examples/ip_fragmentation/main.c

   1 /*-
   2  *   BSD LICENSE
   3  *
   4  *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
   5  *   All rights reserved.
   6  *
   7  *   Redistribution and use in source and binary forms, with or without
   8  *   modification, are permitted provided that the following conditions
   9  *   are met:
  10  *
  11  *     * Redistributions of source code must retain the above copyright
  12  *       notice, this list of conditions and the following disclaimer.
  13  *     * Redistributions in binary form must reproduce the above copyright
  14  *       notice, this list of conditions and the following disclaimer in
  15  *       the documentation and/or other materials provided with the
  16  *       distribution.
  17  *     * Neither the name of Intel Corporation nor the names of its
  18  *       contributors may be used to endorse or promote products derived
  19  *       from this software without specific prior written permission.
  20  *
  21  *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  24  *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  25  *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  26  *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  27  *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  28  *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  29  *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  30  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  31  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32  */
  33
  34 #include <stdio.h>
  35 #include <stdlib.h>
  36 #include <stdint.h>
  37 #include <inttypes.h>
  38 #include <sys/types.h>
  39 #include <sys/param.h>
  40 #include <string.h>
  41 #include <sys/queue.h>
  42 #include <stdarg.h>
  43 #include <errno.h>
  44 #include <getopt.h>
  45
  46 #include <rte_common.h>
  47 #include <rte_byteorder.h>
  48 #include <rte_log.h>
  49 #include <rte_memory.h>
  50 #include <rte_memcpy.h>
  51 #include <rte_memzone.h>
  52 #include <rte_eal.h>
  53 #include <rte_per_lcore.h>
  54 #include <rte_launch.h>
  55 #include <rte_atomic.h>
  56 #include <rte_cycles.h>
  57 #include <rte_prefetch.h>
  58 #include <rte_lcore.h>
  59 #include <rte_per_lcore.h>
  60 #include <rte_branch_prediction.h>
  61 #include <rte_interrupts.h>
  62 #include <rte_pci.h>
  63 #include <rte_random.h>
  64 #include <rte_debug.h>
  65 #include <rte_ether.h>
  66 #include <rte_ethdev.h>
  67 #include <rte_ring.h>
  68 #include <rte_mempool.h>
  69 #include <rte_mbuf.h>
  70 #include <rte_lpm.h>
  71 #include <rte_lpm6.h>
  72 #include <rte_ip.h>
  73 #include <rte_string_fns.h>
  74
  75 #include <rte_ip_frag.h>
  76
  77 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
  78
  79 /* allow max jumbo frame 9.5 KB */
  80 #define JUMBO_FRAME_MAX_SIZE    0x2600
  81
  82 #define ROUNDUP_DIV(a, b)       (((a) + (b) - 1) / (b))
  83
  84 /*
  85  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
  86  * This value includes the size of IPv6 header.
  87  */
  88 #define IPV4_MTU_DEFAULT        ETHER_MTU
  89 #define IPV6_MTU_DEFAULT        ETHER_MTU
  90
  91 /*
  92  * Default payload in bytes for the IPv6 packet.
  93  */
  94 #define IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
  95 #define IPV6_DEFAULT_PAYLOAD    (IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
  96
  97 /*
  98  * Max number of fragments per packet expected - defined by config file.
