examples/ipsec-secgw/ipsec-secgw.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <stdbool.h>
   6 #include <stdio.h>
   7 #include <stdlib.h>
   8 #include <stdint.h>
   9 #include <inttypes.h>
  10 #include <sys/types.h>
  11 #include <netinet/in.h>
  12 #include <netinet/ip.h>
  13 #include <netinet/ip6.h>
  14 #include <string.h>
  15 #include <sys/queue.h>
  16 #include <stdarg.h>
  17 #include <errno.h>
  18 #include <signal.h>
  19 #include <getopt.h>
  20
  21 #include <rte_common.h>
  22 #include <rte_bitmap.h>
  23 #include <rte_byteorder.h>
  24 #include <rte_log.h>
  25 #include <rte_eal.h>
  26 #include <rte_launch.h>
  27 #include <rte_atomic.h>
  28 #include <rte_cycles.h>
  29 #include <rte_prefetch.h>
  30 #include <rte_lcore.h>
  31 #include <rte_per_lcore.h>
  32 #include <rte_branch_prediction.h>
  33 #include <rte_interrupts.h>
  34 #include <rte_random.h>
  35 #include <rte_debug.h>
  36 #include <rte_ether.h>
  37 #include <rte_ethdev.h>
  38 #include <rte_mempool.h>
  39 #include <rte_mbuf.h>
  40 #include <rte_acl.h>
  41 #include <rte_lpm.h>
  42 #include <rte_lpm6.h>
  43 #include <rte_hash.h>
  44 #include <rte_jhash.h>
  45 #include <rte_cryptodev.h>
  46 #include <rte_security.h>
  47 #include <rte_eventdev.h>
  48 #include <rte_ip.h>
  49 #include <rte_ip_frag.h>
  50
  51 #include "event_helper.h"
  52 #include "ipsec.h"
  53 #include "ipsec_worker.h"
  54 #include "parser.h"
  55 #include "sad.h"
  56
  57 volatile bool force_quit;
  58
  59 #define MAX_JUMBO_PKT_LEN  9600
  60
  61 #define MEMPOOL_CACHE_SIZE 256
  62
  63 #define CDEV_QUEUE_DESC 2048
  64 #define CDEV_MAP_ENTRIES 16384
  65 #define CDEV_MP_NB_OBJS 1024
  66 #define CDEV_MP_CACHE_SZ 64
  67 #define MAX_QUEUE_PAIRS 1
  68
  69 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  70
  71 /* Configure how many packets ahead to prefetch, when reading packets */
  72 #define PREFETCH_OFFSET 3
  73
  74 #define MAX_RX_QUEUE_PER_LCORE 16
  75
  76 #define MAX_LCORE_PARAMS 1024
  77
  78 /*
  79  * Configurable number of RX/TX ring descriptors
  80  */
  81 #define IPSEC_SECGW_RX_DESC_DEFAULT 1024
  82 #define IPSEC_SECGW_TX_DESC_DEFAULT 1024
  83 static uint16_t nb_rxd = IPSEC_SECGW_RX_DESC_DEFAULT;
  84 static uint16_t nb_txd = IPSEC_SECGW_TX_DESC_DEFAULT;
  85
  86 #define ETHADDR_TO_UINT64(addr) __BYTES_TO_UINT64( \
  87                 (addr)->addr_bytes[0], (addr)->addr_bytes[1], \
  88                 (addr)->addr_bytes[2], (addr)->addr_bytes[3], \
  89                 (addr)->addr_bytes[4], (addr)->addr_bytes[5], \
  90                 0, 0)
  91
  92 #define FRAG_TBL_BUCKET_ENTRIES 4
  93 #define MAX_FRAG_TTL_NS         (10LL * NS_PER_S)
  94
  95 #define MTU_TO_FRAMELEN(x)      ((x) + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN)
  96
  97 struct ethaddr_info ethaddr_tbl[RTE_MAX_ETHPORTS] = {
  98         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x7e, 0x94, 0x9a) },
  99         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x22, 0xa1, 0xd9) },
 100         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x08, 0x69, 0x26) },
 101         { 0, ETHADDR(0x00, 0x16, 0x3e, 0x49, 0x9e, 0xdd) }
 102 };
 103
 104 struct flow_info flow_info_tbl[RTE_MAX_ETHPORTS];
 105
 106 #define CMD_LINE_OPT_CONFIG             "config"
 107 #define CMD_LINE_OPT_SINGLE_SA          "single-sa"
 108 #define CMD_LINE_OPT_CRYPTODEV_MASK     "cryptodev_mask"
 109 #define CMD_LINE_OPT_TRANSFER_MODE      "transfer-mode"
 110 #define CMD_LINE_OPT_SCHEDULE_TYPE      "event-schedule-type"
 111 #define CMD_LINE_OPT_RX_OFFLOAD         "rxoffload"
 112 #define CMD_LINE_OPT_TX_OFFLOAD         "txoffload"
 113 #define CMD_LINE_OPT_REASSEMBLE         "reassemble"
 114 #define CMD_LINE_OPT_MTU                "mtu"
 115 #define CMD_LINE_OPT_FRAG_TTL           "frag-ttl"
 116
 117 #define CMD_LINE_ARG_EVENT      "event"
 118 #define CMD_LINE_ARG_POLL       "poll"
 119 #define CMD_LINE_ARG_ORDERED    "ordered"
 120 #define CMD_LINE_ARG_ATOMIC     "atomic"
 121 #define CMD_LINE_ARG_PARALLEL   "parallel"
 122
 123 enum {
 124         /* long options mapped to a short option */
 125
 126         /* first long only option value must be >= 256, so that we won't
 127          * conflict with short options
 128          */
 129         CMD_LINE_OPT_MIN_NUM = 256,
 130         CMD_LINE_OPT_CONFIG_NUM,
 131         CMD_LINE_OPT_SINGLE_SA_NUM,
 132         CMD_LINE_OPT_CRYPTODEV_MASK_NUM,
 133         CMD_LINE_OPT_TRANSFER_MODE_NUM,
 134         CMD_LINE_OPT_SCHEDULE_TYPE_NUM,
 135         CMD_LINE_OPT_RX_OFFLOAD_NUM,
 136         CMD_LINE_OPT_TX_OFFLOAD_NUM,
 137         CMD_LINE_OPT_REASSEMBLE_NUM,
 138         CMD_LINE_OPT_MTU_NUM,
 139         CMD_LINE_OPT_FRAG_TTL_NUM,
 140 };
 141
 142 static const struct option lgopts[] = {
 143         {CMD_LINE_OPT_CONFIG, 1, 0, CMD_LINE_OPT_CONFIG_NUM},
 144         {CMD_LINE_OPT_SINGLE_SA, 1, 0, CMD_LINE_OPT_SINGLE_SA_NUM},
 145         {CMD_LINE_OPT_CRYPTODEV_MASK, 1, 0, CMD_LINE_OPT_CRYPTODEV_MASK_NUM},
 146         {CMD_LINE_OPT_TRANSFER_MODE, 1, 0, CMD_LINE_OPT_TRANSFER_MODE_NUM},
 147         {CMD_LINE_OPT_SCHEDULE_TYPE, 1, 0, CMD_LINE_OPT_SCHEDULE_TYPE_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         {CMD_LINE_OPT_REASSEMBLE, 1, 0, CMD_LINE_OPT_REASSEMBLE_NUM},
 151         {CMD_LINE_OPT_MTU, 1, 0, CMD_LINE_OPT_MTU_NUM},
 152         {CMD_LINE_OPT_FRAG_TTL, 1, 0, CMD_LINE_OPT_FRAG_TTL_NUM},
 153         {NULL, 0, 0, 0}
 154 };
 155
 156 uint32_t unprotected_port_mask;
 157 uint32_t single_sa_idx;
 158 /* mask of enabled ports */
 159 static uint32_t enabled_port_mask;
 160 static uint64_t enabled_cryptodev_mask = UINT64_MAX;
 161 static int32_t promiscuous_on = 1;
 162 static int32_t numa_on = 1; /**< NUMA is enabled by default. */
 163 static uint32_t nb_lcores;
 164 static uint32_t single_sa;
 165 static uint32_t nb_bufs_in_pool;
 166
 167 /*
 168  * RX/TX HW offload capabilities to enable/use on ethernet ports.
 169  * By default all capabilities are enabled.
 170  */
 171 static uint64_t dev_rx_offload = UINT64_MAX;
 172 static uint64_t dev_tx_offload = UINT64_MAX;
 173
 174 /*
 175  * global values that determine multi-seg policy
 176  */
 177 static uint32_t frag_tbl_sz;
 178 static uint32_t frame_buf_size = RTE_MBUF_DEFAULT_BUF_SIZE;
 179 static uint32_t mtu_size = RTE_ETHER_MTU;
 180 static uint64_t frag_ttl_ns = MAX_FRAG_TTL_NS;
 181
 182 /* application wide librte_ipsec/SA parameters */
 183 struct app_sa_prm app_sa_prm = {
 184                         .enable = 0,
 185                         .cache_sz = SA_CACHE_SZ
 186                 };
 187 static const char *cfgfile;
 188
 189 struct lcore_rx_queue {
 190         uint16_t port_id;
 191         uint8_t queue_id;
 192 } __rte_cache_aligned;
 193
 194 struct lcore_params {
 195         uint16_t port_id;
 196         uint8_t queue_id;
 197         uint8_t lcore_id;
 198 } __rte_cache_aligned;
 199
 200 static struct lcore_params lcore_params_array[MAX_LCORE_PARAMS];
 201
 202 static struct lcore_params *lcore_params;
 203 static uint16_t nb_lcore_params;
 204
 205 static struct rte_hash *cdev_map_in;
 206 static struct rte_hash *cdev_map_out;
 207
 208 struct buffer {
 209         uint16_t len;
 210         struct rte_mbuf *m_table[MAX_PKT_BURST] __rte_aligned(sizeof(void *));
 211 };
 212
 213 struct lcore_conf {
 214         uint16_t nb_rx_queue;
 215         struct lcore_rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 216         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 217         struct buffer tx_mbufs[RTE_MAX_ETHPORTS];
 218         struct ipsec_ctx inbound;
 219         struct ipsec_ctx outbound;
 220         struct rt_ctx *rt4_ctx;
 221         struct rt_ctx *rt6_ctx;
 222         struct {
 223                 struct rte_ip_frag_tbl *tbl;
 224                 struct rte_mempool *pool_dir;
 225                 struct rte_mempool *pool_indir;
 226                 struct rte_ip_frag_death_row dr;
 227         } frag;
 228 } __rte_cache_aligned;
 229
 230 static struct lcore_conf lcore_conf[RTE_MAX_LCORE];
 231
 232 static struct rte_eth_conf port_conf = {
 233         .rxmode = {
 234                 .mq_mode        = ETH_MQ_RX_RSS,
 235                 .max_rx_pkt_len = RTE_ETHER_MAX_LEN,
 236                 .split_hdr_size = 0,
 237                 .offloads = DEV_RX_OFFLOAD_CHECKSUM,
 238         },
 239         .rx_adv_conf = {
 240                 .rss_conf = {
 241                         .rss_key = NULL,
 242                         .rss_hf = ETH_RSS_IP | ETH_RSS_UDP |
 243                                 ETH_RSS_TCP | ETH_RSS_SCTP,
 244                 },
 245         },
 246         .txmode = {
 247                 .mq_mode = ETH_MQ_TX_NONE,
 248         },
 249 };
 250
 251 struct socket_ctx socket_ctx[NB_SOCKETS];
 252
 253 /*
 254  * Determine is multi-segment support required:
 255  *  - either frame buffer size is smaller then mtu
 256  *  - or reassmeble support is requested
 257  */
 258 static int
 259 multi_seg_required(void)
 260 {
 261         return (MTU_TO_FRAMELEN(mtu_size) + RTE_PKTMBUF_HEADROOM >
 262                 frame_buf_size || frag_tbl_sz != 0);
 263 }
 264
 265 static inline void
 266 adjust_ipv4_pktlen(struct rte_mbuf *m, const struct rte_ipv4_hdr *iph,
 267         uint32_t l2_len)
 268 {
 269         uint32_t plen, trim;
 270
 271         plen = rte_be_to_cpu_16(iph->total_length) + l2_len;
 272         if (plen < m->pkt_len) {
 273                 trim = m->pkt_len - plen;
 274                 rte_pktmbuf_trim(m, trim);
 275         }
 276 }
 277
 278 static inline void
 279 adjust_ipv6_pktlen(struct rte_mbuf *m, const struct rte_ipv6_hdr *iph,
 280         uint32_t l2_len)
 281 {
 282         uint32_t plen, trim;
 283
 284         plen = rte_be_to_cpu_16(iph->payload_len) + sizeof(*iph) + l2_len;
 285         if (plen < m->pkt_len) {
 286                 trim = m->pkt_len - plen;
 287                 rte_pktmbuf_trim(m, trim);
 288         }
 289 }
 290
 291 static inline void
 292 prepare_one_packet(struct rte_mbuf *pkt, struct ipsec_traffic *t)
 293 {
 294         const struct rte_ether_hdr *eth;
 295         const struct rte_ipv4_hdr *iph4;
 296         const struct rte_ipv6_hdr *iph6;
 297
 298         eth = rte_pktmbuf_mtod(pkt, const struct rte_ether_hdr *);
 299         if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
 300
 301                 iph4 = (const struct rte_ipv4_hdr *)rte_pktmbuf_adj(pkt,
 302                         RTE_ETHER_HDR_LEN);
 303                 adjust_ipv4_pktlen(pkt, iph4, 0);
 304
 305                 if (iph4->next_proto_id == IPPROTO_ESP)
 306                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 307                 else {
 308                         t->ip4.data[t->ip4.num] = &iph4->next_proto_id;
 309                         t->ip4.pkts[(t->ip4.num)++] = pkt;
 310                 }
 311                 pkt->l2_len = 0;
 312                 pkt->l3_len = sizeof(*iph4);
 313                 pkt->packet_type |= RTE_PTYPE_L3_IPV4;
 314         } else if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
 315                 int next_proto;
 316                 size_t l3len, ext_len;
 317                 uint8_t *p;
 318
 319                 /* get protocol type */
 320                 iph6 = (const struct rte_ipv6_hdr *)rte_pktmbuf_adj(pkt,
 321                         RTE_ETHER_HDR_LEN);
 322                 adjust_ipv6_pktlen(pkt, iph6, 0);
 323
 324                 next_proto = iph6->proto;
 325
 326                 /* determine l3 header size up to ESP extension */
 327                 l3len = sizeof(struct ip6_hdr);
 328                 p = rte_pktmbuf_mtod(pkt, uint8_t *);
 329                 while (next_proto != IPPROTO_ESP && l3len < pkt->data_len &&
 330                         (next_proto = rte_ipv6_get_next_ext(p + l3len,
 331                                                 next_proto, &ext_len)) >= 0)
 332                         l3len += ext_len;
 333
 334                 /* drop packet when IPv6 header exceeds first segment length */
 335                 if (unlikely(l3len > pkt->data_len)) {
 336                         rte_pktmbuf_free(pkt);
 337                         return;
 338                 }
 339
 340                 if (next_proto == IPPROTO_ESP)
 341                         t->ipsec.pkts[(t->ipsec.num)++] = pkt;
 342                 else {
 343                         t->ip6.data[t->ip6.num] = &iph6->proto;
 344                         t->ip6.pkts[(t->ip6.num)++] = pkt;
 345                 }
 346                 pkt->l2_len = 0;
 347                 pkt->l3_len = l3len;
 348                 pkt->packet_type |= RTE_PTYPE_L3_IPV6;
 349         } else {
 350                 /* Unknown/Unsupported type, drop the packet */
 351                 RTE_LOG(ERR, IPSEC, "Unsupported packet type 0x%x\n",
 352                         rte_be_to_cpu_16(eth->ether_type));
 353                 rte_pktmbuf_free(pkt);
 354                 return;
 355         }
 356
 357         /* Check if the packet has been processed inline. For inline protocol
 358          * processed packets, the metadata in the mbuf can be used to identify
 359          * the security processing done on the packet. The metadata will be
 360          * used to retrieve the application registered userdata associated
 361          * with the security session.
