NUM_FLD)
1414                         return -1;
1415                 for (i = 0; i < _NUM_FLD; i++) {
1416                         errno = 0;
1417                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1418                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1419                                 return -1;
1420                 }
1421                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1422                         printf("exceeded max number of lcore params: %hu\n",
1423                                 nb_lcore_params);
1424                         return -1;
1425                 }
1426                 lcore_params_array[nb_lcore_params].port_id =
1427                         (uint8_t)int_fld[FLD_PORT];
1428                 lcore_params_array[nb_lcore_params].queue_id =
1429                         (uint8_t)int_fld[FLD_QUEUE];
1430                 lcore_params_array[nb_lcore_params].lcore_id =
1431                         (uint8_t)int_fld[FLD_LCORE];
1432                 ++nb_lcore_params;
1433         }
1434         lcore_params = lcore_params_array;
1435         return 0;
1436 }
1437
1438 static void
1439 print_app_sa_prm(const struct app_sa_prm *prm)
1440 {
1441         printf("librte_ipsec usage: %s\n",
1442                 (prm->enable == 0) ? "disabled" : "enabled");
1443
1444         printf("replay window size: %u\n", prm->window_size);
1445         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1446         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1447         printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1448 }
1449
1450 static int32_t
1451 parse_args(int32_t argc, char **argv)
1452 {
1453         int opt;
1454         int64_t ret;
1455         char **argvopt;
1456         int32_t option_index;
1457         char *prgname = argv[0];
1458         int32_t f_present = 0;
1459
1460         argvopt = argv;
1461
1462         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:",
1463                                 lgopts, &option_index)) != EOF) {
1464
1465                 switch (opt) {
1466                 case 'p':
1467                         enabled_port_mask = parse_portmask(optarg);
1468                         if (enabled_port_mask == 0) {
1469                                 printf("invalid portmask\n");
1470                                 print_usage(prgname);
1471                                 return -1;
1472                         }
1473                         break;
1474                 case 'P':
1475                         printf("Promiscuous mode selected\n");
1476                         promiscuous_on = 1;
1477                         break;
1478                 case 'u':
1479                         unprotected_port_mask = parse_portmask(optarg);
1480                         if (unprotected_port_mask == 0) {
1481                                 printf("invalid unprotected portmask\n");
1482                                 print_usage(prgname);
1483                                 return -1;
1484                         }
1485                         break;
1486                 case 'f':
1487                         if (f_present == 1) {
1488                                 printf("\"-f\" option present more than "
1489                                         "once!\n");
1490                                 print_usage(prgname);
1491                                 return -1;
1492                         }
1493                         cfgfile = optarg;
1494                         f_present = 1;
1495                         break;
1496                 case 'j':
1497                         ret = parse_decimal(optarg);
1498                         if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1499                                         ret > UINT16_MAX) {
1500                                 printf("Invalid frame buffer size value: %s\n",
1501                                         optarg);
1502                                 print_usage(prgname);
1503                                 return -1;
1504                         }
1505                         frame_buf_size = ret;
1506                         printf("Custom frame buffer size %u\n", frame_buf_size);
1507                         break;
1508                 case 'l':
1509                         app_sa_prm.enable = 1;
1510                         break;
1511                 case 'w':
1512                         app_sa_prm.window_size = parse_decimal(optarg);
1513                         break;
1514                 case 'e':
1515                         app_sa_prm.enable_esn = 1;
1516                         break;
1517                 case 'a':
1518                         app_sa_prm.enable = 1;
1519                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1520                         break;
1521                 case 'c':
1522                         ret = parse_decimal(optarg);
1523                         if (ret < 0) {
1524                                 printf("Invalid SA cache size: %s\n", optarg);
1525                                 print_usage(prgname);
1526                                 return -1;
1527                         }
1528                         app_sa_prm.cache_sz = ret;
1529                         break;
1530                 case CMD_LINE_OPT_CONFIG_NUM:
1531                         ret = parse_config(optarg);
1532                         if (ret) {
1533                                 printf("Invalid config\n");
