examples/ipsec-secgw/ipsec-secgw.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <stdbool.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <stdint.h>
   9 #include <inttypes.h>
  10 #include <sys/types.h>
  11 #include <netinet/in.h>
  12 #include <netinet/ip.h>
  13 #include <netinet/ip6.h>
  14 #include <string.h>
  15 #include <sys/queue.h>
  16 #include <stdarg.h>
  17 #include <errno.h>
  18 #include <signal.h>
  19 #include <getopt.h>
  20
  21 #include <rte_common.h>
  22 #include <rte_bitmap.h>
  23 #include <rte_byteorder.h>
  24 #include <rte_log.h>
  25 #include <rte_eal.h>
  26 #include <rte_launch.h>
  27 #include <rte_atomic.h>
  28 #include <rte_cycles.h>
  29 #include <rte_prefetch.h>
  30 #include <rte_lcore.h>
  31 #include <rte_per_lcore.h>
  32 #include <rte_branch_prediction.h>
  33 #include <rte_interrupts.h>
  34 #include <rte_random.h>
  35 #include <rte_debug.h>
  36 #include <rte_ether.h>
  37 #include <rte_ethdev.h>
  38 #include <rte_mempool.h>
  39 #include <rte_mbuf.h>
  40 #include <rte_acl.h>
  41 #include <rte_lpm.h>
  42 #include <rte_lpm6.h>
  43 #include <rte_hash.h>
  44 #include <rte_jhash.h>
  45 #include <rte_cryptodev.h>
  46 #include <rte_security.h>
  47 #include <rte_eventdev.h>
  48 #include <rte_ip.h>
  49 #include <rte_ip_frag.h>
  50
  51 #include "event_helper.h"
  52 #include "ipsec.h"
  53 #include "parser.h"
  54 #include "sad.h"
  55
  56 volatile bool force_quit;
  57
  58 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  59
  60 #define MAX_JUMBO_PKT_LEN  9600
  61
  62 #define MEMPOOL_CACHE_SIZE 256
  63
  64 #define NB_MBUF (32000)
  65
  66 #define CDEV_QUEUE_DESC 2048
  67 #define CDEV_MAP_ENTRIES 16384
  68 #define CDEV_MP_NB_OBJS 1024
  69 #define CDEV_MP_CACHE_SZ 64
  70 #define MAX_QUEUE_PAIRS 1
  71
  72 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  73
  74 #define NB_SOCKETS 4
  75
  76 /* Configure how many packets ahead to prefetch, when reading packets */
  77 #define PREFETCH_OFFSET 3
  78
  79 #define MAX_RX_QUEUE_PER_LCORE 16
  80
  81 #define MAX_LCORE_PARAMS 1024
  82
  83 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
  84
  85 /*
  86  * Configurable number of RX/TX ring descriptors
  87  */
  88 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  89 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  90 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  91 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  92
  93 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
  94 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  95         (((uint64_t)((a) & 0xff) << 56) | \
  96         ((uint64_t)((b) & 0xff) << 48) | \
  97         ((uint64_t)((c) & 0xff) << 40) | \
  98         ((uint64_t)((d) & 0xff) << 32) | \
  99         ((uint64_t)((e) & 0xff) << 24) | \
 100         ((uint64_t)((f) & 0xff) << 16) | \
 101         ((uint64_t)((g) & 0xff) << 8)  | \
 102         ((uint64_t)(h) & 0xff))
 103 #else
 104 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
 105         (((uint64_t)((h) & 0xff) << 56) | \
 106         ((uint64_t)((g) & 0xff) << 48) | \
 107         ((uint64_t)((f) & 0xff) << 40) | \
 108         ((uint64_t)((e) & 0xff) << 32) | \
 109         ((uint64_t)((d) & 0xff) << 24) | \
 110         ((uint64_t)((c) & 0xff) << 16) | \
 111         ((uint64_t)((b) & 0xff) << 8) | \
 112         ((uint64_t)(a) & 0xff))
 113 #endif
 114 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 115
 116 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 117                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
 118                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
 119                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
 120                 0, 0)
 121
 122 #define FRAG_TBL_BUCKET_ENTRIES 4
 123 #define MAX_FRAG_TTL_NS         (10LL * NS_PER_S)
 124
 125 #define MTU_TO_FRAMELEN(x)      ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
 126
 127 /* port/source ethernet addr and destination ethernet addr */
 128 struct ethaddr_info {
 129         uint64_t src, dst;
 130 };
 131
 132 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 133         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 134         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 135         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 136         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 137 };
 138
 139 struct flow_info flow_info_tbl[RTE_MAX_ETHPORTS];
 140
 141 #define CMD_LINE_OPT_CONFIG             "config"
 142 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 143 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 144 #define CMD_LINE_OPT_TRANSFER_MODE      "transfer-mode"
 145 #define CMD_LINE_OPT_SCHEDULE_TYPE      "event-schedule-type"
 146 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
 147 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
 148 #define CMD_LINE_OPT_REASSEMBLE         "reassemble"
 149 #define CMD_LINE_OPT_MTU                "mtu"
 150 #define CMD_LINE_OPT_FRAG_TTL           "frag-ttl"
 151
 152 #define CMD_LINE_ARG_EVENT      "event"
 153 #define CMD_LINE_ARG_POLL       "poll"
 154 #define CMD_LINE_ARG_ORDERED    "ordered"
 155 #define CMD_LINE_ARG_ATOMIC     "atomic"
 156 #define CMD_LINE_ARG_PARALLEL   "parallel"
 157
 158 enum {
 159         /* long options mapped to a short option */
 160
 161         /* first long only option value must be >= 256, so that we won't
 162          * conflict with short options
 163          */
 164         CMD_LINE_OPT_MIN_NUM = 256,
 165         CMD_LINE_OPT_CONFIG_NUM,
 166         CMD_LINE_OPT_SINGLE_SA_NUM,
 167         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 168         CMD_LINE_OPT_TRANSFER_MODE_NUM,
 169         CMD_LINE_OPT_SCHEDULE_TYPE_NUM,
 170         CMD_LINE_OPT_RX_OFFLOAD_NUM,
 171         CMD_LINE_OPT_TX_OFFLOAD_NUM,
 172         CMD_LINE_OPT_REASSEMBLE_NUM,
 173         CMD_LINE_OPT_MTU_NUM,
 174         CMD_LINE_OPT_FRAG_TTL_NUM,
 175 };
 176
 177 static const struct option lgopts[] = {
 178         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 179         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 180         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 181         {CMD_LINE_OPT_TRANSFER_MODE, 1, 0, CMD_LINE_OPT_TRANSFER_MODE_NUM},
 182         {CMD_LINE_OPT_SCHEDULE_TYPE, 1, 0, CMD_LINE_OPT_SCHEDULE_TYPE_NUM},
 183         {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
 184         {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
 185         {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
 186         {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
 187         {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
 188         {NULL, 0, 0, 0}
 189 };
 190
 191 /* mask of enabled ports */
 192 static uint32_t enabled_port_mask;
 193 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 194 static uint32_t unprotected_port_mask;
 195 static int32_t promiscuous_on = 1;
 196 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 197 static uint32_t nb_lcores;
 198 static uint32_t single_sa;
 199 static uint32_t single_sa_idx;
 200
 201 /*
 202  * RX/TX HW offload capabilities to enable/use on ethernet ports.
 203  * By default all capabilities are enabled.
 204  */
 205 static uint64_t dev_rx_offload = UINT64_MAX;
 206 static uint64_t dev_tx_offload = UINT64_MAX;
 207
 208 /*
 209  * global values that determine multi-seg policy
 210  */
 211 static uint32_t frag_tbl_sz;
 212 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
 213 static uint32_t mtu_size = RTE_ETHER_MTU;
 214 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
 215
 216 /* application wide librte_ipsec/SA parameters */
 217 struct app_sa_prm app_sa_prm = {
 218                         .enable = 0,
 219                         .cache_sz = SA_CACHE_SZ
 220                 };
 221 static const char *cfgfile;
 222
 223 struct lcore_rx_queue {
 224         uint16_t port_id;
 225         uint8_t queue_id;
 226 } __rte_cache_aligned;
 227
 228 struct lcore_params {
 229         uint16_t port_id;
 230         uint8_t queue_id;
 231         uint8_t lcore_id;
 232 } __rte_cache_aligned;
 233
 234 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 235
 236 static struct lcore_params *lcore_params;
 237 static uint16_t nb_lcore_params;
 238
 239 static struct rte_hash *cdev_map_in;
 240 static struct rte_hash *cdev_map_out;
 241
 242 struct buffer {
 243         uint16_t len;
 244         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 245 };
 246
 247 struct lcore_conf {
 248         uint16_t nb_rx_queue;
 249         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 250         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 251         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 252         struct ipsec_ctx inbound;
 253         struct ipsec_ctx outbound;
 254         struct rt_ctx *rt4_ctx;
 255         struct rt_ctx *rt6_ctx;
 256         struct {
 257                 struct rte_ip_frag_tbl *tbl;
 258                 struct rte_mempool *pool_dir;
 259                 struct rte_mempool *pool_indir;
 260                 struct rte_ip_frag_death_row dr;
 261         } frag;
 262 } __rte_cache_aligned;
 263
 264 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 265
 266 static struct rte_eth_conf port_conf = {
 267         .rxmode = {
 268                 .mq_mode        = ETH_MQ_RX_RSS,
 269                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 270                 .split_hdr_size = 0,
 271                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 272         },
 273         .rx_adv_conf = {
 274                 .rss_conf = {
 275                         .rss_key = NULL,
 276                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 277                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 278                 },
 279         },
 280         .txmode = {
 281                 .mq_mode = ETH_MQ_TX_NONE,
 282         },
 283 };
 284
 285 static struct socket_ctx socket_ctx[NB_SOCKETS];
 286
 287 /*
 288  * Determine is multi-segment support required:
 289  *  - either frame buffer size is smaller then mtu
 290  *  - or reassmeble support is requested
 291  */
 292 static int
 293 multi_seg_required(void)
 294 {
 295         return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
 296                 frame_buf_size || frag_tbl_sz != 0);
 297 }
 298
 299 static inline void
 300 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
 301         uint32_t l2_len)
 302 {
 303         uint32_t plen, trim;
 304
 305         plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
 306         if (plen < m->pkt_len) {
 307                 trim = m->pkt_len - plen;
 308                 rte_pktmbuf_trim(m, trim);
 309         }
 310 }
 311
 312 static inline void
 313 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
 314         uint32_t l2_len)
 315 {
 316         uint32_t plen, trim;
 317
 318         plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
 319         if (plen < m->pkt_len) {
 320                 trim = m->pkt_len - plen;
 321                 rte_pktmbuf_trim(m, trim);
 322         }
 323 }
 324
 325 static inline void
 326 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 327 {
 328         const struct rte_ether_hdr *eth;
 329         const struct rte_ipv4_hdr *iph4;
 330         const struct rte_ipv6_hdr *iph6;
 331
 332         eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
 333         if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 334
 335                 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
 336                         RTE_ETHER_HDR_LEN);
 337                 adjust_ipv4_pktlen(pkt, iph4, 0);
 338
 339                 if (iph4->next_proto_id == IPPROTO_ESP)
 340                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 341                 else {
 342                         t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
 343                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 344                 }
 345                 pkt->l2_len = 0;
 346                 pkt->l3_len = sizeof(*iph4);
 347                 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
 348         } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 349                 int next_proto;
 350                 size_t l3len, ext_len;
 351                 uint8_t *p;
 352
 353                 /* get protocol type */
 354                 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
 355                         RTE_ETHER_HDR_LEN);
 356                 adjust_ipv6_pktlen(pkt, iph6, 0);
 357
 358                 next_proto = iph6->proto;
 359
 360                 /* determine l3 header size up to ESP extension */
 361                 l3len = sizeof(struct ip6_hdr);
 362                 p = rte_pktmbuf_mtod(pkt, uint8_t *);
 363                 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
 364                         (next_proto = rte_ipv6_get_next_ext(p + l3len,
 365                                                 next_proto, &ext_len)) >= 0)
 366                         l3len += ext_len;
 367
 368                 /* drop packet when IPv6 header exceeds first segment length */
 369                 if (unlikely(l3len > pkt->data_len)) {
 370                         rte_pktmbuf_free(pkt);
 371                         return;
 372                 }
 373
 374                 if (next_proto == IPPROTO_ESP)
 375                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 376                 else {
 377                         t->ip6.data[t->ip6.num] = &iph6->proto;
 378                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 379                 }
 380                 pkt->l2_len = 0;
 381                 pkt->l3_len = l3len;
 382                 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
 383         } else {
 384                 /* Unknown/Unsupported type, drop the packet */
 385                 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
 386                         rte_be_to_cpu_16(eth->ether_type));
 387                 rte_pktmbuf_free(pkt);
 388                 return;
 389         }
 390
 391         /* Check if the packet has been processed inline. For inline protocol
 392          * processed packets, the metadata in the mbuf can be used to identify
 393          * the security processing done on the packet. The metadata will be
 394          * used to retrieve the application registered userdata associated
 395          * with the security session.
