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 #include "main.h"
  79
  80 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
  81
  82 #define MBUF_SIZE (2048 + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
  83
  84 /* allow max jumbo frame 9.5 KB */
  85 #define JUMBO_FRAME_MAX_SIZE    0x2600
  86
  87 #define ROUNDUP_DIV(a, b)       (((a) + (b) - 1) / (b))
  88
  89 /*
  90  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
  91  * This value includes the size of IPv6 header.
  92  */
  93 #define IPV4_MTU_DEFAULT        ETHER_MTU
  94 #define IPV6_MTU_DEFAULT        ETHER_MTU
  95
  96 /*
  97  * Default payload in bytes for the IPv6 packet.
  98  */
  99 #define IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct ipv4_hdr))
 100 #define IPV6_DEFAULT_PAYLOAD    (IPV6_MTU_DEFAULT - sizeof(struct ipv6_hdr))
 101
 102 /*
 103  * Max number of fragments per packet expected - defined by config file.
 104  */
 105 #define MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
 106
 107 #define NB_MBUF   8192
 108
 109 #define MAX_PKT_BURST   32
 110 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 111
 112 /* Configure how many packets ahead to prefetch, when reading packets */
 113 #define PREFETCH_OFFSET 3
 114
 115 /*
 116  * Configurable number of RX/TX ring descriptors
 117  */
 118 #define RTE_TEST_RX_DESC_DEFAULT 128
 119 #define RTE_TEST_TX_DESC_DEFAULT 512
 120 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 121 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 122
 123 /* ethernet addresses of ports */
 124 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 125
 126 #ifndef IPv4_BYTES
 127 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
 128 #define IPv4_BYTES(addr) \
 129                 (uint8_t) (((addr) >> 24) & 0xFF),\
 130                 (uint8_t) (((addr) >> 16) & 0xFF),\
 131                 (uint8_t) (((addr) >> 8) & 0xFF),\
 132                 (uint8_t) ((addr) & 0xFF)
 133 #endif
 134
 135 #ifndef IPv6_BYTES
 136 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 137                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 138 #define IPv6_BYTES(addr) \
 139         addr[0],  addr[1], addr[2],  addr[3], \
 140         addr[4],  addr[5], addr[6],  addr[7], \
 141         addr[8],  addr[9], addr[10], addr[11],\
 142         addr[12], addr[13],addr[14], addr[15]
 143 #endif
 144
 145 #define IPV6_ADDR_LEN 16
 146
 147 /* mask of enabled ports */
 148 static int enabled_port_mask = 0;
 149
 150 static int rx_queue_per_lcore = 1;
 151
 152 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
 153
 154 struct mbuf_table {
 155         uint16_t len;
 156         struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
 157 };
 158
 159 struct rx_queue {
 160         struct rte_mempool *direct_pool;
 161         struct rte_mempool *indirect_pool;
 162         struct rte_lpm *lpm;
 163         struct rte_lpm6 *lpm6;
 164         uint8_t portid;
 165 };
 166
 167 #define MAX_RX_QUEUE_PER_LCORE 16
 168 #define MAX_TX_QUEUE_PER_PORT 16
 169 struct lcore_queue_conf {
 170         uint16_t n_rx_queue;
 171         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 172         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 173         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 174 } __rte_cache_aligned;
 175 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 176
 177 static const struct rte_eth_conf port_conf = {
 178         .rxmode = {
 179                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 180                 .split_hdr_size = 0,
 181                 .header_split   = 0, /**< Header Split disabled */
 182                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 183                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 184                 .jumbo_frame    = 1, /**< Jumbo Frame Support enabled */
 185                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 186         },
 187         .txmode = {
 188                 .mq_mode = ETH_MQ_TX_NONE,
 189         },
 190 };
 191
 192 /*
 193  * IPv4 forwarding table
 194  */
 195 struct l3fwd_ipv4_route {
 196         uint32_t ip;
 197         uint8_t  depth;
 198         uint8_t  if_out;
 199 };
 200
 201 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 202                 {IPv4(100,10,0,0), 16, 0},
 203                 {IPv4(100,20,0,0), 16, 1},
 204                 {IPv4(100,30,0,0), 16, 2},
 205                 {IPv4(100,40,0,0), 16, 3},
 206                 {IPv4(100,50,0,0), 16, 4},
 207                 {IPv4(100,60,0,0), 16, 5},
 208                 {IPv4(100,70,0,0), 16, 6},
 209                 {IPv4(100,80,0,0), 16, 7},
 210 };
 211
 212 /*
 213  * IPv6 forwarding table
 214  */
 215
 216 struct l3fwd_ipv6_route {
 217         uint8_t ip[IPV6_ADDR_LEN];
 218         uint8_t depth;
 219         uint8_t if_out;
 220 };
 221
 222 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 223         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 224         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 225         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 226         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 227         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 228         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 229         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 230         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 231 };
