examples/ip_fragmentation/main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <sys/param.h>
  11 #include <string.h>
  12 #include <sys/queue.h>
  13 #include <stdarg.h>
  14 #include <errno.h>
  15 #include <getopt.h>
  16
  17 #include <rte_common.h>
  18 #include <rte_byteorder.h>
  19 #include <rte_log.h>
  20 #include <rte_memory.h>
  21 #include <rte_memcpy.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_mempool.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_lpm.h>
  38 #include <rte_lpm6.h>
  39 #include <rte_ip.h>
  40 #include <rte_string_fns.h>
  41
  42 #include <rte_ip_frag.h>
  43
  44 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
  45
  46 /* allow max jumbo frame 9.5 KB */
  47 #define JUMBO_FRAME_MAX_SIZE    0x2600
  48
  49 #define ROUNDUP_DIV(a, b)       (((a) + (b) - 1) / (b))
  50
  51 /*
  52  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
  53  * This value includes the size of IPv6 header.
  54  */
  55 #define IPV4_MTU_DEFAULT        ETHER_MTU
  56 #define IPV6_MTU_DEFAULT        ETHER_MTU
  57
  58 /*
  59  * The overhead from max frame size to MTU.
  60  * We have to consider the max possible overhead.
  61  */
  62 #define MTU_OVERHEAD    \
  63         (ETHER_HDR_LEN + ETHER_CRC_LEN + 2 * sizeof(struct rte_vlan_hdr))
  64
  65 /*
  66  * Default payload in bytes for the IPv6 packet.
  67  */
  68 #define IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
  69 #define IPV6_DEFAULT_PAYLOAD    (IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
  70
  71 /*
  72  * Max number of fragments per packet expected - defined by config file.
  73  */
  74 #define MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
  75
  76 #define NB_MBUF   8192
  77
  78 #define MAX_PKT_BURST   32
  79 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  80
  81 /* Configure how many packets ahead to prefetch, when reading packets */
  82 #define PREFETCH_OFFSET 3
  83
  84 /*
  85  * Configurable number of RX/TX ring descriptors
  86  */
  87 #define RTE_TEST_RX_DESC_DEFAULT 1024
  88 #define RTE_TEST_TX_DESC_DEFAULT 1024
  89 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
  90 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
  91
  92 /* ethernet addresses of ports */
  93 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
  94
  95 #ifndef IPv4_BYTES
  96 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
  97 #define IPv4_BYTES(addr) \
  98                 (uint8_t) (((addr) >> 24) & 0xFF),\
  99                 (uint8_t) (((addr) >> 16) & 0xFF),\
 100                 (uint8_t) (((addr) >> 8) & 0xFF),\
 101                 (uint8_t) ((addr) & 0xFF)
 102 #endif
 103
 104 #ifndef IPv6_BYTES
 105 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 106                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 107 #define IPv6_BYTES(addr) \
 108         addr[0],  addr[1], addr[2],  addr[3], \
 109         addr[4],  addr[5], addr[6],  addr[7], \
 110         addr[8],  addr[9], addr[10], addr[11],\
 111         addr[12], addr[13],addr[14], addr[15]
 112 #endif
 113
 114 #define IPV6_ADDR_LEN 16
 115
 116 /* mask of enabled ports */
 117 static int enabled_port_mask = 0;
 118
 119 static int rx_queue_per_lcore = 1;
 120
 121 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
 122
 123 struct mbuf_table {
 124         uint16_t len;
 125         struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
 126 };
 127
 128 struct rx_queue {
 129         struct rte_mempool *direct_pool;
 130         struct rte_mempool *indirect_pool;
 131         struct rte_lpm *lpm;
 132         struct rte_lpm6 *lpm6;
 133         uint16_t portid;
 134 };
 135
 136 #define MAX_RX_QUEUE_PER_LCORE 16
 137 #define MAX_TX_QUEUE_PER_PORT 16
 138 struct lcore_queue_conf {
 139         uint16_t n_rx_queue;
 140         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 141         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 142         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 143 } __rte_cache_aligned;
 144 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 145
 146 static struct rte_eth_conf port_conf = {
 147         .rxmode = {
 148                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 149                 .split_hdr_size = 0,
 150                 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
 151                              DEV_RX_OFFLOAD_SCATTER |
 152                              DEV_RX_OFFLOAD_JUMBO_FRAME),
 153         },
 154         .txmode = {
 155                 .mq_mode = ETH_MQ_TX_NONE,
 156                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
 157                              DEV_TX_OFFLOAD_MULTI_SEGS),
 158         },
 159 };
 160
 161 /*
 162  * IPv4 forwarding table
 163  */
 164 struct l3fwd_ipv4_route {
 165         uint32_t ip;
 166         uint8_t  depth;
 167         uint8_t  if_out;
 168 };
 169
 170 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 171                 {IPv4(100,10,0,0), 16, 0},
 172                 {IPv4(100,20,0,0), 16, 1},
 173                 {IPv4(100,30,0,0), 16, 2},