 396          */
 397
 398         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 399                 struct ipsec_sa *sa;
 400                 struct ipsec_mbuf_metadata *priv;
 401                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 402                                                 rte_eth_dev_get_sec_ctx(
 403                                                 pkt->port);
 404
 405                 /* Retrieve the userdata registered. Here, the userdata
 406                  * registered is the SA pointer.
 407                  */
 408
 409                 sa = (struct ipsec_sa *)
 410                                 rte_security_get_userdata(ctx, pkt->udata64);
 411
 412                 if (sa == NULL) {
 413                         /* userdata could not be retrieved */
 414                         return;
 415                 }
 416
 417                 /* Save SA as priv member in mbuf. This will be used in the
 418                  * IPsec selector(SP-SA) check.
 419                  */
 420
 421                 priv = get_priv(pkt);
 422                 priv->sa = sa;
 423         }
 424 }
 425
 426 static inline void
 427 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 428                 uint16_t nb_pkts)
 429 {
 430         int32_t i;
 431
 432         t->ipsec.num = 0;
 433         t->ip4.num = 0;
 434         t->ip6.num = 0;
 435
 436         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 437                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 438                                         void *));
 439                 prepare_one_packet(pkts[i], t);
 440         }
 441         /* Process left packets */
 442         for (; i < nb_pkts; i++)
 443                 prepare_one_packet(pkts[i], t);
 444 }
 445
 446 static inline void
 447 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 448                 const struct lcore_conf *qconf)
 449 {
 450         struct ip *ip;
 451         struct rte_ether_hdr *ethhdr;
 452
 453         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 454
 455         ethhdr = (struct rte_ether_hdr *)
 456                 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
 457
 458         if (ip->ip_v == IPVERSION) {
 459                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 460                 pkt->l3_len = sizeof(struct ip);
 461                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 462
 463                 ip->ip_sum = 0;
 464
 465                 /* calculate IPv4 cksum in SW */
 466                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 467                         ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
 468
 469                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
 470         } else {
 471                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 472                 pkt->l3_len = sizeof(struct ip6_hdr);
 473                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 474
 475                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
 476         }
 477
 478         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 479                         sizeof(struct rte_ether_addr));
 480         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 481                         sizeof(struct rte_ether_addr));
 482 }
 483
 484 static inline void
 485 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 486                 const struct lcore_conf *qconf)
 487 {
 488         int32_t i;
 489         const int32_t prefetch_offset = 2;
 490
 491         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 492                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 493                 prepare_tx_pkt(pkts[i], port, qconf);
 494         }
 495         /* Process left packets */
 496         for (; i < nb_pkts; i++)
 497                 prepare_tx_pkt(pkts[i], port, qconf);
 498 }
 499
 500 /* Send burst of packets on an output interface */
 501 static inline int32_t
 502 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 503 {
 504         struct rte_mbuf **m_table;
 505         int32_t ret;
 506         uint16_t queueid;
 507
 508         queueid = qconf->tx_queue_id[port];
 509         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 510
 511         prepare_tx_burst(m_table, n, port, qconf);
 512
 513         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 514         if (unlikely(ret < n)) {
 515                 do {
 516                         rte_pktmbuf_free(m_table[ret]);
 517                 } while (++ret < n);
 518         }
 519
 520         return 0;
 521 }
 522
 523 /*
 524  * Helper function to fragment and queue for TX one packet.
 525  */
 526 static inline uint32_t
 527 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
 528         uint16_t port, uint8_t proto)
 529 {
 530         struct buffer *tbl;
 531         uint32_t len, n;
 532         int32_t rc;
 533
 534         tbl =  qconf->tx_mbufs + port;
 535         len = tbl->len;
 536
 537         /* free space for new fragments */
 538         if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >=  RTE_DIM(tbl->m_table)) {
 539                 send_burst(qconf, len, port);
 540                 len = 0;
 541         }
 542
 543         n = RTE_DIM(tbl->m_table) - len;
 544
 545         if (proto == IPPROTO_IP)
 546                 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
 547                         n, mtu_size, qconf->frag.pool_dir,
 548                         qconf->frag.pool_indir);
 549         else
 550                 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
 551                         n, mtu_size, qconf->frag.pool_dir,
 552                         qconf->frag.pool_indir);
 553
 554         if (rc >= 0)
 555                 len += rc;
 556         else
 557                 RTE_LOG(ERR, IPSEC,
 558                         "%s: failed to fragment packet with size %u, "
 559                         "error code: %d\n",
 560                         __func__, m->pkt_len, rte_errno);
 561
 562         rte_pktmbuf_free(m);
 563         return len;
 564 }
 565
 566 /* Enqueue a single packet, and send burst if queue is filled */
 567 static inline int32_t
 568 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
 569 {
 570         uint32_t lcore_id;
 571         uint16_t len;
 572         struct lcore_conf *qconf;
 573
 574         lcore_id = rte_lcore_id();
 575
 576         qconf = &lcore_conf[lcore_id];
 577         len = qconf->tx_mbufs[port].len;
 578
 579         if (m->pkt_len <= mtu_size) {
 580                 qconf->tx_mbufs[port].m_table[len] = m;
 581                 len++;
 582
 583         /* need to fragment the packet */
 584         } else if (frag_tbl_sz > 0)
 585                 len = send_fragment_packet(qconf, m, port, proto);
 586         else
 587                 rte_pktmbuf_free(m);
 588
 589         /* enough pkts to be sent */
 590         if (unlikely(len == MAX_PKT_BURST)) {
 591                 send_burst(qconf, MAX_PKT_BURST, port);
 592                 len = 0;
 593         }
 594
 595         qconf->tx_mbufs[port].len = len;
 596         return 0;
 597 }
 598
 599 static inline void
 600 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 601                 uint16_t lim)
 602 {
 603         struct rte_mbuf *m;
 604         uint32_t i, j, res, sa_idx;
 605
 606         if (ip->num == 0 || sp == NULL)
 607                 return;
 608
 609         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 610                         ip->num, DEFAULT_MAX_CATEGORIES);
 611
 612         j = 0;
 613         for (i = 0; i < ip->num; i++) {
 614                 m = ip->pkts[i];
 615                 res = ip->res[i];
 616                 if (res == BYPASS) {
 617                         ip->pkts[j++] = m;
 618                         continue;
 619                 }
 620                 if (res == DISCARD) {
 621                         rte_pktmbuf_free(m);
 622                         continue;
 623                 }
 624
 625                 /* Only check SPI match for processed IPSec packets */
 626                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 627                         rte_pktmbuf_free(m);
 628                         continue;
 629                 }
 630
 631                 sa_idx = res - 1;
 632                 if (!inbound_sa_check(sa, m, sa_idx)) {
 633                         rte_pktmbuf_free(m);
 634                         continue;
 635                 }
 636                 ip->pkts[j++] = m;
 637         }
 638         ip->num = j;
 639 }
 640
 641 static void
 642 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 643 {
 644         uint32_t i, n4, n6;
 645         struct ip *ip;
 646         struct rte_mbuf *m;
 647
 648         n4 = trf->ip4.num;
 649         n6 = trf->ip6.num;
 650
 651         for (i = 0; i < num; i++) {
 652
 653                 m = mb[i];
 654                 ip = rte_pktmbuf_mtod(m, struct ip *);
 655
 656                 if (ip->ip_v == IPVERSION) {
 657                         trf->ip4.pkts[n4] = m;
 658                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 659                                         uint8_t *, offsetof(struct ip, ip_p));
 660                         n4++;
 661                 } else if (ip->ip_v == IP6_VERSION) {
 662                         trf->ip6.pkts[n6] = m;
 663                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 664                                         uint8_t *,
 665                                         offsetof(struct ip6_hdr, ip6_nxt));
 666                         n6++;
 667                 } else
 668                         rte_pktmbuf_free(m);
 669         }
 670
 671         trf->ip4.num = n4;
 672         trf->ip6.num = n6;
 673 }
 674
 675
 676 static inline void
 677 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 678                 struct ipsec_traffic *traffic)
 679 {
 680         uint16_t nb_pkts_in, n_ip4, n_ip6;
 681
 682         n_ip4 = traffic->ip4.num;
 683         n_ip6 = traffic->ip6.num;
 684
 685         if (app_sa_prm.enable == 0) {
 686                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 687                                 traffic->ipsec.num, MAX_PKT_BURST);
 688                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 689         } else {
 690                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 691                         traffic->ipsec.saptr, traffic->ipsec.num);
 692                 ipsec_process(ipsec_ctx, traffic);
 693         }
 694
 695         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 696                         n_ip4);
 697
 698         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 699                         n_ip6);
 700 }
 701
 702 static inline void
 703 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 704                 struct traffic_type *ipsec)
 705 {
 706         struct rte_mbuf *m;
 707         uint32_t i, j, sa_idx;
 708
 709         if (ip->num == 0 || sp == NULL)
 710                 return;
 711
 712         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 713                         ip->num, DEFAULT_MAX_CATEGORIES);
 714
 715         j = 0;
 716         for (i = 0; i < ip->num; i++) {
 717                 m = ip->pkts[i];
 718                 sa_idx = ip->res[i] - 1;
 719                 if (ip->res[i] == DISCARD)
 720                         rte_pktmbuf_free(m);
 721                 else if (ip->res[i] == BYPASS)
 722                         ip->pkts[j++] = m;
 723                 else {
 724                         ipsec->res[ipsec->num] = sa_idx;
 725                         ipsec->pkts[ipsec->num++] = m;
 726                 }
 727         }
 728         ip->num = j;
 729 }
 730
 731 static inline void
 732 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 733                 struct ipsec_traffic *traffic)
 734 {
 735         struct rte_mbuf *m;
 736         uint16_t idx, nb_pkts_out, i;
 737
 738         /* Drop any IPsec traffic from protected ports */
 739         for (i = 0; i < traffic->ipsec.num; i++)
 740                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 741
 742         traffic->ipsec.num = 0;
 743
 744         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 745
 746         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 747
 748         if (app_sa_prm.enable == 0) {
 749
 750                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 751                                 traffic->ipsec.res, traffic->ipsec.num,
 752                                 MAX_PKT_BURST);
 753
 754                 for (i = 0; i < nb_pkts_out; i++) {
 755                         m = traffic->ipsec.pkts[i];
 756                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 757                         if (ip->ip_v == IPVERSION) {
 758                                 idx = traffic->ip4.num++;
 759                                 traffic->ip4.pkts[idx] = m;
 760                         } else {
 761                                 idx = traffic->ip6.num++;
 762                                 traffic->ip6.pkts[idx] = m;
 763                         }
 764                 }
 765         } else {
 766                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 767                         traffic->ipsec.saptr, traffic->ipsec.num);
