examples/ip_fragmentation/main.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <sys/param.h>
  11 #include <string.h>
  12 #include <sys/queue.h>
  13 #include <stdarg.h>
  14 #include <errno.h>
  15 #include <getopt.h>
  16
  17 #include <rte_common.h>
  18 #include <rte_byteorder.h>
  19 #include <rte_log.h>
  20 #include <rte_memory.h>
  21 #include <rte_memcpy.h>
  22 #include <rte_eal.h>
  23 #include <rte_launch.h>
  24 #include <rte_atomic.h>
  25 #include <rte_cycles.h>
  26 #include <rte_prefetch.h>
  27 #include <rte_lcore.h>
  28 #include <rte_per_lcore.h>
  29 #include <rte_branch_prediction.h>
  30 #include <rte_interrupts.h>
  31 #include <rte_random.h>
  32 #include <rte_debug.h>
  33 #include <rte_ether.h>
  34 #include <rte_ethdev.h>
  35 #include <rte_mempool.h>
  36 #include <rte_mbuf.h>
  37 #include <rte_lpm.h>
  38 #include <rte_lpm6.h>
  39 #include <rte_ip.h>
  40 #include <rte_string_fns.h>
  41
  42 #include <rte_ip_frag.h>
  43
  44 #define RTE_LOGTYPE_IP_FRAG RTE_LOGTYPE_USER1
  45
  46 /* allow max jumbo frame 9.5 KB */
  47 #define JUMBO_FRAME_MAX_SIZE    0x2600
  48
  49 #define ROUNDUP_DIV(a, b)       (((a) + (b) - 1) / (b))
  50
  51 /*
  52  * Default byte size for the IPv6 Maximum Transfer Unit (MTU).
  53  * This value includes the size of IPv6 header.
  54  */
  55 #define IPV4_MTU_DEFAULT        RTE_ETHER_MTU
  56 #define IPV6_MTU_DEFAULT        RTE_ETHER_MTU
  57
  58 /*
  59  * The overhead from max frame size to MTU.
  60  * We have to consider the max possible overhead.
  61  */
  62 #define MTU_OVERHEAD    \
  63         (RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN + \
  64                 2 * sizeof(struct rte_vlan_hdr))
  65
  66 /*
  67  * Default payload in bytes for the IPv6 packet.
  68  */
  69 #define IPV4_DEFAULT_PAYLOAD    (IPV4_MTU_DEFAULT - sizeof(struct rte_ipv4_hdr))
  70 #define IPV6_DEFAULT_PAYLOAD    (IPV6_MTU_DEFAULT - sizeof(struct rte_ipv6_hdr))
  71
  72 /*
  73  * Max number of fragments per packet expected - defined by config file.
  74  */
  75 #define MAX_PACKET_FRAG RTE_LIBRTE_IP_FRAG_MAX_FRAG
  76
  77 #define NB_MBUF   8192
  78
  79 #define MAX_PKT_BURST   32
  80 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  81
  82 /* Configure how many packets ahead to prefetch, when reading packets */
  83 #define PREFETCH_OFFSET 3
  84
  85 /*
  86  * Configurable number of RX/TX ring descriptors
  87  */
  88 #define RTE_TEST_RX_DESC_DEFAULT 1024
  89 #define RTE_TEST_TX_DESC_DEFAULT 1024
  90 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
  91 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
  92
  93 /* ethernet addresses of ports */
  94 static struct rte_ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
  95
  96 #ifndef IPv4_BYTES
  97 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
  98 #define IPv4_BYTES(addr) \
  99                 (uint8_t) (((addr) >> 24) & 0xFF),\
 100                 (uint8_t) (((addr) >> 16) & 0xFF),\
 101                 (uint8_t) (((addr) >> 8) & 0xFF),\
 102                 (uint8_t) ((addr) & 0xFF)
 103 #endif
 104
 105 #ifndef IPv6_BYTES
 106 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 107                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 108 #define IPv6_BYTES(addr) \
 109         addr[0],  addr[1], addr[2],  addr[3], \
 110         addr[4],  addr[5], addr[6],  addr[7], \
 111         addr[8],  addr[9], addr[10], addr[11],\
 112         addr[12], addr[13],addr[14], addr[15]
 113 #endif
 114
 115 #define IPV6_ADDR_LEN 16
 116
 117 /* mask of enabled ports */
 118 static int enabled_port_mask = 0;
 119
 120 static int rx_queue_per_lcore = 1;
 121
 122 #define MBUF_TABLE_SIZE  (2 * MAX(MAX_PKT_BURST, MAX_PACKET_FRAG))
 123
 124 struct mbuf_table {
 125         uint16_t len;
 126         struct rte_mbuf *m_table[MBUF_TABLE_SIZE];
 127 };
 128
 129 struct rx_queue {
 130         struct rte_mempool *direct_pool;
 131         struct rte_mempool *indirect_pool;
 132         struct rte_lpm *lpm;
 133         struct rte_lpm6 *lpm6;
 134         uint16_t portid;
 135 };
 136
 137 #define MAX_RX_QUEUE_PER_LCORE 16
 138 #define MAX_TX_QUEUE_PER_PORT 16
 139 struct lcore_queue_conf {
 140         uint16_t n_rx_queue;
 141         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 142         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 143         struct mbuf_table tx_mbufs[RTE_MAX_ETHPORTS];
 144 } __rte_cache_aligned;
 145 struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 146
 147 static struct rte_eth_conf port_conf = {
 148         .rxmode = {
 149                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 150                 .split_hdr_size = 0,
 151                 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
 152                              DEV_RX_OFFLOAD_SCATTER |
 153                              DEV_RX_OFFLOAD_JUMBO_FRAME),
 154         },
 155         .txmode = {
 156                 .mq_mode = ETH_MQ_TX_NONE,
 157                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
 158                              DEV_TX_OFFLOAD_MULTI_SEGS),
 159         },
 160 };
 161
 162 /*
 163  * IPv4 forwarding table
 164  */
 165 struct l3fwd_ipv4_route {
 166         uint32_t ip;
 167         uint8_t  depth;
 168         uint8_t  if_out;
 169 };
 170
 171 /* Default l3fwd_ipv4_route_array table. 8< */
 172 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 173                 {RTE_IPV4(100,10,0,0), 16, 0},
