examples/ip_reassembly/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 <string.h>
  40 #include <sys/queue.h>
  41 #include <stdarg.h>
  42 #include <errno.h>
  43 #include <getopt.h>
  44 #include <signal.h>
  45 #include <sys/param.h>
  46
  47 #include <rte_common.h>
  48 #include <rte_byteorder.h>
  49 #include <rte_log.h>
  50 #include <rte_memory.h>
  51 #include <rte_memcpy.h>
  52 #include <rte_memzone.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_malloc.h>
  72 #include <rte_ip.h>
  73 #include <rte_tcp.h>
  74 #include <rte_udp.h>
  75 #include <rte_string_fns.h>
  76 #include <rte_lpm.h>
  77 #include <rte_lpm6.h>
  78
  79 #include <rte_ip_frag.h>
  80
  81 #define MAX_PKT_BURST 32
  82
  83
  84 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
  85
  86 #define MAX_JUMBO_PKT_LEN  9600
  87
  88 #define BUF_SIZE        RTE_MBUF_DEFAULT_DATAROOM
  89 #define MBUF_SIZE       \
  90         (BUF_SIZE + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
  91
  92 #define NB_MBUF 8192
  93
  94 /* allow max jumbo frame 9.5 KB */
  95 #define JUMBO_FRAME_MAX_SIZE    0x2600
  96
  97 #define MAX_FLOW_NUM    UINT16_MAX
  98 #define MIN_FLOW_NUM    1
  99 #define DEF_FLOW_NUM    0x1000
 100
 101 /* TTL numbers are in ms. */
 102 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
 103 #define MIN_FLOW_TTL    1
 104 #define DEF_FLOW_TTL    MS_PER_S
 105
 106 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
 107
 108 /* Should be power of two. */
 109 #define IP_FRAG_TBL_BUCKET_ENTRIES      16
 110
 111 static uint32_t max_flow_num = DEF_FLOW_NUM;
 112 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
 113
 114 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 115
 116 #define NB_SOCKETS 8
 117
 118 /* Configure how many packets ahead to prefetch, when reading packets */
 119 #define PREFETCH_OFFSET 3
 120
 121 /*
 122  * Configurable number of RX/TX ring descriptors
 123  */
 124 #define RTE_TEST_RX_DESC_DEFAULT 128
 125 #define RTE_TEST_TX_DESC_DEFAULT 512
 126
 127 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 128 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 129
 130 /* ethernet addresses of ports */
 131 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 132
 133 #ifndef IPv4_BYTES
 134 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
 135 #define IPv4_BYTES(addr) \
 136                 (uint8_t) (((addr) >> 24) & 0xFF),\
 137                 (uint8_t) (((addr) >> 16) & 0xFF),\
 138                 (uint8_t) (((addr) >> 8) & 0xFF),\
 139                 (uint8_t) ((addr) & 0xFF)
 140 #endif
 141
 142 #ifndef IPv6_BYTES
 143 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 144                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 145 #define IPv6_BYTES(addr) \
 146         addr[0],  addr[1], addr[2],  addr[3], \
 147         addr[4],  addr[5], addr[6],  addr[7], \
 148         addr[8],  addr[9], addr[10], addr[11],\
 149         addr[12], addr[13],addr[14], addr[15]
 150 #endif
 151
 152 #define IPV6_ADDR_LEN 16
 153
 154 /* mask of enabled ports */
 155 static uint32_t enabled_port_mask = 0;
 156
 157 static int rx_queue_per_lcore = 1;
 158
 159 struct mbuf_table {
 160         uint32_t len;
 161         uint32_t head;
 162         uint32_t tail;
 163         struct rte_mbuf *m_table[0];
 164 };
 165
 166 struct rx_queue {
 167         struct rte_ip_frag_tbl *frag_tbl;
 168         struct rte_mempool *pool;
 169         struct rte_lpm *lpm;
 170         struct rte_lpm6 *lpm6;
 171         uint8_t portid;
 172 };
 173
 174 struct tx_lcore_stat {
 175         uint64_t call;
 176         uint64_t drop;
 177         uint64_t queue;
 178         uint64_t send;
 179 };
 180
 181 #define MAX_RX_QUEUE_PER_LCORE 16
 182 #define MAX_TX_QUEUE_PER_PORT 16
 183 #define MAX_RX_QUEUE_PER_PORT 128
 184
 185 struct lcore_queue_conf {
 186         uint16_t n_rx_queue;
 187         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 188         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 189         struct rte_ip_frag_death_row death_row;
 190         struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
 191         struct tx_lcore_stat tx_stat;
 192 } __rte_cache_aligned;
 193 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 194
 195 static struct rte_eth_conf port_conf = {
 196         .rxmode = {
 197                 .mq_mode        = ETH_MQ_RX_RSS,
 198                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 199                 .split_hdr_size = 0,
 200                 .header_split   = 0, /**< Header Split disabled */
 201                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 202                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 203                 .jumbo_frame    = 1, /**< Jumbo Frame Support disabled */
 204                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 205         },
 206         .rx_adv_conf = {
 207                         .rss_conf = {
 208                                 .rss_key = NULL,
 209                                 .rss_hf = ETH_RSS_IP,
 210                 },
 211         },
 212         .txmode = {
 213                 .mq_mode = ETH_MQ_TX_NONE,
 214         },
 215 };
 216
 217 /*
 218  * IPv4 forwarding table
 219  */
 220 struct l3fwd_ipv4_route {
 221         uint32_t ip;