 174                 {IPv4(100,40,0,0), 16, 3},
 175                 {IPv4(100,50,0,0), 16, 4},
 176                 {IPv4(100,60,0,0), 16, 5},
 177                 {IPv4(100,70,0,0), 16, 6},
 178                 {IPv4(100,80,0,0), 16, 7},
 179 };
 180
 181 /*
 182  * IPv6 forwarding table
 183  */
 184
 185 struct l3fwd_ipv6_route {
 186         uint8_t ip[IPV6_ADDR_LEN];
 187         uint8_t depth;
 188         uint8_t if_out;
 189 };
 190
 191 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 192         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 193         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 194         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 195         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 196         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 197         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 198         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 199         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 200 };
 201
 202 #define LPM_MAX_RULES         1024
 203 #define LPM6_MAX_RULES         1024
 204 #define LPM6_NUMBER_TBL8S (1 << 16)
 205
 206 struct rte_lpm6_config lpm6_config = {
 207                 .max_rules = LPM6_MAX_RULES,
 208                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 209                 .flags = 0
 210 };
 211
 212 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
 213 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
 214 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 215 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 216
 217 /* Send burst of packets on an output interface */
 218 static inline int
 219 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
 220 {
 221         struct rte_mbuf **m_table;
 222         int ret;
 223         uint16_t queueid;
 224
 225         queueid = qconf->tx_queue_id[port];
 226         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 227
 228         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 229         if (unlikely(ret < n)) {
 230                 do {
 231                         rte_pktmbuf_free(m_table[ret]);
 232                 } while (++ret < n);
 233         }
 234
 235         return 0;
 236 }
 237
 238 static inline void
 239 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
 240                 uint8_t queueid, uint16_t port_in)
 241 {
 242         struct rx_queue *rxq;
 243         uint32_t i, len, next_hop;
 244         uint8_t ipv6;
 245         uint16_t port_out;
 246         int32_t len2;
 247
 248         ipv6 = 0;
 249         rxq = &qconf->rx_queue_list[queueid];
 250
 251         /* by default, send everything back to the source port */
 252         port_out = port_in;
 253
 254         /* Remove the Ethernet header and trailer from the input packet */
 255         rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
 256
 257         /* Build transmission burst */
 258         len = qconf->tx_mbufs[port_out].len;
 259
 260         /* if this is an IPv4 packet */
 261         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 262                 struct ipv4_hdr *ip_hdr;
 263                 uint32_t ip_dst;
 264                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 265                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
 266                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 267
 268                 /* Find destination port */
 269                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 270                                 (enabled_port_mask & 1 << next_hop) != 0) {
 271                         port_out = next_hop;
 272
 273                         /* Build transmission burst for new port */
 274                         len = qconf->tx_mbufs[port_out].len;
 275                 }
 276
 277                 /* if we don't need to do any fragmentation */
 278                 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
 279                         qconf->tx_mbufs[port_out].m_table[len] = m;
 280                         len2 = 1;
 281                 } else {
 282                         len2 = rte_ipv4_fragment_packet(m,
 283                                 &qconf->tx_mbufs[port_out].m_table[len],
 284                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 285                                 IPV4_MTU_DEFAULT,
 286                                 rxq->direct_pool, rxq->indirect_pool);
 287
 288                         /* Free input packet */
 289                         rte_pktmbuf_free(m);
 290
 291                         /* If we fail to fragment the packet */
 292                         if (unlikely (len2 < 0))
 293                                 return;
 294                 }
 295         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 296                 /* if this is an IPv6 packet */
 297                 struct ipv6_hdr *ip_hdr;
 298
 299                 ipv6 = 1;
 300
 301                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 302                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
