examples/ip_reassembly/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 <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15 #include <signal.h>
  16 #include <sys/param.h>
  17
  18 #include <rte_common.h>
  19 #include <rte_byteorder.h>
  20 #include <rte_log.h>
  21 #include <rte_memory.h>
  22 #include <rte_memcpy.h>
  23 #include <rte_eal.h>
  24 #include <rte_launch.h>
  25 #include <rte_atomic.h>
  26 #include <rte_cycles.h>
  27 #include <rte_prefetch.h>
  28 #include <rte_lcore.h>
  29 #include <rte_per_lcore.h>
  30 #include <rte_branch_prediction.h>
  31 #include <rte_interrupts.h>
  32 #include <rte_random.h>
  33 #include <rte_debug.h>
  34 #include <rte_ether.h>
  35 #include <rte_ethdev.h>
  36 #include <rte_mempool.h>
  37 #include <rte_mbuf.h>
  38 #include <rte_malloc.h>
  39 #include <rte_ip.h>
  40 #include <rte_tcp.h>
  41 #include <rte_udp.h>
  42 #include <rte_string_fns.h>
  43 #include <rte_lpm.h>
  44 #include <rte_lpm6.h>
  45
  46 #include <rte_ip_frag.h>
  47
  48 #define MAX_PKT_BURST 32
  49
  50
  51 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
  52
  53 #define MAX_JUMBO_PKT_LEN  9600
  54
  55 #define BUF_SIZE        RTE_MBUF_DEFAULT_DATAROOM
  56 #define MBUF_DATA_SIZE  RTE_MBUF_DEFAULT_BUF_SIZE
  57
  58 #define NB_MBUF 8192
  59 #define MEMPOOL_CACHE_SIZE 256
  60
  61 /* allow max jumbo frame 9.5 KB */
  62 #define JUMBO_FRAME_MAX_SIZE    0x2600
  63
  64 #define MAX_FLOW_NUM    UINT16_MAX
  65 #define MIN_FLOW_NUM    1
  66 #define DEF_FLOW_NUM    0x1000
  67
  68 /* TTL numbers are in ms. */
  69 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
  70 #define MIN_FLOW_TTL    1
  71 #define DEF_FLOW_TTL    MS_PER_S
  72
  73 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
  74
  75 /* Should be power of two. */
  76 #define IP_FRAG_TBL_BUCKET_ENTRIES      16
  77
  78 static uint32_t max_flow_num = DEF_FLOW_NUM;
  79 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
  80
  81 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
  82
  83 #define NB_SOCKETS 8
  84
  85 /* Configure how many packets ahead to prefetch, when reading packets */
  86 #define PREFETCH_OFFSET 3
  87
  88 /*
  89  * Configurable number of RX/TX ring descriptors
  90  */
  91 #define RTE_TEST_RX_DESC_DEFAULT 1024
  92 #define RTE_TEST_TX_DESC_DEFAULT 1024
  93
  94 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
  95 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
  96
  97 /* ethernet addresses of ports */
  98 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
  99
 100 #ifndef IPv4_BYTES
 101 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
 102 #define IPv4_BYTES(addr) \
 103                 (uint8_t) (((addr) >> 24) & 0xFF),\
 104                 (uint8_t) (((addr) >> 16) & 0xFF),\
 105                 (uint8_t) (((addr) >> 8) & 0xFF),\
 106                 (uint8_t) ((addr) & 0xFF)
 107 #endif
 108
 109 #ifndef IPv6_BYTES
 110 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 111                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 112 #define IPv6_BYTES(addr) \
 113         addr[0],  addr[1], addr[2],  addr[3], \
 114         addr[4],  addr[5], addr[6],  addr[7], \
 115         addr[8],  addr[9], addr[10], addr[11],\
 116         addr[12], addr[13],addr[14], addr[15]
 117 #endif
 118
 119 #define IPV6_ADDR_LEN 16
 120
 121 /* mask of enabled ports */
 122 static uint32_t enabled_port_mask = 0;
 123
 124 static int rx_queue_per_lcore = 1;
 125
 126 struct mbuf_table {
 127         uint32_t len;
 128         uint32_t head;
 129         uint32_t tail;
 130         struct rte_mbuf *m_table[0];
 131 };
 132
 133 struct rx_queue {
 134         struct rte_ip_frag_tbl *frag_tbl;
 135         struct rte_mempool *pool;
 136         struct rte_lpm *lpm;
 137         struct rte_lpm6 *lpm6;
 138         uint16_t portid;
 139 };
 140