  99  */
 100 #define MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
 101
 102 #define NB_MBUF   8192
 103
 104 #define MAX_PKT_BURST   32
 105 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 106
 107 /* Configure how many packets ahead to prefetch, when reading packets */
 108 #define PREFETCH_OFFSET 3
 109
 110 /*
 111  * Configurable number of RX/TX ring descriptors
 112  */
 113 #define RTE_TEST_RX_DESC_DEFAULT 128
 114 #define RTE_TEST_TX_DESC_DEFAULT 512
 115 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 116 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 117
 118 /* ethernet addresses of ports */
 119 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 120
 121 #ifndef IPv4_BYTES
 122 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
 123 #define IPv4_BYTES(addr) \
 124                 (uint8_t) (((addr) >> 24) & 0xFF),\
 125                 (uint8_t) (((addr) >> 16) & 0xFF),\
 126                 (uint8_t) (((addr) >> 8) & 0xFF),\
 127                 (uint8_t) ((addr) & 0xFF)
 128 #endif
 129
 130 #ifndef IPv6_BYTES
 131 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 132                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 133 #define IPv6_BYTES(addr) \
 134         addr[0],  addr[1], addr[2],  addr[3], \
 135         addr[4],  addr[5], addr[6],  addr[7], \
 136         addr[8],  addr[9], addr[10], addr[11],\
 137         addr[12], addr[13],addr[14], addr[15]
 138 #endif
 139
 140 #define IPV6_ADDR_LEN 16
 141
 142 /* mask of enabled ports */
 143 static int enabled_port_mask = 0;
 144
 145 static int rx_queue_per_lcore = 1;
 146
 147 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
 148
 149 struct mbuf_table {
 150         uint16_t len;
 151         struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
 152 };
 153
 154 struct rx_queue {
 155         struct rte_mempool *direct_pool;
 156         struct rte_mempool *indirect_pool;
 157         struct rte_lpm *lpm;
 158         struct rte_lpm6 *lpm6;
 159         uint8_t portid;
 160 };
 161
 162 #define MAX_RX_QUEUE_PER_LCORE 16
 163 #define MAX_TX_QUEUE_PER_PORT 16
 164 struct lcore_queue_conf {
 165         uint16_t n_rx_queue;
 166         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 167         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 168         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 169 } __rte_cache_aligned;
 170 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 171
 172 static const struct rte_eth_conf port_conf = {
 173         .rxmode = {
 174                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 175                 .split_hdr_size = 0,
 176                 .header_split   = 0, /**< Header Split disabled */
 177                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 178                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 179                 .jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
 180                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 181         },
 182         .txmode = {
 183                 .mq_mode = ETH_MQ_TX_NONE,
 184         },
 185 };
 186
 187 /*
 188  * IPv4 forwarding table
 189  */
 190 struct l3fwd_ipv4_route {
 191         uint32_t ip;
 192         uint8_t  depth;
 193         uint8_t  if_out;
 194 };
 195
 196 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 197                 {IPv4(100,10,0,0), 16, 0},
 198                 {IPv4(100,20,0,0), 16, 1},
 199                 {IPv4(100,30,0,0), 16, 2},
 200                 {IPv4(100,40,0,0), 16, 3},
 201                 {IPv4(100,50,0,0), 16, 4},
 202                 {IPv4(100,60,0,0), 16, 5},
 203                 {IPv4(100,70,0,0), 16, 6},
 204                 {IPv4(100,80,0,0), 16, 7},
 205 };
 206
 207 /*
 208  * IPv6 forwarding table
 209  */
 210
 211 struct l3fwd_ipv6_route {
 212         uint8_t ip[IPV6_ADDR_LEN];
 213         uint8_t depth;
 214         uint8_t if_out;
 215 };
 216
 217 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 218         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 219         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 220         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 221         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 222         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 223         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 224         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 225         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 226 };
 227
 228 #define LPM_MAX_RULES         1024
 229 #define LPM6_MAX_RULES         1024
 230 #define LPM6_NUMBER_TBL8S (1 << 16)
 231
 232 struct rte_lpm6_config lpm6_config = {
 233                 .max_rules = LPM6_MAX_RULES,
 234                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 235                 .flags = 0
 236 };
 237
 238 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
 239 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
 240 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 241 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 242
 243 /* Send burst of packets on an output interface */
 244 static inline int
 245 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint8_t port)
 246 {
 247         struct rte_mbuf **m_table;
 248         int ret;
 249         uint16_t queueid;
 250
 251         queueid = qconf->tx_queue_id[port];
 252         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 253
 254         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 255         if (unlikely(ret < n)) {
 256                 do {
 257                         rte_pktmbuf_free(m_table[ret]);
 258                 } while (++ret < n);
 259         }
 260
 261         return 0;
 262 }
 263