 303
 304                 /* Find destination port */
 305                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
 306                                                 &next_hop) == 0 &&
 307                                 (enabled_port_mask & 1 << next_hop) != 0) {
 308                         port_out = next_hop;
 309
 310                         /* Build transmission burst for new port */
 311                         len = qconf->tx_mbufs[port_out].len;
 312                 }
 313
 314                 /* if we don't need to do any fragmentation */
 315                 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
 316                         qconf->tx_mbufs[port_out].m_table[len] = m;
 317                         len2 = 1;
 318                 } else {
 319                         len2 = rte_ipv6_fragment_packet(m,
 320                                 &qconf->tx_mbufs[port_out].m_table[len],
 321                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 322                                 IPV6_MTU_DEFAULT,
 323                                 rxq->direct_pool, rxq->indirect_pool);
 324
 325                         /* Free input packet */
 326                         rte_pktmbuf_free(m);
 327
 328                         /* If we fail to fragment the packet */
 329                         if (unlikely (len2 < 0))
 330                                 return;
 331                 }
 332         }
 333         /* else, just forward the packet */
 334         else {
 335                 qconf->tx_mbufs[port_out].m_table[len] = m;
 336                 len2 = 1;
 337         }
 338
 339         for (i = len; i < len + len2; i ++) {
 340                 void *d_addr_bytes;
 341
 342                 m = qconf->tx_mbufs[port_out].m_table[i];
 343                 struct rte_ether_hdr *eth_hdr = (struct rte_ether_hdr *)
 344                         rte_pktmbuf_prepend(m,
 345                                 (uint16_t)sizeof(struct rte_ether_hdr));
 346                 if (eth_hdr == NULL) {
 347                         rte_panic("No headroom in mbuf.\n");
 348                 }
 349
 350                 m->l2_len = sizeof(struct rte_ether_hdr);
 351
 352                 /* 02:00:00:00:00:xx */
 353                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 354                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
 355
 356                 /* src addr */
 357                 ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
 358                 if (ipv6)
 359                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 360                 else
 361                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 362         }
 363
 364         len += len2;
 365
 366         if (likely(len < MAX_PKT_BURST)) {
 367                 qconf->tx_mbufs[port_out].len = (uint16_t)len;
 368                 return;
 369         }
 370
 371         /* Transmit packets */
 372         send_burst(qconf, (uint16_t)len, port_out);
 373         qconf->tx_mbufs[port_out].len = 0;
 374 }
 375
 376 /* main processing loop */
 377 static int
 378 main_loop(__attribute__((unused)) void *dummy)
 379 {
 380         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 381         unsigned lcore_id;
 382         uint64_t prev_tsc, diff_tsc, cur_tsc;
 383         int i, j, nb_rx;
 384         uint16_t portid;
 385         struct lcore_queue_conf *qconf;
 386         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 387
 388         prev_tsc = 0;
 389
 390         lcore_id = rte_lcore_id();
 391         qconf = &lcore_queue_conf[lcore_id];
 392
 393         if (qconf->n_rx_queue == 0) {
 394                 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
 395                 return 0;
 396         }
 397
 398         RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
 399
 400         for (i = 0; i < qconf->n_rx_queue; i++) {
 401
 402                 portid = qconf->rx_queue_list[i].portid;
 403                 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
 404                                 portid);
 405         }
 406
 407         while (1) {
 408
 409                 cur_tsc = rte_rdtsc();
 410
 411                 /*
 412                  * TX burst queue drain
 413                  */
 414                 diff_tsc = cur_tsc - prev_tsc;
 415                 if (unlikely(diff_tsc > drain_tsc)) {
 416
 417                         /*
 418                          * This could be optimized (use queueid instead of
 419                          * portid), but it is not called so often
 420                          */
 421                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 422                                 if (qconf->tx_mbufs[portid].len == 0)
 423                                         continue;
 424                                 send_burst(&lcore_queue_conf[lcore_id],
 425                                            qconf->tx_mbufs[portid].len,
 426                                            portid);
