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 ether_hdr *eth;
 237
 238         eth = rte_pktmbuf_mtod(pkt, struct 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 ether_hdr *ethhdr;
 329
 330         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 331
 332         ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
 333
 334         if (ip->ip_v == IPVERSION) {
 335                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 336                 pkt->l3_len = sizeof(struct ip);
 337                 pkt->l2_len = ETHER_HDR_LEN;
 338
 339                 ip->ip_sum = 0;
 340
 341                 /* calculate IPv4 cksum in SW */
 342                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 343                         ip->ip_sum = rte_ipv4_cksum((struct ipv4_hdr *)ip);
 344
 345                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
 346         } else {
 347                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 348                 pkt->l3_len = sizeof(struct ip6_hdr);
 349                 pkt->l2_len = ETHER_HDR_LEN;
 350
 351                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
 352         }
 353
 354         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 355                         sizeof(struct ether_addr));
 356         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 357                         sizeof(struct ether_addr));
 358 }
 359
 360 static inline void
 361 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 362                 const struct lcore_conf *qconf)
 363 {
 364         int32_t i;
 365         const int32_t prefetch_offset = 2;
 366
 367         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 368                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 369                 prepare_tx_pkt(pkts[i], port, qconf);
 370         }
 371         /* Process left packets */
 372         for (; i < nb_pkts; i++)
 373                 prepare_tx_pkt(pkts[i], port, qconf);
 374 }
 375
 376 /* Send burst of packets on an output interface */
 377 static inline int32_t
 378 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 379 {
 380         struct rte_mbuf **m_table;
 381         int32_t ret;
 382         uint16_t queueid;
 383
 384         queueid = qconf->tx_queue_id[port];
 385         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 386
 387         prepare_tx_burst(m_table, n, port, qconf);
 388
 389         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 390         if (unlikely(ret < n)) {
 391                 do {
 392                         rte_pktmbuf_free(m_table[ret]);
 393                 } while (++ret < n);
 394         }
 395
 396         return 0;
 397 }
 398
 399 /* Enqueue a single packet, and send burst if queue is filled */
 400 static inline int32_t
 401 send_single_packet(struct rte_mbuf *m, uint16_t port)
 402 {
 403         uint32_t lcore_id;
 404         uint16_t len;
 405         struct lcore_conf *qconf;
 406
 407         lcore_id = rte_lcore_id();
 408
 409         qconf = &lcore_conf[lcore_id];
 410         len = qconf->tx_mbufs[port].len;
 411         qconf->tx_mbufs[port].m_table[len] = m;
 412         len++;
 413
 414         /* enough pkts to be sent */
 415         if (unlikely(len == MAX_PKT_BURST)) {
 416                 send_burst(qconf, MAX_PKT_BURST, port);
 417                 len = 0;
 418         }
 419
 420         qconf->tx_mbufs[port].len = len;
 421         return 0;
 422 }
 423
 424 static inline void
 425 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 426                 uint16_t lim)
 427 {
 428         struct rte_mbuf *m;
 429         uint32_t i, j, res, sa_idx;
 430
 431         if (ip->num == 0 || sp == NULL)
 432                 return;
 433
 434         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 435                         ip->num, DEFAULT_MAX_CATEGORIES);
 436
 437         j = 0;
 438         for (i = 0; i < ip->num; i++) {
 439                 m = ip->pkts[i];
 440                 res = ip->res[i];
 441                 if (res & BYPASS) {
 442                         ip->pkts[j++] = m;
 443                         continue;
 444                 }
 445                 if (res & DISCARD) {
 446                         rte_pktmbuf_free(m);
 447                         continue;
 448                 }
 449
 450                 /* Only check SPI match for processed IPSec packets */
 451                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 452                         rte_pktmbuf_free(m);
 453                         continue;
 454                 }
 455
 456                 sa_idx = ip->res[i] & PROTECT_MASK;
 457                 if (sa_idx >= IPSEC_SA_MAX_ENTRIES ||
 458                                 !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 = ip->res[i] & PROTECT_MASK;
 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 if (sa_idx < IPSEC_SA_MAX_ENTRIES) {
 550                         ipsec->res[ipsec->num] = sa_idx;
 551                         ipsec->pkts[ipsec->num++] = m;
 552                 } else /* invalid SA idx */
 553                         rte_pktmbuf_free(m);
 554         }
 555         ip->num = j;
 556 }
 557
 558 static inline void
 559 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 560                 struct ipsec_traffic *traffic)
 561 {
 562         struct rte_mbuf *m;
 563         uint16_t idx, nb_pkts_out, i;
 564
 565         /* Drop any IPsec traffic from protected ports */
 566         for (i = 0; i < traffic->ipsec.num; i++)
 567                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 568
 569         traffic->ipsec.num = 0;
 570
 571         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 572
 573         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 574
 575         if (app_sa_prm.enable == 0) {
 576
 577                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 578                                 traffic->ipsec.res, traffic->ipsec.num,
 579                                 MAX_PKT_BURST);
 580
 581                 for (i = 0; i < nb_pkts_out; i++) {
 582                         m = traffic->ipsec.pkts[i];
 583                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 584                         if (ip->ip_v == IPVERSION) {
 585                                 idx = traffic->ip4.num++;
 586                                 traffic->ip4.pkts[idx] = m;
 587                         } else {
 588                                 idx = traffic->ip6.num++;
 589                                 traffic->ip6.pkts[idx] = m;
 590                         }
 591                 }
 592         } else {
 593                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 594                         traffic->ipsec.saptr, traffic->ipsec.num);
 595                 ipsec_process(ipsec_ctx, traffic);
 596         }
 597 }
 598
 599 static inline void
 600 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 601                 struct ipsec_traffic *traffic)
