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\n");
 264                 rte_pktmbuf_free(pkt);
 265         }
 266
 267         /* Check if the packet has been processed inline. For inline protocol
 268          * processed packets, the metadata in the mbuf can be used to identify
 269          * the security processing done on the packet. The metadata will be
 270          * used to retrieve the application registered userdata associated
 271          * with the security session.
 272          */
 273
 274         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 275                 struct ipsec_sa *sa;
 276                 struct ipsec_mbuf_metadata *priv;
 277                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 278                                                 rte_eth_dev_get_sec_ctx(
 279                                                 pkt->port);
 280
 281                 /* Retrieve the userdata registered. Here, the userdata
 282                  * registered is the SA pointer.
 283                  */
 284
 285                 sa = (struct ipsec_sa *)
 286                                 rte_security_get_userdata(ctx, pkt->udata64);
 287
 288                 if (sa == NULL) {
 289                         /* userdata could not be retrieved */
 290                         return;
 291                 }
 292
 293                 /* Save SA as priv member in mbuf. This will be used in the
 294                  * IPsec selector(SP-SA) check.
 295                  */
 296
 297                 priv = get_priv(pkt);
 298                 priv->sa = sa;
 299         }
 300 }
 301
 302 static inline void
 303 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 304                 uint16_t nb_pkts)
 305 {
 306         int32_t i;
 307
 308         t->ipsec.num = 0;
 309         t->ip4.num = 0;
 310         t->ip6.num = 0;
 311
 312         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 313                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 314                                         void *));
 315                 prepare_one_packet(pkts[i], t);
 316         }
 317         /* Process left packets */
 318         for (; i < nb_pkts; i++)
 319                 prepare_one_packet(pkts[i], t);
 320 }
 321
 322 static inline void
 323 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 324                 const struct lcore_conf *qconf)
 325 {
 326         struct ip *ip;
 327         struct ether_hdr *ethhdr;
 328
 329         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 330
 331         ethhdr = (struct ether_hdr *)rte_pktmbuf_prepend(pkt, ETHER_HDR_LEN);
 332
 333         if (ip->ip_v == IPVERSION) {
 334                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 335                 pkt->l3_len = sizeof(struct ip);
 336                 pkt->l2_len = ETHER_HDR_LEN;
 337
 338                 ip->ip_sum = 0;
 339
 340                 /* calculate IPv4 cksum in SW */
 341                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 342                         ip->ip_sum = rte_ipv4_cksum((struct ipv4_hdr *)ip);
 343
 344                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv4);
 345         } else {
 346                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 347                 pkt->l3_len = sizeof(struct ip6_hdr);
 348                 pkt->l2_len = ETHER_HDR_LEN;
 349
 350                 ethhdr->ether_type = rte_cpu_to_be_16(ETHER_TYPE_IPv6);
 351         }
 352
 353         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 354                         sizeof(struct ether_addr));
 355         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 356                         sizeof(struct ether_addr));
 357 }
 358
 359 static inline void
 360 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 361                 const struct lcore_conf *qconf)
 362 {
 363         int32_t i;
 364         const int32_t prefetch_offset = 2;
 365
 366         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 367                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 368                 prepare_tx_pkt(pkts[i], port, qconf);
 369         }
 370         /* Process left packets */
 371         for (; i < nb_pkts; i++)
 372                 prepare_tx_pkt(pkts[i], port, qconf);
 373 }
 374
 375 /* Send burst of packets on an output interface */
 376 static inline int32_t
 377 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 378 {
 379         struct rte_mbuf **m_table;
 380         int32_t ret;
 381         uint16_t queueid;
 382
 383         queueid = qconf->tx_queue_id[port];
 384         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 385
 386         prepare_tx_burst(m_table, n, port, qconf);
 387
 388         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 389         if (unlikely(ret < n)) {
 390                 do {
 391                         rte_pktmbuf_free(m_table[ret]);
 392                 } while (++ret < n);
 393         }
 394
 395         return 0;
 396 }
 397
 398 /* Enqueue a single packet, and send burst if queue is filled */
 399 static inline int32_t
 400 send_single_packet(struct rte_mbuf *m, uint16_t port)
 401 {
 402         uint32_t lcore_id;
 403         uint16_t len;
 404         struct lcore_conf *qconf;
 405
 406         lcore_id = rte_lcore_id();
 407
 408         qconf = &lcore_conf[lcore_id];
 409         len = qconf->tx_mbufs[port].len;
 410         qconf->tx_mbufs[port].m_table[len] = m;
 411         len++;
 412
 413         /* enough pkts to be sent */
 414         if (unlikely(len == MAX_PKT_BURST)) {
 415                 send_burst(qconf, MAX_PKT_BURST, port);
 416                 len = 0;
 417         }
 418
 419         qconf->tx_mbufs[port].len = len;
 420         return 0;
 421 }
 422
 423 static inline void
 424 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 425                 uint16_t lim)
 426 {
 427         struct rte_mbuf *m;
 428         uint32_t i, j, res, sa_idx;
 429
 430         if (ip->num == 0 || sp == NULL)
 431                 return;
 432
 433         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 434                         ip->num, DEFAULT_MAX_CATEGORIES);
 435
 436         j = 0;
 437         for (i = 0; i < ip->num; i++) {
 438                 m = ip->pkts[i];
 439                 res = ip->res[i];
 440                 if (res & BYPASS) {
 441                         ip->pkts[j++] = m;
 442                         continue;
 443                 }
 444                 if (res & DISCARD) {
 445                         rte_pktmbuf_free(m);
 446                         continue;
 447                 }
 448
 449                 /* Only check SPI match for processed IPSec packets */
 450                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 451                         rte_pktmbuf_free(m);
 452                         continue;
 453                 }
 454
 455                 sa_idx = ip->res[i] & PROTECT_MASK;
 456                 if (sa_idx >= IPSEC_SA_MAX_ENTRIES ||
 457                                 !inbound_sa_check(sa, m, sa_idx)) {
 458                         rte_pktmbuf_free(m);
 459                         continue;
 460                 }
