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_cycles.h>
  25 #include <rte_prefetch.h>
  26 #include <rte_lcore.h>
  27 #include <rte_per_lcore.h>
  28 #include <rte_branch_prediction.h>
  29 #include <rte_interrupts.h>
  30 #include <rte_random.h>
  31 #include <rte_debug.h>
  32 #include <rte_ether.h>
  33 #include <rte_ethdev.h>
  34 #include <rte_mempool.h>
  35 #include <rte_mbuf.h>
  36 #include <rte_lpm.h>
  37 #include <rte_lpm6.h>
  38 #include <rte_ip.h>
  39 #include <rte_string_fns.h>
  40
  41 #include <rte_ip_frag.h>
  42
  43 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
  44
  45 /* allow max jumbo frame 9.5 KB */
  46 #define JUMBO_FRAME_MAX_SIZE    0x2600
  47
  48 #define ROUNDUP_DIV(a, b)       (((a) + (b) - 1) / (b))
  49
  50 /*
  51  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
  52  * This value includes the size of IPv6 header.
  53  */
  54 #define IPV4_MTU_DEFAULT        RTE_ETHER_MTU
  55 #define IPV6_MTU_DEFAULT        RTE_ETHER_MTU
  56
  57 /*
  58  * The overhead from max frame size to MTU.
  59  * We have to consider the max possible overhead.
  60  */
  61 #define MTU_OVERHEAD    \
  62         (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + \
  63                 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 rte_ipv4_hdr))
  69 #define IPV6_DEFAULT_PAYLOAD    (IPV6_MTU_DEFAULT - sizeof(struct rte_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                 .mtu = JUMBO_FRAME_MAX_SIZE - RTE_ETHER_HDR_LEN -
 149                         RTE_ETHER_CRC_LEN,
 150                 .split_hdr_size = 0,
 151                 .offloads = (RTE_ETH_RX_OFFLOAD_CHECKSUM |
 152                              RTE_ETH_RX_OFFLOAD_SCATTER),
 153         },
 154         .txmode = {
 155                 .mq_mode = RTE_ETH_MQ_TX_NONE,
 156                 .offloads = (RTE_ETH_TX_OFFLOAD_IPV4_CKSUM |
 157                              RTE_ETH_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 /* Default l3fwd_ipv4_route_array table. 8< */
 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 /* >8 End of default l3fwd_ipv4_route_array table */
 182
 183 /*
 184  * IPv6 forwarding table
 185  */
 186
 187 struct l3fwd_ipv6_route {
 188         uint8_t ip[IPV6_ADDR_LEN];
 189         uint8_t depth;
 190         uint8_t if_out;
 191 };
 192
 193 /* Default l3fwd_ipv6_route_array table. 8< */
 194 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 195         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 196         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 197         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 198         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 199         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 200         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 201         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 202         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 203 };
 204 /* >8 End of default l3fwd_ipv6_route_array table. */
 205
 206 #define LPM_MAX_RULES         1024
 207 #define LPM6_MAX_RULES         1024
 208 #define LPM6_NUMBER_TBL8S (1 << 16)
 209
 210 struct rte_lpm6_config lpm6_config = {
 211                 .max_rules = LPM6_MAX_RULES,
 212                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 213                 .flags = 0
 214 };
 215
 216 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
 217 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
 218 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 219 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 220
 221 /* Send burst of packets on an output interface */
 222 static inline int
 223 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
 224 {
 225         struct rte_mbuf **m_table;
 226         int ret;
 227         uint16_t queueid;
 228
 229         queueid = qconf->tx_queue_id[port];
 230         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 231
 232         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 233         if (unlikely(ret < n)) {
 234                 do {
 235                         rte_pktmbuf_free(m_table[ret]);
 236                 } while (++ret < n);
 237         }
 238
 239         return 0;
 240 }
 241
 242 static inline void
 243 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
 244                 uint8_t queueid, uint16_t port_in)
 245 {
 246         struct rx_queue *rxq;
 247         uint32_t i, len, next_hop;
 248         uint16_t port_out, ether_type;
 249         int32_t len2;
 250         uint64_t ol_flags;
 251         const struct rte_ether_hdr *eth;
 252
 253         ol_flags = 0;
 254         rxq = &qconf->rx_queue_list[queueid];
 255
 256         /* by default, send everything back to the source port */
 257         port_out = port_in;
 258
 259         /* save ether type of the incoming packet */
 260         eth = rte_pktmbuf_mtod(m, const struct rte_ether_hdr *);
 261         ether_type = eth->ether_type;
 262
 263         /* Remove the Ethernet header and trailer from the input packet */
 264         rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
 265
 266         /* Build transmission burst */
 267         len = qconf->tx_mbufs[port_out].len;
 268
 269         /* if this is an IPv4 packet */
 270         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 271                 struct rte_ipv4_hdr *ip_hdr;
 272                 uint32_t ip_dst;
 273                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 274                 ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv4_hdr *);
