examples/ipsec-secgw/ipsec-secgw.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <netinet/in.h>
  11 #include <netinet/ip.h>
  12 #include <netinet/ip6.h>
  13 #include <string.h>
  14 #include <sys/queue.h>
  15 #include <stdarg.h>
  16 #include <errno.h>
  17 #include <getopt.h>
  18
  19 #include <rte_common.h>
  20 #include <rte_byteorder.h>
  21 #include <rte_log.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_mempool.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_acl.h>
  38 #include <rte_lpm.h>
  39 #include <rte_lpm6.h>
  40 #include <rte_hash.h>
  41 #include <rte_jhash.h>
  42 #include <rte_cryptodev.h>
  43 #include <rte_security.h>
  44
  45 #include "ipsec.h"
  46 #include "parser.h"
  47
  48 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  49
  50 #define MAX_JUMBO_PKT_LEN  9600
  51
  52 #define MEMPOOL_CACHE_SIZE 256
  53
  54 #define NB_MBUF (32000)
  55
  56 #define CDEV_QUEUE_DESC 2048
  57 #define CDEV_MAP_ENTRIES 16384
  58 #define CDEV_MP_NB_OBJS 1024
  59 #define CDEV_MP_CACHE_SZ 64
  60 #define MAX_QUEUE_PAIRS 1
  61
  62 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  63
  64 #define NB_SOCKETS 4
  65
  66 /* Configure how many packets ahead to prefetch, when reading packets */
  67 #define PREFETCH_OFFSET 3
  68
  69 #define MAX_RX_QUEUE_PER_LCORE 16
  70
  71 #define MAX_LCORE_PARAMS 1024
  72
  73 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
  74
  75 /*
  76  * Configurable number of RX/TX ring descriptors
  77  */
  78 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  79 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  80 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  81 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  82
  83 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
  84 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  85         (((uint64_t)((a) & 0xff) << 56) | \
  86         ((uint64_t)((b) & 0xff) << 48) | \
  87         ((uint64_t)((c) & 0xff) << 40) | \
  88         ((uint64_t)((d) & 0xff) << 32) | \
  89         ((uint64_t)((e) & 0xff) << 24) | \
  90         ((uint64_t)((f) & 0xff) << 16) | \
  91         ((uint64_t)((g) & 0xff) << 8)  | \
  92         ((uint64_t)(h) & 0xff))
  93 #else
  94 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  95         (((uint64_t)((h) & 0xff) << 56) | \
  96         ((uint64_t)((g) & 0xff) << 48) | \
  97         ((uint64_t)((f) & 0xff) << 40) | \
  98         ((uint64_t)((e) & 0xff) << 32) | \
  99         ((uint64_t)((d) & 0xff) << 24) | \
 100         ((uint64_t)((c) & 0xff) << 16) | \
 101         ((uint64_t)((b) & 0xff) << 8) | \
 102         ((uint64_t)(a) & 0xff))
 103 #endif
 104 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 105
 106 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 107                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
 108                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
 109                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
 110                 0, 0)
 111
 112 /* port/source ethernet addr and destination ethernet addr */
 113 struct ethaddr_info {
 114         uint64_t src, dst;
 115 };
 116
 117 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 118         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 119         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 120         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 121         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 122 };
 123
 124 #define CMD_LINE_OPT_CONFIG             "config"
 125 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 126 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 127 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
 128 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
 129
 130 enum {
 131         /* long options mapped to a short option */
 132
 133         /* first long only option value must be >= 256, so that we won't
 134          * conflict with short options
 135          */
 136         CMD_LINE_OPT_MIN_NUM = 256,
 137         CMD_LINE_OPT_CONFIG_NUM,
 138         CMD_LINE_OPT_SINGLE_SA_NUM,
 139         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 140         CMD_LINE_OPT_RX_OFFLOAD_NUM,
 141         CMD_LINE_OPT_TX_OFFLOAD_NUM,
 142 };
 143
 144 static const struct option lgopts[] = {
 145         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 146         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 147         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 148         {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
 149         {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
 150         {NULL, 0, 0, 0}
 151 };
 152
 153 /* mask of enabled ports */
 154 static uint32_t enabled_port_mask;
 155 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 156 static uint32_t unprotected_port_mask;
 157 static int32_t promiscuous_on = 1;
 158 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 159 static uint32_t nb_lcores;
 160 static uint32_t single_sa;
 161 static uint32_t single_sa_idx;
 162 static uint32_t frame_size;
 163
 164 /*
 165  * RX/TX HW offload capabilities to enable/use on ethernet ports.
 166  * By default all capabilities are enabled.
 167  */
 168 static uint64_t dev_rx_offload = UINT64_MAX;
 169 static uint64_t dev_tx_offload = UINT64_MAX;
 170
 171 /* application wide librte_ipsec/SA parameters */
 172 struct app_sa_prm app_sa_prm = {.enable = 0};
 173
 174 struct lcore_rx_queue {
 175         uint16_t port_id;
 176         uint8_t queue_id;
 177 } __rte_cache_aligned;
 178
 179 struct lcore_params {
 180         uint16_t port_id;
 181         uint8_t queue_id;
 182         uint8_t lcore_id;
 183 } __rte_cache_aligned;
 184
 185 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 186
 187 static struct lcore_params *lcore_params;
 188 static uint16_t nb_lcore_params;
 189
 190 static struct rte_hash *cdev_map_in;
 191 static struct rte_hash *cdev_map_out;
 192
 193 struct buffer {
 194         uint16_t len;
 195         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 196 };
 197
 198 struct lcore_conf {
 199         uint16_t nb_rx_queue;
 200         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 201         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 202         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 203         struct ipsec_ctx inbound;
 204         struct ipsec_ctx outbound;
 205         struct rt_ctx *rt4_ctx;
 206         struct rt_ctx *rt6_ctx;
 207 } __rte_cache_aligned;
 208
 209 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 210
 211 static struct rte_eth_conf port_conf = {
 212         .rxmode = {
 213                 .mq_mode        = ETH_MQ_RX_RSS,
 214                 .max_rx_pkt_len = ETHER_MAX_LEN,
 215                 .split_hdr_size = 0,
 216                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 217         },
 218         .rx_adv_conf = {
 219                 .rss_conf = {
 220                         .rss_key = NULL,
 221                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 222                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 223                 },
 224         },
 225         .txmode = {
 226                 .mq_mode = ETH_MQ_TX_NONE,
 227         },
 228 };
 229
 230 static struct socket_ctx socket_ctx[NB_SOCKETS];
 231
 232 static inline void
 233 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 234 {
 235         uint8_t *nlp;
 236         struct rte_ether_hdr *eth;
 237
 238         eth = rte_pktmbuf_mtod(pkt, struct rte_ether_hdr *);
 239         if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4)) {
 240                 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
 241                 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip, ip_p));
 242                 if (*nlp == IPPROTO_ESP)
 243                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 244                 else {
 245                         t->ip4.data[t->ip4.num] = nlp;
 246                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 247                 }
 248                 pkt->l2_len = 0;
 249                 pkt->l3_len = sizeof(struct ip);
 250         } else if (eth->ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6)) {
 251                 nlp = (uint8_t *)rte_pktmbuf_adj(pkt, ETHER_HDR_LEN);
 252                 nlp = RTE_PTR_ADD(nlp, offsetof(struct ip6_hdr, ip6_nxt));
 253                 if (*nlp == IPPROTO_ESP)
 254                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 255                 else {
 256                         t->ip6.data[t->ip6.num] = nlp;
 257                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 258                 }
 259                 pkt->l2_len = 0;
 260                 pkt->l3_len = sizeof(struct ip6_hdr);
 261         } else {
 262                 /* Unknown/Unsupported type, drop the packet */
 263                 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
 264                         rte_be_to_cpu_16(eth->ether_type));
 265                 rte_pktmbuf_free(pkt);
 266         }
 267
 268         /* Check if the packet has been processed inline. For inline protocol
 269          * processed packets, the metadata in the mbuf can be used to identify
 270          * the security processing done on the packet. The metadata will be
 271          * used to retrieve the application registered userdata associated
 272          * with the security session.
