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 (m->ol_flags & PKT_RX_IPV4_HDR) {
 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         }
 321         /* if this is an IPv6 packet */
 322         else if (m->ol_flags & PKT_RX_IPV6_HDR) {
 323                 struct ipv6_hdr *ip_hdr;
 324
 325                 ipv6 = 1;
 326
 327                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 328                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
 329
 330                 /* Find destination port */
 331                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, &next_hop) == 0 &&
 332                                 (enabled_port_mask & 1 << next_hop) != 0) {
 333                         port_out = next_hop;
 334
 335                         /* Build transmission burst for new port */
 336                         len = qconf->tx_mbufs[port_out].len;
 337                 }
 338
 339                 /* if we don't need to do any fragmentation */
 340                 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
 341                         qconf->tx_mbufs[port_out].m_table[len] = m;
 342                         len2 = 1;
 343                 } else {
 344                         len2 = rte_ipv6_fragment_packet(m,
 345                                 &qconf->tx_mbufs[port_out].m_table[len],
 346                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 347                                 IPV6_MTU_DEFAULT,
 348                                 rxq->direct_pool, rxq->indirect_pool);
 349
 350                         /* Free input packet */
 351                         rte_pktmbuf_free(m);
 352
 353                         /* If we fail to fragment the packet */
 354                         if (unlikely (len2 < 0))
 355                                 return;
 356                 }
 357         }
 358         /* else, just forward the packet */
 359         else {
 360                 qconf->tx_mbufs[port_out].m_table[len] = m;
 361                 len2 = 1;
 362         }
 363
 364         for (i = len; i < len + len2; i ++) {
 365                 void *d_addr_bytes;
 366
 367                 m = qconf->tx_mbufs[port_out].m_table[i];
 368                 struct ether_hdr *eth_hdr = (struct ether_hdr *)
 369                         rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
 370                 if (eth_hdr == NULL) {
 371                         rte_panic("No headroom in mbuf.\n");
 372                 }
 373
 374                 m->l2_len = sizeof(struct ether_hdr);
 375
 376                 /* 02:00:00:00:00:xx */
 377                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 378                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
 379
 380                 /* src addr */
 381                 ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
 382                 if (ipv6)
 383                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 384                 else
 385                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 386         }
 387
 388         len += len2;
 389
 390         if (likely(len < MAX_PKT_BURST)) {
 391                 qconf->tx_mbufs[port_out].len = (uint16_t)len;
 392                 return;
 393         }
 394
 395         /* Transmit packets */
 396         send_burst(qconf, (uint16_t)len, port_out);
 397         qconf->tx_mbufs[port_out].len = 0;
 398 }
 399
 400 /* main processing loop */
 401 static int
 402 main_loop(__attribute__((unused)) void *dummy)
 403 {
 404         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 405         unsigned lcore_id;
 406         uint64_t prev_tsc, diff_tsc, cur_tsc;
 407         int i, j, nb_rx;
 408         uint8_t portid;
 409         struct lcore_queue_conf *qconf;
 410         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 411
 412         prev_tsc = 0;
 413
 414         lcore_id = rte_lcore_id();
 415         qconf = &lcore_queue_conf[lcore_id];
 416
 417         if (qconf->n_rx_queue == 0) {
 418                 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
 419                 return 0;
 420         }
 421
 422         RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
 423
 424         for (i = 0; i < qconf->n_rx_queue; i++) {
 425
 426                 portid = qconf->rx_queue_list[i].portid;
 427                 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
 428                                 (int) portid);
 429         }
 430
 431         while (1) {
 432
 433                 cur_tsc = rte_rdtsc();
 434
 435                 /*
 436                  * TX burst queue drain
 437                  */
 438                 diff_tsc = cur_tsc - prev_tsc;
 439                 if (unlikely(diff_tsc > drain_tsc)) {
 440
 441                         /*
 442                          * This could be optimized (use queueid instead of
 443                          * portid), but it is not called so often
 444                          */
 445                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 446                                 if (qconf->tx_mbufs[portid].len == 0)
 447                                         continue;
 448                                 send_burst(&lcore_queue_conf[lcore_id],
 449                                            qconf->tx_mbufs[portid].len,
 450                                            portid);
 451                                 qconf->tx_mbufs[portid].len = 0;
 452                         }
 453
 454                         prev_tsc = cur_tsc;
 455                 }
 456
 457                 /*
 458                  * Read packet from RX queues
 459                  */
 460                 for (i = 0; i < qconf->n_rx_queue; i++) {
 461
 462                         portid = qconf->rx_queue_list[i].portid;
 463                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 464                                                  MAX_PKT_BURST);
 465
 466                         /* Prefetch first packets */
 467                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 468                                 rte_prefetch0(rte_pktmbuf_mtod(
 469                                                 pkts_burst[j], void *));
 470                         }
 471
 472                         /* Prefetch and forward already prefetched packets */
 473                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 474                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 475                                                 j + PREFETCH_OFFSET], void *));