 275                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 276
 277                 /* Find destination port */
 278                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 279                                 (enabled_port_mask & 1 << next_hop) != 0) {
 280                         port_out = next_hop;
 281
 282                         /* Build transmission burst for new port */
 283                         len = qconf->tx_mbufs[port_out].len;
 284                 }
 285
 286                 /* if we don't need to do any fragmentation */
 287                 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
 288                         qconf->tx_mbufs[port_out].m_table[len] = m;
 289                         len2 = 1;
 290                 } else {
 291                         len2 = rte_ipv4_fragment_packet(m,
 292                                 &qconf->tx_mbufs[port_out].m_table[len],
 293                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 294                                 IPV4_MTU_DEFAULT,
 295                                 rxq->direct_pool, rxq->indirect_pool);
 296
 297                         /* Free input packet */
 298                         rte_pktmbuf_free(m);
 299
 300                         /* request HW to regenerate IPv4 cksum */
 301                         ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
 302
 303                         /* If we fail to fragment the packet */
 304                         if (unlikely (len2 < 0))
 305                                 return;
 306                 }
 307         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 308                 /* if this is an IPv6 packet */
 309                 struct rte_ipv6_hdr *ip_hdr;
 310
 311                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 312                 ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv6_hdr *);
 313
 314                 /* Find destination port */
 315                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
 316                                                 &next_hop) == 0 &&
 317                                 (enabled_port_mask & 1 << next_hop) != 0) {
 318                         port_out = next_hop;
 319
 320                         /* Build transmission burst for new port */
 321                         len = qconf->tx_mbufs[port_out].len;
 322                 }
 323
 324                 /* if we don't need to do any fragmentation */
 325                 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
 326                         qconf->tx_mbufs[port_out].m_table[len] = m;
 327                         len2 = 1;
 328                 } else {
 329                         len2 = rte_ipv6_fragment_packet(m,
 330                                 &qconf->tx_mbufs[port_out].m_table[len],
 331                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 332                                 IPV6_MTU_DEFAULT,
 333                                 rxq->direct_pool, rxq->indirect_pool);
 334
 335                         /* Free input packet */
 336                         rte_pktmbuf_free(m);
 337
 338                         /* If we fail to fragment the packet */
 339                         if (unlikely (len2 < 0))
 340                                 return;
 341                 }
 342         }
 343         /* else, just forward the packet */
 344         else {
 345                 qconf->tx_mbufs[port_out].m_table[len] = m;
 346                 len2 = 1;
 347         }
 348
 349         for (i = len; i < len + len2; i ++) {
 350                 void *d_addr_bytes;
 351
 352                 m = qconf->tx_mbufs[port_out].m_table[i];
 353                 struct rte_ether_hdr *eth_hdr = (struct rte_ether_hdr *)
 354                         rte_pktmbuf_prepend(m,
 355                                 (uint16_t)sizeof(struct rte_ether_hdr));
 356                 if (eth_hdr == NULL) {
 357                         rte_panic("No headroom in mbuf.\n");
 358                 }
 359
 360                 m->ol_flags |= ol_flags;
 361                 m->l2_len = sizeof(struct rte_ether_hdr);
 362
 363                 /* 02:00:00:00:00:xx */
 364                 d_addr_bytes = &eth_hdr->dst_addr.addr_bytes[0];
 365                 *((uint64_t *)d_addr_bytes) = 0x000000000002 +
 366                         ((uint64_t)port_out << 40);
 367
 368                 /* src addr */
 369                 rte_ether_addr_copy(&ports_eth_addr[port_out],
 370                                 &eth_hdr->src_addr);
 371                 eth_hdr->ether_type = ether_type;
 372         }
 373
 374         len += len2;
 375
 376         if (likely(len < MAX_PKT_BURST)) {
 377                 qconf->tx_mbufs[port_out].len = (uint16_t)len;
 378                 return;
 379         }
 380
 381         /* Transmit packets */
 382         send_burst(qconf, (uint16_t)len, port_out);
 383         qconf->tx_mbufs[port_out].len = 0;
 384 }
 385
 386 /* main processing loop */
 387 static int
 388 main_loop(__rte_unused void *dummy)
 389 {
 390         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 391         unsigned lcore_id;
 392         uint64_t prev_tsc, diff_tsc, cur_tsc;
 393         int i, j, nb_rx;
 394         uint16_t portid;
 395         struct lcore_queue_conf *qconf;
 396         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 397
 398         prev_tsc = 0;
 399
 400         lcore_id = rte_lcore_id();
 401         qconf = &lcore_queue_conf[lcore_id];