 174                 {RTE_IPV4(100,20,0,0), 16, 1},
 175                 {RTE_IPV4(100,30,0,0), 16, 2},
 176                 {RTE_IPV4(100,40,0,0), 16, 3},
 177                 {RTE_IPV4(100,50,0,0), 16, 4},
 178                 {RTE_IPV4(100,60,0,0), 16, 5},
 179                 {RTE_IPV4(100,70,0,0), 16, 6},
 180                 {RTE_IPV4(100,80,0,0), 16, 7},
 181 };
 182 /* >8 End of default l3fwd_ipv4_route_array table */
 183
 184 /*
 185  * IPv6 forwarding table
 186  */
 187
 188 struct l3fwd_ipv6_route {
 189         uint8_t ip[IPV6_ADDR_LEN];
 190         uint8_t depth;
 191         uint8_t if_out;
 192 };
 193
 194 /* Default l3fwd_ipv6_route_array table. 8< */
 195 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 196         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 197         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 198         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 199         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 200         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 201         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 202         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 203         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 204 };
 205 /* >8 End of default l3fwd_ipv6_route_array table. */
 206
 207 #define LPM_MAX_RULES         1024
 208 #define LPM6_MAX_RULES         1024
 209 #define LPM6_NUMBER_TBL8S (1 << 16)
 210
 211 struct rte_lpm6_config lpm6_config = {
 212                 .max_rules = LPM6_MAX_RULES,
 213                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 214                 .flags = 0
 215 };
 216
 217 static struct rte_mempool *socket_direct_pool[RTE_MAX_NUMA_NODES];
 218 static struct rte_mempool *socket_indirect_pool[RTE_MAX_NUMA_NODES];
 219 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 220 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 221
 222 /* Send burst of packets on an output interface */
 223 static inline int
 224 send_burst(struct lcore_queue_conf *qconf, uint16_t n, uint16_t port)
 225 {
 226         struct rte_mbuf **m_table;
 227         int ret;
 228         uint16_t queueid;
 229
 230         queueid = qconf->tx_queue_id[port];
 231         m_table = (struct rte_mbuf **)qconf->tx_mbufs[port].m_table;
 232
 233         ret = rte_eth_tx_burst(port, queueid, m_table, n);
 234         if (unlikely(ret < n)) {
 235                 do {
 236                         rte_pktmbuf_free(m_table[ret]);
 237                 } while (++ret < n);
 238         }
 239
 240         return 0;
 241 }
 242
 243 static inline void
 244 l3fwd_simple_forward(struct rte_mbuf *m, struct lcore_queue_conf *qconf,
 245                 uint8_t queueid, uint16_t port_in)
 246 {
 247         struct rx_queue *rxq;
 248         uint32_t i, len, next_hop;
 249         uint16_t port_out, ether_type;
 250         int32_t len2;
 251         uint64_t ol_flags;
 252         const struct rte_ether_hdr *eth;
 253
 254         ol_flags = 0;
 255         rxq = &qconf->rx_queue_list[queueid];
 256
 257         /* by default, send everything back to the source port */
 258         port_out = port_in;
 259
 260         /* save ether type of the incoming packet */
 261         eth = rte_pktmbuf_mtod(m, const struct rte_ether_hdr *);
 262         ether_type = eth->ether_type;
 263
 264         /* Remove the Ethernet header and trailer from the input packet */
 265         rte_pktmbuf_adj(m, (uint16_t)sizeof(struct rte_ether_hdr));
 266
 267         /* Build transmission burst */
 268         len = qconf->tx_mbufs[port_out].len;
 269
 270         /* if this is an IPv4 packet */
 271         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 272                 struct rte_ipv4_hdr *ip_hdr;
 273                 uint32_t ip_dst;
 274                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 275                 ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv4_hdr *);
 276                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 277
 278                 /* Find destination port */
 279                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 280                                 (enabled_port_mask & 1 << next_hop) != 0) {
 281                         port_out = next_hop;
 282
 283                         /* Build transmission burst for new port */
 284                         len = qconf->tx_mbufs[port_out].len;
 285                 }
 286
 287                 /* if we don't need to do any fragmentation */
 288                 if (likely (IPV4_MTU_DEFAULT >= m->pkt_len)) {
 289                         qconf->tx_mbufs[port_out].m_table[len] = m;
 290                         len2 = 1;
 291                 } else {
 292                         len2 = rte_ipv4_fragment_packet(m,
 293                                 &qconf->tx_mbufs[port_out].m_table[len],
 294                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 295                                 IPV4_MTU_DEFAULT,
 296                                 rxq->direct_pool, rxq->indirect_pool);
 297
 298                         /* Free input packet */
 299                         rte_pktmbuf_free(m);
 300
 301                         /* request HW to regenerate IPv4 cksum */
 302                         ol_flags |= (PKT_TX_IPV4 | PKT_TX_IP_CKSUM);
 303
 304                         /* If we fail to fragment the packet */
