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_tailq.h>
  53 #include <rte_eal.h>
  54 #include <rte_per_lcore.h>
  55 #include <rte_launch.h>
  56 #include <rte_atomic.h>
  57 #include <rte_cycles.h>
  58 #include <rte_prefetch.h>
  59 #include <rte_lcore.h>
  60 #include <rte_per_lcore.h>
  61 #include <rte_branch_prediction.h>
  62 #include <rte_interrupts.h>
  63 #include <rte_pci.h>
  64 #include <rte_random.h>
  65 #include <rte_debug.h>
  66 #include <rte_ether.h>
  67 #include <rte_ethdev.h>
  68 #include <rte_ring.h>
  69 #include <rte_mempool.h>
  70 #include <rte_mbuf.h>
  71 #include <rte_lpm.h>
  72 #include <rte_lpm6.h>
  73 #include <rte_ip.h>
  74 #include <rte_string_fns.h>
  75
  76 #include <rte_ip_frag.h>
  77
  78 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
  79
  80 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
  81
  82 /* allow max jumbo frame 9.5 KB */
  83 #define JUMBO_FRAME_MAX_SIZE    0x2600
  84
  85 #define ROUNDUP_DIV(a, b)       (((a) + (b) - 1) / (b))
  86
  87 /*
  88  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
  89  * This value includes the size of IPv6 header.
  90  */
  91 #define IPV4_MTU_DEFAULT        ETHER_MTU
  92 #define IPV6_MTU_DEFAULT        ETHER_MTU
  93
  94 /*
  95  * Default payload in bytes for the IPv6 packet.
  96  */
  97 #define IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
  98 #define IPV6_DEFAULT_PAYLOAD    (IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
  99
 100 /*
 101  * Max number of fragments per packet expected - defined by config file.
 102  */
 103 #define MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
 104
 105 #define NB_MBUF   8192
 106
 107 #define MAX_PKT_BURST   32
 108 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 109
 110 /* Configure how many packets ahead to prefetch, when reading packets */
 111 #define PREFETCH_OFFSET 3
 112
 113 /*
 114  * Configurable number of RX/TX ring descriptors
 115  */
 116 #define RTE_TEST_RX_DESC_DEFAULT 128
 117 #define RTE_TEST_TX_DESC_DEFAULT 512
 118 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 119 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 120
 121 /* ethernet addresses of ports */
 122 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 123
 124 #ifndef IPv4_BYTES
 125 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
 126 #define IPv4_BYTES(addr) \
 127                 (uint8_t) (((addr) >> 24) & 0xFF),\
 128                 (uint8_t) (((addr) >> 16) & 0xFF),\
 129                 (uint8_t) (((addr) >> 8) & 0xFF),\
 130                 (uint8_t) ((addr) & 0xFF)
 131 #endif
 132
 133 #ifndef IPv6_BYTES
 134 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 135                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 136 #define IPv6_BYTES(addr) \
 137         addr[0],  addr[1], addr[2],  addr[3], \
 138         addr[4],  addr[5], addr[6],  addr[7], \
 139         addr[8],  addr[9], addr[10], addr[11],\
 140         addr[12], addr[13],addr[14], addr[15]
 141 #endif
 142
 143 #define IPV6_ADDR_LEN 16
 144
 145 /* mask of enabled ports */
 146 static int enabled_port_mask = 0;
 147
 148 static int rx_queue_per_lcore = 1;
 149
 150 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
 151
 152 struct mbuf_table {
 153         uint16_t len;
 154         struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
 155 };
 156
 157 struct rx_queue {
 158         struct rte_mempool *direct_pool;
 159         struct rte_mempool *indirect_pool;
 160         struct rte_lpm *lpm;
 161         struct rte_lpm6 *lpm6;
 162         uint8_t portid;
 163 };
 164
 165 #define MAX_RX_QUEUE_PER_LCORE 16
 166 #define MAX_TX_QUEUE_PER_PORT 16
 167 struct lcore_queue_conf {
 168         uint16_t n_rx_queue;
 169         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 170         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 171         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 172 } __rte_cache_aligned;
 173 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 174
 175 static const struct rte_eth_conf port_conf = {
 176         .rxmode = {
 177                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 178                 .split_hdr_size = 0,
 179                 .header_split   = 0, /**< Header Split disabled */
 180                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 181                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 182                 .jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
 183                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 184         },
 185         .txmode = {
 186                 .mq_mode = ETH_MQ_TX_NONE,
 187         },
 188 };
 189
 190 /*
 191  * IPv4 forwarding table
 192  */
 193 struct l3fwd_ipv4_route {
 194         uint32_t ip;
 195         uint8_t  depth;
 196         uint8_t  if_out;
 197 };
 198
 199 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 200                 {IPv4(100,10,0,0), 16, 0},
 201                 {IPv4(100,20,0,0), 16, 1},
 202                 {IPv4(100,30,0,0), 16, 2},
 203                 {IPv4(100,40,0,0), 16, 3},
 204                 {IPv4(100,50,0,0), 16, 4},
 205                 {IPv4(100,60,0,0), 16, 5},
 206                 {IPv4(100,70,0,0), 16, 6},