 461                 ip->pkts[j++] = m;
 462         }
 463         ip->num = j;
 464 }
 465
 466 static void
 467 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 468 {
 469         uint32_t i, n4, n6;
 470         struct ip *ip;
 471         struct rte_mbuf *m;
 472
 473         n4 = trf->ip4.num;
 474         n6 = trf->ip6.num;
 475
 476         for (i = 0; i < num; i++) {
 477
 478                 m = mb[i];
 479                 ip = rte_pktmbuf_mtod(m, struct ip *);
 480
 481                 if (ip->ip_v == IPVERSION) {
 482                         trf->ip4.pkts[n4] = m;
 483                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 484                                         uint8_t *, offsetof(struct ip, ip_p));
 485                         n4++;
 486                 } else if (ip->ip_v == IP6_VERSION) {
 487                         trf->ip6.pkts[n6] = m;
 488                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 489                                         uint8_t *,
 490                                         offsetof(struct ip6_hdr, ip6_nxt));
 491                         n6++;
 492                 } else
 493                         rte_pktmbuf_free(m);
 494         }
 495
 496         trf->ip4.num = n4;
 497         trf->ip6.num = n6;
 498 }
 499
 500
 501 static inline void
 502 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 503                 struct ipsec_traffic *traffic)
 504 {
 505         uint16_t nb_pkts_in, n_ip4, n_ip6;
 506
 507         n_ip4 = traffic->ip4.num;
 508         n_ip6 = traffic->ip6.num;
 509
 510         if (app_sa_prm.enable == 0) {
 511                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 512                                 traffic->ipsec.num, MAX_PKT_BURST);
 513                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 514         } else {
 515                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 516                         traffic->ipsec.saptr, traffic->ipsec.num);
 517                 ipsec_process(ipsec_ctx, traffic);
 518         }
 519
 520         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 521                         n_ip4);
 522
 523         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 524                         n_ip6);
 525 }
 526
 527 static inline void
 528 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 529                 struct traffic_type *ipsec)
 530 {
 531         struct rte_mbuf *m;
 532         uint32_t i, j, sa_idx;
 533
 534         if (ip->num == 0 || sp == NULL)
 535                 return;
 536
 537         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 538                         ip->num, DEFAULT_MAX_CATEGORIES);
 539
 540         j = 0;
 541         for (i = 0; i < ip->num; i++) {
 542                 m = ip->pkts[i];
 543                 sa_idx = ip->res[i] & PROTECT_MASK;
 544                 if (ip->res[i] & DISCARD)
 545                         rte_pktmbuf_free(m);
 546                 else if (ip->res[i] & BYPASS)
 547                         ip->pkts[j++] = m;
 548                 else if (sa_idx < IPSEC_SA_MAX_ENTRIES) {
 549                         ipsec->res[ipsec->num] = sa_idx;
 550                         ipsec->pkts[ipsec->num++] = m;
 551                 } else /* invalid SA idx */
 552                         rte_pktmbuf_free(m);
 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(INFO, IPSEC, "lcore %u has nothing to do\n", lcore_id);
 988                 return 0;
 989         }
 990
 991         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
 992
 993         for (i = 0; i < qconf->nb_rx_queue; i++) {
 994                 portid = rxql[i].port_id;
 995                 queueid = rxql[i].queue_id;
 996                 RTE_LOG(INFO, IPSEC,
 997                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 998                         lcore_id, portid, queueid);
 999         }
1000
1001         while (1) {
1002                 cur_tsc = rte_rdtsc();
1003
1004                 /* TX queue buffer drain */
1005                 diff_tsc = cur_tsc - prev_tsc;
1006
1007                 if (unlikely(diff_tsc > drain_tsc)) {
1008                         drain_tx_buffers(qconf);
1009                         drain_crypto_buffers(qconf);
1010                         prev_tsc = cur_tsc;
1011                 }
1012
1013                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1014
1015                         /* Read packets from RX queues */
1016                         portid = rxql[i].port_id;
1017                         queueid = rxql[i].queue_id;
1018                         nb_rx = rte_eth_rx_burst(portid, queueid,
1019                                         pkts, MAX_PKT_BURST);
1020
1021                         if (nb_rx > 0)
1022                                 process_pkts(qconf, pkts, nb_rx, portid);
1023
1024                         /* dequeue and process completed crypto-ops */
1025                         if (UNPROTECTED_PORT(portid))
1026                                 drain_inbound_crypto_queues(qconf,
1027                                         &qconf->inbound);
1028                         else
1029                                 drain_outbound_crypto_queues(qconf,
1030                                         &qconf->outbound);
1031                 }
1032         }
1033 }
1034
1035 static int32_t
1036 check_params(void)
1037 {
1038         uint8_t lcore;
1039         uint16_t portid;
1040         uint16_t i;
1041         int32_t socket_id;
1042
1043         if (lcore_params == NULL) {
1044                 printf("Error: No port/queue/core mappings\n");
1045                 return -1;
1046         }
1047
1048         for (i = 0; i < nb_lcore_params; ++i) {
1049                 lcore = lcore_params[i].lcore_id;
1050                 if (!rte_lcore_is_enabled(lcore)) {
1051                         printf("error: lcore %hhu is not enabled in "
1052                                 "lcore mask\n", lcore);
1053                         return -1;
1054                 }
1055                 socket_id = rte_lcore_to_socket_id(lcore);
1056                 if (socket_id != 0 && numa_on == 0) {
1057                         printf("warning: lcore %hhu is on socket %d "
1058                                 "with numa off\n",
1059                                 lcore, socket_id);
1060                 }
1061                 portid = lcore_params[i].port_id;
1062                 if ((enabled_port_mask & (1 << portid)) == 0) {
1063                         printf("port %u is not enabled in port mask\n", portid);
1064                         return -1;
1065                 }
1066                 if (!rte_eth_dev_is_valid_port(portid)) {
1067                         printf("port %u is not present on the board\n", portid);
1068                         return -1;
1069                 }
1070         }
1071         return 0;
1072 }
1073
1074 static uint8_t
1075 get_port_nb_rx_queues(const uint16_t port)
1076 {
1077         int32_t queue = -1;
1078         uint16_t i;
1079
1080         for (i = 0; i < nb_lcore_params; ++i) {
1081                 if (lcore_params[i].port_id == port &&
1082                                 lcore_params[i].queue_id > queue)