 305                         if (unlikely (len2 < 0))
 306                                 return;
 307                 }
 308         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 309                 /* if this is an IPv6 packet */
 310                 struct rte_ipv6_hdr *ip_hdr;
 311
 312                 /* Read the lookup key (i.e. ip_dst) from the input packet */
 313                 ip_hdr = rte_pktmbuf_mtod(m, struct rte_ipv6_hdr *);
 314
 315                 /* Find destination port */
 316                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
 317                                                 &next_hop) == 0 &&
 318                                 (enabled_port_mask & 1 << next_hop) != 0) {
 319                         port_out = next_hop;
 320
 321                         /* Build transmission burst for new port */
 322                         len = qconf->tx_mbufs[port_out].len;
 323                 }
 324
 325                 /* if we don't need to do any fragmentation */
 326                 if (likely (IPV6_MTU_DEFAULT >= m->pkt_len)) {
 327                         qconf->tx_mbufs[port_out].m_table[len] = m;
 328                         len2 = 1;
 329                 } else {
 330                         len2 = rte_ipv6_fragment_packet(m,
 331                                 &qconf->tx_mbufs[port_out].m_table[len],
 332                                 (uint16_t)(MBUF_TABLE_SIZE - len),
 333                                 IPV6_MTU_DEFAULT,
 334                                 rxq->direct_pool, rxq->indirect_pool);
 335
 336                         /* Free input packet */
 337                         rte_pktmbuf_free(m);
 338
 339                         /* If we fail to fragment the packet */
 340                         if (unlikely (len2 < 0))
 341                                 return;
 342                 }
 343         }
 344         /* else, just forward the packet */
 345         else {
 346                 qconf->tx_mbufs[port_out].m_table[len] = m;
 347                 len2 = 1;
 348         }
 349
 350         for (i = len; i < len + len2; i ++) {
 351                 void *d_addr_bytes;
 352
 353                 m = qconf->tx_mbufs[port_out].m_table[i];
 354                 struct rte_ether_hdr *eth_hdr = (struct rte_ether_hdr *)
 355                         rte_pktmbuf_prepend(m,
 356                                 (uint16_t)sizeof(struct rte_ether_hdr));
 357                 if (eth_hdr == NULL) {
 358                         rte_panic("No headroom in mbuf.\n");
 359                 }
 360
 361                 m->ol_flags |= ol_flags;
 362                 m->l2_len = sizeof(struct rte_ether_hdr);
 363
 364                 /* 02:00:00:00:00:xx */
 365                 d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 366                 *((uint64_t *)d_addr_bytes) = 0x000000000002 +
 367                         ((uint64_t)port_out << 40);
 368
 369                 /* src addr */
 370                 rte_ether_addr_copy(&ports_eth_addr[port_out],
 371                                 &eth_hdr->s_addr);
 372                 eth_hdr->ether_type = ether_type;
 373         }
 374
 375         len += len2;
 376
 377         if (likely(len < MAX_PKT_BURST)) {
 378                 qconf->tx_mbufs[port_out].len = (uint16_t)len;
 379                 return;
 380         }
 381
 382         /* Transmit packets */
 383         send_burst(qconf, (uint16_t)len, port_out);
 384         qconf->tx_mbufs[port_out].len = 0;
 385 }
 386
 387 /* main processing loop */
 388 static int
 389 main_loop(__rte_unused void *dummy)
 390 {
 391         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 392         unsigned lcore_id;
 393         uint64_t prev_tsc, diff_tsc, cur_tsc;
 394         int i, j, nb_rx;
 395         uint16_t portid;
 396         struct lcore_queue_conf *qconf;
 397         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 398
 399         prev_tsc = 0;
 400
 401         lcore_id = rte_lcore_id();
 402         qconf = &lcore_queue_conf[lcore_id];
 403
 404         if (qconf->n_rx_queue == 0) {
 405                 RTE_LOG(INFO, IP_FRAG, "lcore %u has nothing to do\n", lcore_id);
 406                 return 0;
 407         }
 408
 409         RTE_LOG(INFO, IP_FRAG, "entering main loop on lcore %u\n", lcore_id);
 410
 411         for (i = 0; i < qconf->n_rx_queue; i++) {
 412
 413                 portid = qconf->rx_queue_list[i].portid;
 414                 RTE_LOG(INFO, IP_FRAG, " -- lcoreid=%u portid=%d\n", lcore_id,
 415                                 portid);
 416         }
 417
 418         while (1) {
 419
 420                 cur_tsc = rte_rdtsc();
 421
 422                 /*
 423                  * TX burst queue drain
 424                  */
 425                 diff_tsc = cur_tsc - prev_tsc;
 426                 if (unlikely(diff_tsc > drain_tsc)) {
 427
 428                         /*
 429                          * This could be optimized (use queueid instead of
 430                          * portid), but it is not called so often
 431                          */
 432                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 433                                 if (qconf->tx_mbufs[portid].len == 0)
 434                                         continue;
 435                                 send_burst(&lcore_queue_conf[lcore_id],
 436                                            qconf->tx_mbufs[portid].len,