1534                                 print_usage(prgname);
1535                                 return -1;
1536                         }
1537                         break;
1538                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1539                         ret = parse_decimal(optarg);
1540                         if (ret == -1 || ret > UINT32_MAX) {
1541                                 printf("Invalid argument[sa_idx]\n");
1542                                 print_usage(prgname);
1543                                 return -1;
1544                         }
1545
1546                         /* else */
1547                         single_sa = 1;
1548                         single_sa_idx = ret;
1549                         printf("Configured with single SA index %u\n",
1550                                         single_sa_idx);
1551                         break;
1552                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1553                         ret = parse_portmask(optarg);
1554                         if (ret == -1) {
1555                                 printf("Invalid argument[portmask]\n");
1556                                 print_usage(prgname);
1557                                 return -1;
1558                         }
1559
1560                         /* else */
1561                         enabled_cryptodev_mask = ret;
1562                         break;
1563                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1564                         ret = parse_mask(optarg, &dev_rx_offload);
1565                         if (ret != 0) {
1566                                 printf("Invalid argument for \'%s\': %s\n",
1567                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1568                                 print_usage(prgname);
1569                                 return -1;
1570                         }
1571                         break;
1572                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1573                         ret = parse_mask(optarg, &dev_tx_offload);
1574                         if (ret != 0) {
1575                                 printf("Invalid argument for \'%s\': %s\n",
1576                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1577                                 print_usage(prgname);
1578                                 return -1;
1579                         }
1580                         break;
1581                 case CMD_LINE_OPT_REASSEMBLE_NUM:
1582                         ret = parse_decimal(optarg);
1583                         if (ret < 0 || ret > UINT32_MAX) {
1584                                 printf("Invalid argument for \'%s\': %s\n",
1585                                         CMD_LINE_OPT_REASSEMBLE, optarg);
1586                                 print_usage(prgname);
1587                                 return -1;
1588                         }
1589                         frag_tbl_sz = ret;
1590                         break;
1591                 case CMD_LINE_OPT_MTU_NUM:
1592                         ret = parse_decimal(optarg);
1593                         if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1594                                 printf("Invalid argument for \'%s\': %s\n",
1595                                         CMD_LINE_OPT_MTU, optarg);
1596                                 print_usage(prgname);
1597                                 return -1;
1598                         }
1599                         mtu_size = ret;
1600                         break;
1601                 case CMD_LINE_OPT_FRAG_TTL_NUM:
1602                         ret = parse_decimal(optarg);
1603                         if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1604                                 printf("Invalid argument for \'%s\': %s\n",
1605                                         CMD_LINE_OPT_MTU, optarg);
1606                                 print_usage(prgname);
1607                                 return -1;
1608                         }
1609                         frag_ttl_ns = ret;
1610                         break;
1611                 default:
1612                         print_usage(prgname);
1613                         return -1;
1614                 }
1615         }
1616
1617         if (f_present == 0) {
1618                 printf("Mandatory option \"-f\" not present\n");
1619                 return -1;
1620         }
1621
1622         /* check do we need to enable multi-seg support */
1623         if (multi_seg_required()) {
1624                 /* legacy mode doesn't support multi-seg */
1625                 app_sa_prm.enable = 1;
1626                 printf("frame buf size: %u, mtu: %u, "
1627                         "number of reassemble entries: %u\n"
1628                         "multi-segment support is required\n",
1629                         frame_buf_size, mtu_size, frag_tbl_sz);
1630         }
1631
1632         print_app_sa_prm(&app_sa_prm);
1633
1634         if (optind >= 0)
1635                 argv[optind-1] = prgname;
1636
1637         ret = optind-1;
1638         optind = 1; /* reset getopt lib */
1639         return ret;
1640 }
1641
1642 static void
1643 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1644 {
1645         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1646         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1647         printf("%s%s", name, buf);
1648 }
1649
1650 /*
1651  * Update destination ethaddr for the port.