1083                         queue = lcore_params[i].queue_id;
1084         }
1085         return (uint8_t)(++queue);
1086 }
1087
1088 static int32_t
1089 init_lcore_rx_queues(void)
1090 {
1091         uint16_t i, nb_rx_queue;
1092         uint8_t lcore;
1093
1094         for (i = 0; i < nb_lcore_params; ++i) {
1095                 lcore = lcore_params[i].lcore_id;
1096                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1097                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1098                         printf("error: too many queues (%u) for lcore: %u\n",
1099                                         nb_rx_queue + 1, lcore);
1100                         return -1;
1101                 }
1102                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1103                         lcore_params[i].port_id;
1104                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1105                         lcore_params[i].queue_id;
1106                 lcore_conf[lcore].nb_rx_queue++;
1107         }
1108         return 0;
1109 }
1110
1111 /* display usage */
1112 static void
1113 print_usage(const char *prgname)
1114 {
1115         fprintf(stderr, "%s [EAL options] --"
1116                 " -p PORTMASK"
1117                 " [-P]"
1118                 " [-u PORTMASK]"
1119                 " [-j FRAMESIZE]"
1120                 " [-l]"
1121                 " [-w REPLAY_WINDOW_SIZE]"
1122                 " [-e]"
1123                 " [-a]"
1124                 " -f CONFIG_FILE"
1125                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1126                 " [--single-sa SAIDX]"
1127                 " [--cryptodev_mask MASK]"
1128                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1129                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1130                 "\n\n"
1131                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1132                 "  -P : Enable promiscuous mode\n"
1133                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1134                 "  -j FRAMESIZE: Enable jumbo frame with 'FRAMESIZE' as maximum\n"
1135                 "                packet size\n"
1136                 "  -l enables code-path that uses librte_ipsec\n"
1137                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1138                 "     size for each SA\n"
1139                 "  -e enables ESN\n"
1140                 "  -a enables SA SQN atomic behaviour\n"
1141                 "  -f CONFIG_FILE: Configuration file\n"
1142                 "  --config (port,queue,lcore): Rx queue configuration\n"
1143                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1144                 "                     bypassing the SP\n"
1145                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1146                 "                         devices to configure\n"
1147                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1148                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1149                 "                         (DEV_RX_OFFLOAD_*)\n"
1150                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1151                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1152                 "                         (DEV_TX_OFFLOAD_*)\n"
1153                 "\n",
1154                 prgname);
1155 }
1156
1157 static int
1158 parse_mask(const char *str, uint64_t *val)
1159 {
1160         char *end;
1161         unsigned long t;
1162
1163         errno = 0;
1164         t = strtoul(str, &end, 0);
1165         if (errno != 0 || end[0] != 0)
1166                 return -EINVAL;
1167
1168         *val = t;
1169         return 0;
1170 }
1171
1172 static int32_t
1173 parse_portmask(const char *portmask)
1174 {
1175         char *end = NULL;
1176         unsigned long pm;
1177
1178         /* parse hexadecimal string */
1179         pm = strtoul(portmask, &end, 16);
1180         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1181                 return -1;
1182
1183         if ((pm == 0) && errno)
1184                 return -1;
1185
1186         return pm;
1187 }
1188
1189 static int32_t
1190 parse_decimal(const char *str)
1191 {
1192         char *end = NULL;
1193         unsigned long num;
1194
1195         num = strtoul(str, &end, 10);
1196         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
1197                 return -1;
1198
1199         return num;
1200 }
1201
1202 static int32_t
1203 parse_config(const char *q_arg)
1204 {
1205         char s[256];
1206         const char *p, *p0 = q_arg;
1207         char *end;
1208         enum fieldnames {
1209                 FLD_PORT = 0,
1210                 FLD_QUEUE,
1211                 FLD_LCORE,
1212                 _NUM_FLD
1213         };
1214         unsigned long int_fld[_NUM_FLD];
1215         char *str_fld[_NUM_FLD];
1216         int32_t i;
1217         uint32_t size;
1218
1219         nb_lcore_params = 0;
1220
1221         while ((p = strchr(p0, '(')) != NULL) {
1222                 ++p;
1223                 p0 = strchr(p, ')');
1224                 if (p0 == NULL)
1225                         return -1;
1226
1227                 size = p0 - p;
1228                 if (size >= sizeof(s))
1229                         return -1;
1230
1231                 snprintf(s, sizeof(s), "%.*s", size, p);
1232                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1233                                 _NUM_FLD)
1234                         return -1;
1235                 for (i = 0; i < _NUM_FLD; i++) {
1236                         errno = 0;
1237                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1238                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1239                                 return -1;
1240                 }
1241                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1242                         printf("exceeded max number of lcore params: %hu\n",
1243                                 nb_lcore_params);
1244                         return -1;
1245                 }
1246                 lcore_params_array[nb_lcore_params].port_id =
1247                         (uint8_t)int_fld[FLD_PORT];
1248                 lcore_params_array[nb_lcore_params].queue_id =
1249                         (uint8_t)int_fld[FLD_QUEUE];
1250                 lcore_params_array[nb_lcore_params].lcore_id =
1251                         (uint8_t)int_fld[FLD_LCORE];
1252                 ++nb_lcore_params;
1253         }
1254         lcore_params = lcore_params_array;
1255         return 0;
1256 }
1257
1258 static void
1259 print_app_sa_prm(const struct app_sa_prm *prm)
1260 {
1261         printf("librte_ipsec usage: %s\n",
1262                 (prm->enable == 0) ? "disabled" : "enabled");
1263
1264         if (prm->enable == 0)
1265                 return;
1266
1267         printf("replay window size: %u\n", prm->window_size);
1268         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1269         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1270 }
1271
1272 static int32_t
1273 parse_args(int32_t argc, char **argv)
1274 {
1275         int32_t opt, ret;
1276         char **argvopt;