 362          */
 363
 364         if (pkt->ol_flags & PKT_RX_SEC_OFFLOAD) {
 365                 struct ipsec_sa *sa;
 366                 struct ipsec_mbuf_metadata *priv;
 367                 struct rte_security_ctx *ctx = (struct rte_security_ctx *)
 368                                                 rte_eth_dev_get_sec_ctx(
 369                                                 pkt->port);
 370
 371                 /* Retrieve the userdata registered. Here, the userdata
 372                  * registered is the SA pointer.
 373                  */
 374
 375                 sa = (struct ipsec_sa *)
 376                                 rte_security_get_userdata(ctx, pkt->udata64);
 377
 378                 if (sa == NULL) {
 379                         /* userdata could not be retrieved */
 380                         return;
 381                 }
 382
 383                 /* Save SA as priv member in mbuf. This will be used in the
 384                  * IPsec selector(SP-SA) check.
 385                  */
 386
 387                 priv = get_priv(pkt);
 388                 priv->sa = sa;
 389         }
 390 }
 391
 392 static inline void
 393 prepare_traffic(struct rte_mbuf **pkts, struct ipsec_traffic *t,
 394                 uint16_t nb_pkts)
 395 {
 396         int32_t i;
 397
 398         t->ipsec.num = 0;
 399         t->ip4.num = 0;
 400         t->ip6.num = 0;
 401
 402         for (i = 0; i < (nb_pkts - PREFETCH_OFFSET); i++) {
 403                 rte_prefetch0(rte_pktmbuf_mtod(pkts[i + PREFETCH_OFFSET],
 404                                         void *));
 405                 prepare_one_packet(pkts[i], t);
 406         }
 407         /* Process left packets */
 408         for (; i < nb_pkts; i++)
 409                 prepare_one_packet(pkts[i], t);
 410 }
 411
 412 static inline void
 413 prepare_tx_pkt(struct rte_mbuf *pkt, uint16_t port,
 414                 const struct lcore_conf *qconf)
 415 {
 416         struct ip *ip;
 417         struct rte_ether_hdr *ethhdr;
 418
 419         ip = rte_pktmbuf_mtod(pkt, struct ip *);
 420
 421         ethhdr = (struct rte_ether_hdr *)
 422                 rte_pktmbuf_prepend(pkt, RTE_ETHER_HDR_LEN);
 423
 424         if (ip->ip_v == IPVERSION) {
 425                 pkt->ol_flags |= qconf->outbound.ipv4_offloads;
 426                 pkt->l3_len = sizeof(struct ip);
 427                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 428
 429                 ip->ip_sum = 0;
 430
 431                 /* calculate IPv4 cksum in SW */
 432                 if ((pkt->ol_flags & PKT_TX_IP_CKSUM) == 0)
 433                         ip->ip_sum = rte_ipv4_cksum((struct rte_ipv4_hdr *)ip);
 434
 435                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4);
 436         } else {
 437                 pkt->ol_flags |= qconf->outbound.ipv6_offloads;
 438                 pkt->l3_len = sizeof(struct ip6_hdr);
 439                 pkt->l2_len = RTE_ETHER_HDR_LEN;
 440
 441                 ethhdr->ether_type = rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6);
 442         }
 443
 444         memcpy(&ethhdr->s_addr, &ethaddr_tbl[port].src,
 445                         sizeof(struct rte_ether_addr));
 446         memcpy(&ethhdr->d_addr, &ethaddr_tbl[port].dst,
 447                         sizeof(struct rte_ether_addr));
 448 }
 449
 450 static inline void
 451 prepare_tx_burst(struct rte_mbuf *pkts[], uint16_t nb_pkts, uint16_t port,
 452                 const struct lcore_conf *qconf)
 453 {
 454         int32_t i;
 455         const int32_t prefetch_offset = 2;
 456
 457         for (i = 0; i < (nb_pkts - prefetch_offset); i++) {
 458                 rte_mbuf_prefetch_part2(pkts[i + prefetch_offset]);
 459                 prepare_tx_pkt(pkts[i], port, qconf);
 460         }
 461         /* Process left packets */
 462         for (; i < nb_pkts; i++)
 463                 prepare_tx_pkt(pkts[i], port, qconf);
 464 }
 465
 466 /* Send burst of packets on an output interface */
 467 static inline int32_t
 468 send_burst(struct lcore_conf *qconf, uint16_t n, uint16_t port)
 469 {
 470         struct rte_mbuf **m_table;
 471         int32_t ret;
 472         uint16_t queueid;
 473
 474         queueid = qconf->tx_queue_id[port];
 475         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 476
 477         prepare_tx_burst(m_table, n, port, qconf);
 478
 479         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 480         if (unlikely(ret < n)) {
 481                 do {
 482                         rte_pktmbuf_free(m_table[ret]);
 483                 } while (++ret < n);
 484         }
 485
 486         return 0;
 487 }
 488
 489 /*
 490  * Helper function to fragment and queue for TX one packet.
 491  */
 492 static inline uint32_t
 493 send_fragment_packet(struct lcore_conf *qconf, struct rte_mbuf *m,
 494         uint16_t port, uint8_t proto)
 495 {
 496         struct buffer *tbl;
 497         uint32_t len, n;
 498         int32_t rc;
 499
 500         tbl =  qconf->tx_mbufs + port;
 501         len = tbl->len;
 502
 503         /* free space for new fragments */
 504         if (len + RTE_LIBRTE_IP_FRAG_MAX_FRAG >=  RTE_DIM(tbl->m_table)) {
 505                 send_burst(qconf, len, port);
 506                 len = 0;
 507         }
 508
 509         n = RTE_DIM(tbl->m_table) - len;
 510
 511         if (proto == IPPROTO_IP)
 512                 rc = rte_ipv4_fragment_packet(m, tbl->m_table + len,
 513                         n, mtu_size, qconf->frag.pool_dir,
 514                         qconf->frag.pool_indir);
 515         else
 516                 rc = rte_ipv6_fragment_packet(m, tbl->m_table + len,
 517                         n, mtu_size, qconf->frag.pool_dir,
 518                         qconf->frag.pool_indir);
 519
 520         if (rc >= 0)
 521                 len += rc;
 522         else
 523                 RTE_LOG(ERR, IPSEC,
 524                         "%s: failed to fragment packet with size %u, "
 525                         "error code: %d\n",
 526                         __func__, m->pkt_len, rte_errno);
 527
 528         rte_pktmbuf_free(m);
 529         return len;
 530 }
 531
 532 /* Enqueue a single packet, and send burst if queue is filled */
 533 static inline int32_t
 534 send_single_packet(struct rte_mbuf *m, uint16_t port, uint8_t proto)
 535 {
 536         uint32_t lcore_id;
 537         uint16_t len;
 538         struct lcore_conf *qconf;
 539
 540         lcore_id = rte_lcore_id();
 541
 542         qconf = &lcore_conf[lcore_id];
 543         len = qconf->tx_mbufs[port].len;
 544
 545         if (m->pkt_len <= mtu_size) {
 546                 qconf->tx_mbufs[port].m_table[len] = m;
 547                 len++;
 548
 549         /* need to fragment the packet */
 550         } else if (frag_tbl_sz > 0)
 551                 len = send_fragment_packet(qconf, m, port, proto);
 552         else
 553                 rte_pktmbuf_free(m);
 554
 555         /* enough pkts to be sent */
 556         if (unlikely(len == MAX_PKT_BURST)) {
 557                 send_burst(qconf, MAX_PKT_BURST, port);
 558                 len = 0;
 559         }
 560
 561         qconf->tx_mbufs[port].len = len;
 562         return 0;
 563 }
 564
 565 static inline void
 566 inbound_sp_sa(struct sp_ctx *sp, struct sa_ctx *sa, struct traffic_type *ip,
 567                 uint16_t lim)
 568 {
 569         struct rte_mbuf *m;
 570         uint32_t i, j, res, sa_idx;
 571
 572         if (ip->num == 0 || sp == NULL)
 573                 return;
 574
 575         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 576                         ip->num, DEFAULT_MAX_CATEGORIES);
 577
 578         j = 0;
 579         for (i = 0; i < ip->num; i++) {
 580                 m = ip->pkts[i];
 581                 res = ip->res[i];
 582                 if (res == BYPASS) {
 583                         ip->pkts[j++] = m;
 584                         continue;
 585                 }
 586                 if (res == DISCARD) {
 587                         rte_pktmbuf_free(m);
 588                         continue;
 589                 }
 590
 591                 /* Only check SPI match for processed IPSec packets */
 592                 if (i < lim && ((m->ol_flags & PKT_RX_SEC_OFFLOAD) == 0)) {
 593                         rte_pktmbuf_free(m);
 594                         continue;
 595                 }
 596
 597                 sa_idx = res - 1;
 598                 if (!inbound_sa_check(sa, m, sa_idx)) {
 599                         rte_pktmbuf_free(m);
 600                         continue;
 601                 }
 602                 ip->pkts[j++] = m;
 603         }
 604         ip->num = j;
 605 }
 606
 607 static void
 608 split46_traffic(struct ipsec_traffic *trf, struct rte_mbuf *mb[], uint32_t num)
 609 {
 610         uint32_t i, n4, n6;
 611         struct ip *ip;
 612         struct rte_mbuf *m;
 613
 614         n4 = trf->ip4.num;
 615         n6 = trf->ip6.num;
 616
 617         for (i = 0; i < num; i++) {
 618
 619                 m = mb[i];
 620                 ip = rte_pktmbuf_mtod(m, struct ip *);
 621
 622                 if (ip->ip_v == IPVERSION) {
 623                         trf->ip4.pkts[n4] = m;
 624                         trf->ip4.data[n4] = rte_pktmbuf_mtod_offset(m,
 625                                         uint8_t *, offsetof(struct ip, ip_p));
 626                         n4++;
 627                 } else if (ip->ip_v == IP6_VERSION) {
 628                         trf->ip6.pkts[n6] = m;
 629                         trf->ip6.data[n6] = rte_pktmbuf_mtod_offset(m,
 630                                         uint8_t *,
 631                                         offsetof(struct ip6_hdr, ip6_nxt));
 632                         n6++;
 633                 } else
 634                         rte_pktmbuf_free(m);
 635         }
 636
 637         trf->ip4.num = n4;
 638         trf->ip6.num = n6;
 639 }
 640
 641
 642 static inline void
 643 process_pkts_inbound(struct ipsec_ctx *ipsec_ctx,
 644                 struct ipsec_traffic *traffic)
 645 {
 646         uint16_t nb_pkts_in, n_ip4, n_ip6;
 647
 648         n_ip4 = traffic->ip4.num;
 649         n_ip6 = traffic->ip6.num;
 650
 651         if (app_sa_prm.enable == 0) {
 652                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 653                                 traffic->ipsec.num, MAX_PKT_BURST);
 654                 split46_traffic(traffic, traffic->ipsec.pkts, nb_pkts_in);
 655         } else {
 656                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 657                         traffic->ipsec.saptr, traffic->ipsec.num);
 658                 ipsec_process(ipsec_ctx, traffic);
 659         }
 660
 661         inbound_sp_sa(ipsec_ctx->sp4_ctx, ipsec_ctx->sa_ctx, &traffic->ip4,
 662                         n_ip4);
 663
 664         inbound_sp_sa(ipsec_ctx->sp6_ctx, ipsec_ctx->sa_ctx, &traffic->ip6,
 665                         n_ip6);
 666 }
 667
 668 static inline void
 669 outbound_sp(struct sp_ctx *sp, struct traffic_type *ip,
 670                 struct traffic_type *ipsec)
 671 {
 672         struct rte_mbuf *m;
 673         uint32_t i, j, sa_idx;
 674
 675         if (ip->num == 0 || sp == NULL)
 676                 return;
 677
 678         rte_acl_classify((struct rte_acl_ctx *)sp, ip->data, ip->res,
 679                         ip->num, DEFAULT_MAX_CATEGORIES);
 680
 681         j = 0;
 682         for (i = 0; i < ip->num; i++) {
 683                 m = ip->pkts[i];
 684                 sa_idx = ip->res[i] - 1;
 685                 if (ip->res[i] == DISCARD)
 686                         rte_pktmbuf_free(m);
 687                 else if (ip->res[i] == BYPASS)
 688                         ip->pkts[j++] = m;
 689                 else {
 690                         ipsec->res[ipsec->num] = sa_idx;
 691                         ipsec->pkts[ipsec->num++] = m;
 692                 }
 693         }
 694         ip->num = j;
 695 }
 696
 697 static inline void
 698 process_pkts_outbound(struct ipsec_ctx *ipsec_ctx,
 699                 struct ipsec_traffic *traffic)
 700 {
 701         struct rte_mbuf *m;
 702         uint16_t idx, nb_pkts_out, i;
 703
 704         /* Drop any IPsec traffic from protected ports */
 705         for (i = 0; i < traffic->ipsec.num; i++)
 706                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 707
 708         traffic->ipsec.num = 0;
 709
 710         outbound_sp(ipsec_ctx->sp4_ctx, &traffic->ip4, &traffic->ipsec);
 711
 712         outbound_sp(ipsec_ctx->sp6_ctx, &traffic->ip6, &traffic->ipsec);
 713
 714         if (app_sa_prm.enable == 0) {
 715
 716                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 717                                 traffic->ipsec.res, traffic->ipsec.num,
 718                                 MAX_PKT_BURST);
 719
 720                 for (i = 0; i < nb_pkts_out; i++) {
 721                         m = traffic->ipsec.pkts[i];
 722                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 723                         if (ip->ip_v == IPVERSION) {
 724                                 idx = traffic->ip4.num++;
 725                                 traffic->ip4.pkts[idx] = m;
 726                         } else {
 727                                 idx = traffic->ip6.num++;
 728                                 traffic->ip6.pkts[idx] = m;
 729                         }
 730                 }
 731         } else {
 732                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 733                         traffic->ipsec.saptr, traffic->ipsec.num);
 734                 ipsec_process(ipsec_ctx, traffic);
 735         }
 736 }
 737
 738 static inline void
 739 process_pkts_inbound_nosp(struct ipsec_ctx *ipsec_ctx,
 740                 struct ipsec_traffic *traffic)
 741 {
 742         struct rte_mbuf *m;
 743         uint32_t nb_pkts_in, i, idx;
 744
 745         /* Drop any IPv4 traffic from unprotected ports */