 141 struct tx_lcore_stat {
 142         uint64_t call;
 143         uint64_t drop;
 144         uint64_t queue;
 145         uint64_t send;
 146 };
 147
 148 #define MAX_RX_QUEUE_PER_LCORE 16
 149 #define MAX_TX_QUEUE_PER_PORT 16
 150 #define MAX_RX_QUEUE_PER_PORT 128
 151
 152 struct lcore_queue_conf {
 153         uint16_t n_rx_queue;
 154         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 155         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 156         struct rte_ip_frag_death_row death_row;
 157         struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
 158         struct tx_lcore_stat tx_stat;
 159 } __rte_cache_aligned;
 160 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 161
 162 static struct rte_eth_conf port_conf = {
 163         .rxmode = {
 164                 .mq_mode        = ETH_MQ_RX_RSS,
 165                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 166                 .split_hdr_size = 0,
 167                 .offloads = (DEV_RX_OFFLOAD_CHECKSUM |
 168                              DEV_RX_OFFLOAD_JUMBO_FRAME |
 169                              DEV_RX_OFFLOAD_CRC_STRIP),
 170         },
 171         .rx_adv_conf = {
 172                         .rss_conf = {
 173                                 .rss_key = NULL,
 174                                 .rss_hf = ETH_RSS_IP,
 175                 },
 176         },
 177         .txmode = {
 178                 .mq_mode = ETH_MQ_TX_NONE,
 179                 .offloads = (DEV_TX_OFFLOAD_IPV4_CKSUM |
 180                              DEV_TX_OFFLOAD_MULTI_SEGS),
 181         },
 182 };
 183
 184 /*
 185  * IPv4 forwarding table
 186  */
 187 struct l3fwd_ipv4_route {
 188         uint32_t ip;
 189         uint8_t  depth;
 190         uint8_t  if_out;
 191 };
 192
 193 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 194                 {IPv4(100,10,0,0), 16, 0},
 195                 {IPv4(100,20,0,0), 16, 1},
 196                 {IPv4(100,30,0,0), 16, 2},
 197                 {IPv4(100,40,0,0), 16, 3},
 198                 {IPv4(100,50,0,0), 16, 4},
 199                 {IPv4(100,60,0,0), 16, 5},
 200                 {IPv4(100,70,0,0), 16, 6},
 201                 {IPv4(100,80,0,0), 16, 7},
 202 };
 203
 204 /*
 205  * IPv6 forwarding table
 206  */
 207
 208 struct l3fwd_ipv6_route {
 209         uint8_t ip[IPV6_ADDR_LEN];
 210         uint8_t depth;
 211         uint8_t if_out;
 212 };
 213
 214 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 215         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 216         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 217         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 218         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 219         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 220         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 221         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 222         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 223 };
 224
 225 #define LPM_MAX_RULES         1024
 226 #define LPM6_MAX_RULES         1024
 227 #define LPM6_NUMBER_TBL8S (1 << 16)
 228
 229 struct rte_lpm6_config lpm6_config = {
 230                 .max_rules = LPM6_MAX_RULES,
 231                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 232                 .flags = 0
 233 };
 234
 235 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 236 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 237
 238 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
 239 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
 240 #else
 241 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
 242 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
 243
 244 /*
 245  * If number of queued packets reached given threahold, then
 246  * send burst of packets on an output interface.