1652  */
1653 int
1654 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1655 {
1656         if (port >= RTE_DIM(ethaddr_tbl))
1657                 return -EINVAL;
1658
1659         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1660         return 0;
1661 }
1662
1663 /* Check the link status of all ports in up to 9s, and print them finally */
1664 static void
1665 check_all_ports_link_status(uint32_t port_mask)
1666 {
1667 #define CHECK_INTERVAL 100 /* 100ms */
1668 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1669         uint16_t portid;
1670         uint8_t count, all_ports_up, print_flag = 0;
1671         struct rte_eth_link link;
1672         int ret;
1673
1674         printf("\nChecking link status");
1675         fflush(stdout);
1676         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1677                 all_ports_up = 1;
1678                 RTE_ETH_FOREACH_DEV(portid) {
1679                         if ((port_mask & (1 << portid)) == 0)
1680                                 continue;
1681                         memset(&link, 0, sizeof(link));
1682                         ret = rte_eth_link_get_nowait(portid, &link);
1683                         if (ret < 0) {
1684                                 all_ports_up = 0;
1685                                 if (print_flag == 1)
1686                                         printf("Port %u link get failed: %s\n",
1687                                                 portid, rte_strerror(-ret));
1688                                 continue;
1689                         }
1690                         /* print link status if flag set */
1691                         if (print_flag == 1) {
1692                                 if (link.link_status)
1693                                         printf(
1694                                         "Port%d Link Up - speed %u Mbps -%s\n",
1695                                                 portid, link.link_speed,
1696                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1697                                         ("full-duplex") : ("half-duplex\n"));
1698                                 else
1699                                         printf("Port %d Link Down\n", portid);
1700                                 continue;
1701                         }
1702                         /* clear all_ports_up flag if any link down */
1703                         if (link.link_status == ETH_LINK_DOWN) {
1704                                 all_ports_up = 0;
1705                                 break;
1706                         }
1707                 }
1708                 /* after finally printing all link status, get out */
1709                 if (print_flag == 1)
1710                         break;
1711
1712                 if (all_ports_up == 0) {
1713                         printf(".");
1714                         fflush(stdout);
1715                         rte_delay_ms(CHECK_INTERVAL);
1716                 }
1717
1718                 /* set the print_flag if all ports up or timeout */
1719                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1720                         print_flag = 1;
1721                         printf("done\n");
1722                 }
1723         }
1724 }
1725
1726 static int32_t
1727 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1728                 uint16_t qp, struct lcore_params *params,
1729                 struct ipsec_ctx *ipsec_ctx,
1730                 const struct rte_cryptodev_capabilities *cipher,
1731                 const struct rte_cryptodev_capabilities *auth,
1732                 const struct rte_cryptodev_capabilities *aead)
1733 {
1734         int32_t ret = 0;
1735         unsigned long i;
1736         struct cdev_key key = { 0 };
1737
1738         key.lcore_id = params->lcore_id;
1739         if (cipher)
1740                 key.cipher_algo = cipher->sym.cipher.algo;
1741         if (auth)
1742                 key.auth_algo = auth->sym.auth.algo;
1743         if (aead)
1744                 key.aead_algo = aead->sym.aead.algo;
1745
1746         ret = rte_hash_lookup(map, &key);
1747         if (ret != -ENOENT)
1748                 return 0;
1749
1750         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1751                 if (ipsec_ctx->tbl[i].id == cdev_id)
1752                         break;
1753
1754         if (i == ipsec_ctx->nb_qps) {
1755                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1756                         printf("Maximum number of crypto devices assigned to "
1757                                 "a core, increase MAX_QP_PER_LCORE value\n");
1758                         return 0;
1759                 }
1760                 ipsec_ctx->tbl[i].id = cdev_id;
1761                 ipsec_ctx->tbl[i].qp = qp;
1762                 ipsec_ctx->nb_qps++;
1763                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1764                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1765                                 cdev_id, qp, i);
1766         }
1767
1768         ret = rte_hash_add_key_data(map, &key, (void *)i);
1769         if (ret < 0) {
1770                 printf("Faled to insert cdev mapping for (lcore %u, "
1771                                 "cdev %u, qp %u), errno %d\n",
1772                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1773                                 ipsec_ctx->tbl[i].qp, ret);
1774                 return 0;
1775         }
1776
1777         return 1;
1778 }
1779
1780 static int32_t
1781 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1782                 uint16_t qp, struct lcore_params *params)
1783 {
1784         int32_t ret = 0;
1785         const struct rte_cryptodev_capabilities *i, *j;
1786         struct rte_hash *map;
1787         struct lcore_conf *qconf;
1788         struct ipsec_ctx *ipsec_ctx;
1789         const char *str;
1790
1791         qconf = &lcore_conf[params->lcore_id];
1792
1793         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1794                 map = cdev_map_out;
1795                 ipsec_ctx = &qconf->outbound;
1796                 str = "Outbound";
1797         } else {
1798                 map = cdev_map_in;
1799                 ipsec_ctx = &qconf->inbound;
1800                 str = "Inbound";
1801         }
1802
1803         /* Required cryptodevs with operation chainning */
1804         if (!(dev_info->feature_flags &
1805                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1806                 return ret;
1807
1808         for (i = dev_info->capabilities;
1809                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1810                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1811                         continue;
1812
1813                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1814                         ret |= add_mapping(map, str, cdev_id, qp, params,
1815                                         ipsec_ctx, NULL, NULL, i);
1816                         continue;
1817                 }
1818
1819                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1820                         continue;
1821
1822                 for (j = dev_info->capabilities;
1823                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1824                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1825                                 continue;
1826
1827                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1828                                 continue;
1829
1830                         ret |= add_mapping(map, str, cdev_id, qp, params,
1831                                                 ipsec_ctx, i, j, NULL);
1832                 }
1833         }
1834
1835         return ret;
1836 }
1837
1838 /* Check if the device is enabled by cryptodev_mask */
1839 static int
1840 check_cryptodev_mask(uint8_t cdev_id)
1841 {
1842         if (enabled_cryptodev_mask & (1 << cdev_id))
1843                 return 0;
1844
1845         return -1;
1846 }
1847
1848 static int32_t
1849 cryptodevs_init(void)
1850 {
1851         struct rte_cryptodev_config dev_conf;
1852         struct rte_cryptodev_qp_conf qp_conf;
1853         uint16_t idx, max_nb_qps, qp, i;
1854         int16_t cdev_id;
1855         struct rte_hash_parameters params = { 0 };
1856
1857         const uint64_t mseg_flag = multi_seg_required() ?