 273          */
 274
 275         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 276                 struct ipsec_sa *sa;
 277                 struct ipsec_mbuf_metadata *priv;
 278                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 279                                                 rte_eth_dev_get_sec_ctx(
 280                                                 pkt->port);
 281
 282                 /* Retrieve the userdata registered. Here, the userdata
 283                  * registered is the SA pointer.
 284                  */
 285
 286                 sa = (struct ipsec_sa *)
 287                                 rte_security_get_userdata(ctx, pkt->udata64);
 288
 289                 if (sa == NULL) {
 290                         /* userdata could not be retrieved */
 291                         return;
 292                 }
 293
 294                 /* Save SA as priv member in mbuf. This will be used in the
 295                  * IPsec selector(SP-SA) check.
 296                  */
 297
 298                 priv = get_priv(pkt);
 299                 priv->sa = sa;
 300         }
 301 }
 302
 303 static inline void
 304 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 305                 uint16_t nb_pkts)
 306 {
 307         int32_t i;
 308
 309         t->ipsec.num = 0;
 310         t->ip4.num = 0;
 311         t->ip6.num = 0;
 312
 313         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 314                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 315                                         void *));
 316                 prepare_one_packet(pkts[i], t);
 317         }
 318         /* Process left packets */
 319         for (; i < nb_pkts; i++)
 320                 prepare_one_packet(pkts[i], t);
 321 }
 322
 323 static inline void
 324 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 325                 const struct lcore_conf *qconf)
 326 {
 327         struct ip *ip;
 328         struct rte_ether_hdr *ethhdr;
 329
 330         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 331
 332         ethhdr = (struct rte_ether_hdr *)
 333                 rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
 334
 335         if (ip->ip_v == IPVERSION) {
 336                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 337                 pkt->l3_len = sizeof(struct ip);
 338                 pkt->l2_len = ETHER_HDR_LEN;
 339
 340                 ip->ip_sum = 0;
 341
 342                 /* calculate IPv4 cksum in SW */
 343                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 344                         ip->ip_sum = rte_ipv4_cksum((struct ipv4_hdr *)ip);
 345
 346                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
 347         } else {
 348                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 349                 pkt->l3_len = sizeof(struct ip6_hdr);
 350                 pkt->l2_len = ETHER_HDR_LEN;
 351
 352                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
 353         }
 354
 355         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 356                         sizeof(struct rte_ether_addr));
 357         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 358                         sizeof(struct rte_ether_addr));
 359 }
 360
 361 static inline void
 362 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 363                 const struct lcore_conf *qconf)
 364 {
 365         int32_t i;
 366         const int32_t prefetch_offset = 2;
 367
 368         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 369                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 370                 prepare_tx_pkt(pkts[i], port, qconf);
 371         }
 372         /* Process left packets */
 373         for (; i < nb_pkts; i++)
 374                 prepare_tx_pkt(pkts[i], port, qconf);
 375 }
 376
 377 /* Send burst of packets on an output interface */
 378 static inline int32_t
 379 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 380 {
 381         struct rte_mbuf **m_table;
 382         int32_t ret;
 383         uint16_t queueid;
 384
 385         queueid = qconf->tx_queue_id[port];
 386         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 387
 388         prepare_tx_burst(m_table, n, port, qconf);
 389
 390         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 391         if (unlikely(ret < n)) {
 392                 do {
 393                         rte_pktmbuf_free(m_table[ret]);
 394                 } while (++ret < n);
 395         }
 396
 397         return 0;
 398 }
 399
 400 /* Enqueue a single packet, and send burst if queue is filled */
 401 static inline int32_t
 402 send_single_packet(struct rte_mbuf *m, uint16_t port)
 403 {
 404         uint32_t lcore_id;
 405         uint16_t len;
 406         struct lcore_conf *qconf;
 407
 408         lcore_id = rte_lcore_id();
 409
 410         qconf = &lcore_conf[lcore_id];
 411         len = qconf->tx_mbufs[port].len;
 412         qconf->tx_mbufs[port].m_table[len] = m;
 413         len++;
 414
 415         /* enough pkts to be sent */
 416         if (unlikely(len == MAX_PKT_BURST)) {
 417                 send_burst(qconf, MAX_PKT_BURST, port);
 418                 len = 0;
 419         }
 420
 421         qconf->tx_mbufs[port].len = len;
 422         return 0;
 423 }
 424
 425 static inline void
 426 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 427                 uint16_t lim)
 428 {
 429         struct rte_mbuf *m;
 430         uint32_t i, j, res, sa_idx;
 431
 432         if (ip->num == 0 || sp == NULL)
 433                 return;
 434
 435         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 436                         ip->num, DEFAULT_MAX_CATEGORIES);
 437
 438         j = 0;
 439         for (i = 0; i < ip->num; i++) {
 440                 m = ip->pkts[i];
 441                 res = ip->res[i];
 442                 if (res == BYPASS) {
 443                         ip->pkts[j++] = m;
 444                         continue;
 445                 }
 446                 if (res == DISCARD) {
 447                         rte_pktmbuf_free(m);
 448                         continue;
 449                 }
 450
 451                 /* Only check SPI match for processed IPSec packets */
 452                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 453                         rte_pktmbuf_free(m);
 454                         continue;
 455                 }
 456
 457                 sa_idx = SPI2IDX(res);
 458                 if (!inbound_sa_check(sa, m, sa_idx)) {
 459                         rte_pktmbuf_free(m);
 460                         continue;
 461                 }
 462                 ip->pkts[j++] = m;
 463         }
 464         ip->num = j;
 465 }
 466
 467 static void
 468 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 469 {
 470         uint32_t i, n4, n6;
 471         struct ip *ip;
 472         struct rte_mbuf *m;
 473
 474         n4 = trf->ip4.num;
 475         n6 = trf->ip6.num;
 476
 477         for (i = 0; i < num; i++) {
 478
 479                 m = mb[i];
 480                 ip = rte_pktmbuf_mtod(m, struct ip *);
 481
 482                 if (ip->ip_v == IPVERSION) {
 483                         trf->ip4.pkts[n4] = m;
 484                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 485                                         uint8_t *, offsetof(struct ip, ip_p));
 486                         n4++;
 487                 } else if (ip->ip_v == IP6_VERSION) {
 488                         trf->ip6.pkts[n6] = m;
 489                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 490                                         uint8_t *,
 491                                         offsetof(struct ip6_hdr, ip6_nxt));
 492                         n6++;
 493                 } else
 494                         rte_pktmbuf_free(m);
 495         }
 496
 497         trf->ip4.num = n4;
 498         trf->ip6.num = n6;
 499 }
 500
 501
 502 static inline void
 503 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 504                 struct ipsec_traffic *traffic)
 505 {
 506         uint16_t nb_pkts_in, n_ip4, n_ip6;
 507
 508         n_ip4 = traffic->ip4.num;
 509         n_ip6 = traffic->ip6.num;
 510
 511         if (app_sa_prm.enable == 0) {
 512                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 513                                 traffic->ipsec.num, MAX_PKT_BURST);
 514                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 515         } else {
 516                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 517                         traffic->ipsec.saptr, traffic->ipsec.num);
 518                 ipsec_process(ipsec_ctx, traffic);
 519         }
 520
 521         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 522                         n_ip4);
 523
 524         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 525                         n_ip6);
 526 }
 527
 528 static inline void
 529 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 530                 struct traffic_type *ipsec)
 531 {
 532         struct rte_mbuf *m;
 533         uint32_t i, j, sa_idx;
 534
 535         if (ip->num == 0 || sp == NULL)
 536                 return;
 537
 538         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 539                         ip->num, DEFAULT_MAX_CATEGORIES);
 540
 541         j = 0;
 542         for (i = 0; i < ip->num; i++) {
 543                 m = ip->pkts[i];
 544                 sa_idx = SPI2IDX(ip->res[i]);
 545                 if (ip->res[i] == DISCARD)
 546                         rte_pktmbuf_free(m);
 547                 else if (ip->res[i] == BYPASS)
 548                         ip->pkts[j++] = m;
 549                 else {
 550                         ipsec->res[ipsec->num] = sa_idx;
 551                         ipsec->pkts[ipsec->num++] = m;
 552                 }
 553         }
 554         ip->num = j;
 555 }
 556
 557 static inline void
 558 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 559                 struct ipsec_traffic *traffic)
 560 {
 561         struct rte_mbuf *m;
 562         uint16_t idx, nb_pkts_out, i;
 563
 564         /* Drop any IPsec traffic from protected ports */
 565         for (i = 0; i < traffic->ipsec.num; i++)
 566                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 567
 568         traffic->ipsec.num = 0;
 569
 570         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 571
 572         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 573
 574         if (app_sa_prm.enable == 0) {
 575
 576                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 577                                 traffic->ipsec.res, traffic->ipsec.num,
 578                                 MAX_PKT_BURST);
 579
 580                 for (i = 0; i < nb_pkts_out; i++) {
 581                         m = traffic->ipsec.pkts[i];
 582                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 583                         if (ip->ip_v == IPVERSION) {
 584                                 idx = traffic->ip4.num++;
