examples/ipsec-secgw/ipsec-secgw.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <netinet/in.h>
  11 #include <netinet/ip.h>
  12 #include <netinet/ip6.h>
  13 #include <string.h>
  14 #include <sys/queue.h>
  15 #include <stdarg.h>
  16 #include <errno.h>
  17 #include <getopt.h>
  18
  19 #include <rte_common.h>
  20 #include <rte_byteorder.h>
  21 #include <rte_log.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_mempool.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_acl.h>
  38 #include <rte_lpm.h>
  39 #include <rte_lpm6.h>
  40 #include <rte_hash.h>
  41 #include <rte_jhash.h>
  42 #include <rte_cryptodev.h>
  43 #include <rte_security.h>
  44 #include <rte_ip.h>
  45 #include <rte_ip_frag.h>
  46
  47 #include "ipsec.h"
  48 #include "parser.h"
  49
  50 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  51
  52 #define MAX_JUMBO_PKT_LEN  9600
  53
  54 #define MEMPOOL_CACHE_SIZE 256
  55
  56 #define NB_MBUF (32000)
  57
  58 #define CDEV_QUEUE_DESC 2048
  59 #define CDEV_MAP_ENTRIES 16384
  60 #define CDEV_MP_NB_OBJS 1024
  61 #define CDEV_MP_CACHE_SZ 64
  62 #define MAX_QUEUE_PAIRS 1
  63
  64 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  65
  66 #define NB_SOCKETS 4
  67
  68 /* Configure how many packets ahead to prefetch, when reading packets */
  69 #define PREFETCH_OFFSET 3
  70
  71 #define MAX_RX_QUEUE_PER_LCORE 16
  72
  73 #define MAX_LCORE_PARAMS 1024
  74
  75 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
  76
  77 /*
  78  * Configurable number of RX/TX ring descriptors
  79  */
  80 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  81 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  82 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  83 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  84
  85 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
  86 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  87         (((uint64_t)((a) & 0xff) << 56) | \
  88         ((uint64_t)((b) & 0xff) << 48) | \
  89         ((uint64_t)((c) & 0xff) << 40) | \
  90         ((uint64_t)((d) & 0xff) << 32) | \
  91         ((uint64_t)((e) & 0xff) << 24) | \
  92         ((uint64_t)((f) & 0xff) << 16) | \
  93         ((uint64_t)((g) & 0xff) << 8)  | \
  94         ((uint64_t)(h) & 0xff))
  95 #else
  96 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  97         (((uint64_t)((h) & 0xff) << 56) | \
  98         ((uint64_t)((g) & 0xff) << 48) | \
  99         ((uint64_t)((f) & 0xff) << 40) | \
 100         ((uint64_t)((e) & 0xff) << 32) | \
 101         ((uint64_t)((d) & 0xff) << 24) | \
 102         ((uint64_t)((c) & 0xff) << 16) | \
 103         ((uint64_t)((b) & 0xff) << 8) | \
 104         ((uint64_t)(a) & 0xff))
 105 #endif
 106 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 107
 108 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 109                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
 110                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
 111                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
 112                 0, 0)
 113
 114 #define FRAG_TBL_BUCKET_ENTRIES 4
 115 #define FRAG_TTL_MS             (10 * MS_PER_S)
 116
 117 #define MTU_TO_FRAMELEN(x)      ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
 118
 119 /* port/source ethernet addr and destination ethernet addr */
 120 struct ethaddr_info {
 121         uint64_t src, dst;
 122 };
 123
 124 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 125         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 126         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 127         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 128         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 129 };
 130
 131 #define CMD_LINE_OPT_CONFIG             "config"
 132 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 133 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 134 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
 135 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
 136 #define CMD_LINE_OPT_REASSEMBLE         "reassemble"
 137 #define CMD_LINE_OPT_MTU                "mtu"
 138
 139 enum {
 140         /* long options mapped to a short option */
 141
 142         /* first long only option value must be >= 256, so that we won't
 143          * conflict with short options
 144          */
 145         CMD_LINE_OPT_MIN_NUM = 256,
 146         CMD_LINE_OPT_CONFIG_NUM,
 147         CMD_LINE_OPT_SINGLE_SA_NUM,
 148         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 149         CMD_LINE_OPT_RX_OFFLOAD_NUM,
 150         CMD_LINE_OPT_TX_OFFLOAD_NUM,
 151         CMD_LINE_OPT_REASSEMBLE_NUM,
 152         CMD_LINE_OPT_MTU_NUM,
 153 };
 154
 155 static const struct option lgopts[] = {
 156         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 157         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 158         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 159         {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
 160         {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
 161         {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
 162         {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
 163         {NULL, 0, 0, 0}
 164 };
 165
 166 /* mask of enabled ports */
 167 static uint32_t enabled_port_mask;
 168 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 169 static uint32_t unprotected_port_mask;
 170 static int32_t promiscuous_on = 1;
 171 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 172 static uint32_t nb_lcores;
 173 static uint32_t single_sa;
 174 static uint32_t single_sa_idx;
 175
 176 /*
 177  * RX/TX HW offload capabilities to enable/use on ethernet ports.
 178  * By default all capabilities are enabled.
 179  */
 180 static uint64_t dev_rx_offload = UINT64_MAX;
 181 static uint64_t dev_tx_offload = UINT64_MAX;
 182
 183 /*
 184  * global values that determine multi-seg policy
 185  */
 186 static uint32_t frag_tbl_sz;
 187 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
 188 static uint32_t mtu_size = RTE_ETHER_MTU;
 189
 190 /* application wide librte_ipsec/SA parameters */
 191 struct app_sa_prm app_sa_prm = {.enable = 0};
 192
 193 struct lcore_rx_queue {
 194         uint16_t port_id;
 195         uint8_t queue_id;
 196 } __rte_cache_aligned;
 197
 198 struct lcore_params {
 199         uint16_t port_id;
 200         uint8_t queue_id;
 201         uint8_t lcore_id;
 202 } __rte_cache_aligned;
 203
 204 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 205
 206 static struct lcore_params *lcore_params;
 207 static uint16_t nb_lcore_params;
 208
 209 static struct rte_hash *cdev_map_in;
 210 static struct rte_hash *cdev_map_out;
 211
 212 struct buffer {
 213         uint16_t len;
 214         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 215 };
 216
 217 struct lcore_conf {
 218         uint16_t nb_rx_queue;
 219         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 220         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 221         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 222         struct ipsec_ctx inbound;
 223         struct ipsec_ctx outbound;
 224         struct rt_ctx *rt4_ctx;
 225         struct rt_ctx *rt6_ctx;
 226         struct {
 227                 struct rte_ip_frag_tbl *tbl;
 228                 struct rte_mempool *pool_dir;
 229                 struct rte_mempool *pool_indir;
 230                 struct rte_ip_frag_death_row dr;
 231         } frag;
 232 } __rte_cache_aligned;
 233
 234 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 235
 236 static struct rte_eth_conf port_conf = {
 237         .rxmode = {
 238                 .mq_mode        = ETH_MQ_RX_RSS,
 239                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 240                 .split_hdr_size = 0,
 241                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 242         },
 243         .rx_adv_conf = {
 244                 .rss_conf = {
 245                         .rss_key = NULL,
 246                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 247                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 248                 },
 249         },
 250         .txmode = {
 251                 .mq_mode = ETH_MQ_TX_NONE,
 252         },
 253 };
 254
 255 static struct socket_ctx socket_ctx[NB_SOCKETS];
 256
 257 /*
 258  * Determine is multi-segment support required:
 259  *  - either frame buffer size is smaller then mtu
 260  *  - or reassmeble support is requested
 261  */
 262 static int
 263 multi_seg_required(void)
 264 {
 265         return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
 266                 frame_buf_size || frag_tbl_sz != 0);
 267 }
 268
 269 static inline void
 270 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
 271         uint32_t l2_len)
 272 {
 273         uint32_t plen, trim;
 274
 275         plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
 276         if (plen < m->pkt_len) {
 277                 trim = m->pkt_len - plen;
 278                 rte_pktmbuf_trim(m, trim);
 279         }
 280 }
 281
 282 static inline void
 283 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
 284         uint32_t l2_len)
 285 {
 286         uint32_t plen, trim;
 287
 288         plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
 289         if (plen < m->pkt_len) {
 290                 trim = m->pkt_len - plen;
 291                 rte_pktmbuf_trim(m, trim);
 292         }
 293 }
 294
 295 static inline void
 296 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 297 {
 298         const struct rte_ether_hdr *eth;
 299         const struct rte_ipv4_hdr *iph4;
 300         const struct rte_ipv6_hdr *iph6;
 301
 302         eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
 303         if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 304
 305                 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
 306                         RTE_ETHER_HDR_LEN);
 307                 adjust_ipv4_pktlen(pkt, iph4, 0);
 308
 309                 if (iph4->next_proto_id == IPPROTO_ESP)
 310                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 311                 else {
 312                         t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
 313                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 314                 }
 315                 pkt->l2_len = 0;
 316                 pkt->l3_len = sizeof(*iph4);
 317         } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 318                 int next_proto;
 319                 size_t l3len, ext_len;
 320                 uint8_t *p;
 321
 322                 /* get protocol type */
 323                 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
 324                         RTE_ETHER_HDR_LEN);
 325                 adjust_ipv6_pktlen(pkt, iph6, 0);
 326
 327                 next_proto = iph6->proto;
 328
 329                 /* determine l3 header size up to ESP extension */
 330                 l3len = sizeof(struct ip6_hdr);
 331                 p = rte_pktmbuf_mtod(pkt, uint8_t *);
 332                 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
 333                         (next_proto = rte_ipv6_get_next_ext(p + l3len,
 334                                                 next_proto, &ext_len)) >= 0)
 335                         l3len += ext_len;
 336
 337                 /* drop packet when IPv6 header exceeds first segment length */
 338                 if (unlikely(l3len > pkt->data_len)) {
 339                         rte_pktmbuf_free(pkt);
 340                         return;
 341                 }
 342
 343                 if (next_proto == IPPROTO_ESP)
 344                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 345                 else {
 346                         t->ip6.data[t->ip6.num] = &iph6->proto;
 347                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 348                 }
 349                 pkt->l2_len = 0;
 350                 pkt->l3_len = l3len;
 351         } else {
 352                 /* Unknown/Unsupported type, drop the packet */
 353                 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
 354                         rte_be_to_cpu_16(eth->ether_type));
 355                 rte_pktmbuf_free(pkt);
 356         }
 357
 358         /* Check if the packet has been processed inline. For inline protocol
 359          * processed packets, the metadata in the mbuf can be used to identify
 360          * the security processing done on the packet. The metadata will be
 361          * used to retrieve the application registered userdata associated
 362          * with the security session.