 602 {
 603         struct rte_mbuf *m;
 604         uint32_t nb_pkts_in, i, idx;
 605
 606         /* Drop any IPv4 traffic from unprotected ports */
 607         for (i = 0; i < traffic->ip4.num; i++)
 608                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 609
 610         traffic->ip4.num = 0;
 611
 612         /* Drop any IPv6 traffic from unprotected ports */
 613         for (i = 0; i < traffic->ip6.num; i++)
 614                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 615
 616         traffic->ip6.num = 0;
 617
 618         if (app_sa_prm.enable == 0) {
 619
 620                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 621                                 traffic->ipsec.num, MAX_PKT_BURST);
 622
 623                 for (i = 0; i < nb_pkts_in; i++) {
 624                         m = traffic->ipsec.pkts[i];
 625                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 626                         if (ip->ip_v == IPVERSION) {
 627                                 idx = traffic->ip4.num++;
 628                                 traffic->ip4.pkts[idx] = m;
 629                         } else {
 630                                 idx = traffic->ip6.num++;
 631                                 traffic->ip6.pkts[idx] = m;
 632                         }
 633                 }
 634         } else {
 635                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 636                         traffic->ipsec.saptr, traffic->ipsec.num);
 637                 ipsec_process(ipsec_ctx, traffic);
 638         }
 639 }
 640
 641 static inline void
 642 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 643                 struct ipsec_traffic *traffic)
 644 {
 645         struct rte_mbuf *m;
 646         uint32_t nb_pkts_out, i, n;
 647         struct ip *ip;
 648
 649         /* Drop any IPsec traffic from protected ports */
 650         for (i = 0; i < traffic->ipsec.num; i++)
 651                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 652
 653         n = 0;
 654
 655         for (i = 0; i < traffic->ip4.num; i++) {
 656                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 657                 traffic->ipsec.res[n++] = single_sa_idx;
 658         }
 659
 660         for (i = 0; i < traffic->ip6.num; i++) {
 661                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 662                 traffic->ipsec.res[n++] = single_sa_idx;
 663         }
 664
 665         traffic->ip4.num = 0;
 666         traffic->ip6.num = 0;
 667         traffic->ipsec.num = n;
 668
 669         if (app_sa_prm.enable == 0) {
 670
 671                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 672                                 traffic->ipsec.res, traffic->ipsec.num,
 673                                 MAX_PKT_BURST);
 674
 675                 /* They all sue the same SA (ip4 or ip6 tunnel) */
 676                 m = traffic->ipsec.pkts[0];
 677                 ip = rte_pktmbuf_mtod(m, struct ip *);
 678                 if (ip->ip_v == IPVERSION) {
 679                         traffic->ip4.num = nb_pkts_out;
 680                         for (i = 0; i < nb_pkts_out; i++)
 681                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 682                 } else {
 683                         traffic->ip6.num = nb_pkts_out;
 684                         for (i = 0; i < nb_pkts_out; i++)
 685                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 686                 }
 687         } else {
 688                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 689                         traffic->ipsec.saptr, traffic->ipsec.num);
 690                 ipsec_process(ipsec_ctx, traffic);
 691         }
 692 }
 693
 694 static inline int32_t
 695 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 696 {
 697         struct ipsec_mbuf_metadata *priv;
 698         struct ipsec_sa *sa;
 699
 700         priv = get_priv(pkt);
 701
 702         sa = priv->sa;
 703         if (unlikely(sa == NULL)) {
 704                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 705                 goto fail;
 706         }
 707
 708         if (is_ipv6)
 709                 return sa->portid;
 710
 711         /* else */
 712         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 713
 714 fail:
 715         if (is_ipv6)
 716                 return -1;
 717
 718         /* else */
 719         return 0;
 720 }
 721
 722 static inline void
 723 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 724 {
 725         uint32_t hop[MAX_PKT_BURST * 2];
 726         uint32_t dst_ip[MAX_PKT_BURST * 2];
 727         int32_t pkt_hop = 0;
 728         uint16_t i, offset;
 729         uint16_t lpm_pkts = 0;
 730
 731         if (nb_pkts == 0)
 732                 return;
 733
 734         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 735          * have port ID in the SA
 736          */
 737
 738         for (i = 0; i < nb_pkts; i++) {
 739                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 740                         /* Security offload not enabled. So an LPM lookup is
 741                          * required to get the hop
 742                          */
 743                         offset = offsetof(struct ip, ip_dst);
 744                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 745                                         uint32_t *, offset);
 746                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 747                         lpm_pkts++;
 748                 }
 749         }
 750
 751         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 752
 753         lpm_pkts = 0;
 754
 755         for (i = 0; i < nb_pkts; i++) {
 756                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 757                         /* Read hop from the SA */
 758                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 759                 } else {
 760                         /* Need to use hop returned by lookup */
 761                         pkt_hop = hop[lpm_pkts++];
 762                 }
 763
 764                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 765                         rte_pktmbuf_free(pkts[i]);
 766                         continue;
 767                 }
 768                 send_single_packet(pkts[i], pkt_hop & 0xff);
 769         }
 770 }
 771
 772 static inline void
 773 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 774 {
 775         int32_t hop[MAX_PKT_BURST * 2];
 776         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 777         uint8_t *ip6_dst;
 778         int32_t pkt_hop = 0;
 779         uint16_t i, offset;
 780         uint16_t lpm_pkts = 0;
 781
 782         if (nb_pkts == 0)
 783                 return;
 784
 785         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 786          * have port ID in the SA
 787          */
 788
 789         for (i = 0; i < nb_pkts; i++) {