 427                                 qconf->tx_mbufs[portid].len = 0;
 428                         }
 429
 430                         prev_tsc = cur_tsc;
 431                 }
 432
 433                 /*
 434                  * Read packet from RX queues
 435                  */
 436                 for (i = 0; i < qconf->n_rx_queue; i++) {
 437
 438                         portid = qconf->rx_queue_list[i].portid;
 439                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 440                                                  MAX_PKT_BURST);
 441
 442                         /* Prefetch first packets */
 443                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 444                                 rte_prefetch0(rte_pktmbuf_mtod(
 445                                                 pkts_burst[j], void *));
 446                         }
 447
 448                         /* Prefetch and forward already prefetched packets */
 449                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 450                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 451                                                 j + PREFETCH_OFFSET], void *));
 452                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 453                         }
 454
 455                         /* Forward remaining prefetched packets */
 456                         for (; j < nb_rx; j++) {
 457                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 458                         }
 459                 }
 460         }
 461 }
 462
 463 /* display usage */
 464 static void
 465 print_usage(const char *prgname)
 466 {
 467         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
 468                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 469                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
 470                prgname);
 471 }
 472
 473 static int
 474 parse_portmask(const char *portmask)
 475 {
 476         char *end = NULL;
 477         unsigned long pm;
 478
 479         /* parse hexadecimal string */
 480         pm = strtoul(portmask, &end, 16);
 481         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 482                 return -1;
 483
 484         if (pm == 0)
 485                 return -1;
 486
 487         return pm;
 488 }
 489
 490 static int
 491 parse_nqueue(const char *q_arg)
 492 {
 493         char *end = NULL;
 494         unsigned long n;
 495
 496         /* parse hexadecimal string */
 497         n = strtoul(q_arg, &end, 10);
 498         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 499                 return -1;
 500         if (n == 0)
 501                 return -1;
 502         if (n >= MAX_RX_QUEUE_PER_LCORE)
 503                 return -1;
 504
 505         return n;
 506 }
 507
 508 /* Parse the argument given in the command line of the application */
 509 static int
 510 parse_args(int argc, char **argv)
 511 {
 512         int opt, ret;
 513         char **argvopt;
 514         int option_index;
 515         char *prgname = argv[0];
 516         static struct option lgopts[] = {
 517                 {NULL, 0, 0, 0}
 518         };
 519
 520         argvopt = argv;
 521
 522         while ((opt = getopt_long(argc, argvopt, "p:q:",
 523                                   lgopts, &option_index)) != EOF) {
 524
 525                 switch (opt) {
 526                 /* portmask */
 527                 case 'p':
 528                         enabled_port_mask = parse_portmask(optarg);
 529                         if (enabled_port_mask < 0) {
 530                                 printf("invalid portmask\n");
 531                                 print_usage(prgname);
 532                                 return -1;
 533                         }
 534                         break;
 535
 536                 /* nqueue */
 537                 case 'q':
 538                         rx_queue_per_lcore = parse_nqueue(optarg);
 539                         if (rx_queue_per_lcore < 0) {
 540                                 printf("invalid queue number\n");
 541                                 print_usage(prgname);
 542                                 return -1;
 543                         }
 544                         break;
 545
 546                 /* long options */
 547                 case 0:
 548                         print_usage(prgname);
 549                         return -1;
 550
 551                 default:
 552                         print_usage(prgname);
 553                         return -1;
 554                 }
 555         }
 556
 557         if (enabled_port_mask == 0) {
 558                 printf("portmask not specified\n");
 559                 print_usage(prgname);
 560                 return -1;
 561         }
 562
 563         if (optind >= 0)
 564                 argv[optind-1] = prgname;
 565
 566         ret = optind-1;
 567         optind = 1; /* reset getopt lib */
 568         return ret;
 569 }
 570
 571 static void
 572 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr)
 573 {
 574         char buf[ETHER_ADDR_FMT_SIZE];