 768                 ipsec_process(ipsec_ctx, traffic);
 769         }
 770 }
 771
 772 static inline void
 773 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 774                 struct ipsec_traffic *traffic)
 775 {
 776         struct rte_mbuf *m;
 777         uint32_t nb_pkts_in, i, idx;
 778
 779         /* Drop any IPv4 traffic from unprotected ports */
 780         for (i = 0; i < traffic->ip4.num; i++)
 781                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 782
 783         traffic->ip4.num = 0;
 784
 785         /* Drop any IPv6 traffic from unprotected ports */
 786         for (i = 0; i < traffic->ip6.num; i++)
 787                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 788
 789         traffic->ip6.num = 0;
 790
 791         if (app_sa_prm.enable == 0) {
 792
 793                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 794                                 traffic->ipsec.num, MAX_PKT_BURST);
 795
 796                 for (i = 0; i < nb_pkts_in; i++) {
 797                         m = traffic->ipsec.pkts[i];
 798                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 799                         if (ip->ip_v == IPVERSION) {
 800                                 idx = traffic->ip4.num++;
 801                                 traffic->ip4.pkts[idx] = m;
 802                         } else {
 803                                 idx = traffic->ip6.num++;
 804                                 traffic->ip6.pkts[idx] = m;
 805                         }
 806                 }
 807         } else {
 808                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 809                         traffic->ipsec.saptr, traffic->ipsec.num);
 810                 ipsec_process(ipsec_ctx, traffic);
 811         }
 812 }
 813
 814 static inline void
 815 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 816                 struct ipsec_traffic *traffic)
 817 {
 818         struct rte_mbuf *m;
 819         uint32_t nb_pkts_out, i, n;
 820         struct ip *ip;
 821
 822         /* Drop any IPsec traffic from protected ports */
 823         for (i = 0; i < traffic->ipsec.num; i++)
 824                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 825
 826         n = 0;
 827
 828         for (i = 0; i < traffic->ip4.num; i++) {
 829                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 830                 traffic->ipsec.res[n++] = single_sa_idx;
 831         }
 832
 833         for (i = 0; i < traffic->ip6.num; i++) {
 834                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 835                 traffic->ipsec.res[n++] = single_sa_idx;
 836         }
 837
 838         traffic->ip4.num = 0;
 839         traffic->ip6.num = 0;
 840         traffic->ipsec.num = n;
 841
 842         if (app_sa_prm.enable == 0) {
 843
 844                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 845                                 traffic->ipsec.res, traffic->ipsec.num,
 846                                 MAX_PKT_BURST);
 847
 848                 /* They all sue the same SA (ip4 or ip6 tunnel) */
 849                 m = traffic->ipsec.pkts[0];
 850                 ip = rte_pktmbuf_mtod(m, struct ip *);
 851                 if (ip->ip_v == IPVERSION) {
 852                         traffic->ip4.num = nb_pkts_out;
 853                         for (i = 0; i < nb_pkts_out; i++)
 854                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 855                 } else {
 856                         traffic->ip6.num = nb_pkts_out;
 857                         for (i = 0; i < nb_pkts_out; i++)
 858                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 859                 }
 860         } else {
 861                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 862                         traffic->ipsec.saptr, traffic->ipsec.num);
 863                 ipsec_process(ipsec_ctx, traffic);
 864         }
 865 }
 866
 867 static inline int32_t
 868 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 869 {
 870         struct ipsec_mbuf_metadata *priv;
 871         struct ipsec_sa *sa;
 872
 873         priv = get_priv(pkt);
 874
 875         sa = priv->sa;
 876         if (unlikely(sa == NULL)) {
 877                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 878                 goto fail;
 879         }
 880
 881         if (is_ipv6)
 882                 return sa->portid;
 883
 884         /* else */
 885         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 886
 887 fail:
 888         if (is_ipv6)
 889                 return -1;
 890
 891         /* else */
 892         return 0;
 893 }
 894
 895 static inline void
 896 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 897 {
 898         uint32_t hop[MAX_PKT_BURST * 2];
 899         uint32_t dst_ip[MAX_PKT_BURST * 2];
 900         int32_t pkt_hop = 0;
 901         uint16_t i, offset;
 902         uint16_t lpm_pkts = 0;
 903
 904         if (nb_pkts == 0)
 905                 return;
 906
 907         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 908          * have port ID in the SA
 909          */
 910
 911         for (i = 0; i < nb_pkts; i++) {
 912                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 913                         /* Security offload not enabled. So an LPM lookup is
 914                          * required to get the hop
 915                          */
 916                         offset = offsetof(struct ip, ip_dst);
 917                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 918                                         uint32_t *, offset);
 919                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 920                         lpm_pkts++;
 921                 }
 922         }
 923
 924         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 925
 926         lpm_pkts = 0;
 927
 928         for (i = 0; i < nb_pkts; i++) {
 929                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 930                         /* Read hop from the SA */
 931                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 932                 } else {
 933                         /* Need to use hop returned by lookup */
 934                         pkt_hop = hop[lpm_pkts++];
 935                 }
 936
 937                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 938                         rte_pktmbuf_free(pkts[i]);
 939                         continue;
 940                 }
 941                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
 942         }
 943 }
 944
 945 static inline void
 946 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 947 {
 948         int32_t hop[MAX_PKT_BURST * 2];
 949         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 950         uint8_t *ip6_dst;
 951         int32_t pkt_hop = 0;
 952         uint16_t i, offset;
 953         uint16_t lpm_pkts = 0;
 954
 955         if (nb_pkts == 0)
 956                 return;
 957
 958         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 959          * have port ID in the SA
 960          */
 961
 962         for (i = 0; i < nb_pkts; i++) {
 963                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 964                         /* Security offload not enabled. So an LPM lookup is
 965                          * required to get the hop
 966                          */
 967                         offset = offsetof(struct ip6_hdr, ip6_dst);
 968                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 969                                         offset);
 970                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 971                         lpm_pkts++;
 972                 }
 973         }
 974
 975         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 976                         lpm_pkts);
 977
 978         lpm_pkts = 0;
 979
 980         for (i = 0; i < nb_pkts; i++) {
 981                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 982                         /* Read hop from the SA */
 983                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 984                 } else {
 985                         /* Need to use hop returned by lookup */
 986                         pkt_hop = hop[lpm_pkts++];
 987                 }
 988
 989                 if (pkt_hop == -1) {
 990                         rte_pktmbuf_free(pkts[i]);
 991                         continue;
 992                 }
 993                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
 994         }
 995 }
 996
 997 static inline void
 998 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 999                 uint8_t nb_pkts, uint16_t portid)
1000 {
1001         struct ipsec_traffic traffic;
1002
1003         prepare_traffic(pkts, &traffic, nb_pkts);
1004
1005         if (unlikely(single_sa)) {
1006                 if (UNPROTECTED_PORT(portid))
1007                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
1008                 else
1009                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
1010         } else {
1011                 if (UNPROTECTED_PORT(portid))
1012                         process_pkts_inbound(&qconf->inbound, &traffic);
1013                 else
1014                         process_pkts_outbound(&qconf->outbound, &traffic);
1015         }
1016
1017         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
1018         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
1019 }
1020
1021 static inline void
1022 drain_tx_buffers(struct lcore_conf *qconf)
1023 {
1024         struct buffer *buf;
1025         uint32_t portid;
1026
1027         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
1028                 buf = &qconf->tx_mbufs[portid];
1029                 if (buf->len == 0)
1030                         continue;
1031                 send_burst(qconf, buf->len, portid);
1032                 buf->len = 0;
1033         }
1034 }
1035
1036 static inline void
1037 drain_crypto_buffers(struct lcore_conf *qconf)
1038 {
1039         uint32_t i;
1040         struct ipsec_ctx *ctx;
1041
1042         /* drain inbound buffers*/
1043         ctx = &qconf->inbound;
1044         for (i = 0; i != ctx->nb_qps; i++) {
1045                 if (ctx->tbl[i].len != 0)
1046                         enqueue_cop_burst(ctx->tbl  + i);
1047         }
1048
1049         /* drain outbound buffers*/
1050         ctx = &qconf->outbound;
1051         for (i = 0; i != ctx->nb_qps; i++) {
1052                 if (ctx->tbl[i].len != 0)
1053                         enqueue_cop_burst(ctx->tbl  + i);
1054         }
1055 }
1056
1057 static void
1058 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1059                 struct ipsec_ctx *ctx)
1060 {
1061         uint32_t n;
1062         struct ipsec_traffic trf;
1063
1064         if (app_sa_prm.enable == 0) {
1065
1066                 /* dequeue packets from crypto-queue */
1067                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1068                         RTE_DIM(trf.ipsec.pkts));
1069
1070                 trf.ip4.num = 0;
1071                 trf.ip6.num = 0;
1072
1073                 /* split traffic by ipv4-ipv6 */
1074                 split46_traffic(&trf, trf.ipsec.pkts, n);
1075         } else
1076                 ipsec_cqp_process(ctx, &trf);
1077
1078         /* process ipv4 packets */
1079         if (trf.ip4.num != 0) {
1080                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1081                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1082         }
1083
1084         /* process ipv6 packets */
1085         if (trf.ip6.num != 0) {
1086                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1087                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1088         }
1089 }
1090
1091 static void
1092 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1093                 struct ipsec_ctx *ctx)
1094 {
1095         uint32_t n;
1096         struct ipsec_traffic trf;
1097
1098         if (app_sa_prm.enable == 0) {
1099
1100                 /* dequeue packets from crypto-queue */
1101                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1102                         RTE_DIM(trf.ipsec.pkts));
1103
1104                 trf.ip4.num = 0;
1105                 trf.ip6.num = 0;
1106
1107                 /* split traffic by ipv4-ipv6 */
1108                 split46_traffic(&trf, trf.ipsec.pkts, n);
1109         } else
1110                 ipsec_cqp_process(ctx, &trf);
1111
1112         /* process ipv4 packets */
1113         if (trf.ip4.num != 0)
1114                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1115
1116         /* process ipv6 packets */
1117         if (trf.ip6.num != 0)
1118                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1119 }
1120
1121 /* main processing loop */
1122 static int32_t
1123 main_loop(__attribute__((unused)) void *dummy)
1124 {
1125         struct rte_mbuf *pkts[MAX_PKT_BURST];
1126         uint32_t lcore_id;
1127         uint64_t prev_tsc, diff_tsc, cur_tsc;
1128         int32_t i, nb_rx;
1129         uint16_t portid;
1130         uint8_t queueid;
1131         struct lcore_conf *qconf;
1132         int32_t rc, socket_id;
1133         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1134                         / US_PER_S * BURST_TX_DRAIN_US;
