examples/ipsec-secgw/ipsec-secgw.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <netinet/in.h>
  11 #include <netinet/ip.h>
  12 #include <netinet/ip6.h>
  13 #include <string.h>
  14 #include <sys/queue.h>
  15 #include <stdarg.h>
  16 #include <errno.h>
  17 #include <getopt.h>
  18
  19 #include <rte_common.h>
  20 #include <rte_byteorder.h>
  21 #include <rte_log.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_mempool.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_acl.h>
  38 #include <rte_lpm.h>
  39 #include <rte_lpm6.h>
  40 #include <rte_hash.h>
  41 #include <rte_jhash.h>
  42 #include <rte_cryptodev.h>
  43 #include <rte_security.h>
  44 #include <rte_ip.h>
  45
  46 #include "ipsec.h"
  47 #include "parser.h"
  48
  49 #define RTE_LOGTYPE_IPSEC RTE_LOGTYPE_USER1
  50
  51 #define MAX_JUMBO_PKT_LEN  9600
  52
  53 #define MEMPOOL_CACHE_SIZE 256
  54
  55 #define NB_MBUF (32000)
  56
  57 #define CDEV_QUEUE_DESC 2048
  58 #define CDEV_MAP_ENTRIES 16384
  59 #define CDEV_MP_NB_OBJS 1024
  60 #define CDEV_MP_CACHE_SZ 64
  61 #define MAX_QUEUE_PAIRS 1
  62
  63 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  64
  65 #define NB_SOCKETS 4
  66
  67 /* Configure how many packets ahead to prefetch, when reading packets */
  68 #define PREFETCH_OFFSET 3
  69
  70 #define MAX_RX_QUEUE_PER_LCORE 16
  71
  72 #define MAX_LCORE_PARAMS 1024
  73
  74 #define UNPROTECTED_PORT(port) (unprotected_port_mask & (1 << portid))
  75
  76 /*
  77  * Configurable number of RX/TX ring descriptors
  78  */
  79 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  80 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  81 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  82 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  83
  84 #if RTE_BYTE_ORDER != RTE_LITTLE_ENDIAN
  85 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  86         (((uint64_t)((a) & 0xff) << 56) | \
  87         ((uint64_t)((b) & 0xff) << 48) | \
  88         ((uint64_t)((c) & 0xff) << 40) | \
  89         ((uint64_t)((d) & 0xff) << 32) | \
  90         ((uint64_t)((e) & 0xff) << 24) | \
  91         ((uint64_t)((f) & 0xff) << 16) | \
  92         ((uint64_t)((g) & 0xff) << 8)  | \
  93         ((uint64_t)(h) & 0xff))
  94 #else
  95 #define __BYTES_TO_UINT64(a, b, c, d, e, f, g, h) \
  96         (((uint64_t)((h) & 0xff) << 56) | \
  97         ((uint64_t)((g) & 0xff) << 48) | \
  98         ((uint64_t)((f) & 0xff) << 40) | \
  99         ((uint64_t)((e) & 0xff) << 32) | \
 100         ((uint64_t)((d) & 0xff) << 24) | \
 101         ((uint64_t)((c) & 0xff) << 16) | \
 102         ((uint64_t)((b) & 0xff) << 8) | \
 103         ((uint64_t)(a) & 0xff))
 104 #endif
 105 #define ETHADDR(a, b, c, d, e, f) (__BYTES_TO_UINT64(a, b, c, d, e, f, 0, 0))
 106
 107 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
 108                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
 109                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
 110                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
 111                 0, 0)
 112
 113 /* port/source ethernet addr and destination ethernet addr */
 114 struct ethaddr_info {
 115         uint64_t src, dst;
 116 };
 117
 118 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
 119         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
 120         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 121         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 122         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 123 };
 124
 125 #define CMD_LINE_OPT_CONFIG             "config"
 126 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 127 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 128 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
 129 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
 130
 131 enum {
 132         /* long options mapped to a short option */
 133
 134         /* first long only option value must be >= 256, so that we won't
 135          * conflict with short options
 136          */
 137         CMD_LINE_OPT_MIN_NUM = 256,
 138         CMD_LINE_OPT_CONFIG_NUM,
 139         CMD_LINE_OPT_SINGLE_SA_NUM,
 140         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 141         CMD_LINE_OPT_RX_OFFLOAD_NUM,
 142         CMD_LINE_OPT_TX_OFFLOAD_NUM,
 143 };
 144
 145 static const struct option lgopts[] = {
 146         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 147         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 148         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 149         {CMD_LINE_OPT_RX_OFFLOAD, 1, 0, CMD_LINE_OPT_RX_OFFLOAD_NUM},
 150         {CMD_LINE_OPT_TX_OFFLOAD, 1, 0, CMD_LINE_OPT_TX_OFFLOAD_NUM},
 151         {NULL, 0, 0, 0}
 152 };
 153
 154 /* mask of enabled ports */
 155 static uint32_t enabled_port_mask;
 156 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 157 static uint32_t unprotected_port_mask;
 158 static int32_t promiscuous_on = 1;
 159 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 160 static uint32_t nb_lcores;
 161 static uint32_t single_sa;
 162 static uint32_t single_sa_idx;
 163 static uint32_t frame_size;
 164
 165 /*
 166  * RX/TX HW offload capabilities to enable/use on ethernet ports.
 167  * By default all capabilities are enabled.
 168  */
 169 static uint64_t dev_rx_offload = UINT64_MAX;
 170 static uint64_t dev_tx_offload = UINT64_MAX;
 171
 172 /* application wide librte_ipsec/SA parameters */
 173 struct app_sa_prm app_sa_prm = {.enable = 0};
 174
 175 struct lcore_rx_queue {
 176         uint16_t port_id;
 177         uint8_t queue_id;
 178 } __rte_cache_aligned;
 179
 180 struct lcore_params {
 181         uint16_t port_id;
 182         uint8_t queue_id;
 183         uint8_t lcore_id;
 184 } __rte_cache_aligned;
 185
 186 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 187
 188 static struct lcore_params *lcore_params;
 189 static uint16_t nb_lcore_params;
 190
 191 static struct rte_hash *cdev_map_in;
 192 static struct rte_hash *cdev_map_out;
 193
 194 struct buffer {
 195         uint16_t len;
 196         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 197 };
 198
 199 struct lcore_conf {
 200         uint16_t nb_rx_queue;
 201         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 202         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 203         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 204         struct ipsec_ctx inbound;
 205         struct ipsec_ctx outbound;
 206         struct rt_ctx *rt4_ctx;
 207         struct rt_ctx *rt6_ctx;
 208 } __rte_cache_aligned;
 209
 210 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 211
 212 static struct rte_eth_conf port_conf = {
 213         .rxmode = {
 214                 .mq_mode        = ETH_MQ_RX_RSS,
 215                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 216                 .split_hdr_size = 0,
 217                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 218         },
 219         .rx_adv_conf = {
 220                 .rss_conf = {
 221                         .rss_key = NULL,
 222                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 223                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 224                 },
 225         },
 226         .txmode = {
 227                 .mq_mode = ETH_MQ_TX_NONE,
 228         },
 229 };
 230
 231 static struct socket_ctx socket_ctx[NB_SOCKETS];
 232
 233 static inline void
 234 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
 235         uint32_t l2_len)
 236 {
 237         uint32_t plen, trim;
 238
 239         plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
 240         if (plen < m->pkt_len) {
 241                 trim = m->pkt_len - plen;
 242                 rte_pktmbuf_trim(m, trim);
 243         }
 244 }
 245
 246 static inline void
 247 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
 248         uint32_t l2_len)
 249 {
 250         uint32_t plen, trim;
 251
 252         plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
 253         if (plen < m->pkt_len) {
 254                 trim = m->pkt_len - plen;
 255                 rte_pktmbuf_trim(m, trim);
 256         }
 257 }
 258
 259 static inline void
 260 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 261 {
 262         const struct rte_ether_hdr *eth;
 263         const struct rte_ipv4_hdr *iph4;
 264         const struct rte_ipv6_hdr *iph6;
 265
 266         eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
 267         if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 268
 269                 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
 270                         RTE_ETHER_HDR_LEN);
 271                 adjust_ipv4_pktlen(pkt, iph4, 0);
 272
 273                 if (iph4->next_proto_id == IPPROTO_ESP)
 274                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 275                 else {
 276                         t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
 277                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 278                 }
 279                 pkt->l2_len = 0;
 280                 pkt->l3_len = sizeof(*iph4);
 281         } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 282                 int next_proto;
 283                 size_t l3len, ext_len;
 284                 uint8_t *p;
 285
 286                 /* get protocol type */
 287                 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
 288                         RTE_ETHER_HDR_LEN);
 289                 adjust_ipv6_pktlen(pkt, iph6, 0);
 290
 291                 next_proto = iph6->proto;
 292
 293                 /* determine l3 header size up to ESP extension */
 294                 l3len = sizeof(struct ip6_hdr);
 295                 p = rte_pktmbuf_mtod(pkt, uint8_t *);
 296                 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
 297                         (next_proto = rte_ipv6_get_next_ext(p + l3len,
 298                                                 next_proto, &ext_len)) >= 0)
 299                         l3len += ext_len;
 300
 301                 /* drop packet when IPv6 header exceeds first segment length */
 302                 if (unlikely(l3len > pkt->data_len)) {
 303                         rte_pktmbuf_free(pkt);
 304                         return;
 305                 }
 306
 307                 if (next_proto == IPPROTO_ESP)
 308                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 309                 else {
 310                         t->ip6.data[t->ip6.num] = &iph6->proto;
 311                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 312                 }
 313                 pkt->l2_len = 0;
 314                 pkt->l3_len = l3len;
 315         } else {
 316                 /* Unknown/Unsupported type, drop the packet */
 317                 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
 318                         rte_be_to_cpu_16(eth->ether_type));
 319                 rte_pktmbuf_free(pkt);
 320         }
 321
 322         /* Check if the packet has been processed inline. For inline protocol
 323          * processed packets, the metadata in the mbuf can be used to identify
 324          * the security processing done on the packet. The metadata will be
 325          * used to retrieve the application registered userdata associated
 326          * with the security session.