 264 static inline void
 265 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
 266                 uint8_t queueid, uint8_t port_in)
 267 {
 268         struct rx_queue *rxq;
 269         uint32_t i, len;
 270         uint8_t next_hop, port_out, ipv6;
 271         int32_t len2;
 272
 273         ipv6 = 0;
 274         rxq = &qconf->rx_queue_list[queueid];
 275
 276         /* by default, send everything back to the source port */
 277         port_out = port_in;
 278
 279         /* Remove the Ethernet header and trailer from the input packet */
 280         rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
 281
 282         /* Build transmission burst */
 283         len = qconf->tx_mbufs[port_out].len;
 284
 285         /* if this is an IPv4 packet */
 286         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 287                 struct ipv4_hdr *ip_hdr;
 288                 uint32_t ip_dst;
 289                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 290                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
 291                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 292
 293                 /* Find destination port */
 294                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 295                                 (enabled_port_mask & 1 << next_hop) != 0) {
 296                         port_out = next_hop;
 297
 298                         /* Build transmission burst for new port */
 299                         len = qconf->tx_mbufs[port_out].len;
 300                 }
 301
 302                 /* if we don't need to do any fragmentation */
 303                 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
 304                         qconf->tx_mbufs[port_out].m_table[len] = m;
 305                         len2 = 1;
 306                 } else {
 307                         len2 = rte_ipv4_fragment_packet(m,
 308                                 &qconf->tx_mbufs[port_out].m_table[len],
 309                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 310                                 IPV4_MTU_DEFAULT,
 311                                 rxq->direct_pool, rxq->indirect_pool);
 312
 313                         /* Free input packet */
 314                         rte_pktmbuf_free(m);
 315
 316                         /* If we fail to fragment the packet */
 317                         if (unlikely (len2 < 0))
 318                                 return;
 319                 }
 320         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 321                 /* if this is an IPv6 packet */
 322                 struct ipv6_hdr *ip_hdr;
 323
 324                 ipv6 = 1;
 325
 326                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 327                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
 328
 329                 /* Find destination port */
 330                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, &next_hop) == 0 &&
 331                                 (enabled_port_mask & 1 << next_hop) != 0) {
 332                         port_out = next_hop;
 333
 334                         /* Build transmission burst for new port */
 335                         len = qconf->tx_mbufs[port_out].len;
 336                 }
 337
 338                 /* if we don't need to do any fragmentation */
 339                 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
 340                         qconf->tx_mbufs[port_out].m_table[len] = m;
 341                         len2 = 1;
 342                 } else {
 343                         len2 = rte_ipv6_fragment_packet(m,
 344                                 &qconf->tx_mbufs[port_out].m_table[len],
 345                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 346                                 IPV6_MTU_DEFAULT,
 347                                 rxq->direct_pool, rxq->indirect_pool);
 348
 349                         /* Free input packet */
 350                         rte_pktmbuf_free(m);
 351
 352                         /* If we fail to fragment the packet */
 353                         if (unlikely (len2 < 0))
 354                                 return;
 355                 }
 356         }
 357         /* else, just forward the packet */
 358         else {
 359                 qconf->tx_mbufs[port_out].m_table[len] = m;
 360                 len2 = 1;
 361         }
 362
 363         for (i = len; i < len + len2; i ++) {
 364                 void *d_addr_bytes;
 365
 366                 m = qconf->tx_mbufs[port_out].m_table[i];
 367                 struct ether_hdr *eth_hdr = (struct ether_hdr *)
 368                         rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
 369                 if (eth_hdr == NULL) {
 370                         rte_panic("No headroom in mbuf.\n");
 371                 }
 372
 373                 m->l2_len = sizeof(struct ether_hdr);
 374
 375                 /* 02:00:00:00:00:xx */
 376                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 377                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
 378
 379                 /* src addr */
 380                 ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
 381                 if (ipv6)
 382                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 383                 else
 384                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 385         }
 386
 387         len += len2;
 388
 389         if (likely(len < MAX_PKT_BURST)) {
 390                 qconf->tx_mbufs[port_out].len = (uint16_t)len;
 391                 return;
 392         }
 393
 394         /* Transmit packets */
 395         send_burst(qconf, (uint16_t)len, port_out);
 396         qconf->tx_mbufs[port_out].len = 0;
