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_launch.h>
  54 #include <rte_atomic.h>
  55 #include <rte_cycles.h>
  56 #include <rte_prefetch.h>
  57 #include <rte_lcore.h>
  58 #include <rte_per_lcore.h>
  59 #include <rte_branch_prediction.h>
  60 #include <rte_interrupts.h>
  61 #include <rte_pci.h>
  62 #include <rte_random.h>
  63 #include <rte_debug.h>
  64 #include <rte_ether.h>
  65 #include <rte_ethdev.h>
  66 #include <rte_mempool.h>
  67 #include <rte_mbuf.h>
  68 #include <rte_lpm.h>
  69 #include <rte_lpm6.h>
  70 #include <rte_ip.h>
  71 #include <rte_string_fns.h>
  72
  73 #include <rte_ip_frag.h>
  74
  75 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
  76
  77 /* allow max jumbo frame 9.5 KB */
  78 #define JUMBO_FRAME_MAX_SIZE    0x2600
  79
  80 #define ROUNDUP_DIV(a, b)       (((a) + (b) - 1) / (b))
  81
  82 /*
  83  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
  84  * This value includes the size of IPv6 header.
  85  */
  86 #define IPV4_MTU_DEFAULT        ETHER_MTU
  87 #define IPV6_MTU_DEFAULT        ETHER_MTU
  88
  89 /*
  90  * Default payload in bytes for the IPv6 packet.
  91  */
  92 #define IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
  93 #define IPV6_DEFAULT_PAYLOAD    (IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
  94
  95 /*
  96  * Max number of fragments per packet expected - defined by config file.
  97  */
  98 #define MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
  99
 100 #define NB_MBUF   8192
 101
 102 #define MAX_PKT_BURST   32
 103 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 104
 105 /* Configure how many packets ahead to prefetch, when reading packets */
 106 #define PREFETCH_OFFSET 3
 107
 108 /*
 109  * Configurable number of RX/TX ring descriptors
 110  */
 111 #define RTE_TEST_RX_DESC_DEFAULT 128
 112 #define RTE_TEST_TX_DESC_DEFAULT 512
 113 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 114 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 115
 116 /* ethernet addresses of ports */
 117 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 118
 119 #ifndef IPv4_BYTES
 120 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
 121 #define IPv4_BYTES(addr) \
 122                 (uint8_t) (((addr) >> 24) & 0xFF),\
 123                 (uint8_t) (((addr) >> 16) & 0xFF),\
 124                 (uint8_t) (((addr) >> 8) & 0xFF),\
 125                 (uint8_t) ((addr) & 0xFF)
 126 #endif
 127
 128 #ifndef IPv6_BYTES
 129 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 130                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 131 #define IPv6_BYTES(addr) \
 132         addr[0],  addr[1], addr[2],  addr[3], \
 133         addr[4],  addr[5], addr[6],  addr[7], \
 134         addr[8],  addr[9], addr[10], addr[11],\
 135         addr[12], addr[13],addr[14], addr[15]
 136 #endif
 137
 138 #define IPV6_ADDR_LEN 16
 139
 140 /* mask of enabled ports */
 141 static int enabled_port_mask = 0;
 142
 143 static int rx_queue_per_lcore = 1;
 144
 145 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
 146
 147 struct mbuf_table {
 148         uint16_t len;
 149         struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
 150 };
 151
 152 struct rx_queue {
 153         struct rte_mempool *direct_pool;
 154         struct rte_mempool *indirect_pool;
 155         struct rte_lpm *lpm;
 156         struct rte_lpm6 *lpm6;
 157         uint16_t portid;
 158 };
 159
 160 #define MAX_RX_QUEUE_PER_LCORE 16
 161 #define MAX_TX_QUEUE_PER_PORT 16
 162 struct lcore_queue_conf {
 163         uint16_t n_rx_queue;
 164         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 165         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 166         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 167 } __rte_cache_aligned;
 168 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 169
 170 static struct rte_eth_conf port_conf = {
 171         .rxmode = {
 172                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 173                 .split_hdr_size = 0,
 174                 .header_split   = 0, /**< Header Split disabled */
 175                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 176                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 177                 .jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
 178                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 179         },
 180         .txmode = {
 181                 .mq_mode = ETH_MQ_TX_NONE,
 182         },
 183 };
 184
 185 /*
 186  * IPv4 forwarding table
 187  */
 188 struct l3fwd_ipv4_route {
 189         uint32_t ip;
 190         uint8_t  depth;
 191         uint8_t  if_out;
 192 };
 193
 194 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 195                 {IPv4(100,10,0,0), 16, 0},
 196                 {IPv4(100,20,0,0), 16, 1},
 197                 {IPv4(100,30,0,0), 16, 2},
 198                 {IPv4(100,40,0,0), 16, 3},
 199                 {IPv4(100,50,0,0), 16, 4},
 200                 {IPv4(100,60,0,0), 16, 5},
 201                 {IPv4(100,70,0,0), 16, 6},
 202                 {IPv4(100,80,0,0), 16, 7},
 203 };
 204
 205 /*
 206  * IPv6 forwarding table
 207  */
 208
 209 struct l3fwd_ipv6_route {
 210         uint8_t ip[IPV6_ADDR_LEN];
 211         uint8_t depth;
 212         uint8_t if_out;
 213 };
 214