 575         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 576         printf("%s%s", name, buf);
 577 }
 578
 579 /* Check the link status of all ports in up to 9s, and print them finally */
 580 static void
 581 check_all_ports_link_status(uint32_t port_mask)
 582 {
 583 #define CHECK_INTERVAL 100 /* 100ms */
 584 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 585         uint16_t portid;
 586         uint8_t count, all_ports_up, print_flag = 0;
 587         struct rte_eth_link link;
 588
 589         printf("\nChecking link status");
 590         fflush(stdout);
 591         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 592                 all_ports_up = 1;
 593                 RTE_ETH_FOREACH_DEV(portid) {
 594                         if ((port_mask & (1 << portid)) == 0)
 595                                 continue;
 596                         memset(&link, 0, sizeof(link));
 597                         rte_eth_link_get_nowait(portid, &link);
 598                         /* print link status if flag set */
 599                         if (print_flag == 1) {
 600                                 if (link.link_status)
 601                                         printf(
 602                                         "Port%d Link Up .Speed %u Mbps - %s\n",
 603                                                 portid, link.link_speed,
 604                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 605                                         ("full-duplex") : ("half-duplex\n"));
 606                                 else
 607                                         printf("Port %d Link Down\n", portid);
 608                                 continue;
 609                         }
 610                         /* clear all_ports_up flag if any link down */
 611                         if (link.link_status == ETH_LINK_DOWN) {
 612                                 all_ports_up = 0;
 613                                 break;
 614                         }
 615                 }
 616                 /* after finally printing all link status, get out */
 617                 if (print_flag == 1)
 618                         break;
 619
 620                 if (all_ports_up == 0) {
 621                         printf(".");
 622                         fflush(stdout);
 623                         rte_delay_ms(CHECK_INTERVAL);
 624                 }
 625
 626                 /* set the print_flag if all ports up or timeout */
 627                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 628                         print_flag = 1;
 629                         printf("\ndone\n");
 630                 }
 631         }
 632 }
 633
 634 /* Check L3 packet type detection capablity of the NIC port */
 635 static int
 636 check_ptype(int portid)
 637 {
 638         int i, ret;
 639         int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
 640         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
 641
 642         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 643         if (ret <= 0)
 644                 return 0;
 645
 646         uint32_t ptypes[ret];
 647
 648         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 649         for (i = 0; i < ret; ++i) {
 650                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
 651                         ptype_l3_ipv4 = 1;
 652                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
 653                         ptype_l3_ipv6 = 1;
 654         }
 655
 656         if (ptype_l3_ipv4 == 0)
 657                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
 658
 659         if (ptype_l3_ipv6 == 0)
 660                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
 661
 662         if (ptype_l3_ipv4 && ptype_l3_ipv6)
 663                 return 1;
 664
 665         return 0;
 666
 667 }
 668
 669 /* Parse packet type of a packet by SW */
 670 static inline void
 671 parse_ptype(struct rte_mbuf *m)
 672 {
 673         struct rte_ether_hdr *eth_hdr;
 674         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 675         uint16_t ether_type;
 676
 677         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 678         ether_type = eth_hdr->ether_type;
 679         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 680                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 681         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 682                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 683
 684         m->packet_type = packet_type;
 685 }
 686
 687 /* callback function to detect packet type for a queue of a port */
 688 static uint16_t
 689 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 690                    struct rte_mbuf *pkts[], uint16_t nb_pkts,
 691                    uint16_t max_pkts __rte_unused,
 692                    void *user_param __rte_unused)
 693 {
 694         uint16_t i;
 695
 696         for (i = 0; i < nb_pkts; ++i)
 697                 parse_ptype(pkts[i]);
 698
 699         return nb_pkts;
 700 }
 701
 702 static int