 437                                            portid);
 438                                 qconf->tx_mbufs[portid].len = 0;
 439                         }
 440
 441                         prev_tsc = cur_tsc;
 442                 }
 443
 444                 /*
 445                  * Read packet from RX queues
 446                  */
 447                 for (i = 0; i < qconf->n_rx_queue; i++) {
 448
 449                         portid = qconf->rx_queue_list[i].portid;
 450                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 451                                                  MAX_PKT_BURST);
 452
 453                         /* Prefetch first packets */
 454                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 455                                 rte_prefetch0(rte_pktmbuf_mtod(
 456                                                 pkts_burst[j], void *));
 457                         }
 458
 459                         /* Prefetch and forward already prefetched packets */
 460                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 461                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 462                                                 j + PREFETCH_OFFSET], void *));
 463                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 464                         }
 465
 466                         /* Forward remaining prefetched packets */
 467                         for (; j < nb_rx; j++) {
 468                                 l3fwd_simple_forward(pkts_burst[j], qconf, i, portid);
 469                         }
 470                 }
 471         }
 472 }
 473
 474 /* display usage */
 475 static void
 476 print_usage(const char *prgname)
 477 {
 478         printf("%s [EAL options] -- -p PORTMASK [-q NQ]\n"
 479                "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 480                "  -q NQ: number of queue (=ports) per lcore (default is 1)\n",
 481                prgname);
 482 }
 483
 484 static int
 485 parse_portmask(const char *portmask)
 486 {
 487         char *end = NULL;
 488         unsigned long pm;
 489
 490         /* parse hexadecimal string */
 491         pm = strtoul(portmask, &end, 16);
 492         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 493                 return -1;
 494
 495         if (pm == 0)
 496                 return -1;
 497
 498         return pm;
 499 }
 500
 501 static int
 502 parse_nqueue(const char *q_arg)
 503 {
 504         char *end = NULL;
 505         unsigned long n;
 506
 507         /* parse hexadecimal string */
 508         n = strtoul(q_arg, &end, 10);
 509         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 510                 return -1;
 511         if (n == 0)
 512                 return -1;
 513         if (n >= MAX_RX_QUEUE_PER_LCORE)
 514                 return -1;
 515
 516         return n;
 517 }
 518
 519 /* Parse the argument given in the command line of the application */
 520 static int
 521 parse_args(int argc, char **argv)
 522 {
 523         int opt, ret;
 524         char **argvopt;
 525         int option_index;
 526         char *prgname = argv[0];
 527         static struct option lgopts[] = {
 528                 {NULL, 0, 0, 0}
 529         };
 530
 531         argvopt = argv;
 532
 533         while ((opt = getopt_long(argc, argvopt, "p:q:",
 534                                   lgopts, &option_index)) != EOF) {
 535
 536                 switch (opt) {
 537                 /* portmask */
 538                 case 'p':
 539                         enabled_port_mask = parse_portmask(optarg);
 540                         if (enabled_port_mask < 0) {
 541                                 printf("invalid portmask\n");
 542                                 print_usage(prgname);
 543                                 return -1;
 544                         }
 545                         break;
 546
 547                 /* nqueue */
 548                 case 'q':
 549                         rx_queue_per_lcore = parse_nqueue(optarg);
 550                         if (rx_queue_per_lcore < 0) {
 551                                 printf("invalid queue number\n");
 552                                 print_usage(prgname);
 553                                 return -1;
 554                         }
 555                         break;
 556
 557                 /* long options */
 558                 case 0:
 559                         print_usage(prgname);
 560                         return -1;
 561
 562                 default:
 563                         print_usage(prgname);
 564                         return -1;
 565                 }
 566         }
 567
 568         if (enabled_port_mask == 0) {
 569                 printf("portmask not specified\n");
 570                 print_usage(prgname);
 571                 return -1;
 572         }
 573
 574         if (optind >= 0)
 575                 argv[optind-1] = prgname;
 576
 577         ret = optind-1;
 578         optind = 1; /* reset getopt lib */
 579         return ret;
 580 }
 581
 582 static void
 583 print_ethaddr(const char *name, struct rte_ether_addr *eth_addr)
 584 {
 585         char buf[RTE_ETHER_ADDR_FMT_SIZE];
 586         rte_ether_format_addr(buf, RTE_ETHER_ADDR_FMT_SIZE, eth_addr);
 587         printf("%s%s", name, buf);
 588 }
 589
 590 /* Check the link status of all ports in up to 9s, and print them finally */