 215 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 216         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 217         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 218         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 219         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 220         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 221         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 222         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 223         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 224 };
 225
 226 #define LPM_MAX_RULES         1024
 227 #define LPM6_MAX_RULES         1024
 228 #define LPM6_NUMBER_TBL8S (1 << 16)
 229
 230 struct rte_lpm6_config lpm6_config = {
 231                 .max_rules = LPM6_MAX_RULES,
 232                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 233                 .flags = 0
 234 };
 235
 236 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
 237 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
 238 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 239 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 240
 241 /* Send burst of packets on an output interface */
 242 static inline int
 243 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
 244 {
 245         struct rte_mbuf **m_table;
 246         int ret;
 247         uint16_t queueid;
 248
 249         queueid = qconf->tx_queue_id[port];
 250         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 251
 252         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 253         if (unlikely(ret < n)) {
 254                 do {
 255                         rte_pktmbuf_free(m_table[ret]);
 256                 } while (++ret < n);
 257         }
 258
 259         return 0;
 260 }
 261
 262 static inline void
 263 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
 264                 uint8_t queueid, uint16_t port_in)
 265 {
 266         struct rx_queue *rxq;
 267         uint32_t i, len, next_hop;
 268         uint8_t ipv6;
 269         uint16_t port_out;
 270         int32_t len2;
 271
 272         ipv6 = 0;
 273         rxq = &qconf->rx_queue_list[queueid];
 274
 275         /* by default, send everything back to the source port */
 276         port_out = port_in;
 277
 278         /* Remove the Ethernet header and trailer from the input packet */
 279         rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
 280
 281         /* Build transmission burst */
 282         len = qconf->tx_mbufs[port_out].len;
 283
 284         /* if this is an IPv4 packet */
 285         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 286                 struct ipv4_hdr *ip_hdr;
 287                 uint32_t ip_dst;
 288                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 289                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
 290                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 291
 292                 /* Find destination port */
 293                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 294                                 (enabled_port_mask & 1 << next_hop) != 0) {
 295                         port_out = next_hop;
 296
 297                         /* Build transmission burst for new port */
 298                         len = qconf->tx_mbufs[port_out].len;
 299                 }
 300
 301                 /* if we don't need to do any fragmentation */
 302                 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
 303                         qconf->tx_mbufs[port_out].m_table[len] = m;
 304                         len2 = 1;
 305                 } else {
 306                         len2 = rte_ipv4_fragment_packet(m,
 307                                 &qconf->tx_mbufs[port_out].m_table[len],
 308                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 309                                 IPV4_MTU_DEFAULT,
 310                                 rxq->direct_pool, rxq->indirect_pool);
 311
 312                         /* Free input packet */
 313                         rte_pktmbuf_free(m);
 314
 315                         /* If we fail to fragment the packet */
 316                         if (unlikely (len2 < 0))
 317                                 return;
 318                 }
 319         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 320                 /* if this is an IPv6 packet */
 321                 struct ipv6_hdr *ip_hdr;
 322
 323                 ipv6 = 1;
 324
 325                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 326                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
 327
 328                 /* Find destination port */
 329                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
 330                                                 &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         uint16_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                                 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 = 1; /* 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(uint16_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         uint16_t portid;
 609         uint8_t 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(
 625                                         "Port%d Link Up .Speed %u Mbps - %s\n",
 626                                                 portid, 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", portid);
 631                                 continue;
 632                         }
 633                         /* clear all_ports_up flag if any link down */
 634                         if (link.link_status == ETH_LINK_DOWN) {