 207                 {IPv4(100,80,0,0), 16, 7},
 208 };
 209
 210 /*
 211  * IPv6 forwarding table
 212  */
 213
 214 struct l3fwd_ipv6_route {
 215         uint8_t ip[IPV6_ADDR_LEN];
 216         uint8_t depth;
 217         uint8_t if_out;
 218 };
 219
 220 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 221         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 222         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 223         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 224         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 225         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 226         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 227         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 228         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 229 };
 230
 231 #define LPM_MAX_RULES         1024
 232 #define LPM6_MAX_RULES         1024
 233 #define LPM6_NUMBER_TBL8S (1 << 16)
 234
 235 struct rte_lpm6_config lpm6_config = {
 236                 .max_rules = LPM6_MAX_RULES,
 237                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 238                 .flags = 0
 239 };
 240
 241 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
 242 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
 243 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 244 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 245
 246 /* Send burst of packets on an output interface */
 247 static inline int
 248 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint8_t port)
 249 {
 250         struct rte_mbuf **m_table;
 251         int ret;
 252         uint16_t queueid;
 253
 254         queueid = qconf->tx_queue_id[port];
 255         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 256
 257         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 258         if (unlikely(ret < n)) {
 259                 do {
 260                         rte_pktmbuf_free(m_table[ret]);
 261                 } while (++ret < n);
 262         }
 263
 264         return 0;
 265 }
 266
 267 static inline void
 268 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
 269                 uint8_t queueid, uint8_t port_in)
 270 {
 271         struct rx_queue *rxq;
 272         uint32_t i, len;
 273         uint8_t next_hop, port_out, ipv6;
 274         int32_t len2;
 275
 276         ipv6 = 0;
 277         rxq = &qconf->rx_queue_list[queueid];
 278
 279         /* by default, send everything back to the source port */
 280         port_out = port_in;
 281
 282         /* Remove the Ethernet header and trailer from the input packet */
 283         rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
 284
 285         /* Build transmission burst */
 286         len = qconf->tx_mbufs[port_out].len;
 287
 288         /* if this is an IPv4 packet */
 289         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 290                 struct ipv4_hdr *ip_hdr;
 291                 uint32_t ip_dst;
 292                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 293                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
 294                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 295
 296                 /* Find destination port */
 297                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 298                                 (enabled_port_mask & 1 << next_hop) != 0) {
 299                         port_out = next_hop;
 300
 301                         /* Build transmission burst for new port */
 302                         len = qconf->tx_mbufs[port_out].len;
 303                 }
 304
 305                 /* if we don't need to do any fragmentation */
 306                 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
 307                         qconf->tx_mbufs[port_out].m_table[len] = m;
 308                         len2 = 1;
 309                 } else {
 310                         len2 = rte_ipv4_fragment_packet(m,
 311                                 &qconf->tx_mbufs[port_out].m_table[len],
 312                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 313                                 IPV4_MTU_DEFAULT,
 314                                 rxq->direct_pool, rxq->indirect_pool);
 315
 316                         /* Free input packet */
 317                         rte_pktmbuf_free(m);
 318
 319                         /* If we fail to fragment the packet */
 320                         if (unlikely (len2 < 0))
 321                                 return;
 322                 }
 323         }
 324         /* if this is an IPv6 packet */
 325         else if (m->ol_flags & PKT_RX_IPV6_HDR) {
 326                 struct ipv6_hdr *ip_hdr;
 327
 328                 ipv6 = 1;
 329
 330                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 331                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
 332
 333                 /* Find destination port */
 334                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, &next_hop) == 0 &&
 335                                 (enabled_port_mask & 1 << next_hop) != 0) {
 336                         port_out = next_hop;
 337
 338                         /* Build transmission burst for new port */