 397 }
 398
 399 /* main processing loop */
 400 static int
 401 main_loop(__attribute__((unused)) void *dummy)
 402 {
 403         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 404         unsigned lcore_id;
 405         uint64_t prev_tsc, diff_tsc, cur_tsc;
 406         int i, j, nb_rx;
 407         uint8_t portid;
 408         struct lcore_queue_conf *qconf;
 409         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 410
 411         prev_tsc = 0;
 412
 413         lcore_id = rte_lcore_id();
 414         qconf = &lcore_queue_conf[lcore_id];
 415
 416         if (qconf->n_rx_queue == 0) {
 417                 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
 418                 return 0;
 419         }
 420
 421         RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
 422
 423         for (i = 0; i < qconf->n_rx_queue; i++) {
 424
 425                 portid = qconf->rx_queue_list[i].portid;
 426                 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
 427                                 (int) portid);
 428         }
 429
 430         while (1) {
 431
 432                 cur_tsc = rte_rdtsc();
 433
 434                 /*
 435                  * TX burst queue drain
 436                  */
 437                 diff_tsc = cur_tsc - prev_tsc;
 438                 if (unlikely(diff_tsc > drain_tsc)) {
 439
 440                         /*
 441                          * This could be optimized (use queueid instead of
 442                          * portid), but it is not called so often
 443                          */
 444                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 445                                 if (qconf->tx_mbufs[portid].len == 0)
 446                                         continue;
 447                                 send_burst(&lcore_queue_conf[lcore_id],
 448                                            qconf->tx_mbufs[portid].len,
 449                                            portid);
 450                                 qconf->tx_mbufs[portid].len = 0;
 451                         }
 452
 453                         prev_tsc = cur_tsc;
 454                 }
 455
 456                 /*
 457                  * Read packet from RX queues
 458                  */
 459                 for (i = 0; i < qconf->n_rx_queue; i++) {
 460
 461                         portid = qconf->rx_queue_list[i].portid;
 462                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 463                                                  MAX_PKT_BURST);
 464
 465                         /* Prefetch first packets */
 466                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 467                                 rte_prefetch0(rte_pktmbuf_mtod(
 468                                                 pkts_burst[j], void *));
 469                         }
 470
 471                         /* Prefetch and forward already prefetched packets */
 472                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 473                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 474                                                 j + PREFETCH_OFFSET], void *));
 475                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 476                         }
 477
 478                         /* Forward remaining prefetched packets */
 479                         for (; j < nb_rx; j++) {
 480                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 481                         }
 482                 }
 483         }
 484 }
 485
 486 /* display usage */
 487 static void
 488 print_usage(const char *prgname)
 489 {
 490         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
 491                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 492                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
 493                prgname);
 494 }
 495
 496 static int
 497 parse_portmask(const char *portmask)
 498 {
 499         char *end = NULL;
 500         unsigned long pm;
 501
 502         /* parse hexadecimal string */
 503         pm = strtoul(portmask, &end, 16);
 504         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 505                 return -1;
 506
 507         if (pm == 0)
 508                 return -1;
 509
 510         return pm;
 511 }
 512
 513 static int
 514 parse_nqueue(const char *q_arg)
 515 {
 516         char *end = NULL;
 517         unsigned long n;
 518
 519         /* parse hexadecimal string */
 520         n = strtoul(q_arg, &end, 10);
 521         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 522                 return -1;
 523         if (n == 0)
 524                 return -1;
 525         if (n >= MAX_RX_QUEUE_PER_LCORE)
 526                 return -1;
 527
 528         return n;
 529 }
 530
 531 /* Parse the argument given in the command line of the application */
 532 static int
 533 parse_args(int argc, char **argv)
 534 {
 535         int opt, ret;
 536         char **argvopt;
 537         int option_index;
 538         char *prgname = argv[0];
 539         static struct option lgopts[] = {
 540                 {NULL, 0, 0, 0}
 541         };
 542
 543         argvopt = argv;
 544
 545         while ((opt = getopt_long(argc, argvopt, "p:q:",
 546                                   lgopts, &option_index)) != EOF) {
 547
 548                 switch (opt) {
 549                 /* portmask */
 550                 case 'p':
 551                         enabled_port_mask = parse_portmask(optarg);
 552                         if (enabled_port_mask < 0) {
 553                                 printf("invalid portmask\n");
 554                                 print_usage(prgname);
 555                                 return -1;
 556                         }