 790                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 791                         /* Security offload not enabled. So an LPM lookup is
 792                          * required to get the hop
 793                          */
 794                         offset = offsetof(struct ip6_hdr, ip6_dst);
 795                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 796                                         offset);
 797                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 798                         lpm_pkts++;
 799                 }
 800         }
 801
 802         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 803                         lpm_pkts);
 804
 805         lpm_pkts = 0;
 806
 807         for (i = 0; i < nb_pkts; i++) {
 808                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 809                         /* Read hop from the SA */
 810                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 811                 } else {
 812                         /* Need to use hop returned by lookup */
 813                         pkt_hop = hop[lpm_pkts++];
 814                 }
 815
 816                 if (pkt_hop == -1) {
 817                         rte_pktmbuf_free(pkts[i]);
 818                         continue;
 819                 }
 820                 send_single_packet(pkts[i], pkt_hop & 0xff);
 821         }
 822 }
 823
 824 static inline void
 825 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 826                 uint8_t nb_pkts, uint16_t portid)
 827 {
 828         struct ipsec_traffic traffic;
 829
 830         prepare_traffic(pkts, &traffic, nb_pkts);
 831
 832         if (unlikely(single_sa)) {
 833                 if (UNPROTECTED_PORT(portid))
 834                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 835                 else
 836                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 837         } else {
 838                 if (UNPROTECTED_PORT(portid))
 839                         process_pkts_inbound(&qconf->inbound, &traffic);
 840                 else
 841                         process_pkts_outbound(&qconf->outbound, &traffic);
 842         }
 843
 844         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 845         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 846 }
 847
 848 static inline void
 849 drain_tx_buffers(struct lcore_conf *qconf)
 850 {
 851         struct buffer *buf;
 852         uint32_t portid;
 853
 854         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 855                 buf = &qconf->tx_mbufs[portid];
 856                 if (buf->len == 0)
 857                         continue;
 858                 send_burst(qconf, buf->len, portid);
 859                 buf->len = 0;
 860         }
 861 }
 862
 863 static inline void
 864 drain_crypto_buffers(struct lcore_conf *qconf)
 865 {
 866         uint32_t i;
 867         struct ipsec_ctx *ctx;
 868
 869         /* drain inbound buffers*/
 870         ctx = &qconf->inbound;
 871         for (i = 0; i != ctx->nb_qps; i++) {
 872                 if (ctx->tbl[i].len != 0)
 873                         enqueue_cop_burst(ctx->tbl  + i);
 874         }
 875
 876         /* drain outbound buffers*/
 877         ctx = &qconf->outbound;
 878         for (i = 0; i != ctx->nb_qps; i++) {
 879                 if (ctx->tbl[i].len != 0)
 880                         enqueue_cop_burst(ctx->tbl  + i);
 881         }
 882 }
 883
 884 static void
 885 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
 886                 struct ipsec_ctx *ctx)
 887 {
 888         uint32_t n;
 889         struct ipsec_traffic trf;
 890
 891         if (app_sa_prm.enable == 0) {
 892
 893                 /* dequeue packets from crypto-queue */
 894                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
 895                         RTE_DIM(trf.ipsec.pkts));
 896
 897                 trf.ip4.num = 0;
 898                 trf.ip6.num = 0;
 899
 900                 /* split traffic by ipv4-ipv6 */
 901                 split46_traffic(&trf, trf.ipsec.pkts, n);
 902         } else
 903                 ipsec_cqp_process(ctx, &trf);
 904
 905         /* process ipv4 packets */
 906         if (trf.ip4.num != 0) {
 907                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
 908                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
 909         }
 910
 911         /* process ipv6 packets */
 912         if (trf.ip6.num != 0) {
 913                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
 914                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
 915         }
 916 }
 917
 918 static void
 919 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
 920                 struct ipsec_ctx *ctx)
 921 {
 922         uint32_t n;
 923         struct ipsec_traffic trf;
 924
 925         if (app_sa_prm.enable == 0) {
 926
 927                 /* dequeue packets from crypto-queue */
 928                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
 929                         RTE_DIM(trf.ipsec.pkts));
 930
 931                 trf.ip4.num = 0;
 932                 trf.ip6.num = 0;
 933
 934                 /* split traffic by ipv4-ipv6 */
 935                 split46_traffic(&trf, trf.ipsec.pkts, n);
 936         } else
 937                 ipsec_cqp_process(ctx, &trf);
 938
 939         /* process ipv4 packets */
 940         if (trf.ip4.num != 0)
 941                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
 942
 943         /* process ipv6 packets */
 944         if (trf.ip6.num != 0)
 945                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
 946 }
 947
 948 /* main processing loop */
 949 static int32_t
 950 main_loop(__attribute__((unused)) void *dummy)
 951 {
 952         struct rte_mbuf *pkts[MAX_PKT_BURST];
 953         uint32_t lcore_id;
 954         uint64_t prev_tsc, diff_tsc, cur_tsc;
 955         int32_t i, nb_rx;
 956         uint16_t portid;
 957         uint8_t queueid;
 958         struct lcore_conf *qconf;
 959         int32_t socket_id;
 960         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
 961                         / US_PER_S * BURST_TX_DRAIN_US;
 962         struct lcore_rx_queue *rxql;
 963
 964         prev_tsc = 0;
 965         lcore_id = rte_lcore_id();
 966         qconf = &lcore_conf[lcore_id];
 967         rxql = qconf->rx_queue_list;
 968         socket_id = rte_lcore_to_socket_id(lcore_id);
 969
 970         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
 971         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
 972         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
 973         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
 974         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
 975         qconf->inbound.cdev_map = cdev_map_in;
 976         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
 977         qconf->inbound.session_priv_pool =
 978                         socket_ctx[socket_id].session_priv_pool;