 247  */
 248 static inline uint32_t
 249 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port)
 250 {
 251         uint32_t fill, len, k, n;
 252         struct mbuf_table *txmb;
 253
 254         txmb = qconf->tx_mbufs[port];
 255         len = txmb->len;
 256
 257         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
 258                 fill += len;
 259
 260         if (fill >= thresh) {
 261                 n = RTE_MIN(len - txmb->tail, fill);
 262
 263                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
 264                         txmb->m_table + txmb->tail, (uint16_t)n);
 265
 266                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
 267                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
 268
 269                 fill -= k;
 270                 if ((txmb->tail += k) == len)
 271                         txmb->tail = 0;
 272         }
 273
 274         return fill;
 275 }
 276
 277 /* Enqueue a single packet, and send burst if queue is filled */
 278 static inline int
 279 send_single_packet(struct rte_mbuf *m, uint16_t port)
 280 {
 281         uint32_t fill, lcore_id, len;
 282         struct lcore_queue_conf *qconf;
 283         struct mbuf_table *txmb;
 284
 285         lcore_id = rte_lcore_id();
 286         qconf = &lcore_queue_conf[lcore_id];
 287
 288         txmb = qconf->tx_mbufs[port];
 289         len = txmb->len;
 290
 291         fill = send_burst(qconf, MAX_PKT_BURST, port);
 292
 293         if (fill == len - 1) {
 294                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
 295                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
 296                 if (++txmb->tail == len)
 297                         txmb->tail = 0;
 298         }
 299
 300         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
 301         txmb->m_table[txmb->head] = m;
 302         if(++txmb->head == len)
 303                 txmb->head = 0;
 304
 305         return 0;
 306 }
 307
 308 static inline void
 309 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue,
 310         struct lcore_queue_conf *qconf, uint64_t tms)
 311 {
 312         struct ether_hdr *eth_hdr;
 313         struct rte_ip_frag_tbl *tbl;
 314         struct rte_ip_frag_death_row *dr;
 315         struct rx_queue *rxq;
 316         void *d_addr_bytes;
 317         uint32_t next_hop;
 318         uint16_t dst_port;
 319
 320         rxq = &qconf->rx_queue_list[queue];
 321
 322         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 323
 324         dst_port = portid;
 325
 326         /* if packet is IPv4 */
 327         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 328                 struct ipv4_hdr *ip_hdr;
 329                 uint32_t ip_dst;
 330
 331                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 332
 333                  /* if it is a fragmented packet, then try to reassemble. */
 334                 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
 335                         struct rte_mbuf *mo;
 336
 337                         tbl = rxq->frag_tbl;
 338                         dr = &qconf->death_row;
 339
 340                         /* prepare mbuf: setup l2_len/l3_len. */
 341                         m->l2_len = sizeof(*eth_hdr);
 342                         m->l3_len = sizeof(*ip_hdr);
 343
 344                         /* process this fragment. */
 345                         mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
 346                         if (mo == NULL)
 347                                 /* no packet to send out. */
 348                                 return;
 349
 350                         /* we have our packet reassembled. */
 351                         if (mo != m) {
 352                                 m = mo;
 353                                 eth_hdr = rte_pktmbuf_mtod(m,
 354                                         struct ether_hdr *);
 355                                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 356                         }
 357                 }
 358                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 359
 360                 /* Find destination port */
 361                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 362                                 (enabled_port_mask & 1 << next_hop) != 0) {
 363                         dst_port = next_hop;
 364                 }
 365
 366                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 367         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 368                 /* if packet is IPv6 */
 369                 struct ipv6_extension_fragment *frag_hdr;
 370                 struct ipv6_hdr *ip_hdr;
 371
 372                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 373
 374                 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
 375
 376                 if (frag_hdr != NULL) {
 377                         struct rte_mbuf *mo;
 378
 379                         tbl = rxq->frag_tbl;
 380                         dr  = &qconf->death_row;
 381
 382                         /* prepare mbuf: setup l2_len/l3_len. */
 383                         m->l2_len = sizeof(*eth_hdr);
 384                         m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
 385
 386                         mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
 387                         if (mo == NULL)
 388                                 return;
 389
 390                         if (mo != m) {
 391                                 m = mo;
 392                                 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 393                                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 394                         }
 395                 }
 396
 397                 /* Find destination port */
 398                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
 399                                                 &next_hop) == 0 &&
 400                                 (enabled_port_mask & 1 << next_hop) != 0) {
 401                         dst_port = next_hop;
 402                 }
 403
 404                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 405         }
 406         /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