1858                                 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
1859
1860         params.entries = CDEV_MAP_ENTRIES;
1861         params.key_len = sizeof(struct cdev_key);
1862         params.hash_func = rte_jhash;
1863         params.hash_func_init_val = 0;
1864         params.socket_id = rte_socket_id();
1865
1866         params.name = "cdev_map_in";
1867         cdev_map_in = rte_hash_create(&params);
1868         if (cdev_map_in == NULL)
1869                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1870                                 rte_errno);
1871
1872         params.name = "cdev_map_out";
1873         cdev_map_out = rte_hash_create(&params);
1874         if (cdev_map_out == NULL)
1875                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1876                                 rte_errno);
1877
1878         printf("lcore/cryptodev/qp mappings:\n");
1879
1880         idx = 0;
1881         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1882                 struct rte_cryptodev_info cdev_info;
1883
1884                 if (check_cryptodev_mask((uint8_t)cdev_id))
1885                         continue;
1886
1887                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1888
1889                 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
1890                         rte_exit(EXIT_FAILURE,
1891                                 "Device %hd does not support \'%s\' feature\n",
1892                                 cdev_id,
1893                                 rte_cryptodev_get_feature_name(mseg_flag));
1894
1895                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1896                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1897                 else
1898                         max_nb_qps = nb_lcore_params;
1899
1900                 qp = 0;
1901                 i = 0;
1902                 while (qp < max_nb_qps && i < nb_lcore_params) {
1903                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1904                                                 &lcore_params[idx]))
1905                                 qp++;
1906                         idx++;
1907                         idx = idx % nb_lcore_params;
1908                         i++;
1909                 }
1910
1911                 if (qp == 0)
1912                         continue;
1913
1914                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1915                 dev_conf.nb_queue_pairs = qp;
1916                 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
1917
1918                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1919                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1920                         rte_exit(EXIT_FAILURE,
1921                                 "Device does not support at least %u "
1922                                 "sessions", CDEV_MP_NB_OBJS);
1923
1924                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1925                         rte_panic("Failed to initialize cryptodev %u\n",
1926                                         cdev_id);
1927
1928                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1929                 qp_conf.mp_session =
1930                         socket_ctx[dev_conf.socket_id].session_pool;
1931                 qp_conf.mp_session_private =
1932                         socket_ctx[dev_conf.socket_id].session_priv_pool;
1933                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1934                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1935                                         &qp_conf, dev_conf.socket_id))
1936                                 rte_panic("Failed to setup queue %u for "
1937                                                 "cdev_id %u\n", 0, cdev_id);
1938
1939                 if (rte_cryptodev_start(cdev_id))
1940                         rte_panic("Failed to start cryptodev %u\n",
1941                                         cdev_id);
1942         }
1943
1944         printf("\n");
1945
1946         return 0;
1947 }
1948
1949 static void
1950 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
1951 {
1952         uint32_t frame_size;
1953         struct rte_eth_dev_info dev_info;
1954         struct rte_eth_txconf *txconf;
1955         uint16_t nb_tx_queue, nb_rx_queue;
1956         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1957         int32_t ret, socket_id;
1958         struct lcore_conf *qconf;
1959         struct rte_ether_addr ethaddr;
1960         struct rte_eth_conf local_port_conf = port_conf;
1961
1962         ret = rte_eth_dev_info_get(portid, &dev_info);
1963         if (ret != 0)
1964                 rte_exit(EXIT_FAILURE,
1965                         "Error during getting device (port %u) info: %s\n",
1966                         portid, strerror(-ret));
1967
1968         /* limit allowed HW offloafs, as user requested */
1969         dev_info.rx_offload_capa &= dev_rx_offload;
1970         dev_info.tx_offload_capa &= dev_tx_offload;
1971
1972         printf("Configuring device port %u:\n", portid);
1973
1974         ret = rte_eth_macaddr_get(portid, &ethaddr);
1975         if (ret != 0)
1976                 rte_exit(EXIT_FAILURE,
1977                         "Error getting MAC address (port %u): %s\n",
1978                         portid, rte_strerror(-ret));
1979
1980         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
1981         print_ethaddr("Address: ", &ethaddr);
1982         printf("\n");
1983
1984         nb_rx_queue = get_port_nb_rx_queues(portid);
1985         nb_tx_queue = nb_lcores;
1986
1987         if (nb_rx_queue > dev_info.max_rx_queues)
1988                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1989                                 "(max rx queue is %u)\n",
1990                                 nb_rx_queue, dev_info.max_rx_queues);
1991
1992         if (nb_tx_queue > dev_info.max_tx_queues)
1993                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1994                                 "(max tx queue is %u)\n",
1995                                 nb_tx_queue, dev_info.max_tx_queues);
1996
1997         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1998                         nb_rx_queue, nb_tx_queue);
1999
2000         frame_size = MTU_TO_FRAMELEN(mtu_size);
2001         if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