1135         struct lcore_rx_queue *rxql;
1136
1137         prev_tsc = 0;
1138         lcore_id = rte_lcore_id();
1139         qconf = &lcore_conf[lcore_id];
1140         rxql = qconf->rx_queue_list;
1141         socket_id = rte_lcore_to_socket_id(lcore_id);
1142
1143         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1144         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1145         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1146         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1147         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1148         qconf->inbound.cdev_map = cdev_map_in;
1149         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1150         qconf->inbound.session_priv_pool =
1151                         socket_ctx[socket_id].session_priv_pool;
1152         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1153         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1154         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1155         qconf->outbound.cdev_map = cdev_map_out;
1156         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1157         qconf->outbound.session_priv_pool =
1158                         socket_ctx[socket_id].session_priv_pool;
1159         qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1160         qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1161
1162         rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1163         if (rc != 0) {
1164                 RTE_LOG(ERR, IPSEC,
1165                         "SAD cache init on lcore %u, failed with code: %d\n",
1166                         lcore_id, rc);
1167                 return rc;
1168         }
1169
1170         if (qconf->nb_rx_queue == 0) {
1171                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1172                         lcore_id);
1173                 return 0;
1174         }
1175
1176         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1177
1178         for (i = 0; i < qconf->nb_rx_queue; i++) {
1179                 portid = rxql[i].port_id;
1180                 queueid = rxql[i].queue_id;
1181                 RTE_LOG(INFO, IPSEC,
1182                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1183                         lcore_id, portid, queueid);
1184         }
1185
1186         while (1) {
1187                 cur_tsc = rte_rdtsc();
1188
1189                 /* TX queue buffer drain */
1190                 diff_tsc = cur_tsc - prev_tsc;
1191
1192                 if (unlikely(diff_tsc > drain_tsc)) {
1193                         drain_tx_buffers(qconf);
1194                         drain_crypto_buffers(qconf);
1195                         prev_tsc = cur_tsc;
1196                 }
1197
1198                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1199
1200                         /* Read packets from RX queues */
1201                         portid = rxql[i].port_id;
1202                         queueid = rxql[i].queue_id;
1203                         nb_rx = rte_eth_rx_burst(portid, queueid,
1204                                         pkts, MAX_PKT_BURST);
1205
1206                         if (nb_rx > 0)
1207                                 process_pkts(qconf, pkts, nb_rx, portid);
1208
1209                         /* dequeue and process completed crypto-ops */
1210                         if (UNPROTECTED_PORT(portid))
1211                                 drain_inbound_crypto_queues(qconf,
1212                                         &qconf->inbound);
1213                         else
1214                                 drain_outbound_crypto_queues(qconf,
1215                                         &qconf->outbound);
1216                 }
1217         }
1218 }
1219
1220 static int32_t
1221 check_poll_mode_params(struct eh_conf *eh_conf)
1222 {
1223         uint8_t lcore;
1224         uint16_t portid;
1225         uint16_t i;
1226         int32_t socket_id;
1227
1228         if (!eh_conf)
1229                 return -EINVAL;
1230
1231         if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1232                 return 0;
1233
1234         if (lcore_params == NULL) {
1235                 printf("Error: No port/queue/core mappings\n");
1236                 return -1;
1237         }
1238
1239         for (i = 0; i < nb_lcore_params; ++i) {
1240                 lcore = lcore_params[i].lcore_id;
1241                 if (!rte_lcore_is_enabled(lcore)) {
1242                         printf("error: lcore %hhu is not enabled in "
1243                                 "lcore mask\n", lcore);
1244                         return -1;
1245                 }
1246                 socket_id = rte_lcore_to_socket_id(lcore);
1247                 if (socket_id != 0 && numa_on == 0) {
1248                         printf("warning: lcore %hhu is on socket %d "
1249                                 "with numa off\n",
1250                                 lcore, socket_id);
1251                 }
1252                 portid = lcore_params[i].port_id;
1253                 if ((enabled_port_mask & (1 << portid)) == 0) {
1254                         printf("port %u is not enabled in port mask\n", portid);
1255                         return -1;
1256                 }
1257                 if (!rte_eth_dev_is_valid_port(portid)) {
1258                         printf("port %u is not present on the board\n", portid);
1259                         return -1;
1260                 }
1261         }
1262         return 0;
1263 }
1264
1265 static uint8_t
1266 get_port_nb_rx_queues(const uint16_t port)
1267 {
1268         int32_t queue = -1;
1269         uint16_t i;
1270
1271         for (i = 0; i < nb_lcore_params; ++i) {
1272                 if (lcore_params[i].port_id == port &&
1273                                 lcore_params[i].queue_id > queue)
1274                         queue = lcore_params[i].queue_id;
1275         }
1276         return (uint8_t)(++queue);
1277 }
1278
1279 static int32_t
1280 init_lcore_rx_queues(void)
1281 {
1282         uint16_t i, nb_rx_queue;
1283         uint8_t lcore;
1284
1285         for (i = 0; i < nb_lcore_params; ++i) {
1286                 lcore = lcore_params[i].lcore_id;
1287                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1288                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1289                         printf("error: too many queues (%u) for lcore: %u\n",
1290                                         nb_rx_queue + 1, lcore);
1291                         return -1;
1292                 }
1293                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1294                         lcore_params[i].port_id;
1295                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1296                         lcore_params[i].queue_id;
1297                 lcore_conf[lcore].nb_rx_queue++;
1298         }
1299         return 0;
1300 }
1301
1302 /* display usage */
1303 static void
1304 print_usage(const char *prgname)
1305 {
1306         fprintf(stderr, "%s [EAL options] --"
1307                 " -p PORTMASK"
1308                 " [-P]"
1309                 " [-u PORTMASK]"
1310                 " [-j FRAMESIZE]"
1311                 " [-l]"
1312                 " [-w REPLAY_WINDOW_SIZE]"
1313                 " [-e]"
1314                 " [-a]"
1315                 " [-c]"
1316                 " -f CONFIG_FILE"
1317                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1318                 " [--single-sa SAIDX]"
1319                 " [--cryptodev_mask MASK]"
1320                 " [--transfer-mode MODE]"
1321                 " [--event-schedule-type TYPE]"
1322                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1323                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1324                 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1325                 " [--" CMD_LINE_OPT_MTU " MTU]"
1326                 "\n\n"
1327                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1328                 "  -P : Enable promiscuous mode\n"
1329                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1330                 "  -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1331                 "     value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1332                 "  -l enables code-path that uses librte_ipsec\n"
1333                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1334                 "     size for each SA\n"
1335                 "  -e enables ESN\n"
1336                 "  -a enables SA SQN atomic behaviour\n"
1337                 "  -c specifies inbound SAD cache size,\n"
1338                 "     zero value disables the cache (default value: 128)\n"
1339                 "  -f CONFIG_FILE: Configuration file\n"
1340                 "  --config (port,queue,lcore): Rx queue configuration. In poll\n"
1341                 "                               mode determines which queues from\n"
1342                 "                               which ports are mapped to which cores.\n"
1343                 "                               In event mode this option is not used\n"
1344                 "                               as packets are dynamically scheduled\n"
1345                 "                               to cores by HW.\n"
1346                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1347                 "                     bypassing the SP\n"
1348                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1349                 "                         devices to configure\n"
1350                 "  --transfer-mode MODE\n"
1351                 "               \"poll\"  : Packet transfer via polling (default)\n"
1352                 "               \"event\" : Packet transfer via event device\n"
1353                 "  --event-schedule-type TYPE queue schedule type, used only when\n"
1354                 "                             transfer mode is set to event\n"
1355                 "               \"ordered\"  : Ordered (default)\n"
1356                 "               \"atomic\"   : Atomic\n"
1357                 "               \"parallel\" : Parallel\n"
1358                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1359                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1360                 "                         (DEV_RX_OFFLOAD_*)\n"
1361                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1362                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1363                 "                         (DEV_TX_OFFLOAD_*)\n"
1364                 "  --" CMD_LINE_OPT_REASSEMBLE " NUM"
1365                 ": max number of entries in reassemble(fragment) table\n"
1366                 "    (zero (default value) disables reassembly)\n"
1367                 "  --" CMD_LINE_OPT_MTU " MTU"
1368                 ": MTU value on all ports (default value: 1500)\n"
1369                 "    outgoing packets with bigger size will be fragmented\n"
1370                 "    incoming packets with bigger size will be discarded\n"
1371                 "  --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1372                 ": fragments lifetime in nanoseconds, default\n"
1373                 "    and maximum value is 10.000.000.000 ns (10 s)\n"
1374                 "\n",
1375                 prgname);
1376 }
1377
1378 static int
1379 parse_mask(const char *str, uint64_t *val)
1380 {
1381         char *end;
1382         unsigned long t;
1383
1384         errno = 0;
1385         t = strtoul(str, &end, 0);
1386         if (errno != 0 || end[0] != 0)
1387                 return -EINVAL;
1388
1389         *val = t;
1390         return 0;
1391 }
1392
1393 static int32_t
1394 parse_portmask(const char *portmask)
1395 {
1396         char *end = NULL;
1397         unsigned long pm;
1398
1399         /* parse hexadecimal string */
1400         pm = strtoul(portmask, &end, 16);
1401         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1402                 return -1;
1403
1404         if ((pm == 0) && errno)
1405                 return -1;
1406
1407         return pm;
1408 }
1409
1410 static int64_t
1411 parse_decimal(const char *str)
1412 {
1413         char *end = NULL;
1414         uint64_t num;
1415
1416         num = strtoull(str, &end, 10);
1417         if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1418                 || num > INT64_MAX)
1419                 return -1;
1420
1421         return num;
1422 }
1423
1424 static int32_t
1425 parse_config(const char *q_arg)
1426 {
1427         char s[256];
1428         const char *p, *p0 = q_arg;
1429         char *end;
1430         enum fieldnames {
1431                 FLD_PORT = 0,
1432                 FLD_QUEUE,
1433                 FLD_LCORE,
1434                 _NUM_FLD
1435         };
1436         unsigned long int_fld[_NUM_FLD];
1437         char *str_fld[_NUM_FLD];
1438         int32_t i;
1439         uint32_t size;
1440
1441         nb_lcore_params = 0;
1442
1443         while ((p = strchr(p0, '(')) != NULL) {
1444                 ++p;
1445                 p0 = strchr(p, ')');
1446                 if (p0 == NULL)
1447                         return -1;
1448
1449                 size = p0 - p;
1450                 if (size >= sizeof(s))
1451                         return -1;
1452
1453                 snprintf(s, sizeof(s), "%.*s", size, p);