 402
 403         if (qconf->n_rx_queue == 0) {
 404                 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
 405                 return 0;
 406         }
 407
 408         RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
 409
 410         for (i = 0; i < qconf->n_rx_queue; i++) {
 411
 412                 portid = qconf->rx_queue_list[i].portid;
 413                 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
 414                                 portid);
 415         }
 416
 417         while (1) {
 418
 419                 cur_tsc = rte_rdtsc();
 420
 421                 /*
 422                  * TX burst queue drain
 423                  */
 424                 diff_tsc = cur_tsc - prev_tsc;
 425                 if (unlikely(diff_tsc > drain_tsc)) {
 426
 427                         /*
 428                          * This could be optimized (use queueid instead of
 429                          * portid), but it is not called so often
 430                          */
 431                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 432                                 if (qconf->tx_mbufs[portid].len == 0)
 433                                         continue;
 434                                 send_burst(&lcore_queue_conf[lcore_id],
 435                                            qconf->tx_mbufs[portid].len,
 436                                            portid);
 437                                 qconf->tx_mbufs[portid].len = 0;
 438                         }
 439
 440                         prev_tsc = cur_tsc;
 441                 }
 442
 443                 /*
 444                  * Read packet from RX queues
 445                  */
 446                 for (i = 0; i < qconf->n_rx_queue; i++) {
 447
 448                         portid = qconf->rx_queue_list[i].portid;
 449                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 450                                                  MAX_PKT_BURST);
 451
 452                         /* Prefetch first packets */
 453                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 454                                 rte_prefetch0(rte_pktmbuf_mtod(
 455                                                 pkts_burst[j], void *));
 456                         }
 457
 458                         /* Prefetch and forward already prefetched packets */
 459                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 460                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 461                                                 j + PREFETCH_OFFSET], void *));
 462                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 463                         }
 464
 465                         /* Forward remaining prefetched packets */
 466                         for (; j < nb_rx; j++) {
 467                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 468                         }
 469                 }
 470         }
 471 }
 472
 473 /* display usage */
 474 static void
 475 print_usage(const char *prgname)
 476 {
 477         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
 478                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 479                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
 480                prgname);
 481 }
 482
 483 static int
 484 parse_portmask(const char *portmask)
 485 {
 486         char *end = NULL;
 487         unsigned long pm;
 488
 489         /* parse hexadecimal string */
 490         pm = strtoul(portmask, &end, 16);
 491         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 492                 return -1;
 493
 494         if (pm == 0)
 495                 return -1;
 496
 497         return pm;
 498 }
 499
 500 static int
 501 parse_nqueue(const char *q_arg)
 502 {
 503         char *end = NULL;
 504         unsigned long n;
 505
 506         /* parse hexadecimal string */
 507         n = strtoul(q_arg, &end, 10);
 508         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 509                 return -1;
 510         if (n == 0)
 511                 return -1;
 512         if (n >= MAX_RX_QUEUE_PER_LCORE)
 513                 return -1;
 514
 515         return n;
 516 }
 517
 518 /* Parse the argument given in the command line of the application */
 519 static int
 520 parse_args(int argc, char **argv)
 521 {
 522         int opt, ret;
 523         char **argvopt;
 524         int option_index;
 525         char *prgname = argv[0];
 526         static struct option lgopts[] = {
 527                 {NULL, 0, 0, 0}
 528         };
 529
 530         argvopt = argv;
 531
 532         while ((opt = getopt_long(argc, argvopt, "p:q:",
 533                                   lgopts, &option_index)) != EOF) {
 534
 535                 switch (opt) {
 536                 /* portmask */
 537                 case 'p':
 538                         enabled_port_mask = parse_portmask(optarg);
 539                         if (enabled_port_mask < 0) {
 540                                 printf("invalid portmask\n");
 541                                 print_usage(prgname);
 542                                 return -1;
 543                         }
 544                         break;
 545
 546                 /* nqueue */
 547                 case 'q':
 548                         rx_queue_per_lcore = parse_nqueue(optarg);
 549                         if (rx_queue_per_lcore < 0) {
 550                                 printf("invalid queue number\n");