2002                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2003         local_port_conf.rxmode.max_rx_pkt_len = frame_size;
2004
2005         if (multi_seg_required()) {
2006                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
2007                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
2008         }
2009
2010         local_port_conf.rxmode.offloads |= req_rx_offloads;
2011         local_port_conf.txmode.offloads |= req_tx_offloads;
2012
2013         /* Check that all required capabilities are supported */
2014         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2015                         local_port_conf.rxmode.offloads)
2016                 rte_exit(EXIT_FAILURE,
2017                         "Error: port %u required RX offloads: 0x%" PRIx64
2018                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2019                         portid, local_port_conf.rxmode.offloads,
2020                         dev_info.rx_offload_capa);
2021
2022         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2023                         local_port_conf.txmode.offloads)
2024                 rte_exit(EXIT_FAILURE,
2025                         "Error: port %u required TX offloads: 0x%" PRIx64
2026                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2027                         portid, local_port_conf.txmode.offloads,
2028                         dev_info.tx_offload_capa);
2029
2030         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2031                 local_port_conf.txmode.offloads |=
2032                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2033
2034         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
2035                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2036
2037         printf("port %u configurng rx_offloads=0x%" PRIx64
2038                 ", tx_offloads=0x%" PRIx64 "\n",
2039                 portid, local_port_conf.rxmode.offloads,
2040                 local_port_conf.txmode.offloads);
2041
2042         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2043                 dev_info.flow_type_rss_offloads;
2044         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2045                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
2046                 printf("Port %u modified RSS hash function based on hardware support,"
2047                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2048                         portid,
2049                         port_conf.rx_adv_conf.rss_conf.rss_hf,
2050                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2051         }
2052
2053         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2054                         &local_port_conf);
2055         if (ret < 0)
2056                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2057                                 "err=%d, port=%d\n", ret, portid);
2058
2059         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2060         if (ret < 0)
2061                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2062                                 "err=%d, port=%d\n", ret, portid);
2063
2064         /* init one TX queue per lcore */
2065         tx_queueid = 0;
2066         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2067                 if (rte_lcore_is_enabled(lcore_id) == 0)
2068                         continue;
2069
2070                 if (numa_on)
2071                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2072                 else
2073                         socket_id = 0;
2074
2075                 /* init TX queue */
2076                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2077
2078                 txconf = &dev_info.default_txconf;
2079                 txconf->offloads = local_port_conf.txmode.offloads;
2080
2081                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2082                                 socket_id, txconf);
2083                 if (ret < 0)
2084                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2085                                         "err=%d, port=%d\n", ret, portid);
2086
2087                 qconf = &lcore_conf[lcore_id];
2088                 qconf->tx_queue_id[portid] = tx_queueid;
2089
2090                 /* Pre-populate pkt offloads based on capabilities */
2091                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2092                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2093                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2094                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2095
2096                 tx_queueid++;
2097
2098                 /* init RX queues */
2099                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2100                         struct rte_eth_rxconf rxq_conf;
2101
2102                         if (portid != qconf->rx_queue_list[queue].port_id)
2103                                 continue;
2104
2105                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
2106
2107                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2108                                         socket_id);
2109
2110                         rxq_conf = dev_info.default_rxconf;
2111                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
2112                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2113                                         nb_rxd, socket_id, &rxq_conf,
2114                                         socket_ctx[socket_id].mbuf_pool);
2115                         if (ret < 0)
2116                                 rte_exit(EXIT_FAILURE,
2117                                         "rte_eth_rx_queue_setup: err=%d, "
2118                                         "port=%d\n", ret, portid);
2119                 }
2120         }
2121         printf("\n");
2122 }
2123
2124 static size_t
2125 max_session_size(void)
2126 {
2127         size_t max_sz, sz;
2128         void *sec_ctx;
2129         int16_t cdev_id, port_id, n;
2130
2131         max_sz = 0;
2132         n =  rte_cryptodev_count();
2133         for (cdev_id = 0; cdev_id != n; cdev_id++) {
2134                 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2135                 if (sz > max_sz)
2136                         max_sz = sz;
2137                 /*
2138                  * If crypto device is security capable, need to check the
2139                  * size of security session as well.