 557                         break;
 558
 559                 /* nqueue */
 560                 case 'q':
 561                         rx_queue_per_lcore = parse_nqueue(optarg);
 562                         if (rx_queue_per_lcore < 0) {
 563                                 printf("invalid queue number\n");
 564                                 print_usage(prgname);
 565                                 return -1;
 566                         }
 567                         break;
 568
 569                 /* long options */
 570                 case 0:
 571                         print_usage(prgname);
 572                         return -1;
 573
 574                 default:
 575                         print_usage(prgname);
 576                         return -1;
 577                 }
 578         }
 579
 580         if (enabled_port_mask == 0) {
 581                 printf("portmask not specified\n");
 582                 print_usage(prgname);
 583                 return -1;
 584         }
 585
 586         if (optind >= 0)
 587                 argv[optind-1] = prgname;
 588
 589         ret = optind-1;
 590         optind = 0; /* reset getopt lib */
 591         return ret;
 592 }
 593
 594 static void
 595 print_ethaddr(const char *name, struct ether_addr *eth_addr)
 596 {
 597         char buf[ETHER_ADDR_FMT_SIZE];
 598         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 599         printf("%s%s", name, buf);
 600 }
 601
 602 /* Check the link status of all ports in up to 9s, and print them finally */
 603 static void
 604 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 605 {
 606 #define CHECK_INTERVAL 100 /* 100ms */
 607 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 608         uint8_t portid, count, all_ports_up, print_flag = 0;
 609         struct rte_eth_link link;
 610
 611         printf("\nChecking link status");
 612         fflush(stdout);
 613         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 614                 all_ports_up = 1;
 615                 for (portid = 0; portid < port_num; portid++) {
 616                         if ((port_mask & (1 << portid)) == 0)
 617                                 continue;
 618                         memset(&link, 0, sizeof(link));
 619                         rte_eth_link_get_nowait(portid, &link);
 620                         /* print link status if flag set */
 621                         if (print_flag == 1) {
 622                                 if (link.link_status)
 623                                         printf("Port %d Link Up - speed %u "
 624                                                 "Mbps - %s\n", (uint8_t)portid,
 625                                                 (unsigned)link.link_speed,
 626                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 627                                         ("full-duplex") : ("half-duplex\n"));
 628                                 else
 629                                         printf("Port %d Link Down\n",
 630                                                         (uint8_t)portid);
 631                                 continue;
 632                         }
 633                         /* clear all_ports_up flag if any link down */
 634                         if (link.link_status == 0) {
 635                                 all_ports_up = 0;
 636                                 break;
 637                         }
 638                 }
 639                 /* after finally printing all link status, get out */
 640                 if (print_flag == 1)
 641                         break;
 642
 643                 if (all_ports_up == 0) {
 644                         printf(".");
 645                         fflush(stdout);
 646                         rte_delay_ms(CHECK_INTERVAL);
 647                 }
 648
 649                 /* set the print_flag if all ports up or timeout */
 650                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 651                         print_flag = 1;
 652                         printf("\ndone\n");
 653                 }
 654         }
 655 }
 656
 657 static int
 658 init_routing_table(void)
 659 {
 660         struct rte_lpm *lpm;
 661         struct rte_lpm6 *lpm6;
 662         int socket, ret;
 663         unsigned i;
 664
 665         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 666                 if (socket_lpm[socket]) {
 667                         lpm = socket_lpm[socket];
 668                         /* populate the LPM table */
 669                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 670                                 ret = rte_lpm_add(lpm,
 671                                         l3fwd_ipv4_route_array[i].ip,
 672                                         l3fwd_ipv4_route_array[i].depth,
 673                                         l3fwd_ipv4_route_array[i].if_out);
 674
 675                                 if (ret < 0) {
 676                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 677                                                 "LPM table\n", i);
 678                                         return -1;
 679                                 }
 680
 681                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 682                                                 "/%d (port %d)\n",
 683                                         socket,
 684                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 685                                         l3fwd_ipv4_route_array[i].depth,
 686                                         l3fwd_ipv4_route_array[i].if_out);
 687                         }
 688                 }
 689
 690                 if (socket_lpm6[socket]) {
 691                         lpm6 = socket_lpm6[socket];