1454                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1455                                 _NUM_FLD)
1456                         return -1;
1457                 for (i = 0; i < _NUM_FLD; i++) {
1458                         errno = 0;
1459                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1460                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1461                                 return -1;
1462                 }
1463                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1464                         printf("exceeded max number of lcore params: %hu\n",
1465                                 nb_lcore_params);
1466                         return -1;
1467                 }
1468                 lcore_params_array[nb_lcore_params].port_id =
1469                         (uint8_t)int_fld[FLD_PORT];
1470                 lcore_params_array[nb_lcore_params].queue_id =
1471                         (uint8_t)int_fld[FLD_QUEUE];
1472                 lcore_params_array[nb_lcore_params].lcore_id =
1473                         (uint8_t)int_fld[FLD_LCORE];
1474                 ++nb_lcore_params;
1475         }
1476         lcore_params = lcore_params_array;
1477         return 0;
1478 }
1479
1480 static void
1481 print_app_sa_prm(const struct app_sa_prm *prm)
1482 {
1483         printf("librte_ipsec usage: %s\n",
1484                 (prm->enable == 0) ? "disabled" : "enabled");
1485
1486         printf("replay window size: %u\n", prm->window_size);
1487         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1488         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1489         printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1490 }
1491
1492 static int
1493 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1494 {
1495         if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1496                 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1497         else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1498                 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1499         else {
1500                 printf("Unsupported packet transfer mode\n");
1501                 return -EINVAL;
1502         }
1503
1504         return 0;
1505 }
1506
1507 static int
1508 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1509 {
1510         struct eventmode_conf *em_conf = NULL;
1511
1512         /* Get eventmode conf */
1513         em_conf = conf->mode_params;
1514
1515         if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1516                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1517         else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1518                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1519         else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1520                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1521         else {
1522                 printf("Unsupported queue schedule type\n");
1523                 return -EINVAL;
1524         }
1525
1526         return 0;
1527 }
1528
1529 static int32_t
1530 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1531 {
1532         int opt;
1533         int64_t ret;
1534         char **argvopt;
1535         int32_t option_index;
1536         char *prgname = argv[0];
1537         int32_t f_present = 0;
1538
1539         argvopt = argv;
1540
1541         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:",
1542                                 lgopts, &option_index)) != EOF) {
1543
1544                 switch (opt) {
1545                 case 'p':
1546                         enabled_port_mask = parse_portmask(optarg);
1547                         if (enabled_port_mask == 0) {
1548                                 printf("invalid portmask\n");
1549                                 print_usage(prgname);
1550                                 return -1;
1551                         }
1552                         break;
1553                 case 'P':
1554                         printf("Promiscuous mode selected\n");
1555                         promiscuous_on = 1;
1556                         break;
1557                 case 'u':
1558                         unprotected_port_mask = parse_portmask(optarg);
1559                         if (unprotected_port_mask == 0) {
1560                                 printf("invalid unprotected portmask\n");
1561                                 print_usage(prgname);
1562                                 return -1;
1563                         }
1564                         break;
1565                 case 'f':
1566                         if (f_present == 1) {
1567                                 printf("\"-f\" option present more than "
1568                                         "once!\n");
1569                                 print_usage(prgname);
1570                                 return -1;
1571                         }
1572                         cfgfile = optarg;
1573                         f_present = 1;
1574                         break;
1575                 case 'j':
1576                         ret = parse_decimal(optarg);
1577                         if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1578                                         ret > UINT16_MAX) {
1579                                 printf("Invalid frame buffer size value: %s\n",
1580                                         optarg);
1581                                 print_usage(prgname);
1582                                 return -1;
1583                         }
1584                         frame_buf_size = ret;
1585                         printf("Custom frame buffer size %u\n", frame_buf_size);
1586                         break;
1587                 case 'l':
1588                         app_sa_prm.enable = 1;
1589                         break;
1590                 case 'w':
1591                         app_sa_prm.window_size = parse_decimal(optarg);
1592                         break;
1593                 case 'e':
1594                         app_sa_prm.enable_esn = 1;
1595                         break;
1596                 case 'a':
1597                         app_sa_prm.enable = 1;
1598                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1599                         break;
1600                 case 'c':
1601                         ret = parse_decimal(optarg);
1602                         if (ret < 0) {
1603                                 printf("Invalid SA cache size: %s\n", optarg);
1604                                 print_usage(prgname);
1605                                 return -1;
1606                         }
1607                         app_sa_prm.cache_sz = ret;
1608                         break;
1609                 case CMD_LINE_OPT_CONFIG_NUM:
1610                         ret = parse_config(optarg);
1611                         if (ret) {
1612                                 printf("Invalid config\n");
1613                                 print_usage(prgname);
1614                                 return -1;
1615                         }
1616                         break;
1617                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1618                         ret = parse_decimal(optarg);
1619                         if (ret == -1 || ret > UINT32_MAX) {
1620                                 printf("Invalid argument[sa_idx]\n");
1621                                 print_usage(prgname);
1622                                 return -1;
1623                         }
1624
1625                         /* else */
1626                         single_sa = 1;
1627                         single_sa_idx = ret;
1628                         eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1629                         printf("Configured with single SA index %u\n",
1630                                         single_sa_idx);
1631                         break;
1632                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1633                         ret = parse_portmask(optarg);
1634                         if (ret == -1) {
1635                                 printf("Invalid argument[portmask]\n");
1636                                 print_usage(prgname);
1637                                 return -1;
1638                         }
1639
1640                         /* else */
1641                         enabled_cryptodev_mask = ret;
1642                         break;
1643
1644                 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1645                         ret = parse_transfer_mode(eh_conf, optarg);
1646                         if (ret < 0) {
1647                                 printf("Invalid packet transfer mode\n");
1648                                 print_usage(prgname);
1649                                 return -1;
1650                         }
1651                         break;
1652
1653                 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1654                         ret = parse_schedule_type(eh_conf, optarg);
1655                         if (ret < 0) {
1656                                 printf("Invalid queue schedule type\n");
1657                                 print_usage(prgname);
1658                                 return -1;
1659                         }
1660                         break;
1661
1662                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1663                         ret = parse_mask(optarg, &dev_rx_offload);
1664                         if (ret != 0) {
1665                                 printf("Invalid argument for \'%s\': %s\n",
1666                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1667                                 print_usage(prgname);
1668                                 return -1;
1669                         }
1670                         break;
1671                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1672                         ret = parse_mask(optarg, &dev_tx_offload);
1673                         if (ret != 0) {
1674                                 printf("Invalid argument for \'%s\': %s\n",
1675                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1676                                 print_usage(prgname);
1677                                 return -1;
1678                         }
1679                         break;
1680                 case CMD_LINE_OPT_REASSEMBLE_NUM:
1681                         ret = parse_decimal(optarg);
1682                         if (ret < 0 || ret > UINT32_MAX) {
1683                                 printf("Invalid argument for \'%s\': %s\n",
1684                                         CMD_LINE_OPT_REASSEMBLE, optarg);
1685                                 print_usage(prgname);
1686                                 return -1;
1687                         }
1688                         frag_tbl_sz = ret;
1689                         break;
1690                 case CMD_LINE_OPT_MTU_NUM:
1691                         ret = parse_decimal(optarg);
1692                         if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1693                                 printf("Invalid argument for \'%s\': %s\n",
1694                                         CMD_LINE_OPT_MTU, optarg);
1695                                 print_usage(prgname);
1696                                 return -1;
1697                         }
1698                         mtu_size = ret;
1699                         break;
1700                 case CMD_LINE_OPT_FRAG_TTL_NUM:
1701                         ret = parse_decimal(optarg);
1702                         if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1703                                 printf("Invalid argument for \'%s\': %s\n",
1704                                         CMD_LINE_OPT_MTU, optarg);
1705                                 print_usage(prgname);
1706                                 return -1;
1707                         }
1708                         frag_ttl_ns = ret;
1709                         break;
1710                 default:
1711                         print_usage(prgname);
1712                         return -1;
1713                 }
1714         }
1715
1716         if (f_present == 0) {
1717                 printf("Mandatory option \"-f\" not present\n");
1718                 return -1;
1719         }
1720
1721         /* check do we need to enable multi-seg support */
1722         if (multi_seg_required()) {
1723                 /* legacy mode doesn't support multi-seg */
1724                 app_sa_prm.enable = 1;
1725                 printf("frame buf size: %u, mtu: %u, "
1726                         "number of reassemble entries: %u\n"
1727                         "multi-segment support is required\n",
1728                         frame_buf_size, mtu_size, frag_tbl_sz);
1729         }
1730
1731         print_app_sa_prm(&app_sa_prm);
1732
1733         if (optind >= 0)
1734                 argv[optind-1] = prgname;
1735
1736         ret = optind-1;
1737         optind = 1; /* reset getopt lib */
1738         return ret;
1739 }
1740
1741 static void
1742 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1743 {
1744         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1745         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1746         printf("%s%s", name, buf);
1747 }
1748
1749 /*
1750  * Update destination ethaddr for the port.