2140                  */
2141
2142                 /* Get security context of the crypto device */
2143                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2144                 if (sec_ctx == NULL)
2145                         continue;
2146
2147                 /* Get size of security session */
2148                 sz = rte_security_session_get_size(sec_ctx);
2149                 if (sz > max_sz)
2150                         max_sz = sz;
2151         }
2152
2153         RTE_ETH_FOREACH_DEV(port_id) {
2154                 if ((enabled_port_mask & (1 << port_id)) == 0)
2155                         continue;
2156
2157                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2158                 if (sec_ctx == NULL)
2159                         continue;
2160
2161                 sz = rte_security_session_get_size(sec_ctx);
2162                 if (sz > max_sz)
2163                         max_sz = sz;
2164         }
2165
2166         return max_sz;
2167 }
2168
2169 static void
2170 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2171 {
2172         char mp_name[RTE_MEMPOOL_NAMESIZE];
2173         struct rte_mempool *sess_mp;
2174
2175         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2176                         "sess_mp_%u", socket_id);
2177         sess_mp = rte_cryptodev_sym_session_pool_create(
2178                         mp_name, CDEV_MP_NB_OBJS,
2179                         sess_sz, CDEV_MP_CACHE_SZ, 0,
2180                         socket_id);
2181         ctx->session_pool = sess_mp;
2182
2183         if (ctx->session_pool == NULL)
2184                 rte_exit(EXIT_FAILURE,
2185                         "Cannot init session pool on socket %d\n", socket_id);
2186         else
2187                 printf("Allocated session pool on socket %d\n", socket_id);
2188 }
2189
2190 static void
2191 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2192         size_t sess_sz)
2193 {
2194         char mp_name[RTE_MEMPOOL_NAMESIZE];
2195         struct rte_mempool *sess_mp;
2196
2197         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2198                         "sess_mp_priv_%u", socket_id);
2199         sess_mp = rte_mempool_create(mp_name,
2200                         CDEV_MP_NB_OBJS,
2201                         sess_sz,
2202                         CDEV_MP_CACHE_SZ,
2203                         0, NULL, NULL, NULL,
2204                         NULL, socket_id,
2205                         0);
2206         ctx->session_priv_pool = sess_mp;
2207
2208         if (ctx->session_priv_pool == NULL)
2209                 rte_exit(EXIT_FAILURE,
2210                         "Cannot init session priv pool on socket %d\n",
2211                         socket_id);
2212         else
2213                 printf("Allocated session priv pool on socket %d\n",
2214                         socket_id);
2215 }
2216
2217 static void
2218 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2219 {
2220         char s[64];
2221         int32_t ms;
2222
2223         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2224         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2225                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2226                         frame_buf_size, socket_id);
2227
2228         /*
2229          * if multi-segment support is enabled, then create a pool
2230          * for indirect mbufs.
2231          */
2232         ms = multi_seg_required();
2233         if (ms != 0) {
2234                 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2235                 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2236                         MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2237         }
2238
2239         if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2240                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2241                                 socket_id);
2242         else
2243                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2244 }
2245
2246 static inline int
2247 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2248 {
2249         struct ipsec_sa *sa;
2250
2251         /* For inline protocol processing, the metadata in the event will
2252          * uniquely identify the security session which raised the event.
2253          * Application would then need the userdata it had registered with the
2254          * security session to process the event.