 692                         /* populate the LPM6 table */
 693                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 694                                 ret = rte_lpm6_add(lpm6,
 695                                         l3fwd_ipv6_route_array[i].ip,
 696                                         l3fwd_ipv6_route_array[i].depth,
 697                                         l3fwd_ipv6_route_array[i].if_out);
 698
 699                                 if (ret < 0) {
 700                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 701                                                 "LPM6 table\n", i);
 702                                         return -1;
 703                                 }
 704
 705                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 706                                                 "/%d (port %d)\n",
 707                                         socket,
 708                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 709                                         l3fwd_ipv6_route_array[i].depth,
 710                                         l3fwd_ipv6_route_array[i].if_out);
 711                         }
 712                 }
 713         }
 714         return 0;
 715 }
 716
 717 static int
 718 init_mem(void)
 719 {
 720         char buf[PATH_MAX];
 721         struct rte_mempool *mp;
 722         struct rte_lpm *lpm;
 723         struct rte_lpm6 *lpm6;
 724         int socket;
 725         unsigned lcore_id;
 726
 727         /* traverse through lcores and initialize structures on each socket */
 728
 729         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 730
 731                 if (rte_lcore_is_enabled(lcore_id) == 0)
 732                         continue;
 733
 734                 socket = rte_lcore_to_socket_id(lcore_id);
 735
 736                 if (socket == SOCKET_ID_ANY)
 737                         socket = 0;
 738
 739                 if (socket_direct_pool[socket] == NULL) {
 740                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 741                                         socket);
 742                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 743
 744                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
 745                                 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
 746                         if (mp == NULL) {
 747                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 748                                 return -1;
 749                         }
 750                         socket_direct_pool[socket] = mp;
 751                 }
 752
 753                 if (socket_indirect_pool[socket] == NULL) {
 754                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 755                                         socket);
 756                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 757
 758                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
 759                                 socket);
 760                         if (mp == NULL) {
 761                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 762                                 return -1;
 763                         }
 764                         socket_indirect_pool[socket] = mp;
 765                 }
 766
 767                 if (socket_lpm[socket] == NULL) {
 768                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 769                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 770
 771                         lpm = rte_lpm_create(buf, socket, LPM_MAX_RULES, 0);
 772                         if (lpm == NULL) {
 773                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 774                                 return -1;
 775                         }
 776                         socket_lpm[socket] = lpm;
 777                 }
 778
 779                 if (socket_lpm6[socket] == NULL) {
 780                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 781                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 782
 783                         lpm6 = rte_lpm6_create("IP_FRAG_LPM6", socket, &lpm6_config);
 784                         if (lpm6 == NULL) {
 785                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 786                                 return -1;
 787                         }
 788                         socket_lpm6[socket] = lpm6;
 789                 }
 790         }
 791
 792         return 0;
 793 }
 794
 795 int
 796 main(int argc, char **argv)
 797 {
 798         struct lcore_queue_conf *qconf;
 799         struct rte_eth_dev_info dev_info;
 800         struct rte_eth_txconf *txconf;
 801         struct rx_queue *rxq;
 802         int socket, ret;
 803         unsigned nb_ports;
 804         uint16_t queueid = 0;
 805         unsigned lcore_id = 0, rx_lcore_id = 0;
 806         uint32_t n_tx_queue, nb_lcores;
 807         uint8_t portid;
 808
 809         /* init EAL */
 810         ret = rte_eal_init(argc, argv);
 811         if (ret < 0)
 812                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 813         argc -= ret;
 814         argv += ret;
 815
 816         /* parse application arguments (after the EAL ones) */
 817         ret = parse_args(argc, argv);
 818         if (ret < 0)
 819                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 820
 821         nb_ports = rte_eth_dev_count();
 822         if (nb_ports > RTE_MAX_ETHPORTS)
 823                 nb_ports = RTE_MAX_ETHPORTS;
 824         else if (nb_ports == 0)
 825                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 826
 827         nb_lcores = rte_lcore_count();
 828
 829         /* initialize structures (mempools, lpm etc.) */
 830         if (init_mem() < 0)