 746         for (i = 0; i < traffic->ip4.num; i++)
 747                 rte_pktmbuf_free(traffic->ip4.pkts[i]);
 748
 749         traffic->ip4.num = 0;
 750
 751         /* Drop any IPv6 traffic from unprotected ports */
 752         for (i = 0; i < traffic->ip6.num; i++)
 753                 rte_pktmbuf_free(traffic->ip6.pkts[i]);
 754
 755         traffic->ip6.num = 0;
 756
 757         if (app_sa_prm.enable == 0) {
 758
 759                 nb_pkts_in = ipsec_inbound(ipsec_ctx, traffic->ipsec.pkts,
 760                                 traffic->ipsec.num, MAX_PKT_BURST);
 761
 762                 for (i = 0; i < nb_pkts_in; i++) {
 763                         m = traffic->ipsec.pkts[i];
 764                         struct ip *ip = rte_pktmbuf_mtod(m, struct ip *);
 765                         if (ip->ip_v == IPVERSION) {
 766                                 idx = traffic->ip4.num++;
 767                                 traffic->ip4.pkts[idx] = m;
 768                         } else {
 769                                 idx = traffic->ip6.num++;
 770                                 traffic->ip6.pkts[idx] = m;
 771                         }
 772                 }
 773         } else {
 774                 inbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.pkts,
 775                         traffic->ipsec.saptr, traffic->ipsec.num);
 776                 ipsec_process(ipsec_ctx, traffic);
 777         }
 778 }
 779
 780 static inline void
 781 process_pkts_outbound_nosp(struct ipsec_ctx *ipsec_ctx,
 782                 struct ipsec_traffic *traffic)
 783 {
 784         struct rte_mbuf *m;
 785         uint32_t nb_pkts_out, i, n;
 786         struct ip *ip;
 787
 788         /* Drop any IPsec traffic from protected ports */
 789         for (i = 0; i < traffic->ipsec.num; i++)
 790                 rte_pktmbuf_free(traffic->ipsec.pkts[i]);
 791
 792         n = 0;
 793
 794         for (i = 0; i < traffic->ip4.num; i++) {
 795                 traffic->ipsec.pkts[n] = traffic->ip4.pkts[i];
 796                 traffic->ipsec.res[n++] = single_sa_idx;
 797         }
 798
 799         for (i = 0; i < traffic->ip6.num; i++) {
 800                 traffic->ipsec.pkts[n] = traffic->ip6.pkts[i];
 801                 traffic->ipsec.res[n++] = single_sa_idx;
 802         }
 803
 804         traffic->ip4.num = 0;
 805         traffic->ip6.num = 0;
 806         traffic->ipsec.num = n;
 807
 808         if (app_sa_prm.enable == 0) {
 809
 810                 nb_pkts_out = ipsec_outbound(ipsec_ctx, traffic->ipsec.pkts,
 811                                 traffic->ipsec.res, traffic->ipsec.num,
 812                                 MAX_PKT_BURST);
 813
 814                 /* They all sue the same SA (ip4 or ip6 tunnel) */
 815                 m = traffic->ipsec.pkts[0];
 816                 ip = rte_pktmbuf_mtod(m, struct ip *);
 817                 if (ip->ip_v == IPVERSION) {
 818                         traffic->ip4.num = nb_pkts_out;
 819                         for (i = 0; i < nb_pkts_out; i++)
 820                                 traffic->ip4.pkts[i] = traffic->ipsec.pkts[i];
 821                 } else {
 822                         traffic->ip6.num = nb_pkts_out;
 823                         for (i = 0; i < nb_pkts_out; i++)
 824                                 traffic->ip6.pkts[i] = traffic->ipsec.pkts[i];
 825                 }
 826         } else {
 827                 outbound_sa_lookup(ipsec_ctx->sa_ctx, traffic->ipsec.res,
 828                         traffic->ipsec.saptr, traffic->ipsec.num);
 829                 ipsec_process(ipsec_ctx, traffic);
 830         }
 831 }
 832
 833 static inline int32_t
 834 get_hop_for_offload_pkt(struct rte_mbuf *pkt, int is_ipv6)
 835 {
 836         struct ipsec_mbuf_metadata *priv;
 837         struct ipsec_sa *sa;
 838
 839         priv = get_priv(pkt);
 840
 841         sa = priv->sa;
 842         if (unlikely(sa == NULL)) {
 843                 RTE_LOG(ERR, IPSEC, "SA not saved in private data\n");
 844                 goto fail;
 845         }
 846
 847         if (is_ipv6)
 848                 return sa->portid;
 849
 850         /* else */
 851         return (sa->portid | RTE_LPM_LOOKUP_SUCCESS);
 852
 853 fail:
 854         if (is_ipv6)
 855                 return -1;
 856
 857         /* else */
 858         return 0;
 859 }
 860
 861 static inline void
 862 route4_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 863 {
 864         uint32_t hop[MAX_PKT_BURST * 2];
 865         uint32_t dst_ip[MAX_PKT_BURST * 2];
 866         int32_t pkt_hop = 0;
 867         uint16_t i, offset;
 868         uint16_t lpm_pkts = 0;
 869
 870         if (nb_pkts == 0)
 871                 return;
 872
 873         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 874          * have port ID in the SA
 875          */
 876
 877         for (i = 0; i < nb_pkts; i++) {
 878                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 879                         /* Security offload not enabled. So an LPM lookup is
 880                          * required to get the hop
 881                          */
 882                         offset = offsetof(struct ip, ip_dst);
 883                         dst_ip[lpm_pkts] = *rte_pktmbuf_mtod_offset(pkts[i],
 884                                         uint32_t *, offset);
 885                         dst_ip[lpm_pkts] = rte_be_to_cpu_32(dst_ip[lpm_pkts]);
 886                         lpm_pkts++;
 887                 }
 888         }
 889
 890         rte_lpm_lookup_bulk((struct rte_lpm *)rt_ctx, dst_ip, hop, lpm_pkts);
 891
 892         lpm_pkts = 0;
 893
 894         for (i = 0; i < nb_pkts; i++) {
 895                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 896                         /* Read hop from the SA */
 897                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 0);
 898                 } else {
 899                         /* Need to use hop returned by lookup */
 900                         pkt_hop = hop[lpm_pkts++];
 901                 }
 902
 903                 if ((pkt_hop & RTE_LPM_LOOKUP_SUCCESS) == 0) {
 904                         rte_pktmbuf_free(pkts[i]);
 905                         continue;
 906                 }
 907                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IP);
 908         }
 909 }
 910
 911 static inline void
 912 route6_pkts(struct rt_ctx *rt_ctx, struct rte_mbuf *pkts[], uint8_t nb_pkts)
 913 {
 914         int32_t hop[MAX_PKT_BURST * 2];
 915         uint8_t dst_ip[MAX_PKT_BURST * 2][16];
 916         uint8_t *ip6_dst;
 917         int32_t pkt_hop = 0;
 918         uint16_t i, offset;
 919         uint16_t lpm_pkts = 0;
 920
 921         if (nb_pkts == 0)
 922                 return;
 923
 924         /* Need to do an LPM lookup for non-inline packets. Inline packets will
 925          * have port ID in the SA
 926          */
 927
 928         for (i = 0; i < nb_pkts; i++) {
 929                 if (!(pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD)) {
 930                         /* Security offload not enabled. So an LPM lookup is
 931                          * required to get the hop
 932                          */
 933                         offset = offsetof(struct ip6_hdr, ip6_dst);
 934                         ip6_dst = rte_pktmbuf_mtod_offset(pkts[i], uint8_t *,
 935                                         offset);
 936                         memcpy(&dst_ip[lpm_pkts][0], ip6_dst, 16);
 937                         lpm_pkts++;
 938                 }
 939         }
 940
 941         rte_lpm6_lookup_bulk_func((struct rte_lpm6 *)rt_ctx, dst_ip, hop,
 942                         lpm_pkts);
 943
 944         lpm_pkts = 0;
 945
 946         for (i = 0; i < nb_pkts; i++) {
 947                 if (pkts[i]->ol_flags & PKT_TX_SEC_OFFLOAD) {
 948                         /* Read hop from the SA */
 949                         pkt_hop = get_hop_for_offload_pkt(pkts[i], 1);
 950                 } else {
 951                         /* Need to use hop returned by lookup */
 952                         pkt_hop = hop[lpm_pkts++];
 953                 }
 954
 955                 if (pkt_hop == -1) {
 956                         rte_pktmbuf_free(pkts[i]);
 957                         continue;
 958                 }
 959                 send_single_packet(pkts[i], pkt_hop & 0xff, IPPROTO_IPV6);
 960         }
 961 }
 962
 963 static inline void
 964 process_pkts(struct lcore_conf *qconf, struct rte_mbuf **pkts,
 965                 uint8_t nb_pkts, uint16_t portid)
 966 {
 967         struct ipsec_traffic traffic;
 968
 969         prepare_traffic(pkts, &traffic, nb_pkts);
 970
 971         if (unlikely(single_sa)) {
 972                 if (is_unprotected_port(portid))
 973                         process_pkts_inbound_nosp(&qconf->inbound, &traffic);
 974                 else
 975                         process_pkts_outbound_nosp(&qconf->outbound, &traffic);
 976         } else {
 977                 if (is_unprotected_port(portid))
 978                         process_pkts_inbound(&qconf->inbound, &traffic);
 979                 else
 980                         process_pkts_outbound(&qconf->outbound, &traffic);
 981         }
 982
 983         route4_pkts(qconf->rt4_ctx, traffic.ip4.pkts, traffic.ip4.num);
 984         route6_pkts(qconf->rt6_ctx, traffic.ip6.pkts, traffic.ip6.num);
 985 }
 986
 987 static inline void
 988 drain_tx_buffers(struct lcore_conf *qconf)
 989 {
 990         struct buffer *buf;
 991         uint32_t portid;
 992
 993         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 994                 buf = &qconf->tx_mbufs[portid];
 995                 if (buf->len == 0)
 996                         continue;
 997                 send_burst(qconf, buf->len, portid);
 998                 buf->len = 0;
 999         }
1000 }
1001
1002 static inline void
1003 drain_crypto_buffers(struct lcore_conf *qconf)
1004 {
1005         uint32_t i;
1006         struct ipsec_ctx *ctx;
1007
1008         /* drain inbound buffers*/
1009         ctx = &qconf->inbound;
1010         for (i = 0; i != ctx->nb_qps; i++) {
1011                 if (ctx->tbl[i].len != 0)
1012                         enqueue_cop_burst(ctx->tbl  + i);
1013         }
1014
1015         /* drain outbound buffers*/
1016         ctx = &qconf->outbound;
1017         for (i = 0; i != ctx->nb_qps; i++) {
1018                 if (ctx->tbl[i].len != 0)
1019                         enqueue_cop_burst(ctx->tbl  + i);
1020         }
1021 }
1022
1023 static void
1024 drain_inbound_crypto_queues(const struct lcore_conf *qconf,
1025                 struct ipsec_ctx *ctx)
1026 {
1027         uint32_t n;
1028         struct ipsec_traffic trf;
1029
1030         if (app_sa_prm.enable == 0) {
1031
1032                 /* dequeue packets from crypto-queue */
1033                 n = ipsec_inbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1034                         RTE_DIM(trf.ipsec.pkts));
1035
1036                 trf.ip4.num = 0;
1037                 trf.ip6.num = 0;
1038
1039                 /* split traffic by ipv4-ipv6 */
1040                 split46_traffic(&trf, trf.ipsec.pkts, n);
1041         } else
1042                 ipsec_cqp_process(ctx, &trf);
1043
1044         /* process ipv4 packets */
1045         if (trf.ip4.num != 0) {
1046                 inbound_sp_sa(ctx->sp4_ctx, ctx->sa_ctx, &trf.ip4, 0);
1047                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1048         }
1049
1050         /* process ipv6 packets */
1051         if (trf.ip6.num != 0) {
1052                 inbound_sp_sa(ctx->sp6_ctx, ctx->sa_ctx, &trf.ip6, 0);
1053                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1054         }
1055 }
1056
1057 static void
1058 drain_outbound_crypto_queues(const struct lcore_conf *qconf,
1059                 struct ipsec_ctx *ctx)
1060 {
1061         uint32_t n;
1062         struct ipsec_traffic trf;
1063
1064         if (app_sa_prm.enable == 0) {
1065
1066                 /* dequeue packets from crypto-queue */
1067                 n = ipsec_outbound_cqp_dequeue(ctx, trf.ipsec.pkts,
1068                         RTE_DIM(trf.ipsec.pkts));
1069
1070                 trf.ip4.num = 0;
1071                 trf.ip6.num = 0;
1072
1073                 /* split traffic by ipv4-ipv6 */
1074                 split46_traffic(&trf, trf.ipsec.pkts, n);
1075         } else
1076                 ipsec_cqp_process(ctx, &trf);
1077
1078         /* process ipv4 packets */
1079         if (trf.ip4.num != 0)
1080                 route4_pkts(qconf->rt4_ctx, trf.ip4.pkts, trf.ip4.num);
1081
1082         /* process ipv6 packets */
1083         if (trf.ip6.num != 0)
1084                 route6_pkts(qconf->rt6_ctx, trf.ip6.pkts, trf.ip6.num);
1085 }
1086
1087 /* main processing loop */
1088 void
1089 ipsec_poll_mode_worker(void)
1090 {
1091         struct rte_mbuf *pkts[MAX_PKT_BURST];
1092         uint32_t lcore_id;
1093         uint64_t prev_tsc, diff_tsc, cur_tsc;
1094         int32_t i, nb_rx;
1095         uint16_t portid;
1096         uint8_t queueid;
1097         struct lcore_conf *qconf;
1098         int32_t rc, socket_id;
1099         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1)
1100                         / US_PER_S * BURST_TX_DRAIN_US;
1101         struct lcore_rx_queue *rxql;
1102
1103         prev_tsc = 0;
1104         lcore_id = rte_lcore_id();
1105         qconf = &lcore_conf[lcore_id];
1106         rxql = qconf->rx_queue_list;
1107         socket_id = rte_lcore_to_socket_id(lcore_id);
1108
1109         qconf->rt4_ctx = socket_ctx[socket_id].rt_ip4;
1110         qconf->rt6_ctx = socket_ctx[socket_id].rt_ip6;
1111         qconf->inbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_in;
1112         qconf->inbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_in;
1113         qconf->inbound.sa_ctx = socket_ctx[socket_id].sa_in;