 232
 233 #define LPM_MAX_RULES         1024
 234 #define LPM6_MAX_RULES         1024
 235 #define LPM6_NUMBER_TBL8S (1 << 16)
 236
 237 struct rte_lpm6_config lpm6_config = {
 238                 .max_rules = LPM6_MAX_RULES,
 239                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 240                 .flags = 0
 241 };
 242
 243 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
 244 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
 245 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 246 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 247
 248 /* Send burst of packets on an output interface */
 249 static inline int
 250 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint8_t port)
 251 {
 252         struct rte_mbuf **m_table;
 253         int ret;
 254         uint16_t queueid;
 255
 256         queueid = qconf->tx_queue_id[port];
 257         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 258
 259         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 260         if (unlikely(ret < n)) {
 261                 do {
 262                         rte_pktmbuf_free(m_table[ret]);
 263                 } while (++ret < n);
 264         }
 265
 266         return 0;
 267 }
 268
 269 static inline void
 270 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
 271                 uint8_t queueid, uint8_t port_in)
 272 {
 273         struct rx_queue *rxq;
 274         uint32_t i, len;
 275         uint8_t next_hop, port_out, ipv6;
 276         int32_t len2;
 277
 278         ipv6 = 0;
 279         rxq = &qconf->rx_queue_list[queueid];
 280
 281         /* by default, send everything back to the source port */
 282         port_out = port_in;
 283
 284         /* Remove the Ethernet header and trailer from the input packet */
 285         rte_pktmbuf_adj(m, (uint16_t)sizeof(struct ether_hdr));
 286
 287         /* Build transmission burst */
 288         len = qconf->tx_mbufs[port_out].len;
 289
 290         /* if this is an IPv4 packet */
 291         if (m->ol_flags & PKT_RX_IPV4_HDR) {
 292                 struct ipv4_hdr *ip_hdr;
 293                 uint32_t ip_dst;
 294                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 295                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv4_hdr *);
 296                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 297
 298                 /* Find destination port */
 299                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 300                                 (enabled_port_mask & 1 << next_hop) != 0) {
 301                         port_out = next_hop;
 302
 303                         /* Build transmission burst for new port */
 304                         len = qconf->tx_mbufs[port_out].len;
 305                 }
 306
 307                 /* if we don't need to do any fragmentation */
 308                 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
 309                         qconf->tx_mbufs[port_out].m_table[len] = m;
 310                         len2 = 1;
 311                 } else {
 312                         len2 = rte_ipv4_fragment_packet(m,
 313                                 &qconf->tx_mbufs[port_out].m_table[len],
 314                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 315                                 IPV4_MTU_DEFAULT,
 316                                 rxq->direct_pool, rxq->indirect_pool);
 317
 318                         /* Free input packet */
 319                         rte_pktmbuf_free(m);
 320
 321                         /* If we fail to fragment the packet */
 322                         if (unlikely (len2 < 0))
 323                                 return;
 324                 }
 325         }
 326         /* if this is an IPv6 packet */
 327         else if (m->ol_flags & PKT_RX_IPV6_HDR) {
 328                 struct ipv6_hdr *ip_hdr;
 329
 330                 ipv6 = 1;
 331
 332                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 333                 ip_hdr = rte_pktmbuf_mtod(m, struct ipv6_hdr *);
 334
 335                 /* Find destination port */
 336                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, &next_hop) == 0 &&
 337                                 (enabled_port_mask & 1 << next_hop) != 0) {
 338                         port_out = next_hop;
 339
 340                         /* Build transmission burst for new port */
 341                         len = qconf->tx_mbufs[port_out].len;
 342                 }
 343
 344                 /* if we don't need to do any fragmentation */
 345                 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
 346                         qconf->tx_mbufs[port_out].m_table[len] = m;
 347                         len2 = 1;
 348                 } else {
 349                         len2 = rte_ipv6_fragment_packet(m,