 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 /* Check L3 packet type detection capablity of the NIC port */
 658 static int
 659 check_ptype(int portid)
 660 {
 661         int i, ret;
 662         int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
 663         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
 664
 665         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 666         if (ret <= 0)
 667                 return 0;
 668
 669         uint32_t ptypes[ret];
 670
 671         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 672         for (i = 0; i < ret; ++i) {
 673                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
 674                         ptype_l3_ipv4 = 1;
 675                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
 676                         ptype_l3_ipv6 = 1;
 677         }
 678
 679         if (ptype_l3_ipv4 == 0)
 680                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
 681
 682         if (ptype_l3_ipv6 == 0)
 683                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
 684
 685         if (ptype_l3_ipv4 && ptype_l3_ipv6)
 686                 return 1;
 687
 688         return 0;
 689
 690 }
 691
 692 /* Parse packet type of a packet by SW */
 693 static inline void
 694 parse_ptype(struct rte_mbuf *m)
 695 {
 696         struct ether_hdr *eth_hdr;
 697         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 698         uint16_t ether_type;
 699
 700         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 701         ether_type = eth_hdr->ether_type;
 702         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 703                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 704         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 705                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 706
 707         m->packet_type = packet_type;
 708 }
 709
 710 /* callback function to detect packet type for a queue of a port */
 711 static uint16_t
 712 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 713                    struct rte_mbuf *pkts[], uint16_t nb_pkts,
 714                    uint16_t max_pkts __rte_unused,
 715                    void *user_param __rte_unused)
 716 {
 717         uint16_t i;
 718
 719         for (i = 0; i < nb_pkts; ++i)
 720                 parse_ptype(pkts[i]);
 721
 722         return nb_pkts;
 723 }
 724
 725 static int
 726 init_routing_table(void)
 727 {
 728         struct rte_lpm *lpm;
 729         struct rte_lpm6 *lpm6;
 730         int socket, ret;
 731         unsigned i;
 732
 733         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 734                 if (socket_lpm[socket]) {
 735                         lpm = socket_lpm[socket];
 736                         /* populate the LPM table */
 737                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 738                                 ret = rte_lpm_add(lpm,
 739                                         l3fwd_ipv4_route_array[i].ip,
 740                                         l3fwd_ipv4_route_array[i].depth,
 741                                         l3fwd_ipv4_route_array[i].if_out);
 742
 743                                 if (ret < 0) {
 744                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 745                                                 "LPM table\n", i);
 746                                         return -1;
 747                                 }
 748
 749                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 750                                                 "/%d (port %d)\n",
 751                                         socket,
 752                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 753                                         l3fwd_ipv4_route_array[i].depth,
 754                                         l3fwd_ipv4_route_array[i].if_out);
 755                         }
 756                 }
 757
 758                 if (socket_lpm6[socket]) {
 759                         lpm6 = socket_lpm6[socket];
 760                         /* populate the LPM6 table */
 761                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 762                                 ret = rte_lpm6_add(lpm6,
 763                                         l3fwd_ipv6_route_array[i].ip,
 764                                         l3fwd_ipv6_route_array[i].depth,
 765                                         l3fwd_ipv6_route_array[i].if_out);
 766
 767                                 if (ret < 0) {
 768                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 769                                                 "LPM6 table\n", i);
 770                                         return -1;
 771                                 }
 772
 773                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 774                                                 "/%d (port %d)\n",
 775                                         socket,
 776                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 777                                         l3fwd_ipv6_route_array[i].depth,
 778                                         l3fwd_ipv6_route_array[i].if_out);
 779                         }
 780                 }
 781         }
 782         return 0;
 783 }
 784
 785 static int
 786 init_mem(void)
 787 {
 788         char buf[PATH_MAX];
 789         struct rte_mempool *mp;
 790         struct rte_lpm *lpm;
 791         struct rte_lpm6 *lpm6;
 792         struct rte_lpm_config lpm_config;
 793         int socket;
 794         unsigned lcore_id;