 222         uint8_t  depth;
 223         uint8_t  if_out;
 224 };
 225
 226 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 227                 {IPv4(100,10,0,0), 16, 0},
 228                 {IPv4(100,20,0,0), 16, 1},
 229                 {IPv4(100,30,0,0), 16, 2},
 230                 {IPv4(100,40,0,0), 16, 3},
 231                 {IPv4(100,50,0,0), 16, 4},
 232                 {IPv4(100,60,0,0), 16, 5},
 233                 {IPv4(100,70,0,0), 16, 6},
 234                 {IPv4(100,80,0,0), 16, 7},
 235 };
 236
 237 /*
 238  * IPv6 forwarding table
 239  */
 240
 241 struct l3fwd_ipv6_route {
 242         uint8_t ip[IPV6_ADDR_LEN];
 243         uint8_t depth;
 244         uint8_t if_out;
 245 };
 246
 247 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 248         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 249         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 250         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 251         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 252         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 253         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 254         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 255         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 256 };
 257
 258 #define LPM_MAX_RULES         1024
 259 #define LPM6_MAX_RULES         1024
 260 #define LPM6_NUMBER_TBL8S (1 << 16)
 261
 262 struct rte_lpm6_config lpm6_config = {
 263                 .max_rules = LPM6_MAX_RULES,
 264                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 265                 .flags = 0
 266 };
 267
 268 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 269 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 270
 271 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
 272 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
 273 #else
 274 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
 275 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
 276
 277 /*
 278  * If number of queued packets reached given threahold, then
 279  * send burst of packets on an output interface.
 280  */
 281 static inline uint32_t
 282 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint8_t port)
 283 {
 284         uint32_t fill, len, k, n;
 285         struct mbuf_table *txmb;
 286
 287         txmb = qconf->tx_mbufs[port];
 288         len = txmb->len;
 289
 290         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
 291                 fill += len;
 292
 293         if (fill >= thresh) {
 294                 n = RTE_MIN(len - txmb->tail, fill);
 295
 296                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
 297                         txmb->m_table + txmb->tail, (uint16_t)n);
 298
 299                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
 300                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
 301
 302                 fill -= k;
 303                 if ((txmb->tail += k) == len)
 304                         txmb->tail = 0;
 305         }
 306
 307         return (fill);
 308 }
 309
 310 /* Enqueue a single packet, and send burst if queue is filled */
 311 static inline int
 312 send_single_packet(struct rte_mbuf *m, uint8_t port)
 313 {
 314         uint32_t fill, lcore_id, len;
 315         struct lcore_queue_conf *qconf;
 316         struct mbuf_table *txmb;
 317
 318         lcore_id = rte_lcore_id();
 319         qconf = &lcore_queue_conf[lcore_id];
 320
 321         txmb = qconf->tx_mbufs[port];
 322         len = txmb->len;
 323
 324         fill = send_burst(qconf, MAX_PKT_BURST, port);
 325
 326         if (fill == len - 1) {
 327                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
 328                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
 329                 if (++txmb->tail == len)
 330                         txmb->tail = 0;
 331         }
 332
 333         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
 334         txmb->m_table[txmb->head] = m;
 335         if(++txmb->head == len)
 336                 txmb->head = 0;
 337
 338         return (0);
 339 }
 340
 341 static inline void
 342 reassemble(struct rte_mbuf *m, uint8_t portid, uint32_t queue,
 343         struct lcore_queue_conf *qconf, uint64_t tms)
 344 {
 345         struct ether_hdr *eth_hdr;
 346         struct rte_ip_frag_tbl *tbl;
 347         struct rte_ip_frag_death_row *dr;
 348         struct rx_queue *rxq;
 349         void *d_addr_bytes;
 350         uint8_t next_hop, dst_port;
 351
 352         rxq = &qconf->rx_queue_list[queue];
 353
 354         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 355
 356         dst_port = portid;
 357
 358         /* if packet is IPv4 */
 359         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 360                 struct ipv4_hdr *ip_hdr;
 361                 uint32_t ip_dst;
 362
 363                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 364
 365                  /* if it is a fragmented packet, then try to reassemble. */
 366                 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
 367                         struct rte_mbuf *mo;
 368
 369                         tbl = rxq->frag_tbl;
 370                         dr = &qconf->death_row;
 371
 372                         /* prepare mbuf: setup l2_len/l3_len. */