 350                                 &qconf->tx_mbufs[port_out].m_table[len],
 351                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 352                                 IPV6_MTU_DEFAULT,
 353                                 rxq->direct_pool, rxq->indirect_pool);
 354
 355                         /* Free input packet */
 356                         rte_pktmbuf_free(m);
 357
 358                         /* If we fail to fragment the packet */
 359                         if (unlikely (len2 < 0))
 360                                 return;
 361                 }
 362         }
 363         /* else, just forward the packet */
 364         else {
 365                 qconf->tx_mbufs[port_out].m_table[len] = m;
 366                 len2 = 1;
 367         }
 368
 369         for (i = len; i < len + len2; i ++) {
 370                 void *d_addr_bytes;
 371
 372                 m = qconf->tx_mbufs[port_out].m_table[i];
 373                 struct ether_hdr *eth_hdr = (struct ether_hdr *)
 374                         rte_pktmbuf_prepend(m, (uint16_t)sizeof(struct ether_hdr));
 375                 if (eth_hdr == NULL) {
 376                         rte_panic("No headroom in mbuf.\n");
 377                 }
 378
 379                 m->l2_len = sizeof(struct ether_hdr);
 380
 381                 /* 02:00:00:00:00:xx */
 382                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 383                 *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)port_out << 40);
 384
 385                 /* src addr */
 386                 ether_addr_copy(&ports_eth_addr[port_out], &eth_hdr->s_addr);
 387                 if (ipv6)
 388                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 389                 else
 390                         eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 391         }
 392
 393         len += len2;
 394
 395         if (likely(len < MAX_PKT_BURST)) {
 396                 qconf->tx_mbufs[port_out].len = (uint16_t)len;
 397                 return;
 398         }
 399
 400         /* Transmit packets */
 401         send_burst(qconf, (uint16_t)len, port_out);
 402         qconf->tx_mbufs[port_out].len = 0;
 403 }
 404
 405 /* main processing loop */
 406 static int
 407 main_loop(__attribute__((unused)) void *dummy)
 408 {
 409         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 410         unsigned lcore_id;
 411         uint64_t prev_tsc, diff_tsc, cur_tsc;
 412         int i, j, nb_rx;
 413         uint8_t portid;
 414         struct lcore_queue_conf *qconf;
 415         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 416
 417         prev_tsc = 0;
 418
 419         lcore_id = rte_lcore_id();
 420         qconf = &lcore_queue_conf[lcore_id];
 421
 422         if (qconf->n_rx_queue == 0) {
 423                 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
 424                 return 0;
 425         }
 426
 427         RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
 428
 429         for (i = 0; i < qconf->n_rx_queue; i++) {
 430
 431                 portid = qconf->rx_queue_list[i].portid;
 432                 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
 433                                 (int) portid);
 434         }
 435
 436         while (1) {
 437
 438                 cur_tsc = rte_rdtsc();
 439
 440                 /*
 441                  * TX burst queue drain
 442                  */
 443                 diff_tsc = cur_tsc - prev_tsc;
 444                 if (unlikely(diff_tsc > drain_tsc)) {
 445
 446                         /*
 447                          * This could be optimized (use queueid instead of
 448                          * portid), but it is not called so often
 449                          */
 450                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 451                                 if (qconf->tx_mbufs[portid].len == 0)
 452                                         continue;
 453                                 send_burst(&lcore_queue_conf[lcore_id],
 454                                            qconf->tx_mbufs[portid].len,
 455                                            portid);
 456                                 qconf->tx_mbufs[portid].len = 0;
 457                         }
 458
 459                         prev_tsc = cur_tsc;
 460                 }
 461
 462                 /*
 463                  * Read packet from RX queues
 464                  */
 465                 for (i = 0; i < qconf->n_rx_queue; i++) {
 466
 467                         portid = qconf->rx_queue_list[i].portid;
 468                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 469                                                  MAX_PKT_BURST);
 470
 471                         /* Prefetch first packets */
 472                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 473                                 rte_prefetch0(rte_pktmbuf_mtod(
 474                                                 pkts_burst[j], void *));
 475                         }
 476
 477                         /* Prefetch and forward already prefetched packets */