 585                                 traffic->ip4.pkts[idx] = m;
 586                         } else {
 587                                 idx = traffic->ip6.num++;
 588                                 traffic->ip6.pkts[idx] = m;
 589                         }
 590                 }
 591         } else {
 592                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 593                         traffic->ipsec.saptr, traffic->ipsec.num);
 594                 ipsec_process(ipsec_ctx, traffic);
 595         }
 596 }
 597
 598 static inline void
 599 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 600                 struct ipsec_traffic *traffic)
 601 {
 602         struct rte_mbuf *m;
 603         uint32_t nb_pkts_in, i, idx;
 604
 605         /* Drop any IPv4 traffic from unprotected ports */
 606         for (i = 0; i < traffic->ip4.num; i++)
 607                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 608
 609         traffic->ip4.num = 0;
 610
 611         /* Drop any IPv6 traffic from unprotected ports */
 612         for (i = 0; i < traffic->ip6.num; i++)
 613                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 614
 615         traffic->ip6.num = 0;
 616
 617         if (app_sa_prm.enable == 0) {
 618
 619                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 620                                 traffic->ipsec.num, MAX_PKT_BURST);
 621
 622                 for (i = 0; i < nb_pkts_in; i++) {
 623                         m = traffic->ipsec.pkts[i];
 624                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 625                         if (ip->ip_v == IPVERSION) {
 626                                 idx = traffic->ip4.num++;
 627                                 traffic->ip4.pkts[idx] = m;
 628                         } else {
 629                                 idx = traffic->ip6.num++;
 630                                 traffic->ip6.pkts[idx] = m;
 631                         }
 632                 }
 633         } else {
 634                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 635                         traffic->ipsec.saptr, traffic->ipsec.num);
 636                 ipsec_process(ipsec_ctx, traffic);
 637         }
 638 }
 639
 640 static inline void
 641 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 642                 struct ipsec_traffic *traffic)
 643 {
 644         struct rte_mbuf *m;
 645         uint32_t nb_pkts_out, i, n;
 646         struct ip *ip;
 647
 648         /* Drop any IPsec traffic from protected ports */
 649         for (i = 0; i < traffic->ipsec.num; i++)
 650                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 651
 652         n = 0;
 653
 654         for (i = 0; i < traffic->ip4.num; i++) {
 655                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 656                 traffic->ipsec.res[n++] = single_sa_idx;
 657         }
 658
 659         for (i = 0; i < traffic->ip6.num; i++) {
 660                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 661                 traffic->ipsec.res[n++] = single_sa_idx;
 662         }
 663
 664         traffic->ip4.num = 0;
 665         traffic->ip6.num = 0;
 666         traffic->ipsec.num = n;
 667
 668         if (app_sa_prm.enable == 0) {
 669
 670                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 671                                 traffic->ipsec.res, traffic->ipsec.num,
 672                                 MAX_PKT_BURST);
 673
 674                 /* They all sue the same SA (ip4 or ip6 tunnel) */
 675                 m = traffic->ipsec.pkts[0];
 676                 ip = rte_pktmbuf_mtod(m, struct ip *);
 677                 if (ip->ip_v == IPVERSION) {
 678                         traffic->ip4.num = nb_pkts_out;
 679                         for (i = 0; i < nb_pkts_out; i++)
 680                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 681                 } else {
 682                         traffic->ip6.num = nb_pkts_out;
 683                         for (i = 0; i < nb_pkts_out; i++)
 684                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 685                 }
 686         } else {
 687                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 688                         traffic->ipsec.saptr, traffic->ipsec.num);
 689                 ipsec_process(ipsec_ctx, traffic);
 690         }
 691 }
 692
 693 static inline int32_t
 694 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 695 {
 696         struct ipsec_mbuf_metadata *priv;
 697         struct ipsec_sa *sa;
 698
 699         priv = get_priv(pkt);
 700
 701         sa = priv->sa;
 702         if (unlikely(sa == NULL)) {
 703                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 704                 goto fail;
 705         }
 706
 707         if (is_ipv6)
 708                 return sa->portid;
 709
 710         /* else */
 711         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 712
 713 fail:
 714         if (is_ipv6)
 715                 return -1;
 716
 717         /* else */
 718         return 0;
 719 }
 720
 721 static inline void
 722 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 723 {
 724         uint32_t hop[MAX_PKT_BURST * 2];
 725         uint32_t dst_ip[MAX_PKT_BURST * 2];
 726         int32_t pkt_hop = 0;
 727         uint16_t i, offset;
 728         uint16_t lpm_pkts = 0;
 729
 730         if (nb_pkts == 0)
 731                 return;
 732
 733         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 734          * have port ID in the SA
 735          */
 736
 737         for (i = 0; i < nb_pkts; i++) {
 738                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 739                         /* Security offload not enabled. So an LPM lookup is
 740                          * required to get the hop
 741                          */
 742                         offset = offsetof(struct ip, ip_dst);
 743                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 744                                         uint32_t *, offset);
 745                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 746                         lpm_pkts++;
 747                 }
 748         }
 749
 750         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 751
 752         lpm_pkts = 0;
 753
 754         for (i = 0; i < nb_pkts; i++) {
 755                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 756                         /* Read hop from the SA */
 757                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 758                 } else {
 759                         /* Need to use hop returned by lookup */
 760                         pkt_hop = hop[lpm_pkts++];
 761                 }
 762
 763                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 764                         rte_pktmbuf_free(pkts[i]);
 765                         continue;
 766                 }
 767                 send_single_packet(pkts[i], pkt_hop & 0xff);
 768         }
 769 }
 770
 771 static inline void
 772 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 773 {
 774         int32_t hop[MAX_PKT_BURST * 2];
 775         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 776         uint8_t *ip6_dst;
 777         int32_t pkt_hop = 0;
 778         uint16_t i, offset;
 779         uint16_t lpm_pkts = 0;
 780
 781         if (nb_pkts == 0)
 782                 return;
 783
 784         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 785          * have port ID in the SA
 786          */
 787
 788         for (i = 0; i < nb_pkts; i++) {
 789                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 790                         /* Security offload not enabled. So an LPM lookup is
 791                          * required to get the hop
 792                          */
 793                         offset = offsetof(struct ip6_hdr, ip6_dst);
 794                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 795                                         offset);
 796                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 797                         lpm_pkts++;
 798                 }
 799         }
 800
 801         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 802                         lpm_pkts);
 803
 804         lpm_pkts = 0;
 805
 806         for (i = 0; i < nb_pkts; i++) {
 807                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 808                         /* Read hop from the SA */
 809                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 810                 } else {
 811                         /* Need to use hop returned by lookup */
 812                         pkt_hop = hop[lpm_pkts++];
 813                 }
 814
 815                 if (pkt_hop == -1) {
 816                         rte_pktmbuf_free(pkts[i]);
 817                         continue;
 818                 }
 819                 send_single_packet(pkts[i], pkt_hop & 0xff);
 820         }
 821 }
 822
 823 static inline void
 824 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 825                 uint8_t nb_pkts, uint16_t portid)
 826 {
 827         struct ipsec_traffic traffic;
 828
 829         prepare_traffic(pkts, &traffic, nb_pkts);
 830
 831         if (unlikely(single_sa)) {
 832                 if (UNPROTECTED_PORT(portid))
 833                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 834                 else
 835                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 836         } else {
 837                 if (UNPROTECTED_PORT(portid))
 838                         process_pkts_inbound(&qconf->inbound, &traffic);
 839                 else
 840                         process_pkts_outbound(&qconf->outbound, &traffic);
 841         }
 842
 843         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 844         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 845 }
 846
 847 static inline void
 848 drain_tx_buffers(struct lcore_conf *qconf)
 849 {
 850         struct buffer *buf;
 851         uint32_t portid;
 852
 853         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 854                 buf = &qconf->tx_mbufs[portid];
 855                 if (buf->len == 0)
 856                         continue;
 857                 send_burst(qconf, buf->len, portid);
 858                 buf->len = 0;
 859         }
 860 }
 861
 862 static inline void
 863 drain_crypto_buffers(struct lcore_conf *qconf)
 864 {
 865         uint32_t i;
 866         struct ipsec_ctx *ctx;
 867
 868         /* drain inbound buffers*/
 869         ctx = &qconf->inbound;
 870         for (i = 0; i != ctx->nb_qps; i++) {
 871                 if (ctx->tbl[i].len != 0)
 872                         enqueue_cop_burst(ctx->tbl  + i);
 873         }
 874
 875         /* drain outbound buffers*/
 876         ctx = &qconf->outbound;
 877         for (i = 0; i != ctx->nb_qps; i++) {
 878                 if (ctx->tbl[i].len != 0)
 879                         enqueue_cop_burst(ctx->tbl  + i);
 880         }
 881 }
 882
 883 static void
 884 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