 327          */
 328
 329         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 330                 struct ipsec_sa *sa;
 331                 struct ipsec_mbuf_metadata *priv;
 332                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 333                                                 rte_eth_dev_get_sec_ctx(
 334                                                 pkt->port);
 335
 336                 /* Retrieve the userdata registered. Here, the userdata
 337                  * registered is the SA pointer.
 338                  */
 339
 340                 sa = (struct ipsec_sa *)
 341                                 rte_security_get_userdata(ctx, pkt->udata64);
 342
 343                 if (sa == NULL) {
 344                         /* userdata could not be retrieved */
 345                         return;
 346                 }
 347
 348                 /* Save SA as priv member in mbuf. This will be used in the
 349                  * IPsec selector(SP-SA) check.
 350                  */
 351
 352                 priv = get_priv(pkt);
 353                 priv->sa = sa;
 354         }
 355 }
 356
 357 static inline void
 358 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 359                 uint16_t nb_pkts)
 360 {
 361         int32_t i;
 362
 363         t->ipsec.num = 0;
 364         t->ip4.num = 0;
 365         t->ip6.num = 0;
 366
 367         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 368                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 369                                         void *));
 370                 prepare_one_packet(pkts[i], t);
 371         }
 372         /* Process left packets */
 373         for (; i < nb_pkts; i++)
 374                 prepare_one_packet(pkts[i], t);
 375 }
 376
 377 static inline void
 378 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 379                 const struct lcore_conf *qconf)
 380 {
 381         struct ip *ip;
 382         struct rte_ether_hdr *ethhdr;
 383
 384         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 385
 386         ethhdr = (struct rte_ether_hdr *)
 387                 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
 388
 389         if (ip->ip_v == IPVERSION) {
 390                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 391                 pkt->l3_len = sizeof(struct ip);
 392                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 393
 394                 ip->ip_sum = 0;
 395
 396                 /* calculate IPv4 cksum in SW */
 397                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 398                         ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
 399
 400                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
 401         } else {
 402                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 403                 pkt->l3_len = sizeof(struct ip6_hdr);
 404                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 405
 406                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
 407         }
 408
 409         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 410                         sizeof(struct rte_ether_addr));
 411         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 412                         sizeof(struct rte_ether_addr));
 413 }
 414
 415 static inline void
 416 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 417                 const struct lcore_conf *qconf)
 418 {
 419         int32_t i;
 420         const int32_t prefetch_offset = 2;
 421
 422         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 423                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 424                 prepare_tx_pkt(pkts[i], port, qconf);
 425         }
 426         /* Process left packets */
 427         for (; i < nb_pkts; i++)
 428                 prepare_tx_pkt(pkts[i], port, qconf);
 429 }
 430
 431 /* Send burst of packets on an output interface */
 432 static inline int32_t
 433 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 434 {
 435         struct rte_mbuf **m_table;
 436         int32_t ret;
 437         uint16_t queueid;
 438
 439         queueid = qconf->tx_queue_id[port];
 440         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 441
 442         prepare_tx_burst(m_table, n, port, qconf);
 443
 444         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 445         if (unlikely(ret < n)) {
 446                 do {
 447                         rte_pktmbuf_free(m_table[ret]);
 448                 } while (++ret < n);
 449         }
 450
 451         return 0;
 452 }
 453
 454 /* Enqueue a single packet, and send burst if queue is filled */
 455 static inline int32_t
 456 send_single_packet(struct rte_mbuf *m, uint16_t port)
 457 {
 458         uint32_t lcore_id;
 459         uint16_t len;
 460         struct lcore_conf *qconf;
 461
 462         lcore_id = rte_lcore_id();
 463
 464         qconf = &lcore_conf[lcore_id];
 465         len = qconf->tx_mbufs[port].len;
 466         qconf->tx_mbufs[port].m_table[len] = m;
 467         len++;
 468
 469         /* enough pkts to be sent */
 470         if (unlikely(len == MAX_PKT_BURST)) {
 471                 send_burst(qconf, MAX_PKT_BURST, port);
 472                 len = 0;
 473         }
 474
 475         qconf->tx_mbufs[port].len = len;
 476         return 0;
 477 }
 478
 479 static inline void
 480 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 481                 uint16_t lim)
 482 {
 483         struct rte_mbuf *m;
 484         uint32_t i, j, res, sa_idx;
 485
 486         if (ip->num == 0 || sp == NULL)
 487                 return;
 488
 489         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 490                         ip->num, DEFAULT_MAX_CATEGORIES);
 491
 492         j = 0;
 493         for (i = 0; i < ip->num; i++) {
 494                 m = ip->pkts[i];
 495                 res = ip->res[i];
 496                 if (res == BYPASS) {
 497                         ip->pkts[j++] = m;
 498                         continue;
 499                 }
 500                 if (res == DISCARD) {
 501                         rte_pktmbuf_free(m);
 502                         continue;
 503                 }
 504
 505                 /* Only check SPI match for processed IPSec packets */
 506                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 507                         rte_pktmbuf_free(m);
 508                         continue;
 509                 }
 510
 511                 sa_idx = SPI2IDX(res);
 512                 if (!inbound_sa_check(sa, m, sa_idx)) {
 513                         rte_pktmbuf_free(m);
 514                         continue;
 515                 }
 516                 ip->pkts[j++] = m;
 517         }
 518         ip->num = j;
 519 }
 520
 521 static void
 522 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 523 {
 524         uint32_t i, n4, n6;
 525         struct ip *ip;
 526         struct rte_mbuf *m;
 527
 528         n4 = trf->ip4.num;
 529         n6 = trf->ip6.num;
 530
 531         for (i = 0; i < num; i++) {
 532
 533                 m = mb[i];
 534                 ip = rte_pktmbuf_mtod(m, struct ip *);
 535
 536                 if (ip->ip_v == IPVERSION) {
 537                         trf->ip4.pkts[n4] = m;
 538                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 539                                         uint8_t *, offsetof(struct ip, ip_p));
 540                         n4++;
 541                 } else if (ip->ip_v == IP6_VERSION) {
 542                         trf->ip6.pkts[n6] = m;
 543                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 544                                         uint8_t *,
 545                                         offsetof(struct ip6_hdr, ip6_nxt));
 546                         n6++;
 547                 } else
 548                         rte_pktmbuf_free(m);
 549         }
 550
 551         trf->ip4.num = n4;
 552         trf->ip6.num = n6;
 553 }
 554
 555
 556 static inline void
 557 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 558                 struct ipsec_traffic *traffic)
 559 {
 560         uint16_t nb_pkts_in, n_ip4, n_ip6;
 561
 562         n_ip4 = traffic->ip4.num;
 563         n_ip6 = traffic->ip6.num;
 564
 565         if (app_sa_prm.enable == 0) {
 566                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 567                                 traffic->ipsec.num, MAX_PKT_BURST);
 568                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 569         } else {
 570                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 571                         traffic->ipsec.saptr, traffic->ipsec.num);
 572                 ipsec_process(ipsec_ctx, traffic);
 573         }
 574
 575         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 576                         n_ip4);
 577
 578         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 579                         n_ip6);
 580 }
 581
 582 static inline void
 583 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 584                 struct traffic_type *ipsec)
 585 {
 586         struct rte_mbuf *m;
 587         uint32_t i, j, sa_idx;
 588
 589         if (ip->num == 0 || sp == NULL)
 590                 return;
 591
 592         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 593                         ip->num, DEFAULT_MAX_CATEGORIES);
 594
 595         j = 0;
 596         for (i = 0; i < ip->num; i++) {
 597                 m = ip->pkts[i];
 598                 sa_idx = SPI2IDX(ip->res[i]);
 599                 if (ip->res[i] == DISCARD)
 600                         rte_pktmbuf_free(m);
 601                 else if (ip->res[i] == BYPASS)
 602                         ip->pkts[j++] = m;
 603                 else {
 604                         ipsec->res[ipsec->num] = sa_idx;
 605                         ipsec->pkts[ipsec->num++] = m;
 606                 }
 607         }
 608         ip->num = j;
 609 }
 610
 611 static inline void
 612 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 613                 struct ipsec_traffic *traffic)
 614 {
 615         struct rte_mbuf *m;
 616         uint16_t idx, nb_pkts_out, i;
 617
 618         /* Drop any IPsec traffic from protected ports */
 619         for (i = 0; i < traffic->ipsec.num; i++)
 620                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 621
 622         traffic->ipsec.num = 0;
 623
 624         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 625
 626         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 627
 628         if (app_sa_prm.enable == 0) {
 629
 630                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 631                                 traffic->ipsec.res, traffic->ipsec.num,
 632                                 MAX_PKT_BURST);
 633
 634                 for (i = 0; i < nb_pkts_out; i++) {
 635                         m = traffic->ipsec.pkts[i];
 636                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 637                         if (ip->ip_v == IPVERSION) {
 638                                 idx = traffic->ip4.num++;