 476                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 477                         }
 478
 479                         /* Forward remaining prefetched packets */
 480                         for (; j < nb_rx; j++) {
 481                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 482                         }
 483                 }
 484         }
 485 }
 486
 487 /* display usage */
 488 static void
 489 print_usage(const char *prgname)
 490 {
 491         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
 492                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 493                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
 494                prgname);
 495 }
 496
 497 static int
 498 parse_portmask(const char *portmask)
 499 {
 500         char *end = NULL;
 501         unsigned long pm;
 502
 503         /* parse hexadecimal string */
 504         pm = strtoul(portmask, &end, 16);
 505         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 506                 return -1;
 507
 508         if (pm == 0)
 509                 return -1;
 510
 511         return pm;
 512 }
 513
 514 static int
 515 parse_nqueue(const char *q_arg)
 516 {
 517         char *end = NULL;
 518         unsigned long n;
 519
 520         /* parse hexadecimal string */
 521         n = strtoul(q_arg, &end, 10);
 522         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 523                 return -1;
 524         if (n == 0)
 525                 return -1;
 526         if (n >= MAX_RX_QUEUE_PER_LCORE)
 527                 return -1;
 528
 529         return n;
 530 }
 531
 532 /* Parse the argument given in the command line of the application */
 533 static int
 534 parse_args(int argc, char **argv)
 535 {
 536         int opt, ret;
 537         char **argvopt;
 538         int option_index;
 539         char *prgname = argv[0];
 540         static struct option lgopts[] = {
 541                 {NULL, 0, 0, 0}
 542         };
 543
 544         argvopt = argv;
 545
 546         while ((opt = getopt_long(argc, argvopt, "p:q:",
 547                                   lgopts, &option_index)) != EOF) {
 548
 549                 switch (opt) {
 550                 /* portmask */
 551                 case 'p':
 552                         enabled_port_mask = parse_portmask(optarg);
 553                         if (enabled_port_mask < 0) {
 554                                 printf("invalid portmask\n");
 555                                 print_usage(prgname);
 556                                 return -1;
 557                         }
 558                         break;
 559
 560                 /* nqueue */
 561                 case 'q':
 562                         rx_queue_per_lcore = parse_nqueue(optarg);
 563                         if (rx_queue_per_lcore < 0) {
 564                                 printf("invalid queue number\n");
 565                                 print_usage(prgname);
 566                                 return -1;
 567                         }
 568                         break;
 569
 570                 /* long options */
 571                 case 0:
 572                         print_usage(prgname);
 573                         return -1;
 574
 575                 default:
 576                         print_usage(prgname);
 577                         return -1;
 578                 }
 579         }
 580
 581         if (enabled_port_mask == 0) {
 582                 printf("portmask not specified\n");
 583                 print_usage(prgname);
 584                 return -1;
 585         }
 586
 587         if (optind >= 0)
 588                 argv[optind-1] = prgname;
 589
 590         ret = optind-1;
 591         optind = 0; /* reset getopt lib */
 592         return ret;
 593 }
 594
 595 static void
 596 print_ethaddr(const char *name, struct ether_addr *eth_addr)
 597 {
 598         char buf[ETHER_ADDR_FMT_SIZE];
 599         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 600         printf("%s%s", name, buf);
 601 }
 602
 603 /* Check the link status of all ports in up to 9s, and print them finally */
 604 static void
 605 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 606 {
 607 #define CHECK_INTERVAL 100 /* 100ms */
 608 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 609         uint8_t portid, count, all_ports_up, print_flag = 0;
 610         struct rte_eth_link link;
 611
 612         printf("\nChecking link status");
 613         fflush(stdout);
 614         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 615                 all_ports_up = 1;
 616                 for (portid = 0; portid < port_num; portid++) {
 617                         if ((port_mask & (1 << portid)) == 0)
 618                                 continue;
 619                         memset(&link, 0, sizeof(link));
 620                         rte_eth_link_get_nowait(portid, &link);
 621                         /* print link status if flag set */
 622                         if (print_flag == 1) {
 623                                 if (link.link_status)
 624                                         printf("Port %d Link Up - speed %u "
 625                                                 "Mbps - %s\n", (uint8_t)portid,
 626                                                 (unsigned)link.link_speed,
 627                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 628                                         ("full-duplex") : ("half-duplex\n"));
 629                                 else
 630                                         printf("Port %d Link Down\n",
 631                                                         (uint8_t)portid);
 632                                 continue;
 633                         }
 634                         /* clear all_ports_up flag if any link down */
 635                         if (link.link_status == 0) {
 636                                 all_ports_up = 0;
 637                                 break;
 638                         }
 639                 }
 640                 /* after finally printing all link status, get out */
 641                 if (print_flag == 1)
 642                         break;
 643
 644                 if (all_ports_up == 0) {
 645                         printf(".");
 646                         fflush(stdout);
 647                         rte_delay_ms(CHECK_INTERVAL);
 648                 }
 649
 650                 /* set the print_flag if all ports up or timeout */
 651                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 652                         print_flag = 1;
 653                         printf("\ndone\n");
 654                 }
