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_eal.h>
  53 #include <rte_launch.h>
  54 #include <rte_atomic.h>
  55 #include <rte_cycles.h>
  56 #include <rte_prefetch.h>
  57 #include <rte_lcore.h>
  58 #include <rte_per_lcore.h>
  59 #include <rte_branch_prediction.h>
  60 #include <rte_interrupts.h>
  61 #include <rte_pci.h>
  62 #include <rte_random.h>
  63 #include <rte_debug.h>
  64 #include <rte_ether.h>
  65 #include <rte_ethdev.h>
  66 #include <rte_mempool.h>
  67 #include <rte_mbuf.h>
  68 #include <rte_malloc.h>
  69 #include <rte_ip.h>
  70 #include <rte_tcp.h>
  71 #include <rte_udp.h>
  72 #include <rte_string_fns.h>
  73 #include <rte_lpm.h>
  74 #include <rte_lpm6.h>
  75
  76 #include <rte_ip_frag.h>
  77
  78 #define MAX_PKT_BURST 32
  79
  80
  81 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
  82
  83 #define MAX_JUMBO_PKT_LEN  9600
  84
  85 #define BUF_SIZE        RTE_MBUF_DEFAULT_DATAROOM
  86 #define MBUF_DATA_SIZE  RTE_MBUF_DEFAULT_BUF_SIZE
  87
  88 #define NB_MBUF 8192
  89 #define MEMPOOL_CACHE_SIZE 256
  90
  91 /* allow max jumbo frame 9.5 KB */
  92 #define JUMBO_FRAME_MAX_SIZE    0x2600
  93
  94 #define MAX_FLOW_NUM    UINT16_MAX
  95 #define MIN_FLOW_NUM    1
  96 #define DEF_FLOW_NUM    0x1000
  97
  98 /* TTL numbers are in ms. */
  99 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
 100 #define MIN_FLOW_TTL    1
 101 #define DEF_FLOW_TTL    MS_PER_S
 102
 103 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
 104
 105 /* Should be power of two. */
 106 #define IP_FRAG_TBL_BUCKET_ENTRIES      16
 107
 108 static uint32_t max_flow_num = DEF_FLOW_NUM;
 109 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
 110
 111 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 112
 113 #define NB_SOCKETS 8
 114
 115 /* Configure how many packets ahead to prefetch, when reading packets */
 116 #define PREFETCH_OFFSET 3
 117
 118 /*
 119  * Configurable number of RX/TX ring descriptors
 120  */
 121 #define RTE_TEST_RX_DESC_DEFAULT 128
 122 #define RTE_TEST_TX_DESC_DEFAULT 512
 123
 124 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 125 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 126
 127 /* ethernet addresses of ports */
 128 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 129
 130 #ifndef IPv4_BYTES
 131 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
 132 #define IPv4_BYTES(addr) \
 133                 (uint8_t) (((addr) >> 24) & 0xFF),\
 134                 (uint8_t) (((addr) >> 16) & 0xFF),\
 135                 (uint8_t) (((addr) >> 8) & 0xFF),\
 136                 (uint8_t) ((addr) & 0xFF)
 137 #endif
 138
 139 #ifndef IPv6_BYTES
 140 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 141                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 142 #define IPv6_BYTES(addr) \
 143         addr[0],  addr[1], addr[2],  addr[3], \
 144         addr[4],  addr[5], addr[6],  addr[7], \
 145         addr[8],  addr[9], addr[10], addr[11],\
 146         addr[12], addr[13],addr[14], addr[15]
 147 #endif
 148
 149 #define IPV6_ADDR_LEN 16
 150
 151 /* mask of enabled ports */
 152 static uint32_t enabled_port_mask = 0;
 153
 154 static int rx_queue_per_lcore = 1;
 155
 156 struct mbuf_table {
 157         uint32_t len;
 158         uint32_t head;
 159         uint32_t tail;
 160         struct rte_mbuf *m_table[0];
 161 };
 162
 163 struct rx_queue {
 164         struct rte_ip_frag_tbl *frag_tbl;
 165         struct rte_mempool *pool;
 166         struct rte_lpm *lpm;
 167         struct rte_lpm6 *lpm6;
 168         uint16_t portid;
 169 };
 170
 171 struct tx_lcore_stat {
 172         uint64_t call;
 173         uint64_t drop;
 174         uint64_t queue;
 175         uint64_t send;
 176 };
 177