 979         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
 980         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
 981         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
 982         qconf->outbound.cdev_map = cdev_map_out;
 983         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
 984         qconf->outbound.session_priv_pool =
 985                         socket_ctx[socket_id].session_priv_pool;
 986
 987         if (qconf->nb_rx_queue == 0) {
 988                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
 989                         lcore_id);
 990                 return 0;
 991         }
 992
 993         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
 994
 995         for (i = 0; i < qconf->nb_rx_queue; i++) {
 996                 portid = rxql[i].port_id;
 997                 queueid = rxql[i].queue_id;
 998                 RTE_LOG(INFO, IPSEC,
 999                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1000                         lcore_id, portid, queueid);
1001         }
1002
1003         while (1) {
1004                 cur_tsc = rte_rdtsc();
1005
1006                 /* TX queue buffer drain */
1007                 diff_tsc = cur_tsc - prev_tsc;
1008
1009                 if (unlikely(diff_tsc > drain_tsc)) {
1010                         drain_tx_buffers(qconf);
1011                         drain_crypto_buffers(qconf);
1012                         prev_tsc = cur_tsc;
1013                 }
1014
1015                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1016
1017                         /* Read packets from RX queues */
1018                         portid = rxql[i].port_id;
1019                         queueid = rxql[i].queue_id;
1020                         nb_rx = rte_eth_rx_burst(portid, queueid,
1021                                         pkts, MAX_PKT_BURST);
1022
1023                         if (nb_rx > 0)
1024                                 process_pkts(qconf, pkts, nb_rx, portid);
1025
1026                         /* dequeue and process completed crypto-ops */
1027                         if (UNPROTECTED_PORT(portid))
1028                                 drain_inbound_crypto_queues(qconf,
1029                                         &qconf->inbound);
1030                         else
1031                                 drain_outbound_crypto_queues(qconf,
1032                                         &qconf->outbound);
1033                 }
1034         }
1035 }
1036
1037 static int32_t
1038 check_params(void)
1039 {
1040         uint8_t lcore;
1041         uint16_t portid;
1042         uint16_t i;
1043         int32_t socket_id;
1044
1045         if (lcore_params == NULL) {
1046                 printf("Error: No port/queue/core mappings\n");
1047                 return -1;
1048         }
1049
1050         for (i = 0; i < nb_lcore_params; ++i) {
1051                 lcore = lcore_params[i].lcore_id;
1052                 if (!rte_lcore_is_enabled(lcore)) {
1053                         printf("error: lcore %hhu is not enabled in "
1054                                 "lcore mask\n", lcore);
1055                         return -1;
1056                 }
1057                 socket_id = rte_lcore_to_socket_id(lcore);
1058                 if (socket_id != 0 && numa_on == 0) {
1059                         printf("warning: lcore %hhu is on socket %d "
1060                                 "with numa off\n",
1061                                 lcore, socket_id);
1062                 }
1063                 portid = lcore_params[i].port_id;
1064                 if ((enabled_port_mask & (1 << portid)) == 0) {
1065                         printf("port %u is not enabled in port mask\n", portid);
1066                         return -1;
1067                 }
1068                 if (!rte_eth_dev_is_valid_port(portid)) {
1069                         printf("port %u is not present on the board\n", portid);
1070                         return -1;
1071                 }
1072         }
1073         return 0;
1074 }
1075
1076 static uint8_t
1077 get_port_nb_rx_queues(const uint16_t port)
1078 {
1079         int32_t queue = -1;
1080         uint16_t i;
1081
1082         for (i = 0; i < nb_lcore_params; ++i) {
1083                 if (lcore_params[i].port_id == port &&
1084                                 lcore_params[i].queue_id > queue)
1085                         queue = lcore_params[i].queue_id;
1086         }
1087         return (uint8_t)(++queue);
1088 }
1089
1090 static int32_t
1091 init_lcore_rx_queues(void)
1092 {
1093         uint16_t i, nb_rx_queue;
1094         uint8_t lcore;
1095
1096         for (i = 0; i < nb_lcore_params; ++i) {
1097                 lcore = lcore_params[i].lcore_id;
1098                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1099                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1100                         printf("error: too many queues (%u) for lcore: %u\n",
1101                                         nb_rx_queue + 1, lcore);
1102                         return -1;
1103                 }
1104                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1105                         lcore_params[i].port_id;
1106                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1107                         lcore_params[i].queue_id;
1108                 lcore_conf[lcore].nb_rx_queue++;
1109         }
1110         return 0;
1111 }
1112
1113 /* display usage */
1114 static void
1115 print_usage(const char *prgname)
1116 {
1117         fprintf(stderr, "%s [EAL options] --"
1118                 " -p PORTMASK"
1119                 " [-P]"
1120                 " [-u PORTMASK]"
1121                 " [-j FRAMESIZE]"
1122                 " [-l]"
1123                 " [-w REPLAY_WINDOW_SIZE]"
1124                 " [-e]"
1125                 " [-a]"
1126                 " -f CONFIG_FILE"
1127                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1128                 " [--single-sa SAIDX]"
1129                 " [--cryptodev_mask MASK]"
1130                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1131                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1132                 "\n\n"
1133                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1134                 "  -P : Enable promiscuous mode\n"
1135                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1136                 "  -j FRAMESIZE: Enable jumbo frame with 'FRAMESIZE' as maximum\n"
1137                 "                packet size\n"
1138                 "  -l enables code-path that uses librte_ipsec\n"
1139                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1140                 "     size for each SA\n"
1141                 "  -e enables ESN\n"
1142                 "  -a enables SA SQN atomic behaviour\n"
1143                 "  -f CONFIG_FILE: Configuration file\n"
1144                 "  --config (port,queue,lcore): Rx queue configuration\n"