 478                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 479                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 480                                                 j + PREFETCH_OFFSET], void *));
 481                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 482                         }
 483
 484                         /* Forward remaining prefetched packets */
 485                         for (; j < nb_rx; j++) {
 486                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 487                         }
 488                 }
 489         }
 490 }
 491
 492 /* display usage */
 493 static void
 494 print_usage(const char *prgname)
 495 {
 496         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
 497                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 498                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
 499                prgname);
 500 }
 501
 502 static int
 503 parse_portmask(const char *portmask)
 504 {
 505         char *end = NULL;
 506         unsigned long pm;
 507
 508         /* parse hexadecimal string */
 509         pm = strtoul(portmask, &end, 16);
 510         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 511                 return -1;
 512
 513         if (pm == 0)
 514                 return -1;
 515
 516         return pm;
 517 }
 518
 519 static int
 520 parse_nqueue(const char *q_arg)
 521 {
 522         char *end = NULL;
 523         unsigned long n;
 524
 525         /* parse hexadecimal string */
 526         n = strtoul(q_arg, &end, 10);
 527         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 528                 return -1;
 529         if (n == 0)
 530                 return -1;
 531         if (n >= MAX_RX_QUEUE_PER_LCORE)
 532                 return -1;
 533
 534         return n;
 535 }
 536
 537 /* Parse the argument given in the command line of the application */
 538 static int
 539 parse_args(int argc, char **argv)
 540 {
 541         int opt, ret;
 542         char **argvopt;
 543         int option_index;
 544         char *prgname = argv[0];
 545         static struct option lgopts[] = {
 546                 {NULL, 0, 0, 0}
 547         };
 548
 549         argvopt = argv;
 550
 551         while ((opt = getopt_long(argc, argvopt, "p:q:",
 552                                   lgopts, &option_index)) != EOF) {
 553
 554                 switch (opt) {
 555                 /* portmask */
 556                 case 'p':
 557                         enabled_port_mask = parse_portmask(optarg);
 558                         if (enabled_port_mask < 0) {
 559                                 printf("invalid portmask\n");
 560                                 print_usage(prgname);
 561                                 return -1;
 562                         }
 563                         break;
 564
 565                 /* nqueue */
 566                 case 'q':
 567                         rx_queue_per_lcore = parse_nqueue(optarg);
 568                         if (rx_queue_per_lcore < 0) {
 569                                 printf("invalid queue number\n");
 570                                 print_usage(prgname);
 571                                 return -1;
 572                         }
 573                         break;
 574
 575                 /* long options */
 576                 case 0:
 577                         print_usage(prgname);
 578                         return -1;
 579
 580                 default:
 581                         print_usage(prgname);
 582                         return -1;
 583                 }
 584         }
 585
 586         if (enabled_port_mask == 0) {
 587                 printf("portmask not specified\n");
 588                 print_usage(prgname);
 589                 return -1;
 590         }
 591
 592         if (optind >= 0)
 593                 argv[optind-1] = prgname;
 594
 595         ret = optind-1;
 596         optind = 0; /* reset getopt lib */
 597         return ret;
 598 }
 599
 600 static void
 601 print_ethaddr(const char *name, struct ether_addr *eth_addr)
 602 {
 603         char buf[ETHER_ADDR_FMT_SIZE];
 604         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 605         printf("%s%s", name, buf);
 606 }
 607
 608 /* Check the link status of all ports in up to 9s, and print them finally */
 609 static void
 610 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 611 {
 612 #define CHECK_INTERVAL 100 /* 100ms */
 613 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 614         uint8_t portid, count, all_ports_up, print_flag = 0;
 615         struct rte_eth_link link;
 616
 617         printf("\nChecking link status");
 618         fflush(stdout);
 619         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 620                 all_ports_up = 1;
 621                 for (portid = 0; portid < port_num; portid++) {
 622                         if ((port_mask & (1 << portid)) == 0)
 623                                 continue;
 624                         memset(&link, 0, sizeof(link));
 625                         rte_eth_link_get_nowait(portid, &link);
 626                         /* print link status if flag set */
 627                         if (print_flag == 1) {