 373                         m->l2_len = sizeof(*eth_hdr);
 374                         m->l3_len = sizeof(*ip_hdr);
 375
 376                         /* process this fragment. */
 377                         mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
 378                         if (mo == NULL)
 379                                 /* no packet to send out. */
 380                                 return;
 381
 382                         /* we have our packet reassembled. */
 383                         if (mo != m) {
 384                                 m = mo;
 385                                 eth_hdr = rte_pktmbuf_mtod(m,
 386                                         struct ether_hdr *);
 387                                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 388                         }
 389                 }
 390                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 391
 392                 /* Find destination port */
 393                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 394                                 (enabled_port_mask & 1 << next_hop) != 0) {
 395                         dst_port = next_hop;
 396                 }
 397
 398                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 399         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 400                 /* if packet is IPv6 */
 401                 struct ipv6_extension_fragment *frag_hdr;
 402                 struct ipv6_hdr *ip_hdr;
 403
 404                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 405
 406                 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
 407
 408                 if (frag_hdr != NULL) {
 409                         struct rte_mbuf *mo;
 410
 411                         tbl = rxq->frag_tbl;
 412                         dr  = &qconf->death_row;
 413
 414                         /* prepare mbuf: setup l2_len/l3_len. */
 415                         m->l2_len = sizeof(*eth_hdr);
 416                         m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
 417
 418                         mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
 419                         if (mo == NULL)
 420                                 return;
 421
 422                         if (mo != m) {
 423                                 m = mo;
 424                                 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 425                                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 426                         }
 427                 }
 428
 429                 /* Find destination port */
 430                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, &next_hop) == 0 &&
 431                                 (enabled_port_mask & 1 << next_hop) != 0) {
 432                         dst_port = next_hop;
 433                 }
 434
 435                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 436         }
 437         /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
 438
 439         /* 02:00:00:00:00:xx */
 440         d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 441         *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 442
 443         /* src addr */
 444         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 445
 446         send_single_packet(m, dst_port);
 447 }
 448
 449 /* main processing loop */
 450 static int
 451 main_loop(__attribute__((unused)) void *dummy)
 452 {
 453         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 454         unsigned lcore_id;
 455         uint64_t diff_tsc, cur_tsc, prev_tsc;
 456         int i, j, nb_rx;
 457         uint8_t portid;
 458         struct lcore_queue_conf *qconf;
 459         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 460
 461         prev_tsc = 0;
 462
 463         lcore_id = rte_lcore_id();
 464         qconf = &lcore_queue_conf[lcore_id];
 465
 466         if (qconf->n_rx_queue == 0) {
 467                 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
 468                 return 0;
 469         }
 470
 471         RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
 472
 473         for (i = 0; i < qconf->n_rx_queue; i++) {
 474
 475                 portid = qconf->rx_queue_list[i].portid;
 476                 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%hhu\n", lcore_id,
 477                         portid);
 478         }
 479
 480         while (1) {
 481
 482                 cur_tsc = rte_rdtsc();
 483
 484                 /*
 485                  * TX burst queue drain
 486                  */
 487                 diff_tsc = cur_tsc - prev_tsc;
 488                 if (unlikely(diff_tsc > drain_tsc)) {
 489
 490                         /*
 491                          * This could be optimized (use queueid instead of
 492                          * portid), but it is not called so often
 493                          */
 494                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 495                                 if ((enabled_port_mask & (1 << portid)) != 0)
 496                                         send_burst(qconf, 1, portid);
 497                         }
 498
 499                         prev_tsc = cur_tsc;
 500                 }
 501
 502                 /*