1751  */
1752 int
1753 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1754 {
1755         if (port >= RTE_DIM(ethaddr_tbl))
1756                 return -EINVAL;
1757
1758         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1759         return 0;
1760 }
1761
1762 /* Check the link status of all ports in up to 9s, and print them finally */
1763 static void
1764 check_all_ports_link_status(uint32_t port_mask)
1765 {
1766 #define CHECK_INTERVAL 100 /* 100ms */
1767 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1768         uint16_t portid;
1769         uint8_t count, all_ports_up, print_flag = 0;
1770         struct rte_eth_link link;
1771         int ret;
1772
1773         printf("\nChecking link status");
1774         fflush(stdout);
1775         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1776                 all_ports_up = 1;
1777                 RTE_ETH_FOREACH_DEV(portid) {
1778                         if ((port_mask & (1 << portid)) == 0)
1779                                 continue;
1780                         memset(&link, 0, sizeof(link));
1781                         ret = rte_eth_link_get_nowait(portid, &link);
1782                         if (ret < 0) {
1783                                 all_ports_up = 0;
1784                                 if (print_flag == 1)
1785                                         printf("Port %u link get failed: %s\n",
1786                                                 portid, rte_strerror(-ret));
1787                                 continue;
1788                         }
1789                         /* print link status if flag set */
1790                         if (print_flag == 1) {
1791                                 if (link.link_status)
1792                                         printf(
1793                                         "Port%d Link Up - speed %u Mbps -%s\n",
1794                                                 portid, link.link_speed,
1795                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1796                                         ("full-duplex") : ("half-duplex\n"));
1797                                 else
1798                                         printf("Port %d Link Down\n", portid);
1799                                 continue;
1800                         }
1801                         /* clear all_ports_up flag if any link down */
1802                         if (link.link_status == ETH_LINK_DOWN) {
1803                                 all_ports_up = 0;
1804                                 break;
1805                         }
1806                 }
1807                 /* after finally printing all link status, get out */
1808                 if (print_flag == 1)
1809                         break;
1810
1811                 if (all_ports_up == 0) {
1812                         printf(".");
1813                         fflush(stdout);
1814                         rte_delay_ms(CHECK_INTERVAL);
1815                 }
1816
1817                 /* set the print_flag if all ports up or timeout */
1818                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1819                         print_flag = 1;
1820                         printf("done\n");
1821                 }
1822         }
1823 }
1824
1825 static int32_t
1826 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1827                 uint16_t qp, struct lcore_params *params,
1828                 struct ipsec_ctx *ipsec_ctx,
1829                 const struct rte_cryptodev_capabilities *cipher,
1830                 const struct rte_cryptodev_capabilities *auth,
1831                 const struct rte_cryptodev_capabilities *aead)
1832 {
1833         int32_t ret = 0;
1834         unsigned long i;
1835         struct cdev_key key = { 0 };
1836
1837         key.lcore_id = params->lcore_id;
1838         if (cipher)
1839                 key.cipher_algo = cipher->sym.cipher.algo;
1840         if (auth)
1841                 key.auth_algo = auth->sym.auth.algo;
1842         if (aead)
1843                 key.aead_algo = aead->sym.aead.algo;
1844
1845         ret = rte_hash_lookup(map, &key);
1846         if (ret != -ENOENT)
1847                 return 0;
1848
1849         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1850                 if (ipsec_ctx->tbl[i].id == cdev_id)
1851                         break;
1852
1853         if (i == ipsec_ctx->nb_qps) {
1854                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1855                         printf("Maximum number of crypto devices assigned to "
1856                                 "a core, increase MAX_QP_PER_LCORE value\n");
1857                         return 0;
1858                 }
1859                 ipsec_ctx->tbl[i].id = cdev_id;
1860                 ipsec_ctx->tbl[i].qp = qp;
1861                 ipsec_ctx->nb_qps++;
1862                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1863                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1864                                 cdev_id, qp, i);
1865         }
1866
1867         ret = rte_hash_add_key_data(map, &key, (void *)i);
1868         if (ret < 0) {
1869                 printf("Faled to insert cdev mapping for (lcore %u, "
1870                                 "cdev %u, qp %u), errno %d\n",
1871                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1872                                 ipsec_ctx->tbl[i].qp, ret);
1873                 return 0;
1874         }
1875
1876         return 1;
1877 }
1878
1879 static int32_t
1880 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1881                 uint16_t qp, struct lcore_params *params)
1882 {
1883         int32_t ret = 0;
1884         const struct rte_cryptodev_capabilities *i, *j;
1885         struct rte_hash *map;
1886         struct lcore_conf *qconf;
1887         struct ipsec_ctx *ipsec_ctx;
1888         const char *str;
1889
1890         qconf = &lcore_conf[params->lcore_id];
1891
1892         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1893                 map = cdev_map_out;
1894                 ipsec_ctx = &qconf->outbound;
1895                 str = "Outbound";
1896         } else {
1897                 map = cdev_map_in;
1898                 ipsec_ctx = &qconf->inbound;
1899                 str = "Inbound";
1900         }
1901
1902         /* Required cryptodevs with operation chainning */
1903         if (!(dev_info->feature_flags &
1904                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1905                 return ret;
1906
1907         for (i = dev_info->capabilities;
1908                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1909                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1910                         continue;
1911
1912                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1913                         ret |= add_mapping(map, str, cdev_id, qp, params,
1914                                         ipsec_ctx, NULL, NULL, i);
1915                         continue;
1916                 }
1917
1918                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1919                         continue;
1920
1921                 for (j = dev_info->capabilities;
1922                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1923                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1924                                 continue;
1925
1926                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1927                                 continue;
1928
1929                         ret |= add_mapping(map, str, cdev_id, qp, params,
1930                                                 ipsec_ctx, i, j, NULL);
1931                 }
1932         }
1933
1934         return ret;
1935 }
1936
1937 /* Check if the device is enabled by cryptodev_mask */
1938 static int
1939 check_cryptodev_mask(uint8_t cdev_id)
1940 {
1941         if (enabled_cryptodev_mask & (1 << cdev_id))
1942                 return 0;
1943
1944         return -1;
1945 }
1946
1947 static int32_t
1948 cryptodevs_init(void)
1949 {
1950         struct rte_cryptodev_config dev_conf;
1951         struct rte_cryptodev_qp_conf qp_conf;
1952         uint16_t idx, max_nb_qps, qp, i;
1953         int16_t cdev_id;
1954         struct rte_hash_parameters params = { 0 };
1955
1956         const uint64_t mseg_flag = multi_seg_required() ?
1957                                 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
1958
1959         params.entries = CDEV_MAP_ENTRIES;
1960         params.key_len = sizeof(struct cdev_key);
1961         params.hash_func = rte_jhash;
1962         params.hash_func_init_val = 0;
1963         params.socket_id = rte_socket_id();
1964
1965         params.name = "cdev_map_in";
1966         cdev_map_in = rte_hash_create(&params);
1967         if (cdev_map_in == NULL)
1968                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1969                                 rte_errno);
1970
1971         params.name = "cdev_map_out";
1972         cdev_map_out = rte_hash_create(&params);
1973         if (cdev_map_out == NULL)
1974                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1975                                 rte_errno);
1976
1977         printf("lcore/cryptodev/qp mappings:\n");
1978
1979         idx = 0;
1980         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1981                 struct rte_cryptodev_info cdev_info;
1982
1983                 if (check_cryptodev_mask((uint8_t)cdev_id))
1984                         continue;
1985
1986                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1987
1988                 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
1989                         rte_exit(EXIT_FAILURE,
1990                                 "Device %hd does not support \'%s\' feature\n",
1991                                 cdev_id,
1992                                 rte_cryptodev_get_feature_name(mseg_flag));
1993
1994                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1995                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1996                 else
1997                         max_nb_qps = nb_lcore_params;
1998
1999                 qp = 0;
2000                 i = 0;
2001                 while (qp < max_nb_qps && i < nb_lcore_params) {
2002                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2003                                                 &lcore_params[idx]))
2004                                 qp++;
2005                         idx++;
2006                         idx = idx % nb_lcore_params;
2007                         i++;
2008                 }
2009
2010                 if (qp == 0)
2011                         continue;
2012
2013                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2014                 dev_conf.nb_queue_pairs = qp;
2015                 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2016
2017                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2018                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
2019                         rte_exit(EXIT_FAILURE,
2020                                 "Device does not support at least %u "
2021                                 "sessions", CDEV_MP_NB_OBJS);
2022
2023                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2024                         rte_panic("Failed to initialize cryptodev %u\n",
2025                                         cdev_id);
2026
2027                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2028                 qp_conf.mp_session =
2029                         socket_ctx[dev_conf.socket_id].session_pool;
2030                 qp_conf.mp_session_private =
2031                         socket_ctx[dev_conf.socket_id].session_priv_pool;
2032                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2033                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2034                                         &qp_conf, dev_conf.socket_id))
2035                                 rte_panic("Failed to setup queue %u for "
2036                                                 "cdev_id %u\n", 0, cdev_id);
2037
2038                 if (rte_cryptodev_start(cdev_id))
2039                         rte_panic("Failed to start cryptodev %u\n",
2040                                         cdev_id);
2041         }
2042
2043         printf("\n");
2044
2045         return 0;
2046 }
2047
2048 static void
2049 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2050 {
2051         uint32_t frame_size;
2052         struct rte_eth_dev_info dev_info;
2053         struct rte_eth_txconf *txconf;
2054         uint16_t nb_tx_queue, nb_rx_queue;
2055         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2056         int32_t ret, socket_id;
2057         struct lcore_conf *qconf;
2058         struct rte_ether_addr ethaddr;
2059         struct rte_eth_conf local_port_conf = port_conf;
2060
2061         ret = rte_eth_dev_info_get(portid, &dev_info);
2062         if (ret != 0)
2063                 rte_exit(EXIT_FAILURE,
2064                         "Error during getting device (port %u) info: %s\n",
2065                         portid, strerror(-ret));
2066
2067         /* limit allowed HW offloafs, as user requested */
2068         dev_info.rx_offload_capa &= dev_rx_offload;
2069         dev_info.tx_offload_capa &= dev_tx_offload;
2070
2071         printf("Configuring device port %u:\n", portid);
2072
2073         ret = rte_eth_macaddr_get(portid, &ethaddr);
2074         if (ret != 0)
2075                 rte_exit(EXIT_FAILURE,
2076                         "Error getting MAC address (port %u): %s\n",
2077                         portid, rte_strerror(-ret));
2078
2079         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
2080         print_ethaddr("Address: ", &ethaddr);
2081         printf("\n");
2082
2083         nb_rx_queue = get_port_nb_rx_queues(portid);
2084         nb_tx_queue = nb_lcores;
2085
2086         if (nb_rx_queue > dev_info.max_rx_queues)
2087                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2088                                 "(max rx queue is %u)\n",
2089                                 nb_rx_queue, dev_info.max_rx_queues);
2090
2091         if (nb_tx_queue > dev_info.max_tx_queues)
2092                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2093                                 "(max tx queue is %u)\n",
2094                                 nb_tx_queue, dev_info.max_tx_queues);
2095
2096         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2097                         nb_rx_queue, nb_tx_queue);
2098
2099         frame_size = MTU_TO_FRAMELEN(mtu_size);
2100         if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
2101                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2102         local_port_conf.rxmode.max_rx_pkt_len = frame_size;
2103
2104         if (multi_seg_required()) {
2105                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
2106                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
2107         }
2108
2109         local_port_conf.rxmode.offloads |= req_rx_offloads;
2110         local_port_conf.txmode.offloads |= req_tx_offloads;
2111
2112         /* Check that all required capabilities are supported */
2113         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2114                         local_port_conf.rxmode.offloads)
2115                 rte_exit(EXIT_FAILURE,
2116                         "Error: port %u required RX offloads: 0x%" PRIx64
2117                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2118                         portid, local_port_conf.rxmode.offloads,
2119                         dev_info.rx_offload_capa);
2120
2121         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2122                         local_port_conf.txmode.offloads)
2123                 rte_exit(EXIT_FAILURE,
2124                         "Error: port %u required TX offloads: 0x%" PRIx64
2125                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2126                         portid, local_port_conf.txmode.offloads,
2127                         dev_info.tx_offload_capa);
2128
2129         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2130                 local_port_conf.txmode.offloads |=
2131                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2132
2133         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
2134                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2135
2136         printf("port %u configurng rx_offloads=0x%" PRIx64
2137                 ", tx_offloads=0x%" PRIx64 "\n",
2138                 portid, local_port_conf.rxmode.offloads,
2139                 local_port_conf.txmode.offloads);
2140
2141         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2142                 dev_info.flow_type_rss_offloads;
2143         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2144                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
2145                 printf("Port %u modified RSS hash function based on hardware support,"
2146                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2147                         portid,
2148                         port_conf.rx_adv_conf.rss_conf.rss_hf,
2149                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2150         }
2151
2152         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2153                         &local_port_conf);
2154         if (ret < 0)
2155                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2156                                 "err=%d, port=%d\n", ret, portid);
2157
2158         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2159         if (ret < 0)
2160                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2161                                 "err=%d, port=%d\n", ret, portid);
2162
2163         /* init one TX queue per lcore */
2164         tx_queueid = 0;
2165         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2166                 if (rte_lcore_is_enabled(lcore_id) == 0)
2167                         continue;
2168
2169                 if (numa_on)
2170                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2171                 else
2172                         socket_id = 0;
2173
2174                 /* init TX queue */
2175                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2176
2177                 txconf = &dev_info.default_txconf;
2178                 txconf->offloads = local_port_conf.txmode.offloads;
2179
2180                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2181                                 socket_id, txconf);
2182                 if (ret < 0)
2183                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2184                                         "err=%d, port=%d\n", ret, portid);
2185
2186                 qconf = &lcore_conf[lcore_id];
2187                 qconf->tx_queue_id[portid] = tx_queueid;
2188
2189                 /* Pre-populate pkt offloads based on capabilities */
2190                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2191                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2192                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2193                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2194
2195                 tx_queueid++;
2196
2197                 /* init RX queues */
2198                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2199                         struct rte_eth_rxconf rxq_conf;
2200
2201                         if (portid != qconf->rx_queue_list[queue].port_id)
2202                                 continue;
2203
2204                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
2205
2206                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2207                                         socket_id);
2208
2209                         rxq_conf = dev_info.default_rxconf;
2210                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
2211                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2212                                         nb_rxd, socket_id, &rxq_conf,
2213                                         socket_ctx[socket_id].mbuf_pool);
2214                         if (ret < 0)
2215                                 rte_exit(EXIT_FAILURE,
2216                                         "rte_eth_rx_queue_setup: err=%d, "
2217                                         "port=%d\n", ret, portid);
2218                 }
2219         }
2220         printf("\n");
2221 }
2222
2223 static size_t
2224 max_session_size(void)
2225 {
2226         size_t max_sz, sz;
2227         void *sec_ctx;
2228         int16_t cdev_id, port_id, n;
2229
2230         max_sz = 0;
2231         n =  rte_cryptodev_count();
2232         for (cdev_id = 0; cdev_id != n; cdev_id++) {
2233                 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2234                 if (sz > max_sz)
2235                         max_sz = sz;
2236                 /*
2237                  * If crypto device is security capable, need to check the
2238                  * size of security session as well.