 628                                 if (link.link_status)
 629                                         printf("Port %d Link Up - speed %u "
 630                                                 "Mbps - %s\n", (uint8_t)portid,
 631                                                 (unsigned)link.link_speed,
 632                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 633                                         ("full-duplex") : ("half-duplex\n"));
 634                                 else
 635                                         printf("Port %d Link Down\n",
 636                                                         (uint8_t)portid);
 637                                 continue;
 638                         }
 639                         /* clear all_ports_up flag if any link down */
 640                         if (link.link_status == 0) {
 641                                 all_ports_up = 0;
 642                                 break;
 643                         }
 644                 }
 645                 /* after finally printing all link status, get out */
 646                 if (print_flag == 1)
 647                         break;
 648
 649                 if (all_ports_up == 0) {
 650                         printf(".");
 651                         fflush(stdout);
 652                         rte_delay_ms(CHECK_INTERVAL);
 653                 }
 654
 655                 /* set the print_flag if all ports up or timeout */
 656                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 657                         print_flag = 1;
 658                         printf("\ndone\n");
 659                 }
 660         }
 661 }
 662
 663 static int
 664 init_routing_table(void)
 665 {
 666         struct rte_lpm *lpm;
 667         struct rte_lpm6 *lpm6;
 668         int socket, ret;
 669         unsigned i;
 670
 671         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 672                 if (socket_lpm[socket]) {
 673                         lpm = socket_lpm[socket];
 674                         /* populate the LPM table */
 675                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 676                                 ret = rte_lpm_add(lpm,
 677                                         l3fwd_ipv4_route_array[i].ip,
 678                                         l3fwd_ipv4_route_array[i].depth,
 679                                         l3fwd_ipv4_route_array[i].if_out);
 680
 681                                 if (ret < 0) {
 682                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 683                                                 "LPM table\n", i);
 684                                         return -1;
 685                                 }
 686
 687                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 688                                                 "/%d (port %d)\n",
 689                                         socket,
 690                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 691                                         l3fwd_ipv4_route_array[i].depth,
 692                                         l3fwd_ipv4_route_array[i].if_out);
 693                         }
 694                 }
 695
 696                 if (socket_lpm6[socket]) {
 697                         lpm6 = socket_lpm6[socket];
 698                         /* populate the LPM6 table */
 699                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 700                                 ret = rte_lpm6_add(lpm6,
 701                                         l3fwd_ipv6_route_array[i].ip,
 702                                         l3fwd_ipv6_route_array[i].depth,
 703                                         l3fwd_ipv6_route_array[i].if_out);
 704
 705                                 if (ret < 0) {
 706                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 707                                                 "LPM6 table\n", i);
 708                                         return -1;
 709                                 }
 710
 711                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 712                                                 "/%d (port %d)\n",
 713                                         socket,
 714                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 715                                         l3fwd_ipv6_route_array[i].depth,
 716                                         l3fwd_ipv6_route_array[i].if_out);
 717                         }
 718                 }
 719         }
 720         return 0;
 721 }
 722
 723 static int
 724 init_mem(void)
 725 {
 726         char buf[PATH_MAX];
 727         struct rte_mempool *mp;
 728         struct rte_lpm *lpm;
 729         struct rte_lpm6 *lpm6;
 730         int socket;
 731         unsigned lcore_id;
 732
 733         /* traverse through lcores and initialize structures on each socket */
 734
 735         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 736
 737                 if (rte_lcore_is_enabled(lcore_id) == 0)
 738                         continue;
 739
 740                 socket = rte_lcore_to_socket_id(lcore_id);
 741
 742                 if (socket == SOCKET_ID_ANY)