1277         int32_t option_index;
1278         char *prgname = argv[0];
1279         int32_t f_present = 0;
1280
1281         argvopt = argv;
1282
1283         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:",
1284                                 lgopts, &option_index)) != EOF) {
1285
1286                 switch (opt) {
1287                 case 'p':
1288                         enabled_port_mask = parse_portmask(optarg);
1289                         if (enabled_port_mask == 0) {
1290                                 printf("invalid portmask\n");
1291                                 print_usage(prgname);
1292                                 return -1;
1293                         }
1294                         break;
1295                 case 'P':
1296                         printf("Promiscuous mode selected\n");
1297                         promiscuous_on = 1;
1298                         break;
1299                 case 'u':
1300                         unprotected_port_mask = parse_portmask(optarg);
1301                         if (unprotected_port_mask == 0) {
1302                                 printf("invalid unprotected portmask\n");
1303                                 print_usage(prgname);
1304                                 return -1;
1305                         }
1306                         break;
1307                 case 'f':
1308                         if (f_present == 1) {
1309                                 printf("\"-f\" option present more than "
1310                                         "once!\n");
1311                                 print_usage(prgname);
1312                                 return -1;
1313                         }
1314                         if (parse_cfg_file(optarg) < 0) {
1315                                 printf("parsing file \"%s\" failed\n",
1316                                         optarg);
1317                                 print_usage(prgname);
1318                                 return -1;
1319                         }
1320                         f_present = 1;
1321                         break;
1322                 case 'j':
1323                         {
1324                                 int32_t size = parse_decimal(optarg);
1325                                 if (size <= 1518) {
1326                                         printf("Invalid jumbo frame size\n");
1327                                         if (size < 0) {
1328                                                 print_usage(prgname);
1329                                                 return -1;
1330                                         }
1331                                         printf("Using default value 9000\n");
1332                                         frame_size = 9000;
1333                                 } else {
1334                                         frame_size = size;
1335                                 }
1336                         }
1337                         printf("Enabled jumbo frames size %u\n", frame_size);
1338                         break;
1339                 case 'l':
1340                         app_sa_prm.enable = 1;
1341                         break;
1342                 case 'w':
1343                         app_sa_prm.enable = 1;
1344                         app_sa_prm.window_size = parse_decimal(optarg);
1345                         break;
1346                 case 'e':
1347                         app_sa_prm.enable = 1;
1348                         app_sa_prm.enable_esn = 1;
1349                         break;
1350                 case 'a':
1351                         app_sa_prm.enable = 1;
1352                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1353                         break;
1354                 case CMD_LINE_OPT_CONFIG_NUM:
1355                         ret = parse_config(optarg);
1356                         if (ret) {
1357                                 printf("Invalid config\n");
1358                                 print_usage(prgname);
1359                                 return -1;
1360                         }
1361                         break;
1362                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1363                         ret = parse_decimal(optarg);
1364                         if (ret == -1) {
1365                                 printf("Invalid argument[sa_idx]\n");
1366                                 print_usage(prgname);
1367                                 return -1;
1368                         }
1369
1370                         /* else */
1371                         single_sa = 1;
1372                         single_sa_idx = ret;
1373                         printf("Configured with single SA index %u\n",
1374                                         single_sa_idx);
1375                         break;
1376                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1377                         ret = parse_portmask(optarg);
1378                         if (ret == -1) {
1379                                 printf("Invalid argument[portmask]\n");
1380                                 print_usage(prgname);
1381                                 return -1;
1382                         }
1383
1384                         /* else */
1385                         enabled_cryptodev_mask = ret;
1386                         break;
1387                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1388                         ret = parse_mask(optarg, &dev_rx_offload);
1389                         if (ret != 0) {
1390                                 printf("Invalid argument for \'%s\': %s\n",
1391                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1392                                 print_usage(prgname);
1393                                 return -1;
1394                         }
1395                         break;
1396                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1397                         ret = parse_mask(optarg, &dev_tx_offload);
1398                         if (ret != 0) {
1399                                 printf("Invalid argument for \'%s\': %s\n",
1400                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1401                                 print_usage(prgname);
1402                                 return -1;
1403                         }
1404                         break;
1405                 default:
1406                         print_usage(prgname);
1407                         return -1;
1408                 }
1409         }
1410
1411         if (f_present == 0) {
1412                 printf("Mandatory option \"-f\" not present\n");
1413                 return -1;
1414         }
1415
1416         print_app_sa_prm(&app_sa_prm);
1417
1418         if (optind >= 0)
1419                 argv[optind-1] = prgname;
1420
1421         ret = optind-1;
1422         optind = 1; /* reset getopt lib */
1423         return ret;
1424 }
1425
1426 static void
1427 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
1428 {
1429         char buf[ETHER_ADDR_FMT_SIZE];
1430         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
1431         printf("%s%s", name, buf);
1432 }
1433
1434 /*
1435  * Update destination ethaddr for the port.