 551                                 print_usage(prgname);
 552                                 return -1;
 553                         }
 554                         break;
 555
 556                 /* long options */
 557                 case 0:
 558                         print_usage(prgname);
 559                         return -1;
 560
 561                 default:
 562                         print_usage(prgname);
 563                         return -1;
 564                 }
 565         }
 566
 567         if (enabled_port_mask == 0) {
 568                 printf("portmask not specified\n");
 569                 print_usage(prgname);
 570                 return -1;
 571         }
 572
 573         if (optind >= 0)
 574                 argv[optind-1] = prgname;
 575
 576         ret = optind-1;
 577         optind = 1; /* reset getopt lib */
 578         return ret;
 579 }
 580
 581 static void
 582 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr)
 583 {
 584         char buf[RTE_ETHER_ADDR_FMT_SIZE];
 585         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
 586         printf("%s%s", name, buf);
 587 }
 588
 589 /* Check the link status of all ports in up to 9s, and print them finally */
 590 static void
 591 check_all_ports_link_status(uint32_t port_mask)
 592 {
 593 #define CHECK_INTERVAL 100 /* 100ms */
 594 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 595         uint16_t portid;
 596         uint8_t count, all_ports_up, print_flag = 0;
 597         struct rte_eth_link link;
 598         int ret;
 599         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
 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                         ret = rte_eth_link_get_nowait(portid, &link);
 610                         if (ret < 0) {
 611                                 all_ports_up = 0;
 612                                 if (print_flag == 1)
 613                                         printf("Port %u link get failed: %s\n",
 614                                                 portid, rte_strerror(-ret));
 615                                 continue;
 616                         }
 617                         /* print link status if flag set */
 618                         if (print_flag == 1) {
 619                                 rte_eth_link_to_str(link_status_text,
 620                                         sizeof(link_status_text), &link);
 621                                 printf("Port %d %s\n", portid,
 622                                        link_status_text);
 623                                 continue;
 624                         }
 625                         /* clear all_ports_up flag if any link down */
 626                         if (link.link_status == RTE_ETH_LINK_DOWN) {
 627                                 all_ports_up = 0;
 628                                 break;
 629                         }
 630                 }
 631                 /* after finally printing all link status, get out */
 632                 if (print_flag == 1)
 633                         break;
 634
 635                 if (all_ports_up == 0) {
 636                         printf(".");
 637                         fflush(stdout);
 638                         rte_delay_ms(CHECK_INTERVAL);
 639                 }
 640
 641                 /* set the print_flag if all ports up or timeout */
 642                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 643                         print_flag = 1;
 644                         printf("\ndone\n");
 645                 }
 646         }
 647 }
 648
 649 /* Check L3 packet type detection capability of the NIC port */
 650 static int
 651 check_ptype(int portid)
 652 {
 653         int i, ret;
 654         int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
 655         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
 656
 657         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 658         if (ret <= 0)
 659                 return 0;
 660
 661         uint32_t ptypes[ret];
 662
 663         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 664         for (i = 0; i < ret; ++i) {
 665                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
 666                         ptype_l3_ipv4 = 1;
 667                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
 668                         ptype_l3_ipv6 = 1;
 669         }
 670
 671         if (ptype_l3_ipv4 == 0)
 672                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
 673
 674         if (ptype_l3_ipv6 == 0)
 675                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
 676
 677         if (ptype_l3_ipv4 && ptype_l3_ipv6)
 678                 return 1;
 679
 680         return 0;
 681
 682 }
 683
 684 /* Parse packet type of a packet by SW */
 685 static inline void
 686 parse_ptype(struct rte_mbuf *m)
 687 {
 688         struct rte_ether_hdr *eth_hdr;
 689         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 690         uint16_t ether_type;
 691
 692         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 693         ether_type = eth_hdr->ether_type;
 694         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 695                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 696         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 697                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 698
 699         m->packet_type = packet_type;
 700 }
 701
 702 /* callback function to detect packet type for a queue of a port */
 703 static uint16_t