 178 #define MAX_RX_QUEUE_PER_LCORE 16
 179 #define MAX_TX_QUEUE_PER_PORT 16
 180 #define MAX_RX_QUEUE_PER_PORT 128
 181
 182 struct lcore_queue_conf {
 183         uint16_t n_rx_queue;
 184         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 185         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 186         struct rte_ip_frag_death_row death_row;
 187         struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
 188         struct tx_lcore_stat tx_stat;
 189 } __rte_cache_aligned;
 190 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 191
 192 static struct rte_eth_conf port_conf = {
 193         .rxmode = {
 194                 .mq_mode        = ETH_MQ_RX_RSS,
 195                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 196                 .split_hdr_size = 0,
 197                 .header_split   = 0, /**< Header Split disabled */
 198                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 199                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 200                 .jumbo_frame    = 1, /**< Jumbo Frame Support disabled */
 201                 .hw_strip_crc   = 1, /**< CRC stripped by hardware */
 202         },
 203         .rx_adv_conf = {
 204                         .rss_conf = {
 205                                 .rss_key = NULL,
 206                                 .rss_hf = ETH_RSS_IP,
 207                 },
 208         },
 209         .txmode = {
 210                 .mq_mode = ETH_MQ_TX_NONE,
 211         },
 212 };
 213
 214 /*
 215  * IPv4 forwarding table
 216  */
 217 struct l3fwd_ipv4_route {
 218         uint32_t ip;
 219         uint8_t  depth;
 220         uint8_t  if_out;
 221 };
 222
 223 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 224                 {IPv4(100,10,0,0), 16, 0},
 225                 {IPv4(100,20,0,0), 16, 1},
 226                 {IPv4(100,30,0,0), 16, 2},
 227                 {IPv4(100,40,0,0), 16, 3},
 228                 {IPv4(100,50,0,0), 16, 4},
 229                 {IPv4(100,60,0,0), 16, 5},
 230                 {IPv4(100,70,0,0), 16, 6},
 231                 {IPv4(100,80,0,0), 16, 7},
 232 };
 233
 234 /*
 235  * IPv6 forwarding table
 236  */
 237
 238 struct l3fwd_ipv6_route {
 239         uint8_t ip[IPV6_ADDR_LEN];
 240         uint8_t depth;
 241         uint8_t if_out;
 242 };
 243
 244 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 245         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 246         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 247         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 248         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 249         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 250         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 251         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 252         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 253 };
 254
 255 #define LPM_MAX_RULES         1024
 256 #define LPM6_MAX_RULES         1024
 257 #define LPM6_NUMBER_TBL8S (1 << 16)
 258
 259 struct rte_lpm6_config lpm6_config = {
 260                 .max_rules = LPM6_MAX_RULES,
 261                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 262                 .flags = 0
 263 };
 264
 265 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 266 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 267
 268 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
 269 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
 270 #else
 271 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
 272 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
 273
 274 /*
 275  * If number of queued packets reached given threahold, then
 276  * send burst of packets on an output interface.