2239                  */
2240
2241                 /* Get security context of the crypto device */
2242                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2243                 if (sec_ctx == NULL)
2244                         continue;
2245
2246                 /* Get size of security session */
2247                 sz = rte_security_session_get_size(sec_ctx);
2248                 if (sz > max_sz)
2249                         max_sz = sz;
2250         }
2251
2252         RTE_ETH_FOREACH_DEV(port_id) {
2253                 if ((enabled_port_mask & (1 << port_id)) == 0)
2254                         continue;
2255
2256                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2257                 if (sec_ctx == NULL)
2258                         continue;
2259
2260                 sz = rte_security_session_get_size(sec_ctx);
2261                 if (sz > max_sz)
2262                         max_sz = sz;
2263         }
2264
2265         return max_sz;
2266 }
2267
2268 static void
2269 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2270 {
2271         char mp_name[RTE_MEMPOOL_NAMESIZE];
2272         struct rte_mempool *sess_mp;
2273
2274         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2275                         "sess_mp_%u", socket_id);
2276         sess_mp = rte_cryptodev_sym_session_pool_create(
2277                         mp_name, CDEV_MP_NB_OBJS,
2278                         sess_sz, CDEV_MP_CACHE_SZ, 0,
2279                         socket_id);
2280         ctx->session_pool = sess_mp;
2281
2282         if (ctx->session_pool == NULL)
2283                 rte_exit(EXIT_FAILURE,
2284                         "Cannot init session pool on socket %d\n", socket_id);
2285         else
2286                 printf("Allocated session pool on socket %d\n", socket_id);
2287 }
2288
2289 static void
2290 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2291         size_t sess_sz)
2292 {
2293         char mp_name[RTE_MEMPOOL_NAMESIZE];
2294         struct rte_mempool *sess_mp;
2295
2296         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2297                         "sess_mp_priv_%u", socket_id);
2298         sess_mp = rte_mempool_create(mp_name,
2299                         CDEV_MP_NB_OBJS,
2300                         sess_sz,
2301                         CDEV_MP_CACHE_SZ,
2302                         0, NULL, NULL, NULL,
2303                         NULL, socket_id,
2304                         0);
2305         ctx->session_priv_pool = sess_mp;
2306
2307         if (ctx->session_priv_pool == NULL)
2308                 rte_exit(EXIT_FAILURE,
2309                         "Cannot init session priv pool on socket %d\n",
2310                         socket_id);
2311         else
2312                 printf("Allocated session priv pool on socket %d\n",
2313                         socket_id);
2314 }
2315
2316 static void
2317 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2318 {
2319         char s[64];
2320         int32_t ms;
2321
2322         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2323         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2324                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2325                         frame_buf_size, socket_id);
2326
2327         /*
2328          * if multi-segment support is enabled, then create a pool
2329          * for indirect mbufs.
2330          */
2331         ms = multi_seg_required();
2332         if (ms != 0) {
2333                 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2334                 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2335                         MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2336         }
2337
2338         if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2339                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2340                                 socket_id);
2341         else
2342                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2343 }
2344
2345 static inline int
2346 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2347 {
2348         struct ipsec_sa *sa;
2349
2350         /* For inline protocol processing, the metadata in the event will
2351          * uniquely identify the security session which raised the event.
2352          * Application would then need the userdata it had registered with the
2353          * security session to process the event.
2354          */
2355
2356         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2357
2358         if (sa == NULL) {
2359                 /* userdata could not be retrieved */
2360                 return -1;
2361         }
2362
2363         /* Sequence number over flow. SA need to be re-established */
2364         RTE_SET_USED(sa);
2365         return 0;
2366 }
2367
2368 static int
2369 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2370                  void *param, void *ret_param)
2371 {
2372         uint64_t md;
2373         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2374         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2375                                         rte_eth_dev_get_sec_ctx(port_id);
2376
2377         RTE_SET_USED(param);
2378
2379         if (type != RTE_ETH_EVENT_IPSEC)
2380                 return -1;
2381
2382         event_desc = ret_param;
2383         if (event_desc == NULL) {
2384                 printf("Event descriptor not set\n");
2385                 return -1;
2386         }
2387
2388         md = event_desc->metadata;
2389
2390         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2391                 return inline_ipsec_event_esn_overflow(ctx, md);
2392         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2393                 printf("Invalid IPsec event reported\n");
2394                 return -1;
2395         }
2396
2397         return -1;
2398 }
2399
2400 static uint16_t
2401 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2402         struct rte_mbuf *pkt[], uint16_t nb_pkts,
2403         __rte_unused uint16_t max_pkts, void *user_param)
2404 {
2405         uint64_t tm;
2406         uint32_t i, k;
2407         struct lcore_conf *lc;
2408         struct rte_mbuf *mb;
2409         struct rte_ether_hdr *eth;
2410
2411         lc = user_param;
2412         k = 0;
2413         tm = 0;
2414
2415         for (i = 0; i != nb_pkts; i++) {
2416
2417                 mb = pkt[i];
2418                 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2419                 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2420
2421                         struct rte_ipv4_hdr *iph;
2422
2423                         iph = (struct rte_ipv4_hdr *)(eth + 1);
2424                         if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2425
2426                                 mb->l2_len = sizeof(*eth);
2427                                 mb->l3_len = sizeof(*iph);
2428                                 tm = (tm != 0) ? tm : rte_rdtsc();
2429                                 mb = rte_ipv4_frag_reassemble_packet(
2430                                         lc->frag.tbl, &lc->frag.dr,
2431                                         mb, tm, iph);
2432
2433                                 if (mb != NULL) {
2434                                         /* fix ip cksum after reassemble. */
2435                                         iph = rte_pktmbuf_mtod_offset(mb,
2436                                                 struct rte_ipv4_hdr *,
2437                                                 mb->l2_len);
2438                                         iph->hdr_checksum = 0;
2439                                         iph->hdr_checksum = rte_ipv4_cksum(iph);
2440                                 }
2441                         }
2442                 } else if (eth->ether_type ==
2443                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2444
2445                         struct rte_ipv6_hdr *iph;
2446                         struct ipv6_extension_fragment *fh;
2447
2448                         iph = (struct rte_ipv6_hdr *)(eth + 1);
2449                         fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2450                         if (fh != NULL) {
2451                                 mb->l2_len = sizeof(*eth);
2452                                 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2453                                         sizeof(*fh);
2454                                 tm = (tm != 0) ? tm : rte_rdtsc();
2455                                 mb = rte_ipv6_frag_reassemble_packet(
2456                                         lc->frag.tbl, &lc->frag.dr,
2457                                         mb, tm, iph, fh);
2458                                 if (mb != NULL)
2459                                         /* fix l3_len after reassemble. */
2460                                         mb->l3_len = mb->l3_len - sizeof(*fh);
2461                         }
2462                 }
2463
2464                 pkt[k] = mb;
2465                 k += (mb != NULL);
2466         }
2467
2468         /* some fragments were encountered, drain death row */
2469         if (tm != 0)
2470                 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2471
2472         return k;
2473 }
2474
2475
2476 static int
2477 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2478 {
2479         int32_t sid;
2480         uint32_t i;
2481         uint64_t frag_cycles;
2482         const struct lcore_rx_queue *rxq;
2483         const struct rte_eth_rxtx_callback *cb;
2484
2485         /* create fragment table */
2486         sid = rte_lcore_to_socket_id(cid);
2487         frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2488                 NS_PER_S * frag_ttl_ns;
2489
2490         lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2491                 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2492         if (lc->frag.tbl == NULL) {
2493                 printf("%s(%u): failed to create fragment table of size: %u, "
2494                         "error code: %d\n",
2495                         __func__, cid, frag_tbl_sz, rte_errno);
2496                 return -ENOMEM;
2497         }
2498
2499         /* setup reassemble RX callbacks for all queues */
2500         for (i = 0; i != lc->nb_rx_queue; i++) {
2501
2502                 rxq = lc->rx_queue_list + i;
2503                 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2504                         rx_callback, lc);
2505                 if (cb == NULL) {
2506                         printf("%s(%u): failed to install RX callback for "
2507                                 "portid=%u, queueid=%u, error code: %d\n",
2508                                 __func__, cid,
2509                                 rxq->port_id, rxq->queue_id, rte_errno);
2510                         return -ENOMEM;
2511                 }
2512         }
2513
2514         return 0;
2515 }
2516
2517 static int
2518 reassemble_init(void)
2519 {
2520         int32_t rc;
2521         uint32_t i, lc;
2522
2523         rc = 0;
2524         for (i = 0; i != nb_lcore_params; i++) {
2525                 lc = lcore_params[i].lcore_id;
2526                 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2527                 if (rc != 0)
2528                         break;
2529         }
2530
2531         return rc;
2532 }
2533
2534 static void
2535 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2536 {
2537         struct rte_flow_action action[2];
2538         struct rte_flow_item pattern[2];
2539         struct rte_flow_attr attr = {0};
2540         struct rte_flow_error err;
2541         struct rte_flow *flow;
2542         int ret;
2543
2544         if (!(rx_offloads & DEV_RX_OFFLOAD_SECURITY))
2545                 return;
2546
2547         /* Add the default rte_flow to enable SECURITY for all ESP packets */
2548
2549         pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2550         pattern[0].spec = NULL;
2551         pattern[0].mask = NULL;
2552         pattern[0].last = NULL;
2553         pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2554
2555         action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2556         action[0].conf = NULL;
2557         action[1].type = RTE_FLOW_ACTION_TYPE_END;
2558         action[1].conf = NULL;
2559
2560         attr.ingress = 1;
2561
2562         ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2563         if (ret)
2564                 return;
2565
2566         flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2567         if (flow == NULL)
2568                 return;
2569
2570         flow_info_tbl[port_id].rx_def_flow = flow;
2571         RTE_LOG(INFO, IPSEC,
2572                 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2573                 port_id);
2574 }
2575
2576 static void
2577 signal_handler(int signum)
2578 {
2579         if (signum == SIGINT || signum == SIGTERM) {
2580                 printf("\n\nSignal %d received, preparing to exit...\n",
2581                                 signum);
2582                 force_quit = true;
2583         }
2584 }
2585
2586 static void
2587 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2588 {
2589         struct rte_ipsec_session *ips;
2590         int32_t i;
2591
2592         if (!sa || !nb_sa)
2593                 return;
2594
2595         for (i = 0; i < nb_sa; i++) {
2596                 ips = ipsec_get_primary_session(&sa[i]);
2597                 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2598                         rte_exit(EXIT_FAILURE, "Event mode supports only "
2599                                  "inline protocol sessions\n");
2600         }
2601
2602 }
2603
2604 static int32_t
2605 check_event_mode_params(struct eh_conf *eh_conf)
2606 {
2607         struct eventmode_conf *em_conf = NULL;
2608         struct lcore_params *params;
2609         uint16_t portid;
2610
2611         if (!eh_conf || !eh_conf->mode_params)
2612                 return -EINVAL;
2613
2614         /* Get eventmode conf */
2615         em_conf = eh_conf->mode_params;
2616
2617         if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2618             em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2619                 printf("error: option --event-schedule-type applies only to "
2620                        "event mode\n");
2621                 return -EINVAL;
2622         }
2623
2624         if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2625                 return 0;
2626
2627         /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2628         if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2629                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2630
2631         /*
2632          * Event mode currently supports only inline protocol sessions.