1436  */
1437 int
1438 add_dst_ethaddr(uint16_t port, const struct ether_addr *addr)
1439 {
1440         if (port > RTE_DIM(ethaddr_tbl))
1441                 return -EINVAL;
1442
1443         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1444         return 0;
1445 }
1446
1447 /* Check the link status of all ports in up to 9s, and print them finally */
1448 static void
1449 check_all_ports_link_status(uint32_t port_mask)
1450 {
1451 #define CHECK_INTERVAL 100 /* 100ms */
1452 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1453         uint16_t portid;
1454         uint8_t count, all_ports_up, print_flag = 0;
1455         struct rte_eth_link link;
1456
1457         printf("\nChecking link status");
1458         fflush(stdout);
1459         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1460                 all_ports_up = 1;
1461                 RTE_ETH_FOREACH_DEV(portid) {
1462                         if ((port_mask & (1 << portid)) == 0)
1463                                 continue;
1464                         memset(&link, 0, sizeof(link));
1465                         rte_eth_link_get_nowait(portid, &link);
1466                         /* print link status if flag set */
1467                         if (print_flag == 1) {
1468                                 if (link.link_status)
1469                                         printf(
1470                                         "Port%d Link Up - speed %u Mbps -%s\n",
1471                                                 portid, link.link_speed,
1472                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1473                                         ("full-duplex") : ("half-duplex\n"));
1474                                 else
1475                                         printf("Port %d Link Down\n", portid);
1476                                 continue;
1477                         }
1478                         /* clear all_ports_up flag if any link down */
1479                         if (link.link_status == ETH_LINK_DOWN) {
1480                                 all_ports_up = 0;
1481                                 break;
1482                         }
1483                 }
1484                 /* after finally printing all link status, get out */
1485                 if (print_flag == 1)
1486                         break;
1487
1488                 if (all_ports_up == 0) {
1489                         printf(".");
1490                         fflush(stdout);
1491                         rte_delay_ms(CHECK_INTERVAL);
1492                 }
1493
1494                 /* set the print_flag if all ports up or timeout */
1495                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1496                         print_flag = 1;
1497                         printf("done\n");
1498                 }
1499         }
1500 }
1501
1502 static int32_t
1503 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1504                 uint16_t qp, struct lcore_params *params,
1505                 struct ipsec_ctx *ipsec_ctx,
1506                 const struct rte_cryptodev_capabilities *cipher,
1507                 const struct rte_cryptodev_capabilities *auth,
1508                 const struct rte_cryptodev_capabilities *aead)
1509 {
1510         int32_t ret = 0;
1511         unsigned long i;
1512         struct cdev_key key = { 0 };
1513
1514         key.lcore_id = params->lcore_id;
1515         if (cipher)
1516                 key.cipher_algo = cipher->sym.cipher.algo;
1517         if (auth)
1518                 key.auth_algo = auth->sym.auth.algo;
1519         if (aead)
1520                 key.aead_algo = aead->sym.aead.algo;
1521
1522         ret = rte_hash_lookup(map, &key);
1523         if (ret != -ENOENT)
1524                 return 0;
1525
1526         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1527                 if (ipsec_ctx->tbl[i].id == cdev_id)
1528                         break;
1529
1530         if (i == ipsec_ctx->nb_qps) {
1531                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1532                         printf("Maximum number of crypto devices assigned to "
1533                                 "a core, increase MAX_QP_PER_LCORE value\n");
1534                         return 0;
1535                 }
1536                 ipsec_ctx->tbl[i].id = cdev_id;
1537                 ipsec_ctx->tbl[i].qp = qp;
1538                 ipsec_ctx->nb_qps++;
1539                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1540                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1541                                 cdev_id, qp, i);
1542         }
1543
1544         ret = rte_hash_add_key_data(map, &key, (void *)i);
1545         if (ret < 0) {
1546                 printf("Faled to insert cdev mapping for (lcore %u, "
1547                                 "cdev %u, qp %u), errno %d\n",
1548                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1549                                 ipsec_ctx->tbl[i].qp, ret);
1550                 return 0;
1551         }
1552
1553         return 1;
1554 }
1555
1556 static int32_t
1557 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1558                 uint16_t qp, struct lcore_params *params)
1559 {
1560         int32_t ret = 0;
1561         const struct rte_cryptodev_capabilities *i, *j;
1562         struct rte_hash *map;
1563         struct lcore_conf *qconf;
1564         struct ipsec_ctx *ipsec_ctx;
1565         const char *str;
1566
1567         qconf = &lcore_conf[params->lcore_id];
1568
1569         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1570                 map = cdev_map_out;
1571                 ipsec_ctx = &qconf->outbound;
1572                 str = "Outbound";
1573         } else {
1574                 map = cdev_map_in;
1575                 ipsec_ctx = &qconf->inbound;
1576                 str = "Inbound";
1577         }
1578
1579         /* Required cryptodevs with operation chainning */
1580         if (!(dev_info->feature_flags &
1581                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1582                 return ret;
1583
1584         for (i = dev_info->capabilities;
1585                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1586                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1587                         continue;
1588
1589                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1590                         ret |= add_mapping(map, str, cdev_id, qp, params,
1591                                         ipsec_ctx, NULL, NULL, i);
1592                         continue;
1593                 }
1594
1595                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1596                         continue;
1597
1598                 for (j = dev_info->capabilities;
1599                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1600                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1601                                 continue;
1602
1603                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1604                                 continue;
1605
1606                         ret |= add_mapping(map, str, cdev_id, qp, params,
1607                                                 ipsec_ctx, i, j, NULL);
1608                 }
1609         }
1610
1611         return ret;
1612 }
1613
1614 /* Check if the device is enabled by cryptodev_mask */
1615 static int
1616 check_cryptodev_mask(uint8_t cdev_id)
1617 {
1618         if (enabled_cryptodev_mask & (1 << cdev_id))
1619                 return 0;
1620
1621         return -1;
1622 }
1623
1624 static int32_t
1625 cryptodevs_init(void)
1626 {
1627         struct rte_cryptodev_config dev_conf;
1628         struct rte_cryptodev_qp_conf qp_conf;
1629         uint16_t idx, max_nb_qps, qp, i;
1630         int16_t cdev_id, port_id;
1631         struct rte_hash_parameters params = { 0 };
1632
1633         params.entries = CDEV_MAP_ENTRIES;
1634         params.key_len = sizeof(struct cdev_key);
1635         params.hash_func = rte_jhash;
1636         params.hash_func_init_val = 0;
1637         params.socket_id = rte_socket_id();
1638
1639         params.name = "cdev_map_in";
1640         cdev_map_in = rte_hash_create(&params);
1641         if (cdev_map_in == NULL)
1642                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1643                                 rte_errno);
1644
1645         params.name = "cdev_map_out";
1646         cdev_map_out = rte_hash_create(&params);
1647         if (cdev_map_out == NULL)
1648                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1649                                 rte_errno);
1650
1651         printf("lcore/cryptodev/qp mappings:\n");
1652
1653         uint32_t max_sess_sz = 0, sess_sz;
1654         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1655                 void *sec_ctx;
1656
1657                 /* Get crypto priv session size */
1658                 sess_sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
1659                 if (sess_sz > max_sess_sz)
1660                         max_sess_sz = sess_sz;
1661
1662                 /*
1663                  * If crypto device is security capable, need to check the
1664                  * size of security session as well.