 795
 796         /* traverse through lcores and initialize structures on each socket */
 797
 798         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 799
 800                 if (rte_lcore_is_enabled(lcore_id) == 0)
 801                         continue;
 802
 803                 socket = rte_lcore_to_socket_id(lcore_id);
 804
 805                 if (socket == SOCKET_ID_ANY)
 806                         socket = 0;
 807
 808                 if (socket_direct_pool[socket] == NULL) {
 809                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 810                                         socket);
 811                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 812
 813                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
 814                                 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
 815                         if (mp == NULL) {
 816                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 817                                 return -1;
 818                         }
 819                         socket_direct_pool[socket] = mp;
 820                 }
 821
 822                 if (socket_indirect_pool[socket] == NULL) {
 823                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 824                                         socket);
 825                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 826
 827                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
 828                                 socket);
 829                         if (mp == NULL) {
 830                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 831                                 return -1;
 832                         }
 833                         socket_indirect_pool[socket] = mp;
 834                 }
 835
 836                 if (socket_lpm[socket] == NULL) {
 837                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 838                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 839
 840                         lpm_config.max_rules = LPM_MAX_RULES;
 841                         lpm_config.number_tbl8s = 256;
 842                         lpm_config.flags = 0;
 843
 844                         lpm = rte_lpm_create(buf, socket, &lpm_config);
 845                         if (lpm == NULL) {
 846                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 847                                 return -1;
 848                         }
 849                         socket_lpm[socket] = lpm;
 850                 }
 851
 852                 if (socket_lpm6[socket] == NULL) {
 853                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 854                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 855
 856                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
 857                         if (lpm6 == NULL) {
 858                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 859                                 return -1;
 860                         }
 861                         socket_lpm6[socket] = lpm6;
 862                 }
 863         }
 864
 865         return 0;
 866 }
 867
 868 int
 869 main(int argc, char **argv)
 870 {
 871         struct lcore_queue_conf *qconf;
 872         struct rte_eth_dev_info dev_info;
 873         struct rte_eth_txconf *txconf;
 874         struct rx_queue *rxq;
 875         int socket, ret;
 876         unsigned nb_ports;
 877         uint16_t queueid = 0;
 878         unsigned lcore_id = 0, rx_lcore_id = 0;
 879         uint32_t n_tx_queue, nb_lcores;
 880         uint16_t portid;
 881
 882         /* init EAL */
 883         ret = rte_eal_init(argc, argv);
 884         if (ret < 0)
 885                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 886         argc -= ret;
 887         argv += ret;
 888
 889         /* parse application arguments (after the EAL ones) */
 890         ret = parse_args(argc, argv);
 891         if (ret < 0)
 892                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 893
 894         nb_ports = rte_eth_dev_count();
 895         if (nb_ports == 0)
 896                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 897
 898         nb_lcores = rte_lcore_count();
 899
 900         /* initialize structures (mempools, lpm etc.) */
 901         if (init_mem() < 0)
 902                 rte_panic("Cannot initialize memory structures!\n");
 903
 904         /* check if portmask has non-existent ports */
 905         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 906                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 907
 908         /* initialize all ports */
 909         for (portid = 0; portid < nb_ports; portid++) {
 910                 /* skip ports that are not enabled */
 911                 if ((enabled_port_mask & (1 << portid)) == 0) {
 912                         printf("Skipping disabled port %d\n", portid);
 913                         continue;
 914                 }
 915
 916                 qconf = &lcore_queue_conf[rx_lcore_id];
 917
 918                 /* limit the frame size to the maximum supported by NIC */
 919                 rte_eth_dev_info_get(portid, &dev_info);
 920                 port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
 921                     dev_info.max_rx_pktlen, port_conf.rxmode.max_rx_pkt_len);
 922
 923                 /* get the lcore_id for this port */
 924                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 925                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 926