 743                         socket = 0;
 744
 745                 if (socket_direct_pool[socket] == NULL) {
 746                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 747                                         socket);
 748                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 749
 750                         mp = rte_mempool_create(buf, NB_MBUF,
 751                                                    MBUF_SIZE, 32,
 752                                                    sizeof(struct rte_pktmbuf_pool_private),
 753                                                    rte_pktmbuf_pool_init, NULL,
 754                                                    rte_pktmbuf_init, NULL,
 755                                                    socket, 0);
 756                         if (mp == NULL) {
 757                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 758                                 return -1;
 759                         }
 760                         socket_direct_pool[socket] = mp;
 761                 }
 762
 763                 if (socket_indirect_pool[socket] == NULL) {
 764                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 765                                         socket);
 766                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 767
 768                         mp = rte_mempool_create(buf, NB_MBUF,
 769                                                            sizeof(struct rte_mbuf), 32,
 770                                                            0,
 771                                                            NULL, NULL,
 772                                                            rte_pktmbuf_init, NULL,
 773                                                            socket, 0);
 774                         if (mp == NULL) {
 775                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 776                                 return -1;
 777                         }
 778                         socket_indirect_pool[socket] = mp;
 779                 }
 780
 781                 if (socket_lpm[socket] == NULL) {
 782                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 783                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 784
 785                         lpm = rte_lpm_create(buf, socket, LPM_MAX_RULES, 0);
 786                         if (lpm == NULL) {
 787                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 788                                 return -1;
 789                         }
 790                         socket_lpm[socket] = lpm;
 791                 }
 792
 793                 if (socket_lpm6[socket] == NULL) {
 794                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 795                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 796
 797                         lpm6 = rte_lpm6_create("IP_FRAG_LPM6", socket, &lpm6_config);
 798                         if (lpm6 == NULL) {
 799                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 800                                 return -1;
 801                         }
 802                         socket_lpm6[socket] = lpm6;
 803                 }
 804         }
 805
 806         return 0;
 807 }
 808
 809 int
 810 MAIN(int argc, char **argv)
 811 {
 812         struct lcore_queue_conf *qconf;
 813         struct rte_eth_dev_info dev_info;
 814         struct rte_eth_txconf *txconf;
 815         struct rx_queue *rxq;
 816         int socket, ret;
 817         unsigned nb_ports;
 818         uint16_t queueid = 0;
 819         unsigned lcore_id = 0, rx_lcore_id = 0;
 820         uint32_t n_tx_queue, nb_lcores;
 821         uint8_t portid;
 822
 823         /* init EAL */
 824         ret = rte_eal_init(argc, argv);
 825         if (ret < 0)
 826                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 827         argc -= ret;
 828         argv += ret;
 829
 830         /* parse application arguments (after the EAL ones) */
 831         ret = parse_args(argc, argv);
 832         if (ret < 0)
 833                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 834
 835         nb_ports = rte_eth_dev_count();
 836         if (nb_ports > RTE_MAX_ETHPORTS)
 837                 nb_ports = RTE_MAX_ETHPORTS;
 838         else if (nb_ports == 0)
 839                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 840
 841         nb_lcores = rte_lcore_count();
 842
 843         /* initialize structures (mempools, lpm etc.) */
 844         if (init_mem() < 0)
 845                 rte_panic("Cannot initialize memory structures!\n");
 846
 847         /* check if portmask has non-existent ports */
 848         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 849                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 850
 851         /* initialize all ports */
 852         for (portid = 0; portid < nb_ports; portid++) {
 853                 /* skip ports that are not enabled */
 854                 if ((enabled_port_mask & (1 << portid)) == 0) {
 855                         printf("Skipping disabled port %d\n", portid);
 856                         continue;