 591 static void
 592 check_all_ports_link_status(uint32_t port_mask)
 593 {
 594 #define CHECK_INTERVAL 100 /* 100ms */
 595 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 596         uint16_t portid;
 597         uint8_t count, all_ports_up, print_flag = 0;
 598         struct rte_eth_link link;
 599         int ret;
 600         char link_status_text[RTE_ETH_LINK_MAX_STR_LEN];
 601
 602         printf("\nChecking link status");
 603         fflush(stdout);
 604         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 605                 all_ports_up = 1;
 606                 RTE_ETH_FOREACH_DEV(portid) {
 607                         if ((port_mask & (1 << portid)) == 0)
 608                                 continue;
 609                         memset(&link, 0, sizeof(link));
 610                         ret = rte_eth_link_get_nowait(portid, &link);
 611                         if (ret < 0) {
 612                                 all_ports_up = 0;
 613                                 if (print_flag == 1)
 614                                         printf("Port %u link get failed: %s\n",
 615                                                 portid, rte_strerror(-ret));
 616                                 continue;
 617                         }
 618                         /* print link status if flag set */
 619                         if (print_flag == 1) {
 620                                 rte_eth_link_to_str(link_status_text,
 621                                         sizeof(link_status_text), &link);
 622                                 printf("Port %d %s\n", portid,
 623                                        link_status_text);
 624                                 continue;
 625                         }
 626                         /* clear all_ports_up flag if any link down */
 627                         if (link.link_status == ETH_LINK_DOWN) {
 628                                 all_ports_up = 0;
 629                                 break;
 630                         }
 631                 }
 632                 /* after finally printing all link status, get out */
 633                 if (print_flag == 1)
 634                         break;
 635
 636                 if (all_ports_up == 0) {
 637                         printf(".");
 638                         fflush(stdout);
 639                         rte_delay_ms(CHECK_INTERVAL);
 640                 }
 641
 642                 /* set the print_flag if all ports up or timeout */
 643                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 644                         print_flag = 1;
 645                         printf("\ndone\n");
 646                 }
 647         }
 648 }
 649
 650 /* Check L3 packet type detection capability of the NIC port */
 651 static int
 652 check_ptype(int portid)
 653 {
 654         int i, ret;
 655         int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
 656         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
 657
 658         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 659         if (ret <= 0)
 660                 return 0;
 661
 662         uint32_t ptypes[ret];
 663
 664         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 665         for (i = 0; i < ret; ++i) {
 666                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
 667                         ptype_l3_ipv4 = 1;
 668                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
 669                         ptype_l3_ipv6 = 1;
 670         }
 671
 672         if (ptype_l3_ipv4 == 0)
 673                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
 674
 675         if (ptype_l3_ipv6 == 0)
 676                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
 677
 678         if (ptype_l3_ipv4 && ptype_l3_ipv6)
 679                 return 1;
 680
 681         return 0;
 682
 683 }
 684
 685 /* Parse packet type of a packet by SW */
 686 static inline void
 687 parse_ptype(struct rte_mbuf *m)
 688 {
 689         struct rte_ether_hdr *eth_hdr;
 690         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 691         uint16_t ether_type;
 692
 693         eth_hdr = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
 694         ether_type = eth_hdr->ether_type;
 695         if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV4))
 696                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 697         else if (ether_type == rte_cpu_to_be_16(RTE_ETHER_TYPE_IPV6))
 698                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 699
 700         m->packet_type = packet_type;
 701 }
 702
 703 /* callback function to detect packet type for a queue of a port */
 704 static uint16_t
 705 cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 706                    struct rte_mbuf *pkts[], uint16_t nb_pkts,
 707                    uint16_t max_pkts __rte_unused,
 708                    void *user_param __rte_unused)
 709 {
 710         uint16_t i;
 711
 712         for (i = 0; i < nb_pkts; ++i)
 713                 parse_ptype(pkts[i]);
 714
 715         return nb_pkts;
 716 }
 717
 718 static int
 719 init_routing_table(void)
 720 {
 721         struct rte_lpm *lpm;
 722         struct rte_lpm6 *lpm6;
 723         int socket, ret;
 724         unsigned i;
 725
 726         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 727                 if (socket_lpm[socket]) {