 363          */
 364
 365         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 366                 struct ipsec_sa *sa;
 367                 struct ipsec_mbuf_metadata *priv;
 368                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 369                                                 rte_eth_dev_get_sec_ctx(
 370                                                 pkt->port);
 371
 372                 /* Retrieve the userdata registered. Here, the userdata
 373                  * registered is the SA pointer.
 374                  */
 375
 376                 sa = (struct ipsec_sa *)
 377                                 rte_security_get_userdata(ctx, pkt->udata64);
 378
 379                 if (sa == NULL) {
 380                         /* userdata could not be retrieved */
 381                         return;
 382                 }
 383
 384                 /* Save SA as priv member in mbuf. This will be used in the
 385                  * IPsec selector(SP-SA) check.
 386                  */
 387
 388                 priv = get_priv(pkt);
 389                 priv->sa = sa;
 390         }
 391 }
 392
 393 static inline void
 394 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 395                 uint16_t nb_pkts)
 396 {
 397         int32_t i;
 398
 399         t->ipsec.num = 0;
 400         t->ip4.num = 0;
 401         t->ip6.num = 0;
 402
 403         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 404                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 405                                         void *));
 406                 prepare_one_packet(pkts[i], t);
 407         }
 408         /* Process left packets */
 409         for (; i < nb_pkts; i++)
 410                 prepare_one_packet(pkts[i], t);
 411 }
 412
 413 static inline void
 414 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 415                 const struct lcore_conf *qconf)
 416 {
 417         struct ip *ip;
 418         struct rte_ether_hdr *ethhdr;
 419
 420         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 421
 422         ethhdr = (struct rte_ether_hdr *)
 423                 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
 424
 425         if (ip->ip_v == IPVERSION) {
 426                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 427                 pkt->l3_len = sizeof(struct ip);
 428                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 429
 430                 ip->ip_sum = 0;
 431
 432                 /* calculate IPv4 cksum in SW */
 433                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 434                         ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
 435
 436                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
 437         } else {
 438                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 439                 pkt->l3_len = sizeof(struct ip6_hdr);
 440                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 441
 442                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
 443         }
 444
 445         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 446                         sizeof(struct rte_ether_addr));
 447         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 448                         sizeof(struct rte_ether_addr));
 449 }
 450
 451 static inline void
 452 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 453                 const struct lcore_conf *qconf)
 454 {
 455         int32_t i;
 456         const int32_t prefetch_offset = 2;
 457
 458         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 459                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 460                 prepare_tx_pkt(pkts[i], port, qconf);
 461         }
 462         /* Process left packets */
 463         for (; i < nb_pkts; i++)
 464                 prepare_tx_pkt(pkts[i], port, qconf);
 465 }
 466
 467 /* Send burst of packets on an output interface */
 468 static inline int32_t
 469 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 470 {
 471         struct rte_mbuf **m_table;
 472         int32_t ret;
 473         uint16_t queueid;
 474
 475         queueid = qconf->tx_queue_id[port];
 476         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 477
 478         prepare_tx_burst(m_table, n, port, qconf);
 479
 480         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 481         if (unlikely(ret < n)) {
 482                 do {
 483                         rte_pktmbuf_free(m_table[ret]);
 484                 } while (++ret < n);
 485         }
 486
 487         return 0;
 488 }
 489
 490 /*
 491  * Helper function to fragment and queue for TX one packet.
 492  */
 493 static inline uint32_t
 494 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
 495         uint16_t port, uint8_t proto)
 496 {
 497         struct buffer *tbl;
 498         uint32_t len, n;
 499         int32_t rc;
 500
 501         tbl =  qconf->tx_mbufs + port;
 502         len = tbl->len;
 503
 504         /* free space for new fragments */
 505         if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >=  RTE_DIM(tbl->m_table)) {
 506                 send_burst(qconf, len, port);
 507                 len = 0;
 508         }
 509
 510         n = RTE_DIM(tbl->m_table) - len;
 511
 512         if (proto == IPPROTO_IP)
 513                 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
 514                         n, mtu_size, qconf->frag.pool_dir,
 515                         qconf->frag.pool_indir);
 516         else
 517                 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
 518                         n, mtu_size, qconf->frag.pool_dir,
 519                         qconf->frag.pool_indir);
 520
 521         if (rc >= 0)
 522                 len += rc;
 523         else
 524                 RTE_LOG(ERR, IPSEC,
 525                         "%s: failed to fragment packet with size %u, "
 526                         "error code: %d\n",
 527                         __func__, m->pkt_len, rte_errno);
 528
 529         rte_pktmbuf_free(m);
 530         return len;
 531 }
 532
 533 /* Enqueue a single packet, and send burst if queue is filled */
 534 static inline int32_t
 535 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
 536 {
 537         uint32_t lcore_id;
 538         uint16_t len;
 539         struct lcore_conf *qconf;
 540
 541         lcore_id = rte_lcore_id();
 542
 543         qconf = &lcore_conf[lcore_id];
 544         len = qconf->tx_mbufs[port].len;
 545
 546         if (m->pkt_len <= mtu_size) {
 547                 qconf->tx_mbufs[port].m_table[len] = m;
 548                 len++;
 549
 550         /* need to fragment the packet */
 551         } else
 552                 len = send_fragment_packet(qconf, m, port, proto);
 553
 554         /* enough pkts to be sent */
 555         if (unlikely(len == MAX_PKT_BURST)) {
 556                 send_burst(qconf, MAX_PKT_BURST, port);
 557                 len = 0;
 558         }
 559
 560         qconf->tx_mbufs[port].len = len;
 561         return 0;
 562 }
 563
 564 static inline void
 565 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 566                 uint16_t lim)
 567 {
 568         struct rte_mbuf *m;
 569         uint32_t i, j, res, sa_idx;
 570
 571         if (ip->num == 0 || sp == NULL)
 572                 return;
 573
 574         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 575                         ip->num, DEFAULT_MAX_CATEGORIES);
 576
 577         j = 0;
 578         for (i = 0; i < ip->num; i++) {
 579                 m = ip->pkts[i];
 580                 res = ip->res[i];
 581                 if (res == BYPASS) {
 582                         ip->pkts[j++] = m;
 583                         continue;
 584                 }
 585                 if (res == DISCARD) {
 586                         rte_pktmbuf_free(m);
 587                         continue;
 588                 }
 589
 590                 /* Only check SPI match for processed IPSec packets */
 591                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 592                         rte_pktmbuf_free(m);
 593                         continue;
 594                 }
 595
 596                 sa_idx = SPI2IDX(res);
 597                 if (!inbound_sa_check(sa, m, sa_idx)) {
 598                         rte_pktmbuf_free(m);
 599                         continue;
 600                 }
 601                 ip->pkts[j++] = m;
 602         }
 603         ip->num = j;
 604 }
 605
 606 static void
 607 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 608 {
 609         uint32_t i, n4, n6;
 610         struct ip *ip;
 611         struct rte_mbuf *m;
 612
 613         n4 = trf->ip4.num;
 614         n6 = trf->ip6.num;
 615
 616         for (i = 0; i < num; i++) {
 617
 618                 m = mb[i];
 619                 ip = rte_pktmbuf_mtod(m, struct ip *);
 620
 621                 if (ip->ip_v == IPVERSION) {
 622                         trf->ip4.pkts[n4] = m;
 623                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 624                                         uint8_t *, offsetof(struct ip, ip_p));
 625                         n4++;
 626                 } else if (ip->ip_v == IP6_VERSION) {
 627                         trf->ip6.pkts[n6] = m;
 628                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 629                                         uint8_t *,
 630                                         offsetof(struct ip6_hdr, ip6_nxt));
 631                         n6++;
 632                 } else
 633                         rte_pktmbuf_free(m);
 634         }
 635
 636         trf->ip4.num = n4;
 637         trf->ip6.num = n6;
 638 }
 639
 640
 641 static inline void
 642 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 643                 struct ipsec_traffic *traffic)
 644 {
 645         uint16_t nb_pkts_in, n_ip4, n_ip6;
 646
 647         n_ip4 = traffic->ip4.num;
 648         n_ip6 = traffic->ip6.num;
 649
 650         if (app_sa_prm.enable == 0) {
 651                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 652                                 traffic->ipsec.num, MAX_PKT_BURST);
 653                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 654         } else {
 655                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 656                         traffic->ipsec.saptr, traffic->ipsec.num);
 657                 ipsec_process(ipsec_ctx, traffic);
 658         }
 659
 660         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 661                         n_ip4);
 662
 663         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 664                         n_ip6);
 665 }
 666
 667 static inline void
 668 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 669                 struct traffic_type *ipsec)
 670 {
 671         struct rte_mbuf *m;
 672         uint32_t i, j, sa_idx;
 673
 674         if (ip->num == 0 || sp == NULL)
 675                 return;
 676
 677         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 678                         ip->num, DEFAULT_MAX_CATEGORIES);
 679
 680         j = 0;
 681         for (i = 0; i < ip->num; i++) {
 682                 m = ip->pkts[i];
 683                 sa_idx = SPI2IDX(ip->res[i]);
 684                 if (ip->res[i] == DISCARD)
 685                         rte_pktmbuf_free(m);
 686                 else if (ip->res[i] == BYPASS)
 687                         ip->pkts[j++] = m;
 688                 else {
 689                         ipsec->res[ipsec->num] = sa_idx;
 690                         ipsec->pkts[ipsec->num++] = m;
 691                 }
 692         }
 693         ip->num = j;
 694 }
 695
 696 static inline void
 697 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 698                 struct ipsec_traffic *traffic)
 699 {
 700         struct rte_mbuf *m;
 701         uint16_t idx, nb_pkts_out, i;
 702
 703         /* Drop any IPsec traffic from protected ports */
 704         for (i = 0; i < traffic->ipsec.num; i++)
 705                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 706