 503                  * Read packet from RX queues
 504                  */
 505                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 506
 507                         portid = qconf->rx_queue_list[i].portid;
 508
 509                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 510                                 MAX_PKT_BURST);
 511
 512                         /* Prefetch first packets */
 513                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 514                                 rte_prefetch0(rte_pktmbuf_mtod(
 515                                                 pkts_burst[j], void *));
 516                         }
 517
 518                         /* Prefetch and forward already prefetched packets */
 519                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 520                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 521                                         j + PREFETCH_OFFSET], void *));
 522                                 reassemble(pkts_burst[j], portid,
 523                                         i, qconf, cur_tsc);
 524                         }
 525
 526                         /* Forward remaining prefetched packets */
 527                         for (; j < nb_rx; j++) {
 528                                 reassemble(pkts_burst[j], portid,
 529                                         i, qconf, cur_tsc);
 530                         }
 531
 532                         rte_ip_frag_free_death_row(&qconf->death_row,
 533                                 PREFETCH_OFFSET);
 534                 }
 535         }
 536 }
 537
 538 /* display usage */
 539 static void
 540 print_usage(const char *prgname)
 541 {
 542         printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
 543                 "  [--max-pkt-len PKTLEN]"
 544                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
 545                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 546                 "  -q NQ: number of RX queues per lcore\n"
 547                 "  --maxflows=<flows>: optional, maximum number of flows "
 548                 "supported\n"
 549                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
 550                 "flow\n",
 551                 prgname);
 552 }
 553
 554 static uint32_t
 555 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 556 {
 557         char *end;
 558         uint64_t v;
 559
 560         /* parse decimal string */
 561         errno = 0;
 562         v = strtoul(str, &end, 10);
 563         if (errno != 0 || *end != '\0')
 564                 return (-EINVAL);
 565
 566         if (v < min || v > max)
 567                 return (-EINVAL);
 568
 569         *val = (uint32_t)v;
 570         return (0);
 571 }
 572
 573 static int
 574 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 575 {
 576         char *end;
 577         uint64_t v;
 578
 579         static const char frmt_sec[] = "s";
 580         static const char frmt_msec[] = "ms";
 581
 582         /* parse decimal string */
 583         errno = 0;
 584         v = strtoul(str, &end, 10);
 585         if (errno != 0)
 586                 return (-EINVAL);
 587
 588         if (*end != '\0') {
 589                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
 590                         v *= MS_PER_S;
 591                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
 592                         return (-EINVAL);
 593         }
 594
 595         if (v < min || v > max)
 596                 return (-EINVAL);
 597
 598         *val = (uint32_t)v;
 599         return (0);
 600 }
 601
 602 static int
 603 parse_portmask(const char *portmask)
 604 {
 605         char *end = NULL;
 606         unsigned long pm;
 607
 608         /* parse hexadecimal string */
 609         pm = strtoul(portmask, &end, 16);
 610         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 611                 return -1;
 612
 613         if (pm == 0)
 614                 return -1;
 615
 616         return pm;
 617 }
 618
 619 static int
 620 parse_nqueue(const char *q_arg)
 621 {
 622         char *end = NULL;
 623         unsigned long n;
 624
 625         printf("%p\n", q_arg);
 626
 627         /* parse hexadecimal string */
 628         n = strtoul(q_arg, &end, 10);
 629         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 630                 return -1;
 631         if (n == 0)
 632                 return -1;
 633         if (n >= MAX_RX_QUEUE_PER_LCORE)
 634                 return -1;
 635
 636         return n;
 637 }
 638
 639 /* Parse the argument given in the command line of the application */
 640 static int
 641 parse_args(int argc, char **argv)
 642 {
 643         int opt, ret;
 644         char **argvopt;
 645         int option_index;
 646         char *prgname = argv[0];
 647         static struct option lgopts[] = {
 648                 {"max-pkt-len", 1, 0, 0},
 649                 {"maxflows", 1, 0, 0},
 650                 {"flowttl", 1, 0, 0},
 651                 {NULL, 0, 0, 0}
 652         };
 653
 654         argvopt = argv;
 655
 656         while ((opt = getopt_long(argc, argvopt, "p:q:",
 657                                 lgopts, &option_index)) != EOF) {
 658
 659                 switch (opt) {
 660                 /* portmask */
 661                 case 'p':