 728                         lpm = socket_lpm[socket];
 729                         /* populate the LPM table */
 730                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 731                                 ret = rte_lpm_add(lpm,
 732                                         l3fwd_ipv4_route_array[i].ip,
 733                                         l3fwd_ipv4_route_array[i].depth,
 734                                         l3fwd_ipv4_route_array[i].if_out);
 735
 736                                 if (ret < 0) {
 737                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 738                                                 "LPM table\n", i);
 739                                         return -1;
 740                                 }
 741
 742                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv4_BYTES_FMT
 743                                                 "/%d (port %d)\n",
 744                                         socket,
 745                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 746                                         l3fwd_ipv4_route_array[i].depth,
 747                                         l3fwd_ipv4_route_array[i].if_out);
 748                         }
 749                 }
 750
 751                 if (socket_lpm6[socket]) {
 752                         lpm6 = socket_lpm6[socket];
 753                         /* populate the LPM6 table */
 754                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 755                                 ret = rte_lpm6_add(lpm6,
 756                                         l3fwd_ipv6_route_array[i].ip,
 757                                         l3fwd_ipv6_route_array[i].depth,
 758                                         l3fwd_ipv6_route_array[i].if_out);
 759
 760                                 if (ret < 0) {
 761                                         RTE_LOG(ERR, IP_FRAG, "Unable to add entry %i to the l3fwd "
 762                                                 "LPM6 table\n", i);
 763                                         return -1;
 764                                 }
 765
 766                                 RTE_LOG(INFO, IP_FRAG, "Socket %i: adding route " IPv6_BYTES_FMT
 767                                                 "/%d (port %d)\n",
 768                                         socket,
 769                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 770                                         l3fwd_ipv6_route_array[i].depth,
 771                                         l3fwd_ipv6_route_array[i].if_out);
 772                         }
 773                 }
 774         }
 775         return 0;
 776 }
 777
 778 static int
 779 init_mem(void)
 780 {
 781         char buf[PATH_MAX];
 782         struct rte_mempool *mp;
 783         struct rte_lpm *lpm;
 784         struct rte_lpm6 *lpm6;
 785         struct rte_lpm_config lpm_config;
 786         int socket;
 787         unsigned lcore_id;
 788
 789         /* traverse through lcores and initialize structures on each socket */
 790
 791         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 792
 793                 if (rte_lcore_is_enabled(lcore_id) == 0)
 794                         continue;
 795
 796                 socket = rte_lcore_to_socket_id(lcore_id);
 797
 798                 if (socket == SOCKET_ID_ANY)
 799                         socket = 0;
 800
 801                 if (socket_direct_pool[socket] == NULL) {
 802                         RTE_LOG(INFO, IP_FRAG, "Creating direct mempool on socket %i\n",
 803                                         socket);
 804                         snprintf(buf, sizeof(buf), "pool_direct_%i", socket);
 805
 806                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32,
 807                                 0, RTE_MBUF_DEFAULT_BUF_SIZE, socket);
 808                         if (mp == NULL) {
 809                                 RTE_LOG(ERR, IP_FRAG, "Cannot create direct mempool\n");
 810                                 return -1;
 811                         }
 812                         socket_direct_pool[socket] = mp;
 813                 }
 814
 815                 if (socket_indirect_pool[socket] == NULL) {
 816                         RTE_LOG(INFO, IP_FRAG, "Creating indirect mempool on socket %i\n",
 817                                         socket);
 818                         snprintf(buf, sizeof(buf), "pool_indirect_%i", socket);
 819
 820                         mp = rte_pktmbuf_pool_create(buf, NB_MBUF, 32, 0, 0,
 821                                 socket);
 822                         if (mp == NULL) {
 823                                 RTE_LOG(ERR, IP_FRAG, "Cannot create indirect mempool\n");
 824                                 return -1;
 825                         }
 826                         socket_indirect_pool[socket] = mp;
 827                 }
 828
 829                 if (socket_lpm[socket] == NULL) {
 830                         RTE_LOG(INFO, IP_FRAG, "Creating LPM table on socket %i\n", socket);
 831                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 832
 833                         lpm_config.max_rules = LPM_MAX_RULES;
 834                         lpm_config.number_tbl8s = 256;
 835                         lpm_config.flags = 0;
 836
 837                         lpm = rte_lpm_create(buf, socket, &lpm_config);
 838                         if (lpm == NULL) {