 707         traffic->ipsec.num = 0;
 708
 709         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 710
 711         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 712
 713         if (app_sa_prm.enable == 0) {
 714
 715                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 716                                 traffic->ipsec.res, traffic->ipsec.num,
 717                                 MAX_PKT_BURST);
 718
 719                 for (i = 0; i < nb_pkts_out; i++) {
 720                         m = traffic->ipsec.pkts[i];
 721                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 722                         if (ip->ip_v == IPVERSION) {
 723                                 idx = traffic->ip4.num++;
 724                                 traffic->ip4.pkts[idx] = m;
 725                         } else {
 726                                 idx = traffic->ip6.num++;
 727                                 traffic->ip6.pkts[idx] = m;
 728                         }
 729                 }
 730         } else {
 731                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 732                         traffic->ipsec.saptr, traffic->ipsec.num);
 733                 ipsec_process(ipsec_ctx, traffic);
 734         }
 735 }
 736
 737 static inline void
 738 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 739                 struct ipsec_traffic *traffic)
 740 {
 741         struct rte_mbuf *m;
 742         uint32_t nb_pkts_in, i, idx;
 743
 744         /* Drop any IPv4 traffic from unprotected ports */
 745         for (i = 0; i < traffic->ip4.num; i++)
 746                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 747
 748         traffic->ip4.num = 0;
 749
 750         /* Drop any IPv6 traffic from unprotected ports */
 751         for (i = 0; i < traffic->ip6.num; i++)
 752                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 753
 754         traffic->ip6.num = 0;
 755
 756         if (app_sa_prm.enable == 0) {
 757
 758                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 759                                 traffic->ipsec.num, MAX_PKT_BURST);
 760
 761                 for (i = 0; i < nb_pkts_in; i++) {
 762                         m = traffic->ipsec.pkts[i];
 763                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 764                         if (ip->ip_v == IPVERSION) {
 765                                 idx = traffic->ip4.num++;
 766                                 traffic->ip4.pkts[idx] = m;
 767                         } else {
 768                                 idx = traffic->ip6.num++;
 769                                 traffic->ip6.pkts[idx] = m;
 770                         }
 771                 }
 772         } else {
 773                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 774                         traffic->ipsec.saptr, traffic->ipsec.num);
 775                 ipsec_process(ipsec_ctx, traffic);
 776         }
 777 }
 778
 779 static inline void
 780 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 781                 struct ipsec_traffic *traffic)
 782 {
 783         struct rte_mbuf *m;
 784         uint32_t nb_pkts_out, i, n;
 785         struct ip *ip;
 786
 787         /* Drop any IPsec traffic from protected ports */
 788         for (i = 0; i < traffic->ipsec.num; i++)
 789                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 790
 791         n = 0;
 792
 793         for (i = 0; i < traffic->ip4.num; i++) {
 794                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 795                 traffic->ipsec.res[n++] = single_sa_idx;
 796         }
 797
 798         for (i = 0; i < traffic->ip6.num; i++) {
 799                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 800                 traffic->ipsec.res[n++] = single_sa_idx;
 801         }
 802
 803         traffic->ip4.num = 0;
 804         traffic->ip6.num = 0;
 805         traffic->ipsec.num = n;
 806
 807         if (app_sa_prm.enable == 0) {
 808
 809                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 810                                 traffic->ipsec.res, traffic->ipsec.num,
 811                                 MAX_PKT_BURST);
 812
 813                 /* They all sue the same SA (ip4 or ip6 tunnel) */
 814                 m = traffic->ipsec.pkts[0];
 815                 ip = rte_pktmbuf_mtod(m, struct ip *);
 816                 if (ip->ip_v == IPVERSION) {
 817                         traffic->ip4.num = nb_pkts_out;
 818                         for (i = 0; i < nb_pkts_out; i++)
 819                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 820                 } else {
 821                         traffic->ip6.num = nb_pkts_out;
 822                         for (i = 0; i < nb_pkts_out; i++)
 823                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 824                 }
 825         } else {
 826                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 827                         traffic->ipsec.saptr, traffic->ipsec.num);
 828                 ipsec_process(ipsec_ctx, traffic);
 829         }
 830 }
 831
 832 static inline int32_t
 833 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 834 {
 835         struct ipsec_mbuf_metadata *priv;
 836         struct ipsec_sa *sa;
 837
 838         priv = get_priv(pkt);
 839
 840         sa = priv->sa;
 841         if (unlikely(sa == NULL)) {
 842                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 843                 goto fail;
 844         }
 845
 846         if (is_ipv6)
 847                 return sa->portid;
 848
 849         /* else */
 850         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 851
 852 fail:
 853         if (is_ipv6)
 854                 return -1;
 855
 856         /* else */
 857         return 0;
 858 }
 859
 860 static inline void
 861 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 862 {
 863         uint32_t hop[MAX_PKT_BURST * 2];
 864         uint32_t dst_ip[MAX_PKT_BURST * 2];
 865         int32_t pkt_hop = 0;
 866         uint16_t i, offset;
 867         uint16_t lpm_pkts = 0;
 868
 869         if (nb_pkts == 0)
 870                 return;
 871
 872         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 873          * have port ID in the SA
 874          */
 875
 876         for (i = 0; i < nb_pkts; i++) {
 877                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 878                         /* Security offload not enabled. So an LPM lookup is
 879                          * required to get the hop
 880                          */
 881                         offset = offsetof(struct ip, ip_dst);
 882                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 883                                         uint32_t *, offset);
 884                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 885                         lpm_pkts++;
 886                 }
 887         }
 888
 889         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 890
 891         lpm_pkts = 0;
 892
 893         for (i = 0; i < nb_pkts; i++) {
 894                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 895                         /* Read hop from the SA */
 896                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 897                 } else {
 898                         /* Need to use hop returned by lookup */
 899                         pkt_hop = hop[lpm_pkts++];
 900                 }
 901
 902                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 903                         rte_pktmbuf_free(pkts[i]);
 904                         continue;
 905                 }
 906                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
 907         }
 908 }
 909
 910 static inline void
 911 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 912 {
 913         int32_t hop[MAX_PKT_BURST * 2];
 914         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 915         uint8_t *ip6_dst;
 916         int32_t pkt_hop = 0;
 917         uint16_t i, offset;
 918         uint16_t lpm_pkts = 0;
 919
 920         if (nb_pkts == 0)
 921                 return;
 922
 923         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 924          * have port ID in the SA
 925          */
 926
 927         for (i = 0; i < nb_pkts; i++) {
 928                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 929                         /* Security offload not enabled. So an LPM lookup is
 930                          * required to get the hop
 931                          */
 932                         offset = offsetof(struct ip6_hdr, ip6_dst);
 933                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 934                                         offset);
 935                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 936                         lpm_pkts++;
 937                 }
 938         }
 939
 940         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 941                         lpm_pkts);
 942
 943         lpm_pkts = 0;
 944
 945         for (i = 0; i < nb_pkts; i++) {
 946                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 947                         /* Read hop from the SA */
 948                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 949                 } else {
 950                         /* Need to use hop returned by lookup */
 951                         pkt_hop = hop[lpm_pkts++];
 952                 }
 953
 954                 if (pkt_hop == -1) {
 955                         rte_pktmbuf_free(pkts[i]);
 956                         continue;
 957                 }
 958                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
 959         }
 960 }
 961
 962 static inline void
 963 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 964                 uint8_t nb_pkts, uint16_t portid)
 965 {
 966         struct ipsec_traffic traffic;
 967
 968         prepare_traffic(pkts, &traffic, nb_pkts);
 969
 970         if (unlikely(single_sa)) {
 971                 if (UNPROTECTED_PORT(portid))
 972                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 973                 else
 974                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 975         } else {
 976                 if (UNPROTECTED_PORT(portid))
 977                         process_pkts_inbound(&qconf->inbound, &traffic);
 978                 else
 979                         process_pkts_outbound(&qconf->outbound, &traffic);
 980         }
 981
 982         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 983         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 984 }
 985
 986 static inline void
 987 drain_tx_buffers(struct lcore_conf *qconf)
 988 {
 989         struct buffer *buf;
 990         uint32_t portid;
 991
 992         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 993                 buf = &qconf->tx_mbufs[portid];
 994                 if (buf->len == 0)
 995                         continue;
 996                 send_burst(qconf, buf->len, portid);
 997                 buf->len = 0;
 998         }
 999 }
1000
1001 static inline void
1002 drain_crypto_buffers(struct lcore_conf *qconf)
1003 {
1004         uint32_t i;
1005         struct ipsec_ctx *ctx;
1006
1007         /* drain inbound buffers*/
1008         ctx = &qconf->inbound;
1009         for (i = 0; i != ctx->nb_qps; i++) {
1010                 if (ctx->tbl[i].len != 0)
1011                         enqueue_cop_burst(ctx->tbl  + i);
1012         }
1013
1014         /* drain outbound buffers*/
1015         ctx = &qconf->outbound;
1016         for (i = 0; i != ctx->nb_qps; i++) {
1017                 if (ctx->tbl[i].len != 0)
1018                         enqueue_cop_burst(ctx->tbl  + i);
1019         }
1020 }
1021
1022 static void
1023 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1024                 struct ipsec_ctx *ctx)
1025 {
1026         uint32_t n;
1027         struct ipsec_traffic trf;
1028
1029         if (app_sa_prm.enable == 0) {