 639                                 traffic->ip4.pkts[idx] = m;
 640                         } else {
 641                                 idx = traffic->ip6.num++;
 642                                 traffic->ip6.pkts[idx] = m;
 643                         }
 644                 }
 645         } else {
 646                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 647                         traffic->ipsec.saptr, traffic->ipsec.num);
 648                 ipsec_process(ipsec_ctx, traffic);
 649         }
 650 }
 651
 652 static inline void
 653 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 654                 struct ipsec_traffic *traffic)
 655 {
 656         struct rte_mbuf *m;
 657         uint32_t nb_pkts_in, i, idx;
 658
 659         /* Drop any IPv4 traffic from unprotected ports */
 660         for (i = 0; i < traffic->ip4.num; i++)
 661                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 662
 663         traffic->ip4.num = 0;
 664
 665         /* Drop any IPv6 traffic from unprotected ports */
 666         for (i = 0; i < traffic->ip6.num; i++)
 667                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 668
 669         traffic->ip6.num = 0;
 670
 671         if (app_sa_prm.enable == 0) {
 672
 673                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 674                                 traffic->ipsec.num, MAX_PKT_BURST);
 675
 676                 for (i = 0; i < nb_pkts_in; i++) {
 677                         m = traffic->ipsec.pkts[i];
 678                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 679                         if (ip->ip_v == IPVERSION) {
 680                                 idx = traffic->ip4.num++;
 681                                 traffic->ip4.pkts[idx] = m;
 682                         } else {
 683                                 idx = traffic->ip6.num++;
 684                                 traffic->ip6.pkts[idx] = m;
 685                         }
 686                 }
 687         } else {
 688                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 689                         traffic->ipsec.saptr, traffic->ipsec.num);
 690                 ipsec_process(ipsec_ctx, traffic);
 691         }
 692 }
 693
 694 static inline void
 695 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 696                 struct ipsec_traffic *traffic)
 697 {
 698         struct rte_mbuf *m;
 699         uint32_t nb_pkts_out, i, n;
 700         struct ip *ip;
 701
 702         /* Drop any IPsec traffic from protected ports */
 703         for (i = 0; i < traffic->ipsec.num; i++)
 704                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 705
 706         n = 0;
 707
 708         for (i = 0; i < traffic->ip4.num; i++) {
 709                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 710                 traffic->ipsec.res[n++] = single_sa_idx;
 711         }
 712
 713         for (i = 0; i < traffic->ip6.num; i++) {
 714                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 715                 traffic->ipsec.res[n++] = single_sa_idx;
 716         }
 717
 718         traffic->ip4.num = 0;
 719         traffic->ip6.num = 0;
 720         traffic->ipsec.num = n;
 721
 722         if (app_sa_prm.enable == 0) {
 723
 724                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 725                                 traffic->ipsec.res, traffic->ipsec.num,
 726                                 MAX_PKT_BURST);
 727
 728                 /* They all sue the same SA (ip4 or ip6 tunnel) */
 729                 m = traffic->ipsec.pkts[0];
 730                 ip = rte_pktmbuf_mtod(m, struct ip *);
 731                 if (ip->ip_v == IPVERSION) {
 732                         traffic->ip4.num = nb_pkts_out;
 733                         for (i = 0; i < nb_pkts_out; i++)
 734                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 735                 } else {
 736                         traffic->ip6.num = nb_pkts_out;
 737                         for (i = 0; i < nb_pkts_out; i++)
 738                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 739                 }
 740         } else {
 741                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 742                         traffic->ipsec.saptr, traffic->ipsec.num);
 743                 ipsec_process(ipsec_ctx, traffic);
 744         }
 745 }
 746
 747 static inline int32_t
 748 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 749 {
 750         struct ipsec_mbuf_metadata *priv;
 751         struct ipsec_sa *sa;
 752
 753         priv = get_priv(pkt);
 754
 755         sa = priv->sa;
 756         if (unlikely(sa == NULL)) {
 757                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 758                 goto fail;
 759         }
 760
 761         if (is_ipv6)
 762                 return sa->portid;
 763
 764         /* else */
 765         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 766
 767 fail:
 768         if (is_ipv6)
 769                 return -1;
 770
 771         /* else */
 772         return 0;
 773 }
 774
 775 static inline void
 776 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 777 {
 778         uint32_t hop[MAX_PKT_BURST * 2];
 779         uint32_t dst_ip[MAX_PKT_BURST * 2];
 780         int32_t pkt_hop = 0;
 781         uint16_t i, offset;
 782         uint16_t lpm_pkts = 0;
 783
 784         if (nb_pkts == 0)
 785                 return;
 786
 787         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 788          * have port ID in the SA
 789          */
 790
 791         for (i = 0; i < nb_pkts; i++) {
 792                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 793                         /* Security offload not enabled. So an LPM lookup is
 794                          * required to get the hop
 795                          */
 796                         offset = offsetof(struct ip, ip_dst);
 797                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 798                                         uint32_t *, offset);
 799                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 800                         lpm_pkts++;
 801                 }
 802         }
 803
 804         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 805
 806         lpm_pkts = 0;
 807
 808         for (i = 0; i < nb_pkts; i++) {
 809                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 810                         /* Read hop from the SA */
 811                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 812                 } else {
 813                         /* Need to use hop returned by lookup */
 814                         pkt_hop = hop[lpm_pkts++];
 815                 }
 816
 817                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 818                         rte_pktmbuf_free(pkts[i]);
 819                         continue;
 820                 }
 821                 send_single_packet(pkts[i], pkt_hop & 0xff);
 822         }
 823 }
 824
 825 static inline void
 826 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 827 {
 828         int32_t hop[MAX_PKT_BURST * 2];
 829         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 830         uint8_t *ip6_dst;
 831         int32_t pkt_hop = 0;
 832         uint16_t i, offset;
 833         uint16_t lpm_pkts = 0;
 834
 835         if (nb_pkts == 0)
 836                 return;
 837
 838         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 839          * have port ID in the SA
 840          */
 841
 842         for (i = 0; i < nb_pkts; i++) {
 843                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 844                         /* Security offload not enabled. So an LPM lookup is
 845                          * required to get the hop
 846                          */
 847                         offset = offsetof(struct ip6_hdr, ip6_dst);
 848                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 849                                         offset);
 850                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 851                         lpm_pkts++;
 852                 }
 853         }
 854
 855         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 856                         lpm_pkts);
 857
 858         lpm_pkts = 0;
 859
 860         for (i = 0; i < nb_pkts; i++) {
 861                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 862                         /* Read hop from the SA */
 863                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 864                 } else {
 865                         /* Need to use hop returned by lookup */
 866                         pkt_hop = hop[lpm_pkts++];
 867                 }
 868
 869                 if (pkt_hop == -1) {
 870                         rte_pktmbuf_free(pkts[i]);
 871                         continue;
 872                 }
 873                 send_single_packet(pkts[i], pkt_hop & 0xff);
 874         }
 875 }
 876
 877 static inline void
 878 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 879                 uint8_t nb_pkts, uint16_t portid)
 880 {
 881         struct ipsec_traffic traffic;
 882
 883         prepare_traffic(pkts, &traffic, nb_pkts);
 884
 885         if (unlikely(single_sa)) {
 886                 if (UNPROTECTED_PORT(portid))
 887                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 888                 else
 889                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 890         } else {
 891                 if (UNPROTECTED_PORT(portid))
 892                         process_pkts_inbound(&qconf->inbound, &traffic);
 893                 else
 894                         process_pkts_outbound(&qconf->outbound, &traffic);
 895         }
 896
 897         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 898         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 899 }
 900
 901 static inline void
 902 drain_tx_buffers(struct lcore_conf *qconf)
 903 {
 904         struct buffer *buf;
 905         uint32_t portid;
 906
 907         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 908                 buf = &qconf->tx_mbufs[portid];
 909                 if (buf->len == 0)
 910                         continue;
 911                 send_burst(qconf, buf->len, portid);
 912                 buf->len = 0;
 913         }
 914 }
 915
 916 static inline void
 917 drain_crypto_buffers(struct lcore_conf *qconf)
 918 {
 919         uint32_t i;
 920         struct ipsec_ctx *ctx;
 921
 922         /* drain inbound buffers*/
 923         ctx = &qconf->inbound;
 924         for (i = 0; i != ctx->nb_qps; i++) {
 925                 if (ctx->tbl[i].len != 0)
 926                         enqueue_cop_burst(ctx->tbl  + i);
 927         }
 928
 929         /* drain outbound buffers*/
 930         ctx = &qconf->outbound;
 931         for (i = 0; i != ctx->nb_qps; i++) {
 932                 if (ctx->tbl[i].len != 0)
 933                         enqueue_cop_burst(ctx->tbl  + i);
 934         }
 935 }
 936
 937 static void
 938 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