 662                         enabled_port_mask = parse_portmask(optarg);
 663                         if (enabled_port_mask == 0) {
 664                                 printf("invalid portmask\n");
 665                                 print_usage(prgname);
 666                                 return -1;
 667                         }
 668                         break;
 669
 670                 /* nqueue */
 671                 case 'q':
 672                         rx_queue_per_lcore = parse_nqueue(optarg);
 673                         if (rx_queue_per_lcore < 0) {
 674                                 printf("invalid queue number\n");
 675                                 print_usage(prgname);
 676                                 return -1;
 677                         }
 678                         break;
 679
 680                 /* long options */
 681                 case 0:
 682                         if (!strncmp(lgopts[option_index].name,
 683                                         "maxflows", 8)) {
 684                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
 685                                                 MAX_FLOW_NUM,
 686                                                 &max_flow_num)) != 0) {
 687                                         printf("invalid value: \"%s\" for "
 688                                                 "parameter %s\n",
 689                                                 optarg,
 690                                                 lgopts[option_index].name);
 691                                         print_usage(prgname);
 692                                         return (ret);
 693                                 }
 694                         }
 695
 696                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
 697                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
 698                                                 MAX_FLOW_TTL,
 699                                                 &max_flow_ttl)) != 0) {
 700                                         printf("invalid value: \"%s\" for "
 701                                                 "parameter %s\n",
 702                                                 optarg,
 703                                                 lgopts[option_index].name);
 704                                         print_usage(prgname);
 705                                         return (ret);
 706                                 }
 707                         }
 708
 709                         break;
 710
 711                 default:
 712                         print_usage(prgname);
 713                         return -1;
 714                 }
 715         }
 716
 717         if (optind >= 0)
 718                 argv[optind-1] = prgname;
 719
 720         ret = optind-1;
 721         optind = 0; /* reset getopt lib */
 722         return ret;
 723 }
 724
 725 static void
 726 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 727 {
 728         char buf[ETHER_ADDR_FMT_SIZE];
 729         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 730         printf("%s%s", name, buf);
 731 }
 732
 733 /* Check the link status of all ports in up to 9s, and print them finally */
 734 static void
 735 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 736 {
 737 #define CHECK_INTERVAL 100 /* 100ms */
 738 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 739         uint8_t portid, count, all_ports_up, print_flag = 0;
 740         struct rte_eth_link link;
 741
 742         printf("\nChecking link status");
 743         fflush(stdout);
 744         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 745                 all_ports_up = 1;
 746                 for (portid = 0; portid < port_num; portid++) {
 747                         if ((port_mask & (1 << portid)) == 0)
 748                                 continue;
 749                         memset(&link, 0, sizeof(link));
 750                         rte_eth_link_get_nowait(portid, &link);
 751                         /* print link status if flag set */
 752                         if (print_flag == 1) {
 753                                 if (link.link_status)
 754                                         printf("Port %d Link Up - speed %u "
 755                                                 "Mbps - %s\n", (uint8_t)portid,
 756                                                 (unsigned)link.link_speed,
 757                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 758                                         ("full-duplex") : ("half-duplex\n"));
 759                                 else
 760                                         printf("Port %d Link Down\n",
 761                                                 (uint8_t)portid);
 762                                 continue;
 763                         }
 764                         /* clear all_ports_up flag if any link down */
 765                         if (link.link_status == 0) {
 766                                 all_ports_up = 0;
 767                                 break;
 768                         }
 769                 }
 770                 /* after finally printing all link status, get out */
 771                 if (print_flag == 1)
 772                         break;
 773
 774                 if (all_ports_up == 0) {