 839                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 840                                 return -1;
 841                         }
 842                         socket_lpm[socket] = lpm;
 843                 }
 844
 845                 if (socket_lpm6[socket] == NULL) {
 846                         RTE_LOG(INFO, IP_FRAG, "Creating LPM6 table on socket %i\n", socket);
 847                         snprintf(buf, sizeof(buf), "IP_FRAG_LPM_%i", socket);
 848
 849                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
 850                         if (lpm6 == NULL) {
 851                                 RTE_LOG(ERR, IP_FRAG, "Cannot create LPM table\n");
 852                                 return -1;
 853                         }
 854                         socket_lpm6[socket] = lpm6;
 855                 }
 856         }
 857
 858         return 0;
 859 }
 860
 861 int
 862 main(int argc, char **argv)
 863 {
 864         struct lcore_queue_conf *qconf;
 865         struct rte_eth_dev_info dev_info;
 866         struct rte_eth_txconf *txconf;
 867         struct rx_queue *rxq;
 868         int socket, ret;
 869         uint16_t nb_ports;
 870         uint16_t queueid = 0;
 871         unsigned lcore_id = 0, rx_lcore_id = 0;
 872         uint32_t n_tx_queue, nb_lcores;
 873         uint16_t portid;
 874
 875         /* init EAL */
 876         ret = rte_eal_init(argc, argv);
 877         if (ret < 0)
 878                 rte_exit(EXIT_FAILURE, "rte_eal_init failed");
 879         argc -= ret;
 880         argv += ret;
 881
 882         /* parse application arguments (after the EAL ones) */
 883         ret = parse_args(argc, argv);
 884         if (ret < 0)
 885                 rte_exit(EXIT_FAILURE, "Invalid arguments");
 886
 887         nb_ports = rte_eth_dev_count_avail();
 888         if (nb_ports == 0)
 889                 rte_exit(EXIT_FAILURE, "No ports found!\n");
 890
 891         nb_lcores = rte_lcore_count();
 892
 893         /* initialize structures (mempools, lpm etc.) */
 894         if (init_mem() < 0)
 895                 rte_panic("Cannot initialize memory structures!\n");
 896
 897         /* check if portmask has non-existent ports */
 898         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
 899                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
 900
 901         /* initialize all ports */
 902         RTE_ETH_FOREACH_DEV(portid) {
 903                 struct rte_eth_conf local_port_conf = port_conf;
 904                 struct rte_eth_rxconf rxq_conf;
 905
 906                 /* skip ports that are not enabled */
 907                 if ((enabled_port_mask & (1 << portid)) == 0) {
 908                         printf("Skipping disabled port %d\n", portid);
 909                         continue;
 910                 }
 911
 912                 qconf = &lcore_queue_conf[rx_lcore_id];
 913
 914                 /* limit the frame size to the maximum supported by NIC */
 915                 ret = rte_eth_dev_info_get(portid, &dev_info);
 916                 if (ret != 0)
 917                         rte_exit(EXIT_FAILURE,
 918                                 "Error during getting device (port %u) info: %s\n",
 919                                 portid, strerror(-ret));
 920
 921                 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
 922                     dev_info.max_rx_pktlen,
 923                     local_port_conf.rxmode.max_rx_pkt_len);
 924
 925                 /* get the lcore_id for this port */
 926                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
 927                        qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
 928
 929                         rx_lcore_id ++;
 930                         if (rx_lcore_id >= RTE_MAX_LCORE)
 931                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
 932
 933                         qconf = &lcore_queue_conf[rx_lcore_id];
 934                 }
 935
 936                 socket = (int) rte_lcore_to_socket_id(rx_lcore_id);
 937                 if (socket == SOCKET_ID_ANY)
 938                         socket = 0;
 939
 940                 rxq = &qconf->rx_queue_list[qconf->n_rx_queue];
 941                 rxq->portid = portid;
 942                 rxq->direct_pool = socket_direct_pool[socket];
 943                 rxq->indirect_pool = socket_indirect_pool[socket];
 944                 rxq->lpm = socket_lpm[socket];
 945                 rxq->lpm6 = socket_lpm6[socket];
 946                 qconf->n_rx_queue++;
 947
 948                 /* init port */
 949                 printf("Initializing port %d on lcore %u...", portid,
 950                        rx_lcore_id);
 951                 fflush(stdout);
 952
 953                 n_tx_queue = nb_lcores;
 954                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
 955                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
 956                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
 957                                             &local_port_conf);
 958                 if (ret < 0) {
 959                         printf("\n");
 960                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
 961                                 "err=%d, port=%d\n",
 962                                 ret, portid);
 963                 }
 964
 965                 /* set the mtu to the maximum received packet size */