1665                  */
1666
1667                 /* Get security context of the crypto device */
1668                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
1669                 if (sec_ctx == NULL)
1670                         continue;
1671
1672                 /* Get size of security session */
1673                 sess_sz = rte_security_session_get_size(sec_ctx);
1674                 if (sess_sz > max_sess_sz)
1675                         max_sess_sz = sess_sz;
1676         }
1677         RTE_ETH_FOREACH_DEV(port_id) {
1678                 void *sec_ctx;
1679
1680                 if ((enabled_port_mask & (1 << port_id)) == 0)
1681                         continue;
1682
1683                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
1684                 if (sec_ctx == NULL)
1685                         continue;
1686
1687                 sess_sz = rte_security_session_get_size(sec_ctx);
1688                 if (sess_sz > max_sess_sz)
1689                         max_sess_sz = sess_sz;
1690         }
1691
1692         idx = 0;
1693         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1694                 struct rte_cryptodev_info cdev_info;
1695
1696                 if (check_cryptodev_mask((uint8_t)cdev_id))
1697                         continue;
1698
1699                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1700
1701                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1702                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1703                 else
1704                         max_nb_qps = nb_lcore_params;
1705
1706                 qp = 0;
1707                 i = 0;
1708                 while (qp < max_nb_qps && i < nb_lcore_params) {
1709                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1710                                                 &lcore_params[idx]))
1711                                 qp++;
1712                         idx++;
1713                         idx = idx % nb_lcore_params;
1714                         i++;
1715                 }
1716
1717                 if (qp == 0)
1718                         continue;
1719
1720                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1721                 dev_conf.nb_queue_pairs = qp;
1722
1723                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1724                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1725                         rte_exit(EXIT_FAILURE,
1726                                 "Device does not support at least %u "
1727                                 "sessions", CDEV_MP_NB_OBJS);
1728
1729                 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1730                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1731                         struct rte_mempool *sess_mp;
1732
1733                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1734                                         "sess_mp_%u", dev_conf.socket_id);
1735                         sess_mp = rte_cryptodev_sym_session_pool_create(
1736                                         mp_name, CDEV_MP_NB_OBJS,
1737                                         0, CDEV_MP_CACHE_SZ, 0,
1738                                         dev_conf.socket_id);
1739                         socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1740                 }
1741
1742                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool) {
1743                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1744                         struct rte_mempool *sess_mp;
1745
1746                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1747                                         "sess_mp_priv_%u", dev_conf.socket_id);
1748                         sess_mp = rte_mempool_create(mp_name,
1749                                         CDEV_MP_NB_OBJS,
1750                                         max_sess_sz,
1751                                         CDEV_MP_CACHE_SZ,
1752                                         0, NULL, NULL, NULL,
1753                                         NULL, dev_conf.socket_id,
1754                                         0);
1755                         socket_ctx[dev_conf.socket_id].session_priv_pool =
1756                                         sess_mp;
1757                 }
1758
1759                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool ||
1760                                 !socket_ctx[dev_conf.socket_id].session_pool)
1761                         rte_exit(EXIT_FAILURE,
1762                                 "Cannot create session pool on socket %d\n",
1763                                 dev_conf.socket_id);
1764                 else
1765                         printf("Allocated session pool on socket %d\n",
1766                                         dev_conf.socket_id);
1767
1768                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1769                         rte_panic("Failed to initialize cryptodev %u\n",
1770                                         cdev_id);
1771
1772                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1773                 qp_conf.mp_session =
1774                         socket_ctx[dev_conf.socket_id].session_pool;
1775                 qp_conf.mp_session_private =
1776                         socket_ctx[dev_conf.socket_id].session_priv_pool;
1777                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1778                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1779                                         &qp_conf, dev_conf.socket_id))
1780                                 rte_panic("Failed to setup queue %u for "
1781                                                 "cdev_id %u\n", 0, cdev_id);
1782
1783                 if (rte_cryptodev_start(cdev_id))
1784                         rte_panic("Failed to start cryptodev %u\n",
1785                                         cdev_id);
1786         }
1787
1788         /* create session pools for eth devices that implement security */
1789         RTE_ETH_FOREACH_DEV(port_id) {
1790                 if ((enabled_port_mask & (1 << port_id)) &&
1791                                 rte_eth_dev_get_sec_ctx(port_id)) {
1792                         int socket_id = rte_eth_dev_socket_id(port_id);
1793
1794                         if (!socket_ctx[socket_id].session_pool) {
1795                                 char mp_name[RTE_MEMPOOL_NAMESIZE];
1796                                 struct rte_mempool *sess_mp;
1797
1798                                 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1799                                                 "sess_mp_%u", socket_id);
1800                                 sess_mp = rte_mempool_create(mp_name,
1801                                                 (CDEV_MP_NB_OBJS * 2),
1802                                                 max_sess_sz,
1803                                                 CDEV_MP_CACHE_SZ,
1804                                                 0, NULL, NULL, NULL,
1805                                                 NULL, socket_id,
1806                                                 0);
1807                                 if (sess_mp == NULL)
1808                                         rte_exit(EXIT_FAILURE,
1809                                                 "Cannot create session pool "
1810                                                 "on socket %d\n", socket_id);
1811                                 else
1812                                         printf("Allocated session pool "
1813                                                 "on socket %d\n", socket_id);
1814                                 socket_ctx[socket_id].session_pool = sess_mp;
1815                         }
1816                 }
1817         }
1818
1819
1820         printf("\n");
1821
1822         return 0;
1823 }
1824
1825 static void
1826 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
1827 {
1828         struct rte_eth_dev_info dev_info;
1829         struct rte_eth_txconf *txconf;
1830         uint16_t nb_tx_queue, nb_rx_queue;
1831         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1832         int32_t ret, socket_id;
1833         struct lcore_conf *qconf;