 885                 struct ipsec_ctx *ctx)
 886 {
 887         uint32_t n;
 888         struct ipsec_traffic trf;
 889
 890         if (app_sa_prm.enable == 0) {
 891
 892                 /* dequeue packets from crypto-queue */
 893                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
 894                         RTE_DIM(trf.ipsec.pkts));
 895
 896                 trf.ip4.num = 0;
 897                 trf.ip6.num = 0;
 898
 899                 /* split traffic by ipv4-ipv6 */
 900                 split46_traffic(&trf, trf.ipsec.pkts, n);
 901         } else
 902                 ipsec_cqp_process(ctx, &trf);
 903
 904         /* process ipv4 packets */
 905         if (trf.ip4.num != 0) {
 906                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
 907                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
 908         }
 909
 910         /* process ipv6 packets */
 911         if (trf.ip6.num != 0) {
 912                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
 913                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
 914         }
 915 }
 916
 917 static void
 918 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
 919                 struct ipsec_ctx *ctx)
 920 {
 921         uint32_t n;
 922         struct ipsec_traffic trf;
 923
 924         if (app_sa_prm.enable == 0) {
 925
 926                 /* dequeue packets from crypto-queue */
 927                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
 928                         RTE_DIM(trf.ipsec.pkts));
 929
 930                 trf.ip4.num = 0;
 931                 trf.ip6.num = 0;
 932
 933                 /* split traffic by ipv4-ipv6 */
 934                 split46_traffic(&trf, trf.ipsec.pkts, n);
 935         } else
 936                 ipsec_cqp_process(ctx, &trf);
 937
 938         /* process ipv4 packets */
 939         if (trf.ip4.num != 0)
 940                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
 941
 942         /* process ipv6 packets */
 943         if (trf.ip6.num != 0)
 944                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
 945 }
 946
 947 /* main processing loop */
 948 static int32_t
 949 main_loop(__attribute__((unused)) void *dummy)
 950 {
 951         struct rte_mbuf *pkts[MAX_PKT_BURST];
 952         uint32_t lcore_id;
 953         uint64_t prev_tsc, diff_tsc, cur_tsc;
 954         int32_t i, nb_rx;
 955         uint16_t portid;
 956         uint8_t queueid;
 957         struct lcore_conf *qconf;
 958         int32_t socket_id;
 959         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
 960                         / US_PER_S * BURST_TX_DRAIN_US;
 961         struct lcore_rx_queue *rxql;
 962
 963         prev_tsc = 0;
 964         lcore_id = rte_lcore_id();
 965         qconf = &lcore_conf[lcore_id];
 966         rxql = qconf->rx_queue_list;
 967         socket_id = rte_lcore_to_socket_id(lcore_id);
 968
 969         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
 970         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
 971         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
 972         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
 973         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
 974         qconf->inbound.cdev_map = cdev_map_in;
 975         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
 976         qconf->inbound.session_priv_pool =
 977                         socket_ctx[socket_id].session_priv_pool;
 978         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
 979         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
 980         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
 981         qconf->outbound.cdev_map = cdev_map_out;
 982         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
 983         qconf->outbound.session_priv_pool =
 984                         socket_ctx[socket_id].session_priv_pool;
 985
 986         if (qconf->nb_rx_queue == 0) {
 987                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
 988                         lcore_id);
 989                 return 0;
 990         }
 991
 992         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
 993
 994         for (i = 0; i < qconf->nb_rx_queue; i++) {
 995                 portid = rxql[i].port_id;
 996                 queueid = rxql[i].queue_id;
 997                 RTE_LOG(INFO, IPSEC,
 998                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 999                         lcore_id, portid, queueid);
1000         }
1001
1002         while (1) {
1003                 cur_tsc = rte_rdtsc();
1004
1005                 /* TX queue buffer drain */
1006                 diff_tsc = cur_tsc - prev_tsc;
1007
1008                 if (unlikely(diff_tsc > drain_tsc)) {
1009                         drain_tx_buffers(qconf);
1010                         drain_crypto_buffers(qconf);
1011                         prev_tsc = cur_tsc;
1012                 }
1013
1014                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1015
1016                         /* Read packets from RX queues */
1017                         portid = rxql[i].port_id;
1018                         queueid = rxql[i].queue_id;
1019                         nb_rx = rte_eth_rx_burst(portid, queueid,
1020                                         pkts, MAX_PKT_BURST);
1021
1022                         if (nb_rx > 0)
1023                                 process_pkts(qconf, pkts, nb_rx, portid);
1024
1025                         /* dequeue and process completed crypto-ops */
1026                         if (UNPROTECTED_PORT(portid))
1027                                 drain_inbound_crypto_queues(qconf,
1028                                         &qconf->inbound);
1029                         else
1030                                 drain_outbound_crypto_queues(qconf,
1031                                         &qconf->outbound);
1032                 }
1033         }
1034 }
1035
1036 static int32_t
1037 check_params(void)
1038 {
1039         uint8_t lcore;
1040         uint16_t portid;
1041         uint16_t i;
1042         int32_t socket_id;
1043
1044         if (lcore_params == NULL) {
1045                 printf("Error: No port/queue/core mappings\n");
1046                 return -1;
1047         }
1048
1049         for (i = 0; i < nb_lcore_params; ++i) {
1050                 lcore = lcore_params[i].lcore_id;
1051                 if (!rte_lcore_is_enabled(lcore)) {
1052                         printf("error: lcore %hhu is not enabled in "
1053                                 "lcore mask\n", lcore);
1054                         return -1;
1055                 }
1056                 socket_id = rte_lcore_to_socket_id(lcore);
1057                 if (socket_id != 0 && numa_on == 0) {
1058                         printf("warning: lcore %hhu is on socket %d "
1059                                 "with numa off\n",
1060                                 lcore, socket_id);
1061                 }
1062                 portid = lcore_params[i].port_id;
1063                 if ((enabled_port_mask & (1 << portid)) == 0) {
1064                         printf("port %u is not enabled in port mask\n", portid);
1065                         return -1;
1066                 }
1067                 if (!rte_eth_dev_is_valid_port(portid)) {
1068                         printf("port %u is not present on the board\n", portid);
1069                         return -1;
1070                 }
1071         }
1072         return 0;
1073 }
1074
1075 static uint8_t
1076 get_port_nb_rx_queues(const uint16_t port)
1077 {
1078         int32_t queue = -1;
1079         uint16_t i;
1080
1081         for (i = 0; i < nb_lcore_params; ++i) {
1082                 if (lcore_params[i].port_id == port &&
1083                                 lcore_params[i].queue_id > queue)
1084                         queue = lcore_params[i].queue_id;
1085         }
1086         return (uint8_t)(++queue);
1087 }
1088
1089 static int32_t
1090 init_lcore_rx_queues(void)
1091 {
1092         uint16_t i, nb_rx_queue;
1093         uint8_t lcore;
1094
1095         for (i = 0; i < nb_lcore_params; ++i) {
1096                 lcore = lcore_params[i].lcore_id;
1097                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1098                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1099                         printf("error: too many queues (%u) for lcore: %u\n",
1100                                         nb_rx_queue + 1, lcore);
1101                         return -1;
1102                 }
1103                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1104                         lcore_params[i].port_id;
1105                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1106                         lcore_params[i].queue_id;
1107                 lcore_conf[lcore].nb_rx_queue++;
1108         }
1109         return 0;
1110 }
1111
1112 /* display usage */
1113 static void
1114 print_usage(const char *prgname)
1115 {
1116         fprintf(stderr, "%s [EAL options] --"
1117                 " -p PORTMASK"
1118                 " [-P]"
1119                 " [-u PORTMASK]"
1120                 " [-j FRAMESIZE]"
1121                 " [-l]"
1122                 " [-w REPLAY_WINDOW_SIZE]"
1123                 " [-e]"
1124                 " [-a]"
1125                 " -f CONFIG_FILE"
1126                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1127                 " [--single-sa SAIDX]"
1128                 " [--cryptodev_mask MASK]"
1129                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1130                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1131                 "\n\n"
1132                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1133                 "  -P : Enable promiscuous mode\n"
1134                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1135                 "  -j FRAMESIZE: Enable jumbo frame with 'FRAMESIZE' as maximum\n"
1136                 "                packet size\n"
1137                 "  -l enables code-path that uses librte_ipsec\n"
1138                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1139                 "     size for each SA\n"
1140                 "  -e enables ESN\n"
1141                 "  -a enables SA SQN atomic behaviour\n"
1142                 "  -f CONFIG_FILE: Configuration file\n"
1143                 "  --config (port,queue,lcore): Rx queue configuration\n"
1144                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1145                 "                     bypassing the SP\n"
1146                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1147                 "                         devices to configure\n"
1148                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1149                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1150                 "                         (DEV_RX_OFFLOAD_*)\n"
1151                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1152                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1153                 "                         (DEV_TX_OFFLOAD_*)\n"