 339                         len = qconf->tx_mbufs[port_out].len;
 340                 }
 341
 342                 /* if we don't need to do any fragmentation */
 343                 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
 344                         qconf->tx_mbufs[port_out].m_table[len] = m;
 345                         len2 = 1;
 346                 } else {
 347                         len2 = rte_ipv6_fragment_packet(m,
 348                                 &qconf->tx_mbufs[port_out].m_table[len],
 349                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 350                                 IPV6_MTU_DEFAULT,
 351                                 rxq->direct_pool, rxq->indirect_pool);
 352
 353                         /* Free input packet */
 354                         rte_pktmbuf_free(m);
 355
 356                         /* If we fail to fragment the packet */
 357                         if (unlikely (len2 < 0))
 358                                 return;
 359                 }
 360         }
 361         /* else, just forward the packet */
 362         else {
 363                 qconf->tx_mbufs[port_out].m_table[len] = m;
 364                 len2 = 1;
 365         }
 366
 367         for (i = len; i < len + len2; i ++) {
 368                 void *d_addr_bytes;
 369
 370                 m = qconf->tx_mbufs[port_out].m_table[i];
 371                 struct ether_hdr *eth_hdr = (struct ether_hdr *)
 372                         rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
 373                 if (eth_hdr == NULL) {
 374                         rte_panic("No headroom in mbuf.\n");
 375                 }
 376
 377                 m->l2_len = sizeof(struct ether_hdr);
 378
 379                 /* 02:00:00:00:00:xx */
 380                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 381                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
 382
 383                 /* src addr */
 384                 ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
 385                 if (ipv6)
 386                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 387                 else
 388                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 389         }
 390
 391         len += len2;
 392
 393         if (likely(len < MAX_PKT_BURST)) {
 394                 qconf->tx_mbufs[port_out].len = (uint16_t)len;
 395                 return;
 396         }
 397
 398         /* Transmit packets */
 399         send_burst(qconf, (uint16_t)len, port_out);
 400         qconf->tx_mbufs[port_out].len = 0;
 401 }
 402
 403 /* main processing loop */
 404 static int
 405 main_loop(__attribute__((unused)) void *dummy)
 406 {
 407         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 408         unsigned lcore_id;
 409         uint64_t prev_tsc, diff_tsc, cur_tsc;
 410         int i, j, nb_rx;
 411         uint8_t portid;
 412         struct lcore_queue_conf *qconf;
 413         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 414
 415         prev_tsc = 0;
 416
 417         lcore_id = rte_lcore_id();
 418         qconf = &lcore_queue_conf[lcore_id];
 419
 420         if (qconf->n_rx_queue == 0) {
 421                 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
 422                 return 0;
 423         }
 424
 425         RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
 426
 427         for (i = 0; i < qconf->n_rx_queue; i++) {
 428
 429                 portid = qconf->rx_queue_list[i].portid;
 430                 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
 431                                 (int) portid);
 432         }
 433
 434         while (1) {
 435
 436                 cur_tsc = rte_rdtsc();
 437
 438                 /*
 439                  * TX burst queue drain
 440                  */
 441                 diff_tsc = cur_tsc - prev_tsc;
 442                 if (unlikely(diff_tsc > drain_tsc)) {
 443
 444                         /*
 445                          * This could be optimized (use queueid instead of
 446                          * portid), but it is not called so often
 447                          */
 448                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 449                                 if (qconf->tx_mbufs[portid].len == 0)
 450                                         continue;
 451                                 send_burst(&lcore_queue_conf[lcore_id],
 452                                            qconf->tx_mbufs[portid].len,
 453                                            portid);
 454                                 qconf->tx_mbufs[portid].len = 0;
 455                         }
 456
 457                         prev_tsc = cur_tsc;
 458                 }
 459
 460                 /*
 461                  * Read packet from RX queues
 462                  */
 463                 for (i = 0; i < qconf->n_rx_queue; i++) {
 464
 465                         portid = qconf->rx_queue_list[i].portid;
 466                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 467                                                  MAX_PKT_BURST);
 468
 469                         /* Prefetch first packets */
 470                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 471                                 rte_prefetch0(rte_pktmbuf_mtod(