 927                         rx_lcore_id ++;
 928                         if (rx_lcore_id >= RTE_MAX_LCORE)
 929                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 930
 931                         qconf = &lcore_queue_conf[rx_lcore_id];
 932                 }
 933
 934                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 935                 if (socket == SOCKET_ID_ANY)
 936                         socket = 0;
 937
 938                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 939                 rxq->portid = portid;
 940                 rxq->direct_pool = socket_direct_pool[socket];
 941                 rxq->indirect_pool = socket_indirect_pool[socket];
 942                 rxq->lpm = socket_lpm[socket];
 943                 rxq->lpm6 = socket_lpm6[socket];
 944                 qconf->n_rx_queue++;
 945
 946                 /* init port */
 947                 printf("Initializing port %d on lcore %u...", portid,
 948                        rx_lcore_id);
 949                 fflush(stdout);
 950
 951                 n_tx_queue = nb_lcores;
 952                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 953                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 954                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 955                                             &port_conf);
 956                 if (ret < 0) {
 957                         printf("\n");
 958                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 959                                 "err=%d, port=%d\n",
 960                                 ret, portid);
 961                 }
 962
 963                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
 964                                             &nb_txd);
 965                 if (ret < 0) {
 966                         printf("\n");
 967                         rte_exit(EXIT_FAILURE, "Cannot adjust number of "
 968                                 "descriptors: err=%d, port=%d\n", ret, portid);
 969                 }
 970
 971                 /* init one RX queue */
 972                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 973                                              socket, NULL,
 974                                              socket_direct_pool[socket]);
 975                 if (ret < 0) {
 976                         printf("\n");
 977                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 978                                 "err=%d, port=%d\n",
 979                                 ret, portid);
 980                 }
 981
 982                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 983                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
 984                 printf("\n");
 985
 986                 /* init one TX queue per couple (lcore,port) */
 987                 queueid = 0;
 988                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 989                         if (rte_lcore_is_enabled(lcore_id) == 0)
 990                                 continue;
 991
 992                         socket = (int) rte_lcore_to_socket_id(lcore_id);
 993                         printf("txq=%u,%d ", lcore_id, queueid);
 994                         fflush(stdout);
 995
 996                         txconf = &dev_info.default_txconf;
 997                         txconf->txq_flags = 0;
 998                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
 999                                                      socket, txconf);
1000                         if (ret < 0) {
1001                                 printf("\n");
1002                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1003                                         "err=%d, port=%d\n", ret, portid);
1004                         }
1005
1006                         qconf = &lcore_queue_conf[lcore_id];
1007                         qconf->tx_queue_id[portid] = queueid;
1008                         queueid++;
1009                 }
1010
1011                 printf("\n");
1012         }
1013
1014         printf("\n");
1015
1016         /* start ports */
1017         for (portid = 0; portid < nb_ports; portid++) {
1018                 if ((enabled_port_mask & (1 << portid)) == 0) {
1019                         continue;
1020                 }
1021                 /* Start device */
1022                 ret = rte_eth_dev_start(portid);
1023                 if (ret < 0)
1024                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1025                                 ret, portid);
1026
1027                 rte_eth_promiscuous_enable(portid);
1028
1029                 if (check_ptype(portid) == 0) {
1030                         rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1031                         printf("Add Rx callback function to detect L3 packet type by SW :"
1032                                 " port = %d\n", portid);
1033                 }
1034         }
1035
1036         if (init_routing_table() < 0)
1037                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1038
1039         check_all_ports_link_status(nb_ports, enabled_port_mask);
1040
1041         /* launch per-lcore init on every lcore */
1042         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1043         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1044                 if (rte_eal_wait_lcore(lcore_id) < 0)
1045                         return -1;
1046         }
1047
1048         return 0;
1049 }