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        RTE_ETHER_MTU
  56 #define IPV6_MTU_DEFAULT        RTE_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         (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + \
  64                 2 * sizeof(struct rte_vlan_hdr))
  65
  66 /*
  67  * Default payload in bytes for the IPv6 packet.
  68  */
  69 #define IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct rte_ipv4_hdr))
  70 #define IPV6_DEFAULT_PAYLOAD    (IPV6_MTU_DEFAULT - sizeof(struct rte_ipv6_hdr))
  71
  72 /*
  73  * Max number of fragments per packet expected - defined by config file.
  74  */
  75 #define MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
  76
  77 #define NB_MBUF   8192
  78
  79 #define MAX_PKT_BURST   32
  80 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  81
  82 /* Configure how many packets ahead to prefetch, when reading packets */
  83 #define PREFETCH_OFFSET 3
  84
  85 /*
  86  * Configurable number of RX/TX ring descriptors
  87  */
  88 #define RTE_TEST_RX_DESC_DEFAULT 1024
  89 #define RTE_TEST_TX_DESC_DEFAULT 1024
  90 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
  91 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
  92
  93 /* ethernet addresses of ports */
  94 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
  95
  96 #ifndef IPv4_BYTES
  97 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
  98 #define IPv4_BYTES(addr) \
  99                 (uint8_t) (((addr) >> 24) & 0xFF),\
 100                 (uint8_t) (((addr) >> 16) & 0xFF),\
 101                 (uint8_t) (((addr) >> 8) & 0xFF),\
 102                 (uint8_t) ((addr) & 0xFF)
 103 #endif
 104
 105 #ifndef IPv6_BYTES
 106 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 107                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 108 #define IPv6_BYTES(addr) \
 109         addr[0],  addr[1], addr[2],  addr[3], \
 110         addr[4],  addr[5], addr[6],  addr[7], \
 111         addr[8],  addr[9], addr[10], addr[11],\
 112         addr[12], addr[13],addr[14], addr[15]
 113 #endif
 114
 115 #define IPV6_ADDR_LEN 16
 116
 117 /* mask of enabled ports */
 118 static int enabled_port_mask = 0;
 119
 120 static int rx_queue_per_lcore = 1;
 121
 122 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
 123
 124 struct mbuf_table {
 125         uint16_t len;
 126         struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
 127 };
 128
 129 struct rx_queue {
 130         struct rte_mempool *direct_pool;
 131         struct rte_mempool *indirect_pool;
 132         struct rte_lpm *lpm;
 133         struct rte_lpm6 *lpm6;
 134         uint16_t portid;
 135 };
 136
 137 #define MAX_RX_QUEUE_PER_LCORE 16
 138 #define MAX_TX_QUEUE_PER_PORT 16
 139 struct lcore_queue_conf {
 140         uint16_t n_rx_queue;
 141         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 142         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 143         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 144 } __rte_cache_aligned;
 145 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 146
 147 static struct rte_eth_conf port_conf = {
 148         .rxmode = {
 149                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 150                 .split_hdr_size = 0,
 151                 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
 152                              DEV_RX_OFFLOAD_SCATTER |
 153                              DEV_RX_OFFLOAD_JUMBO_FRAME),
 154         },
 155         .txmode = {
 156                 .mq_mode = ETH_MQ_TX_NONE,
 157                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
 158                              DEV_TX_OFFLOAD_MULTI_SEGS),
 159         },
 160 };
 161
 162 /*
 163  * IPv4 forwarding table
 164  */
 165 struct l3fwd_ipv4_route {
 166         uint32_t ip;
 167         uint8_t  depth;
 168         uint8_t  if_out;
 169 };
 170
 171 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 172                 {RTE_IPV4(100,10,0,0), 16, 0},
 173                 {RTE_IPV4(100,20,0,0), 16, 1},
 174                 {RTE_IPV4(100,30,0,0), 16, 2},
 175                 {RTE_IPV4(100,40,0,0), 16, 3},
 176                 {RTE_IPV4(100,50,0,0), 16, 4},
 177                 {RTE_IPV4(100,60,0,0), 16, 5},
 178                 {RTE_IPV4(100,70,0,0), 16, 6},
 179                 {RTE_IPV4(100,80,0,0), 16, 7},
 180 };
 181
 182 /*
 183  * IPv6 forwarding table
 184  */
 185
 186 struct l3fwd_ipv6_route {
 187         uint8_t ip[IPV6_ADDR_LEN];
 188         uint8_t depth;
 189         uint8_t if_out;
 190 };
 191
 192 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 193         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 194         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 195         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 196         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 197         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 198         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 199         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 200         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 201 };
 202
 203 #define LPM_MAX_RULES         1024
 204 #define LPM6_MAX_RULES         1024
 205 #define LPM6_NUMBER_TBL8S (1 << 16)
 206
 207 struct rte_lpm6_config lpm6_config = {
 208                 .max_rules = LPM6_MAX_RULES,
 209                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 210                 .flags = 0
 211 };
 212
 213 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
 214 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
 215 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 216 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 217
 218 /* Send burst of packets on an output interface */