1114         qconf->inbound.cdev_map = cdev_map_in;
1115         qconf->inbound.session_pool = socket_ctx[socket_id].session_pool;
1116         qconf->inbound.session_priv_pool =
1117                         socket_ctx[socket_id].session_priv_pool;
1118         qconf->outbound.sp4_ctx = socket_ctx[socket_id].sp_ip4_out;
1119         qconf->outbound.sp6_ctx = socket_ctx[socket_id].sp_ip6_out;
1120         qconf->outbound.sa_ctx = socket_ctx[socket_id].sa_out;
1121         qconf->outbound.cdev_map = cdev_map_out;
1122         qconf->outbound.session_pool = socket_ctx[socket_id].session_pool;
1123         qconf->outbound.session_priv_pool =
1124                         socket_ctx[socket_id].session_priv_pool;
1125         qconf->frag.pool_dir = socket_ctx[socket_id].mbuf_pool;
1126         qconf->frag.pool_indir = socket_ctx[socket_id].mbuf_pool_indir;
1127
1128         rc = ipsec_sad_lcore_cache_init(app_sa_prm.cache_sz);
1129         if (rc != 0) {
1130                 RTE_LOG(ERR, IPSEC,
1131                         "SAD cache init on lcore %u, failed with code: %d\n",
1132                         lcore_id, rc);
1133                 return;
1134         }
1135
1136         if (qconf->nb_rx_queue == 0) {
1137                 RTE_LOG(DEBUG, IPSEC, "lcore %u has nothing to do\n",
1138                         lcore_id);
1139                 return;
1140         }
1141
1142         RTE_LOG(INFO, IPSEC, "entering main loop on lcore %u\n", lcore_id);
1143
1144         for (i = 0; i < qconf->nb_rx_queue; i++) {
1145                 portid = rxql[i].port_id;
1146                 queueid = rxql[i].queue_id;
1147                 RTE_LOG(INFO, IPSEC,
1148                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
1149                         lcore_id, portid, queueid);
1150         }
1151
1152         while (!force_quit) {
1153                 cur_tsc = rte_rdtsc();
1154
1155                 /* TX queue buffer drain */
1156                 diff_tsc = cur_tsc - prev_tsc;
1157
1158                 if (unlikely(diff_tsc > drain_tsc)) {
1159                         drain_tx_buffers(qconf);
1160                         drain_crypto_buffers(qconf);
1161                         prev_tsc = cur_tsc;
1162                 }
1163
1164                 for (i = 0; i < qconf->nb_rx_queue; ++i) {
1165
1166                         /* Read packets from RX queues */
1167                         portid = rxql[i].port_id;
1168                         queueid = rxql[i].queue_id;
1169                         nb_rx = rte_eth_rx_burst(portid, queueid,
1170                                         pkts, MAX_PKT_BURST);
1171
1172                         if (nb_rx > 0)
1173                                 process_pkts(qconf, pkts, nb_rx, portid);
1174
1175                         /* dequeue and process completed crypto-ops */
1176                         if (is_unprotected_port(portid))
1177                                 drain_inbound_crypto_queues(qconf,
1178                                         &qconf->inbound);
1179                         else
1180                                 drain_outbound_crypto_queues(qconf,
1181                                         &qconf->outbound);
1182                 }
1183         }
1184 }
1185
1186 int
1187 check_flow_params(uint16_t fdir_portid, uint8_t fdir_qid)
1188 {
1189         uint16_t i;
1190         uint16_t portid;
1191         uint8_t queueid;
1192
1193         for (i = 0; i < nb_lcore_params; ++i) {
1194                 portid = lcore_params_array[i].port_id;
1195                 if (portid == fdir_portid) {
1196                         queueid = lcore_params_array[i].queue_id;
1197                         if (queueid == fdir_qid)
1198                                 break;
1199                 }
1200
1201                 if (i == nb_lcore_params - 1)
1202                         return -1;
1203         }
1204
1205         return 1;
1206 }
1207
1208 static int32_t
1209 check_poll_mode_params(struct eh_conf *eh_conf)
1210 {
1211         uint8_t lcore;
1212         uint16_t portid;
1213         uint16_t i;
1214         int32_t socket_id;
1215
1216         if (!eh_conf)
1217                 return -EINVAL;
1218
1219         if (eh_conf->mode != EH_PKT_TRANSFER_MODE_POLL)
1220                 return 0;
1221
1222         if (lcore_params == NULL) {
1223                 printf("Error: No port/queue/core mappings\n");
1224                 return -1;
1225         }
1226
1227         for (i = 0; i < nb_lcore_params; ++i) {
1228                 lcore = lcore_params[i].lcore_id;
1229                 if (!rte_lcore_is_enabled(lcore)) {
1230                         printf("error: lcore %hhu is not enabled in "
1231                                 "lcore mask\n", lcore);
1232                         return -1;
1233                 }
1234                 socket_id = rte_lcore_to_socket_id(lcore);
1235                 if (socket_id != 0 && numa_on == 0) {
1236                         printf("warning: lcore %hhu is on socket %d "
1237                                 "with numa off\n",
1238                                 lcore, socket_id);
1239                 }
1240                 portid = lcore_params[i].port_id;
1241                 if ((enabled_port_mask & (1 << portid)) == 0) {
1242                         printf("port %u is not enabled in port mask\n", portid);
1243                         return -1;
1244                 }
1245                 if (!rte_eth_dev_is_valid_port(portid)) {
1246                         printf("port %u is not present on the board\n", portid);
1247                         return -1;
1248                 }
1249         }
1250         return 0;
1251 }
1252
1253 static uint8_t
1254 get_port_nb_rx_queues(const uint16_t port)
1255 {
1256         int32_t queue = -1;
1257         uint16_t i;
1258
1259         for (i = 0; i < nb_lcore_params; ++i) {
1260                 if (lcore_params[i].port_id == port &&
1261                                 lcore_params[i].queue_id > queue)
1262                         queue = lcore_params[i].queue_id;
1263         }
1264         return (uint8_t)(++queue);
1265 }
1266
1267 static int32_t
1268 init_lcore_rx_queues(void)
1269 {
1270         uint16_t i, nb_rx_queue;
1271         uint8_t lcore;
1272
1273         for (i = 0; i < nb_lcore_params; ++i) {
1274                 lcore = lcore_params[i].lcore_id;
1275                 nb_rx_queue = lcore_conf[lcore].nb_rx_queue;
1276                 if (nb_rx_queue >= MAX_RX_QUEUE_PER_LCORE) {
1277                         printf("error: too many queues (%u) for lcore: %u\n",
1278                                         nb_rx_queue + 1, lcore);
1279                         return -1;
1280                 }
1281                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].port_id =
1282                         lcore_params[i].port_id;
1283                 lcore_conf[lcore].rx_queue_list[nb_rx_queue].queue_id =
1284                         lcore_params[i].queue_id;
1285                 lcore_conf[lcore].nb_rx_queue++;
1286         }
1287         return 0;
1288 }
1289
1290 /* display usage */
1291 static void
1292 print_usage(const char *prgname)
1293 {
1294         fprintf(stderr, "%s [EAL options] --"
1295                 " -p PORTMASK"
1296                 " [-P]"
1297                 " [-u PORTMASK]"
1298                 " [-j FRAMESIZE]"
1299                 " [-l]"
1300                 " [-w REPLAY_WINDOW_SIZE]"
1301                 " [-e]"
1302                 " [-a]"
1303                 " [-c]"
1304                 " [-s NUMBER_OF_MBUFS_IN_PKT_POOL]"
1305                 " -f CONFIG_FILE"
1306                 " --config (port,queue,lcore)[,(port,queue,lcore)]"
1307                 " [--single-sa SAIDX]"
1308                 " [--cryptodev_mask MASK]"
1309                 " [--transfer-mode MODE]"
1310                 " [--event-schedule-type TYPE]"
1311                 " [--" CMD_LINE_OPT_RX_OFFLOAD " RX_OFFLOAD_MASK]"
1312                 " [--" CMD_LINE_OPT_TX_OFFLOAD " TX_OFFLOAD_MASK]"
1313                 " [--" CMD_LINE_OPT_REASSEMBLE " REASSEMBLE_TABLE_SIZE]"
1314                 " [--" CMD_LINE_OPT_MTU " MTU]"
1315                 "\n\n"
1316                 "  -p PORTMASK: Hexadecimal bitmask of ports to configure\n"
1317                 "  -P : Enable promiscuous mode\n"
1318                 "  -u PORTMASK: Hexadecimal bitmask of unprotected ports\n"
1319                 "  -j FRAMESIZE: Data buffer size, minimum (and default)\n"
1320                 "     value: RTE_MBUF_DEFAULT_BUF_SIZE\n"
1321                 "  -l enables code-path that uses librte_ipsec\n"
1322                 "  -w REPLAY_WINDOW_SIZE specifies IPsec SQN replay window\n"
1323                 "     size for each SA\n"
1324                 "  -e enables ESN\n"
1325                 "  -a enables SA SQN atomic behaviour\n"
1326                 "  -c specifies inbound SAD cache size,\n"
1327                 "     zero value disables the cache (default value: 128)\n"
1328                 "  -s number of mbufs in packet pool, if not specified number\n"
1329                 "     of mbufs will be calculated based on number of cores,\n"
1330                 "     ports and crypto queues\n"
1331                 "  -f CONFIG_FILE: Configuration file\n"
1332                 "  --config (port,queue,lcore): Rx queue configuration. In poll\n"
1333                 "                               mode determines which queues from\n"
1334                 "                               which ports are mapped to which cores.\n"
1335                 "                               In event mode this option is not used\n"
1336                 "                               as packets are dynamically scheduled\n"
1337                 "                               to cores by HW.\n"
1338                 "  --single-sa SAIDX: In poll mode use single SA index for\n"
1339                 "                     outbound traffic, bypassing the SP\n"
1340                 "                     In event mode selects driver submode,\n"
1341                 "                     SA index value is ignored\n"
1342                 "  --cryptodev_mask MASK: Hexadecimal bitmask of the crypto\n"
1343                 "                         devices to configure\n"
1344                 "  --transfer-mode MODE\n"
1345                 "               \"poll\"  : Packet transfer via polling (default)\n"
1346                 "               \"event\" : Packet transfer via event device\n"
1347                 "  --event-schedule-type TYPE queue schedule type, used only when\n"
1348                 "                             transfer mode is set to event\n"
1349                 "               \"ordered\"  : Ordered (default)\n"
1350                 "               \"atomic\"   : Atomic\n"
1351                 "               \"parallel\" : Parallel\n"
1352                 "  --" CMD_LINE_OPT_RX_OFFLOAD
1353                 ": bitmask of the RX HW offload capabilities to enable/use\n"
1354                 "                         (DEV_RX_OFFLOAD_*)\n"
1355                 "  --" CMD_LINE_OPT_TX_OFFLOAD
1356                 ": bitmask of the TX HW offload capabilities to enable/use\n"
1357                 "                         (DEV_TX_OFFLOAD_*)\n"
1358                 "  --" CMD_LINE_OPT_REASSEMBLE " NUM"
1359                 ": max number of entries in reassemble(fragment) table\n"
1360                 "    (zero (default value) disables reassembly)\n"
1361                 "  --" CMD_LINE_OPT_MTU " MTU"
1362                 ": MTU value on all ports (default value: 1500)\n"
1363                 "    outgoing packets with bigger size will be fragmented\n"
1364                 "    incoming packets with bigger size will be discarded\n"
1365                 "  --" CMD_LINE_OPT_FRAG_TTL " FRAG_TTL_NS"
1366                 ": fragments lifetime in nanoseconds, default\n"
1367                 "    and maximum value is 10.000.000.000 ns (10 s)\n"
1368                 "\n",
1369                 prgname);
1370 }
1371
1372 static int
1373 parse_mask(const char *str, uint64_t *val)
1374 {
1375         char *end;
1376         unsigned long t;
1377
1378         errno = 0;
1379         t = strtoul(str, &end, 0);
1380         if (errno != 0 || end[0] != 0)
1381                 return -EINVAL;
1382
1383         *val = t;
1384         return 0;
1385 }
1386
1387 static int32_t
1388 parse_portmask(const char *portmask)
1389 {
1390         char *end = NULL;
1391         unsigned long pm;
1392
1393         /* parse hexadecimal string */
1394         pm = strtoul(portmask, &end, 16);
1395         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
1396                 return -1;
1397
1398         if ((pm == 0) && errno)
1399                 return -1;
1400
1401         return pm;
1402 }
1403
1404 static int64_t
1405 parse_decimal(const char *str)
1406 {
1407         char *end = NULL;
1408         uint64_t num;
1409
1410         num = strtoull(str, &end, 10);
1411         if ((str[0] == '\0') || (end == NULL) || (*end != '\0')
1412                 || num > INT64_MAX)
1413                 return -1;
1414
1415         return num;
1416 }
1417
1418 static int32_t
1419 parse_config(const char *q_arg)
1420 {
1421         char s[256];
1422         const char *p, *p0 = q_arg;
1423         char *end;
1424         enum fieldnames {
1425                 FLD_PORT = 0,
1426                 FLD_QUEUE,
1427                 FLD_LCORE,
1428                 _NUM_FLD
1429         };
1430         unsigned long int_fld[_NUM_FLD];
1431         char *str_fld[_NUM_FLD];
1432         int32_t i;
1433         uint32_t size;
1434
1435         nb_lcore_params = 0;
1436
1437         while ((p = strchr(p0, '(')) != NULL) {
1438                 ++p;
1439                 p0 = strchr(p, ')');
1440                 if (p0 == NULL)
1441                         return -1;
1442
1443                 size = p0 - p;
1444                 if (size >= sizeof(s))