1154                 "\n",
1155                 prgname);
1156 }
1157
1158 static int
1159 parse_mask(const char *str, uint64_t *val)
1160 {
1161         char *end;
1162         unsigned long t;
1163
1164         errno = 0;
1165         t = strtoul(str, &end, 0);
1166         if (errno != 0 || end[0] != 0)
1167                 return -EINVAL;
1168
1169         *val = t;
1170         return 0;
1171 }
1172
1173 static int32_t
1174 parse_portmask(const char *portmask)
1175 {
1176         char *end = NULL;
1177         unsigned long pm;
1178
1179         /* parse hexadecimal string */
1180         pm = strtoul(portmask, &end, 16);
1181         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1182                 return -1;
1183
1184         if ((pm == 0) && errno)
1185                 return -1;
1186
1187         return pm;
1188 }
1189
1190 static int32_t
1191 parse_decimal(const char *str)
1192 {
1193         char *end = NULL;
1194         unsigned long num;
1195
1196         num = strtoul(str, &end, 10);
1197         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
1198                 return -1;
1199
1200         return num;
1201 }
1202
1203 static int32_t
1204 parse_config(const char *q_arg)
1205 {
1206         char s[256];
1207         const char *p, *p0 = q_arg;
1208         char *end;
1209         enum fieldnames {
1210                 FLD_PORT = 0,
1211                 FLD_QUEUE,
1212                 FLD_LCORE,
1213                 _NUM_FLD
1214         };
1215         unsigned long int_fld[_NUM_FLD];
1216         char *str_fld[_NUM_FLD];
1217         int32_t i;
1218         uint32_t size;
1219
1220         nb_lcore_params = 0;
1221
1222         while ((p = strchr(p0, '(')) != NULL) {
1223                 ++p;
1224                 p0 = strchr(p, ')');
1225                 if (p0 == NULL)
1226                         return -1;
1227
1228                 size = p0 - p;
1229                 if (size >= sizeof(s))
1230                         return -1;
1231
1232                 snprintf(s, sizeof(s), "%.*s", size, p);
1233                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1234                                 _NUM_FLD)
1235                         return -1;
1236                 for (i = 0; i < _NUM_FLD; i++) {
1237                         errno = 0;
1238                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1239                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1240                                 return -1;
1241                 }
1242                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1243                         printf("exceeded max number of lcore params: %hu\n",
1244                                 nb_lcore_params);
1245                         return -1;
1246                 }
1247                 lcore_params_array[nb_lcore_params].port_id =
1248                         (uint8_t)int_fld[FLD_PORT];
1249                 lcore_params_array[nb_lcore_params].queue_id =
1250                         (uint8_t)int_fld[FLD_QUEUE];
1251                 lcore_params_array[nb_lcore_params].lcore_id =
1252                         (uint8_t)int_fld[FLD_LCORE];
1253                 ++nb_lcore_params;
1254         }
1255         lcore_params = lcore_params_array;
1256         return 0;
1257 }
1258
1259 static void
1260 print_app_sa_prm(const struct app_sa_prm *prm)
1261 {
1262         printf("librte_ipsec usage: %s\n",
1263                 (prm->enable == 0) ? "disabled" : "enabled");
1264
1265         if (prm->enable == 0)
1266                 return;
1267
1268         printf("replay window size: %u\n", prm->window_size);
1269         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1270         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1271 }
1272
1273 static int32_t
1274 parse_args(int32_t argc, char **argv)
1275 {
1276         int32_t opt, ret;
1277         char **argvopt;
1278         int32_t option_index;
1279         char *prgname = argv[0];
1280         int32_t f_present = 0;
1281
1282         argvopt = argv;
1283
1284         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:",
1285                                 lgopts, &option_index)) != EOF) {
1286
1287                 switch (opt) {
1288                 case 'p':
1289                         enabled_port_mask = parse_portmask(optarg);
1290                         if (enabled_port_mask == 0) {
1291                                 printf("invalid portmask\n");
1292                                 print_usage(prgname);
1293                                 return -1;
1294                         }
1295                         break;
1296                 case 'P':
1297                         printf("Promiscuous mode selected\n");
1298                         promiscuous_on = 1;
1299                         break;
1300                 case 'u':
1301                         unprotected_port_mask = parse_portmask(optarg);
1302                         if (unprotected_port_mask == 0) {
1303                                 printf("invalid unprotected portmask\n");
1304                                 print_usage(prgname);
1305                                 return -1;
1306                         }
1307                         break;
1308                 case 'f':
1309                         if (f_present == 1) {
1310                                 printf("\"-f\" option present more than "
1311                                         "once!\n");
1312                                 print_usage(prgname);
1313                                 return -1;
1314                         }
1315                         if (parse_cfg_file(optarg) < 0) {
1316                                 printf("parsing file \"%s\" failed\n",
1317                                         optarg);
1318                                 print_usage(prgname);
1319                                 return -1;
1320                         }
1321                         f_present = 1;
1322                         break;
1323                 case 'j':
1324                         {
1325                                 int32_t size = parse_decimal(optarg);
1326                                 if (size <= 1518) {
1327                                         printf("Invalid jumbo frame size\n");
1328                                         if (size < 0) {
1329                                                 print_usage(prgname);
1330                                                 return -1;
1331                                         }
1332                                         printf("Using default value 9000\n");
1333                                         frame_size = 9000;
1334                                 } else {
1335                                         frame_size = size;
1336                                 }
1337                         }
1338                         printf("Enabled jumbo frames size %u\n", frame_size);
1339                         break;
1340                 case 'l':
1341                         app_sa_prm.enable = 1;
1342                         break;
1343                 case 'w':
1344                         app_sa_prm.enable = 1;
1345                         app_sa_prm.window_size = parse_decimal(optarg);
1346                         break;
1347                 case 'e':
1348                         app_sa_prm.enable = 1;
1349                         app_sa_prm.enable_esn = 1;
1350                         break;
1351                 case 'a':
1352                         app_sa_prm.enable = 1;
1353                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1354                         break;
1355                 case CMD_LINE_OPT_CONFIG_NUM:
1356                         ret = parse_config(optarg);
1357                         if (ret) {
1358                                 printf("Invalid config\n");
1359                                 print_usage(prgname);
1360                                 return -1;
1361                         }
1362                         break;
1363                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1364                         ret = parse_decimal(optarg);
1365                         if (ret == -1) {
1366                                 printf("Invalid argument[sa_idx]\n");
1367                                 print_usage(prgname);
1368                                 return -1;
1369                         }
1370
1371                         /* else */
1372                         single_sa = 1;
1373                         single_sa_idx = ret;
1374                         printf("Configured with single SA index %u\n",
1375                                         single_sa_idx);
1376                         break;
1377                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1378                         ret = parse_portmask(optarg);
1379                         if (ret == -1) {
1380                                 printf("Invalid argument[portmask]\n");
1381                                 print_usage(prgname);
1382                                 return -1;
1383                         }
1384
1385                         /* else */
1386                         enabled_cryptodev_mask = ret;
1387                         break;
1388                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1389                         ret = parse_mask(optarg, &dev_rx_offload);
1390                         if (ret != 0) {
1391                                 printf("Invalid argument for \'%s\': %s\n",
1392                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1393                                 print_usage(prgname);
1394                                 return -1;
1395                         }
1396                         break;
1397                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1398                         ret = parse_mask(optarg, &dev_tx_offload);
1399                         if (ret != 0) {
1400                                 printf("Invalid argument for \'%s\': %s\n",
1401                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1402                                 print_usage(prgname);
1403                                 return -1;
1404                         }
1405                         break;
1406                 default:
1407                         print_usage(prgname);
1408                         return -1;
1409                 }
1410         }
1411
1412         if (f_present == 0) {
1413                 printf("Mandatory option \"-f\" not present\n");
1414                 return -1;
1415         }
1416
1417         print_app_sa_prm(&app_sa_prm);
1418
1419         if (optind >= 0)
1420                 argv[optind-1] = prgname;
1421
1422         ret = optind-1;
1423         optind = 1; /* reset getopt lib */
1424         return ret;
1425 }
1426
1427 static void
1428 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1429 {
1430         char buf[ETHER_ADDR_FMT_SIZE];
1431         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1432         printf("%s%s", name, buf);
1433 }
1434
1435 /*
1436  * Update destination ethaddr for the port.