 966                 ret = rte_eth_dev_set_mtu(portid,
 967                         local_port_conf.rxmode.max_rx_pkt_len - MTU_OVERHEAD);
 968                 if (ret < 0) {
 969                         printf("\n");
 970                         rte_exit(EXIT_FAILURE, "Set MTU failed: "
 971                                 "err=%d, port=%d\n",
 972                         ret, portid);
 973                 }
 974
 975                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
 976                                             &nb_txd);
 977                 if (ret < 0) {
 978                         printf("\n");
 979                         rte_exit(EXIT_FAILURE, "Cannot adjust number of "
 980                                 "descriptors: err=%d, port=%d\n", ret, portid);
 981                 }
 982
 983                 /* init one RX queue */
 984                 rxq_conf = dev_info.default_rxconf;
 985                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
 986                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
 987                                              socket, &rxq_conf,
 988                                              socket_direct_pool[socket]);
 989                 if (ret < 0) {
 990                         printf("\n");
 991                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
 992                                 "err=%d, port=%d\n",
 993                                 ret, portid);
 994                 }
 995
 996                 ret = rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
 997                 if (ret < 0) {
 998                         printf("\n");
 999                         rte_exit(EXIT_FAILURE,
1000                                 "rte_eth_macaddr_get: err=%d, port=%d\n",
1001                                 ret, portid);
1002                 }
1003
1004                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1005                 printf("\n");
1006
1007                 /* init one TX queue per couple (lcore,port) */
1008                 ret = rte_eth_dev_info_get(portid, &dev_info);
1009                 if (ret != 0)
1010                         rte_exit(EXIT_FAILURE,
1011                                 "Error during getting device (port %u) info: %s\n",
1012                                 portid, strerror(-ret));
1013
1014                 queueid = 0;
1015                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1016                         if (rte_lcore_is_enabled(lcore_id) == 0)
1017                                 continue;
1018
1019                         if (queueid >= dev_info.nb_tx_queues)
1020                                 break;
1021
1022                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1023                         printf("txq=%u,%d ", lcore_id, queueid);
1024                         fflush(stdout);
1025
1026                         txconf = &dev_info.default_txconf;
1027                         txconf->offloads = local_port_conf.txmode.offloads;
1028                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1029                                                      socket, txconf);
1030                         if (ret < 0) {
1031                                 printf("\n");
1032                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: "
1033                                         "err=%d, port=%d\n", ret, portid);
1034                         }
1035
1036                         qconf = &lcore_queue_conf[lcore_id];
1037                         qconf->tx_queue_id[portid] = queueid;
1038                         queueid++;
1039                 }
1040
1041                 printf("\n");
1042         }
1043
1044         printf("\n");
1045
1046         /* start ports */
1047         RTE_ETH_FOREACH_DEV(portid) {
1048                 if ((enabled_port_mask & (1 << portid)) == 0) {
1049                         continue;
1050                 }
1051                 /* Start device */
1052                 ret = rte_eth_dev_start(portid);
1053                 if (ret < 0)
1054                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1055                                 ret, portid);
1056
1057                 ret = rte_eth_promiscuous_enable(portid);
1058                 if (ret != 0)
1059                         rte_exit(EXIT_FAILURE,
1060                                 "rte_eth_promiscuous_enable: err=%s, port=%d\n",
1061                                 rte_strerror(-ret), portid);
1062
1063                 if (check_ptype(portid) == 0) {
1064                         rte_eth_add_rx_callback(portid, 0, cb_parse_ptype, NULL);
1065                         printf("Add Rx callback function to detect L3 packet type by SW :"
1066                                 " port = %d\n", portid);
1067                 }
1068         }
1069
1070         if (init_routing_table() < 0)
1071                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1072
1073         check_all_ports_link_status(enabled_port_mask);
1074
1075         /* launch per-lcore init on every lcore */
1076         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MAIN);
1077         RTE_LCORE_FOREACH_WORKER(lcore_id) {
1078                 if (rte_eal_wait_lcore(lcore_id) < 0)
1079                         return -1;
1080         }
1081
1082         /* clean up the EAL */
1083         rte_eal_cleanup();
1084
1085         return 0;
1086 }