1445                         return -1;
1446
1447                 snprintf(s, sizeof(s), "%.*s", size, p);
1448                 if (rte_strsplit(s, sizeof(s), str_fld, _NUM_FLD, ',') !=
1449                                 _NUM_FLD)
1450                         return -1;
1451                 for (i = 0; i < _NUM_FLD; i++) {
1452                         errno = 0;
1453                         int_fld[i] = strtoul(str_fld[i], &end, 0);
1454                         if (errno != 0 || end == str_fld[i] || int_fld[i] > 255)
1455                                 return -1;
1456                 }
1457                 if (nb_lcore_params >= MAX_LCORE_PARAMS) {
1458                         printf("exceeded max number of lcore params: %hu\n",
1459                                 nb_lcore_params);
1460                         return -1;
1461                 }
1462                 lcore_params_array[nb_lcore_params].port_id =
1463                         (uint8_t)int_fld[FLD_PORT];
1464                 lcore_params_array[nb_lcore_params].queue_id =
1465                         (uint8_t)int_fld[FLD_QUEUE];
1466                 lcore_params_array[nb_lcore_params].lcore_id =
1467                         (uint8_t)int_fld[FLD_LCORE];
1468                 ++nb_lcore_params;
1469         }
1470         lcore_params = lcore_params_array;
1471         return 0;
1472 }
1473
1474 static void
1475 print_app_sa_prm(const struct app_sa_prm *prm)
1476 {
1477         printf("librte_ipsec usage: %s\n",
1478                 (prm->enable == 0) ? "disabled" : "enabled");
1479
1480         printf("replay window size: %u\n", prm->window_size);
1481         printf("ESN: %s\n", (prm->enable_esn == 0) ? "disabled" : "enabled");
1482         printf("SA flags: %#" PRIx64 "\n", prm->flags);
1483         printf("Frag TTL: %" PRIu64 " ns\n", frag_ttl_ns);
1484 }
1485
1486 static int
1487 parse_transfer_mode(struct eh_conf *conf, const char *optarg)
1488 {
1489         if (!strcmp(CMD_LINE_ARG_POLL, optarg))
1490                 conf->mode = EH_PKT_TRANSFER_MODE_POLL;
1491         else if (!strcmp(CMD_LINE_ARG_EVENT, optarg))
1492                 conf->mode = EH_PKT_TRANSFER_MODE_EVENT;
1493         else {
1494                 printf("Unsupported packet transfer mode\n");
1495                 return -EINVAL;
1496         }
1497
1498         return 0;
1499 }
1500
1501 static int
1502 parse_schedule_type(struct eh_conf *conf, const char *optarg)
1503 {
1504         struct eventmode_conf *em_conf = NULL;
1505
1506         /* Get eventmode conf */
1507         em_conf = conf->mode_params;
1508
1509         if (!strcmp(CMD_LINE_ARG_ORDERED, optarg))
1510                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
1511         else if (!strcmp(CMD_LINE_ARG_ATOMIC, optarg))
1512                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ATOMIC;
1513         else if (!strcmp(CMD_LINE_ARG_PARALLEL, optarg))
1514                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_PARALLEL;
1515         else {
1516                 printf("Unsupported queue schedule type\n");
1517                 return -EINVAL;
1518         }
1519
1520         return 0;
1521 }
1522
1523 static int32_t
1524 parse_args(int32_t argc, char **argv, struct eh_conf *eh_conf)
1525 {
1526         int opt;
1527         int64_t ret;
1528         char **argvopt;
1529         int32_t option_index;
1530         char *prgname = argv[0];
1531         int32_t f_present = 0;
1532
1533         argvopt = argv;
1534
1535         while ((opt = getopt_long(argc, argvopt, "aelp:Pu:f:j:w:c:s:",
1536                                 lgopts, &option_index)) != EOF) {
1537
1538                 switch (opt) {
1539                 case 'p':
1540                         enabled_port_mask = parse_portmask(optarg);
1541                         if (enabled_port_mask == 0) {
1542                                 printf("invalid portmask\n");
1543                                 print_usage(prgname);
1544                                 return -1;
1545                         }
1546                         break;
1547                 case 'P':
1548                         printf("Promiscuous mode selected\n");
1549                         promiscuous_on = 1;
1550                         break;
1551                 case 'u':
1552                         unprotected_port_mask = parse_portmask(optarg);
1553                         if (unprotected_port_mask == 0) {
1554                                 printf("invalid unprotected portmask\n");
1555                                 print_usage(prgname);
1556                                 return -1;
1557                         }
1558                         break;
1559                 case 'f':
1560                         if (f_present == 1) {
1561                                 printf("\"-f\" option present more than "
1562                                         "once!\n");
1563                                 print_usage(prgname);
1564                                 return -1;
1565                         }
1566                         cfgfile = optarg;
1567                         f_present = 1;
1568                         break;
1569
1570                 case 's':
1571                         ret = parse_decimal(optarg);
1572                         if (ret < 0) {
1573                                 printf("Invalid number of buffers in a pool: "
1574                                         "%s\n", optarg);
1575                                 print_usage(prgname);
1576                                 return -1;
1577                         }
1578
1579                         nb_bufs_in_pool = ret;
1580                         break;
1581
1582                 case 'j':
1583                         ret = parse_decimal(optarg);
1584                         if (ret < RTE_MBUF_DEFAULT_BUF_SIZE ||
1585                                         ret > UINT16_MAX) {
1586                                 printf("Invalid frame buffer size value: %s\n",
1587                                         optarg);
1588                                 print_usage(prgname);
1589                                 return -1;
1590                         }
1591                         frame_buf_size = ret;
1592                         printf("Custom frame buffer size %u\n", frame_buf_size);
1593                         break;
1594                 case 'l':
1595                         app_sa_prm.enable = 1;
1596                         break;
1597                 case 'w':
1598                         app_sa_prm.window_size = parse_decimal(optarg);
1599                         break;
1600                 case 'e':
1601                         app_sa_prm.enable_esn = 1;
1602                         break;
1603                 case 'a':
1604                         app_sa_prm.enable = 1;
1605                         app_sa_prm.flags |= RTE_IPSEC_SAFLAG_SQN_ATOM;
1606                         break;
1607                 case 'c':
1608                         ret = parse_decimal(optarg);
1609                         if (ret < 0) {
1610                                 printf("Invalid SA cache size: %s\n", optarg);
1611                                 print_usage(prgname);
1612                                 return -1;
1613                         }
1614                         app_sa_prm.cache_sz = ret;
1615                         break;
1616                 case CMD_LINE_OPT_CONFIG_NUM:
1617                         ret = parse_config(optarg);
1618                         if (ret) {
1619                                 printf("Invalid config\n");
1620                                 print_usage(prgname);
1621                                 return -1;
1622                         }
1623                         break;
1624                 case CMD_LINE_OPT_SINGLE_SA_NUM:
1625                         ret = parse_decimal(optarg);
1626                         if (ret == -1 || ret > UINT32_MAX) {
1627                                 printf("Invalid argument[sa_idx]\n");
1628                                 print_usage(prgname);
1629                                 return -1;
1630                         }
1631
1632                         /* else */
1633                         single_sa = 1;
1634                         single_sa_idx = ret;
1635                         eh_conf->ipsec_mode = EH_IPSEC_MODE_TYPE_DRIVER;
1636                         printf("Configured with single SA index %u\n",
1637                                         single_sa_idx);
1638                         break;
1639                 case CMD_LINE_OPT_CRYPTODEV_MASK_NUM:
1640                         ret = parse_portmask(optarg);
1641                         if (ret == -1) {
1642                                 printf("Invalid argument[portmask]\n");
1643                                 print_usage(prgname);
1644                                 return -1;
1645                         }
1646
1647                         /* else */
1648                         enabled_cryptodev_mask = ret;
1649                         break;
1650
1651                 case CMD_LINE_OPT_TRANSFER_MODE_NUM:
1652                         ret = parse_transfer_mode(eh_conf, optarg);
1653                         if (ret < 0) {
1654                                 printf("Invalid packet transfer mode\n");
1655                                 print_usage(prgname);
1656                                 return -1;
1657                         }
1658                         break;
1659
1660                 case CMD_LINE_OPT_SCHEDULE_TYPE_NUM:
1661                         ret = parse_schedule_type(eh_conf, optarg);
1662                         if (ret < 0) {
1663                                 printf("Invalid queue schedule type\n");
1664                                 print_usage(prgname);
1665                                 return -1;
1666                         }
1667                         break;
1668
1669                 case CMD_LINE_OPT_RX_OFFLOAD_NUM:
1670                         ret = parse_mask(optarg, &dev_rx_offload);
1671                         if (ret != 0) {
1672                                 printf("Invalid argument for \'%s\': %s\n",
1673                                         CMD_LINE_OPT_RX_OFFLOAD, optarg);
1674                                 print_usage(prgname);
1675                                 return -1;
1676                         }
1677                         break;
1678                 case CMD_LINE_OPT_TX_OFFLOAD_NUM:
1679                         ret = parse_mask(optarg, &dev_tx_offload);
1680                         if (ret != 0) {
1681                                 printf("Invalid argument for \'%s\': %s\n",
1682                                         CMD_LINE_OPT_TX_OFFLOAD, optarg);
1683                                 print_usage(prgname);
1684                                 return -1;
1685                         }
1686                         break;
1687                 case CMD_LINE_OPT_REASSEMBLE_NUM:
1688                         ret = parse_decimal(optarg);
1689                         if (ret < 0 || ret > UINT32_MAX) {
1690                                 printf("Invalid argument for \'%s\': %s\n",
1691                                         CMD_LINE_OPT_REASSEMBLE, optarg);
1692                                 print_usage(prgname);
1693                                 return -1;
1694                         }
1695                         frag_tbl_sz = ret;
1696                         break;
1697                 case CMD_LINE_OPT_MTU_NUM:
1698                         ret = parse_decimal(optarg);
1699                         if (ret < 0 || ret > RTE_IPV4_MAX_PKT_LEN) {
1700                                 printf("Invalid argument for \'%s\': %s\n",
1701                                         CMD_LINE_OPT_MTU, optarg);
1702                                 print_usage(prgname);
1703                                 return -1;
1704                         }
1705                         mtu_size = ret;
1706                         break;
1707                 case CMD_LINE_OPT_FRAG_TTL_NUM:
1708                         ret = parse_decimal(optarg);
1709                         if (ret < 0 || ret > MAX_FRAG_TTL_NS) {
1710                                 printf("Invalid argument for \'%s\': %s\n",
1711                                         CMD_LINE_OPT_MTU, optarg);
1712                                 print_usage(prgname);
1713                                 return -1;
1714                         }
1715                         frag_ttl_ns = ret;
1716                         break;
1717                 default:
1718                         print_usage(prgname);
1719                         return -1;
1720                 }
1721         }
1722
1723         if (f_present == 0) {
1724                 printf("Mandatory option \"-f\" not present\n");
1725                 return -1;
1726         }
1727
1728         /* check do we need to enable multi-seg support */
1729         if (multi_seg_required()) {
1730                 /* legacy mode doesn't support multi-seg */
1731                 app_sa_prm.enable = 1;
1732                 printf("frame buf size: %u, mtu: %u, "
1733                         "number of reassemble entries: %u\n"
1734                         "multi-segment support is required\n",
1735                         frame_buf_size, mtu_size, frag_tbl_sz);
1736         }
1737
1738         print_app_sa_prm(&app_sa_prm);
1739
1740         if (optind >= 0)
1741                 argv[optind-1] = prgname;
1742
1743         ret = optind-1;
1744         optind = 1; /* reset getopt lib */
1745         return ret;
1746 }
1747
1748 static void
1749 print_ethaddr(const char *name, const struct rte_ether_addr *eth_addr)
1750 {
1751         char buf[RTE_ETHER_ADDR_FMT_SIZE];
1752         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
1753         printf("%s%s", name, buf);
1754 }
1755
1756 /*
1757  * Update destination ethaddr for the port.