2255          */
2256
2257         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2258
2259         if (sa == NULL) {
2260                 /* userdata could not be retrieved */
2261                 return -1;
2262         }
2263
2264         /* Sequence number over flow. SA need to be re-established */
2265         RTE_SET_USED(sa);
2266         return 0;
2267 }
2268
2269 static int
2270 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2271                  void *param, void *ret_param)
2272 {
2273         uint64_t md;
2274         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2275         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2276                                         rte_eth_dev_get_sec_ctx(port_id);
2277
2278         RTE_SET_USED(param);
2279
2280         if (type != RTE_ETH_EVENT_IPSEC)
2281                 return -1;
2282
2283         event_desc = ret_param;
2284         if (event_desc == NULL) {
2285                 printf("Event descriptor not set\n");
2286                 return -1;
2287         }
2288
2289         md = event_desc->metadata;
2290
2291         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2292                 return inline_ipsec_event_esn_overflow(ctx, md);
2293         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2294                 printf("Invalid IPsec event reported\n");
2295                 return -1;
2296         }
2297
2298         return -1;
2299 }
2300
2301 static uint16_t
2302 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2303         struct rte_mbuf *pkt[], uint16_t nb_pkts,
2304         __rte_unused uint16_t max_pkts, void *user_param)
2305 {
2306         uint64_t tm;
2307         uint32_t i, k;
2308         struct lcore_conf *lc;
2309         struct rte_mbuf *mb;
2310         struct rte_ether_hdr *eth;
2311
2312         lc = user_param;
2313         k = 0;
2314         tm = 0;
2315
2316         for (i = 0; i != nb_pkts; i++) {
2317
2318                 mb = pkt[i];
2319                 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2320                 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2321
2322                         struct rte_ipv4_hdr *iph;
2323
2324                         iph = (struct rte_ipv4_hdr *)(eth + 1);
2325                         if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2326
2327                                 mb->l2_len = sizeof(*eth);
2328                                 mb->l3_len = sizeof(*iph);
2329                                 tm = (tm != 0) ? tm : rte_rdtsc();
2330                                 mb = rte_ipv4_frag_reassemble_packet(
2331                                         lc->frag.tbl, &lc->frag.dr,
2332                                         mb, tm, iph);
2333
2334                                 if (mb != NULL) {
2335                                         /* fix ip cksum after reassemble. */
2336                                         iph = rte_pktmbuf_mtod_offset(mb,
2337                                                 struct rte_ipv4_hdr *,
2338                                                 mb->l2_len);
2339                                         iph->hdr_checksum = 0;
2340                                         iph->hdr_checksum = rte_ipv4_cksum(iph);
2341                                 }
2342                         }
2343                 } else if (eth->ether_type ==
2344                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2345
2346                         struct rte_ipv6_hdr *iph;
2347                         struct ipv6_extension_fragment *fh;
2348
2349                         iph = (struct rte_ipv6_hdr *)(eth + 1);
2350                         fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2351                         if (fh != NULL) {
2352                                 mb->l2_len = sizeof(*eth);
2353                                 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2354                                         sizeof(*fh);
2355                                 tm = (tm != 0) ? tm : rte_rdtsc();
2356                                 mb = rte_ipv6_frag_reassemble_packet(
2357                                         lc->frag.tbl, &lc->frag.dr,
2358                                         mb, tm, iph, fh);
2359                                 if (mb != NULL)
2360                                         /* fix l3_len after reassemble. */
2361                                         mb->l3_len = mb->l3_len - sizeof(*fh);
2362                         }
2363                 }
2364
2365                 pkt[k] = mb;
2366                 k += (mb != NULL);
2367         }
2368
2369         /* some fragments were encountered, drain death row */
2370         if (tm != 0)
2371                 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2372
2373         return k;
2374 }
2375
2376
2377 static int
2378 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2379 {
2380         int32_t sid;
2381         uint32_t i;
2382         uint64_t frag_cycles;
2383         const struct lcore_rx_queue *rxq;
2384         const struct rte_eth_rxtx_callback *cb;
2385
2386         /* create fragment table */
2387         sid = rte_lcore_to_socket_id(cid);
2388         frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2389                 NS_PER_S * frag_ttl_ns;
2390
2391         lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2392                 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2393         if (lc->frag.tbl == NULL) {
2394                 printf("%s(%u): failed to create fragment table of size: %u, "
2395                         "error code: %d\n",
2396                         __func__, cid, frag_tbl_sz, rte_errno);
2397                 return -ENOMEM;
2398         }
2399
2400         /* setup reassemble RX callbacks for all queues */
2401         for (i = 0; i != lc->nb_rx_queue; i++) {
2402
2403                 rxq = lc->rx_queue_list + i;
2404                 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2405                         rx_callback, lc);
2406                 if (cb == NULL) {
2407                         printf("%s(%u): failed to install RX callback for "
2408                                 "portid=%u, queueid=%u, error code: %d\n",
2409                                 __func__, cid,