 407
 408         /* 02:00:00:00:00:xx */
 409         d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 410         *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 411
 412         /* src addr */
 413         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 414
 415         send_single_packet(m, dst_port);
 416 }
 417
 418 /* main processing loop */
 419 static int
 420 main_loop(__attribute__((unused)) void *dummy)
 421 {
 422         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 423         unsigned lcore_id;
 424         uint64_t diff_tsc, cur_tsc, prev_tsc;
 425         int i, j, nb_rx;
 426         uint16_t portid;
 427         struct lcore_queue_conf *qconf;
 428         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 429
 430         prev_tsc = 0;
 431
 432         lcore_id = rte_lcore_id();
 433         qconf = &lcore_queue_conf[lcore_id];
 434
 435         if (qconf->n_rx_queue == 0) {
 436                 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
 437                 return 0;
 438         }
 439
 440         RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
 441
 442         for (i = 0; i < qconf->n_rx_queue; i++) {
 443
 444                 portid = qconf->rx_queue_list[i].portid;
 445                 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
 446                         portid);
 447         }
 448
 449         while (1) {
 450
 451                 cur_tsc = rte_rdtsc();
 452
 453                 /*
 454                  * TX burst queue drain
 455                  */
 456                 diff_tsc = cur_tsc - prev_tsc;
 457                 if (unlikely(diff_tsc > drain_tsc)) {
 458
 459                         /*
 460                          * This could be optimized (use queueid instead of
 461                          * portid), but it is not called so often
 462                          */
 463                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 464                                 if ((enabled_port_mask & (1 << portid)) != 0)
 465                                         send_burst(qconf, 1, portid);
 466                         }
 467
 468                         prev_tsc = cur_tsc;
 469                 }
 470
 471                 /*
 472                  * Read packet from RX queues
 473                  */
 474                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 475
 476                         portid = qconf->rx_queue_list[i].portid;
 477
 478                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 479                                 MAX_PKT_BURST);
 480
 481                         /* Prefetch first packets */
 482                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 483                                 rte_prefetch0(rte_pktmbuf_mtod(
 484                                                 pkts_burst[j], void *));
 485                         }
 486
 487                         /* Prefetch and forward already prefetched packets */
 488                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 489                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 490                                         j + PREFETCH_OFFSET], void *));
 491                                 reassemble(pkts_burst[j], portid,
 492                                         i, qconf, cur_tsc);
 493                         }
 494
 495                         /* Forward remaining prefetched packets */
 496                         for (; j < nb_rx; j++) {
 497                                 reassemble(pkts_burst[j], portid,
 498                                         i, qconf, cur_tsc);
 499                         }
 500
 501                         rte_ip_frag_free_death_row(&qconf->death_row,
 502                                 PREFETCH_OFFSET);
 503                 }
 504         }
 505 }
 506
 507 /* display usage */
 508 static void
 509 print_usage(const char *prgname)
 510 {
 511         printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
 512                 "  [--max-pkt-len PKTLEN]"
 513                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
 514                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 515                 "  -q NQ: number of RX queues per lcore\n"
 516                 "  --maxflows=<flows>: optional, maximum number of flows "
 517                 "supported\n"
 518                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
 519                 "flow\n",
 520                 prgname);
 521 }
 522
 523 static uint32_t
 524 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 525 {
 526         char *end;
 527         uint64_t v;
 528
 529         /* parse decimal string */
 530         errno = 0;
 531         v = strtoul(str, &end, 10);
 532         if (errno != 0 || *end != '\0')
 533                 return -EINVAL;
 534
 535         if (v < min || v > max)
 536                 return -EINVAL;
 537
 538         *val = (uint32_t)v;
 539         return 0;
 540 }
 541
 542 static int
 543 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 544 {
 545         char *end;
 546         uint64_t v;
 547
 548         static const char frmt_sec[] = "s";
 549         static const char frmt_msec[] = "ms";
 550
 551         /* parse decimal string */
 552         errno = 0;
 553         v = strtoul(str, &end, 10);
 554         if (errno != 0)
 555                 return -EINVAL;
 556
 557         if (*end != '\0') {
 558                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
 559                         v *= MS_PER_S;
 560                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
 561                         return -EINVAL;
 562         }
 563
 564         if (v < min || v > max)
 565                 return -EINVAL;
 566
 567         *val = (uint32_t)v;
 568         return 0;
 569 }
 570
 571 static int
 572 parse_portmask(const char *portmask)
 573 {
 574         char *end = NULL;
 575         unsigned long pm;
 576
 577         /* parse hexadecimal string */
 578         pm = strtoul(portmask, &end, 16);
 579         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 580                 return -1;
 581
 582         if (pm == 0)
 583                 return -1;
 584