1758  */
1759 int
1760 add_dst_ethaddr(uint16_t port, const struct rte_ether_addr *addr)
1761 {
1762         if (port >= RTE_DIM(ethaddr_tbl))
1763                 return -EINVAL;
1764
1765         ethaddr_tbl[port].dst = ETHADDR_TO_UINT64(addr);
1766         return 0;
1767 }
1768
1769 /* Check the link status of all ports in up to 9s, and print them finally */
1770 static void
1771 check_all_ports_link_status(uint32_t port_mask)
1772 {
1773 #define CHECK_INTERVAL 100 /* 100ms */
1774 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
1775         uint16_t portid;
1776         uint8_t count, all_ports_up, print_flag = 0;
1777         struct rte_eth_link link;
1778         int ret;
1779
1780         printf("\nChecking link status");
1781         fflush(stdout);
1782         for (count = 0; count <= MAX_CHECK_TIME; count++) {
1783                 all_ports_up = 1;
1784                 RTE_ETH_FOREACH_DEV(portid) {
1785                         if ((port_mask & (1 << portid)) == 0)
1786                                 continue;
1787                         memset(&link, 0, sizeof(link));
1788                         ret = rte_eth_link_get_nowait(portid, &link);
1789                         if (ret < 0) {
1790                                 all_ports_up = 0;
1791                                 if (print_flag == 1)
1792                                         printf("Port %u link get failed: %s\n",
1793                                                 portid, rte_strerror(-ret));
1794                                 continue;
1795                         }
1796                         /* print link status if flag set */
1797                         if (print_flag == 1) {
1798                                 if (link.link_status)
1799                                         printf(
1800                                         "Port%d Link Up - speed %u Mbps -%s\n",
1801                                                 portid, link.link_speed,
1802                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
1803                                         ("full-duplex") : ("half-duplex\n"));
1804                                 else
1805                                         printf("Port %d Link Down\n", portid);
1806                                 continue;
1807                         }
1808                         /* clear all_ports_up flag if any link down */
1809                         if (link.link_status == ETH_LINK_DOWN) {
1810                                 all_ports_up = 0;
1811                                 break;
1812                         }
1813                 }
1814                 /* after finally printing all link status, get out */
1815                 if (print_flag == 1)
1816                         break;
1817
1818                 if (all_ports_up == 0) {
1819                         printf(".");
1820                         fflush(stdout);
1821                         rte_delay_ms(CHECK_INTERVAL);
1822                 }
1823
1824                 /* set the print_flag if all ports up or timeout */
1825                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
1826                         print_flag = 1;
1827                         printf("done\n");
1828                 }
1829         }
1830 }
1831
1832 static int32_t
1833 add_mapping(struct rte_hash *map, const char *str, uint16_t cdev_id,
1834                 uint16_t qp, struct lcore_params *params,
1835                 struct ipsec_ctx *ipsec_ctx,
1836                 const struct rte_cryptodev_capabilities *cipher,
1837                 const struct rte_cryptodev_capabilities *auth,
1838                 const struct rte_cryptodev_capabilities *aead)
1839 {
1840         int32_t ret = 0;
1841         unsigned long i;
1842         struct cdev_key key = { 0 };
1843
1844         key.lcore_id = params->lcore_id;
1845         if (cipher)
1846                 key.cipher_algo = cipher->sym.cipher.algo;
1847         if (auth)
1848                 key.auth_algo = auth->sym.auth.algo;
1849         if (aead)
1850                 key.aead_algo = aead->sym.aead.algo;
1851
1852         ret = rte_hash_lookup(map, &key);
1853         if (ret != -ENOENT)
1854                 return 0;
1855
1856         for (i = 0; i < ipsec_ctx->nb_qps; i++)
1857                 if (ipsec_ctx->tbl[i].id == cdev_id)
1858                         break;
1859
1860         if (i == ipsec_ctx->nb_qps) {
1861                 if (ipsec_ctx->nb_qps == MAX_QP_PER_LCORE) {
1862                         printf("Maximum number of crypto devices assigned to "
1863                                 "a core, increase MAX_QP_PER_LCORE value\n");
1864                         return 0;
1865                 }
1866                 ipsec_ctx->tbl[i].id = cdev_id;
1867                 ipsec_ctx->tbl[i].qp = qp;
1868                 ipsec_ctx->nb_qps++;
1869                 printf("%s cdev mapping: lcore %u using cdev %u qp %u "
1870                                 "(cdev_id_qp %lu)\n", str, key.lcore_id,
1871                                 cdev_id, qp, i);
1872         }
1873
1874         ret = rte_hash_add_key_data(map, &key, (void *)i);
1875         if (ret < 0) {
1876                 printf("Faled to insert cdev mapping for (lcore %u, "
1877                                 "cdev %u, qp %u), errno %d\n",
1878                                 key.lcore_id, ipsec_ctx->tbl[i].id,
1879                                 ipsec_ctx->tbl[i].qp, ret);
1880                 return 0;
1881         }
1882
1883         return 1;
1884 }
1885
1886 static int32_t
1887 add_cdev_mapping(struct rte_cryptodev_info *dev_info, uint16_t cdev_id,
1888                 uint16_t qp, struct lcore_params *params)
1889 {
1890         int32_t ret = 0;
1891         const struct rte_cryptodev_capabilities *i, *j;
1892         struct rte_hash *map;
1893         struct lcore_conf *qconf;
1894         struct ipsec_ctx *ipsec_ctx;
1895         const char *str;
1896
1897         qconf = &lcore_conf[params->lcore_id];
1898
1899         if ((unprotected_port_mask & (1 << params->port_id)) == 0) {
1900                 map = cdev_map_out;
1901                 ipsec_ctx = &qconf->outbound;
1902                 str = "Outbound";
1903         } else {
1904                 map = cdev_map_in;
1905                 ipsec_ctx = &qconf->inbound;
1906                 str = "Inbound";
1907         }
1908
1909         /* Required cryptodevs with operation chainning */
1910         if (!(dev_info->feature_flags &
1911                                 RTE_CRYPTODEV_FF_SYM_OPERATION_CHAINING))
1912                 return ret;
1913
1914         for (i = dev_info->capabilities;
1915                         i->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; i++) {
1916                 if (i->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1917                         continue;
1918
1919                 if (i->sym.xform_type == RTE_CRYPTO_SYM_XFORM_AEAD) {
1920                         ret |= add_mapping(map, str, cdev_id, qp, params,
1921                                         ipsec_ctx, NULL, NULL, i);
1922                         continue;
1923                 }
1924
1925                 if (i->sym.xform_type != RTE_CRYPTO_SYM_XFORM_CIPHER)
1926                         continue;
1927
1928                 for (j = dev_info->capabilities;
1929                                 j->op != RTE_CRYPTO_OP_TYPE_UNDEFINED; j++) {
1930                         if (j->op != RTE_CRYPTO_OP_TYPE_SYMMETRIC)
1931                                 continue;
1932
1933                         if (j->sym.xform_type != RTE_CRYPTO_SYM_XFORM_AUTH)
1934                                 continue;
1935
1936                         ret |= add_mapping(map, str, cdev_id, qp, params,
1937                                                 ipsec_ctx, i, j, NULL);
1938                 }
1939         }
1940
1941         return ret;
1942 }
1943
1944 /* Check if the device is enabled by cryptodev_mask */
1945 static int
1946 check_cryptodev_mask(uint8_t cdev_id)
1947 {
1948         if (enabled_cryptodev_mask & (1 << cdev_id))
1949                 return 0;
1950
1951         return -1;
1952 }
1953
1954 static uint16_t
1955 cryptodevs_init(uint16_t req_queue_num)
1956 {
1957         struct rte_cryptodev_config dev_conf;
1958         struct rte_cryptodev_qp_conf qp_conf;
1959         uint16_t idx, max_nb_qps, qp, total_nb_qps, i;
1960         int16_t cdev_id;
1961         struct rte_hash_parameters params = { 0 };
1962
1963         const uint64_t mseg_flag = multi_seg_required() ?
1964                                 RTE_CRYPTODEV_FF_IN_PLACE_SGL : 0;
1965
1966         params.entries = CDEV_MAP_ENTRIES;
1967         params.key_len = sizeof(struct cdev_key);
1968         params.hash_func = rte_jhash;
1969         params.hash_func_init_val = 0;
1970         params.socket_id = rte_socket_id();
1971
1972         params.name = "cdev_map_in";
1973         cdev_map_in = rte_hash_create(&params);
1974         if (cdev_map_in == NULL)
1975                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1976                                 rte_errno);
1977
1978         params.name = "cdev_map_out";
1979         cdev_map_out = rte_hash_create(&params);
1980         if (cdev_map_out == NULL)
1981                 rte_panic("Failed to create cdev_map hash table, errno = %d\n",
1982                                 rte_errno);
1983
1984         printf("lcore/cryptodev/qp mappings:\n");
1985
1986         idx = 0;
1987         total_nb_qps = 0;
1988         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
1989                 struct rte_cryptodev_info cdev_info;
1990
1991                 if (check_cryptodev_mask((uint8_t)cdev_id))
1992                         continue;
1993
1994                 rte_cryptodev_info_get(cdev_id, &cdev_info);
1995
1996                 if ((mseg_flag & cdev_info.feature_flags) != mseg_flag)
1997                         rte_exit(EXIT_FAILURE,
1998                                 "Device %hd does not support \'%s\' feature\n",
1999                                 cdev_id,
2000                                 rte_cryptodev_get_feature_name(mseg_flag));
2001
2002                 if (nb_lcore_params > cdev_info.max_nb_queue_pairs)
2003                         max_nb_qps = cdev_info.max_nb_queue_pairs;
2004                 else
2005                         max_nb_qps = nb_lcore_params;
2006
2007                 qp = 0;
2008                 i = 0;
2009                 while (qp < max_nb_qps && i < nb_lcore_params) {
2010                         if (add_cdev_mapping(&cdev_info, cdev_id, qp,
2011                                                 &lcore_params[idx]))
2012                                 qp++;
2013                         idx++;
2014                         idx = idx % nb_lcore_params;
2015                         i++;
2016                 }
2017
2018                 qp = RTE_MIN(max_nb_qps, RTE_MAX(req_queue_num, qp));
2019                 if (qp == 0)
2020                         continue;
2021
2022                 total_nb_qps += qp;
2023                 dev_conf.socket_id = rte_cryptodev_socket_id(cdev_id);
2024                 dev_conf.nb_queue_pairs = qp;
2025                 dev_conf.ff_disable = RTE_CRYPTODEV_FF_ASYMMETRIC_CRYPTO;
2026
2027                 uint32_t dev_max_sess = cdev_info.sym.max_nb_sessions;
2028                 if (dev_max_sess != 0 && dev_max_sess < CDEV_MP_NB_OBJS)
2029                         rte_exit(EXIT_FAILURE,
2030                                 "Device does not support at least %u "
2031                                 "sessions", CDEV_MP_NB_OBJS);
2032
2033                 if (rte_cryptodev_configure(cdev_id, &dev_conf))
2034                         rte_panic("Failed to initialize cryptodev %u\n",
2035                                         cdev_id);
2036
2037                 qp_conf.nb_descriptors = CDEV_QUEUE_DESC;
2038                 qp_conf.mp_session =
2039                         socket_ctx[dev_conf.socket_id].session_pool;
2040                 qp_conf.mp_session_private =
2041                         socket_ctx[dev_conf.socket_id].session_priv_pool;
2042                 for (qp = 0; qp < dev_conf.nb_queue_pairs; qp++)
2043                         if (rte_cryptodev_queue_pair_setup(cdev_id, qp,
2044                                         &qp_conf, dev_conf.socket_id))
2045                                 rte_panic("Failed to setup queue %u for "
2046                                                 "cdev_id %u\n", 0, cdev_id);
2047
2048                 if (rte_cryptodev_start(cdev_id))
2049                         rte_panic("Failed to start cryptodev %u\n",
2050                                         cdev_id);
2051         }
2052
2053         printf("\n");
2054
2055         return total_nb_qps;
2056 }
2057
2058 static void
2059 port_init(uint16_t portid, uint64_t req_rx_offloads, uint64_t req_tx_offloads)
2060 {
2061         uint32_t frame_size;
2062         struct rte_eth_dev_info dev_info;
2063         struct rte_eth_txconf *txconf;
2064         uint16_t nb_tx_queue, nb_rx_queue;
2065         uint16_t tx_queueid, rx_queueid, queue, lcore_id;
2066         int32_t ret, socket_id;
2067         struct lcore_conf *qconf;
2068         struct rte_ether_addr ethaddr;
2069         struct rte_eth_conf local_port_conf = port_conf;
2070
2071         ret = rte_eth_dev_info_get(portid, &dev_info);
2072         if (ret != 0)
2073                 rte_exit(EXIT_FAILURE,
2074                         "Error during getting device (port %u) info: %s\n",
2075                         portid, strerror(-ret));
2076
2077         /* limit allowed HW offloafs, as user requested */
2078         dev_info.rx_offload_capa &= dev_rx_offload;
2079         dev_info.tx_offload_capa &= dev_tx_offload;
2080
2081         printf("Configuring device port %u:\n", portid);
2082
2083         ret = rte_eth_macaddr_get(portid, &ethaddr);
2084         if (ret != 0)
2085                 rte_exit(EXIT_FAILURE,
2086                         "Error getting MAC address (port %u): %s\n",
2087                         portid, rte_strerror(-ret));
2088
2089         ethaddr_tbl[portid].src = ETHADDR_TO_UINT64(&ethaddr);
2090         print_ethaddr("Address: ", &ethaddr);
2091         printf("\n");
2092
2093         nb_rx_queue = get_port_nb_rx_queues(portid);
2094         nb_tx_queue = nb_lcores;
2095
2096         if (nb_rx_queue > dev_info.max_rx_queues)
2097                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2098                                 "(max rx queue is %u)\n",
2099                                 nb_rx_queue, dev_info.max_rx_queues);
2100
2101         if (nb_tx_queue > dev_info.max_tx_queues)
2102                 rte_exit(EXIT_FAILURE, "Error: queue %u not available "
2103                                 "(max tx queue is %u)\n",
2104                                 nb_tx_queue, dev_info.max_tx_queues);
2105
2106         printf("Creating queues: nb_rx_queue=%d nb_tx_queue=%u...\n",
2107                         nb_rx_queue, nb_tx_queue);
2108
2109         frame_size = MTU_TO_FRAMELEN(mtu_size);
2110         if (frame_size > local_port_conf.rxmode.max_rx_pkt_len)
2111                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_JUMBO_FRAME;
2112         local_port_conf.rxmode.max_rx_pkt_len = frame_size;
2113
2114         if (multi_seg_required()) {
2115                 local_port_conf.rxmode.offloads |= DEV_RX_OFFLOAD_SCATTER;
2116                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_MULTI_SEGS;
2117         }
2118
2119         local_port_conf.rxmode.offloads |= req_rx_offloads;
2120         local_port_conf.txmode.offloads |= req_tx_offloads;
2121
2122         /* Check that all required capabilities are supported */
2123         if ((local_port_conf.rxmode.offloads & dev_info.rx_offload_capa) !=
2124                         local_port_conf.rxmode.offloads)
2125                 rte_exit(EXIT_FAILURE,
2126                         "Error: port %u required RX offloads: 0x%" PRIx64
2127                         ", avaialbe RX offloads: 0x%" PRIx64 "\n",
2128                         portid, local_port_conf.rxmode.offloads,
2129                         dev_info.rx_offload_capa);
2130
2131         if ((local_port_conf.txmode.offloads & dev_info.tx_offload_capa) !=
2132                         local_port_conf.txmode.offloads)
2133                 rte_exit(EXIT_FAILURE,
2134                         "Error: port %u required TX offloads: 0x%" PRIx64
2135                         ", avaialbe TX offloads: 0x%" PRIx64 "\n",
2136                         portid, local_port_conf.txmode.offloads,
2137                         dev_info.tx_offload_capa);
2138
2139         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
2140                 local_port_conf.txmode.offloads |=
2141                         DEV_TX_OFFLOAD_MBUF_FAST_FREE;
2142
2143         if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_IPV4_CKSUM)
2144                 local_port_conf.txmode.offloads |= DEV_TX_OFFLOAD_IPV4_CKSUM;
2145
2146         printf("port %u configurng rx_offloads=0x%" PRIx64
2147                 ", tx_offloads=0x%" PRIx64 "\n",
2148                 portid, local_port_conf.rxmode.offloads,
2149                 local_port_conf.txmode.offloads);
2150
2151         local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
2152                 dev_info.flow_type_rss_offloads;
2153         if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
2154                         port_conf.rx_adv_conf.rss_conf.rss_hf) {
2155                 printf("Port %u modified RSS hash function based on hardware support,"
2156                         "requested:%#"PRIx64" configured:%#"PRIx64"\n",
2157                         portid,
2158                         port_conf.rx_adv_conf.rss_conf.rss_hf,
2159                         local_port_conf.rx_adv_conf.rss_conf.rss_hf);
2160         }
2161
2162         ret = rte_eth_dev_configure(portid, nb_rx_queue, nb_tx_queue,
2163                         &local_port_conf);
2164         if (ret < 0)
2165                 rte_exit(EXIT_FAILURE, "Cannot configure device: "
2166                                 "err=%d, port=%d\n", ret, portid);
2167
2168         ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd, &nb_txd);
2169         if (ret < 0)
2170                 rte_exit(EXIT_FAILURE, "Cannot adjust number of descriptors: "
2171                                 "err=%d, port=%d\n", ret, portid);
2172
2173         /* init one TX queue per lcore */
2174         tx_queueid = 0;
2175         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2176                 if (rte_lcore_is_enabled(lcore_id) == 0)
2177                         continue;
2178
2179                 if (numa_on)
2180                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2181                 else
2182                         socket_id = 0;
2183
2184                 /* init TX queue */
2185                 printf("Setup txq=%u,%d,%d\n", lcore_id, tx_queueid, socket_id);
2186
2187                 txconf = &dev_info.default_txconf;
2188                 txconf->offloads = local_port_conf.txmode.offloads;
2189
2190                 ret = rte_eth_tx_queue_setup(portid, tx_queueid, nb_txd,
2191                                 socket_id, txconf);
2192                 if (ret < 0)
2193                         rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
2194                                         "err=%d, port=%d\n", ret, portid);
2195
2196                 qconf = &lcore_conf[lcore_id];
2197                 qconf->tx_queue_id[portid] = tx_queueid;
2198
2199                 /* Pre-populate pkt offloads based on capabilities */
2200                 qconf->outbound.ipv4_offloads = PKT_TX_IPV4;
2201                 qconf->outbound.ipv6_offloads = PKT_TX_IPV6;
2202                 if (local_port_conf.txmode.offloads & DEV_TX_OFFLOAD_IPV4_CKSUM)
2203                         qconf->outbound.ipv4_offloads |= PKT_TX_IP_CKSUM;
2204
2205                 tx_queueid++;
2206
2207                 /* init RX queues */
2208                 for (queue = 0; queue < qconf->nb_rx_queue; ++queue) {
2209                         struct rte_eth_rxconf rxq_conf;
2210
2211                         if (portid != qconf->rx_queue_list[queue].port_id)
2212                                 continue;
2213
2214                         rx_queueid = qconf->rx_queue_list[queue].queue_id;
2215
2216                         printf("Setup rxq=%d,%d,%d\n", portid, rx_queueid,
2217                                         socket_id);
2218
2219                         rxq_conf = dev_info.default_rxconf;
2220                         rxq_conf.offloads = local_port_conf.rxmode.offloads;
2221                         ret = rte_eth_rx_queue_setup(portid, rx_queueid,
2222                                         nb_rxd, socket_id, &rxq_conf,
2223                                         socket_ctx[socket_id].mbuf_pool);
2224                         if (ret < 0)
2225                                 rte_exit(EXIT_FAILURE,
2226                                         "rte_eth_rx_queue_setup: err=%d, "
2227                                         "port=%d\n", ret, portid);
2228                 }
2229         }
2230         printf("\n");
2231 }
2232
2233 static size_t
2234 max_session_size(void)
2235 {
2236         size_t max_sz, sz;
2237         void *sec_ctx;
2238         int16_t cdev_id, port_id, n;
2239
2240         max_sz = 0;
2241         n =  rte_cryptodev_count();
2242         for (cdev_id = 0; cdev_id != n; cdev_id++) {
2243                 sz = rte_cryptodev_sym_get_private_session_size(cdev_id);
2244                 if (sz > max_sz)
2245                         max_sz = sz;
2246                 /*
2247                  * If crypto device is security capable, need to check the
2248                  * size of security session as well.