 939                 struct ipsec_ctx *ctx)
 940 {
 941         uint32_t n;
 942         struct ipsec_traffic trf;
 943
 944         if (app_sa_prm.enable == 0) {
 945
 946                 /* dequeue packets from crypto-queue */
 947                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
 948                         RTE_DIM(trf.ipsec.pkts));
 949
 950                 trf.ip4.num = 0;
 951                 trf.ip6.num = 0;
 952
 953                 /* split traffic by ipv4-ipv6 */
 954                 split46_traffic(&trf, trf.ipsec.pkts, n);
 955         } else
 956                 ipsec_cqp_process(ctx, &trf);
 957
 958         /* process ipv4 packets */
 959         if (trf.ip4.num != 0) {
 960                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
 961                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
 962         }
 963
 964         /* process ipv6 packets */
 965         if (trf.ip6.num != 0) {
 966                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
 967                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
 968         }
 969 }
 970
 971 static void
 972 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
 973                 struct ipsec_ctx *ctx)
 974 {
 975         uint32_t n;
 976         struct ipsec_traffic trf;
 977
 978         if (app_sa_prm.enable == 0) {
 979
 980                 /* dequeue packets from crypto-queue */
 981                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
 982                         RTE_DIM(trf.ipsec.pkts));
 983
 984                 trf.ip4.num = 0;
 985                 trf.ip6.num = 0;
 986
 987                 /* split traffic by ipv4-ipv6 */
 988                 split46_traffic(&trf, trf.ipsec.pkts, n);
 989         } else
 990                 ipsec_cqp_process(ctx, &trf);
 991
 992         /* process ipv4 packets */
 993         if (trf.ip4.num != 0)
 994                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
 995
 996         /* process ipv6 packets */
 997         if (trf.ip6.num != 0)
 998                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
 999 }
1000
1001 /* main processing loop */
1002 static int32_t
1003 main_loop(__attribute__((unused)) void *dummy)
1004 {
1005         struct rte_mbuf *pkts[MAX_PKT_BURST];
1006         uint32_t lcore_id;
1007         uint64_t prev_tsc, diff_tsc, cur_tsc;
1008         int32_t i, nb_rx;
1009         uint16_t portid;
1010         uint8_t queueid;
1011         struct lcore_conf *qconf;
1012         int32_t socket_id;
1013         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1014                         / US_PER_S * BURST_TX_DRAIN_US;
1015         struct lcore_rx_queue *rxql;
1016
1017         prev_tsc = 0;
1018         lcore_id = rte_lcore_id();
1019         qconf = &lcore_conf[lcore_id];
1020         rxql = qconf->rx_queue_list;
1021         socket_id = rte_lcore_to_socket_id(lcore_id);
1022
1023         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1024         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1025         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1026         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1027         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1028         qconf->inbound.cdev_map = cdev_map_in;
1029         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1030         qconf->inbound.session_priv_pool =
1031                         socket_ctx[socket_id].session_priv_pool;
1032         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1033         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1034         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1035         qconf->outbound.cdev_map = cdev_map_out;
1036         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1037         qconf->outbound.session_priv_pool =
1038                         socket_ctx[socket_id].session_priv_pool;
1039
1040         if (qconf->nb_rx_queue == 0) {
1041                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1042                         lcore_id);
1043                 return 0;
1044         }
1045
1046         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1047
1048         for (i = 0; i < qconf->nb_rx_queue; i++) {
1049                 portid = rxql[i].port_id;
1050                 queueid = rxql[i].queue_id;
1051                 RTE_LOG(INFO, IPSEC,
1052                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1053                         lcore_id, portid, queueid);
1054         }
1055
1056         while (1) {
1057                 cur_tsc = rte_rdtsc();
1058
1059                 /* TX queue buffer drain */
1060                 diff_tsc = cur_tsc - prev_tsc;
1061
1062                 if (unlikely(diff_tsc > drain_tsc)) {
1063                         drain_tx_buffers(qconf);
1064                         drain_crypto_buffers(qconf);
1065                         prev_tsc = cur_tsc;
1066                 }
1067
1068                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1069
1070                         /* Read packets from RX queues */
1071                         portid = rxql[i].port_id;
1072                         queueid = rxql[i].queue_id;
1073                         nb_rx = rte_eth_rx_burst(portid, queueid,
1074                                         pkts, MAX_PKT_BURST);
1075
1076                         if (nb_rx > 0)
1077                                 process_pkts(qconf, pkts, nb_rx, portid);
1078
1079                         /* dequeue and process completed crypto-ops */
1080                         if (UNPROTECTED_PORT(portid))
1081                                 drain_inbound_crypto_queues(qconf,
1082                                         &qconf->inbound);
1083                         else
1084                                 drain_outbound_crypto_queues(qconf,
1085                                         &qconf->outbound);
1086                 }
1087         }
1088 }
1089
1090 static int32_t
1091 check_params(void)
1092 {
1093         uint8_t lcore;
1094         uint16_t portid;
1095         uint16_t i;
1096         int32_t socket_id;
1097
1098         if (lcore_params == NULL) {
1099                 printf("Error: No port/queue/core mappings\n");
1100                 return -1;
1101         }
1102
1103         for (i = 0; i < nb_lcore_params; ++i) {
1104                 lcore = lcore_params[i].lcore_id;
1105                 if (!rte_lcore_is_enabled(lcore)) {
1106                         printf("error: lcore %hhu is not enabled in "
1107                                 "lcore mask\n", lcore);
1108                         return -1;
1109                 }
1110                 socket_id = rte_lcore_to_socket_id(lcore);
1111                 if (socket_id != 0 && numa_on == 0) {
1112                         printf("warning: lcore %hhu is on socket %d "
1113                                 "with numa off\n",
1114                                 lcore, socket_id);
1115                 }
1116                 portid = lcore_params[i].port_id;
1117                 if ((enabled_port_mask & (1 << portid)) == 0) {
1118                         printf("port %u is not enabled in port mask\n", portid);
1119                         return -1;
1120                 }
1121                 if (!rte_eth_dev_is_valid_port(portid)) {
1122                         printf("port %u is not present on the board\n", portid);
1123                         return -1;
1124                 }
1125         }
1126         return 0;
1127 }
1128
1129 static uint8_t
1130 get_port_nb_rx_queues(const uint16_t port)
1131 {
1132         int32_t queue = -1;
1133         uint16_t i;
1134
1135         for (i = 0; i < nb_lcore_params; ++i) {
1136                 if (lcore_params[i].port_id == port &&
1137                                 lcore_params[i].queue_id > queue)
1138                         queue = lcore_params[i].queue_id;
1139         }
1140         return (uint8_t)(++queue);
1141 }
1142
1143 static int32_t
1144 init_lcore_rx_queues(void)
1145 {
1146         uint16_t i, nb_rx_queue;
1147         uint8_t lcore;
1148
1149         for (i = 0; i < nb_lcore_params; ++i) {
1150                 lcore = lcore_params[i].lcore_id;
1151                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1152                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1153                         printf("error: too many queues (%u) for lcore: %u\n",
1154                                         nb_rx_queue + 1, lcore);
1155                         return -1;
1156                 }
1157                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1158                         lcore_params[i].port_id;
1159                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1160                         lcore_params[i].queue_id;
1161                 lcore_conf[lcore].nb_rx_queue++;
1162         }
1163         return 0;
1164 }
1165
1166 /* display usage */
1167 static void
1168 print_usage(const char *prgname)
1169 {
1170         fprintf(stderr, "%s [EAL options] --"
1171                 " -p PORTMASK"
1172                 " [-P]"
1173                 " [-u PORTMASK]"
1174                 " [-j FRAMESIZE]"
1175                 " [-l]"
1176                 " [-w REPLAY_WINDOW_SIZE]"
1177                 " [-e]"
1178                 " [-a]"
1179                 " -f CONFIG_FILE"
1180                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1181                 " [--single-sa SAIDX]"
1182                 " [--cryptodev_mask MASK]"
1183                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1184                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1185                 "\n\n"
1186                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1187                 "  -P : Enable promiscuous mode\n"
1188                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1189                 "  -j FRAMESIZE: Enable jumbo frame with 'FRAMESIZE' as maximum\n"
1190                 "                packet size\n"
1191                 "  -l enables code-path that uses librte_ipsec\n"
1192                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1193                 "     size for each SA\n"
1194                 "  -e enables ESN\n"
1195                 "  -a enables SA SQN atomic behaviour\n"
1196                 "  -f CONFIG_FILE: Configuration file\n"
1197                 "  --config (port,queue,lcore): Rx queue configuration\n"
1198                 "  --single-sa SAIDX: Use single SA index for outbound traffic,\n"
1199                 "                     bypassing the SP\n"
1200                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1201                 "                         devices to configure\n"
1202                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1203                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1204                 "                         (DEV_RX_OFFLOAD_*)\n"
1205                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1206                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1207                 "                         (DEV_TX_OFFLOAD_*)\n"