 219 static inline int
 220 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
 221 {
 222         struct rte_mbuf **m_table;
 223         int ret;
 224         uint16_t queueid;
 225
 226         queueid = qconf->tx_queue_id[port];
 227         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 228
 229         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 230         if (unlikely(ret < n)) {
 231                 do {
 232                         rte_pktmbuf_free(m_table[ret]);
 233                 } while (++ret < n);
 234         }
 235
 236         return 0;
 237 }
 238
 239 static inline void
 240 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
 241                 uint8_t queueid, uint16_t port_in)
 242 {
 243         struct rx_queue *rxq;
 244         uint32_t i, len, next_hop;
 245         uint8_t ipv6;
 246         uint16_t port_out;
 247         int32_t len2;
 248         uint64_t ol_flags;
 249
 250         ipv6 = 0;
 251         ol_flags = 0;
 252         rxq = &qconf->rx_queue_list[queueid];
 253
 254         /* by default, send everything back to the source port */
 255         port_out = port_in;
 256
 257         /* Remove the Ethernet header and trailer from the input packet */
 258         rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
 259
 260         /* Build transmission burst */
 261         len = qconf->tx_mbufs[port_out].len;
 262
 263         /* if this is an IPv4 packet */
 264         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 265                 struct rte_ipv4_hdr *ip_hdr;
 266                 uint32_t ip_dst;
 267                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 268                 ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv4_hdr *);
 269                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 270
 271                 /* Find destination port */
 272                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 273                                 (enabled_port_mask & 1 << next_hop) != 0) {
 274                         port_out = next_hop;
 275
 276                         /* Build transmission burst for new port */
 277                         len = qconf->tx_mbufs[port_out].len;
 278                 }
 279
 280                 /* if we don't need to do any fragmentation */
 281                 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
 282                         qconf->tx_mbufs[port_out].m_table[len] = m;
 283                         len2 = 1;
 284                 } else {
 285                         len2 = rte_ipv4_fragment_packet(m,
 286                                 &qconf->tx_mbufs[port_out].m_table[len],
 287                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 288                                 IPV4_MTU_DEFAULT,
 289                                 rxq->direct_pool, rxq->indirect_pool);
 290
 291                         /* Free input packet */
 292                         rte_pktmbuf_free(m);
 293
 294                         /* request HW to regenerate IPv4 cksum */
 295                         ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
 296
 297                         /* If we fail to fragment the packet */
 298                         if (unlikely (len2 < 0))
 299                                 return;
 300                 }
 301         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 302                 /* if this is an IPv6 packet */
 303                 struct rte_ipv6_hdr *ip_hdr;
 304
 305                 ipv6 = 1;
 306
 307                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 308                 ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv6_hdr *);
 309
 310                 /* Find destination port */
 311                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
 312                                                 &next_hop) == 0 &&
 313                                 (enabled_port_mask & 1 << next_hop) != 0) {
 314                         port_out = next_hop;
 315
 316                         /* Build transmission burst for new port */
 317                         len = qconf->tx_mbufs[port_out].len;
 318                 }
 319
 320                 /* if we don't need to do any fragmentation */
 321                 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
 322                         qconf->tx_mbufs[port_out].m_table[len] = m;
 323                         len2 = 1;
 324                 } else {
 325                         len2 = rte_ipv6_fragment_packet(m,
 326                                 &qconf->tx_mbufs[port_out].m_table[len],
 327                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 328                                 IPV6_MTU_DEFAULT,
 329                                 rxq->direct_pool, rxq->indirect_pool);
 330
 331                         /* Free input packet */
 332                         rte_pktmbuf_free(m);
 333
 334                         /* If we fail to fragment the packet */
 335                         if (unlikely (len2 < 0))
 336                                 return;
 337                 }
 338         }
 339         /* else, just forward the packet */
 340         else {
 341                 qconf->tx_mbufs[port_out].m_table[len] = m;
 342                 len2 = 1;
 343         }
 344
 345         for (i = len; i < len + len2; i ++) {
 346                 void *d_addr_bytes;
 347
 348                 m = qconf->tx_mbufs[port_out].m_table[i];
 349                 struct rte_ether_hdr *eth_hdr = (struct rte_ether_hdr *)
 350                         rte_pktmbuf_prepend(m,
 351                                 (uint16_t)sizeof(struct rte_ether_hdr));
 352                 if (eth_hdr == NULL) {