1030
1031                 /* dequeue packets from crypto-queue */
1032                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1033                         RTE_DIM(trf.ipsec.pkts));
1034
1035                 trf.ip4.num = 0;
1036                 trf.ip6.num = 0;
1037
1038                 /* split traffic by ipv4-ipv6 */
1039                 split46_traffic(&trf, trf.ipsec.pkts, n);
1040         } else
1041                 ipsec_cqp_process(ctx, &trf);
1042
1043         /* process ipv4 packets */
1044         if (trf.ip4.num != 0) {
1045                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1046                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1047         }
1048
1049         /* process ipv6 packets */
1050         if (trf.ip6.num != 0) {
1051                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1052                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1053         }
1054 }
1055
1056 static void
1057 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1058                 struct ipsec_ctx *ctx)
1059 {
1060         uint32_t n;
1061         struct ipsec_traffic trf;
1062
1063         if (app_sa_prm.enable == 0) {
1064
1065                 /* dequeue packets from crypto-queue */
1066                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1067                         RTE_DIM(trf.ipsec.pkts));
1068
1069                 trf.ip4.num = 0;
1070                 trf.ip6.num = 0;
1071
1072                 /* split traffic by ipv4-ipv6 */
1073                 split46_traffic(&trf, trf.ipsec.pkts, n);
1074         } else
1075                 ipsec_cqp_process(ctx, &trf);
1076
1077         /* process ipv4 packets */
1078         if (trf.ip4.num != 0)
1079                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1080
1081         /* process ipv6 packets */
1082         if (trf.ip6.num != 0)
1083                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1084 }
1085
1086 /* main processing loop */
1087 static int32_t
1088 main_loop(__attribute__((unused)) void *dummy)
1089 {
1090         struct rte_mbuf *pkts[MAX_PKT_BURST];
1091         uint32_t lcore_id;
1092         uint64_t prev_tsc, diff_tsc, cur_tsc;
1093         int32_t i, nb_rx;
1094         uint16_t portid;
1095         uint8_t queueid;
1096         struct lcore_conf *qconf;
1097         int32_t socket_id;
1098         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1099                         / US_PER_S * BURST_TX_DRAIN_US;
1100         struct lcore_rx_queue *rxql;
1101
1102         prev_tsc = 0;
1103         lcore_id = rte_lcore_id();
1104         qconf = &lcore_conf[lcore_id];
1105         rxql = qconf->rx_queue_list;
1106         socket_id = rte_lcore_to_socket_id(lcore_id);
1107
1108         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1109         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1110         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1111         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1112         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1113         qconf->inbound.cdev_map = cdev_map_in;
1114         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1115         qconf->inbound.session_priv_pool =
1116                         socket_ctx[socket_id].session_priv_pool;
1117         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1118         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1119         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1120         qconf->outbound.cdev_map = cdev_map_out;
1121         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1122         qconf->outbound.session_priv_pool =
1123                         socket_ctx[socket_id].session_priv_pool;
1124         qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1125         qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1126
1127         if (qconf->nb_rx_queue == 0) {
1128                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1129                         lcore_id);
1130                 return 0;
1131         }
1132
1133         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1134
1135         for (i = 0; i < qconf->nb_rx_queue; i++) {
1136                 portid = rxql[i].port_id;
1137                 queueid = rxql[i].queue_id;
1138                 RTE_LOG(INFO, IPSEC,
1139                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1140                         lcore_id, portid, queueid);
1141         }
1142
1143         while (1) {
1144                 cur_tsc = rte_rdtsc();
1145
1146                 /* TX queue buffer drain */
1147                 diff_tsc = cur_tsc - prev_tsc;
1148
1149                 if (unlikely(diff_tsc > drain_tsc)) {
1150                         drain_tx_buffers(qconf);
1151                         drain_crypto_buffers(qconf);
1152                         prev_tsc = cur_tsc;
1153                 }
1154
1155                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1156
1157                         /* Read packets from RX queues */
1158                         portid = rxql[i].port_id;
1159                         queueid = rxql[i].queue_id;
1160                         nb_rx = rte_eth_rx_burst(portid, queueid,
1161                                         pkts, MAX_PKT_BURST);
1162
1163                         if (nb_rx > 0)
1164                                 process_pkts(qconf, pkts, nb_rx, portid);
1165
1166                         /* dequeue and process completed crypto-ops */
1167                         if (UNPROTECTED_PORT(portid))
1168                                 drain_inbound_crypto_queues(qconf,
1169                                         &qconf->inbound);
1170                         else
1171                                 drain_outbound_crypto_queues(qconf,
1172                                         &qconf->outbound);
1173                 }
1174         }
1175 }
1176
1177 static int32_t
1178 check_params(void)
1179 {
1180         uint8_t lcore;
1181         uint16_t portid;
1182         uint16_t i;
1183         int32_t socket_id;
1184
1185         if (lcore_params == NULL) {
1186                 printf("Error: No port/queue/core mappings\n");
1187                 return -1;
1188         }
1189
1190         for (i = 0; i < nb_lcore_params; ++i) {
1191                 lcore = lcore_params[i].lcore_id;
1192                 if (!rte_lcore_is_enabled(lcore)) {
1193                         printf("error: lcore %hhu is not enabled in "
1194                                 "lcore mask\n", lcore);
1195                         return -1;
1196                 }
1197                 socket_id = rte_lcore_to_socket_id(lcore);
1198                 if (socket_id != 0 && numa_on == 0) {
1199                         printf("warning: lcore %hhu is on socket %d "
1200                                 "with numa off\n",
1201                                 lcore, socket_id);
1202                 }
1203                 portid = lcore_params[i].port_id;
1204                 if ((enabled_port_mask & (1 << portid)) == 0) {
1205                         printf("port %u is not enabled in port mask\n", portid);
1206                         return -1;
1207                 }
1208                 if (!rte_eth_dev_is_valid_port(portid)) {
1209                         printf("port %u is not present on the board\n", portid);
1210                         return -1;
1211                 }
1212         }
1213         return 0;
1214 }
1215
1216 static uint8_t
1217 get_port_nb_rx_queues(const uint16_t port)
1218 {
1219         int32_t queue = -1;
1220         uint16_t i;
1221
1222         for (i = 0; i < nb_lcore_params; ++i) {
1223                 if (lcore_params[i].port_id == port &&
1224                                 lcore_params[i].queue_id > queue)
1225                         queue = lcore_params[i].queue_id;
1226         }
1227         return (uint8_t)(++queue);
1228 }
1229
1230 static int32_t
1231 init_lcore_rx_queues(void)
1232 {
1233         uint16_t i, nb_rx_queue;
1234         uint8_t lcore;
1235
1236         for (i = 0; i < nb_lcore_params; ++i) {
1237                 lcore = lcore_params[i].lcore_id;
1238                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1239                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1240                         printf("error: too many queues (%u) for lcore: %u\n",
1241                                         nb_rx_queue + 1, lcore);
1242                         return -1;
1243                 }
1244                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1245                         lcore_params[i].port_id;
1246                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1247                         lcore_params[i].queue_id;
1248                 lcore_conf[lcore].nb_rx_queue++;
1249         }
1250         return 0;
1251 }
1252
1253 /* display usage */
1254 static void
1255 print_usage(const char *prgname)
1256 {
1257         fprintf(stderr, "%s [EAL options] --"
1258                 " -p PORTMASK"
1259                 " [-P]"
1260                 " [-u PORTMASK]"
1261                 " [-j FRAMESIZE]"
1262                 " [-l]"
1263                 " [-w REPLAY_WINDOW_SIZE]"
1264                 " [-e]"
1265                 " [-a]"
1266                 " -f CONFIG_FILE"
1267                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1268                 " [--single-sa SAIDX]"
1269                 " [--cryptodev_mask MASK]"
1270                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1271                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1272                 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1273                 " [--" CMD_LINE_OPT_MTU " MTU]"
1274                 "\n\n"
1275                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1276                 "  -P : Enable promiscuous mode\n"
1277                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1278                 "  -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1279                 "     value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1280                 "  -l enables code-path that uses librte_ipsec\n"
1281                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1282                 "     size for each SA\n"
1283                 "  -e enables ESN\n"
1284                 "  -a enables SA SQN atomic behaviour\n"
1285                 "  -f CONFIG_FILE: Configuration file\n"
1286                 "  --config (port,queue,lcore): Rx queue configuration\n"
1287                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1288                 "                     bypassing the SP\n"
1289                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1290                 "                         devices to configure\n"
1291                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1292                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1293                 "                         (DEV_RX_OFFLOAD_*)\n"
1294                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1295                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1296                 "                         (DEV_TX_OFFLOAD_*)\n"
1297                 "  --" CMD_LINE_OPT_REASSEMBLE " NUM"
1298                 ": max number of entries in reassemble(fragment) table\n"
1299                 "    (zero (default value) disables reassembly)\n"
1300                 "  --" CMD_LINE_OPT_MTU " MTU"
1301                 ": MTU value on all ports (default value: 1500)\n"
1302                 "    outgoing packets with bigger size will be fragmented\n"
1303                 "    incoming packets with bigger size will be discarded\n"
1304                 "\n",
1305                 prgname);
1306 }
1307
1308 static int
1309 parse_mask(const char *str, uint64_t *val)
1310 {
1311         char *end;
1312         unsigned long t;
1313
1314         errno = 0;
1315         t = strtoul(str, &end, 0);
1316         if (errno != 0 || end[0] != 0)