1437  */
1438 int
1439 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1440 {
1441         if (port >= RTE_DIM(ethaddr_tbl))
1442                 return -EINVAL;
1443
1444         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1445         return 0;
1446 }
1447
1448 /* Check the link status of all ports in up to 9s, and print them finally */
1449 static void
1450 check_all_ports_link_status(uint32_t port_mask)
1451 {
1452 #define CHECK_INTERVAL 100 /* 100ms */
1453 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1454         uint16_t portid;
1455         uint8_t count, all_ports_up, print_flag = 0;
1456         struct rte_eth_link link;
1457
1458         printf("\nChecking link status");
1459         fflush(stdout);
1460         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1461                 all_ports_up = 1;
1462                 RTE_ETH_FOREACH_DEV(portid) {
1463                         if ((port_mask & (1 << portid)) == 0)
1464                                 continue;
1465                         memset(&link, 0, sizeof(link));
1466                         rte_eth_link_get_nowait(portid, &link);
1467                         /* print link status if flag set */
1468                         if (print_flag == 1) {
1469                                 if (link.link_status)
1470                                         printf(
1471                                         "Port%d Link Up - speed %u Mbps -%s\n",
1472                                                 portid, link.link_speed,
1473                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1474                                         ("full-duplex") : ("half-duplex\n"));
1475                                 else
1476                                         printf("Port %d Link Down\n", portid);
1477                                 continue;
1478                         }
1479                         /* clear all_ports_up flag if any link down */
1480                         if (link.link_status == ETH_LINK_DOWN) {
1481                                 all_ports_up = 0;
1482                                 break;
1483                         }
1484                 }
1485                 /* after finally printing all link status, get out */
1486                 if (print_flag == 1)
1487                         break;
1488
1489                 if (all_ports_up == 0) {
1490                         printf(".");
1491                         fflush(stdout);
1492                         rte_delay_ms(CHECK_INTERVAL);
1493                 }
1494
1495                 /* set the print_flag if all ports up or timeout */
1496                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1497                         print_flag = 1;
1498                         printf("done\n");
1499                 }
1500         }
1501 }
1502
1503 static int32_t
1504 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1505                 uint16_t qp, struct lcore_params *params,
1506                 struct ipsec_ctx *ipsec_ctx,
1507                 const struct rte_cryptodev_capabilities *cipher,
1508                 const struct rte_cryptodev_capabilities *auth,
1509                 const struct rte_cryptodev_capabilities *aead)
1510 {
1511         int32_t ret = 0;
1512         unsigned long i;
1513         struct cdev_key key = { 0 };
1514
1515         key.lcore_id = params->lcore_id;
1516         if (cipher)
1517                 key.cipher_algo = cipher->sym.cipher.algo;
1518         if (auth)
1519                 key.auth_algo = auth->sym.auth.algo;
1520         if (aead)
1521                 key.aead_algo = aead->sym.aead.algo;
1522
1523         ret = rte_hash_lookup(map, &key);
1524         if (ret != -ENOENT)
1525                 return 0;
1526
1527         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1528                 if (ipsec_ctx->tbl[i].id == cdev_id)
1529                         break;
1530
1531         if (i == ipsec_ctx->nb_qps) {
1532                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1533                         printf("Maximum number of crypto devices assigned to "
1534                                 "a core, increase MAX_QP_PER_LCORE value\n");
1535                         return 0;
1536                 }
1537                 ipsec_ctx->tbl[i].id = cdev_id;
1538                 ipsec_ctx->tbl[i].qp = qp;
1539                 ipsec_ctx->nb_qps++;
1540                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1541                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1542                                 cdev_id, qp, i);
1543         }
1544
1545         ret = rte_hash_add_key_data(map, &key, (void *)i);
1546         if (ret < 0) {
1547                 printf("Faled to insert cdev mapping for (lcore %u, "
1548                                 "cdev %u, qp %u), errno %d\n",
1549                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1550                                 ipsec_ctx->tbl[i].qp, ret);
1551                 return 0;
1552         }
1553
1554         return 1;
1555 }
1556
1557 static int32_t
1558 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1559                 uint16_t qp, struct lcore_params *params)
1560 {
1561         int32_t ret = 0;
1562         const struct rte_cryptodev_capabilities *i, *j;
1563         struct rte_hash *map;
1564         struct lcore_conf *qconf;
1565         struct ipsec_ctx *ipsec_ctx;
1566         const char *str;
1567
1568         qconf = &lcore_conf[params->lcore_id];
1569
1570         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1571                 map = cdev_map_out;
1572                 ipsec_ctx = &qconf->outbound;
1573                 str = "Outbound";
1574         } else {
1575                 map = cdev_map_in;
1576                 ipsec_ctx = &qconf->inbound;
1577                 str = "Inbound";
1578         }
1579
1580         /* Required cryptodevs with operation chainning */
1581         if (!(dev_info->feature_flags &
1582                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1583                 return ret;
1584
1585         for (i = dev_info->capabilities;
1586                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1587                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1588                         continue;
1589
1590                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1591                         ret |= add_mapping(map, str, cdev_id, qp, params,
1592                                         ipsec_ctx, NULL, NULL, i);
1593                         continue;
1594                 }
1595
1596                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1597                         continue;
1598
1599                 for (j = dev_info->capabilities;
1600                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1601                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1602                                 continue;
1603
1604                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1605                                 continue;
1606
1607                         ret |= add_mapping(map, str, cdev_id, qp, params,
1608                                                 ipsec_ctx, i, j, NULL);
1609                 }
1610         }
1611
1612         return ret;
1613 }
1614
1615 /* Check if the device is enabled by cryptodev_mask */
1616 static int
1617 check_cryptodev_mask(uint8_t cdev_id)
1618 {
1619         if (enabled_cryptodev_mask & (1 << cdev_id))
1620                 return 0;
1621
1622         return -1;
1623 }
1624
1625 static int32_t
1626 cryptodevs_init(void)
1627 {
1628         struct rte_cryptodev_config dev_conf;
1629         struct rte_cryptodev_qp_conf qp_conf;
1630         uint16_t idx, max_nb_qps, qp, i;
1631         int16_t cdev_id, port_id;
1632         struct rte_hash_parameters params = { 0 };
1633
1634         params.entries = CDEV_MAP_ENTRIES;
1635         params.key_len = sizeof(struct cdev_key);
1636         params.hash_func = rte_jhash;
1637         params.hash_func_init_val = 0;
1638         params.socket_id = rte_socket_id();
1639
1640         params.name = "cdev_map_in";
1641         cdev_map_in = rte_hash_create(&params);
1642         if (cdev_map_in == NULL)
1643                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1644                                 rte_errno);
1645
1646         params.name = "cdev_map_out";
1647         cdev_map_out = rte_hash_create(&params);
1648         if (cdev_map_out == NULL)
1649                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1650                                 rte_errno);
1651
1652         printf("lcore/cryptodev/qp mappings:\n");
1653
1654         uint32_t max_sess_sz = 0, sess_sz;
1655         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1656                 void *sec_ctx;
1657
1658                 /* Get crypto priv session size */
1659                 sess_sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
1660                 if (sess_sz > max_sess_sz)
1661                         max_sess_sz = sess_sz;
1662
1663                 /*
1664                  * If crypto device is security capable, need to check the
1665                  * size of security session as well.