 353                         rte_panic("No headroom in mbuf.\n");
 354                 }
 355
 356                 m->ol_flags |= ol_flags;
 357                 m->l2_len = sizeof(struct rte_ether_hdr);
 358
 359                 /* 02:00:00:00:00:xx */
 360                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 361                 *((uint64_t *)d_addr_bytes) = 0x000000000002 +
 362                         ((uint64_t)port_out << 40);
 363
 364                 /* src addr */
 365                 rte_ether_addr_copy(&ports_eth_addr[port_out],
 366                                 &eth_hdr->s_addr);
 367                 if (ipv6) {
 368                         eth_hdr->ether_type =
 369                                 rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV6);
 370                 } else {
 371                         eth_hdr->ether_type =
 372                                 rte_be_to_cpu_16(RTE_ETHER_TYPE_IPV4);
 373                 }
 374         }
 375
 376         len += len2;
 377
 378         if (likely(len < MAX_PKT_BURST)) {
 379                 qconf->tx_mbufs[port_out].len = (uint16_t)len;
 380                 return;
 381         }
 382
 383         /* Transmit packets */
 384         send_burst(qconf, (uint16_t)len, port_out);
 385         qconf->tx_mbufs[port_out].len = 0;
 386 }
 387
 388 /* main processing loop */
 389 static int
 390 main_loop(__attribute__((unused)) void *dummy)
 391 {
 392         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 393         unsigned lcore_id;
 394         uint64_t prev_tsc, diff_tsc, cur_tsc;
 395         int i, j, nb_rx;
 396         uint16_t portid;
 397         struct lcore_queue_conf *qconf;
 398         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 399
 400         prev_tsc = 0;
 401
 402         lcore_id = rte_lcore_id();
 403         qconf = &lcore_queue_conf[lcore_id];
 404
 405         if (qconf->n_rx_queue == 0) {
 406                 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
 407                 return 0;
 408         }
 409
 410         RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
 411
 412         for (i = 0; i < qconf->n_rx_queue; i++) {
 413
 414                 portid = qconf->rx_queue_list[i].portid;
 415                 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
 416                                 portid);
 417         }
 418
 419         while (1) {
 420
 421                 cur_tsc = rte_rdtsc();
 422
 423                 /*
 424                  * TX burst queue drain
 425                  */
 426                 diff_tsc = cur_tsc - prev_tsc;
 427                 if (unlikely(diff_tsc > drain_tsc)) {
 428
 429                         /*
 430                          * This could be optimized (use queueid instead of
 431                          * portid), but it is not called so often
 432                          */
 433                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 434                                 if (qconf->tx_mbufs[portid].len == 0)
 435                                         continue;
 436                                 send_burst(&lcore_queue_conf[lcore_id],
 437                                            qconf->tx_mbufs[portid].len,
 438                                            portid);
 439                                 qconf->tx_mbufs[portid].len = 0;
 440                         }
 441
 442                         prev_tsc = cur_tsc;
 443                 }
 444
 445                 /*
 446                  * Read packet from RX queues
 447                  */
 448                 for (i = 0; i < qconf->n_rx_queue; i++) {
 449
 450                         portid = qconf->rx_queue_list[i].portid;
 451                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 452                                                  MAX_PKT_BURST);
 453
 454                         /* Prefetch first packets */
 455                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 456                                 rte_prefetch0(rte_pktmbuf_mtod(
 457                                                 pkts_burst[j], void *));
 458                         }
 459
 460                         /* Prefetch and forward already prefetched packets */
 461                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 462                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 463                                                 j + PREFETCH_OFFSET], void *));
 464                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 465                         }
 466
 467                         /* Forward remaining prefetched packets */
 468                         for (; j < nb_rx; j++) {
 469                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 470                         }
 471                 }
 472         }
 473 }
 474
 475 /* display usage */
 476 static void
 477 print_usage(const char *prgname)
 478 {
 479         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
 480                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 481                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
 482                prgname);
 483 }
 484
 485 static int
 486 parse_portmask(const char *portmask)
 487 {
 488         char *end = NULL;
 489         unsigned long pm;
 490
 491         /* parse hexadecimal string */
 492         pm = strtoul(portmask, &end, 16);
 493         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 494                 return -1;
 495
 496         if (pm == 0)
 497                 return -1;
 498
 499         return pm;
 500 }
 501
 502 static int
 503 parse_nqueue(const char *q_arg)
 504 {
 505         char *end = NULL;
 506         unsigned long n;
 507
 508         /* parse hexadecimal string */