1317                 return -EINVAL;
1318
1319         *val = t;
1320         return 0;
1321 }
1322
1323 static int32_t
1324 parse_portmask(const char *portmask)
1325 {
1326         char *end = NULL;
1327         unsigned long pm;
1328
1329         /* parse hexadecimal string */
1330         pm = strtoul(portmask, &end, 16);
1331         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1332                 return -1;
1333
1334         if ((pm == 0) && errno)
1335                 return -1;
1336
1337         return pm;
1338 }
1339
1340 static int32_t
1341 parse_decimal(const char *str)
1342 {
1343         char *end = NULL;
1344         unsigned long num;
1345
1346         num = strtoul(str, &end, 10);
1347         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
1348                 return -1;
1349
1350         return num;
1351 }
1352
1353 static int32_t
1354 parse_config(const char *q_arg)
1355 {
1356         char s[256];
1357         const char *p, *p0 = q_arg;
1358         char *end;
1359         enum fieldnames {
1360                 FLD_PORT = 0,
1361                 FLD_QUEUE,
1362                 FLD_LCORE,
1363                 _NUM_FLD
1364         };
1365         unsigned long int_fld[_NUM_FLD];
1366         char *str_fld[_NUM_FLD];
1367         int32_t i;
1368         uint32_t size;
1369
1370         nb_lcore_params = 0;
1371
1372         while ((p = strchr(p0, '(')) != NULL) {
1373                 ++p;
1374                 p0 = strchr(p, ')');
1375                 if (p0 == NULL)
1376                         return -1;
1377
1378                 size = p0 - p;
1379                 if (size >= sizeof(s))
1380                         return -1;
1381
1382                 snprintf(s, sizeof(s), "%.*s", size, p);
1383                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1384                                 _NUM_FLD)
1385                         return -1;
1386                 for (i = 0; i < _NUM_FLD; i++) {
1387                         errno = 0;
1388                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1389                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1390                                 return -1;
1391                 }
1392                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1393                         printf("exceeded max number of lcore params: %hu\n",
1394                                 nb_lcore_params);
1395                         return -1;
1396                 }
1397                 lcore_params_array[nb_lcore_params].port_id =
1398                         (uint8_t)int_fld[FLD_PORT];
1399                 lcore_params_array[nb_lcore_params].queue_id =
1400                         (uint8_t)int_fld[FLD_QUEUE];
1401                 lcore_params_array[nb_lcore_params].lcore_id =
1402                         (uint8_t)int_fld[FLD_LCORE];
1403                 ++nb_lcore_params;
1404         }
1405         lcore_params = lcore_params_array;
1406         return 0;
1407 }
1408
1409 static void
1410 print_app_sa_prm(const struct app_sa_prm *prm)
1411 {
1412         printf("librte_ipsec usage: %s\n",
1413                 (prm->enable == 0) ? "disabled" : "enabled");
1414
1415         if (prm->enable == 0)
1416                 return;
1417
1418         printf("replay window size: %u\n", prm->window_size);
1419         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1420         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1421 }
1422
1423 static int32_t
1424 parse_args(int32_t argc, char **argv)
1425 {
1426         int32_t opt, ret;
1427         char **argvopt;
1428         int32_t option_index;
1429         char *prgname = argv[0];
1430         int32_t f_present = 0;
1431
1432         argvopt = argv;
1433
1434         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:",
1435                                 lgopts, &option_index)) != EOF) {
1436
1437                 switch (opt) {
1438                 case 'p':
1439                         enabled_port_mask = parse_portmask(optarg);
1440                         if (enabled_port_mask == 0) {
1441                                 printf("invalid portmask\n");
1442                                 print_usage(prgname);
1443                                 return -1;
1444                         }
1445                         break;
1446                 case 'P':
1447                         printf("Promiscuous mode selected\n");
1448                         promiscuous_on = 1;
1449                         break;
1450                 case 'u':
1451                         unprotected_port_mask = parse_portmask(optarg);
1452                         if (unprotected_port_mask == 0) {
1453                                 printf("invalid unprotected portmask\n");
1454                                 print_usage(prgname);
1455                                 return -1;
1456                         }
1457                         break;
1458                 case 'f':
1459                         if (f_present == 1) {
1460                                 printf("\"-f\" option present more than "
1461                                         "once!\n");
1462                                 print_usage(prgname);
1463                                 return -1;
1464                         }
1465                         if (parse_cfg_file(optarg) < 0) {
1466                                 printf("parsing file \"%s\" failed\n",
1467                                         optarg);
1468                                 print_usage(prgname);
1469                                 return -1;
1470                         }
1471                         f_present = 1;
1472                         break;
1473                 case 'j':
1474                         ret = parse_decimal(optarg);
1475                         if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1476                                         ret > UINT16_MAX) {
1477                                 printf("Invalid frame buffer size value: %s\n",
1478                                         optarg);
1479                                 print_usage(prgname);
1480                                 return -1;
1481                         }
1482                         frame_buf_size = ret;
1483                         printf("Custom frame buffer size %u\n", frame_buf_size);
1484                         break;
1485                 case 'l':
1486                         app_sa_prm.enable = 1;
1487                         break;
1488                 case 'w':
1489                         app_sa_prm.enable = 1;
1490                         app_sa_prm.window_size = parse_decimal(optarg);
1491                         break;
1492                 case 'e':
1493                         app_sa_prm.enable = 1;
1494                         app_sa_prm.enable_esn = 1;
1495                         break;
1496                 case 'a':
1497                         app_sa_prm.enable = 1;
1498                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1499                         break;
1500                 case CMD_LINE_OPT_CONFIG_NUM:
1501                         ret = parse_config(optarg);
1502                         if (ret) {
1503                                 printf("Invalid config\n");
1504                                 print_usage(prgname);
1505                                 return -1;
1506                         }
1507                         break;
1508                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1509                         ret = parse_decimal(optarg);
1510                         if (ret == -1) {
1511                                 printf("Invalid argument[sa_idx]\n");
1512                                 print_usage(prgname);
1513                                 return -1;
1514                         }
1515
1516                         /* else */
1517                         single_sa = 1;
1518                         single_sa_idx = ret;
1519                         printf("Configured with single SA index %u\n",
1520                                         single_sa_idx);
1521                         break;
1522                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1523                         ret = parse_portmask(optarg);
1524                         if (ret == -1) {
1525                                 printf("Invalid argument[portmask]\n");
1526                                 print_usage(prgname);
1527                                 return -1;
1528                         }
1529
1530                         /* else */
1531                         enabled_cryptodev_mask = ret;
1532                         break;
1533                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1534                         ret = parse_mask(optarg, &dev_rx_offload);
1535                         if (ret != 0) {
1536                                 printf("Invalid argument for \'%s\': %s\n",
1537                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1538                                 print_usage(prgname);
1539                                 return -1;
1540                         }
1541                         break;
1542                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1543                         ret = parse_mask(optarg, &dev_tx_offload);
1544                         if (ret != 0) {
1545                                 printf("Invalid argument for \'%s\': %s\n",
1546                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1547                                 print_usage(prgname);
1548                                 return -1;
1549                         }
1550                         break;
1551                 case CMD_LINE_OPT_REASSEMBLE_NUM:
1552                         ret = parse_decimal(optarg);
1553                         if (ret < 0) {
1554                                 printf("Invalid argument for \'%s\': %s\n",
1555                                         CMD_LINE_OPT_REASSEMBLE, optarg);
1556                                 print_usage(prgname);
1557                                 return -1;
1558                         }
1559                         frag_tbl_sz = ret;
1560                         break;
1561                 case CMD_LINE_OPT_MTU_NUM:
1562                         ret = parse_decimal(optarg);
1563                         if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1564                                 printf("Invalid argument for \'%s\': %s\n",
1565                                         CMD_LINE_OPT_MTU, optarg);
1566                                 print_usage(prgname);
1567                                 return -1;
1568                         }
1569                         mtu_size = ret;
1570                         break;
1571                 default:
1572                         print_usage(prgname);
1573                         return -1;
1574                 }
1575         }
1576
1577         if (f_present == 0) {
1578                 printf("Mandatory option \"-f\" not present\n");
1579                 return -1;
1580         }
1581
1582         /* check do we need to enable multi-seg support */
1583         if (multi_seg_required()) {
1584                 /* legacy mode doesn't support multi-seg */
1585                 app_sa_prm.enable = 1;
1586                 printf("frame buf size: %u, mtu: %u, "
1587                         "number of reassemble entries: %u\n"
1588                         "multi-segment support is required\n",
1589                         frame_buf_size, mtu_size, frag_tbl_sz);
1590         }
1591
1592         print_app_sa_prm(&app_sa_prm);
1593
1594         if (optind >= 0)
1595                 argv[optind-1] = prgname;
1596
1597         ret = optind-1;
1598         optind = 1; /* reset getopt lib */
1599         return ret;
1600 }
1601
1602 static void
1603 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1604 {
1605         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1606         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1607         printf("%s%s", name, buf);
1608 }
1609
1610 /*
1611  * Update destination ethaddr for the port.