 704 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 705                    struct rte_mbuf *pkts[], uint16_t nb_pkts,
 706                    uint16_t max_pkts __rte_unused,
 707                    void *user_param __rte_unused)
 708 {
 709         uint16_t i;
 710
 711         for (i = 0; i < nb_pkts; ++i)
 712                 parse_ptype(pkts[i]);
 713
 714         return nb_pkts;
 715 }
 716
 717 static int
 718 init_routing_table(void)
 719 {
 720         struct rte_lpm *lpm;
 721         struct rte_lpm6 *lpm6;
 722         int socket, ret;
 723         unsigned i;
 724
 725         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 726                 if (socket_lpm[socket]) {
 727                         lpm = socket_lpm[socket];
 728                         /* populate the LPM table */
 729                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 730                                 ret = rte_lpm_add(lpm,
 731                                         l3fwd_ipv4_route_array[i].ip,
 732                                         l3fwd_ipv4_route_array[i].depth,
 733                                         l3fwd_ipv4_route_array[i].if_out);
 734
 735                                 if (ret < 0) {
 736                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 737                                                 "LPM table\n", i);
 738                                         return -1;
 739                                 }
 740
 741                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 742                                                 "/%d (port %d)\n",
 743                                         socket,
 744                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 745                                         l3fwd_ipv4_route_array[i].depth,
 746                                         l3fwd_ipv4_route_array[i].if_out);
 747                         }
 748                 }
 749
 750                 if (socket_lpm6[socket]) {
 751                         lpm6 = socket_lpm6[socket];
 752                         /* populate the LPM6 table */
 753                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 754                                 ret = rte_lpm6_add(lpm6,
 755                                         l3fwd_ipv6_route_array[i].ip,
 756                                         l3fwd_ipv6_route_array[i].depth,
 757                                         l3fwd_ipv6_route_array[i].if_out);
 758
 759                                 if (ret < 0) {
 760                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 761                                                 "LPM6 table\n", i);
 762                                         return -1;
 763                                 }
 764
 765                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 766                                                 "/%d (port %d)\n",
 767                                         socket,
 768                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 769                                         l3fwd_ipv6_route_array[i].depth,
 770                                         l3fwd_ipv6_route_array[i].if_out);
 771                         }
 772                 }
 773         }
 774         return 0;
 775 }
 776
 777 static int
 778 init_mem(void)
 779 {
 780         char buf[PATH_MAX];
 781         struct rte_mempool *mp;
 782         struct rte_lpm *lpm;
 783         struct rte_lpm6 *lpm6;
 784         struct rte_lpm_config lpm_config;
 785         int socket;
 786         unsigned lcore_id;
 787
 788         /* traverse through lcores and initialize structures on each socket */
 789
 790         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 791
 792                 if (rte_lcore_is_enabled(lcore_id) == 0)
 793                         continue;
 794
 795                 socket = rte_lcore_to_socket_id(lcore_id);
 796
 797                 if (socket == SOCKET_ID_ANY)
 798                         socket = 0;
 799
 800                 if (socket_direct_pool[socket] == NULL) {
 801                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 802                                         socket);
 803                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 804
 805                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
 806                                 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
 807                         if (mp == NULL) {
 808                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 809                                 return -1;
 810                         }
 811                         socket_direct_pool[socket] = mp;
 812                 }
 813
 814                 if (socket_indirect_pool[socket] == NULL) {
 815                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 816                                         socket);
 817                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 818
 819                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
 820                                 socket);
 821                         if (mp == NULL) {
 822                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 823                                 return -1;
 824                         }
 825                         socket_indirect_pool[socket] = mp;
 826                 }
 827
 828                 if (socket_lpm[socket] == NULL) {
 829                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 830                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 831