1834         struct ether_addr ethaddr;
1835         struct rte_eth_conf local_port_conf = port_conf;
1836
1837         rte_eth_dev_info_get(portid, &dev_info);
1838
1839         /* limit allowed HW offloafs, as user requested */
1840         dev_info.rx_offload_capa &= dev_rx_offload;
1841         dev_info.tx_offload_capa &= dev_tx_offload;
1842
1843         printf("Configuring device port %u:\n", portid);
1844
1845         rte_eth_macaddr_get(portid, &ethaddr);
1846         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
1847         print_ethaddr("Address: ", &ethaddr);
1848         printf("\n");
1849
1850         nb_rx_queue = get_port_nb_rx_queues(portid);
1851         nb_tx_queue = nb_lcores;
1852
1853         if (nb_rx_queue > dev_info.max_rx_queues)
1854                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1855                                 "(max rx queue is %u)\n",
1856                                 nb_rx_queue, dev_info.max_rx_queues);
1857
1858         if (nb_tx_queue > dev_info.max_tx_queues)
1859                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1860                                 "(max tx queue is %u)\n",
1861                                 nb_tx_queue, dev_info.max_tx_queues);
1862
1863         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1864                         nb_rx_queue, nb_tx_queue);
1865
1866         if (frame_size) {
1867                 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
1868                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1869         }
1870
1871         local_port_conf.rxmode.offloads |= req_rx_offloads;
1872         local_port_conf.txmode.offloads |= req_tx_offloads;
1873
1874         /* Check that all required capabilities are supported */
1875         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
1876                         local_port_conf.rxmode.offloads)
1877                 rte_exit(EXIT_FAILURE,
1878                         "Error: port %u required RX offloads: 0x%" PRIx64
1879                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
1880                         portid, local_port_conf.rxmode.offloads,
1881                         dev_info.rx_offload_capa);
1882
1883         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
1884                         local_port_conf.txmode.offloads)
1885                 rte_exit(EXIT_FAILURE,
1886                         "Error: port %u required TX offloads: 0x%" PRIx64
1887                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
1888                         portid, local_port_conf.txmode.offloads,
1889                         dev_info.tx_offload_capa);
1890
1891         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1892                 local_port_conf.txmode.offloads |=
1893                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1894
1895         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
1896                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
1897
1898         printf("port %u configurng rx_offloads=0x%" PRIx64
1899                 ", tx_offloads=0x%" PRIx64 "\n",
1900                 portid, local_port_conf.rxmode.offloads,
1901                 local_port_conf.txmode.offloads);
1902
1903         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1904                 dev_info.flow_type_rss_offloads;
1905         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1906                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
1907                 printf("Port %u modified RSS hash function based on hardware support,"
1908                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1909                         portid,
1910                         port_conf.rx_adv_conf.rss_conf.rss_hf,
1911                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1912         }
1913
1914         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1915                         &local_port_conf);
1916         if (ret < 0)
1917                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1918                                 "err=%d, port=%d\n", ret, portid);
1919
1920         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
1921         if (ret < 0)
1922                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
1923                                 "err=%d, port=%d\n", ret, portid);
1924
1925         /* init one TX queue per lcore */
1926         tx_queueid = 0;
1927         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1928                 if (rte_lcore_is_enabled(lcore_id) == 0)
1929                         continue;
1930
1931                 if (numa_on)
1932                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1933                 else
1934                         socket_id = 0;
1935
1936                 /* init TX queue */
1937                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1938
1939                 txconf = &dev_info.default_txconf;
1940                 txconf->offloads = local_port_conf.txmode.offloads;
1941
1942                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
1943                                 socket_id, txconf);
1944                 if (ret < 0)
1945                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1946                                         "err=%d, port=%d\n", ret, portid);
1947
1948                 qconf = &lcore_conf[lcore_id];
1949                 qconf->tx_queue_id[portid] = tx_queueid;
1950
1951                 /* Pre-populate pkt offloads based on capabilities */
1952                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
1953                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
1954                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
1955                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
1956
1957                 tx_queueid++;
1958
1959                 /* init RX queues */
1960                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
1961                         struct rte_eth_rxconf rxq_conf;
1962
1963                         if (portid != qconf->rx_queue_list[queue].port_id)
1964                                 continue;
1965
1966                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
1967
1968                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
1969                                         socket_id);
1970
1971                         rxq_conf = dev_info.default_rxconf;
1972                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
1973                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
1974                                         nb_rxd, socket_id, &rxq_conf,
1975                                         socket_ctx[socket_id].mbuf_pool);
1976                         if (ret < 0)
1977                                 rte_exit(EXIT_FAILURE,
1978                                         "rte_eth_rx_queue_setup: err=%d, "
1979                                         "port=%d\n", ret, portid);
1980                 }
1981         }
1982         printf("\n");
1983 }
1984
1985 static void
1986 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
1987 {
1988         char s[64];
1989         uint32_t buff_size = frame_size ? (frame_size + RTE_PKTMBUF_HEADROOM) :
1990                         RTE_MBUF_DEFAULT_BUF_SIZE;
1991
1992
1993         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
1994         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
1995                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
1996                         buff_size,
1997                         socket_id);
1998         if (ctx->mbuf_pool == NULL)
1999                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2000                                 socket_id);
2001         else
2002                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2003 }
2004
2005 static inline int
2006 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2007 {
2008         struct ipsec_sa *sa;
2009
2010         /* For inline protocol processing, the metadata in the event will
2011          * uniquely identify the security session which raised the event.