 277  */
 278 static inline uint32_t
 279 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint16_t port)
 280 {
 281         uint32_t fill, len, k, n;
 282         struct mbuf_table *txmb;
 283
 284         txmb = qconf->tx_mbufs[port];
 285         len = txmb->len;
 286
 287         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
 288                 fill += len;
 289
 290         if (fill >= thresh) {
 291                 n = RTE_MIN(len - txmb->tail, fill);
 292
 293                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
 294                         txmb->m_table + txmb->tail, (uint16_t)n);
 295
 296                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
 297                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
 298
 299                 fill -= k;
 300                 if ((txmb->tail += k) == len)
 301                         txmb->tail = 0;
 302         }
 303
 304         return fill;
 305 }
 306
 307 /* Enqueue a single packet, and send burst if queue is filled */
 308 static inline int
 309 send_single_packet(struct rte_mbuf *m, uint16_t port)
 310 {
 311         uint32_t fill, lcore_id, len;
 312         struct lcore_queue_conf *qconf;
 313         struct mbuf_table *txmb;
 314
 315         lcore_id = rte_lcore_id();
 316         qconf = &lcore_queue_conf[lcore_id];
 317
 318         txmb = qconf->tx_mbufs[port];
 319         len = txmb->len;
 320
 321         fill = send_burst(qconf, MAX_PKT_BURST, port);
 322
 323         if (fill == len - 1) {
 324                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
 325                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
 326                 if (++txmb->tail == len)
 327                         txmb->tail = 0;
 328         }
 329
 330         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
 331         txmb->m_table[txmb->head] = m;
 332         if(++txmb->head == len)
 333                 txmb->head = 0;
 334
 335         return 0;
 336 }
 337
 338 static inline void
 339 reassemble(struct rte_mbuf *m, uint16_t portid, uint32_t queue,
 340         struct lcore_queue_conf *qconf, uint64_t tms)
 341 {
 342         struct ether_hdr *eth_hdr;
 343         struct rte_ip_frag_tbl *tbl;
 344         struct rte_ip_frag_death_row *dr;
 345         struct rx_queue *rxq;
 346         void *d_addr_bytes;
 347         uint32_t next_hop;
 348         uint16_t dst_port;
 349
 350         rxq = &qconf->rx_queue_list[queue];
 351
 352         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 353
 354         dst_port = portid;
 355
 356         /* if packet is IPv4 */
 357         if (RTE_ETH_IS_IPV4_HDR(m->packet_type)) {
 358                 struct ipv4_hdr *ip_hdr;
 359                 uint32_t ip_dst;
 360
 361                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 362
 363                  /* if it is a fragmented packet, then try to reassemble. */
 364                 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
 365                         struct rte_mbuf *mo;
 366
 367                         tbl = rxq->frag_tbl;
 368                         dr = &qconf->death_row;
 369
 370                         /* prepare mbuf: setup l2_len/l3_len. */
 371                         m->l2_len = sizeof(*eth_hdr);
 372                         m->l3_len = sizeof(*ip_hdr);
 373
 374                         /* process this fragment. */
 375                         mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
 376                         if (mo == NULL)
 377                                 /* no packet to send out. */
 378                                 return;
 379
 380                         /* we have our packet reassembled. */
 381                         if (mo != m) {
 382                                 m = mo;
 383                                 eth_hdr = rte_pktmbuf_mtod(m,
 384                                         struct ether_hdr *);
 385                                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 386                         }
 387                 }
 388                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 389
 390                 /* Find destination port */
 391                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 392                                 (enabled_port_mask & 1 << next_hop) != 0) {
 393                         dst_port = next_hop;
 394                 }
 395
 396                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 397         } else if (RTE_ETH_IS_IPV6_HDR(m->packet_type)) {
 398                 /* if packet is IPv6 */
 399                 struct ipv6_extension_fragment *frag_hdr;
 400                 struct ipv6_hdr *ip_hdr;
 401
 402                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 403
 404                 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
 405
 406                 if (frag_hdr != NULL) {
 407                         struct rte_mbuf *mo;
 408
 409                         tbl = rxq->frag_tbl;
 410                         dr  = &qconf->death_row;
 411
 412                         /* prepare mbuf: setup l2_len/l3_len. */
 413                         m->l2_len = sizeof(*eth_hdr);
 414                         m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
 415
 416                         mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
 417                         if (mo == NULL)
 418                                 return;
 419
 420                         if (mo != m) {
 421                                 m = mo;
 422                                 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 423                                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 424                         }
 425                 }
 426
 427                 /* Find destination port */