 775                         printf(".");
 776                         fflush(stdout);
 777                         rte_delay_ms(CHECK_INTERVAL);
 778                 }
 779
 780                 /* set the print_flag if all ports up or timeout */
 781                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 782                         print_flag = 1;
 783                         printf("\ndone\n");
 784                 }
 785         }
 786 }
 787
 788 static int
 789 init_routing_table(void)
 790 {
 791         struct rte_lpm *lpm;
 792         struct rte_lpm6 *lpm6;
 793         int socket, ret;
 794         unsigned i;
 795
 796         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 797                 if (socket_lpm[socket]) {
 798                         lpm = socket_lpm[socket];
 799                         /* populate the LPM table */
 800                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 801                                 ret = rte_lpm_add(lpm,
 802                                         l3fwd_ipv4_route_array[i].ip,
 803                                         l3fwd_ipv4_route_array[i].depth,
 804                                         l3fwd_ipv4_route_array[i].if_out);
 805
 806                                 if (ret < 0) {
 807                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
 808                                                 "LPM table\n", i);
 809                                         return -1;
 810                                 }
 811
 812                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
 813                                                 "/%d (port %d)\n",
 814                                         socket,
 815                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 816                                         l3fwd_ipv4_route_array[i].depth,
 817                                         l3fwd_ipv4_route_array[i].if_out);
 818                         }
 819                 }
 820
 821                 if (socket_lpm6[socket]) {
 822                         lpm6 = socket_lpm6[socket];
 823                         /* populate the LPM6 table */
 824                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 825                                 ret = rte_lpm6_add(lpm6,
 826                                         l3fwd_ipv6_route_array[i].ip,
 827                                         l3fwd_ipv6_route_array[i].depth,
 828                                         l3fwd_ipv6_route_array[i].if_out);
 829
 830                                 if (ret < 0) {
 831                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
 832                                                 "LPM6 table\n", i);
 833                                         return -1;
 834                                 }
 835
 836                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
 837                                                 "/%d (port %d)\n",
 838                                         socket,
 839                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 840                                         l3fwd_ipv6_route_array[i].depth,
 841                                         l3fwd_ipv6_route_array[i].if_out);
 842                         }
 843                 }
 844         }
 845         return 0;
 846 }
 847
 848 static int
 849 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
 850         uint32_t port)
 851 {
 852         struct mbuf_table *mtb;
 853         uint32_t n;
 854         size_t sz;
 855
 856         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
 857         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
 858
 859         if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
 860                         socket)) == NULL) {
 861                 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
 862                         "failed to allocate %zu bytes\n",
 863                         __func__, lcore, port, sz);
 864                 return -1;
 865         }
 866
 867         mtb->len = n;
 868         qconf->tx_mbufs[port] = mtb;
 869
 870         return 0;
 871 }
 872
 873 static int
 874 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
 875 {
 876         int socket;
 877         uint32_t nb_mbuf;
 878         uint64_t frag_cycles;
 879         char buf[RTE_MEMPOOL_NAMESIZE];
 880
 881         socket = rte_lcore_to_socket_id(lcore);
 882         if (socket == SOCKET_ID_ANY)
 883                 socket = 0;
 884
 885         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
 886                 max_flow_ttl;
 887
 888         if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
 889                         IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
 890                         socket)) == NULL) {
 891                 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
 892                         "lcore: %u for queue: %u failed\n",
 893                         max_flow_num, lcore, queue);
 894                 return -1;
 895         }
 896
 897         /*
 898          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
 899          * mbufs could be stored int the fragment table.
 900          * Plus, each TX queue can hold up to <max_flow_num> packets.