1666                  */
1667
1668                 /* Get security context of the crypto device */
1669                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
1670                 if (sec_ctx == NULL)
1671                         continue;
1672
1673                 /* Get size of security session */
1674                 sess_sz = rte_security_session_get_size(sec_ctx);
1675                 if (sess_sz > max_sess_sz)
1676                         max_sess_sz = sess_sz;
1677         }
1678         RTE_ETH_FOREACH_DEV(port_id) {
1679                 void *sec_ctx;
1680
1681                 if ((enabled_port_mask & (1 << port_id)) == 0)
1682                         continue;
1683
1684                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
1685                 if (sec_ctx == NULL)
1686                         continue;
1687
1688                 sess_sz = rte_security_session_get_size(sec_ctx);
1689                 if (sess_sz > max_sess_sz)
1690                         max_sess_sz = sess_sz;
1691         }
1692
1693         idx = 0;
1694         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1695                 struct rte_cryptodev_info cdev_info;
1696
1697                 if (check_cryptodev_mask((uint8_t)cdev_id))
1698                         continue;
1699
1700                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1701
1702                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1703                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1704                 else
1705                         max_nb_qps = nb_lcore_params;
1706
1707                 qp = 0;
1708                 i = 0;
1709                 while (qp < max_nb_qps && i < nb_lcore_params) {
1710                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1711                                                 &lcore_params[idx]))
1712                                 qp++;
1713                         idx++;
1714                         idx = idx % nb_lcore_params;
1715                         i++;
1716                 }
1717
1718                 if (qp == 0)
1719                         continue;
1720
1721                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1722                 dev_conf.nb_queue_pairs = qp;
1723
1724                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1725                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1726                         rte_exit(EXIT_FAILURE,
1727                                 "Device does not support at least %u "
1728                                 "sessions", CDEV_MP_NB_OBJS);
1729
1730                 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1731                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1732                         struct rte_mempool *sess_mp;
1733
1734                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1735                                         "sess_mp_%u", dev_conf.socket_id);
1736                         sess_mp = rte_cryptodev_sym_session_pool_create(
1737                                         mp_name, CDEV_MP_NB_OBJS,
1738                                         0, CDEV_MP_CACHE_SZ, 0,
1739                                         dev_conf.socket_id);
1740                         socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1741                 }
1742
1743                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool) {
1744                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1745                         struct rte_mempool *sess_mp;
1746
1747                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1748                                         "sess_mp_priv_%u", dev_conf.socket_id);
1749                         sess_mp = rte_mempool_create(mp_name,
1750                                         CDEV_MP_NB_OBJS,
1751                                         max_sess_sz,
1752                                         CDEV_MP_CACHE_SZ,
1753                                         0, NULL, NULL, NULL,
1754                                         NULL, dev_conf.socket_id,
1755                                         0);
1756                         socket_ctx[dev_conf.socket_id].session_priv_pool =
1757                                         sess_mp;
1758                 }
1759
1760                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool ||
1761                                 !socket_ctx[dev_conf.socket_id].session_pool)
1762                         rte_exit(EXIT_FAILURE,
1763                                 "Cannot create session pool on socket %d\n",
1764                                 dev_conf.socket_id);
1765                 else
1766                         printf("Allocated session pool on socket %d\n",
1767                                         dev_conf.socket_id);
1768
1769                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1770                         rte_panic("Failed to initialize cryptodev %u\n",
1771                                         cdev_id);
1772
1773                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1774                 qp_conf.mp_session =
1775                         socket_ctx[dev_conf.socket_id].session_pool;
1776                 qp_conf.mp_session_private =
1777                         socket_ctx[dev_conf.socket_id].session_priv_pool;
1778                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1779                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1780                                         &qp_conf, dev_conf.socket_id))
1781                                 rte_panic("Failed to setup queue %u for "
1782                                                 "cdev_id %u\n", 0, cdev_id);
1783
1784                 if (rte_cryptodev_start(cdev_id))
1785                         rte_panic("Failed to start cryptodev %u\n",
1786                                         cdev_id);
1787         }
1788
1789         /* create session pools for eth devices that implement security */
1790         RTE_ETH_FOREACH_DEV(port_id) {
1791                 if ((enabled_port_mask & (1 << port_id)) &&
1792                                 rte_eth_dev_get_sec_ctx(port_id)) {
1793                         int socket_id = rte_eth_dev_socket_id(port_id);
1794
1795                         if (!socket_ctx[socket_id].session_priv_pool) {
1796                                 char mp_name[RTE_MEMPOOL_NAMESIZE];
1797                                 struct rte_mempool *sess_mp;
1798
1799                                 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1800                                                 "sess_mp_%u", socket_id);
1801                                 sess_mp = rte_mempool_create(mp_name,
1802                                                 (CDEV_MP_NB_OBJS * 2),
1803                                                 max_sess_sz,
1804                                                 CDEV_MP_CACHE_SZ,
1805                                                 0, NULL, NULL, NULL,
1806                                                 NULL, socket_id,
1807                                                 0);
1808                                 if (sess_mp == NULL)
1809                                         rte_exit(EXIT_FAILURE,
1810                                                 "Cannot create session pool "
1811                                                 "on socket %d\n", socket_id);
1812                                 else
1813                                         printf("Allocated session pool "
1814                                                 "on socket %d\n", socket_id);
1815                                 socket_ctx[socket_id].session_priv_pool =
1816                                                 sess_mp;
1817                         }
1818                 }
1819         }
1820
1821
1822         printf("\n");
1823
1824         return 0;
1825 }
1826
1827 static void
1828 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
1829 {
1830         struct rte_eth_dev_info dev_info;
1831         struct rte_eth_txconf *txconf;
1832         uint16_t nb_tx_queue, nb_rx_queue;
1833         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1834         int32_t ret, socket_id;
1835         struct lcore_conf *qconf;
1836         struct rte_ether_addr ethaddr;
1837         struct rte_eth_conf local_port_conf = port_conf;
1838
1839         rte_eth_dev_info_get(portid, &dev_info);
1840
1841         /* limit allowed HW offloafs, as user requested */
1842         dev_info.rx_offload_capa &= dev_rx_offload;
1843         dev_info.tx_offload_capa &= dev_tx_offload;
1844
1845         printf("Configuring device port %u:\n", portid);
1846
1847         rte_eth_macaddr_get(portid, &ethaddr);
1848         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
1849         print_ethaddr("Address: ", &ethaddr);
1850         printf("\n");
1851
1852         nb_rx_queue = get_port_nb_rx_queues(portid);
1853         nb_tx_queue = nb_lcores;
1854
1855         if (nb_rx_queue > dev_info.max_rx_queues)
1856                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1857                                 "(max rx queue is %u)\n",
1858                                 nb_rx_queue, dev_info.max_rx_queues);
1859
1860         if (nb_tx_queue > dev_info.max_tx_queues)
1861                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1862                                 "(max tx queue is %u)\n",
1863                                 nb_tx_queue, dev_info.max_tx_queues);
1864
1865         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1866                         nb_rx_queue, nb_tx_queue);
1867
1868         if (frame_size) {
1869                 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
1870                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1871         }
1872
1873         local_port_conf.rxmode.offloads |= req_rx_offloads;
1874         local_port_conf.txmode.offloads |= req_tx_offloads;
1875
1876         /* Check that all required capabilities are supported */
1877         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
1878                         local_port_conf.rxmode.offloads)
1879                 rte_exit(EXIT_FAILURE,
1880                         "Error: port %u required RX offloads: 0x%" PRIx64
1881                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
1882                         portid, local_port_conf.rxmode.offloads,
1883                         dev_info.rx_offload_capa);
1884
1885         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
1886                         local_port_conf.txmode.offloads)
1887                 rte_exit(EXIT_FAILURE,
1888                         "Error: port %u required TX offloads: 0x%" PRIx64
1889                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
1890                         portid, local_port_conf.txmode.offloads,
1891                         dev_info.tx_offload_capa);
1892