 428                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr,
 429                                                 &next_hop) == 0 &&
 430                                 (enabled_port_mask & 1 << next_hop) != 0) {
 431                         dst_port = next_hop;
 432                 }
 433
 434                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 435         }
 436         /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
 437
 438         /* 02:00:00:00:00:xx */
 439         d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 440         *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 441
 442         /* src addr */
 443         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 444
 445         send_single_packet(m, dst_port);
 446 }
 447
 448 /* main processing loop */
 449 static int
 450 main_loop(__attribute__((unused)) void *dummy)
 451 {
 452         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 453         unsigned lcore_id;
 454         uint64_t diff_tsc, cur_tsc, prev_tsc;
 455         int i, j, nb_rx;
 456         uint16_t portid;
 457         struct lcore_queue_conf *qconf;
 458         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 459
 460         prev_tsc = 0;
 461
 462         lcore_id = rte_lcore_id();
 463         qconf = &lcore_queue_conf[lcore_id];
 464
 465         if (qconf->n_rx_queue == 0) {
 466                 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
 467                 return 0;
 468         }
 469
 470         RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
 471
 472         for (i = 0; i < qconf->n_rx_queue; i++) {
 473
 474                 portid = qconf->rx_queue_list[i].portid;
 475                 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%u\n", lcore_id,
 476                         portid);
 477         }
 478
 479         while (1) {
 480
 481                 cur_tsc = rte_rdtsc();
 482
 483                 /*
 484                  * TX burst queue drain
 485                  */
 486                 diff_tsc = cur_tsc - prev_tsc;
 487                 if (unlikely(diff_tsc > drain_tsc)) {
 488
 489                         /*
 490                          * This could be optimized (use queueid instead of
 491                          * portid), but it is not called so often
 492                          */
 493                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 494                                 if ((enabled_port_mask & (1 << portid)) != 0)
 495                                         send_burst(qconf, 1, portid);
 496                         }
 497
 498                         prev_tsc = cur_tsc;
 499                 }
 500
 501                 /*
 502                  * Read packet from RX queues
 503                  */
 504                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 505
 506                         portid = qconf->rx_queue_list[i].portid;
 507
 508                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 509                                 MAX_PKT_BURST);
 510
 511                         /* Prefetch first packets */
 512                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 513                                 rte_prefetch0(rte_pktmbuf_mtod(
 514                                                 pkts_burst[j], void *));
 515                         }
 516
 517                         /* Prefetch and forward already prefetched packets */
 518                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 519                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 520                                         j + PREFETCH_OFFSET], void *));
 521                                 reassemble(pkts_burst[j], portid,
 522                                         i, qconf, cur_tsc);
 523                         }
 524
 525                         /* Forward remaining prefetched packets */
 526                         for (; j < nb_rx; j++) {
 527                                 reassemble(pkts_burst[j], portid,
 528                                         i, qconf, cur_tsc);
 529                         }
 530
 531                         rte_ip_frag_free_death_row(&qconf->death_row,
 532                                 PREFETCH_OFFSET);
 533                 }
 534         }
 535 }
 536
 537 /* display usage */
 538 static void
 539 print_usage(const char *prgname)
 540 {
 541         printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
 542                 "  [--max-pkt-len PKTLEN]"
 543                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
 544                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 545                 "  -q NQ: number of RX queues per lcore\n"
 546                 "  --maxflows=<flows>: optional, maximum number of flows "
 547                 "supported\n"
 548                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
 549                 "flow\n",
 550                 prgname);
 551 }
 552
 553 static uint32_t
 554 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 555 {
 556         char *end;
 557         uint64_t v;
 558
 559         /* parse decimal string */
 560         errno = 0;
 561         v = strtoul(str, &end, 10);
 562         if (errno != 0 || *end != '\0')
 563                 return -EINVAL;
 564
 565         if (v < min || v > max)
 566                 return -EINVAL;
 567
 568         *val = (uint32_t)v;
 569         return 0;
 570 }
 571
 572 static int
 573 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 574 {
 575         char *end;
 576         uint64_t v;
 577
 578         static const char frmt_sec[] = "s";
 579         static const char frmt_msec[] = "ms";
 580
 581         /* parse decimal string */
 582         errno = 0;
 583         v = strtoul(str, &end, 10);
 584         if (errno != 0)
 585                 return -EINVAL;
 586