2249                  */
2250
2251                 /* Get security context of the crypto device */
2252                 sec_ctx = rte_cryptodev_get_sec_ctx(cdev_id);
2253                 if (sec_ctx == NULL)
2254                         continue;
2255
2256                 /* Get size of security session */
2257                 sz = rte_security_session_get_size(sec_ctx);
2258                 if (sz > max_sz)
2259                         max_sz = sz;
2260         }
2261
2262         RTE_ETH_FOREACH_DEV(port_id) {
2263                 if ((enabled_port_mask & (1 << port_id)) == 0)
2264                         continue;
2265
2266                 sec_ctx = rte_eth_dev_get_sec_ctx(port_id);
2267                 if (sec_ctx == NULL)
2268                         continue;
2269
2270                 sz = rte_security_session_get_size(sec_ctx);
2271                 if (sz > max_sz)
2272                         max_sz = sz;
2273         }
2274
2275         return max_sz;
2276 }
2277
2278 static void
2279 session_pool_init(struct socket_ctx *ctx, int32_t socket_id, size_t sess_sz)
2280 {
2281         char mp_name[RTE_MEMPOOL_NAMESIZE];
2282         struct rte_mempool *sess_mp;
2283
2284         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2285                         "sess_mp_%u", socket_id);
2286         sess_mp = rte_cryptodev_sym_session_pool_create(
2287                         mp_name, CDEV_MP_NB_OBJS,
2288                         sess_sz, CDEV_MP_CACHE_SZ, 0,
2289                         socket_id);
2290         ctx->session_pool = sess_mp;
2291
2292         if (ctx->session_pool == NULL)
2293                 rte_exit(EXIT_FAILURE,
2294                         "Cannot init session pool on socket %d\n", socket_id);
2295         else
2296                 printf("Allocated session pool on socket %d\n", socket_id);
2297 }
2298
2299 static void
2300 session_priv_pool_init(struct socket_ctx *ctx, int32_t socket_id,
2301         size_t sess_sz)
2302 {
2303         char mp_name[RTE_MEMPOOL_NAMESIZE];
2304         struct rte_mempool *sess_mp;
2305
2306         snprintf(mp_name, RTE_MEMPOOL_NAMESIZE,
2307                         "sess_mp_priv_%u", socket_id);
2308         sess_mp = rte_mempool_create(mp_name,
2309                         CDEV_MP_NB_OBJS,
2310                         sess_sz,
2311                         CDEV_MP_CACHE_SZ,
2312                         0, NULL, NULL, NULL,
2313                         NULL, socket_id,
2314                         0);
2315         ctx->session_priv_pool = sess_mp;
2316
2317         if (ctx->session_priv_pool == NULL)
2318                 rte_exit(EXIT_FAILURE,
2319                         "Cannot init session priv pool on socket %d\n",
2320                         socket_id);
2321         else
2322                 printf("Allocated session priv pool on socket %d\n",
2323                         socket_id);
2324 }
2325
2326 static void
2327 pool_init(struct socket_ctx *ctx, int32_t socket_id, uint32_t nb_mbuf)
2328 {
2329         char s[64];
2330         int32_t ms;
2331
2332         snprintf(s, sizeof(s), "mbuf_pool_%d", socket_id);
2333         ctx->mbuf_pool = rte_pktmbuf_pool_create(s, nb_mbuf,
2334                         MEMPOOL_CACHE_SIZE, ipsec_metadata_size(),
2335                         frame_buf_size, socket_id);
2336
2337         /*
2338          * if multi-segment support is enabled, then create a pool
2339          * for indirect mbufs.
2340          */
2341         ms = multi_seg_required();
2342         if (ms != 0) {
2343                 snprintf(s, sizeof(s), "mbuf_pool_indir_%d", socket_id);
2344                 ctx->mbuf_pool_indir = rte_pktmbuf_pool_create(s, nb_mbuf,
2345                         MEMPOOL_CACHE_SIZE, 0, 0, socket_id);
2346         }
2347
2348         if (ctx->mbuf_pool == NULL || (ms != 0 && ctx->mbuf_pool_indir == NULL))
2349                 rte_exit(EXIT_FAILURE, "Cannot init mbuf pool on socket %d\n",
2350                                 socket_id);
2351         else
2352                 printf("Allocated mbuf pool on socket %d\n", socket_id);
2353 }
2354
2355 static inline int
2356 inline_ipsec_event_esn_overflow(struct rte_security_ctx *ctx, uint64_t md)
2357 {
2358         struct ipsec_sa *sa;
2359
2360         /* For inline protocol processing, the metadata in the event will
2361          * uniquely identify the security session which raised the event.
2362          * Application would then need the userdata it had registered with the
2363          * security session to process the event.
2364          */
2365
2366         sa = (struct ipsec_sa *)rte_security_get_userdata(ctx, md);
2367
2368         if (sa == NULL) {
2369                 /* userdata could not be retrieved */
2370                 return -1;
2371         }
2372
2373         /* Sequence number over flow. SA need to be re-established */
2374         RTE_SET_USED(sa);
2375         return 0;
2376 }
2377
2378 static int
2379 inline_ipsec_event_callback(uint16_t port_id, enum rte_eth_event_type type,
2380                  void *param, void *ret_param)
2381 {
2382         uint64_t md;
2383         struct rte_eth_event_ipsec_desc *event_desc = NULL;
2384         struct rte_security_ctx *ctx = (struct rte_security_ctx *)
2385                                         rte_eth_dev_get_sec_ctx(port_id);
2386
2387         RTE_SET_USED(param);
2388
2389         if (type != RTE_ETH_EVENT_IPSEC)
2390                 return -1;
2391
2392         event_desc = ret_param;
2393         if (event_desc == NULL) {
2394                 printf("Event descriptor not set\n");
2395                 return -1;
2396         }
2397
2398         md = event_desc->metadata;
2399
2400         if (event_desc->subtype == RTE_ETH_EVENT_IPSEC_ESN_OVERFLOW)
2401                 return inline_ipsec_event_esn_overflow(ctx, md);
2402         else if (event_desc->subtype >= RTE_ETH_EVENT_IPSEC_MAX) {
2403                 printf("Invalid IPsec event reported\n");
2404                 return -1;
2405         }
2406
2407         return -1;
2408 }
2409
2410 static uint16_t
2411 rx_callback(__rte_unused uint16_t port, __rte_unused uint16_t queue,
2412         struct rte_mbuf *pkt[], uint16_t nb_pkts,
2413         __rte_unused uint16_t max_pkts, void *user_param)
2414 {
2415         uint64_t tm;
2416         uint32_t i, k;
2417         struct lcore_conf *lc;
2418         struct rte_mbuf *mb;
2419         struct rte_ether_hdr *eth;
2420
2421         lc = user_param;
2422         k = 0;
2423         tm = 0;
2424
2425         for (i = 0; i != nb_pkts; i++) {
2426
2427                 mb = pkt[i];
2428                 eth = rte_pktmbuf_mtod(mb, struct rte_ether_hdr *);
2429                 if (eth->ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4)) {
2430
2431                         struct rte_ipv4_hdr *iph;
2432
2433                         iph = (struct rte_ipv4_hdr *)(eth + 1);
2434                         if (rte_ipv4_frag_pkt_is_fragmented(iph)) {
2435
2436                                 mb->l2_len = sizeof(*eth);
2437                                 mb->l3_len = sizeof(*iph);
2438                                 tm = (tm != 0) ? tm : rte_rdtsc();
2439                                 mb = rte_ipv4_frag_reassemble_packet(
2440                                         lc->frag.tbl, &lc->frag.dr,
2441                                         mb, tm, iph);
2442
2443                                 if (mb != NULL) {
2444                                         /* fix ip cksum after reassemble. */
2445                                         iph = rte_pktmbuf_mtod_offset(mb,
2446                                                 struct rte_ipv4_hdr *,
2447                                                 mb->l2_len);
2448                                         iph->hdr_checksum = 0;
2449                                         iph->hdr_checksum = rte_ipv4_cksum(iph);
2450                                 }
2451                         }
2452                 } else if (eth->ether_type ==
2453                                 rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6)) {
2454
2455                         struct rte_ipv6_hdr *iph;
2456                         struct ipv6_extension_fragment *fh;
2457
2458                         iph = (struct rte_ipv6_hdr *)(eth + 1);
2459                         fh = rte_ipv6_frag_get_ipv6_fragment_header(iph);
2460                         if (fh != NULL) {
2461                                 mb->l2_len = sizeof(*eth);
2462                                 mb->l3_len = (uintptr_t)fh - (uintptr_t)iph +
2463                                         sizeof(*fh);
2464                                 tm = (tm != 0) ? tm : rte_rdtsc();
2465                                 mb = rte_ipv6_frag_reassemble_packet(
2466                                         lc->frag.tbl, &lc->frag.dr,
2467                                         mb, tm, iph, fh);
2468                                 if (mb != NULL)
2469                                         /* fix l3_len after reassemble. */
2470                                         mb->l3_len = mb->l3_len - sizeof(*fh);
2471                         }
2472                 }
2473
2474                 pkt[k] = mb;
2475                 k += (mb != NULL);
2476         }
2477
2478         /* some fragments were encountered, drain death row */
2479         if (tm != 0)
2480                 rte_ip_frag_free_death_row(&lc->frag.dr, 0);
2481
2482         return k;
2483 }
2484
2485
2486 static int
2487 reassemble_lcore_init(struct lcore_conf *lc, uint32_t cid)
2488 {
2489         int32_t sid;
2490         uint32_t i;
2491         uint64_t frag_cycles;
2492         const struct lcore_rx_queue *rxq;
2493         const struct rte_eth_rxtx_callback *cb;
2494
2495         /* create fragment table */
2496         sid = rte_lcore_to_socket_id(cid);
2497         frag_cycles = (rte_get_tsc_hz() + NS_PER_S - 1) /
2498                 NS_PER_S * frag_ttl_ns;
2499
2500         lc->frag.tbl = rte_ip_frag_table_create(frag_tbl_sz,
2501                 FRAG_TBL_BUCKET_ENTRIES, frag_tbl_sz, frag_cycles, sid);
2502         if (lc->frag.tbl == NULL) {
2503                 printf("%s(%u): failed to create fragment table of size: %u, "
2504                         "error code: %d\n",
2505                         __func__, cid, frag_tbl_sz, rte_errno);
2506                 return -ENOMEM;
2507         }
2508
2509         /* setup reassemble RX callbacks for all queues */
2510         for (i = 0; i != lc->nb_rx_queue; i++) {
2511
2512                 rxq = lc->rx_queue_list + i;
2513                 cb = rte_eth_add_rx_callback(rxq->port_id, rxq->queue_id,
2514                         rx_callback, lc);
2515                 if (cb == NULL) {
2516                         printf("%s(%u): failed to install RX callback for "
2517                                 "portid=%u, queueid=%u, error code: %d\n",
2518                                 __func__, cid,
2519                                 rxq->port_id, rxq->queue_id, rte_errno);
2520                         return -ENOMEM;
2521                 }
2522         }
2523
2524         return 0;
2525 }
2526
2527 static int
2528 reassemble_init(void)
2529 {
2530         int32_t rc;
2531         uint32_t i, lc;
2532
2533         rc = 0;
2534         for (i = 0; i != nb_lcore_params; i++) {
2535                 lc = lcore_params[i].lcore_id;
2536                 rc = reassemble_lcore_init(lcore_conf + lc, lc);
2537                 if (rc != 0)
2538                         break;
2539         }
2540
2541         return rc;
2542 }
2543
2544 static void
2545 create_default_ipsec_flow(uint16_t port_id, uint64_t rx_offloads)
2546 {
2547         struct rte_flow_action action[2];
2548         struct rte_flow_item pattern[2];
2549         struct rte_flow_attr attr = {0};
2550         struct rte_flow_error err;
2551         struct rte_flow *flow;
2552         int ret;
2553
2554         if (!(rx_offloads & DEV_RX_OFFLOAD_SECURITY))
2555                 return;
2556
2557         /* Add the default rte_flow to enable SECURITY for all ESP packets */
2558
2559         pattern[0].type = RTE_FLOW_ITEM_TYPE_ESP;
2560         pattern[0].spec = NULL;
2561         pattern[0].mask = NULL;
2562         pattern[0].last = NULL;
2563         pattern[1].type = RTE_FLOW_ITEM_TYPE_END;
2564
2565         action[0].type = RTE_FLOW_ACTION_TYPE_SECURITY;
2566         action[0].conf = NULL;
2567         action[1].type = RTE_FLOW_ACTION_TYPE_END;
2568         action[1].conf = NULL;
2569
2570         attr.ingress = 1;
2571
2572         ret = rte_flow_validate(port_id, &attr, pattern, action, &err);
2573         if (ret)
2574                 return;
2575
2576         flow = rte_flow_create(port_id, &attr, pattern, action, &err);
2577         if (flow == NULL)
2578                 return;
2579
2580         flow_info_tbl[port_id].rx_def_flow = flow;
2581         RTE_LOG(INFO, IPSEC,
2582                 "Created default flow enabling SECURITY for all ESP traffic on port %d\n",
2583                 port_id);
2584 }
2585
2586 static void
2587 signal_handler(int signum)
2588 {
2589         if (signum == SIGINT || signum == SIGTERM) {
2590                 printf("\n\nSignal %d received, preparing to exit...\n",
2591                                 signum);
2592                 force_quit = true;
2593         }
2594 }
2595
2596 static void
2597 ev_mode_sess_verify(struct ipsec_sa *sa, int nb_sa)
2598 {
2599         struct rte_ipsec_session *ips;
2600         int32_t i;
2601
2602         if (!sa || !nb_sa)
2603                 return;
2604
2605         for (i = 0; i < nb_sa; i++) {
2606                 ips = ipsec_get_primary_session(&sa[i]);
2607                 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL)
2608                         rte_exit(EXIT_FAILURE, "Event mode supports only "
2609                                  "inline protocol sessions\n");
2610         }
2611
2612 }
2613
2614 static int32_t
2615 check_event_mode_params(struct eh_conf *eh_conf)
2616 {
2617         struct eventmode_conf *em_conf = NULL;
2618         struct lcore_params *params;
2619         uint16_t portid;
2620
2621         if (!eh_conf || !eh_conf->mode_params)
2622                 return -EINVAL;
2623
2624         /* Get eventmode conf */
2625         em_conf = eh_conf->mode_params;
2626
2627         if (eh_conf->mode == EH_PKT_TRANSFER_MODE_POLL &&
2628             em_conf->ext_params.sched_type != SCHED_TYPE_NOT_SET) {
2629                 printf("error: option --event-schedule-type applies only to "
2630                        "event mode\n");
2631                 return -EINVAL;
2632         }
2633
2634         if (eh_conf->mode != EH_PKT_TRANSFER_MODE_EVENT)
2635                 return 0;
2636
2637         /* Set schedule type to ORDERED if it wasn't explicitly set by user */
2638         if (em_conf->ext_params.sched_type == SCHED_TYPE_NOT_SET)
2639                 em_conf->ext_params.sched_type = RTE_SCHED_TYPE_ORDERED;
2640
2641         /*
2642          * Event mode currently supports only inline protocol sessions.