1612  */
1613 int
1614 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1615 {
1616         if (port >= RTE_DIM(ethaddr_tbl))
1617                 return -EINVAL;
1618
1619         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1620         return 0;
1621 }
1622
1623 /* Check the link status of all ports in up to 9s, and print them finally */
1624 static void
1625 check_all_ports_link_status(uint32_t port_mask)
1626 {
1627 #define CHECK_INTERVAL 100 /* 100ms */
1628 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1629         uint16_t portid;
1630         uint8_t count, all_ports_up, print_flag = 0;
1631         struct rte_eth_link link;
1632
1633         printf("\nChecking link status");
1634         fflush(stdout);
1635         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1636                 all_ports_up = 1;
1637                 RTE_ETH_FOREACH_DEV(portid) {
1638                         if ((port_mask & (1 << portid)) == 0)
1639                                 continue;
1640                         memset(&link, 0, sizeof(link));
1641                         rte_eth_link_get_nowait(portid, &link);
1642                         /* print link status if flag set */
1643                         if (print_flag == 1) {
1644                                 if (link.link_status)
1645                                         printf(
1646                                         "Port%d Link Up - speed %u Mbps -%s\n",
1647                                                 portid, link.link_speed,
1648                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1649                                         ("full-duplex") : ("half-duplex\n"));
1650                                 else
1651                                         printf("Port %d Link Down\n", portid);
1652                                 continue;
1653                         }
1654                         /* clear all_ports_up flag if any link down */
1655                         if (link.link_status == ETH_LINK_DOWN) {
1656                                 all_ports_up = 0;
1657                                 break;
1658                         }
1659                 }
1660                 /* after finally printing all link status, get out */
1661                 if (print_flag == 1)
1662                         break;
1663
1664                 if (all_ports_up == 0) {
1665                         printf(".");
1666                         fflush(stdout);
1667                         rte_delay_ms(CHECK_INTERVAL);
1668                 }
1669
1670                 /* set the print_flag if all ports up or timeout */
1671                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1672                         print_flag = 1;
1673                         printf("done\n");
1674                 }
1675         }
1676 }
1677
1678 static int32_t
1679 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1680                 uint16_t qp, struct lcore_params *params,
1681                 struct ipsec_ctx *ipsec_ctx,
1682                 const struct rte_cryptodev_capabilities *cipher,
1683                 const struct rte_cryptodev_capabilities *auth,
1684                 const struct rte_cryptodev_capabilities *aead)
1685 {
1686         int32_t ret = 0;
1687         unsigned long i;
1688         struct cdev_key key = { 0 };
1689
1690         key.lcore_id = params->lcore_id;
1691         if (cipher)
1692                 key.cipher_algo = cipher->sym.cipher.algo;
1693         if (auth)
1694                 key.auth_algo = auth->sym.auth.algo;
1695         if (aead)
1696                 key.aead_algo = aead->sym.aead.algo;
1697
1698         ret = rte_hash_lookup(map, &key);
1699         if (ret != -ENOENT)
1700                 return 0;
1701
1702         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1703                 if (ipsec_ctx->tbl[i].id == cdev_id)
1704                         break;
1705
1706         if (i == ipsec_ctx->nb_qps) {
1707                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1708                         printf("Maximum number of crypto devices assigned to "
1709                                 "a core, increase MAX_QP_PER_LCORE value\n");
1710                         return 0;
1711                 }
1712                 ipsec_ctx->tbl[i].id = cdev_id;
1713                 ipsec_ctx->tbl[i].qp = qp;
1714                 ipsec_ctx->nb_qps++;
1715                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1716                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1717                                 cdev_id, qp, i);
1718         }
1719
1720         ret = rte_hash_add_key_data(map, &key, (void *)i);
1721         if (ret < 0) {
1722                 printf("Faled to insert cdev mapping for (lcore %u, "
1723                                 "cdev %u, qp %u), errno %d\n",
1724                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1725                                 ipsec_ctx->tbl[i].qp, ret);
1726                 return 0;
1727         }
1728
1729         return 1;
1730 }
1731
1732 static int32_t
1733 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1734                 uint16_t qp, struct lcore_params *params)
1735 {
1736         int32_t ret = 0;
1737         const struct rte_cryptodev_capabilities *i, *j;
1738         struct rte_hash *map;
1739         struct lcore_conf *qconf;
1740         struct ipsec_ctx *ipsec_ctx;
1741         const char *str;
1742
1743         qconf = &lcore_conf[params->lcore_id];
1744
1745         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1746                 map = cdev_map_out;
1747                 ipsec_ctx = &qconf->outbound;
1748                 str = "Outbound";
1749         } else {
1750                 map = cdev_map_in;
1751                 ipsec_ctx = &qconf->inbound;
1752                 str = "Inbound";
1753         }
1754
1755         /* Required cryptodevs with operation chainning */
1756         if (!(dev_info->feature_flags &
1757                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1758                 return ret;
1759
1760         for (i = dev_info->capabilities;
1761                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1762                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1763                         continue;
1764
1765                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1766                         ret |= add_mapping(map, str, cdev_id, qp, params,
1767                                         ipsec_ctx, NULL, NULL, i);
1768                         continue;
1769                 }
1770
1771                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1772                         continue;
1773
1774                 for (j = dev_info->capabilities;
1775                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1776                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1777                                 continue;
1778
1779                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1780                                 continue;
1781
1782                         ret |= add_mapping(map, str, cdev_id, qp, params,
1783                                                 ipsec_ctx, i, j, NULL);
1784                 }
1785         }
1786
1787         return ret;
1788 }
1789
1790 /* Check if the device is enabled by cryptodev_mask */
1791 static int
1792 check_cryptodev_mask(uint8_t cdev_id)
1793 {
1794         if (enabled_cryptodev_mask & (1 << cdev_id))
1795                 return 0;
1796
1797         return -1;
1798 }
1799
1800 static int32_t
1801 cryptodevs_init(void)
1802 {
1803         struct rte_cryptodev_config dev_conf;
1804         struct rte_cryptodev_qp_conf qp_conf;
1805         uint16_t idx, max_nb_qps, qp, i;
1806         int16_t cdev_id, port_id;
1807         struct rte_hash_parameters params = { 0 };
1808
1809         const uint64_t mseg_flag = multi_seg_required() ?
1810                                 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
1811
1812         params.entries = CDEV_MAP_ENTRIES;
1813         params.key_len = sizeof(struct cdev_key);
1814         params.hash_func = rte_jhash;
1815         params.hash_func_init_val = 0;
1816         params.socket_id = rte_socket_id();
1817
1818         params.name = "cdev_map_in";
1819         cdev_map_in = rte_hash_create(&params);
1820         if (cdev_map_in == NULL)
1821                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1822                                 rte_errno);
1823
1824         params.name = "cdev_map_out";
1825         cdev_map_out = rte_hash_create(&params);
1826         if (cdev_map_out == NULL)
1827                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1828                                 rte_errno);
1829
1830         printf("lcore/cryptodev/qp mappings:\n");
1831
1832         uint32_t max_sess_sz = 0, sess_sz;
1833         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1834                 void *sec_ctx;
1835
1836                 /* Get crypto priv session size */
1837                 sess_sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
1838                 if (sess_sz > max_sess_sz)
1839                         max_sess_sz = sess_sz;
1840
1841                 /*
1842                  * If crypto device is security capable, need to check the
1843                  * size of security session as well.