 585         return pm;
 586 }
 587
 588 static int
 589 parse_nqueue(const char *q_arg)
 590 {
 591         char *end = NULL;
 592         unsigned long n;
 593
 594         printf("%p\n", q_arg);
 595
 596         /* parse hexadecimal string */
 597         n = strtoul(q_arg, &end, 10);
 598         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 599                 return -1;
 600         if (n == 0)
 601                 return -1;
 602         if (n >= MAX_RX_QUEUE_PER_LCORE)
 603                 return -1;
 604
 605         return n;
 606 }
 607
 608 /* Parse the argument given in the command line of the application */
 609 static int
 610 parse_args(int argc, char **argv)
 611 {
 612         int opt, ret;
 613         char **argvopt;
 614         int option_index;
 615         char *prgname = argv[0];
 616         static struct option lgopts[] = {
 617                 {"max-pkt-len", 1, 0, 0},
 618                 {"maxflows", 1, 0, 0},
 619                 {"flowttl", 1, 0, 0},
 620                 {NULL, 0, 0, 0}
 621         };
 622
 623         argvopt = argv;
 624
 625         while ((opt = getopt_long(argc, argvopt, "p:q:",
 626                                 lgopts, &option_index)) != EOF) {
 627
 628                 switch (opt) {
 629                 /* portmask */
 630                 case 'p':
 631                         enabled_port_mask = parse_portmask(optarg);
 632                         if (enabled_port_mask == 0) {
 633                                 printf("invalid portmask\n");
 634                                 print_usage(prgname);
 635                                 return -1;
 636                         }
 637                         break;
 638
 639                 /* nqueue */
 640                 case 'q':
 641                         rx_queue_per_lcore = parse_nqueue(optarg);
 642                         if (rx_queue_per_lcore < 0) {
 643                                 printf("invalid queue number\n");
 644                                 print_usage(prgname);
 645                                 return -1;
 646                         }
 647                         break;
 648
 649                 /* long options */
 650                 case 0:
 651                         if (!strncmp(lgopts[option_index].name,
 652                                         "maxflows", 8)) {
 653                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
 654                                                 MAX_FLOW_NUM,
 655                                                 &max_flow_num)) != 0) {
 656                                         printf("invalid value: \"%s\" for "
 657                                                 "parameter %s\n",
 658                                                 optarg,
 659                                                 lgopts[option_index].name);
 660                                         print_usage(prgname);
 661                                         return ret;
 662                                 }
 663                         }
 664
 665                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
 666                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
 667                                                 MAX_FLOW_TTL,
 668                                                 &max_flow_ttl)) != 0) {
 669                                         printf("invalid value: \"%s\" for "
 670                                                 "parameter %s\n",
 671                                                 optarg,
 672                                                 lgopts[option_index].name);
 673                                         print_usage(prgname);
 674                                         return ret;
 675                                 }
 676                         }
 677
 678                         break;
 679
 680                 default:
 681                         print_usage(prgname);
 682                         return -1;
 683                 }
 684         }
 685
 686         if (optind >= 0)
 687                 argv[optind-1] = prgname;
 688
 689         ret = optind-1;
 690         optind = 1; /* reset getopt lib */
 691         return ret;
 692 }
 693
 694 static void
 695 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 696 {
 697         char buf[ETHER_ADDR_FMT_SIZE];
 698         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 699         printf("%s%s", name, buf);
 700 }
 701
 702 /* Check the link status of all ports in up to 9s, and print them finally */
 703 static void
 704 check_all_ports_link_status(uint32_t port_mask)
 705 {
 706 #define CHECK_INTERVAL 100 /* 100ms */
 707 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 708         uint16_t portid;
 709         uint8_t count, all_ports_up, print_flag = 0;
 710         struct rte_eth_link link;
 711
 712         printf("\nChecking link status");
 713         fflush(stdout);
 714         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 715                 all_ports_up = 1;
 716                 RTE_ETH_FOREACH_DEV(portid) {
 717                         if ((port_mask & (1 << portid)) == 0)
 718                                 continue;
 719                         memset(&link, 0, sizeof(link));
 720                         rte_eth_link_get_nowait(portid, &link);
 721                         /* print link status if flag set */
 722                         if (print_flag == 1) {
 723                                 if (link.link_status)
 724                                         printf(
 725                                         "Port%d Link Up. Speed %u Mbps - %s\n",
 726                                                 portid, link.link_speed,
 727                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 728                                         ("full-duplex") : ("half-duplex\n"));
 729                                 else
 730                                         printf("Port %d Link Down\n", portid);
 731                                 continue;
 732                         }
 733                         /* clear all_ports_up flag if any link down */
 734                         if (link.link_status == ETH_LINK_DOWN) {
 735                                 all_ports_up = 0;