 509         n = strtoul(q_arg, &end, 10);
 510         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 511                 return -1;
 512         if (n == 0)
 513                 return -1;
 514         if (n >= MAX_RX_QUEUE_PER_LCORE)
 515                 return -1;
 516
 517         return n;
 518 }
 519
 520 /* Parse the argument given in the command line of the application */
 521 static int
 522 parse_args(int argc, char **argv)
 523 {
 524         int opt, ret;
 525         char **argvopt;
 526         int option_index;
 527         char *prgname = argv[0];
 528         static struct option lgopts[] = {
 529                 {NULL, 0, 0, 0}
 530         };
 531
 532         argvopt = argv;
 533
 534         while ((opt = getopt_long(argc, argvopt, "p:q:",
 535                                   lgopts, &option_index)) != EOF) {
 536
 537                 switch (opt) {
 538                 /* portmask */
 539                 case 'p':
 540                         enabled_port_mask = parse_portmask(optarg);
 541                         if (enabled_port_mask < 0) {
 542                                 printf("invalid portmask\n");
 543                                 print_usage(prgname);
 544                                 return -1;
 545                         }
 546                         break;
 547
 548                 /* nqueue */
 549                 case 'q':
 550                         rx_queue_per_lcore = parse_nqueue(optarg);
 551                         if (rx_queue_per_lcore < 0) {
 552                                 printf("invalid queue number\n");
 553                                 print_usage(prgname);
 554                                 return -1;
 555                         }
 556                         break;
 557
 558                 /* long options */
 559                 case 0:
 560                         print_usage(prgname);
 561                         return -1;
 562
 563                 default:
 564                         print_usage(prgname);
 565                         return -1;
 566                 }
 567         }
 568
 569         if (enabled_port_mask == 0) {
 570                 printf("portmask not specified\n");
 571                 print_usage(prgname);
 572                 return -1;
 573         }
 574
 575         if (optind >= 0)
 576                 argv[optind-1] = prgname;
 577
 578         ret = optind-1;
 579         optind = 1; /* reset getopt lib */
 580         return ret;
 581 }
 582
 583 static void
 584 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr)
 585 {
 586         char buf[RTE_ETHER_ADDR_FMT_SIZE];
 587         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
 588         printf("%s%s", name, buf);
 589 }
 590
 591 /* Check the link status of all ports in up to 9s, and print them finally */
 592 static void
 593 check_all_ports_link_status(uint32_t port_mask)
 594 {
 595 #define CHECK_INTERVAL 100 /* 100ms */
 596 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 597         uint16_t portid;
 598         uint8_t count, all_ports_up, print_flag = 0;
 599         struct rte_eth_link link;
 600
 601         printf("\nChecking link status");
 602         fflush(stdout);
 603         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 604                 all_ports_up = 1;
 605                 RTE_ETH_FOREACH_DEV(portid) {
 606                         if ((port_mask & (1 << portid)) == 0)
 607                                 continue;
 608                         memset(&link, 0, sizeof(link));
 609                         rte_eth_link_get_nowait(portid, &link);
 610                         /* print link status if flag set */
 611                         if (print_flag == 1) {
 612                                 if (link.link_status)
 613                                         printf(
 614                                         "Port%d Link Up .Speed %u Mbps - %s\n",
 615                                                 portid, link.link_speed,
 616                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 617                                         ("full-duplex") : ("half-duplex\n"));
 618                                 else
 619                                         printf("Port %d Link Down\n", portid);
 620                                 continue;
 621                         }
 622                         /* clear all_ports_up flag if any link down */
 623                         if (link.link_status == ETH_LINK_DOWN) {
 624                                 all_ports_up = 0;
 625                                 break;
 626                         }
 627                 }
 628                 /* after finally printing all link status, get out */
 629                 if (print_flag == 1)
 630                         break;
 631
 632                 if (all_ports_up == 0) {
 633                         printf(".");
 634                         fflush(stdout);
 635                         rte_delay_ms(CHECK_INTERVAL);
 636                 }
 637
 638                 /* set the print_flag if all ports up or timeout */
 639                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 640                         print_flag = 1;
 641                         printf("\ndone\n");
 642                 }
 643         }
 644 }
 645
 646 /* Check L3 packet type detection capablity of the NIC port */
 647 static int
 648 check_ptype(int portid)
 649 {
 650         int i, ret;
 651         int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
 652         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
 653
 654         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 655         if (ret <= 0)
 656                 return 0;
 657
 658         uint32_t ptypes[ret];
 659