1844                  */
1845
1846                 /* Get security context of the crypto device */
1847                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
1848                 if (sec_ctx == NULL)
1849                         continue;
1850
1851                 /* Get size of security session */
1852                 sess_sz = rte_security_session_get_size(sec_ctx);
1853                 if (sess_sz > max_sess_sz)
1854                         max_sess_sz = sess_sz;
1855         }
1856         RTE_ETH_FOREACH_DEV(port_id) {
1857                 void *sec_ctx;
1858
1859                 if ((enabled_port_mask & (1 << port_id)) == 0)
1860                         continue;
1861
1862                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
1863                 if (sec_ctx == NULL)
1864                         continue;
1865
1866                 sess_sz = rte_security_session_get_size(sec_ctx);
1867                 if (sess_sz > max_sess_sz)
1868                         max_sess_sz = sess_sz;
1869         }
1870
1871         idx = 0;
1872         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1873                 struct rte_cryptodev_info cdev_info;
1874
1875                 if (check_cryptodev_mask((uint8_t)cdev_id))
1876                         continue;
1877
1878                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1879
1880                 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
1881                         rte_exit(EXIT_FAILURE,
1882                                 "Device %hd does not support \'%s\' feature\n",
1883                                 cdev_id,
1884                                 rte_cryptodev_get_feature_name(mseg_flag));
1885
1886                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1887                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1888                 else
1889                         max_nb_qps = nb_lcore_params;
1890
1891                 qp = 0;
1892                 i = 0;
1893                 while (qp < max_nb_qps && i < nb_lcore_params) {
1894                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1895                                                 &lcore_params[idx]))
1896                                 qp++;
1897                         idx++;
1898                         idx = idx % nb_lcore_params;
1899                         i++;
1900                 }
1901
1902                 if (qp == 0)
1903                         continue;
1904
1905                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1906                 dev_conf.nb_queue_pairs = qp;
1907                 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
1908
1909                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1910                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1911                         rte_exit(EXIT_FAILURE,
1912                                 "Device does not support at least %u "
1913                                 "sessions", CDEV_MP_NB_OBJS);
1914
1915                 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1916                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1917                         struct rte_mempool *sess_mp;
1918
1919                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1920                                         "sess_mp_%u", dev_conf.socket_id);
1921                         sess_mp = rte_cryptodev_sym_session_pool_create(
1922                                         mp_name, CDEV_MP_NB_OBJS,
1923                                         0, CDEV_MP_CACHE_SZ, 0,
1924                                         dev_conf.socket_id);
1925                         socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1926                 }
1927
1928                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool) {
1929                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1930                         struct rte_mempool *sess_mp;
1931
1932                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1933                                         "sess_mp_priv_%u", dev_conf.socket_id);
1934                         sess_mp = rte_mempool_create(mp_name,
1935                                         CDEV_MP_NB_OBJS,
1936                                         max_sess_sz,
1937                                         CDEV_MP_CACHE_SZ,
1938                                         0, NULL, NULL, NULL,
1939                                         NULL, dev_conf.socket_id,
1940                                         0);
1941                         socket_ctx[dev_conf.socket_id].session_priv_pool =
1942                                         sess_mp;
1943                 }
1944
1945                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool ||
1946                                 !socket_ctx[dev_conf.socket_id].session_pool)
1947                         rte_exit(EXIT_FAILURE,
1948                                 "Cannot create session pool on socket %d\n",
1949                                 dev_conf.socket_id);
1950                 else
1951                         printf("Allocated session pool on socket %d\n",
1952                                         dev_conf.socket_id);
1953
1954                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1955                         rte_panic("Failed to initialize cryptodev %u\n",
1956                                         cdev_id);
1957
1958                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1959                 qp_conf.mp_session =
1960                         socket_ctx[dev_conf.socket_id].session_pool;
1961                 qp_conf.mp_session_private =
1962                         socket_ctx[dev_conf.socket_id].session_priv_pool;
1963                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1964                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1965                                         &qp_conf, dev_conf.socket_id))
1966                                 rte_panic("Failed to setup queue %u for "
1967                                                 "cdev_id %u\n", 0, cdev_id);
1968
1969                 if (rte_cryptodev_start(cdev_id))
1970                         rte_panic("Failed to start cryptodev %u\n",
1971                                         cdev_id);
1972         }
1973
1974         /* create session pools for eth devices that implement security */
1975         RTE_ETH_FOREACH_DEV(port_id) {
1976                 if ((enabled_port_mask & (1 << port_id)) &&
1977                                 rte_eth_dev_get_sec_ctx(port_id)) {
1978                         int socket_id = rte_eth_dev_socket_id(port_id);
1979
1980                         if (!socket_ctx[socket_id].session_priv_pool) {
1981                                 char mp_name[RTE_MEMPOOL_NAMESIZE];
1982                                 struct rte_mempool *sess_mp;
1983
1984                                 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1985                                                 "sess_mp_%u", socket_id);
1986                                 sess_mp = rte_mempool_create(mp_name,
1987                                                 (CDEV_MP_NB_OBJS * 2),
1988                                                 max_sess_sz,
1989                                                 CDEV_MP_CACHE_SZ,
1990                                                 0, NULL, NULL, NULL,
1991                                                 NULL, socket_id,
1992                                                 0);
1993                                 if (sess_mp == NULL)
1994                                         rte_exit(EXIT_FAILURE,
1995                                                 "Cannot create session pool "
1996                                                 "on socket %d\n", socket_id);
1997                                 else
1998                                         printf("Allocated session pool "
1999                                                 "on socket %d\n", socket_id);
2000                                 socket_ctx[socket_id].session_priv_pool =
2001                                                 sess_mp;
2002                         }
2003                 }
2004         }
2005
2006
2007         printf("\n");
2008
2009         return 0;
2010 }
2011
2012 static void
2013 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2014 {
2015         uint32_t frame_size;
2016         struct rte_eth_dev_info dev_info;
2017         struct rte_eth_txconf *txconf;
2018         uint16_t nb_tx_queue, nb_rx_queue;
2019         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2020         int32_t ret, socket_id;
2021         struct lcore_conf *qconf;
2022         struct rte_ether_addr ethaddr;
2023         struct rte_eth_conf local_port_conf = port_conf;
2024
2025         rte_eth_dev_info_get(portid, &dev_info);
2026
2027         /* limit allowed HW offloafs, as user requested */
2028         dev_info.rx_offload_capa &= dev_rx_offload;
2029         dev_info.tx_offload_capa &= dev_tx_offload;
2030
2031         printf("Configuring device port %u:\n", portid);
2032
2033         rte_eth_macaddr_get(portid, &ethaddr);
2034         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
2035         print_ethaddr("Address: ", &ethaddr);
2036         printf("\n");
2037
2038         nb_rx_queue = get_port_nb_rx_queues(portid);
2039         nb_tx_queue = nb_lcores;
2040
2041         if (nb_rx_queue > dev_info.max_rx_queues)
2042                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2043                                 "(max rx queue is %u)\n",
2044                                 nb_rx_queue, dev_info.max_rx_queues);
2045
2046         if (nb_tx_queue > dev_info.max_tx_queues)
2047                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2048                                 "(max tx queue is %u)\n",
2049                                 nb_tx_queue, dev_info.max_tx_queues);
2050
2051         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2052                         nb_rx_queue, nb_tx_queue);
2053
2054         frame_size = MTU_TO_FRAMELEN(mtu_size);
2055         if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
2056                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2057         local_port_conf.rxmode.max_rx_pkt_len = frame_size;
2058
2059         if (multi_seg_required()) {
2060                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
2061                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
2062         }
2063
2064         local_port_conf.rxmode.offloads |= req_rx_offloads;
2065         local_port_conf.txmode.offloads |= req_tx_offloads;
2066
2067         /* Check that all required capabilities are supported */
2068         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2069                         local_port_conf.rxmode.offloads)
2070                 rte_exit(EXIT_FAILURE,
2071                         "Error: port %u required RX offloads: 0x%" PRIx64
2072                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2073                         portid, local_port_conf.rxmode.offloads,
2074                         dev_info.rx_offload_capa);
2075
2076         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2077                         local_port_conf.txmode.offloads)
2078                 rte_exit(EXIT_FAILURE,
2079                         "Error: port %u required TX offloads: 0x%" PRIx64
2080                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2081                         portid, local_port_conf.txmode.offloads,
2082                         dev_info.tx_offload_capa);
2083
2084         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2085                 local_port_conf.txmode.offloads |=
2086                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2087
2088         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
2089                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2090
2091         printf("port %u configurng rx_offloads=0x%" PRIx64
2092                 ", tx_offloads=0x%" PRIx64 "\n",
2093                 portid, local_port_conf.rxmode.offloads,
2094                 local_port_conf.txmode.offloads);
2095
2096         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2097                 dev_info.flow_type_rss_offloads;
2098         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2099                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
2100                 printf("Port %u modified RSS hash function based on hardware support,"
2101                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2102                         portid,
2103                         port_conf.rx_adv_conf.rss_conf.rss_hf,
2104                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2105         }
2106
2107         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2108                         &local_port_conf);
2109         if (ret < 0)
2110                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2111                                 "err=%d, port=%d\n", ret, portid);
2112
2113         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2114         if (ret < 0)
2115                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2116                                 "err=%d, port=%d\n", ret, portid);
2117
2118         /* init one TX queue per lcore */
2119         tx_queueid = 0;
2120         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2121                 if (rte_lcore_is_enabled(lcore_id) == 0)
2122                         continue;
2123
2124                 if (numa_on)
2125                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2126                 else
2127                         socket_id = 0;
2128
2129                 /* init TX queue */
2130                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2131
2132                 txconf = &dev_info.default_txconf;
2133                 txconf->offloads = local_port_conf.txmode.offloads;
2134
2135                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2136                                 socket_id, txconf);
2137                 if (ret < 0)
2138                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2139                                         "err=%d, port=%d\n", ret, portid);
2140
2141                 qconf = &lcore_conf[lcore_id];
2142                 qconf->tx_queue_id[portid] = tx_queueid;
2143
2144                 /* Pre-populate pkt offloads based on capabilities */
2145                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2146                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2147                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2148                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2149
2150                 tx_queueid++;
2151
2152                 /* init RX queues */
2153                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2154                         struct rte_eth_rxconf rxq_conf;
2155
2156                         if (portid != qconf->rx_queue_list[queue].port_id)
2157                                 continue;
2158
2159                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
2160
2161                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2162                                         socket_id);
2163
2164                         rxq_conf = dev_info.default_rxconf;
2165                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
2166                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2167                                         nb_rxd, socket_id, &rxq_conf,
2168                                         socket_ctx[socket_id].mbuf_pool);
2169                         if (ret < 0)
2170                                 rte_exit(EXIT_FAILURE,
2171                                         "rte_eth_rx_queue_setup: err=%d, "
2172                                         "port=%d\n", ret, portid);
2173                 }
2174         }
2175         printf("\n");
2176 }
2177
2178 static void
2179 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2180 {
2181         char s[64];
2182         int32_t ms;
2183
2184         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2185         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2186                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2187                         frame_buf_size, socket_id);
2188
2189         /*
2190          * if multi-segment support is enabled, then create a pool
2191          * for indirect mbufs.