 832                         lpm_config.max_rules = LPM_MAX_RULES;
 833                         lpm_config.number_tbl8s = 256;
 834                         lpm_config.flags = 0;
 835
 836                         lpm = rte_lpm_create(buf, socket, &lpm_config);
 837                         if (lpm == NULL) {
 838                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 839                                 return -1;
 840                         }
 841                         socket_lpm[socket] = lpm;
 842                 }
 843
 844                 if (socket_lpm6[socket] == NULL) {
 845                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 846                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 847
 848                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
 849                         if (lpm6 == NULL) {
 850                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 851                                 return -1;
 852                         }
 853                         socket_lpm6[socket] = lpm6;
 854                 }
 855         }
 856
 857         return 0;
 858 }
 859
 860 int
 861 main(int argc, char **argv)
 862 {
 863         struct lcore_queue_conf *qconf;
 864         struct rte_eth_dev_info dev_info;
 865         struct rte_eth_txconf *txconf;
 866         struct rx_queue *rxq;
 867         int socket, ret;
 868         uint16_t nb_ports;
 869         uint16_t queueid = 0;
 870         unsigned lcore_id = 0, rx_lcore_id = 0;
 871         uint32_t n_tx_queue, nb_lcores;
 872         uint16_t portid;
 873
 874         /* init EAL */
 875         ret = rte_eal_init(argc, argv);
 876         if (ret < 0)
 877                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 878         argc -= ret;
 879         argv += ret;
 880
 881         /* parse application arguments (after the EAL ones) */
 882         ret = parse_args(argc, argv);
 883         if (ret < 0)
 884                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 885
 886         nb_ports = rte_eth_dev_count_avail();
 887         if (nb_ports == 0)
 888                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 889
 890         nb_lcores = rte_lcore_count();
 891
 892         /* initialize structures (mempools, lpm etc.) */
 893         if (init_mem() < 0)
 894                 rte_panic("Cannot initialize memory structures!\n");
 895
 896         /* check if portmask has non-existent ports */
 897         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 898                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 899
 900         /* initialize all ports */
 901         RTE_ETH_FOREACH_DEV(portid) {
 902                 struct rte_eth_conf local_port_conf = port_conf;
 903                 struct rte_eth_rxconf rxq_conf;
 904
 905                 /* skip ports that are not enabled */
 906                 if ((enabled_port_mask & (1 << portid)) == 0) {
 907                         printf("Skipping disabled port %d\n", portid);
 908                         continue;
 909                 }
 910
 911                 qconf = &lcore_queue_conf[rx_lcore_id];
 912
 913                 /* limit the frame size to the maximum supported by NIC */
 914                 ret = rte_eth_dev_info_get(portid, &dev_info);
 915                 if (ret != 0)
 916                         rte_exit(EXIT_FAILURE,
 917                                 "Error during getting device (port %u) info: %s\n",
 918                                 portid, strerror(-ret));
 919
 920                 local_port_conf.rxmode.mtu = RTE_MIN(
 921                     dev_info.max_mtu,
 922                     local_port_conf.rxmode.mtu);
 923
 924                 /* get the lcore_id for this port */
 925                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 926                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 927
 928                         rx_lcore_id ++;
 929                         if (rx_lcore_id >= RTE_MAX_LCORE)
 930                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 931
 932                         qconf = &lcore_queue_conf[rx_lcore_id];
 933                 }
 934
 935                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 936                 if (socket == SOCKET_ID_ANY)
 937                         socket = 0;
 938
 939                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 940                 rxq->portid = portid;
 941                 rxq->direct_pool = socket_direct_pool[socket];
 942                 rxq->indirect_pool = socket_indirect_pool[socket];
 943                 rxq->lpm = socket_lpm[socket];
 944                 rxq->lpm6 = socket_lpm6[socket];
 945                 qconf->n_rx_queue++;
 946
 947                 /* init port */
 948                 printf("Initializing port %d on lcore %u...", portid,
 949                        rx_lcore_id);
 950                 fflush(stdout);
 951
 952                 n_tx_queue = nb_lcores;
 953                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 954                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 955                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 956                                             &local_port_conf);
 957                 if (ret < 0) {
 958                         printf("\n");
 959                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 960                                 "err=%d, port=%d\n",