 703 init_routing_table(void)
 704 {
 705         struct rte_lpm *lpm;
 706         struct rte_lpm6 *lpm6;
 707         int socket, ret;
 708         unsigned i;
 709
 710         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 711                 if (socket_lpm[socket]) {
 712                         lpm = socket_lpm[socket];
 713                         /* populate the LPM table */
 714                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 715                                 ret = rte_lpm_add(lpm,
 716                                         l3fwd_ipv4_route_array[i].ip,
 717                                         l3fwd_ipv4_route_array[i].depth,
 718                                         l3fwd_ipv4_route_array[i].if_out);
 719
 720                                 if (ret < 0) {
 721                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 722                                                 "LPM table\n", i);
 723                                         return -1;
 724                                 }
 725
 726                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 727                                                 "/%d (port %d)\n",
 728                                         socket,
 729                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 730                                         l3fwd_ipv4_route_array[i].depth,
 731                                         l3fwd_ipv4_route_array[i].if_out);
 732                         }
 733                 }
 734
 735                 if (socket_lpm6[socket]) {
 736                         lpm6 = socket_lpm6[socket];
 737                         /* populate the LPM6 table */
 738                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 739                                 ret = rte_lpm6_add(lpm6,
 740                                         l3fwd_ipv6_route_array[i].ip,
 741                                         l3fwd_ipv6_route_array[i].depth,
 742                                         l3fwd_ipv6_route_array[i].if_out);
 743
 744                                 if (ret < 0) {
 745                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 746                                                 "LPM6 table\n", i);
 747                                         return -1;
 748                                 }
 749
 750                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 751                                                 "/%d (port %d)\n",
 752                                         socket,
 753                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 754                                         l3fwd_ipv6_route_array[i].depth,
 755                                         l3fwd_ipv6_route_array[i].if_out);
 756                         }
 757                 }
 758         }
 759         return 0;
 760 }
 761
 762 static int
 763 init_mem(void)
 764 {
 765         char buf[PATH_MAX];
 766         struct rte_mempool *mp;
 767         struct rte_lpm *lpm;
 768         struct rte_lpm6 *lpm6;
 769         struct rte_lpm_config lpm_config;
 770         int socket;
 771         unsigned lcore_id;
 772
 773         /* traverse through lcores and initialize structures on each socket */
 774
 775         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 776
 777                 if (rte_lcore_is_enabled(lcore_id) == 0)
 778                         continue;
 779
 780                 socket = rte_lcore_to_socket_id(lcore_id);
 781
 782                 if (socket == SOCKET_ID_ANY)
 783                         socket = 0;
 784
 785                 if (socket_direct_pool[socket] == NULL) {
 786                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 787                                         socket);
 788                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 789
 790                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
 791                                 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
 792                         if (mp == NULL) {
 793                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 794                                 return -1;
 795                         }
 796                         socket_direct_pool[socket] = mp;
 797                 }
 798
 799                 if (socket_indirect_pool[socket] == NULL) {
 800                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 801                                         socket);
 802                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 803
 804                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
 805                                 socket);
 806                         if (mp == NULL) {
 807                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 808                                 return -1;
 809                         }
 810                         socket_indirect_pool[socket] = mp;
 811                 }
 812
 813                 if (socket_lpm[socket] == NULL) {