1145                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1146                 "                     bypassing the SP\n"
1147                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1148                 "                         devices to configure\n"
1149                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1150                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1151                 "                         (DEV_RX_OFFLOAD_*)\n"
1152                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1153                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1154                 "                         (DEV_TX_OFFLOAD_*)\n"
1155                 "\n",
1156                 prgname);
1157 }
1158
1159 static int
1160 parse_mask(const char *str, uint64_t *val)
1161 {
1162         char *end;
1163         unsigned long t;
1164
1165         errno = 0;
1166         t = strtoul(str, &end, 0);
1167         if (errno != 0 || end[0] != 0)
1168                 return -EINVAL;
1169
1170         *val = t;
1171         return 0;
1172 }
1173
1174 static int32_t
1175 parse_portmask(const char *portmask)
1176 {
1177         char *end = NULL;
1178         unsigned long pm;
1179
1180         /* parse hexadecimal string */
1181         pm = strtoul(portmask, &end, 16);
1182         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1183                 return -1;
1184
1185         if ((pm == 0) && errno)
1186                 return -1;
1187
1188         return pm;
1189 }
1190
1191 static int32_t
1192 parse_decimal(const char *str)
1193 {
1194         char *end = NULL;
1195         unsigned long num;
1196
1197         num = strtoul(str, &end, 10);
1198         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
1199                 return -1;
1200
1201         return num;
1202 }
1203
1204 static int32_t
1205 parse_config(const char *q_arg)
1206 {
1207         char s[256];
1208         const char *p, *p0 = q_arg;
1209         char *end;
1210         enum fieldnames {
1211                 FLD_PORT = 0,
1212                 FLD_QUEUE,
1213                 FLD_LCORE,
1214                 _NUM_FLD
1215         };
1216         unsigned long int_fld[_NUM_FLD];
1217         char *str_fld[_NUM_FLD];
1218         int32_t i;
1219         uint32_t size;
1220
1221         nb_lcore_params = 0;
1222
1223         while ((p = strchr(p0, '(')) != NULL) {
1224                 ++p;
1225                 p0 = strchr(p, ')');
1226                 if (p0 == NULL)
1227                         return -1;
1228
1229                 size = p0 - p;
1230                 if (size >= sizeof(s))
1231                         return -1;
1232
1233                 snprintf(s, sizeof(s), "%.*s", size, p);
1234                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1235                                 _NUM_FLD)
1236                         return -1;
1237                 for (i = 0; i < _NUM_FLD; i++) {
1238                         errno = 0;
1239                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1240                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1241                                 return -1;
1242                 }
1243                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1244                         printf("exceeded max number of lcore params: %hu\n",
1245                                 nb_lcore_params);
1246                         return -1;
1247                 }
1248                 lcore_params_array[nb_lcore_params].port_id =
1249                         (uint8_t)int_fld[FLD_PORT];
1250                 lcore_params_array[nb_lcore_params].queue_id =
1251                         (uint8_t)int_fld[FLD_QUEUE];
1252                 lcore_params_array[nb_lcore_params].lcore_id =
1253                         (uint8_t)int_fld[FLD_LCORE];
1254                 ++nb_lcore_params;
1255         }
1256         lcore_params = lcore_params_array;
1257         return 0;
1258 }
1259
1260 static void
1261 print_app_sa_prm(const struct app_sa_prm *prm)
1262 {
1263         printf("librte_ipsec usage: %s\n",
1264                 (prm->enable == 0) ? "disabled" : "enabled");
1265
1266         if (prm->enable == 0)
1267                 return;
1268
1269         printf("replay window size: %u\n", prm->window_size);
1270         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1271         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1272 }
1273
1274 static int32_t
1275 parse_args(int32_t argc, char **argv)
1276 {
1277         int32_t opt, ret;
1278         char **argvopt;
1279         int32_t option_index;
1280         char *prgname = argv[0];
1281         int32_t f_present = 0;
1282
1283         argvopt = argv;
1284
1285         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:",
1286                                 lgopts, &option_index)) != EOF) {
1287
1288                 switch (opt) {
1289                 case 'p':
1290                         enabled_port_mask = parse_portmask(optarg);
1291                         if (enabled_port_mask == 0) {
1292                                 printf("invalid portmask\n");
1293                                 print_usage(prgname);
1294                                 return -1;
1295                         }
1296                         break;
1297                 case 'P':
1298                         printf("Promiscuous mode selected\n");
1299                         promiscuous_on = 1;
1300                         break;
1301                 case 'u':
1302                         unprotected_port_mask = parse_portmask(optarg);
1303                         if (unprotected_port_mask == 0) {
1304                                 printf("invalid unprotected portmask\n");
1305                                 print_usage(prgname);
1306                                 return -1;
1307                         }
1308                         break;
1309                 case 'f':
1310                         if (f_present == 1) {
1311                                 printf("\"-f\" option present more than "
1312                                         "once!\n");
1313                                 print_usage(prgname);
1314                                 return -1;
1315                         }
1316                         if (parse_cfg_file(optarg) < 0) {
1317                                 printf("parsing file \"%s\" failed\n",
1318                                         optarg);
1319                                 print_usage(prgname);
1320                                 return -1;
1321                         }
1322                         f_present = 1;
1323                         break;
1324                 case 'j':
1325                         {
1326                                 int32_t size = parse_decimal(optarg);
1327                                 if (size <= 1518) {
1328                                         printf("Invalid jumbo frame size\n");
1329                                         if (size < 0) {
1330                                                 print_usage(prgname);