 736                                 break;
 737                         }
 738                 }
 739                 /* after finally printing all link status, get out */
 740                 if (print_flag == 1)
 741                         break;
 742
 743                 if (all_ports_up == 0) {
 744                         printf(".");
 745                         fflush(stdout);
 746                         rte_delay_ms(CHECK_INTERVAL);
 747                 }
 748
 749                 /* set the print_flag if all ports up or timeout */
 750                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 751                         print_flag = 1;
 752                         printf("\ndone\n");
 753                 }
 754         }
 755 }
 756
 757 static int
 758 init_routing_table(void)
 759 {
 760         struct rte_lpm *lpm;
 761         struct rte_lpm6 *lpm6;
 762         int socket, ret;
 763         unsigned i;
 764
 765         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 766                 if (socket_lpm[socket]) {
 767                         lpm = socket_lpm[socket];
 768                         /* populate the LPM table */
 769                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 770                                 ret = rte_lpm_add(lpm,
 771                                         l3fwd_ipv4_route_array[i].ip,
 772                                         l3fwd_ipv4_route_array[i].depth,
 773                                         l3fwd_ipv4_route_array[i].if_out);
 774
 775                                 if (ret < 0) {
 776                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
 777                                                 "LPM table\n", i);
 778                                         return -1;
 779                                 }
 780
 781                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
 782                                                 "/%d (port %d)\n",
 783                                         socket,
 784                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 785                                         l3fwd_ipv4_route_array[i].depth,
 786                                         l3fwd_ipv4_route_array[i].if_out);
 787                         }
 788                 }
 789
 790                 if (socket_lpm6[socket]) {
 791                         lpm6 = socket_lpm6[socket];
 792                         /* populate the LPM6 table */
 793                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 794                                 ret = rte_lpm6_add(lpm6,
 795                                         l3fwd_ipv6_route_array[i].ip,
 796                                         l3fwd_ipv6_route_array[i].depth,
 797                                         l3fwd_ipv6_route_array[i].if_out);
 798
 799                                 if (ret < 0) {
 800                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
 801                                                 "LPM6 table\n", i);
 802                                         return -1;
 803                                 }
 804
 805                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
 806                                                 "/%d (port %d)\n",
 807                                         socket,
 808                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 809                                         l3fwd_ipv6_route_array[i].depth,
 810                                         l3fwd_ipv6_route_array[i].if_out);
 811                         }
 812                 }
 813         }
 814         return 0;
 815 }
 816
 817 static int
 818 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
 819         uint32_t port)
 820 {
 821         struct mbuf_table *mtb;
 822         uint32_t n;
 823         size_t sz;
 824
 825         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
 826         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
 827
 828         if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
 829                         socket)) == NULL) {
 830                 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
 831                         "failed to allocate %zu bytes\n",
 832                         __func__, lcore, port, sz);
 833                 return -1;
 834         }
 835
 836         mtb->len = n;
 837         qconf->tx_mbufs[port] = mtb;
 838
 839         return 0;
 840 }
 841
 842 static int
 843 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
 844 {
 845         int socket;
 846         uint32_t nb_mbuf;
 847         uint64_t frag_cycles;
 848         char buf[RTE_MEMPOOL_NAMESIZE];
 849
 850         socket = rte_lcore_to_socket_id(lcore);
 851         if (socket == SOCKET_ID_ANY)
 852                 socket = 0;
 853
 854         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
 855                 max_flow_ttl;
 856
 857         if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
 858                         IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
 859                         socket)) == NULL) {
 860                 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
 861                         "lcore: %u for queue: %u failed\n",
 862                         max_flow_num, lcore, queue);
 863                 return -1;
 864         }
 865
 866         /*
 867          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
 868          * mbufs could be stored int the fragment table.
 869          * Plus, each TX queue can hold up to <max_flow_num> packets.
 870          */
 871
 872         nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
 873         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
 874         nb_mbuf *= 2; /* ipv4 and ipv6 */
 875         nb_mbuf += nb_rxd + nb_txd;
 876
 877         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
 878
 879         snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
 880
 881         rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
 882                                             MBUF_DATA_SIZE, socket);
 883         if (rxq->pool == NULL) {
 884                 RTE_LOG(ERR, IP_RSMBL,
 885                         "rte_pktmbuf_pool_create(%s) failed", buf);