 472                                                 pkts_burst[j], void *));
 473                         }
 474
 475                         /* Prefetch and forward already prefetched packets */
 476                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 477                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 478                                                 j + PREFETCH_OFFSET], void *));
 479                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 480                         }
 481
 482                         /* Forward remaining prefetched packets */
 483                         for (; j < nb_rx; j++) {
 484                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 485                         }
 486                 }
 487         }
 488 }
 489
 490 /* display usage */
 491 static void
 492 print_usage(const char *prgname)
 493 {
 494         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
 495                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 496                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
 497                prgname);
 498 }
 499
 500 static int
 501 parse_portmask(const char *portmask)
 502 {
 503         char *end = NULL;
 504         unsigned long pm;
 505
 506         /* parse hexadecimal string */
 507         pm = strtoul(portmask, &end, 16);
 508         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 509                 return -1;
 510
 511         if (pm == 0)
 512                 return -1;
 513
 514         return pm;
 515 }
 516
 517 static int
 518 parse_nqueue(const char *q_arg)
 519 {
 520         char *end = NULL;
 521         unsigned long n;
 522
 523         /* parse hexadecimal string */
 524         n = strtoul(q_arg, &end, 10);
 525         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 526                 return -1;
 527         if (n == 0)
 528                 return -1;
 529         if (n >= MAX_RX_QUEUE_PER_LCORE)
 530                 return -1;
 531
 532         return n;
 533 }
 534
 535 /* Parse the argument given in the command line of the application */
 536 static int
 537 parse_args(int argc, char **argv)
 538 {
 539         int opt, ret;
 540         char **argvopt;
 541         int option_index;
 542         char *prgname = argv[0];
 543         static struct option lgopts[] = {
 544                 {NULL, 0, 0, 0}
 545         };
 546
 547         argvopt = argv;
 548
 549         while ((opt = getopt_long(argc, argvopt, "p:q:",
 550                                   lgopts, &option_index)) != EOF) {
 551
 552                 switch (opt) {
 553                 /* portmask */
 554                 case 'p':
 555                         enabled_port_mask = parse_portmask(optarg);
 556                         if (enabled_port_mask < 0) {
 557                                 printf("invalid portmask\n");
 558                                 print_usage(prgname);
 559                                 return -1;
 560                         }
 561                         break;
 562
 563                 /* nqueue */
 564                 case 'q':
 565                         rx_queue_per_lcore = parse_nqueue(optarg);
 566                         if (rx_queue_per_lcore < 0) {
 567                                 printf("invalid queue number\n");
 568                                 print_usage(prgname);
 569                                 return -1;
 570                         }
 571                         break;
 572
 573                 /* long options */
 574                 case 0:
 575                         print_usage(prgname);
 576                         return -1;
 577
 578                 default:
 579                         print_usage(prgname);
 580                         return -1;
 581                 }
 582         }
 583
 584         if (enabled_port_mask == 0) {
 585                 printf("portmask not specified\n");
 586                 print_usage(prgname);
 587                 return -1;
 588         }
 589
 590         if (optind >= 0)
 591                 argv[optind-1] = prgname;
 592
 593         ret = optind-1;
 594         optind = 0; /* reset getopt lib */
 595         return ret;
 596 }
 597
 598 static void
 599 print_ethaddr(const char *name, struct ether_addr *eth_addr)
 600 {
 601         char buf[ETHER_ADDR_FMT_SIZE];
 602         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 603         printf("%s%s", name, buf);
 604 }
 605
 606 /* Check the link status of all ports in up to 9s, and print them finally */
 607 static void
 608 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 609 {
 610 #define CHECK_INTERVAL 100 /* 100ms */
 611 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 612         uint8_t portid, count, all_ports_up, print_flag = 0;
 613         struct rte_eth_link link;
 614
 615         printf("\nChecking link status");
 616         fflush(stdout);
 617         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 618                 all_ports_up = 1;
 619                 for (portid = 0; portid < port_num; portid++) {
 620                         if ((port_mask & (1 << portid)) == 0)
 621                                 continue;