 587         if (*end != '\0') {
 588                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
 589                         v *= MS_PER_S;
 590                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
 591                         return -EINVAL;
 592         }
 593
 594         if (v < min || v > max)
 595                 return -EINVAL;
 596
 597         *val = (uint32_t)v;
 598         return 0;
 599 }
 600
 601 static int
 602 parse_portmask(const char *portmask)
 603 {
 604         char *end = NULL;
 605         unsigned long pm;
 606
 607         /* parse hexadecimal string */
 608         pm = strtoul(portmask, &end, 16);
 609         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 610                 return -1;
 611
 612         if (pm == 0)
 613                 return -1;
 614
 615         return pm;
 616 }
 617
 618 static int
 619 parse_nqueue(const char *q_arg)
 620 {
 621         char *end = NULL;
 622         unsigned long n;
 623
 624         printf("%p\n", q_arg);
 625
 626         /* parse hexadecimal string */
 627         n = strtoul(q_arg, &end, 10);
 628         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 629                 return -1;
 630         if (n == 0)
 631                 return -1;
 632         if (n >= MAX_RX_QUEUE_PER_LCORE)
 633                 return -1;
 634
 635         return n;
 636 }
 637
 638 /* Parse the argument given in the command line of the application */
 639 static int
 640 parse_args(int argc, char **argv)
 641 {
 642         int opt, ret;
 643         char **argvopt;
 644         int option_index;
 645         char *prgname = argv[0];
 646         static struct option lgopts[] = {
 647                 {"max-pkt-len", 1, 0, 0},
 648                 {"maxflows", 1, 0, 0},
 649                 {"flowttl", 1, 0, 0},
 650                 {NULL, 0, 0, 0}
 651         };
 652
 653         argvopt = argv;
 654
 655         while ((opt = getopt_long(argc, argvopt, "p:q:",
 656                                 lgopts, &option_index)) != EOF) {
 657
 658                 switch (opt) {
 659                 /* portmask */
 660                 case 'p':
 661                         enabled_port_mask = parse_portmask(optarg);
 662                         if (enabled_port_mask == 0) {
 663                                 printf("invalid portmask\n");
 664                                 print_usage(prgname);
 665                                 return -1;
 666                         }
 667                         break;
 668
 669                 /* nqueue */
 670                 case 'q':
 671                         rx_queue_per_lcore = parse_nqueue(optarg);
 672                         if (rx_queue_per_lcore < 0) {
 673                                 printf("invalid queue number\n");
 674                                 print_usage(prgname);
 675                                 return -1;
 676                         }
 677                         break;
 678
 679                 /* long options */
 680                 case 0:
 681                         if (!strncmp(lgopts[option_index].name,
 682                                         "maxflows", 8)) {
 683                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
 684                                                 MAX_FLOW_NUM,
 685                                                 &max_flow_num)) != 0) {
 686                                         printf("invalid value: \"%s\" for "
 687                                                 "parameter %s\n",
 688                                                 optarg,
 689                                                 lgopts[option_index].name);
 690                                         print_usage(prgname);
 691                                         return ret;
 692                                 }
 693                         }
 694
 695                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
 696                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
 697                                                 MAX_FLOW_TTL,
 698                                                 &max_flow_ttl)) != 0) {
 699                                         printf("invalid value: \"%s\" for "
 700                                                 "parameter %s\n",
 701                                                 optarg,
 702                                                 lgopts[option_index].name);
 703                                         print_usage(prgname);
 704                                         return ret;
 705                                 }
 706                         }
 707
 708                         break;
 709
 710                 default:
 711                         print_usage(prgname);
 712                         return -1;
 713                 }
 714         }
 715
 716         if (optind >= 0)
 717                 argv[optind-1] = prgname;
 718
 719         ret = optind-1;
 720         optind = 1; /* reset getopt lib */
 721         return ret;
 722 }
 723
 724 static void
 725 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 726 {
 727         char buf[ETHER_ADDR_FMT_SIZE];
 728         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 729         printf("%s%s", name, buf);
 730 }
 731
 732 /* Check the link status of all ports in up to 9s, and print them finally */
 733 static void
 734 check_all_ports_link_status(uint16_t port_num, uint32_t port_mask)
 735 {
 736 #define CHECK_INTERVAL 100 /* 100ms */
 737 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 738         uint16_t portid;
 739         uint8_t 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(
 755                                         "Port%d Link Up. Speed %u Mbps - %s\n",
 756                                                 portid, 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", portid);