 901          */
 902
 903         nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
 904         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
 905         nb_mbuf *= 2; /* ipv4 and ipv6 */
 906         nb_mbuf += RTE_TEST_RX_DESC_DEFAULT + RTE_TEST_TX_DESC_DEFAULT;
 907
 908         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
 909
 910         snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
 911
 912         if ((rxq->pool = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0,
 913                         sizeof(struct rte_pktmbuf_pool_private),
 914                         rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
 915                         socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL) {
 916                 RTE_LOG(ERR, IP_RSMBL, "mempool_create(%s) failed", buf);
 917                 return -1;
 918         }
 919
 920         return 0;
 921 }
 922
 923 static int
 924 init_mem(void)
 925 {
 926         char buf[PATH_MAX];
 927         struct rte_lpm *lpm;
 928         struct rte_lpm6 *lpm6;
 929         int socket;
 930         unsigned lcore_id;
 931
 932         /* traverse through lcores and initialize structures on each socket */
 933
 934         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 935
 936                 if (rte_lcore_is_enabled(lcore_id) == 0)
 937                         continue;
 938
 939                 socket = rte_lcore_to_socket_id(lcore_id);
 940
 941                 if (socket == SOCKET_ID_ANY)
 942                         socket = 0;
 943
 944                 if (socket_lpm[socket] == NULL) {
 945                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
 946                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
 947
 948                         lpm = rte_lpm_create(buf, socket, LPM_MAX_RULES, 0);
 949                         if (lpm == NULL) {
 950                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
 951                                 return -1;
 952                         }
 953                         socket_lpm[socket] = lpm;
 954                 }
 955
 956                 if (socket_lpm6[socket] == NULL) {
 957                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
 958                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
 959
 960                         lpm6 = rte_lpm6_create("IP_RSMBL_LPM6", socket, &lpm6_config);
 961                         if (lpm6 == NULL) {
 962                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
 963                                 return -1;
 964                         }
 965                         socket_lpm6[socket] = lpm6;
 966                 }
 967         }
 968
 969         return 0;
 970 }
 971
 972 static void
 973 queue_dump_stat(void)
 974 {
 975         uint32_t i, lcore;
 976         const struct lcore_queue_conf *qconf;
 977
 978         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
 979                 if (rte_lcore_is_enabled(lcore) == 0)
 980                         continue;
 981
 982                 qconf = &lcore_queue_conf[lcore];
 983                 for (i = 0; i < qconf->n_rx_queue; i++) {
 984
 985                         fprintf(stdout, " -- lcoreid=%u portid=%hhu "
 986                                 "frag tbl stat:\n",
 987                                 lcore,  qconf->rx_queue_list[i].portid);
 988                         rte_ip_frag_table_statistics_dump(stdout,
 989                                         qconf->rx_queue_list[i].frag_tbl);
 990                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
 991                                 "TX packets _queued:\t%" PRIu64 "\n"
 992                                 "TX packets dropped:\t%" PRIu64 "\n"
 993                                 "TX packets send:\t%" PRIu64 "\n",
 994                                 qconf->tx_stat.call,
 995                                 qconf->tx_stat.queue,
 996                                 qconf->tx_stat.drop,
 997                                 qconf->tx_stat.send);
 998                 }
 999         }
1000 }
1001
1002 static void
1003 signal_handler(int signum)
1004 {
1005         queue_dump_stat();
1006         if (signum != SIGUSR1)
1007                 rte_exit(0, "received signal: %d, exiting\n", signum);
1008 }
1009
1010 int
1011 main(int argc, char **argv)
1012 {
1013         struct lcore_queue_conf *qconf;
1014         struct rte_eth_dev_info dev_info;
1015         struct rte_eth_txconf *txconf;
1016         struct rx_queue *rxq;
1017         int ret, socket;
1018         unsigned nb_ports;
1019         uint16_t queueid;
1020         unsigned lcore_id = 0, rx_lcore_id = 0;
1021         uint32_t n_tx_queue, nb_lcores;
1022         uint8_t portid;
1023
1024         /* init EAL */
1025         ret = rte_eal_init(argc, argv);
1026         if (ret < 0)
1027                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1028         argc -= ret;
1029         argv += ret;
1030
1031         /* parse application arguments (after the EAL ones) */
1032         ret = parse_args(argc, argv);
1033         if (ret < 0)
1034                 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1035
1036         nb_ports = rte_eth_dev_count();
1037         if (nb_ports > RTE_MAX_ETHPORTS)
1038                 nb_ports = RTE_MAX_ETHPORTS;
1039         else if (nb_ports == 0)
1040                 rte_exit(EXIT_FAILURE, "No ports found!\n");
1041
1042         nb_lcores = rte_lcore_count();
1043
1044         /* initialize structures (mempools, lpm etc.) */
1045         if (init_mem() < 0)
1046                 rte_panic("Cannot initialize memory structures!\n");
1047
1048         /* check if portmask has non-existent ports */