 622                         memset(&link, 0, sizeof(link));
 623                         rte_eth_link_get_nowait(portid, &link);
 624                         /* print link status if flag set */
 625                         if (print_flag == 1) {
 626                                 if (link.link_status)
 627                                         printf("Port %d Link Up - speed %u "
 628                                                 "Mbps - %s\n", (uint8_t)portid,
 629                                                 (unsigned)link.link_speed,
 630                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 631                                         ("full-duplex") : ("half-duplex\n"));
 632                                 else
 633                                         printf("Port %d Link Down\n",
 634                                                         (uint8_t)portid);
 635                                 continue;
 636                         }
 637                         /* clear all_ports_up flag if any link down */
 638                         if (link.link_status == 0) {
 639                                 all_ports_up = 0;
 640                                 break;
 641                         }
 642                 }
 643                 /* after finally printing all link status, get out */
 644                 if (print_flag == 1)
 645                         break;
 646
 647                 if (all_ports_up == 0) {
 648                         printf(".");
 649                         fflush(stdout);
 650                         rte_delay_ms(CHECK_INTERVAL);
 651                 }
 652
 653                 /* set the print_flag if all ports up or timeout */
 654                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 655                         print_flag = 1;
 656                         printf("\ndone\n");
 657                 }
 658         }
 659 }
 660
 661 static int
 662 init_routing_table(void)
 663 {
 664         struct rte_lpm *lpm;
 665         struct rte_lpm6 *lpm6;
 666         int socket, ret;
 667         unsigned i;
 668
 669         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 670                 if (socket_lpm[socket]) {
 671                         lpm = socket_lpm[socket];
 672                         /* populate the LPM table */
 673                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 674                                 ret = rte_lpm_add(lpm,
 675                                         l3fwd_ipv4_route_array[i].ip,
 676                                         l3fwd_ipv4_route_array[i].depth,
 677                                         l3fwd_ipv4_route_array[i].if_out);
 678
 679                                 if (ret < 0) {
 680                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 681                                                 "LPM table\n", i);
 682                                         return -1;
 683                                 }
 684
 685                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 686                                                 "/%d (port %d)\n",
 687                                         socket,
 688                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 689                                         l3fwd_ipv4_route_array[i].depth,
 690                                         l3fwd_ipv4_route_array[i].if_out);
 691                         }
 692                 }
 693
 694                 if (socket_lpm6[socket]) {
 695                         lpm6 = socket_lpm6[socket];
 696                         /* populate the LPM6 table */
 697                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 698                                 ret = rte_lpm6_add(lpm6,
 699                                         l3fwd_ipv6_route_array[i].ip,
 700                                         l3fwd_ipv6_route_array[i].depth,
 701                                         l3fwd_ipv6_route_array[i].if_out);
 702
 703                                 if (ret < 0) {
 704                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 705                                                 "LPM6 table\n", i);
 706                                         return -1;
 707                                 }
 708
 709                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 710                                                 "/%d (port %d)\n",
 711                                         socket,
 712                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 713                                         l3fwd_ipv6_route_array[i].depth,
 714                                         l3fwd_ipv6_route_array[i].if_out);
 715                         }
 716                 }
 717         }
 718         return 0;
 719 }
 720
 721 static int
 722 init_mem(void)
 723 {
 724         char buf[PATH_MAX];
 725         struct rte_mempool *mp;
 726         struct rte_lpm *lpm;
 727         struct rte_lpm6 *lpm6;
 728         int socket;
 729         unsigned lcore_id;
 730
 731         /* traverse through lcores and initialize structures on each socket */
 732
 733         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 734