 761                                 continue;
 762                         }
 763                         /* clear all_ports_up flag if any link down */
 764                         if (link.link_status == ETH_LINK_DOWN) {
 765                                 all_ports_up = 0;
 766                                 break;
 767                         }
 768                 }
 769                 /* after finally printing all link status, get out */
 770                 if (print_flag == 1)
 771                         break;
 772
 773                 if (all_ports_up == 0) {
 774                         printf(".");
 775                         fflush(stdout);
 776                         rte_delay_ms(CHECK_INTERVAL);
 777                 }
 778
 779                 /* set the print_flag if all ports up or timeout */
 780                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 781                         print_flag = 1;
 782                         printf("\ndone\n");
 783                 }
 784         }
 785 }
 786
 787 static int
 788 init_routing_table(void)
 789 {
 790         struct rte_lpm *lpm;
 791         struct rte_lpm6 *lpm6;
 792         int socket, ret;
 793         unsigned i;
 794
 795         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 796                 if (socket_lpm[socket]) {
 797                         lpm = socket_lpm[socket];
 798                         /* populate the LPM table */
 799                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 800                                 ret = rte_lpm_add(lpm,
 801                                         l3fwd_ipv4_route_array[i].ip,
 802                                         l3fwd_ipv4_route_array[i].depth,
 803                                         l3fwd_ipv4_route_array[i].if_out);
 804
 805                                 if (ret < 0) {
 806                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
 807                                                 "LPM table\n", i);
 808                                         return -1;
 809                                 }
 810
 811                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
 812                                                 "/%d (port %d)\n",
 813                                         socket,
 814                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 815                                         l3fwd_ipv4_route_array[i].depth,
 816                                         l3fwd_ipv4_route_array[i].if_out);
 817                         }
 818                 }
 819
 820                 if (socket_lpm6[socket]) {
 821                         lpm6 = socket_lpm6[socket];
 822                         /* populate the LPM6 table */
 823                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 824                                 ret = rte_lpm6_add(lpm6,
 825                                         l3fwd_ipv6_route_array[i].ip,
 826                                         l3fwd_ipv6_route_array[i].depth,
 827                                         l3fwd_ipv6_route_array[i].if_out);
 828
 829                                 if (ret < 0) {
 830                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
 831                                                 "LPM6 table\n", i);
 832                                         return -1;
 833                                 }
 834
 835                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
 836                                                 "/%d (port %d)\n",
 837                                         socket,
 838                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 839                                         l3fwd_ipv6_route_array[i].depth,
 840                                         l3fwd_ipv6_route_array[i].if_out);
 841                         }
 842                 }
 843         }
 844         return 0;
 845 }
 846
 847 static int
 848 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
 849         uint32_t port)
 850 {
 851         struct mbuf_table *mtb;
 852         uint32_t n;
 853         size_t sz;
 854
 855         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
 856         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
 857
 858         if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
 859                         socket)) == NULL) {
 860                 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
 861                         "failed to allocate %zu bytes\n",
 862                         __func__, lcore, port, sz);
 863                 return -1;
 864         }
 865
 866         mtb->len = n;
 867         qconf->tx_mbufs[port] = mtb;
 868
 869         return 0;
 870 }
 871
 872 static int
 873 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
 874 {
 875         int socket;
 876         uint32_t nb_mbuf;
 877         uint64_t frag_cycles;
 878         char buf[RTE_MEMPOOL_NAMESIZE];
 879
 880         socket = rte_lcore_to_socket_id(lcore);
 881         if (socket == SOCKET_ID_ANY)
 882                 socket = 0;
 883
 884         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
 885                 max_flow_ttl;
 886
 887         if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
 888                         IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
 889                         socket)) == NULL) {
 890                 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
 891                         "lcore: %u for queue: %u failed\n",
 892                         max_flow_num, lcore, queue);
 893                 return -1;
 894         }
 895
 896         /*
 897          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
 898          * mbufs could be stored int the fragment table.
 899          * Plus, each TX queue can hold up to <max_flow_num> packets.