2643          * If there are other types of sessions configured then exit with
2644          * error.
2645          */
2646         ev_mode_sess_verify(sa_in, nb_sa_in);
2647         ev_mode_sess_verify(sa_out, nb_sa_out);
2648
2649
2650         /* Option --config does not apply to event mode */
2651         if (nb_lcore_params > 0) {
2652                 printf("error: option --config applies only to poll mode\n");
2653                 return -EINVAL;
2654         }
2655
2656         /*
2657          * In order to use the same port_init routine for both poll and event
2658          * modes initialize lcore_params with one queue for each eth port
2659          */
2660         lcore_params = lcore_params_array;
2661         RTE_ETH_FOREACH_DEV(portid) {
2662                 if ((enabled_port_mask & (1 << portid)) == 0)
2663                         continue;
2664
2665                 params = &lcore_params[nb_lcore_params++];
2666                 params->port_id = portid;
2667                 params->queue_id = 0;
2668                 params->lcore_id = rte_get_next_lcore(0, 0, 1);
2669         }
2670
2671         return 0;
2672 }
2673
2674 static void
2675 inline_sessions_free(struct sa_ctx *sa_ctx)
2676 {
2677         struct rte_ipsec_session *ips;
2678         struct ipsec_sa *sa;
2679         int32_t ret;
2680         uint32_t i;
2681
2682         if (!sa_ctx)
2683                 return;
2684
2685         for (i = 0; i < sa_ctx->nb_sa; i++) {
2686
2687                 sa = &sa_ctx->sa[i];
2688                 if (!sa->spi)
2689                         continue;
2690
2691                 ips = ipsec_get_primary_session(sa);
2692                 if (ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_PROTOCOL &&
2693                     ips->type != RTE_SECURITY_ACTION_TYPE_INLINE_CRYPTO)
2694                         continue;
2695
2696                 if (!rte_eth_dev_is_valid_port(sa->portid))
2697                         continue;
2698
2699                 ret = rte_security_session_destroy(
2700                                 rte_eth_dev_get_sec_ctx(sa->portid),
2701                                 ips->security.ses);
2702                 if (ret)
2703                         RTE_LOG(ERR, IPSEC, "Failed to destroy security "
2704                                             "session type %d, spi %d\n",
2705                                             ips->type, sa->spi);
2706         }
2707 }
2708
2709 static uint32_t
2710 calculate_nb_mbufs(uint16_t nb_ports, uint16_t nb_crypto_qp, uint32_t nb_rxq,
2711                 uint32_t nb_txq)
2712 {
2713         return RTE_MAX((nb_rxq * nb_rxd +
2714                         nb_ports * nb_lcores * MAX_PKT_BURST +
2715                         nb_ports * nb_txq * nb_txd +
2716                         nb_lcores * MEMPOOL_CACHE_SIZE +
2717                         nb_crypto_qp * CDEV_QUEUE_DESC +
2718                         nb_lcores * frag_tbl_sz *
2719                         FRAG_TBL_BUCKET_ENTRIES),
2720                        8192U);
2721 }
2722
2723 int32_t
2724 main(int32_t argc, char **argv)
2725 {
2726         int32_t ret;
2727         uint32_t lcore_id, nb_txq, nb_rxq = 0;
2728         uint32_t cdev_id;
2729         uint32_t i;
2730         uint8_t socket_id;
2731         uint16_t portid, nb_crypto_qp, nb_ports = 0;
2732         uint64_t req_rx_offloads[RTE_MAX_ETHPORTS];
2733         uint64_t req_tx_offloads[RTE_MAX_ETHPORTS];
2734         struct eh_conf *eh_conf = NULL;
2735         size_t sess_sz;
2736
2737         nb_bufs_in_pool = 0;
2738
2739         /* init EAL */
2740         ret = rte_eal_init(argc, argv);
2741         if (ret < 0)
2742                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
2743         argc -= ret;
2744         argv += ret;
2745
2746         force_quit = false;
2747         signal(SIGINT, signal_handler);
2748         signal(SIGTERM, signal_handler);
2749
2750         /* initialize event helper configuration */
2751         eh_conf = eh_conf_init();
2752         if (eh_conf == NULL)
2753                 rte_exit(EXIT_FAILURE, "Failed to init event helper config");
2754
2755         /* parse application arguments (after the EAL ones) */
2756         ret = parse_args(argc, argv, eh_conf);
2757         if (ret < 0)
2758                 rte_exit(EXIT_FAILURE, "Invalid parameters\n");
2759
2760         /* parse configuration file */
2761         if (parse_cfg_file(cfgfile) < 0) {
2762                 printf("parsing file \"%s\" failed\n",
2763                         optarg);
2764                 print_usage(argv[0]);
2765                 return -1;
2766         }
2767
2768         if ((unprotected_port_mask & enabled_port_mask) !=
2769                         unprotected_port_mask)
2770                 rte_exit(EXIT_FAILURE, "Invalid unprotected portmask 0x%x\n",
2771                                 unprotected_port_mask);
2772
2773         if (check_poll_mode_params(eh_conf) < 0)
2774                 rte_exit(EXIT_FAILURE, "check_poll_mode_params failed\n");
2775
2776         if (check_event_mode_params(eh_conf) < 0)
2777                 rte_exit(EXIT_FAILURE, "check_event_mode_params failed\n");
2778
2779         ret = init_lcore_rx_queues();
2780         if (ret < 0)
2781                 rte_exit(EXIT_FAILURE, "init_lcore_rx_queues failed\n");
2782
2783         nb_lcores = rte_lcore_count();
2784
2785         sess_sz = max_session_size();
2786
2787         /*
2788          * In event mode request minimum number of crypto queues
2789          * to be reserved equal to number of ports.
2790          */
2791         if (eh_conf->mode == EH_PKT_TRANSFER_MODE_EVENT)
2792                 nb_crypto_qp = rte_eth_dev_count_avail();
2793         else
2794                 nb_crypto_qp = 0;
2795
2796         nb_crypto_qp = cryptodevs_init(nb_crypto_qp);
2797
2798         if (nb_bufs_in_pool == 0) {
2799                 RTE_ETH_FOREACH_DEV(portid) {
2800                         if ((enabled_port_mask & (1 << portid)) == 0)
2801                                 continue;
2802                         nb_ports++;
2803                         nb_rxq += get_port_nb_rx_queues(portid);
2804                 }
2805
2806                 nb_txq = nb_lcores;
2807
2808                 nb_bufs_in_pool = calculate_nb_mbufs(nb_ports, nb_crypto_qp,
2809                                                 nb_rxq, nb_txq);
2810         }
2811
2812         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
2813                 if (rte_lcore_is_enabled(lcore_id) == 0)
2814                         continue;
2815
2816                 if (numa_on)
2817                         socket_id = (uint8_t)rte_lcore_to_socket_id(lcore_id);
2818                 else
2819                         socket_id = 0;
2820
2821                 /* mbuf_pool is initialised by the pool_init() function*/
2822                 if (socket_ctx[socket_id].mbuf_pool)
2823                         continue;
2824
2825                 pool_init(&socket_ctx[socket_id], socket_id, nb_bufs_in_pool);
2826                 session_pool_init(&socket_ctx[socket_id], socket_id, sess_sz);
2827                 session_priv_pool_init(&socket_ctx[socket_id], socket_id,
2828                         sess_sz);
2829         }
2830         printf("Number of mbufs in packet pool %d\n", nb_bufs_in_pool);
2831
2832         RTE_ETH_FOREACH_DEV(portid) {
2833                 if ((enabled_port_mask & (1 << portid)) == 0)
2834                         continue;
2835
2836                 sa_check_offloads(portid, &req_rx_offloads[portid],
2837                                 &req_tx_offloads[portid]);
2838                 port_init(portid, req_rx_offloads[portid],
2839                                 req_tx_offloads[portid]);
2840         }
2841
2842         /*
2843          * Set the enabled port mask in helper config for use by helper
2844          * sub-system. This will be used while initializing devices using
2845          * helper sub-system.
2846          */
2847         eh_conf->eth_portmask = enabled_port_mask;
2848
2849         /* Initialize eventmode components */
2850         ret = eh_devs_init(eh_conf);
2851         if (ret < 0)
2852                 rte_exit(EXIT_FAILURE, "eh_devs_init failed, err=%d\n", ret);
2853
2854         /* start ports */
2855         RTE_ETH_FOREACH_DEV(portid) {
2856                 if ((enabled_port_mask & (1 << portid)) == 0)
2857                         continue;
2858
2859                 /* Create flow before starting the device */
2860                 create_default_ipsec_flow(portid, req_rx_offloads[portid]);
2861
2862                 ret = rte_eth_dev_start(portid);
2863                 if (ret < 0)
2864                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: "
2865                                         "err=%d, port=%d\n", ret, portid);
2866                 /*
2867                  * If enabled, put device in promiscuous mode.
2868                  * This allows IO forwarding mode to forward packets
2869                  * to itself through 2 cross-connected  ports of the
2870                  * target machine.
2871                  */
2872                 if (promiscuous_on) {
2873                         ret = rte_eth_promiscuous_enable(portid);
2874                         if (ret != 0)
2875                                 rte_exit(EXIT_FAILURE,
2876                                         "rte_eth_promiscuous_enable: err=%s, port=%d\n",
2877                                         rte_strerror(-ret), portid);
2878                 }
2879
2880                 rte_eth_dev_callback_register(portid,
2881                         RTE_ETH_EVENT_IPSEC, inline_ipsec_event_callback, NULL);
2882         }
2883
2884         /* fragment reassemble is enabled */
2885         if (frag_tbl_sz != 0) {
2886                 ret = reassemble_init();
2887                 if (ret != 0)
2888                         rte_exit(EXIT_FAILURE, "failed at reassemble init");
2889         }
2890
2891         /* Replicate each context per socket */
2892         for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2893                 socket_id = rte_socket_id_by_idx(i);
2894                 if ((socket_ctx[socket_id].mbuf_pool != NULL) &&
2895                         (socket_ctx[socket_id].sa_in == NULL) &&
2896                         (socket_ctx[socket_id].sa_out == NULL)) {
2897                         sa_init(&socket_ctx[socket_id], socket_id);
2898                         sp4_init(&socket_ctx[socket_id], socket_id);
2899                         sp6_init(&socket_ctx[socket_id], socket_id);
2900                         rt_init(&socket_ctx[socket_id], socket_id);
2901                 }
2902         }
2903
2904         check_all_ports_link_status(enabled_port_mask);
2905
2906         /* launch per-lcore init on every lcore */
2907         rte_eal_mp_remote_launch(ipsec_launch_one_lcore, eh_conf, CALL_MASTER);
2908         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
2909                 if (rte_eal_wait_lcore(lcore_id) < 0)
2910                         return -1;
2911         }
2912
2913         /* Uninitialize eventmode components */
2914         ret = eh_devs_uninit(eh_conf);
2915         if (ret < 0)
2916                 rte_exit(EXIT_FAILURE, "eh_devs_uninit failed, err=%d\n", ret);
2917
2918         /* Free eventmode configuration memory */
2919         eh_conf_uninit(eh_conf);
2920
2921         /* Destroy inline inbound and outbound sessions */
2922         for (i = 0; i < NB_SOCKETS && i < rte_socket_count(); i++) {
2923                 socket_id = rte_socket_id_by_idx(i);
2924                 inline_sessions_free(socket_ctx[socket_id].sa_in);
2925                 inline_sessions_free(socket_ctx[socket_id].sa_out);
2926         }
2927
2928         for (cdev_id = 0; cdev_id < rte_cryptodev_count(); cdev_id++) {
2929                 printf("Closing cryptodev %d...", cdev_id);
2930                 rte_cryptodev_stop(cdev_id);
2931                 rte_cryptodev_close(cdev_id);
2932                 printf(" Done\n");
2933         }
2934
2935         RTE_ETH_FOREACH_DEV(portid) {
2936                 if ((enabled_port_mask & (1 << portid)) == 0)
2937                         continue;
2938
2939                 printf("Closing port %d...", portid);
2940                 if (flow_info_tbl[portid].rx_def_flow) {
2941                         struct rte_flow_error err;
2942
2943                         ret = rte_flow_destroy(portid,
2944                                 flow_info_tbl[portid].rx_def_flow, &err);
2945                         if (ret)
2946                                 RTE_LOG(ERR, IPSEC, "Failed to destroy flow "
2947                                         " for port %u, err msg: %s\n", portid,
2948                                         err.message);
2949                 }
2950                 rte_eth_dev_stop(portid);
2951                 rte_eth_dev_close(portid);
2952                 printf(" Done\n");
2953         }
2954         printf("Bye...\n");
2955
2956         return 0;
2957 }