 735                 if (rte_lcore_is_enabled(lcore_id) == 0)
 736                         continue;
 737
 738                 socket = rte_lcore_to_socket_id(lcore_id);
 739
 740                 if (socket == SOCKET_ID_ANY)
 741                         socket = 0;
 742
 743                 if (socket_direct_pool[socket] == NULL) {
 744                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 745                                         socket);
 746                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 747
 748                         mp = rte_mempool_create(buf, NB_MBUF,
 749                                                    MBUF_SIZE, 32,
 750                                                    sizeof(struct rte_pktmbuf_pool_private),
 751                                                    rte_pktmbuf_pool_init, NULL,
 752                                                    rte_pktmbuf_init, NULL,
 753                                                    socket, 0);
 754                         if (mp == NULL) {
 755                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 756                                 return -1;
 757                         }
 758                         socket_direct_pool[socket] = mp;
 759                 }
 760
 761                 if (socket_indirect_pool[socket] == NULL) {
 762                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 763                                         socket);
 764                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 765
 766                         mp = rte_mempool_create(buf, NB_MBUF,
 767                                                            sizeof(struct rte_mbuf), 32,
 768                                                            0,
 769                                                            NULL, NULL,
 770                                                            rte_pktmbuf_init, NULL,
 771                                                            socket, 0);
 772                         if (mp == NULL) {
 773                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 774                                 return -1;
 775                         }
 776                         socket_indirect_pool[socket] = mp;
 777                 }
 778
 779                 if (socket_lpm[socket] == NULL) {
 780                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 781                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 782
 783                         lpm = rte_lpm_create(buf, socket, LPM_MAX_RULES, 0);
 784                         if (lpm == NULL) {
 785                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 786                                 return -1;
 787                         }
 788                         socket_lpm[socket] = lpm;
 789                 }
 790
 791                 if (socket_lpm6[socket] == NULL) {
 792                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 793                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 794
 795                         lpm6 = rte_lpm6_create("IP_FRAG_LPM6", socket, &lpm6_config);
 796                         if (lpm6 == NULL) {
 797                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 798                                 return -1;
 799                         }
 800                         socket_lpm6[socket] = lpm6;
 801                 }
 802         }
 803
 804         return 0;
 805 }
 806
 807 int
 808 main(int argc, char **argv)
 809 {
 810         struct lcore_queue_conf *qconf;
 811         struct rte_eth_dev_info dev_info;
 812         struct rte_eth_txconf *txconf;
 813         struct rx_queue *rxq;
 814         int socket, ret;
 815         unsigned nb_ports;
 816         uint16_t queueid = 0;
 817         unsigned lcore_id = 0, rx_lcore_id = 0;
 818         uint32_t n_tx_queue, nb_lcores;
 819         uint8_t portid;
 820
 821         /* init EAL */
 822         ret = rte_eal_init(argc, argv);
 823         if (ret < 0)
 824                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 825         argc -= ret;
 826         argv += ret;
 827
 828         /* parse application arguments (after the EAL ones) */
 829         ret = parse_args(argc, argv);
 830         if (ret < 0)
 831                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 832
 833         nb_ports = rte_eth_dev_count();
 834         if (nb_ports > RTE_MAX_ETHPORTS)
 835                 nb_ports = RTE_MAX_ETHPORTS;
 836         else if (nb_ports == 0)
 837                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 838
 839         nb_lcores = rte_lcore_count();
 840
 841         /* initialize structures (mempools, lpm etc.) */
 842         if (init_mem() < 0)
 843                 rte_panic("Cannot initialize memory structures!\n");
 844
 845         /* check if portmask has non-existent ports */
 846         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 847                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 848
 849         /* initialize all ports */
 850         for (portid = 0; portid < nb_ports; portid++) {
 851                 /* skip ports that are not enabled */
 852                 if ((enabled_port_mask & (1 << portid)) == 0) {