 655         }
 656 }
 657
 658 static int
 659 init_routing_table(void)
 660 {
 661         struct rte_lpm *lpm;
 662         struct rte_lpm6 *lpm6;
 663         int socket, ret;
 664         unsigned i;
 665
 666         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 667                 if (socket_lpm[socket]) {
 668                         lpm = socket_lpm[socket];
 669                         /* populate the LPM table */
 670                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 671                                 ret = rte_lpm_add(lpm,
 672                                         l3fwd_ipv4_route_array[i].ip,
 673                                         l3fwd_ipv4_route_array[i].depth,
 674                                         l3fwd_ipv4_route_array[i].if_out);
 675
 676                                 if (ret < 0) {
 677                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 678                                                 "LPM table\n", i);
 679                                         return -1;
 680                                 }
 681
 682                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 683                                                 "/%d (port %d)\n",
 684                                         socket,
 685                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 686                                         l3fwd_ipv4_route_array[i].depth,
 687                                         l3fwd_ipv4_route_array[i].if_out);
 688                         }
 689                 }
 690
 691                 if (socket_lpm6[socket]) {
 692                         lpm6 = socket_lpm6[socket];
 693                         /* populate the LPM6 table */
 694                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 695                                 ret = rte_lpm6_add(lpm6,
 696                                         l3fwd_ipv6_route_array[i].ip,
 697                                         l3fwd_ipv6_route_array[i].depth,
 698                                         l3fwd_ipv6_route_array[i].if_out);
 699
 700                                 if (ret < 0) {
 701                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 702                                                 "LPM6 table\n", i);
 703                                         return -1;
 704                                 }
 705
 706                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 707                                                 "/%d (port %d)\n",
 708                                         socket,
 709                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 710                                         l3fwd_ipv6_route_array[i].depth,
 711                                         l3fwd_ipv6_route_array[i].if_out);
 712                         }
 713                 }
 714         }
 715         return 0;
 716 }
 717
 718 static int
 719 init_mem(void)
 720 {
 721         char buf[PATH_MAX];
 722         struct rte_mempool *mp;
 723         struct rte_lpm *lpm;
 724         struct rte_lpm6 *lpm6;
 725         int socket;
 726         unsigned lcore_id;
 727
 728         /* traverse through lcores and initialize structures on each socket */
 729
 730         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 731
 732                 if (rte_lcore_is_enabled(lcore_id) == 0)
 733                         continue;
 734
 735                 socket = rte_lcore_to_socket_id(lcore_id);
 736
 737                 if (socket == SOCKET_ID_ANY)
 738                         socket = 0;
 739
 740                 if (socket_direct_pool[socket] == NULL) {
 741                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 742                                         socket);
 743                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 744
 745                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
 746                                 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
 747                         if (mp == NULL) {
 748                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 749                                 return -1;
 750                         }
 751                         socket_direct_pool[socket] = mp;
 752                 }
 753
 754                 if (socket_indirect_pool[socket] == NULL) {
 755                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 756                                         socket);
 757                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 758
 759                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
 760                                 socket);
 761                         if (mp == NULL) {
 762                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 763                                 return -1;
 764                         }
 765                         socket_indirect_pool[socket] = mp;
 766                 }
 767
 768                 if (socket_lpm[socket] == NULL) {
 769                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 770                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 771
 772                         lpm = rte_lpm_create(buf, socket, LPM_MAX_RULES, 0);
 773                         if (lpm == NULL) {
 774                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 775                                 return -1;
 776                         }
 777                         socket_lpm[socket] = lpm;
 778                 }
 779
 780                 if (socket_lpm6[socket] == NULL) {
 781                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 782                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 783
 784                         lpm6 = rte_lpm6_create("IP_FRAG_LPM6", socket, &lpm6_config);
 785                         if (lpm6 == NULL) {
 786                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 787                                 return -1;
 788                         }
 789                         socket_lpm6[socket] = lpm6;
 790                 }
 791         }
 792
 793         return 0;
 794 }
 795
 796 int
 797 main(int argc, char **argv)
 798 {
 799         struct lcore_queue_conf *qconf;
 800         struct rte_eth_dev_info dev_info;
 801         struct rte_eth_txconf *txconf;
 802         struct rx_queue *rxq;
 803         int socket, ret;
 804         unsigned nb_ports;
 805         uint16_t queueid = 0;
 806         unsigned lcore_id = 0, rx_lcore_id = 0;
 807         uint32_t n_tx_queue, nb_lcores;
 808         uint8_t portid;