2633          * If there are other types of sessions configured then exit with
2634          * error.
2635          */
2636         ev_mode_sess_verify(sa_in, nb_sa_in);
2637         ev_mode_sess_verify(sa_out, nb_sa_out);
2638
2639
2640         /* Option --config does not apply to event mode */
2641         if (nb_lcore_params > 0) {
2642                 printf("error: option --config applies only to poll mode\n");
2643                 return -EINVAL;
2644         }
2645
2646         /*
2647          * In order to use the same port_init routine for both poll and event
2648          * modes initialize lcore_params with one queue for each eth port
2649          */
2650         lcore_params = lcore_params_array;
2651         RTE_ETH_FOREACH_DEV(portid) {
2652                 if ((enabled_port_mask & (1 << portid)) == 0)
2653                         continue;
2654
2655                 params = &lcore_params[nb_lcore_params++];
2656                 params->port_id = portid;
2657                 params->queue_id = 0;
2658                 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2659         }
2660
2661         return 0;
2662 }
2663
2664 static void
2665 inline_sessions_free(struct sa_ctx *sa_ctx)
2666 {
2667         struct rte_ipsec_session *ips;
2668         struct ipsec_sa *sa;
2669         int32_t ret;
2670         uint32_t i;
2671
2672         if (!sa_ctx)
2673                 return;
2674
2675         for (i = 0; i < sa_ctx->nb_sa; i++) {
2676
2677                 sa = &sa_ctx->sa[i];
2678                 if (!sa->spi)
2679                         continue;
2680
2681                 ips = ipsec_get_primary_session(sa);
2682                 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2683                     ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2684                         continue;
2685
2686                 if (!rte_eth_dev_is_valid_port(sa->portid))
2687                         continue;
2688
2689                 ret = rte_security_session_destroy(
2690                                 rte_eth_dev_get_sec_ctx(sa->portid),
2691                                 ips->security.ses);
2692                 if (ret)
2693                         RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2694                                             "session type %d, spi %d\n",
2695                                             ips->type, sa->spi);
2696         }
2697 }
2698
2699 int32_t
2700 main(int32_t argc, char **argv)
2701 {
2702         int32_t ret;
2703         uint32_t lcore_id;
2704         uint32_t cdev_id;
2705         uint32_t i;
2706         uint8_t socket_id;
2707         uint16_t portid;
2708         uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
2709         uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
2710         struct eh_conf *eh_conf = NULL;
2711         size_t sess_sz;
2712
2713         /* init EAL */
2714         ret = rte_eal_init(argc, argv);
2715         if (ret < 0)
2716                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2717         argc -= ret;
2718         argv += ret;
2719
2720         force_quit = false;
2721         signal(SIGINT, signal_handler);
2722         signal(SIGTERM, signal_handler);
2723
2724         /* initialize event helper configuration */
2725         eh_conf = eh_conf_init();
2726         if (eh_conf == NULL)
2727                 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
2728
2729         /* parse application arguments (after the EAL ones) */
2730         ret = parse_args(argc, argv, eh_conf);
2731         if (ret < 0)
2732                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2733
2734         /* parse configuration file */
2735         if (parse_cfg_file(cfgfile) < 0) {
2736                 printf("parsing file \"%s\" failed\n",
2737                         optarg);
2738                 print_usage(argv[0]);
2739                 return -1;
2740         }
2741
2742         if ((unprotected_port_mask & enabled_port_mask) !=
2743                         unprotected_port_mask)
2744                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2745                                 unprotected_port_mask);
2746
2747         if (check_poll_mode_params(eh_conf) < 0)
2748                 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
2749
2750         if (check_event_mode_params(eh_conf) < 0)
2751                 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
2752
2753         ret = init_lcore_rx_queues();
2754         if (ret < 0)
2755                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2756
2757         nb_lcores = rte_lcore_count();
2758
2759         sess_sz = max_session_size();
2760
2761         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2762                 if (rte_lcore_is_enabled(lcore_id) == 0)
2763                         continue;
2764
2765                 if (numa_on)
2766                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2767                 else
2768                         socket_id = 0;
2769
2770                 /* mbuf_pool is initialised by the pool_init() function*/
2771                 if (socket_ctx[socket_id].mbuf_pool)
2772                         continue;
2773
2774                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2775                 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2776                 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2777                         sess_sz);
2778         }
2779
2780         RTE_ETH_FOREACH_DEV(portid) {
2781                 if ((enabled_port_mask & (1 << portid)) == 0)
2782                         continue;
2783
2784                 sa_check_offloads(portid, &req_rx_offloads[portid],
2785                                 &req_tx_offloads[portid]);
2786                 port_init(portid, req_rx_offloads[portid],
2787                                 req_tx_offloads[portid]);
2788         }
2789
2790         cryptodevs_init();
2791
2792         /*
2793          * Set the enabled port mask in helper config for use by helper
2794          * sub-system. This will be used while initializing devices using
2795          * helper sub-system.
2796          */
2797         eh_conf->eth_portmask = enabled_port_mask;
2798
2799         /* Initialize eventmode components */
2800         ret = eh_devs_init(eh_conf);
2801         if (ret < 0)
2802                 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
2803
2804         /* start ports */
2805         RTE_ETH_FOREACH_DEV(portid) {
2806                 if ((enabled_port_mask & (1 << portid)) == 0)
2807                         continue;
2808
2809                 /* Create flow before starting the device */
2810                 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
2811
2812                 ret = rte_eth_dev_start(portid);
2813                 if (ret < 0)
2814                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2815                                         "err=%d, port=%d\n", ret, portid);
2816                 /*
2817                  * If enabled, put device in promiscuous mode.
2818                  * This allows IO forwarding mode to forward packets
2819                  * to itself through 2 cross-connected  ports of the
2820                  * target machine.
2821                  */
2822                 if (promiscuous_on) {
2823                         ret = rte_eth_promiscuous_enable(portid);
2824                         if (ret != 0)
2825                                 rte_exit(EXIT_FAILURE,
2826                                         "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2827                                         rte_strerror(-ret), portid);
2828                 }
2829
2830                 rte_eth_dev_callback_register(portid,
2831                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2832         }
2833
2834         /* fragment reassemble is enabled */
2835         if (frag_tbl_sz != 0) {
2836                 ret = reassemble_init();
2837                 if (ret != 0)
2838                         rte_exit(EXIT_FAILURE, "failed at reassemble init");
2839         }
2840
2841         /* Replicate each context per socket */
2842         for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2843                 socket_id = rte_socket_id_by_idx(i);
2844                 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2845                         (socket_ctx[socket_id].sa_in == NULL) &&
2846                         (socket_ctx[socket_id].sa_out == NULL)) {
2847                         sa_init(&socket_ctx[socket_id], socket_id);
2848                         sp4_init(&socket_ctx[socket_id], socket_id);
2849                         sp6_init(&socket_ctx[socket_id], socket_id);
2850                         rt_init(&socket_ctx[socket_id], socket_id);
2851                 }
2852         }
2853
2854         check_all_ports_link_status(enabled_port_mask);
2855
2856         /* launch per-lcore init on every lcore */
2857         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2858         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2859                 if (rte_eal_wait_lcore(lcore_id) < 0)
2860                         return -1;
2861         }
2862
2863         /* Uninitialize eventmode components */
2864         ret = eh_devs_uninit(eh_conf);
2865         if (ret < 0)
2866                 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
2867
2868         /* Free eventmode configuration memory */
2869         eh_conf_uninit(eh_conf);
2870
2871         /* Destroy inline inbound and outbound sessions */
2872         for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2873                 socket_id = rte_socket_id_by_idx(i);
2874                 inline_sessions_free(socket_ctx[socket_id].sa_in);
2875                 inline_sessions_free(socket_ctx[socket_id].sa_out);
2876         }
2877
2878         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2879                 printf("Closing cryptodev %d...", cdev_id);
2880                 rte_cryptodev_stop(cdev_id);
2881                 rte_cryptodev_close(cdev_id);
2882                 printf(" Done\n");
2883         }
2884
2885         RTE_ETH_FOREACH_DEV(portid) {
2886                 if ((enabled_port_mask & (1 << portid)) == 0)
2887                         continue;
2888
2889                 printf("Closing port %d...", portid);
2890                 if (flow_info_tbl[portid].rx_def_flow) {
2891                         struct rte_flow_error err;
2892
2893                         ret = rte_flow_destroy(portid,
2894                                 flow_info_tbl[portid].rx_def_flow, &err);
2895                         if (ret)
2896                                 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
2897                                         " for port %u, err msg: %s\n", portid,
2898                                         err.message);
2899                 }
2900                 rte_eth_dev_stop(portid);
2901                 rte_eth_dev_close(portid);
2902                 printf(" Done\n");
2903         }
2904         printf("Bye...\n");
2905
2906         return 0;
2907 }