1893         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1894                 local_port_conf.txmode.offloads |=
1895                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1896
1897         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
1898                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
1899
1900         printf("port %u configurng rx_offloads=0x%" PRIx64
1901                 ", tx_offloads=0x%" PRIx64 "\n",
1902                 portid, local_port_conf.rxmode.offloads,
1903                 local_port_conf.txmode.offloads);
1904
1905         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1906                 dev_info.flow_type_rss_offloads;
1907         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1908                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
1909                 printf("Port %u modified RSS hash function based on hardware support,"
1910                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1911                         portid,
1912                         port_conf.rx_adv_conf.rss_conf.rss_hf,
1913                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1914         }
1915
1916         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1917                         &local_port_conf);
1918         if (ret < 0)
1919                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1920                                 "err=%d, port=%d\n", ret, portid);
1921
1922         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
1923         if (ret < 0)
1924                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
1925                                 "err=%d, port=%d\n", ret, portid);
1926
1927         /* init one TX queue per lcore */
1928         tx_queueid = 0;
1929         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1930                 if (rte_lcore_is_enabled(lcore_id) == 0)
1931                         continue;
1932
1933                 if (numa_on)
1934                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1935                 else
1936                         socket_id = 0;
1937
1938                 /* init TX queue */
1939                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1940
1941                 txconf = &dev_info.default_txconf;
1942                 txconf->offloads = local_port_conf.txmode.offloads;
1943
1944                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1945                                 socket_id, txconf);
1946                 if (ret < 0)
1947                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1948                                         "err=%d, port=%d\n", ret, portid);
1949
1950                 qconf = &lcore_conf[lcore_id];
1951                 qconf->tx_queue_id[portid] = tx_queueid;
1952
1953                 /* Pre-populate pkt offloads based on capabilities */
1954                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
1955                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
1956                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
1957                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
1958
1959                 tx_queueid++;
1960
1961                 /* init RX queues */
1962                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1963                         struct rte_eth_rxconf rxq_conf;
1964
1965                         if (portid != qconf->rx_queue_list[queue].port_id)
1966                                 continue;
1967
1968                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
1969
1970                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1971                                         socket_id);
1972
1973                         rxq_conf = dev_info.default_rxconf;
1974                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
1975                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1976                                         nb_rxd, socket_id, &rxq_conf,
1977                                         socket_ctx[socket_id].mbuf_pool);
1978                         if (ret < 0)
1979                                 rte_exit(EXIT_FAILURE,
1980                                         "rte_eth_rx_queue_setup: err=%d, "
1981                                         "port=%d\n", ret, portid);
1982                 }
1983         }
1984         printf("\n");
1985 }
1986
1987 static void
1988 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1989 {
1990         char s[64];
1991         uint32_t buff_size = frame_size ? (frame_size + RTE_PKTMBUF_HEADROOM) :
1992                         RTE_MBUF_DEFAULT_BUF_SIZE;
1993
1994
1995         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1996         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1997                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1998                         buff_size,
1999                         socket_id);
2000         if (ctx->mbuf_pool == NULL)
2001                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2002                                 socket_id);
2003         else
2004                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2005 }
2006
2007 static inline int
2008 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2009 {
2010         struct ipsec_sa *sa;
2011
2012         /* For inline protocol processing, the metadata in the event will
2013          * uniquely identify the security session which raised the event.
2014          * Application would then need the userdata it had registered with the
2015          * security session to process the event.
2016          */
2017
2018         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2019
2020         if (sa == NULL) {
2021                 /* userdata could not be retrieved */
2022                 return -1;
2023         }
2024
2025         /* Sequence number over flow. SA need to be re-established */
2026         RTE_SET_USED(sa);
2027         return 0;
2028 }
2029
2030 static int
2031 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2032                  void *param, void *ret_param)
2033 {
2034         uint64_t md;
2035         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2036         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2037                                         rte_eth_dev_get_sec_ctx(port_id);
2038
2039         RTE_SET_USED(param);
2040
2041         if (type != RTE_ETH_EVENT_IPSEC)
2042                 return -1;
2043
2044         event_desc = ret_param;
2045         if (event_desc == NULL) {
2046                 printf("Event descriptor not set\n");
2047                 return -1;
2048         }
2049
2050         md = event_desc->metadata;
2051
2052         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2053                 return inline_ipsec_event_esn_overflow(ctx, md);
2054         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2055                 printf("Invalid IPsec event reported\n");
2056                 return -1;
2057         }
2058
2059         return -1;
2060 }
2061
2062 int32_t
2063 main(int32_t argc, char **argv)
2064 {
2065         int32_t ret;
2066         uint32_t lcore_id;
2067         uint8_t socket_id;
2068         uint16_t portid;
2069         uint64_t req_rx_offloads, req_tx_offloads;
2070
2071         /* init EAL */
2072         ret = rte_eal_init(argc, argv);
2073         if (ret < 0)
2074                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2075         argc -= ret;
2076         argv += ret;
2077
2078         /* parse application arguments (after the EAL ones) */
2079         ret = parse_args(argc, argv);
2080         if (ret < 0)
2081                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2082
2083         if ((unprotected_port_mask & enabled_port_mask) !=
2084                         unprotected_port_mask)
2085                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2086                                 unprotected_port_mask);
2087
2088         if (check_params() < 0)
2089                 rte_exit(EXIT_FAILURE, "check_params failed\n");
2090
2091         ret = init_lcore_rx_queues();
2092         if (ret < 0)
2093                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2094
2095         nb_lcores = rte_lcore_count();
2096
2097         /* Replicate each context per socket */
2098         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2099                 if (rte_lcore_is_enabled(lcore_id) == 0)
2100                         continue;
2101
2102                 if (numa_on)
2103                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2104                 else
2105                         socket_id = 0;
2106
2107                 if (socket_ctx[socket_id].mbuf_pool)
2108                         continue;
2109
2110                 /* initilaze SPD */
2111                 sp4_init(&socket_ctx[socket_id], socket_id);
2112
2113                 sp6_init(&socket_ctx[socket_id], socket_id);
2114
2115                 /* initilaze SAD */
2116                 sa_init(&socket_ctx[socket_id], socket_id);
2117
2118                 rt_init(&socket_ctx[socket_id], socket_id);
2119
2120                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2121         }
2122
2123         RTE_ETH_FOREACH_DEV(portid) {
2124                 if ((enabled_port_mask & (1 << portid)) == 0)
2125                         continue;
2126
2127                 sa_check_offloads(portid, &req_rx_offloads, &req_tx_offloads);
2128                 port_init(portid, req_rx_offloads, req_tx_offloads);
2129         }
2130
2131         cryptodevs_init();
2132
2133         /* start ports */
2134         RTE_ETH_FOREACH_DEV(portid) {
2135                 if ((enabled_port_mask & (1 << portid)) == 0)
2136                         continue;
2137
2138                 /* Start device */
2139                 ret = rte_eth_dev_start(portid);
2140                 if (ret < 0)
2141                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2142                                         "err=%d, port=%d\n", ret, portid);
2143                 /*
2144                  * If enabled, put device in promiscuous mode.
2145                  * This allows IO forwarding mode to forward packets
2146                  * to itself through 2 cross-connected  ports of the
2147                  * target machine.
2148                  */
2149                 if (promiscuous_on)
2150                         rte_eth_promiscuous_enable(portid);
2151
2152                 rte_eth_dev_callback_register(portid,
2153                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2154         }
2155
2156         check_all_ports_link_status(enabled_port_mask);
2157
2158         /* launch per-lcore init on every lcore */
2159         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2160         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2161                 if (rte_eal_wait_lcore(lcore_id) < 0)
2162                         return -1;
2163         }
2164
2165         return 0;
2166 }