1331                                                 return -1;
1332                                         }
1333                                         printf("Using default value 9000\n");
1334                                         frame_size = 9000;
1335                                 } else {
1336                                         frame_size = size;
1337                                 }
1338                         }
1339                         printf("Enabled jumbo frames size %u\n", frame_size);
1340                         break;
1341                 case 'l':
1342                         app_sa_prm.enable = 1;
1343                         break;
1344                 case 'w':
1345                         app_sa_prm.enable = 1;
1346                         app_sa_prm.window_size = parse_decimal(optarg);
1347                         break;
1348                 case 'e':
1349                         app_sa_prm.enable = 1;
1350                         app_sa_prm.enable_esn = 1;
1351                         break;
1352                 case 'a':
1353                         app_sa_prm.enable = 1;
1354                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1355                         break;
1356                 case CMD_LINE_OPT_CONFIG_NUM:
1357                         ret = parse_config(optarg);
1358                         if (ret) {
1359                                 printf("Invalid config\n");
1360                                 print_usage(prgname);
1361                                 return -1;
1362                         }
1363                         break;
1364                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1365                         ret = parse_decimal(optarg);
1366                         if (ret == -1) {
1367                                 printf("Invalid argument[sa_idx]\n");
1368                                 print_usage(prgname);
1369                                 return -1;
1370                         }
1371
1372                         /* else */
1373                         single_sa = 1;
1374                         single_sa_idx = ret;
1375                         printf("Configured with single SA index %u\n",
1376                                         single_sa_idx);
1377                         break;
1378                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1379                         ret = parse_portmask(optarg);
1380                         if (ret == -1) {
1381                                 printf("Invalid argument[portmask]\n");
1382                                 print_usage(prgname);
1383                                 return -1;
1384                         }
1385
1386                         /* else */
1387                         enabled_cryptodev_mask = ret;
1388                         break;
1389                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1390                         ret = parse_mask(optarg, &dev_rx_offload);
1391                         if (ret != 0) {
1392                                 printf("Invalid argument for \'%s\': %s\n",
1393                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1394                                 print_usage(prgname);
1395                                 return -1;
1396                         }
1397                         break;
1398                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1399                         ret = parse_mask(optarg, &dev_tx_offload);
1400                         if (ret != 0) {
1401                                 printf("Invalid argument for \'%s\': %s\n",
1402                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1403                                 print_usage(prgname);
1404                                 return -1;
1405                         }
1406                         break;
1407                 default:
1408                         print_usage(prgname);
1409                         return -1;
1410                 }
1411         }
1412
1413         if (f_present == 0) {
1414                 printf("Mandatory option \"-f\" not present\n");
1415                 return -1;
1416         }
1417
1418         print_app_sa_prm(&app_sa_prm);
1419
1420         if (optind >= 0)
1421                 argv[optind-1] = prgname;
1422
1423         ret = optind-1;
1424         optind = 1; /* reset getopt lib */
1425         return ret;
1426 }
1427
1428 static void
1429 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1430 {
1431         char buf[ETHER_ADDR_FMT_SIZE];
1432         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1433         printf("%s%s", name, buf);
1434 }
1435
1436 /*
1437  * Update destination ethaddr for the port.
1438  */
1439 int
1440 add_dst_ethaddr(uint16_t port, const struct ether_addr *addr)
1441 {
1442         if (port >= RTE_DIM(ethaddr_tbl))
1443                 return -EINVAL;
1444
1445         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1446         return 0;
1447 }
1448
1449 /* Check the link status of all ports in up to 9s, and print them finally */
1450 static void
1451 check_all_ports_link_status(uint32_t port_mask)
1452 {
1453 #define CHECK_INTERVAL 100 /* 100ms */
1454 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1455         uint16_t portid;
1456         uint8_t count, all_ports_up, print_flag = 0;
1457         struct rte_eth_link link;
1458
1459         printf("\nChecking link status");
1460         fflush(stdout);
1461         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1462                 all_ports_up = 1;
1463                 RTE_ETH_FOREACH_DEV(portid) {
1464                         if ((port_mask & (1 << portid)) == 0)
1465                                 continue;
1466                         memset(&link, 0, sizeof(link));
1467                         rte_eth_link_get_nowait(portid, &link);
1468                         /* print link status if flag set */
1469                         if (print_flag == 1) {
1470                                 if (link.link_status)
1471                                         printf(
1472                                         "Port%d Link Up - speed %u Mbps -%s\n",
1473                                                 portid, link.link_speed,
1474                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1475                                         ("full-duplex") : ("half-duplex\n"));
1476                                 else
1477                                         printf("Port %d Link Down\n", portid);
1478                                 continue;
1479                         }
1480                         /* clear all_ports_up flag if any link down */
1481                         if (link.link_status == ETH_LINK_DOWN) {
1482                                 all_ports_up = 0;
1483                                 break;
1484                         }
1485                 }
1486                 /* after finally printing all link status, get out */