1208                 "\n",
1209                 prgname);
1210 }
1211
1212 static int
1213 parse_mask(const char *str, uint64_t *val)
1214 {
1215         char *end;
1216         unsigned long t;
1217
1218         errno = 0;
1219         t = strtoul(str, &end, 0);
1220         if (errno != 0 || end[0] != 0)
1221                 return -EINVAL;
1222
1223         *val = t;
1224         return 0;
1225 }
1226
1227 static int32_t
1228 parse_portmask(const char *portmask)
1229 {
1230         char *end = NULL;
1231         unsigned long pm;
1232
1233         /* parse hexadecimal string */
1234         pm = strtoul(portmask, &end, 16);
1235         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1236                 return -1;
1237
1238         if ((pm == 0) && errno)
1239                 return -1;
1240
1241         return pm;
1242 }
1243
1244 static int32_t
1245 parse_decimal(const char *str)
1246 {
1247         char *end = NULL;
1248         unsigned long num;
1249
1250         num = strtoul(str, &end, 10);
1251         if ((str[0] == '\0') || (end == NULL) || (*end != '\0'))
1252                 return -1;
1253
1254         return num;
1255 }
1256
1257 static int32_t
1258 parse_config(const char *q_arg)
1259 {
1260         char s[256];
1261         const char *p, *p0 = q_arg;
1262         char *end;
1263         enum fieldnames {
1264                 FLD_PORT = 0,
1265                 FLD_QUEUE,
1266                 FLD_LCORE,
1267                 _NUM_FLD
1268         };
1269         unsigned long int_fld[_NUM_FLD];
1270         char *str_fld[_NUM_FLD];
1271         int32_t i;
1272         uint32_t size;
1273
1274         nb_lcore_params = 0;
1275
1276         while ((p = strchr(p0, '(')) != NULL) {
1277                 ++p;
1278                 p0 = strchr(p, ')');
1279                 if (p0 == NULL)
1280                         return -1;
1281
1282                 size = p0 - p;
1283                 if (size >= sizeof(s))
1284                         return -1;
1285
1286                 snprintf(s, sizeof(s), "%.*s", size, p);
1287                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1288                                 _NUM_FLD)
1289                         return -1;
1290                 for (i = 0; i < _NUM_FLD; i++) {
1291                         errno = 0;
1292                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1293                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1294                                 return -1;
1295                 }
1296                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1297                         printf("exceeded max number of lcore params: %hu\n",
1298                                 nb_lcore_params);
1299                         return -1;
1300                 }
1301                 lcore_params_array[nb_lcore_params].port_id =
1302                         (uint8_t)int_fld[FLD_PORT];
1303                 lcore_params_array[nb_lcore_params].queue_id =
1304                         (uint8_t)int_fld[FLD_QUEUE];
1305                 lcore_params_array[nb_lcore_params].lcore_id =
1306                         (uint8_t)int_fld[FLD_LCORE];
1307                 ++nb_lcore_params;
1308         }
1309         lcore_params = lcore_params_array;
1310         return 0;
1311 }
1312
1313 static void
1314 print_app_sa_prm(const struct app_sa_prm *prm)
1315 {
1316         printf("librte_ipsec usage: %s\n",
1317                 (prm->enable == 0) ? "disabled" : "enabled");
1318
1319         if (prm->enable == 0)
1320                 return;
1321
1322         printf("replay window size: %u\n", prm->window_size);
1323         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1324         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1325 }
1326
1327 static int32_t
1328 parse_args(int32_t argc, char **argv)
1329 {
1330         int32_t opt, ret;
1331         char **argvopt;
1332         int32_t option_index;
1333         char *prgname = argv[0];
1334         int32_t f_present = 0;
1335
1336         argvopt = argv;
1337
1338         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:",
1339                                 lgopts, &option_index)) != EOF) {
1340
1341                 switch (opt) {
1342                 case 'p':
1343                         enabled_port_mask = parse_portmask(optarg);
1344                         if (enabled_port_mask == 0) {
1345                                 printf("invalid portmask\n");
1346                                 print_usage(prgname);
1347                                 return -1;
1348                         }
1349                         break;
1350                 case 'P':
1351                         printf("Promiscuous mode selected\n");
1352                         promiscuous_on = 1;
1353                         break;
1354                 case 'u':
1355                         unprotected_port_mask = parse_portmask(optarg);
1356                         if (unprotected_port_mask == 0) {
1357                                 printf("invalid unprotected portmask\n");
1358                                 print_usage(prgname);
1359                                 return -1;
1360                         }
1361                         break;
1362                 case 'f':
1363                         if (f_present == 1) {
1364                                 printf("\"-f\" option present more than "
1365                                         "once!\n");
1366                                 print_usage(prgname);
1367                                 return -1;
1368                         }
1369                         if (parse_cfg_file(optarg) < 0) {
1370                                 printf("parsing file \"%s\" failed\n",
1371                                         optarg);
1372                                 print_usage(prgname);
1373                                 return -1;
1374                         }
1375                         f_present = 1;
1376                         break;
1377                 case 'j':
1378                         {
1379                                 int32_t size = parse_decimal(optarg);
1380                                 if (size <= 1518) {
1381                                         printf("Invalid jumbo frame size\n");
1382                                         if (size < 0) {
1383                                                 print_usage(prgname);
1384                                                 return -1;
1385                                         }
1386                                         printf("Using default value 9000\n");
1387                                         frame_size = 9000;
1388                                 } else {
1389                                         frame_size = size;
1390                                 }
1391                         }
1392                         printf("Enabled jumbo frames size %u\n", frame_size);
1393                         break;
1394                 case 'l':
1395                         app_sa_prm.enable = 1;
1396                         break;
1397                 case 'w':
1398                         app_sa_prm.enable = 1;
1399                         app_sa_prm.window_size = parse_decimal(optarg);
1400                         break;
1401                 case 'e':
1402                         app_sa_prm.enable = 1;
1403                         app_sa_prm.enable_esn = 1;
1404                         break;
1405                 case 'a':
1406                         app_sa_prm.enable = 1;
1407                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1408                         break;
1409                 case CMD_LINE_OPT_CONFIG_NUM:
1410                         ret = parse_config(optarg);
1411                         if (ret) {
1412                                 printf("Invalid config\n");
1413                                 print_usage(prgname);
1414                                 return -1;
1415                         }
1416                         break;
1417                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1418                         ret = parse_decimal(optarg);
1419                         if (ret == -1) {
1420                                 printf("Invalid argument[sa_idx]\n");
1421                                 print_usage(prgname);
1422                                 return -1;
1423                         }
1424
1425                         /* else */
1426                         single_sa = 1;
1427                         single_sa_idx = ret;
1428                         printf("Configured with single SA index %u\n",
1429                                         single_sa_idx);
1430                         break;
1431                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1432                         ret = parse_portmask(optarg);
1433                         if (ret == -1) {
1434                                 printf("Invalid argument[portmask]\n");
1435                                 print_usage(prgname);
1436                                 return -1;
1437                         }
1438
1439                         /* else */
1440                         enabled_cryptodev_mask = ret;
1441                         break;
1442                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1443                         ret = parse_mask(optarg, &dev_rx_offload);
1444                         if (ret != 0) {
1445                                 printf("Invalid argument for \'%s\': %s\n",
1446                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1447                                 print_usage(prgname);
1448                                 return -1;
1449                         }
1450                         break;
1451                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1452                         ret = parse_mask(optarg, &dev_tx_offload);
1453                         if (ret != 0) {
1454                                 printf("Invalid argument for \'%s\': %s\n",
1455                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1456                                 print_usage(prgname);
1457                                 return -1;
1458                         }
1459                         break;
1460                 default:
1461                         print_usage(prgname);
1462                         return -1;
1463                 }
1464         }
1465
1466         if (f_present == 0) {
1467                 printf("Mandatory option \"-f\" not present\n");
1468                 return -1;
1469         }
1470
1471         print_app_sa_prm(&app_sa_prm);
1472
1473         if (optind >= 0)
1474                 argv[optind-1] = prgname;
1475
1476         ret = optind-1;
1477         optind = 1; /* reset getopt lib */
1478         return ret;
1479 }
1480
1481 static void
1482 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1483 {
1484         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1485         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1486         printf("%s%s", name, buf);
1487 }
1488
1489 /*
1490  * Update destination ethaddr for the port.