1049         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1050                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1051
1052         /* initialize all ports */
1053         for (portid = 0; portid < nb_ports; portid++) {
1054                 /* skip ports that are not enabled */
1055                 if ((enabled_port_mask & (1 << portid)) == 0) {
1056                         printf("\nSkipping disabled port %d\n", portid);
1057                         continue;
1058                 }
1059
1060                 qconf = &lcore_queue_conf[rx_lcore_id];
1061
1062                 /* get the lcore_id for this port */
1063                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1064                            qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1065
1066                         rx_lcore_id++;
1067                         if (rx_lcore_id >= RTE_MAX_LCORE)
1068                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1069
1070                         qconf = &lcore_queue_conf[rx_lcore_id];
1071                 }
1072
1073                 socket = rte_lcore_to_socket_id(portid);
1074                 if (socket == SOCKET_ID_ANY)
1075                         socket = 0;
1076
1077                 queueid = qconf->n_rx_queue;
1078                 rxq = &qconf->rx_queue_list[queueid];
1079                 rxq->portid = portid;
1080                 rxq->lpm = socket_lpm[socket];
1081                 rxq->lpm6 = socket_lpm6[socket];
1082                 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1083                         rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1084                 qconf->n_rx_queue++;
1085
1086                 /* init port */
1087                 printf("Initializing port %d ... ", portid );
1088                 fflush(stdout);
1089
1090                 n_tx_queue = nb_lcores;
1091                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1092                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1093                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1094                                             &port_conf);
1095                 if (ret < 0) {
1096                         printf("\n");
1097                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1098                                 "err=%d, port=%d\n",
1099                                 ret, portid);
1100                 }
1101
1102                 /* init one RX queue */
1103                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1104                                              socket, NULL,
1105                                              rxq->pool);
1106                 if (ret < 0) {
1107                         printf("\n");
1108                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1109                                 "err=%d, port=%d\n",
1110                                 ret, portid);
1111                 }
1112
1113                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1114                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1115                 printf("\n");
1116
1117                 /* init one TX queue per couple (lcore,port) */
1118                 queueid = 0;
1119                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1120                         if (rte_lcore_is_enabled(lcore_id) == 0)
1121                                 continue;
1122
1123                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1124
1125                         printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1126                         fflush(stdout);
1127
1128                         rte_eth_dev_info_get(portid, &dev_info);
1129                         txconf = &dev_info.default_txconf;
1130                         txconf->txq_flags = 0;
1131
1132                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1133                                         socket, txconf);
1134                         if (ret < 0)
1135                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1136                                         "port=%d\n", ret, portid);
1137
1138                         qconf = &lcore_queue_conf[lcore_id];
1139                         qconf->tx_queue_id[portid] = queueid;
1140                         setup_port_tbl(qconf, lcore_id, socket, portid);
1141                         queueid++;
1142                 }
1143                 printf("\n");
1144         }
1145
1146         printf("\n");
1147
1148         /* start ports */
1149         for (portid = 0; portid < nb_ports; portid++) {
1150                 if ((enabled_port_mask & (1 << portid)) == 0) {
1151                         continue;
1152                 }
1153                 /* Start device */
1154                 ret = rte_eth_dev_start(portid);
1155                 if (ret < 0)
1156                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1157                                 ret, portid);
1158
1159                 rte_eth_promiscuous_enable(portid);
1160         }
1161
1162         if (init_routing_table() < 0)
1163                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1164
1165         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1166
1167         signal(SIGUSR1, signal_handler);
1168         signal(SIGTERM, signal_handler);
1169         signal(SIGINT, signal_handler);
1170
1171         /* launch per-lcore init on every lcore */
1172         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1173         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1174                 if (rte_eal_wait_lcore(lcore_id) < 0)
1175                         return -1;
1176         }
1177
1178         return 0;
1179 }