1487                 if (print_flag == 1)
1488                         break;
1489
1490                 if (all_ports_up == 0) {
1491                         printf(".");
1492                         fflush(stdout);
1493                         rte_delay_ms(CHECK_INTERVAL);
1494                 }
1495
1496                 /* set the print_flag if all ports up or timeout */
1497                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1498                         print_flag = 1;
1499                         printf("done\n");
1500                 }
1501         }
1502 }
1503
1504 static int32_t
1505 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1506                 uint16_t qp, struct lcore_params *params,
1507                 struct ipsec_ctx *ipsec_ctx,
1508                 const struct rte_cryptodev_capabilities *cipher,
1509                 const struct rte_cryptodev_capabilities *auth,
1510                 const struct rte_cryptodev_capabilities *aead)
1511 {
1512         int32_t ret = 0;
1513         unsigned long i;
1514         struct cdev_key key = { 0 };
1515
1516         key.lcore_id = params->lcore_id;
1517         if (cipher)
1518                 key.cipher_algo = cipher->sym.cipher.algo;
1519         if (auth)
1520                 key.auth_algo = auth->sym.auth.algo;
1521         if (aead)
1522                 key.aead_algo = aead->sym.aead.algo;
1523
1524         ret = rte_hash_lookup(map, &key);
1525         if (ret != -ENOENT)
1526                 return 0;
1527
1528         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1529                 if (ipsec_ctx->tbl[i].id == cdev_id)
1530                         break;
1531
1532         if (i == ipsec_ctx->nb_qps) {
1533                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1534                         printf("Maximum number of crypto devices assigned to "
1535                                 "a core, increase MAX_QP_PER_LCORE value\n");
1536                         return 0;
1537                 }
1538                 ipsec_ctx->tbl[i].id = cdev_id;
1539                 ipsec_ctx->tbl[i].qp = qp;
1540                 ipsec_ctx->nb_qps++;
1541                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1542                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1543                                 cdev_id, qp, i);
1544         }
1545
1546         ret = rte_hash_add_key_data(map, &key, (void *)i);
1547         if (ret < 0) {
1548                 printf("Faled to insert cdev mapping for (lcore %u, "
1549                                 "cdev %u, qp %u), errno %d\n",
1550                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1551                                 ipsec_ctx->tbl[i].qp, ret);
1552                 return 0;
1553         }
1554
1555         return 1;
1556 }
1557
1558 static int32_t
1559 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1560                 uint16_t qp, struct lcore_params *params)
1561 {
1562         int32_t ret = 0;
1563         const struct rte_cryptodev_capabilities *i, *j;
1564         struct rte_hash *map;
1565         struct lcore_conf *qconf;
1566         struct ipsec_ctx *ipsec_ctx;
1567         const char *str;
1568
1569         qconf = &lcore_conf[params->lcore_id];
1570
1571         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1572                 map = cdev_map_out;
1573                 ipsec_ctx = &qconf->outbound;
1574                 str = "Outbound";
1575         } else {
1576                 map = cdev_map_in;
1577                 ipsec_ctx = &qconf->inbound;
1578                 str = "Inbound";
1579         }
1580
1581         /* Required cryptodevs with operation chainning */
1582         if (!(dev_info->feature_flags &
1583                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1584                 return ret;
1585
1586         for (i = dev_info->capabilities;
1587                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1588                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1589                         continue;
1590
1591                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1592                         ret |= add_mapping(map, str, cdev_id, qp, params,
1593                                         ipsec_ctx, NULL, NULL, i);
1594                         continue;
1595                 }
1596
1597                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1598                         continue;
1599
1600                 for (j = dev_info->capabilities;
1601                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1602                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1603                                 continue;
1604
1605                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1606                                 continue;
1607
1608                         ret |= add_mapping(map, str, cdev_id, qp, params,
1609                                                 ipsec_ctx, i, j, NULL);
1610                 }
1611         }
1612
1613         return ret;
1614 }
1615
1616 /* Check if the device is enabled by cryptodev_mask */
1617 static int
1618 check_cryptodev_mask(uint8_t cdev_id)
1619 {
1620         if (enabled_cryptodev_mask & (1 << cdev_id))
1621                 return 0;
1622
1623         return -1;
1624 }
1625
1626 static int32_t
1627 cryptodevs_init(void)
1628 {
1629         struct rte_cryptodev_config dev_conf;
1630         struct rte_cryptodev_qp_conf qp_conf;
1631         uint16_t idx, max_nb_qps, qp, i;
1632         int16_t cdev_id, port_id;
1633         struct rte_hash_parameters params = { 0 };
1634
1635         params.entries = CDEV_MAP_ENTRIES;
1636         params.key_len = sizeof(struct cdev_key);
1637         params.hash_func = rte_jhash;
1638         params.hash_func_init_val = 0;
1639         params.socket_id = rte_socket_id();
1640
1641         params.name = "cdev_map_in";
1642         cdev_map_in = rte_hash_create(&params);
1643         if (cdev_map_in == NULL)
1644                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1645                                 rte_errno);
1646
1647         params.name = "cdev_map_out";
1648         cdev_map_out = rte_hash_create(&params);
1649         if (cdev_map_out == NULL)
1650                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1651                                 rte_errno);
1652
1653         printf("lcore/cryptodev/qp mappings:\n");
1654
1655         uint32_t max_sess_sz = 0, sess_sz;
1656         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1657                 void *sec_ctx;
1658
1659                 /* Get crypto priv session size */
1660                 sess_sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
1661                 if (sess_sz > max_sess_sz)
1662                         max_sess_sz = sess_sz;
1663
1664                 /*
1665                  * If crypto device is security capable, need to check the
1666                  * size of security session as well.