 660         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 661         for (i = 0; i < ret; ++i) {
 662                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
 663                         ptype_l3_ipv4 = 1;
 664                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
 665                         ptype_l3_ipv6 = 1;
 666         }
 667
 668         if (ptype_l3_ipv4 == 0)
 669                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
 670
 671         if (ptype_l3_ipv6 == 0)
 672                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
 673
 674         if (ptype_l3_ipv4 && ptype_l3_ipv6)
 675                 return 1;
 676
 677         return 0;
 678
 679 }
 680
 681 /* Parse packet type of a packet by SW */
 682 static inline void
 683 parse_ptype(struct rte_mbuf *m)
 684 {
 685         struct rte_ether_hdr *eth_hdr;
 686         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 687         uint16_t ether_type;
 688
 689         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 690         ether_type = eth_hdr->ether_type;
 691         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 692                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 693         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 694                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 695
 696         m->packet_type = packet_type;
 697 }
 698
 699 /* callback function to detect packet type for a queue of a port */
 700 static uint16_t
 701 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 702                    struct rte_mbuf *pkts[], uint16_t nb_pkts,
 703                    uint16_t max_pkts __rte_unused,
 704                    void *user_param __rte_unused)
 705 {
 706         uint16_t i;
 707
 708         for (i = 0; i < nb_pkts; ++i)
 709                 parse_ptype(pkts[i]);
 710
 711         return nb_pkts;
 712 }
 713
 714 static int
 715 init_routing_table(void)
 716 {
 717         struct rte_lpm *lpm;
 718         struct rte_lpm6 *lpm6;
 719         int socket, ret;
 720         unsigned i;
 721
 722         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 723                 if (socket_lpm[socket]) {
 724                         lpm = socket_lpm[socket];
 725                         /* populate the LPM table */
 726                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 727                                 ret = rte_lpm_add(lpm,
 728                                         l3fwd_ipv4_route_array[i].ip,
 729                                         l3fwd_ipv4_route_array[i].depth,
 730                                         l3fwd_ipv4_route_array[i].if_out);
 731
 732                                 if (ret < 0) {
 733                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 734                                                 "LPM table\n", i);
 735                                         return -1;
 736                                 }
 737
 738                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 739                                                 "/%d (port %d)\n",
 740                                         socket,
 741                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 742                                         l3fwd_ipv4_route_array[i].depth,
 743                                         l3fwd_ipv4_route_array[i].if_out);
 744                         }
 745                 }
 746
 747                 if (socket_lpm6[socket]) {
 748                         lpm6 = socket_lpm6[socket];
 749                         /* populate the LPM6 table */
 750                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 751                                 ret = rte_lpm6_add(lpm6,
 752                                         l3fwd_ipv6_route_array[i].ip,
 753                                         l3fwd_ipv6_route_array[i].depth,
 754                                         l3fwd_ipv6_route_array[i].if_out);
 755
 756                                 if (ret < 0) {
 757                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 758                                                 "LPM6 table\n", i);
 759                                         return -1;
 760                                 }
 761
 762                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 763                                                 "/%d (port %d)\n",
 764                                         socket,
 765                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 766                                         l3fwd_ipv6_route_array[i].depth,
 767                                         l3fwd_ipv6_route_array[i].if_out);
 768                         }
 769                 }
 770         }
 771         return 0;
 772 }
 773
 774 static int
 775 init_mem(void)
 776 {
 777         char buf[PATH_MAX];
 778         struct rte_mempool *mp;
 779         struct rte_lpm *lpm;
 780         struct rte_lpm6 *lpm6;
 781         struct rte_lpm_config lpm_config;
 782         int socket;
 783         unsigned lcore_id;
 784
 785         /* traverse through lcores and initialize structures on each socket */
 786
 787         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 788
 789                 if (rte_lcore_is_enabled(lcore_id) == 0)
 790                         continue;
 791
 792                 socket = rte_lcore_to_socket_id(lcore_id);
 793
 794                 if (socket == SOCKET_ID_ANY)
 795                         socket = 0;
 796
 797                 if (socket_direct_pool[socket] == NULL) {
 798                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 799                                         socket);