 886                 return -1;
 887         }
 888
 889         return 0;
 890 }
 891
 892 static int
 893 init_mem(void)
 894 {
 895         char buf[PATH_MAX];
 896         struct rte_lpm *lpm;
 897         struct rte_lpm6 *lpm6;
 898         struct rte_lpm_config lpm_config;
 899         int socket;
 900         unsigned lcore_id;
 901
 902         /* traverse through lcores and initialize structures on each socket */
 903
 904         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 905
 906                 if (rte_lcore_is_enabled(lcore_id) == 0)
 907                         continue;
 908
 909                 socket = rte_lcore_to_socket_id(lcore_id);
 910
 911                 if (socket == SOCKET_ID_ANY)
 912                         socket = 0;
 913
 914                 if (socket_lpm[socket] == NULL) {
 915                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
 916                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
 917
 918                         lpm_config.max_rules = LPM_MAX_RULES;
 919                         lpm_config.number_tbl8s = 256;
 920                         lpm_config.flags = 0;
 921
 922                         lpm = rte_lpm_create(buf, socket, &lpm_config);
 923                         if (lpm == NULL) {
 924                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
 925                                 return -1;
 926                         }
 927                         socket_lpm[socket] = lpm;
 928                 }
 929
 930                 if (socket_lpm6[socket] == NULL) {
 931                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
 932                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
 933
 934                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
 935                         if (lpm6 == NULL) {
 936                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
 937                                 return -1;
 938                         }
 939                         socket_lpm6[socket] = lpm6;
 940                 }
 941         }
 942
 943         return 0;
 944 }
 945
 946 static void
 947 queue_dump_stat(void)
 948 {
 949         uint32_t i, lcore;
 950         const struct lcore_queue_conf *qconf;
 951
 952         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
 953                 if (rte_lcore_is_enabled(lcore) == 0)
 954                         continue;
 955
 956                 qconf = &lcore_queue_conf[lcore];
 957                 for (i = 0; i < qconf->n_rx_queue; i++) {
 958
 959                         fprintf(stdout, " -- lcoreid=%u portid=%u "
 960                                 "frag tbl stat:\n",
 961                                 lcore,  qconf->rx_queue_list[i].portid);
 962                         rte_ip_frag_table_statistics_dump(stdout,
 963                                         qconf->rx_queue_list[i].frag_tbl);
 964                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
 965                                 "TX packets _queued:\t%" PRIu64 "\n"
 966                                 "TX packets dropped:\t%" PRIu64 "\n"
 967                                 "TX packets send:\t%" PRIu64 "\n",
 968                                 qconf->tx_stat.call,
 969                                 qconf->tx_stat.queue,
 970                                 qconf->tx_stat.drop,
 971                                 qconf->tx_stat.send);
 972                 }
 973         }
 974 }
 975
 976 static void
 977 signal_handler(int signum)
 978 {
 979         queue_dump_stat();
 980         if (signum != SIGUSR1)
 981                 rte_exit(0, "received signal: %d, exiting\n", signum);
 982 }
 983
 984 int
 985 main(int argc, char **argv)
 986 {
 987         struct lcore_queue_conf *qconf;
 988         struct rte_eth_dev_info dev_info;
 989         struct rte_eth_txconf *txconf;
 990         struct rx_queue *rxq;
 991         int ret, socket;
 992         unsigned nb_ports;
 993         uint16_t queueid;
 994         unsigned lcore_id = 0, rx_lcore_id = 0;
 995         uint32_t n_tx_queue, nb_lcores;
 996         uint16_t portid;
 997
 998         /* init EAL */
 999         ret = rte_eal_init(argc, argv);
1000         if (ret < 0)
1001                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1002         argc -= ret;
1003         argv += ret;
1004
1005         /* parse application arguments (after the EAL ones) */
1006         ret = parse_args(argc, argv);
1007         if (ret < 0)
1008                 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1009
1010         nb_ports = rte_eth_dev_count_avail();
1011         if (nb_ports == 0)
1012                 rte_exit(EXIT_FAILURE, "No ports found!\n");
1013
1014         nb_lcores = rte_lcore_count();
1015
1016         /* initialize structures (mempools, lpm etc.) */
1017         if (init_mem() < 0)
1018                 rte_panic("Cannot initialize memory structures!\n");
1019
1020         /* check if portmask has non-existent ports */
1021         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1022                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1023
1024         /* initialize all ports */
1025         RTE_ETH_FOREACH_DEV(portid) {
1026                 struct rte_eth_rxconf rxq_conf;
1027                 struct rte_eth_conf local_port_conf = port_conf;
1028
1029                 /* skip ports that are not enabled */
1030                 if ((enabled_port_mask & (1 << portid)) == 0) {
1031                         printf("\nSkipping disabled port %d\n", portid);
1032                         continue;
1033                 }
1034
1035                 qconf = &lcore_queue_conf[rx_lcore_id];
1036
1037                 /* limit the frame size to the maximum supported by NIC */
1038                 rte_eth_dev_info_get(portid, &dev_info);
1039                 local_port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
1040                     dev_info.max_rx_pktlen,
1041                     local_port_conf.rxmode.max_rx_pkt_len);
1042
1043                 /* get the lcore_id for this port */
1044                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1045                            qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1046
1047                         rx_lcore_id++;