2410                                 rxq->port_id, rxq->queue_id, rte_errno);
2411                         return -ENOMEM;
2412                 }
2413         }
2414
2415         return 0;
2416 }
2417
2418 static int
2419 reassemble_init(void)
2420 {
2421         int32_t rc;
2422         uint32_t i, lc;
2423
2424         rc = 0;
2425         for (i = 0; i != nb_lcore_params; i++) {
2426                 lc = lcore_params[i].lcore_id;
2427                 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2428                 if (rc != 0)
2429                         break;
2430         }
2431
2432         return rc;
2433 }
2434
2435 int32_t
2436 main(int32_t argc, char **argv)
2437 {
2438         int32_t ret;
2439         uint32_t lcore_id;
2440         uint32_t i;
2441         uint8_t socket_id;
2442         uint16_t portid;
2443         uint64_t req_rx_offloads, req_tx_offloads;
2444         size_t sess_sz;
2445
2446         /* init EAL */
2447         ret = rte_eal_init(argc, argv);
2448         if (ret < 0)
2449                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2450         argc -= ret;
2451         argv += ret;
2452
2453         /* parse application arguments (after the EAL ones) */
2454         ret = parse_args(argc, argv);
2455         if (ret < 0)
2456                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2457
2458         /* parse configuration file */
2459         if (parse_cfg_file(cfgfile) < 0) {
2460                 printf("parsing file \"%s\" failed\n",
2461                         optarg);
2462                 print_usage(argv[0]);
2463                 return -1;
2464         }
2465
2466         if ((unprotected_port_mask & enabled_port_mask) !=
2467                         unprotected_port_mask)
2468                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2469                                 unprotected_port_mask);
2470
2471         if (check_params() < 0)
2472                 rte_exit(EXIT_FAILURE, "check_params failed\n");
2473
2474         ret = init_lcore_rx_queues();
2475         if (ret < 0)
2476                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2477
2478         nb_lcores = rte_lcore_count();
2479
2480         sess_sz = max_session_size();
2481
2482         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2483                 if (rte_lcore_is_enabled(lcore_id) == 0)
2484                         continue;
2485
2486                 if (numa_on)
2487                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2488                 else
2489                         socket_id = 0;
2490
2491                 /* mbuf_pool is initialised by the pool_init() function*/
2492                 if (socket_ctx[socket_id].mbuf_pool)
2493                         continue;
2494
2495                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2496                 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2497                 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2498                         sess_sz);
2499         }
2500
2501         RTE_ETH_FOREACH_DEV(portid) {
2502                 if ((enabled_port_mask & (1 << portid)) == 0)
2503                         continue;
2504
2505                 sa_check_offloads(portid, &req_rx_offloads, &req_tx_offloads);
2506                 port_init(portid, req_rx_offloads, req_tx_offloads);
2507         }
2508
2509         cryptodevs_init();
2510
2511         /* start ports */
2512         RTE_ETH_FOREACH_DEV(portid) {
2513                 if ((enabled_port_mask & (1 << portid)) == 0)
2514                         continue;
2515
2516                 /*
2517                  * Start device
2518                  * note: device must be started before a flow rule
2519                  * can be installed.
2520                  */
2521                 ret = rte_eth_dev_start(portid);
2522                 if (ret < 0)
2523                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2524                                         "err=%d, port=%d\n", ret, portid);
2525                 /*
2526                  * If enabled, put device in promiscuous mode.
2527                  * This allows IO forwarding mode to forward packets
2528                  * to itself through 2 cross-connected  ports of the
2529                  * target machine.
2530                  */
2531                 if (promiscuous_on) {
2532                         ret = rte_eth_promiscuous_enable(portid);
2533                         if (ret != 0)
2534                                 rte_exit(EXIT_FAILURE,
2535                                         "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2536                                         rte_strerror(-ret), portid);
2537                 }
2538
2539                 rte_eth_dev_callback_register(portid,
2540                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2541         }
2542
2543         /* fragment reassemble is enabled */
2544         if (frag_tbl_sz != 0) {
2545                 ret = reassemble_init();
2546                 if (ret != 0)
2547                         rte_exit(EXIT_FAILURE, "failed at reassemble init");
2548         }
2549
2550         /* Replicate each context per socket */
2551         for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2552                 socket_id = rte_socket_id_by_idx(i);
2553                 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2554                         (socket_ctx[socket_id].sa_in == NULL) &&
2555                         (socket_ctx[socket_id].sa_out == NULL)) {
2556                         sa_init(&socket_ctx[socket_id], socket_id);
2557                         sp4_init(&socket_ctx[socket_id], socket_id);
2558                         sp6_init(&socket_ctx[socket_id], socket_id);
2559                         rt_init(&socket_ctx[socket_id], socket_id);
2560                 }
2561         }
2562
2563         check_all_ports_link_status(enabled_port_mask);
2564
2565         /* launch per-lcore init on every lcore */
2566         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2567         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2568                 if (rte_eal_wait_lcore(lcore_id) < 0)
2569                         return -1;
2570         }
2571
2572         return 0;
2573 }