 814                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 815                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 816
 817                         lpm_config.max_rules = LPM_MAX_RULES;
 818                         lpm_config.number_tbl8s = 256;
 819                         lpm_config.flags = 0;
 820
 821                         lpm = rte_lpm_create(buf, socket, &lpm_config);
 822                         if (lpm == NULL) {
 823                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 824                                 return -1;
 825                         }
 826                         socket_lpm[socket] = lpm;
 827                 }
 828
 829                 if (socket_lpm6[socket] == NULL) {
 830                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 831                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 832
 833                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
 834                         if (lpm6 == NULL) {
 835                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 836                                 return -1;
 837                         }
 838                         socket_lpm6[socket] = lpm6;
 839                 }
 840         }
 841
 842         return 0;
 843 }
 844
 845 int
 846 main(int argc, char **argv)
 847 {
 848         struct lcore_queue_conf *qconf;
 849         struct rte_eth_dev_info dev_info;
 850         struct rte_eth_txconf *txconf;
 851         struct rx_queue *rxq;
 852         int socket, ret;
 853         uint16_t nb_ports;
 854         uint16_t queueid = 0;
 855         unsigned lcore_id = 0, rx_lcore_id = 0;
 856         uint32_t n_tx_queue, nb_lcores;
 857         uint16_t portid;
 858
 859         /* init EAL */
 860         ret = rte_eal_init(argc, argv);
 861         if (ret < 0)
 862                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 863         argc -= ret;
 864         argv += ret;
 865
 866         /* parse application arguments (after the EAL ones) */
 867         ret = parse_args(argc, argv);
 868         if (ret < 0)
 869                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 870
 871         nb_ports = rte_eth_dev_count_avail();
 872         if (nb_ports == 0)
 873                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 874
 875         nb_lcores = rte_lcore_count();
 876
 877         /* initialize structures (mempools, lpm etc.) */
 878         if (init_mem() < 0)
 879                 rte_panic("Cannot initialize memory structures!\n");
 880
 881         /* check if portmask has non-existent ports */
 882         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 883                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 884
 885         /* initialize all ports */
 886         RTE_ETH_FOREACH_DEV(portid) {
 887                 struct rte_eth_conf local_port_conf = port_conf;
 888                 struct rte_eth_rxconf rxq_conf;
 889
 890                 /* skip ports that are not enabled */
 891                 if ((enabled_port_mask & (1 << portid)) == 0) {
 892                         printf("Skipping disabled port %d\n", portid);
 893                         continue;
 894                 }
 895
 896                 qconf = &lcore_queue_conf[rx_lcore_id];
 897
 898                 /* limit the frame size to the maximum supported by NIC */
 899                 rte_eth_dev_info_get(portid, &dev_info);
 900                 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
 901                     dev_info.max_rx_pktlen,
 902                     local_port_conf.rxmode.max_rx_pkt_len);
 903
 904                 /* get the lcore_id for this port */
 905                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 906                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 907
 908                         rx_lcore_id ++;
 909                         if (rx_lcore_id >= RTE_MAX_LCORE)
 910                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 911
 912                         qconf = &lcore_queue_conf[rx_lcore_id];
 913                 }
 914
 915                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 916                 if (socket == SOCKET_ID_ANY)
 917                         socket = 0;
 918
 919                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 920                 rxq->portid = portid;
 921                 rxq->direct_pool = socket_direct_pool[socket];
 922                 rxq->indirect_pool = socket_indirect_pool[socket];
 923                 rxq->lpm = socket_lpm[socket];
 924                 rxq->lpm6 = socket_lpm6[socket];
 925                 qconf->n_rx_queue++;
 926
 927                 /* init port */
 928                 printf("Initializing port %d on lcore %u...", portid,
 929                        rx_lcore_id);
 930                 fflush(stdout);
 931
 932                 n_tx_queue = nb_lcores;
 933                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 934                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 935                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 936                         local_port_conf.txmode.offloads |=