1491  */
1492 int
1493 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1494 {
1495         if (port >= RTE_DIM(ethaddr_tbl))
1496                 return -EINVAL;
1497
1498         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1499         return 0;
1500 }
1501
1502 /* Check the link status of all ports in up to 9s, and print them finally */
1503 static void
1504 check_all_ports_link_status(uint32_t port_mask)
1505 {
1506 #define CHECK_INTERVAL 100 /* 100ms */
1507 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1508         uint16_t portid;
1509         uint8_t count, all_ports_up, print_flag = 0;
1510         struct rte_eth_link link;
1511
1512         printf("\nChecking link status");
1513         fflush(stdout);
1514         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1515                 all_ports_up = 1;
1516                 RTE_ETH_FOREACH_DEV(portid) {
1517                         if ((port_mask & (1 << portid)) == 0)
1518                                 continue;
1519                         memset(&link, 0, sizeof(link));
1520                         rte_eth_link_get_nowait(portid, &link);
1521                         /* print link status if flag set */
1522                         if (print_flag == 1) {
1523                                 if (link.link_status)
1524                                         printf(
1525                                         "Port%d Link Up - speed %u Mbps -%s\n",
1526                                                 portid, link.link_speed,
1527                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1528                                         ("full-duplex") : ("half-duplex\n"));
1529                                 else
1530                                         printf("Port %d Link Down\n", portid);
1531                                 continue;
1532                         }
1533                         /* clear all_ports_up flag if any link down */
1534                         if (link.link_status == ETH_LINK_DOWN) {
1535                                 all_ports_up = 0;
1536                                 break;
1537                         }
1538                 }
1539                 /* after finally printing all link status, get out */
1540                 if (print_flag == 1)
1541                         break;
1542
1543                 if (all_ports_up == 0) {
1544                         printf(".");
1545                         fflush(stdout);
1546                         rte_delay_ms(CHECK_INTERVAL);
1547                 }
1548
1549                 /* set the print_flag if all ports up or timeout */
1550                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1551                         print_flag = 1;
1552                         printf("done\n");
1553                 }
1554         }
1555 }
1556
1557 static int32_t
1558 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1559                 uint16_t qp, struct lcore_params *params,
1560                 struct ipsec_ctx *ipsec_ctx,
1561                 const struct rte_cryptodev_capabilities *cipher,
1562                 const struct rte_cryptodev_capabilities *auth,
1563                 const struct rte_cryptodev_capabilities *aead)
1564 {
1565         int32_t ret = 0;
1566         unsigned long i;
1567         struct cdev_key key = { 0 };
1568
1569         key.lcore_id = params->lcore_id;
1570         if (cipher)
1571                 key.cipher_algo = cipher->sym.cipher.algo;
1572         if (auth)
1573                 key.auth_algo = auth->sym.auth.algo;
1574         if (aead)
1575                 key.aead_algo = aead->sym.aead.algo;
1576
1577         ret = rte_hash_lookup(map, &key);
1578         if (ret != -ENOENT)
1579                 return 0;
1580
1581         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1582                 if (ipsec_ctx->tbl[i].id == cdev_id)
1583                         break;
1584
1585         if (i == ipsec_ctx->nb_qps) {
1586                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1587                         printf("Maximum number of crypto devices assigned to "
1588                                 "a core, increase MAX_QP_PER_LCORE value\n");
1589                         return 0;
1590                 }
1591                 ipsec_ctx->tbl[i].id = cdev_id;
1592                 ipsec_ctx->tbl[i].qp = qp;
1593                 ipsec_ctx->nb_qps++;
1594                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1595                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1596                                 cdev_id, qp, i);
1597         }
1598
1599         ret = rte_hash_add_key_data(map, &key, (void *)i);
1600         if (ret < 0) {
1601                 printf("Faled to insert cdev mapping for (lcore %u, "
1602                                 "cdev %u, qp %u), errno %d\n",
1603                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1604                                 ipsec_ctx->tbl[i].qp, ret);
1605                 return 0;
1606         }
1607
1608         return 1;
1609 }
1610
1611 static int32_t
1612 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1613                 uint16_t qp, struct lcore_params *params)
1614 {
1615         int32_t ret = 0;
1616         const struct rte_cryptodev_capabilities *i, *j;
1617         struct rte_hash *map;
1618         struct lcore_conf *qconf;
1619         struct ipsec_ctx *ipsec_ctx;
1620         const char *str;
1621
1622         qconf = &lcore_conf[params->lcore_id];
1623
1624         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1625                 map = cdev_map_out;
1626                 ipsec_ctx = &qconf->outbound;
1627                 str = "Outbound";
1628         } else {
1629                 map = cdev_map_in;
1630                 ipsec_ctx = &qconf->inbound;
1631                 str = "Inbound";
1632         }
1633
1634         /* Required cryptodevs with operation chainning */
1635         if (!(dev_info->feature_flags &
1636                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1637                 return ret;
1638
1639         for (i = dev_info->capabilities;
1640                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1641                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1642                         continue;
1643
1644                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1645                         ret |= add_mapping(map, str, cdev_id, qp, params,
1646                                         ipsec_ctx, NULL, NULL, i);
1647                         continue;
1648                 }
1649
1650                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1651                         continue;
1652
1653                 for (j = dev_info->capabilities;
1654                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1655                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1656                                 continue;
1657
1658                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1659                                 continue;
1660
1661                         ret |= add_mapping(map, str, cdev_id, qp, params,
1662                                                 ipsec_ctx, i, j, NULL);
1663                 }
1664         }
1665
1666         return ret;
1667 }
1668
1669 /* Check if the device is enabled by cryptodev_mask */
1670 static int
1671 check_cryptodev_mask(uint8_t cdev_id)
1672 {
1673         if (enabled_cryptodev_mask & (1 << cdev_id))
1674                 return 0;
1675
1676         return -1;
1677 }
1678
1679 static int32_t
1680 cryptodevs_init(void)
1681 {
1682         struct rte_cryptodev_config dev_conf;
1683         struct rte_cryptodev_qp_conf qp_conf;
1684         uint16_t idx, max_nb_qps, qp, i;
1685         int16_t cdev_id, port_id;
1686         struct rte_hash_parameters params = { 0 };
1687
1688         params.entries = CDEV_MAP_ENTRIES;
1689         params.key_len = sizeof(struct cdev_key);
1690         params.hash_func = rte_jhash;
1691         params.hash_func_init_val = 0;
1692         params.socket_id = rte_socket_id();
1693
1694         params.name = "cdev_map_in";
1695         cdev_map_in = rte_hash_create(&params);
1696         if (cdev_map_in == NULL)
1697                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1698                                 rte_errno);
1699
1700         params.name = "cdev_map_out";
1701         cdev_map_out = rte_hash_create(&params);
1702         if (cdev_map_out == NULL)
1703                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1704                                 rte_errno);
1705
1706         printf("lcore/cryptodev/qp mappings:\n");
1707
1708         uint32_t max_sess_sz = 0, sess_sz;
1709         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1710                 void *sec_ctx;
1711
1712                 /* Get crypto priv session size */
1713                 sess_sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
1714                 if (sess_sz > max_sess_sz)
1715                         max_sess_sz = sess_sz;
1716
1717                 /*
1718                  * If crypto device is security capable, need to check the
1719                  * size of security session as well.