 857                 }
 858
 859                 qconf = &lcore_queue_conf[rx_lcore_id];
 860
 861                 /* get the lcore_id for this port */
 862                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 863                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 864
 865                         rx_lcore_id ++;
 866                         if (rx_lcore_id >= RTE_MAX_LCORE)
 867                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 868
 869                         qconf = &lcore_queue_conf[rx_lcore_id];
 870                 }
 871
 872                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 873                 if (socket == SOCKET_ID_ANY)
 874                         socket = 0;
 875
 876                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 877                 rxq->portid = portid;
 878                 rxq->direct_pool = socket_direct_pool[socket];
 879                 rxq->indirect_pool = socket_indirect_pool[socket];
 880                 rxq->lpm = socket_lpm[socket];
 881                 rxq->lpm6 = socket_lpm6[socket];
 882                 qconf->n_rx_queue++;
 883
 884                 /* init port */
 885                 printf("Initializing port %d on lcore %u...", portid,
 886                        rx_lcore_id);
 887                 fflush(stdout);
 888
 889                 n_tx_queue = nb_lcores;
 890                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 891                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 892                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 893                                             &port_conf);
 894                 if (ret < 0) {
 895                         printf("\n");
 896                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 897                                 "err=%d, port=%d\n",
 898                                 ret, portid);
 899                 }
 900
 901                 /* init one RX queue */
 902                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 903                                              socket, NULL,
 904                                              socket_direct_pool[socket]);
 905                 if (ret < 0) {
 906                         printf("\n");
 907                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 908                                 "err=%d, port=%d\n",
 909                                 ret, portid);
 910                 }
 911
 912                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 913                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
 914                 printf("\n");
 915
 916                 /* init one TX queue per couple (lcore,port) */
 917                 queueid = 0;
 918                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 919                         if (rte_lcore_is_enabled(lcore_id) == 0)
 920                                 continue;
 921
 922                         socket = (int) rte_lcore_to_socket_id(lcore_id);
 923                         printf("txq=%u,%d ", lcore_id, queueid);
 924                         fflush(stdout);
 925
 926                         rte_eth_dev_info_get(portid, &dev_info);
 927                         txconf = &dev_info.default_txconf;
 928                         txconf->txq_flags = 0;
 929                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
 930                                                      socket, txconf);
 931                         if (ret < 0) {
 932                                 printf("\n");
 933                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
 934                                         "err=%d, port=%d\n", ret, portid);
 935                         }
 936
 937                         qconf = &lcore_queue_conf[lcore_id];
 938                         qconf->tx_queue_id[portid] = queueid;
 939                         queueid++;
 940                 }
 941
 942                 printf("\n");
 943         }
 944
 945         printf("\n");
 946
 947         /* start ports */
 948         for (portid = 0; portid < nb_ports; portid++) {
 949                 if ((enabled_port_mask & (1 << portid)) == 0) {
 950                         continue;
 951                 }
 952                 /* Start device */
 953                 ret = rte_eth_dev_start(portid);
 954                 if (ret < 0)
 955                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
 956                                 ret, portid);
 957
 958                 rte_eth_promiscuous_enable(portid);
 959         }
 960
 961         if (init_routing_table() < 0)
 962                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
 963
 964         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
 965
 966         /* launch per-lcore init on every lcore */
 967         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
 968         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
 969                 if (rte_eal_wait_lcore(lcore_id) < 0)
 970                         return -1;
 971         }
 972
 973         return 0;
 974 }