 800                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 801
 802                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
 803                                 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
 804                         if (mp == NULL) {
 805                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 806                                 return -1;
 807                         }
 808                         socket_direct_pool[socket] = mp;
 809                 }
 810
 811                 if (socket_indirect_pool[socket] == NULL) {
 812                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 813                                         socket);
 814                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 815
 816                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
 817                                 socket);
 818                         if (mp == NULL) {
 819                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 820                                 return -1;
 821                         }
 822                         socket_indirect_pool[socket] = mp;
 823                 }
 824
 825                 if (socket_lpm[socket] == NULL) {
 826                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 827                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 828
 829                         lpm_config.max_rules = LPM_MAX_RULES;
 830                         lpm_config.number_tbl8s = 256;
 831                         lpm_config.flags = 0;
 832
 833                         lpm = rte_lpm_create(buf, socket, &lpm_config);
 834                         if (lpm == NULL) {
 835                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 836                                 return -1;
 837                         }
 838                         socket_lpm[socket] = lpm;
 839                 }
 840
 841                 if (socket_lpm6[socket] == NULL) {
 842                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 843                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 844
 845                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
 846                         if (lpm6 == NULL) {
 847                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 848                                 return -1;
 849                         }
 850                         socket_lpm6[socket] = lpm6;
 851                 }
 852         }
 853
 854         return 0;
 855 }
 856
 857 int
 858 main(int argc, char **argv)
 859 {
 860         struct lcore_queue_conf *qconf;
 861         struct rte_eth_dev_info dev_info;
 862         struct rte_eth_txconf *txconf;
 863         struct rx_queue *rxq;
 864         int socket, ret;
 865         uint16_t nb_ports;
 866         uint16_t queueid = 0;
 867         unsigned lcore_id = 0, rx_lcore_id = 0;
 868         uint32_t n_tx_queue, nb_lcores;
 869         uint16_t portid;
 870
 871         /* init EAL */
 872         ret = rte_eal_init(argc, argv);
 873         if (ret < 0)
 874                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 875         argc -= ret;
 876         argv += ret;
 877
 878         /* parse application arguments (after the EAL ones) */
 879         ret = parse_args(argc, argv);
 880         if (ret < 0)
 881                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 882
 883         nb_ports = rte_eth_dev_count_avail();
 884         if (nb_ports == 0)
 885                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 886
 887         nb_lcores = rte_lcore_count();
 888
 889         /* initialize structures (mempools, lpm etc.) */
 890         if (init_mem() < 0)
 891                 rte_panic("Cannot initialize memory structures!\n");
 892
 893         /* check if portmask has non-existent ports */
 894         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 895                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 896
 897         /* initialize all ports */
 898         RTE_ETH_FOREACH_DEV(portid) {
 899                 struct rte_eth_conf local_port_conf = port_conf;
 900                 struct rte_eth_rxconf rxq_conf;
 901
 902                 /* skip ports that are not enabled */
 903                 if ((enabled_port_mask & (1 << portid)) == 0) {
 904                         printf("Skipping disabled port %d\n", portid);
 905                         continue;
 906                 }
 907
 908                 qconf = &lcore_queue_conf[rx_lcore_id];
 909
 910                 /* limit the frame size to the maximum supported by NIC */
 911                 rte_eth_dev_info_get(portid, &dev_info);
 912                 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
 913                     dev_info.max_rx_pktlen,
 914                     local_port_conf.rxmode.max_rx_pkt_len);
 915
 916                 /* get the lcore_id for this port */
 917                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 918                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 919
 920                         rx_lcore_id ++;
 921                         if (rx_lcore_id >= RTE_MAX_LCORE)
 922                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 923
 924                         qconf = &lcore_queue_conf[rx_lcore_id];
 925                 }
 926
 927                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 928                 if (socket == SOCKET_ID_ANY)
 929                         socket = 0;
 930
 931                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 932                 rxq->portid = portid;
 933                 rxq->direct_pool = socket_direct_pool[socket];
 934                 rxq->indirect_pool = socket_indirect_pool[socket];