1720                  */
1721
1722                 /* Get security context of the crypto device */
1723                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
1724                 if (sec_ctx == NULL)
1725                         continue;
1726
1727                 /* Get size of security session */
1728                 sess_sz = rte_security_session_get_size(sec_ctx);
1729                 if (sess_sz > max_sess_sz)
1730                         max_sess_sz = sess_sz;
1731         }
1732         RTE_ETH_FOREACH_DEV(port_id) {
1733                 void *sec_ctx;
1734
1735                 if ((enabled_port_mask & (1 << port_id)) == 0)
1736                         continue;
1737
1738                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
1739                 if (sec_ctx == NULL)
1740                         continue;
1741
1742                 sess_sz = rte_security_session_get_size(sec_ctx);
1743                 if (sess_sz > max_sess_sz)
1744                         max_sess_sz = sess_sz;
1745         }
1746
1747         idx = 0;
1748         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1749                 struct rte_cryptodev_info cdev_info;
1750
1751                 if (check_cryptodev_mask((uint8_t)cdev_id))
1752                         continue;
1753
1754                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1755
1756                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
1757                         max_nb_qps = cdev_info.max_nb_queue_pairs;
1758                 else
1759                         max_nb_qps = nb_lcore_params;
1760
1761                 qp = 0;
1762                 i = 0;
1763                 while (qp < max_nb_qps && i < nb_lcore_params) {
1764                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
1765                                                 &lcore_params[idx]))
1766                                 qp++;
1767                         idx++;
1768                         idx = idx % nb_lcore_params;
1769                         i++;
1770                 }
1771
1772                 if (qp == 0)
1773                         continue;
1774
1775                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
1776                 dev_conf.nb_queue_pairs = qp;
1777                 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
1778
1779                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
1780                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
1781                         rte_exit(EXIT_FAILURE,
1782                                 "Device does not support at least %u "
1783                                 "sessions", CDEV_MP_NB_OBJS);
1784
1785                 if (!socket_ctx[dev_conf.socket_id].session_pool) {
1786                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1787                         struct rte_mempool *sess_mp;
1788
1789                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1790                                         "sess_mp_%u", dev_conf.socket_id);
1791                         sess_mp = rte_cryptodev_sym_session_pool_create(
1792                                         mp_name, CDEV_MP_NB_OBJS,
1793                                         0, CDEV_MP_CACHE_SZ, 0,
1794                                         dev_conf.socket_id);
1795                         socket_ctx[dev_conf.socket_id].session_pool = sess_mp;
1796                 }
1797
1798                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool) {
1799                         char mp_name[RTE_MEMPOOL_NAMESIZE];
1800                         struct rte_mempool *sess_mp;
1801
1802                         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1803                                         "sess_mp_priv_%u", dev_conf.socket_id);
1804                         sess_mp = rte_mempool_create(mp_name,
1805                                         CDEV_MP_NB_OBJS,
1806                                         max_sess_sz,
1807                                         CDEV_MP_CACHE_SZ,
1808                                         0, NULL, NULL, NULL,
1809                                         NULL, dev_conf.socket_id,
1810                                         0);
1811                         socket_ctx[dev_conf.socket_id].session_priv_pool =
1812                                         sess_mp;
1813                 }
1814
1815                 if (!socket_ctx[dev_conf.socket_id].session_priv_pool ||
1816                                 !socket_ctx[dev_conf.socket_id].session_pool)
1817                         rte_exit(EXIT_FAILURE,
1818                                 "Cannot create session pool on socket %d\n",
1819                                 dev_conf.socket_id);
1820                 else
1821                         printf("Allocated session pool on socket %d\n",
1822                                         dev_conf.socket_id);
1823
1824                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
1825                         rte_panic("Failed to initialize cryptodev %u\n",
1826                                         cdev_id);
1827
1828                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
1829                 qp_conf.mp_session =
1830                         socket_ctx[dev_conf.socket_id].session_pool;
1831                 qp_conf.mp_session_private =
1832                         socket_ctx[dev_conf.socket_id].session_priv_pool;
1833                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
1834                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
1835                                         &qp_conf, dev_conf.socket_id))
1836                                 rte_panic("Failed to setup queue %u for "
1837                                                 "cdev_id %u\n", 0, cdev_id);
1838
1839                 if (rte_cryptodev_start(cdev_id))
1840                         rte_panic("Failed to start cryptodev %u\n",
1841                                         cdev_id);
1842         }
1843
1844         /* create session pools for eth devices that implement security */
1845         RTE_ETH_FOREACH_DEV(port_id) {
1846                 if ((enabled_port_mask & (1 << port_id)) &&
1847                                 rte_eth_dev_get_sec_ctx(port_id)) {
1848                         int socket_id = rte_eth_dev_socket_id(port_id);
1849
1850                         if (!socket_ctx[socket_id].session_priv_pool) {
1851                                 char mp_name[RTE_MEMPOOL_NAMESIZE];
1852                                 struct rte_mempool *sess_mp;
1853
1854                                 snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
1855                                                 "sess_mp_%u", socket_id);
1856                                 sess_mp = rte_mempool_create(mp_name,
1857                                                 (CDEV_MP_NB_OBJS * 2),
1858                                                 max_sess_sz,
1859                                                 CDEV_MP_CACHE_SZ,
1860                                                 0, NULL, NULL, NULL,
1861                                                 NULL, socket_id,
1862                                                 0);
1863                                 if (sess_mp == NULL)
1864                                         rte_exit(EXIT_FAILURE,
1865                                                 "Cannot create session pool "
1866                                                 "on socket %d\n", socket_id);
1867                                 else
1868                                         printf("Allocated session pool "
1869                                                 "on socket %d\n", socket_id);
1870                                 socket_ctx[socket_id].session_priv_pool =
1871                                                 sess_mp;
1872                         }
1873                 }
1874         }
1875
1876
1877         printf("\n");
1878
1879         return 0;
1880 }
1881
1882 static void
1883 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
1884 {
1885         struct rte_eth_dev_info dev_info;
1886         struct rte_eth_txconf *txconf;
1887         uint16_t nb_tx_queue, nb_rx_queue;
1888         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
1889         int32_t ret, socket_id;
1890         struct lcore_conf *qconf;
1891         struct rte_ether_addr ethaddr;
1892         struct rte_eth_conf local_port_conf = port_conf;
1893
1894         rte_eth_dev_info_get(portid, &dev_info);
1895
1896         /* limit allowed HW offloafs, as user requested */
1897         dev_info.rx_offload_capa &= dev_rx_offload;
1898         dev_info.tx_offload_capa &= dev_tx_offload;
1899
1900         printf("Configuring device port %u:\n", portid);
1901
1902         rte_eth_macaddr_get(portid, &ethaddr);
1903         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
1904         print_ethaddr("Address: ", &ethaddr);
1905         printf("\n");
1906
1907         nb_rx_queue = get_port_nb_rx_queues(portid);
1908         nb_tx_queue = nb_lcores;
1909
1910         if (nb_rx_queue > dev_info.max_rx_queues)
1911                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1912                                 "(max rx queue is %u)\n",
1913                                 nb_rx_queue, dev_info.max_rx_queues);
1914
1915         if (nb_tx_queue > dev_info.max_tx_queues)
1916                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
1917                                 "(max tx queue is %u)\n",
1918                                 nb_tx_queue, dev_info.max_tx_queues);
1919
1920         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
1921                         nb_rx_queue, nb_tx_queue);
1922
1923         if (frame_size) {
1924                 local_port_conf.rxmode.max_rx_pkt_len = frame_size;
1925                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
1926         }
1927
1928         local_port_conf.rxmode.offloads |= req_rx_offloads;
1929         local_port_conf.txmode.offloads |= req_tx_offloads;
1930
1931         /* Check that all required capabilities are supported */
1932         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
1933                         local_port_conf.rxmode.offloads)
1934                 rte_exit(EXIT_FAILURE,
1935                         "Error: port %u required RX offloads: 0x%" PRIx64
1936                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
1937                         portid, local_port_conf.rxmode.offloads,
1938                         dev_info.rx_offload_capa);
1939
1940         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
1941                         local_port_conf.txmode.offloads)
1942                 rte_exit(EXIT_FAILURE,
1943                         "Error: port %u required TX offloads: 0x%" PRIx64
1944                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
1945                         portid, local_port_conf.txmode.offloads,
1946                         dev_info.tx_offload_capa);
1947