1048                         if (rx_lcore_id >= RTE_MAX_LCORE)
1049                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1050
1051                         qconf = &lcore_queue_conf[rx_lcore_id];
1052                 }
1053
1054                 socket = rte_lcore_to_socket_id(portid);
1055                 if (socket == SOCKET_ID_ANY)
1056                         socket = 0;
1057
1058                 queueid = qconf->n_rx_queue;
1059                 rxq = &qconf->rx_queue_list[queueid];
1060                 rxq->portid = portid;
1061                 rxq->lpm = socket_lpm[socket];
1062                 rxq->lpm6 = socket_lpm6[socket];
1063
1064                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1065                                                        &nb_txd);
1066                 if (ret < 0)
1067                         rte_exit(EXIT_FAILURE,
1068                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1069                                  ret, portid);
1070
1071                 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1072                         rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1073                 qconf->n_rx_queue++;
1074
1075                 /* init port */
1076                 printf("Initializing port %d ... ", portid );
1077                 fflush(stdout);
1078
1079                 n_tx_queue = nb_lcores;
1080                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1081                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1082                 if (dev_info.tx_offload_capa & DEV_TX_OFFLOAD_MBUF_FAST_FREE)
1083                         local_port_conf.txmode.offloads |=
1084                                 DEV_TX_OFFLOAD_MBUF_FAST_FREE;
1085
1086                 local_port_conf.rx_adv_conf.rss_conf.rss_hf &=
1087                         dev_info.flow_type_rss_offloads;
1088                 if (local_port_conf.rx_adv_conf.rss_conf.rss_hf !=
1089                                 port_conf.rx_adv_conf.rss_conf.rss_hf) {
1090                         printf("Port %u modified RSS hash function based on hardware support,"
1091                                 "requested:%#"PRIx64" configured:%#"PRIx64"\n",
1092                                 portid,
1093                                 port_conf.rx_adv_conf.rss_conf.rss_hf,
1094                                 local_port_conf.rx_adv_conf.rss_conf.rss_hf);
1095                 }
1096
1097                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1098                                             &local_port_conf);
1099                 if (ret < 0) {
1100                         printf("\n");
1101                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1102                                 "err=%d, port=%d\n",
1103                                 ret, portid);
1104                 }
1105
1106                 /* init one RX queue */
1107                 rxq_conf = dev_info.default_rxconf;
1108                 rxq_conf.offloads = local_port_conf.rxmode.offloads;
1109                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1110                                              socket, &rxq_conf,
1111                                              rxq->pool);
1112                 if (ret < 0) {
1113                         printf("\n");
1114                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1115                                 "err=%d, port=%d\n",
1116                                 ret, portid);
1117                 }
1118
1119                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1120                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1121                 printf("\n");
1122
1123                 /* init one TX queue per couple (lcore,port) */
1124                 queueid = 0;
1125                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1126                         if (rte_lcore_is_enabled(lcore_id) == 0)
1127                                 continue;
1128
1129                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1130
1131                         printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1132                         fflush(stdout);
1133
1134                         txconf = &dev_info.default_txconf;
1135                         txconf->offloads = local_port_conf.txmode.offloads;
1136
1137                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1138                                         socket, txconf);
1139                         if (ret < 0)
1140                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1141                                         "port=%d\n", ret, portid);
1142
1143                         qconf = &lcore_queue_conf[lcore_id];
1144                         qconf->tx_queue_id[portid] = queueid;
1145                         setup_port_tbl(qconf, lcore_id, socket, portid);
1146                         queueid++;
1147                 }
1148                 printf("\n");
1149         }
1150
1151         printf("\n");
1152
1153         /* start ports */
1154         RTE_ETH_FOREACH_DEV(portid) {
1155                 if ((enabled_port_mask & (1 << portid)) == 0) {
1156                         continue;
1157                 }
1158                 /* Start device */
1159                 ret = rte_eth_dev_start(portid);
1160                 if (ret < 0)
1161                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1162                                 ret, portid);
1163
1164                 rte_eth_promiscuous_enable(portid);
1165         }
1166
1167         if (init_routing_table() < 0)
1168                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1169
1170         check_all_ports_link_status(enabled_port_mask);
1171
1172         signal(SIGUSR1, signal_handler);
1173         signal(SIGTERM, signal_handler);
1174         signal(SIGINT, signal_handler);
1175
1176         /* launch per-lcore init on every lcore */
1177         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1178         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1179                 if (rte_eal_wait_lcore(lcore_id) < 0)
1180                         return -1;
1181         }
1182
1183         return 0;
1184 }