 935                 rxq->lpm = socket_lpm[socket];
 936                 rxq->lpm6 = socket_lpm6[socket];
 937                 qconf->n_rx_queue++;
 938
 939                 /* init port */
 940                 printf("Initializing port %d on lcore %u...", portid,
 941                        rx_lcore_id);
 942                 fflush(stdout);
 943
 944                 n_tx_queue = nb_lcores;
 945                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 946                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 947                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
 948                         local_port_conf.txmode.offloads |=
 949                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
 950                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 951                                             &local_port_conf);
 952                 if (ret < 0) {
 953                         printf("\n");
 954                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 955                                 "err=%d, port=%d\n",
 956                                 ret, portid);
 957                 }
 958
 959                 /* set the mtu to the maximum received packet size */
 960                 ret = rte_eth_dev_set_mtu(portid,
 961                         local_port_conf.rxmode.max_rx_pkt_len - MTU_OVERHEAD);
 962                 if (ret < 0) {
 963                         printf("\n");
 964                         rte_exit(EXIT_FAILURE, "Set MTU failed: "
 965                                 "err=%d, port=%d\n",
 966                         ret, portid);
 967                 }
 968
 969                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
 970                                             &nb_txd);
 971                 if (ret < 0) {
 972                         printf("\n");
 973                         rte_exit(EXIT_FAILURE, "Cannot adjust number of "
 974                                 "descriptors: err=%d, port=%d\n", ret, portid);
 975                 }
 976
 977                 /* init one RX queue */
 978                 rxq_conf = dev_info.default_rxconf;
 979                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
 980                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 981                                              socket, &rxq_conf,
 982                                              socket_direct_pool[socket]);
 983                 if (ret < 0) {
 984                         printf("\n");
 985                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 986                                 "err=%d, port=%d\n",
 987                                 ret, portid);
 988                 }
 989
 990                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 991                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
 992                 printf("\n");
 993
 994                 /* init one TX queue per couple (lcore,port) */
 995                 queueid = 0;
 996                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 997                         if (rte_lcore_is_enabled(lcore_id) == 0)
 998                                 continue;
 999
1000                         if (queueid >= dev_info.nb_tx_queues)
1001                                 break;
1002
1003                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1004                         printf("txq=%u,%d ", lcore_id, queueid);
1005                         fflush(stdout);
1006
1007                         txconf = &dev_info.default_txconf;
1008                         txconf->offloads = local_port_conf.txmode.offloads;
1009                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1010                                                      socket, txconf);
1011                         if (ret < 0) {
1012                                 printf("\n");
1013                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1014                                         "err=%d, port=%d\n", ret, portid);
1015                         }
1016
1017                         qconf = &lcore_queue_conf[lcore_id];
1018                         qconf->tx_queue_id[portid] = queueid;
1019                         queueid++;
1020                 }
1021
1022                 printf("\n");
1023         }
1024
1025         printf("\n");
1026
1027         /* start ports */
1028         RTE_ETH_FOREACH_DEV(portid) {
1029                 if ((enabled_port_mask & (1 << portid)) == 0) {
1030                         continue;
1031                 }
1032                 /* Start device */
1033                 ret = rte_eth_dev_start(portid);
1034                 if (ret < 0)
1035                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1036                                 ret, portid);
1037
1038                 rte_eth_promiscuous_enable(portid);
1039
1040                 if (check_ptype(portid) == 0) {
1041                         rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1042                         printf("Add Rx callback function to detect L3 packet type by SW :"
1043                                 " port = %d\n", portid);
1044                 }
1045         }
1046
1047         if (init_routing_table() < 0)
1048                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1049
1050         check_all_ports_link_status(enabled_port_mask);
1051
1052         /* launch per-lcore init on every lcore */
1053         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1054         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1055                 if (rte_eal_wait_lcore(lcore_id) < 0)
1056                         return -1;
1057         }
1058
1059         return 0;
1060 }