2012          * Application would then need the userdata it had registered with the
2013          * security session to process the event.
2014          */
2015
2016         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2017
2018         if (sa == NULL) {
2019                 /* userdata could not be retrieved */
2020                 return -1;
2021         }
2022
2023         /* Sequence number over flow. SA need to be re-established */
2024         RTE_SET_USED(sa);
2025         return 0;
2026 }
2027
2028 static int
2029 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2030                  void *param, void *ret_param)
2031 {
2032         uint64_t md;
2033         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2034         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2035                                         rte_eth_dev_get_sec_ctx(port_id);
2036
2037         RTE_SET_USED(param);
2038
2039         if (type != RTE_ETH_EVENT_IPSEC)
2040                 return -1;
2041
2042         event_desc = ret_param;
2043         if (event_desc == NULL) {
2044                 printf("Event descriptor not set\n");
2045                 return -1;
2046         }
2047
2048         md = event_desc->metadata;
2049
2050         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2051                 return inline_ipsec_event_esn_overflow(ctx, md);
2052         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2053                 printf("Invalid IPsec event reported\n");
2054                 return -1;
2055         }
2056
2057         return -1;
2058 }
2059
2060 int32_t
2061 main(int32_t argc, char **argv)
2062 {
2063         int32_t ret;
2064         uint32_t lcore_id;
2065         uint8_t socket_id;
2066         uint16_t portid;
2067         uint64_t req_rx_offloads, req_tx_offloads;
2068
2069         /* init EAL */
2070         ret = rte_eal_init(argc, argv);
2071         if (ret < 0)
2072                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2073         argc -= ret;
2074         argv += ret;
2075
2076         /* parse application arguments (after the EAL ones) */
2077         ret = parse_args(argc, argv);
2078         if (ret < 0)
2079                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2080
2081         if ((unprotected_port_mask & enabled_port_mask) !=
2082                         unprotected_port_mask)
2083                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2084                                 unprotected_port_mask);
2085
2086         if (check_params() < 0)
2087                 rte_exit(EXIT_FAILURE, "check_params failed\n");
2088
2089         ret = init_lcore_rx_queues();
2090         if (ret < 0)
2091                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2092
2093         nb_lcores = rte_lcore_count();
2094
2095         /* Replicate each context per socket */
2096         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2097                 if (rte_lcore_is_enabled(lcore_id) == 0)
2098                         continue;
2099
2100                 if (numa_on)
2101                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2102                 else
2103                         socket_id = 0;
2104
2105                 if (socket_ctx[socket_id].mbuf_pool)
2106                         continue;
2107
2108                 /* initilaze SPD */
2109                 sp4_init(&socket_ctx[socket_id], socket_id);
2110
2111                 sp6_init(&socket_ctx[socket_id], socket_id);
2112
2113                 /* initilaze SAD */
2114                 sa_init(&socket_ctx[socket_id], socket_id);
2115
2116                 rt_init(&socket_ctx[socket_id], socket_id);
2117
2118                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2119         }
2120
2121         RTE_ETH_FOREACH_DEV(portid) {
2122                 if ((enabled_port_mask & (1 << portid)) == 0)
2123                         continue;
2124
2125                 sa_check_offloads(portid, &req_rx_offloads, &req_tx_offloads);
2126                 port_init(portid, req_rx_offloads, req_tx_offloads);
2127         }
2128
2129         cryptodevs_init();
2130
2131         /* start ports */
2132         RTE_ETH_FOREACH_DEV(portid) {
2133                 if ((enabled_port_mask & (1 << portid)) == 0)
2134                         continue;
2135
2136                 /* Start device */
2137                 ret = rte_eth_dev_start(portid);
2138                 if (ret < 0)
2139                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2140                                         "err=%d, port=%d\n", ret, portid);
2141                 /*
2142                  * If enabled, put device in promiscuous mode.
2143                  * This allows IO forwarding mode to forward packets
2144                  * to itself through 2 cross-connected  ports of the
2145                  * target machine.
2146                  */
2147                 if (promiscuous_on)
2148                         rte_eth_promiscuous_enable(portid);
2149
2150                 rte_eth_dev_callback_register(portid,
2151                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2152         }
2153
2154         check_all_ports_link_status(enabled_port_mask);
2155
2156         /* launch per-lcore init on every lcore */
2157         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2158         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2159                 if (rte_eal_wait_lcore(lcore_id) < 0)
2160                         return -1;
2161         }
2162
2163         return 0;
2164 }