 937                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 938                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 939                                             &local_port_conf);
 940                 if (ret < 0) {
 941                         printf("\n");
 942                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 943                                 "err=%d, port=%d\n",
 944                                 ret, portid);
 945                 }
 946
 947                 /* set the mtu to the maximum received packet size */
 948                 ret = rte_eth_dev_set_mtu(portid,
 949                         local_port_conf.rxmode.max_rx_pkt_len - MTU_OVERHEAD);
 950                 if (ret < 0) {
 951                         printf("\n");
 952                         rte_exit(EXIT_FAILURE, "Set MTU failed: "
 953                                 "err=%d, port=%d\n",
 954                         ret, portid);
 955                 }
 956
 957                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
 958                                             &nb_txd);
 959                 if (ret < 0) {
 960                         printf("\n");
 961                         rte_exit(EXIT_FAILURE, "Cannot adjust number of "
 962                                 "descriptors: err=%d, port=%d\n", ret, portid);
 963                 }
 964
 965                 /* init one RX queue */
 966                 rxq_conf = dev_info.default_rxconf;
 967                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
 968                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 969                                              socket, &rxq_conf,
 970                                              socket_direct_pool[socket]);
 971                 if (ret < 0) {
 972                         printf("\n");
 973                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 974                                 "err=%d, port=%d\n",
 975                                 ret, portid);
 976                 }
 977
 978                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 979                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
 980                 printf("\n");
 981
 982                 /* init one TX queue per couple (lcore,port) */
 983                 queueid = 0;
 984                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 985                         if (rte_lcore_is_enabled(lcore_id) == 0)
 986                                 continue;
 987
 988                         socket = (int) rte_lcore_to_socket_id(lcore_id);
 989                         printf("txq=%u,%d ", lcore_id, queueid);
 990                         fflush(stdout);
 991
 992                         txconf = &dev_info.default_txconf;
 993                         txconf->offloads = local_port_conf.txmode.offloads;
 994                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
 995                                                      socket, txconf);
 996                         if (ret < 0) {
 997                                 printf("\n");
 998                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
 999                                         "err=%d, port=%d\n", ret, portid);
1000                         }
1001
1002                         qconf = &lcore_queue_conf[lcore_id];
1003                         qconf->tx_queue_id[portid] = queueid;
1004                         queueid++;
1005                 }
1006
1007                 printf("\n");
1008         }
1009
1010         printf("\n");
1011
1012         /* start ports */
1013         RTE_ETH_FOREACH_DEV(portid) {
1014                 if ((enabled_port_mask & (1 << portid)) == 0) {
1015                         continue;
1016                 }
1017                 /* Start device */
1018                 ret = rte_eth_dev_start(portid);
1019                 if (ret < 0)
1020                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1021                                 ret, portid);
1022
1023                 rte_eth_promiscuous_enable(portid);
1024
1025                 if (check_ptype(portid) == 0) {
1026                         rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1027                         printf("Add Rx callback function to detect L3 packet type by SW :"
1028                                 " port = %d\n", portid);
1029                 }
1030         }
1031
1032         if (init_routing_table() < 0)
1033                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1034
1035         check_all_ports_link_status(enabled_port_mask);
1036
1037         /* launch per-lcore init on every lcore */
1038         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1039         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1040                 if (rte_eal_wait_lcore(lcore_id) < 0)
1041                         return -1;
1042         }
1043
1044         return 0;
1045 }