1667                  */
1668
1669                 /* Get security context of the crypto device */
1670                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
1671                 if (sec_ctx == NULL)
1672                         continue;
1673
1674                 /* Get size of security session */
1675                 sess_sz = rte_security_session_get_size(sec_ctx);
1676                 if (sess_sz > max_sess_sz)
1677                         max_sess_sz = sess_sz;
1678         }
1679         RTE_ETH_FOREACH_DEV(port_id) {
1680                 void *sec_ctx;
1681
1682                 if ((enabled_port_mask & (1 << port_id)) == 0)
1683                         continue;
1684
1685                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
1686                 if (sec_ctx == NULL)
1687                         continue;
1688
1689                 sess_sz = rte_security_session_get_size(sec_ctx);
1690                 if (sess_sz > max_sess_sz)
1691                         max_sess_sz = sess_sz;
1692         }
1693
1694         idx = 0;
1695         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1696                 struct rte_cryptodev_info cdev_info;
1697
1698                 if (check_cryptodev_mask((uint8_t)cdev_id))
1699                         continue;
1700
1701                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1702
1703                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1704                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1705                 else
1706                         max_nb_qps = nb_lcore_params;
1707
1708                 qp = 0;
1709                 i = 0;
1710                 while (qp < max_nb_qps && i < nb_lcore_params) {
1711                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1712                                                 &lcore_params[idx]))
1713                                 qp++;
1714                         idx++;
1715                         idx = idx % nb_lcore_params;
1716                         i++;
1717                 }
1718
1719                 if (qp == 0)
1720                         continue;
1721
1722                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1723                 dev_conf.nb_queue_pairs = qp;
1724
1725                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1726                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1727                         rte_exit(EXIT_FAILURE,
1728                                 "Device does not support at least %u "
1729                                 "sessions", CDEV_MP_NB_OBJS);
1730
1731                 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1732                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1733                         struct rte_mempool *sess_mp;
1734
1735                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1736                                         "sess_mp_%u", dev_conf.socket_id);
1737                         sess_mp = rte_cryptodev_sym_session_pool_create(
1738                                         mp_name, CDEV_MP_NB_OBJS,
1739                                         0, CDEV_MP_CACHE_SZ, 0,
1740                                         dev_conf.socket_id);
1741                         socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1742                 }
1743
1744                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool) {
1745                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1746                         struct rte_mempool *sess_mp;
1747
1748                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1749                                         "sess_mp_priv_%u", dev_conf.socket_id);
1750                         sess_mp = rte_mempool_create(mp_name,
1751                                         CDEV_MP_NB_OBJS,
1752                                         max_sess_sz,
1753                                         CDEV_MP_CACHE_SZ,
1754                                         0, NULL, NULL, NULL,
1755                                         NULL, dev_conf.socket_id,
1756                                         0);
1757                         socket_ctx[dev_conf.socket_id].session_priv_pool =
1758                                         sess_mp;
1759                 }
1760
1761                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool ||
1762                                 !socket_ctx[dev_conf.socket_id].session_pool)
1763                         rte_exit(EXIT_FAILURE,
1764                                 "Cannot create session pool on socket %d\n",
1765                                 dev_conf.socket_id);
1766                 else
1767                         printf("Allocated session pool on socket %d\n",
1768                                         dev_conf.socket_id);
1769
1770                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1771                         rte_panic("Failed to initialize cryptodev %u\n",
1772                                         cdev_id);
1773
1774                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1775                 qp_conf.mp_session =
1776                         socket_ctx[dev_conf.socket_id].session_pool;
1777                 qp_conf.mp_session_private =
1778                         socket_ctx[dev_conf.socket_id].session_priv_pool;
1779                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1780                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1781                                         &qp_conf, dev_conf.socket_id))
1782                                 rte_panic("Failed to setup queue %u for "
1783                                                 "cdev_id %u\n", 0, cdev_id);
1784
1785                 if (rte_cryptodev_start(cdev_id))
1786                         rte_panic("Failed to start cryptodev %u\n",
1787                                         cdev_id);
1788         }
1789
1790         /* create session pools for eth devices that implement security */
1791         RTE_ETH_FOREACH_DEV(port_id) {
1792                 if ((enabled_port_mask & (1 << port_id)) &&
1793                                 rte_eth_dev_get_sec_ctx(port_id)) {
1794                         int socket_id = rte_eth_dev_socket_id(port_id);
1795
1796                         if (!socket_ctx[socket_id].session_pool) {
1797                                 char mp_name[RTE_MEMPOOL_NAMESIZE];
1798                                 struct rte_mempool *sess_mp;
1799
1800                                 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1801                                                 "sess_mp_%u", socket_id);
1802                                 sess_mp = rte_mempool_create(mp_name,
1803                                                 (CDEV_MP_NB_OBJS * 2),
1804                                                 max_sess_sz,
1805                                                 CDEV_MP_CACHE_SZ,
1806                                                 0, NULL, NULL, NULL,
1807                                                 NULL, socket_id,
1808                                                 0);
1809                                 if (sess_mp == NULL)
1810                                         rte_exit(EXIT_FAILURE,
1811                                                 "Cannot create session pool "
1812                                                 "on socket %d\n", socket_id);
1813                                 else
1814                                         printf("Allocated session pool "
1815                                                 "on socket %d\n", socket_id);
1816                                 socket_ctx[socket_id].session_pool = 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 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 }