 809
 810         /* init EAL */
 811         ret = rte_eal_init(argc, argv);
 812         if (ret < 0)
 813                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 814         argc -= ret;
 815         argv += ret;
 816
 817         /* parse application arguments (after the EAL ones) */
 818         ret = parse_args(argc, argv);
 819         if (ret < 0)
 820                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 821
 822         nb_ports = rte_eth_dev_count();
 823         if (nb_ports > RTE_MAX_ETHPORTS)
 824                 nb_ports = RTE_MAX_ETHPORTS;
 825         else if (nb_ports == 0)
 826                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 827
 828         nb_lcores = rte_lcore_count();
 829
 830         /* initialize structures (mempools, lpm etc.) */
 831         if (init_mem() < 0)
 832                 rte_panic("Cannot initialize memory structures!\n");
 833
 834         /* check if portmask has non-existent ports */
 835         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 836                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 837
 838         /* initialize all ports */
 839         for (portid = 0; portid < nb_ports; portid++) {
 840                 /* skip ports that are not enabled */
 841                 if ((enabled_port_mask & (1 << portid)) == 0) {
 842                         printf("Skipping disabled port %d\n", portid);
 843                         continue;
 844                 }
 845
 846                 qconf = &lcore_queue_conf[rx_lcore_id];
 847
 848                 /* get the lcore_id for this port */
 849                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 850                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 851
 852                         rx_lcore_id ++;
 853                         if (rx_lcore_id >= RTE_MAX_LCORE)
 854                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 855
 856                         qconf = &lcore_queue_conf[rx_lcore_id];
 857                 }
 858
 859                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 860                 if (socket == SOCKET_ID_ANY)
 861                         socket = 0;
 862
 863                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 864                 rxq->portid = portid;
 865                 rxq->direct_pool = socket_direct_pool[socket];
 866                 rxq->indirect_pool = socket_indirect_pool[socket];
 867                 rxq->lpm = socket_lpm[socket];
 868                 rxq->lpm6 = socket_lpm6[socket];
 869                 qconf->n_rx_queue++;
 870
 871                 /* init port */
 872                 printf("Initializing port %d on lcore %u...", portid,
 873                        rx_lcore_id);
 874                 fflush(stdout);
 875
 876                 n_tx_queue = nb_lcores;
 877                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 878                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 879                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 880                                             &port_conf);
 881                 if (ret < 0) {
 882                         printf("\n");
 883                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 884                                 "err=%d, port=%d\n",
 885                                 ret, portid);
 886                 }
 887
 888                 /* init one RX queue */
 889                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 890                                              socket, NULL,
 891                                              socket_direct_pool[socket]);
 892                 if (ret < 0) {
 893                         printf("\n");
 894                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 895                                 "err=%d, port=%d\n",
 896                                 ret, portid);
 897                 }
 898
 899                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 900                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
 901                 printf("\n");
 902
 903                 /* init one TX queue per couple (lcore,port) */
 904                 queueid = 0;
 905                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 906                         if (rte_lcore_is_enabled(lcore_id) == 0)
 907                                 continue;
 908
 909                         socket = (int) rte_lcore_to_socket_id(lcore_id);
 910                         printf("txq=%u,%d ", lcore_id, queueid);
 911                         fflush(stdout);
 912
 913                         rte_eth_dev_info_get(portid, &dev_info);
 914                         txconf = &dev_info.default_txconf;
 915                         txconf->txq_flags = 0;
 916                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
 917                                                      socket, txconf);
 918                         if (ret < 0) {
 919                                 printf("\n");
 920                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
 921                                         "err=%d, port=%d\n", ret, portid);
 922                         }
 923
 924                         qconf = &lcore_queue_conf[lcore_id];
 925                         qconf->tx_queue_id[portid] = queueid;
 926                         queueid++;
 927                 }
 928
 929                 printf("\n");
 930         }
 931
 932         printf("\n");
 933
 934         /* start ports */
 935         for (portid = 0; portid < nb_ports; portid++) {
 936                 if ((enabled_port_mask & (1 << portid)) == 0) {
 937                         continue;
 938                 }
 939                 /* Start device */
 940                 ret = rte_eth_dev_start(portid);
 941                 if (ret < 0)
 942                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
 943                                 ret, portid);
 944
 945                 rte_eth_promiscuous_enable(portid);
 946         }
 947
 948         if (init_routing_table() < 0)
 949                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
 950
 951         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
 952
 953         /* launch per-lcore init on every lcore */
 954         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
 955         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 956                 if (rte_eal_wait_lcore(lcore_id) < 0)
 957                         return -1;
 958         }
 959
 960         return 0;
 961 }