 831                 rte_panic("Cannot initialize memory structures!\n");
 832
 833         /* check if portmask has non-existent ports */
 834         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 835                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 836
 837         /* initialize all ports */
 838         for (portid = 0; portid < nb_ports; portid++) {
 839                 /* skip ports that are not enabled */
 840                 if ((enabled_port_mask & (1 << portid)) == 0) {
 841                         printf("Skipping disabled port %d\n", portid);
 842                         continue;
 843                 }
 844
 845                 qconf = &lcore_queue_conf[rx_lcore_id];
 846
 847                 /* get the lcore_id for this port */
 848                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 849                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 850
 851                         rx_lcore_id ++;
 852                         if (rx_lcore_id >= RTE_MAX_LCORE)
 853                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 854
 855                         qconf = &lcore_queue_conf[rx_lcore_id];
 856                 }
 857
 858                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 859                 if (socket == SOCKET_ID_ANY)
 860                         socket = 0;
 861
 862                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 863                 rxq->portid = portid;
 864                 rxq->direct_pool = socket_direct_pool[socket];
 865                 rxq->indirect_pool = socket_indirect_pool[socket];
 866                 rxq->lpm = socket_lpm[socket];
 867                 rxq->lpm6 = socket_lpm6[socket];
 868                 qconf->n_rx_queue++;
 869
 870                 /* init port */
 871                 printf("Initializing port %d on lcore %u...", portid,
 872                        rx_lcore_id);
 873                 fflush(stdout);
 874
 875                 n_tx_queue = nb_lcores;
 876                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 877                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 878                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 879                                             &port_conf);
 880                 if (ret < 0) {
 881                         printf("\n");
 882                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 883                                 "err=%d, port=%d\n",
 884                                 ret, portid);
 885                 }
 886
 887                 /* init one RX queue */
 888                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 889                                              socket, NULL,
 890                                              socket_direct_pool[socket]);
 891                 if (ret < 0) {
 892                         printf("\n");
 893                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 894                                 "err=%d, port=%d\n",
 895                                 ret, portid);
 896                 }
 897
 898                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 899                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
 900                 printf("\n");
 901
 902                 /* init one TX queue per couple (lcore,port) */
 903                 queueid = 0;
 904                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 905                         if (rte_lcore_is_enabled(lcore_id) == 0)
 906                                 continue;
 907
 908                         socket = (int) rte_lcore_to_socket_id(lcore_id);
 909                         printf("txq=%u,%d ", lcore_id, queueid);
 910                         fflush(stdout);
 911
 912                         rte_eth_dev_info_get(portid, &dev_info);
 913                         txconf = &dev_info.default_txconf;
 914                         txconf->txq_flags = 0;
 915                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
 916                                                      socket, txconf);
 917                         if (ret < 0) {
 918                                 printf("\n");
 919                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
 920                                         "err=%d, port=%d\n", ret, portid);
 921                         }
 922
 923                         qconf = &lcore_queue_conf[lcore_id];
 924                         qconf->tx_queue_id[portid] = queueid;
 925                         queueid++;
 926                 }
 927
 928                 printf("\n");
 929         }
 930
 931         printf("\n");
 932
 933         /* start ports */
 934         for (portid = 0; portid < nb_ports; portid++) {
 935                 if ((enabled_port_mask & (1 << portid)) == 0) {
 936                         continue;
 937                 }
 938                 /* Start device */
 939                 ret = rte_eth_dev_start(portid);
 940                 if (ret < 0)
 941                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
 942                                 ret, portid);
 943
 944                 rte_eth_promiscuous_enable(portid);
 945         }
 946
 947         if (init_routing_table() < 0)
 948                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
 949
 950         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
 951
 952         /* launch per-lcore init on every lcore */
 953         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
 954         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 955                 if (rte_eal_wait_lcore(lcore_id) < 0)
 956                         return -1;
 957         }
 958
 959         return 0;
 960 }