 853                         printf("Skipping disabled port %d\n", portid);
 854                         continue;
 855                 }
 856
 857                 qconf = &lcore_queue_conf[rx_lcore_id];
 858
 859                 /* get the lcore_id for this port */
 860                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 861                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 862
 863                         rx_lcore_id ++;
 864                         if (rx_lcore_id >= RTE_MAX_LCORE)
 865                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 866
 867                         qconf = &lcore_queue_conf[rx_lcore_id];
 868                 }
 869
 870                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 871                 if (socket == SOCKET_ID_ANY)
 872                         socket = 0;
 873
 874                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 875                 rxq->portid = portid;
 876                 rxq->direct_pool = socket_direct_pool[socket];
 877                 rxq->indirect_pool = socket_indirect_pool[socket];
 878                 rxq->lpm = socket_lpm[socket];
 879                 rxq->lpm6 = socket_lpm6[socket];
 880                 qconf->n_rx_queue++;
 881
 882                 /* init port */
 883                 printf("Initializing port %d on lcore %u...", portid,
 884                        rx_lcore_id);
 885                 fflush(stdout);
 886
 887                 n_tx_queue = nb_lcores;
 888                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 889                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 890                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 891                                             &port_conf);
 892                 if (ret < 0) {
 893                         printf("\n");
 894                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 895                                 "err=%d, port=%d\n",
 896                                 ret, portid);
 897                 }
 898
 899                 /* init one RX queue */
 900                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 901                                              socket, NULL,
 902                                              socket_direct_pool[socket]);
 903                 if (ret < 0) {
 904                         printf("\n");
 905                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 906                                 "err=%d, port=%d\n",
 907                                 ret, portid);
 908                 }
 909
 910                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 911                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
 912                 printf("\n");
 913
 914                 /* init one TX queue per couple (lcore,port) */
 915                 queueid = 0;
 916                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 917                         if (rte_lcore_is_enabled(lcore_id) == 0)
 918                                 continue;
 919
 920                         socket = (int) rte_lcore_to_socket_id(lcore_id);
 921                         printf("txq=%u,%d ", lcore_id, queueid);
 922                         fflush(stdout);
 923
 924                         rte_eth_dev_info_get(portid, &dev_info);
 925                         txconf = &dev_info.default_txconf;
 926                         txconf->txq_flags = 0;
 927                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
 928                                                      socket, txconf);
 929                         if (ret < 0) {
 930                                 printf("\n");
 931                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
 932                                         "err=%d, port=%d\n", ret, portid);
 933                         }
 934
 935                         qconf = &lcore_queue_conf[lcore_id];
 936                         qconf->tx_queue_id[portid] = queueid;
 937                         queueid++;
 938                 }
 939
 940                 printf("\n");
 941         }
 942
 943         printf("\n");
 944
 945         /* start ports */
 946         for (portid = 0; portid < nb_ports; portid++) {
 947                 if ((enabled_port_mask & (1 << portid)) == 0) {
 948                         continue;
 949                 }
 950                 /* Start device */
 951                 ret = rte_eth_dev_start(portid);
 952                 if (ret < 0)
 953                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
 954                                 ret, portid);
 955
 956                 rte_eth_promiscuous_enable(portid);
 957         }
 958
 959         if (init_routing_table() < 0)
 960                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
 961
 962         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
 963
 964         /* launch per-lcore init on every lcore */
 965         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
 966         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 967                 if (rte_eal_wait_lcore(lcore_id) < 0)
 968                         return -1;
 969         }
 970
 971         return 0;
 972 }