2192          */
2193         ms = multi_seg_required();
2194         if (ms != 0) {
2195                 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2196                 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2197                         MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2198         }
2199
2200         if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2201                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2202                                 socket_id);
2203         else
2204                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2205 }
2206
2207 static inline int
2208 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2209 {
2210         struct ipsec_sa *sa;
2211
2212         /* For inline protocol processing, the metadata in the event will
2213          * uniquely identify the security session which raised the event.
2214          * Application would then need the userdata it had registered with the
2215          * security session to process the event.
2216          */
2217
2218         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2219
2220         if (sa == NULL) {
2221                 /* userdata could not be retrieved */
2222                 return -1;
2223         }
2224
2225         /* Sequence number over flow. SA need to be re-established */
2226         RTE_SET_USED(sa);
2227         return 0;
2228 }
2229
2230 static int
2231 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2232                  void *param, void *ret_param)
2233 {
2234         uint64_t md;
2235         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2236         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2237                                         rte_eth_dev_get_sec_ctx(port_id);
2238
2239         RTE_SET_USED(param);
2240
2241         if (type != RTE_ETH_EVENT_IPSEC)
2242                 return -1;
2243
2244         event_desc = ret_param;
2245         if (event_desc == NULL) {
2246                 printf("Event descriptor not set\n");
2247                 return -1;
2248         }
2249
2250         md = event_desc->metadata;
2251
2252         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2253                 return inline_ipsec_event_esn_overflow(ctx, md);
2254         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2255                 printf("Invalid IPsec event reported\n");
2256                 return -1;
2257         }
2258
2259         return -1;
2260 }
2261
2262 static uint16_t
2263 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2264         struct rte_mbuf *pkt[], uint16_t nb_pkts,
2265         __rte_unused uint16_t max_pkts, void *user_param)
2266 {
2267         uint64_t tm;
2268         uint32_t i, k;
2269         struct lcore_conf *lc;
2270         struct rte_mbuf *mb;
2271         struct rte_ether_hdr *eth;
2272
2273         lc = user_param;
2274         k = 0;
2275         tm = 0;
2276
2277         for (i = 0; i != nb_pkts; i++) {
2278
2279                 mb = pkt[i];
2280                 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2281                 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2282
2283                         struct rte_ipv4_hdr *iph;
2284
2285                         iph = (struct rte_ipv4_hdr *)(eth + 1);
2286                         if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2287
2288                                 mb->l2_len = sizeof(*eth);
2289                                 mb->l3_len = sizeof(*iph);
2290                                 tm = (tm != 0) ? tm : rte_rdtsc();
2291                                 mb = rte_ipv4_frag_reassemble_packet(
2292                                         lc->frag.tbl, &lc->frag.dr,
2293                                         mb, tm, iph);
2294
2295                                 if (mb != NULL) {
2296                                         /* fix ip cksum after reassemble. */
2297                                         iph = rte_pktmbuf_mtod_offset(mb,
2298                                                 struct rte_ipv4_hdr *,
2299                                                 mb->l2_len);
2300                                         iph->hdr_checksum = 0;
2301                                         iph->hdr_checksum = rte_ipv4_cksum(iph);
2302                                 }
2303                         }
2304                 } else if (eth->ether_type ==
2305                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2306
2307                         struct rte_ipv6_hdr *iph;
2308                         struct ipv6_extension_fragment *fh;
2309
2310                         iph = (struct rte_ipv6_hdr *)(eth + 1);
2311                         fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2312                         if (fh != NULL) {
2313                                 mb->l2_len = sizeof(*eth);
2314                                 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2315                                         sizeof(*fh);
2316                                 tm = (tm != 0) ? tm : rte_rdtsc();
2317                                 mb = rte_ipv6_frag_reassemble_packet(
2318                                         lc->frag.tbl, &lc->frag.dr,
2319                                         mb, tm, iph, fh);
2320                                 if (mb != NULL)
2321                                         /* fix l3_len after reassemble. */
2322                                         mb->l3_len = mb->l3_len - sizeof(*fh);
2323                         }
2324                 }
2325
2326                 pkt[k] = mb;
2327                 k += (mb != NULL);
2328         }
2329
2330         /* some fragments were encountered, drain death row */
2331         if (tm != 0)
2332                 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2333
2334         return k;
2335 }
2336
2337
2338 static int
2339 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2340 {
2341         int32_t sid;
2342         uint32_t i;
2343         uint64_t frag_cycles;
2344         const struct lcore_rx_queue *rxq;
2345         const struct rte_eth_rxtx_callback *cb;
2346
2347         /* create fragment table */
2348         sid = rte_lcore_to_socket_id(cid);
2349         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) /
2350                 MS_PER_S * FRAG_TTL_MS;
2351
2352         lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2353                 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2354         if (lc->frag.tbl == NULL) {
2355                 printf("%s(%u): failed to create fragment table of size: %u, "
2356                         "error code: %d\n",
2357                         __func__, cid, frag_tbl_sz, rte_errno);
2358                 return -ENOMEM;
2359         }
2360
2361         /* setup reassemble RX callbacks for all queues */
2362         for (i = 0; i != lc->nb_rx_queue; i++) {
2363
2364                 rxq = lc->rx_queue_list + i;
2365                 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2366                         rx_callback, lc);
2367                 if (cb == NULL) {
2368                         printf("%s(%u): failed to install RX callback for "
2369                                 "portid=%u, queueid=%u, error code: %d\n",
2370                                 __func__, cid,
2371                                 rxq->port_id, rxq->queue_id, rte_errno);
2372                         return -ENOMEM;
2373                 }
2374         }
2375
2376         return 0;
2377 }
2378
2379 static int
2380 reassemble_init(void)
2381 {
2382         int32_t rc;
2383         uint32_t i, lc;
2384
2385         rc = 0;
2386         for (i = 0; i != nb_lcore_params; i++) {
2387                 lc = lcore_params[i].lcore_id;
2388                 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2389                 if (rc != 0)
2390                         break;
2391         }
2392
2393         return rc;
2394 }
2395
2396 int32_t
2397 main(int32_t argc, char **argv)
2398 {
2399         int32_t ret;
2400         uint32_t lcore_id;
2401         uint8_t socket_id;
2402         uint16_t portid;
2403         uint64_t req_rx_offloads, req_tx_offloads;
2404
2405         /* init EAL */
2406         ret = rte_eal_init(argc, argv);
2407         if (ret < 0)
2408                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2409         argc -= ret;
2410         argv += ret;
2411
2412         /* parse application arguments (after the EAL ones) */
2413         ret = parse_args(argc, argv);
2414         if (ret < 0)
2415                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2416
2417         if ((unprotected_port_mask & enabled_port_mask) !=
2418                         unprotected_port_mask)
2419                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2420                                 unprotected_port_mask);
2421
2422         if (check_params() < 0)
2423                 rte_exit(EXIT_FAILURE, "check_params failed\n");
2424
2425         ret = init_lcore_rx_queues();
2426         if (ret < 0)
2427                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2428
2429         nb_lcores = rte_lcore_count();
2430
2431         /* Replicate each context per socket */
2432         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2433                 if (rte_lcore_is_enabled(lcore_id) == 0)
2434                         continue;
2435
2436                 if (numa_on)
2437                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2438                 else
2439                         socket_id = 0;
2440
2441                 if (socket_ctx[socket_id].mbuf_pool)
2442                         continue;
2443
2444                 /* initilaze SPD */
2445                 sp4_init(&socket_ctx[socket_id], socket_id);
2446
2447                 sp6_init(&socket_ctx[socket_id], socket_id);
2448
2449                 /* initilaze SAD */
2450                 sa_init(&socket_ctx[socket_id], socket_id);
2451
2452                 rt_init(&socket_ctx[socket_id], socket_id);
2453
2454                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2455         }
2456
2457         RTE_ETH_FOREACH_DEV(portid) {
2458                 if ((enabled_port_mask & (1 << portid)) == 0)
2459                         continue;
2460
2461                 sa_check_offloads(portid, &req_rx_offloads, &req_tx_offloads);
2462                 port_init(portid, req_rx_offloads, req_tx_offloads);
2463         }
2464
2465         cryptodevs_init();
2466
2467         /* start ports */
2468         RTE_ETH_FOREACH_DEV(portid) {
2469                 if ((enabled_port_mask & (1 << portid)) == 0)
2470                         continue;
2471
2472                 /* Start device */
2473                 ret = rte_eth_dev_start(portid);
2474                 if (ret < 0)
2475                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2476                                         "err=%d, port=%d\n", ret, portid);
2477                 /*
2478                  * If enabled, put device in promiscuous mode.
2479                  * This allows IO forwarding mode to forward packets
2480                  * to itself through 2 cross-connected  ports of the
2481                  * target machine.
2482                  */
2483                 if (promiscuous_on)
2484                         rte_eth_promiscuous_enable(portid);
2485
2486                 rte_eth_dev_callback_register(portid,
2487                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2488         }
2489
2490         /* fragment reassemble is enabled */
2491         if (frag_tbl_sz != 0) {
2492                 ret = reassemble_init();
2493                 if (ret != 0)
2494                         rte_exit(EXIT_FAILURE, "failed at reassemble init");
2495         }
2496
2497         check_all_ports_link_status(enabled_port_mask);
2498
2499         /* launch per-lcore init on every lcore */
2500         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2501         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2502                 if (rte_eal_wait_lcore(lcore_id) < 0)
2503                         return -1;
2504         }
2505
2506         return 0;
2507 }