1948         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1949                 local_port_conf.txmode.offloads |=
1950                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1951
1952         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
1953                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
1954
1955         printf("port %u configurng rx_offloads=0x%" PRIx64
1956                 ", tx_offloads=0x%" PRIx64 "\n",
1957                 portid, local_port_conf.rxmode.offloads,
1958                 local_port_conf.txmode.offloads);
1959
1960         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1961                 dev_info.flow_type_rss_offloads;
1962         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1963                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
1964                 printf("Port %u modified RSS hash function based on hardware support,"
1965                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1966                         portid,
1967                         port_conf.rx_adv_conf.rss_conf.rss_hf,
1968                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1969         }
1970
1971         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
1972                         &local_port_conf);
1973         if (ret < 0)
1974                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
1975                                 "err=%d, port=%d\n", ret, portid);
1976
1977         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
1978         if (ret < 0)
1979                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
1980                                 "err=%d, port=%d\n", ret, portid);
1981
1982         /* init one TX queue per lcore */
1983         tx_queueid = 0;
1984         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1985                 if (rte_lcore_is_enabled(lcore_id) == 0)
1986                         continue;
1987
1988                 if (numa_on)
1989                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
1990                 else
1991                         socket_id = 0;
1992
1993                 /* init TX queue */
1994                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
1995
1996                 txconf = &dev_info.default_txconf;
1997                 txconf->offloads = local_port_conf.txmode.offloads;
1998
1999                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2000                                 socket_id, txconf);
2001                 if (ret < 0)
2002                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2003                                         "err=%d, port=%d\n", ret, portid);
2004
2005                 qconf = &lcore_conf[lcore_id];
2006                 qconf->tx_queue_id[portid] = tx_queueid;
2007
2008                 /* Pre-populate pkt offloads based on capabilities */
2009                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2010                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2011                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2012                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2013
2014                 tx_queueid++;
2015
2016                 /* init RX queues */
2017                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2018                         struct rte_eth_rxconf rxq_conf;
2019
2020                         if (portid != qconf->rx_queue_list[queue].port_id)
2021                                 continue;
2022
2023                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
2024
2025                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2026                                         socket_id);
2027
2028                         rxq_conf = dev_info.default_rxconf;
2029                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
2030                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2031                                         nb_rxd, socket_id, &rxq_conf,
2032                                         socket_ctx[socket_id].mbuf_pool);
2033                         if (ret < 0)
2034                                 rte_exit(EXIT_FAILURE,
2035                                         "rte_eth_rx_queue_setup: err=%d, "
2036                                         "port=%d\n", ret, portid);
2037                 }
2038         }
2039         printf("\n");
2040 }
2041
2042 static void
2043 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2044 {
2045         char s[64];
2046         uint32_t buff_size = frame_size ? (frame_size + RTE_PKTMBUF_HEADROOM) :
2047                         RTE_MBUF_DEFAULT_BUF_SIZE;
2048
2049
2050         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2051         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2052                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2053                         buff_size,
2054                         socket_id);
2055         if (ctx->mbuf_pool == NULL)
2056                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2057                                 socket_id);
2058         else
2059                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2060 }
2061
2062 static inline int
2063 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2064 {
2065         struct ipsec_sa *sa;
2066
2067         /* For inline protocol processing, the metadata in the event will
2068          * uniquely identify the security session which raised the event.
2069          * Application would then need the userdata it had registered with the
2070          * security session to process the event.
2071          */
2072
2073         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2074
2075         if (sa == NULL) {
2076                 /* userdata could not be retrieved */
2077                 return -1;
2078         }
2079
2080         /* Sequence number over flow. SA need to be re-established */
2081         RTE_SET_USED(sa);
2082         return 0;
2083 }
2084
2085 static int
2086 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2087                  void *param, void *ret_param)
2088 {
2089         uint64_t md;
2090         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2091         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2092                                         rte_eth_dev_get_sec_ctx(port_id);
2093
2094         RTE_SET_USED(param);
2095
2096         if (type != RTE_ETH_EVENT_IPSEC)
2097                 return -1;
2098
2099         event_desc = ret_param;
2100         if (event_desc == NULL) {
2101                 printf("Event descriptor not set\n");
2102                 return -1;
2103         }
2104
2105         md = event_desc->metadata;
2106
2107         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2108                 return inline_ipsec_event_esn_overflow(ctx, md);
2109         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2110                 printf("Invalid IPsec event reported\n");
2111                 return -1;
2112         }
2113
2114         return -1;
2115 }
2116
2117 int32_t
2118 main(int32_t argc, char **argv)
2119 {
2120         int32_t ret;
2121         uint32_t lcore_id;
2122         uint8_t socket_id;
2123         uint16_t portid;
2124         uint64_t req_rx_offloads, req_tx_offloads;
2125
2126         /* init EAL */
2127         ret = rte_eal_init(argc, argv);
2128         if (ret < 0)
2129                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2130         argc -= ret;
2131         argv += ret;
2132
2133         /* parse application arguments (after the EAL ones) */
2134         ret = parse_args(argc, argv);
2135         if (ret < 0)
2136                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2137
2138         if ((unprotected_port_mask & enabled_port_mask) !=
2139                         unprotected_port_mask)
2140                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2141                                 unprotected_port_mask);
2142
2143         if (check_params() < 0)
2144                 rte_exit(EXIT_FAILURE, "check_params failed\n");
2145
2146         ret = init_lcore_rx_queues();
2147         if (ret < 0)
2148                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2149
2150         nb_lcores = rte_lcore_count();
2151
2152         /* Replicate each context per socket */
2153         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2154                 if (rte_lcore_is_enabled(lcore_id) == 0)
2155                         continue;
2156
2157                 if (numa_on)
2158                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2159                 else
2160                         socket_id = 0;
2161
2162                 if (socket_ctx[socket_id].mbuf_pool)
2163                         continue;
2164
2165                 /* initilaze SPD */
2166                 sp4_init(&socket_ctx[socket_id], socket_id);
2167
2168                 sp6_init(&socket_ctx[socket_id], socket_id);
2169
2170                 /* initilaze SAD */
2171                 sa_init(&socket_ctx[socket_id], socket_id);
2172
2173                 rt_init(&socket_ctx[socket_id], socket_id);
2174
2175                 pool_init(&socket_ctx[socket_id], socket_id, NB_MBUF);
2176         }
2177
2178         RTE_ETH_FOREACH_DEV(portid) {
2179                 if ((enabled_port_mask & (1 << portid)) == 0)
2180                         continue;
2181
2182                 sa_check_offloads(portid, &req_rx_offloads, &req_tx_offloads);
2183                 port_init(portid, req_rx_offloads, req_tx_offloads);
2184         }
2185
2186         cryptodevs_init();
2187
2188         /* start ports */
2189         RTE_ETH_FOREACH_DEV(portid) {
2190                 if ((enabled_port_mask & (1 << portid)) == 0)
2191                         continue;
2192
2193                 /* Start device */
2194                 ret = rte_eth_dev_start(portid);
2195                 if (ret < 0)
2196                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2197                                         "err=%d, port=%d\n", ret, portid);
2198                 /*
2199                  * If enabled, put device in promiscuous mode.
2200                  * This allows IO forwarding mode to forward packets
2201                  * to itself through 2 cross-connected  ports of the
2202                  * target machine.
2203                  */
2204                 if (promiscuous_on)
2205                         rte_eth_promiscuous_enable(portid);
2206
2207                 rte_eth_dev_callback_register(portid,
2208                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2209         }
2210
2211         check_all_ports_link_status(enabled_port_mask);
2212
2213         /* launch per-lcore init on every lcore */
2214         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
2215         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2216                 if (rte_eal_wait_lcore(lcore_id) < 0)
2217                         return -1;
2218         }
2219
2220         return 0;
2221 }