 961                                 ret, portid);
 962                 }
 963
 964                 /* set the mtu to the maximum received packet size */
 965                 ret = rte_eth_dev_set_mtu(portid, local_port_conf.rxmode.mtu);
 966                 if (ret < 0) {
 967                         printf("\n");
 968                         rte_exit(EXIT_FAILURE, "Set MTU failed: "
 969                                 "err=%d, port=%d\n",
 970                         ret, portid);
 971                 }
 972
 973                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
 974                                             &nb_txd);
 975                 if (ret < 0) {
 976                         printf("\n");
 977                         rte_exit(EXIT_FAILURE, "Cannot adjust number of "
 978                                 "descriptors: err=%d, port=%d\n", ret, portid);
 979                 }
 980
 981                 /* init one RX queue */
 982                 rxq_conf = dev_info.default_rxconf;
 983                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
 984                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 985                                              socket, &rxq_conf,
 986                                              socket_direct_pool[socket]);
 987                 if (ret < 0) {
 988                         printf("\n");
 989                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 990                                 "err=%d, port=%d\n",
 991                                 ret, portid);
 992                 }
 993
 994                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 995                 if (ret < 0) {
 996                         printf("\n");
 997                         rte_exit(EXIT_FAILURE,
 998                                 "rte_eth_macaddr_get: err=%d, port=%d\n",
 999                                 ret, portid);
1000                 }
1001
1002                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1003                 printf("\n");
1004
1005                 /* init one TX queue per couple (lcore,port) */
1006                 ret = rte_eth_dev_info_get(portid, &dev_info);
1007                 if (ret != 0)
1008                         rte_exit(EXIT_FAILURE,
1009                                 "Error during getting device (port %u) info: %s\n",
1010                                 portid, strerror(-ret));
1011
1012                 queueid = 0;
1013                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1014                         if (rte_lcore_is_enabled(lcore_id) == 0)
1015                                 continue;
1016
1017                         if (queueid >= dev_info.nb_tx_queues)
1018                                 break;
1019
1020                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1021                         printf("txq=%u,%d ", lcore_id, queueid);
1022                         fflush(stdout);
1023
1024                         txconf = &dev_info.default_txconf;
1025                         txconf->offloads = local_port_conf.txmode.offloads;
1026                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1027                                                      socket, txconf);
1028                         if (ret < 0) {
1029                                 printf("\n");
1030                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1031                                         "err=%d, port=%d\n", ret, portid);
1032                         }
1033
1034                         qconf = &lcore_queue_conf[lcore_id];
1035                         qconf->tx_queue_id[portid] = queueid;
1036                         queueid++;
1037                 }
1038
1039                 printf("\n");
1040         }
1041
1042         printf("\n");
1043
1044         /* start ports */
1045         RTE_ETH_FOREACH_DEV(portid) {
1046                 if ((enabled_port_mask & (1 << portid)) == 0) {
1047                         continue;
1048                 }
1049                 /* Start device */
1050                 ret = rte_eth_dev_start(portid);
1051                 if (ret < 0)
1052                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1053                                 ret, portid);
1054
1055                 ret = rte_eth_promiscuous_enable(portid);
1056                 if (ret != 0)
1057                         rte_exit(EXIT_FAILURE,
1058                                 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
1059                                 rte_strerror(-ret), portid);
1060
1061                 if (check_ptype(portid) == 0) {
1062                         rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1063                         printf("Add Rx callback function to detect L3 packet type by SW :"
1064                                 " port = %d\n", portid);
1065                 }
1066         }
1067
1068         if (init_routing_table() < 0)
1069                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1070
1071         check_all_ports_link_status(enabled_port_mask);
1072
1073         /* launch per-lcore init on every lcore */
1074         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MAIN);
1075         RTE_LCORE_FOREACH_WORKER(lcore_id) {
1076                 if (rte_eal_wait_lcore(lcore_id) < 0)
1077                         return -1;
1078         }
1079
1080         /* clean up the EAL */
1081         rte_eal_cleanup();
1082
1083         return 0;
1084 }