 900          */
 901
 902         nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
 903         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
 904         nb_mbuf *= 2; /* ipv4 and ipv6 */
 905         nb_mbuf += nb_rxd + nb_txd;
 906
 907         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
 908
 909         snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
 910
 911         rxq->pool = rte_pktmbuf_pool_create(buf, nb_mbuf, MEMPOOL_CACHE_SIZE, 0,
 912                                             MBUF_DATA_SIZE, socket);
 913         if (rxq->pool == NULL) {
 914                 RTE_LOG(ERR, IP_RSMBL,
 915                         "rte_pktmbuf_pool_create(%s) failed", buf);
 916                 return -1;
 917         }
 918
 919         return 0;
 920 }
 921
 922 static int
 923 init_mem(void)
 924 {
 925         char buf[PATH_MAX];
 926         struct rte_lpm *lpm;
 927         struct rte_lpm6 *lpm6;
 928         struct rte_lpm_config lpm_config;
 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_config.max_rules = LPM_MAX_RULES;
 949                         lpm_config.number_tbl8s = 256;
 950                         lpm_config.flags = 0;
 951
 952                         lpm = rte_lpm_create(buf, socket, &lpm_config);
 953                         if (lpm == NULL) {
 954                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
 955                                 return -1;
 956                         }
 957                         socket_lpm[socket] = lpm;
 958                 }
 959
 960                 if (socket_lpm6[socket] == NULL) {
 961                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
 962                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
 963
 964                         lpm6 = rte_lpm6_create(buf, socket, &lpm6_config);
 965                         if (lpm6 == NULL) {
 966                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
 967                                 return -1;
 968                         }
 969                         socket_lpm6[socket] = lpm6;
 970                 }
 971         }
 972
 973         return 0;
 974 }
 975
 976 static void
 977 queue_dump_stat(void)
 978 {
 979         uint32_t i, lcore;
 980         const struct lcore_queue_conf *qconf;
 981
 982         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
 983                 if (rte_lcore_is_enabled(lcore) == 0)
 984                         continue;
 985
 986                 qconf = &lcore_queue_conf[lcore];
 987                 for (i = 0; i < qconf->n_rx_queue; i++) {
 988
 989                         fprintf(stdout, " -- lcoreid=%u portid=%u "
 990                                 "frag tbl stat:\n",
 991                                 lcore,  qconf->rx_queue_list[i].portid);
 992                         rte_ip_frag_table_statistics_dump(stdout,
 993                                         qconf->rx_queue_list[i].frag_tbl);
 994                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
 995                                 "TX packets _queued:\t%" PRIu64 "\n"
 996                                 "TX packets dropped:\t%" PRIu64 "\n"
 997                                 "TX packets send:\t%" PRIu64 "\n",
 998                                 qconf->tx_stat.call,
 999                                 qconf->tx_stat.queue,
1000                                 qconf->tx_stat.drop,
1001                                 qconf->tx_stat.send);
1002                 }
1003         }
1004 }
1005
1006 static void
1007 signal_handler(int signum)
1008 {
1009         queue_dump_stat();
1010         if (signum != SIGUSR1)
1011                 rte_exit(0, "received signal: %d, exiting\n", signum);
1012 }
1013
1014 int
1015 main(int argc, char **argv)
1016 {
1017         struct lcore_queue_conf *qconf;
1018         struct rte_eth_dev_info dev_info;
1019         struct rte_eth_txconf *txconf;
1020         struct rx_queue *rxq;
1021         int ret, socket;
1022         unsigned nb_ports;
1023         uint16_t queueid;
1024         unsigned lcore_id = 0, rx_lcore_id = 0;
1025         uint32_t n_tx_queue, nb_lcores;
1026         uint16_t portid;
1027
1028         /* init EAL */
1029         ret = rte_eal_init(argc, argv);
1030         if (ret < 0)
1031                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1032         argc -= ret;
1033         argv += ret;
1034
1035         /* parse application arguments (after the EAL ones) */
1036         ret = parse_args(argc, argv);
1037         if (ret < 0)
1038                 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1039
1040         nb_ports = rte_eth_dev_count();
1041         if (nb_ports == 0)
1042                 rte_exit(EXIT_FAILURE, "No ports found!\n");
1043
1044         nb_lcores = rte_lcore_count();
1045
1046         /* initialize structures (mempools, lpm etc.) */
1047         if (init_mem() < 0)
1048                 rte_panic("Cannot initialize memory structures!\n");
1049
1050         /* check if portmask has non-existent ports */
1051         if (enabled_port_mask & ~(RTE_LEN2MASK(nb_ports, unsigned)))
1052                 rte_exit(EXIT_FAILURE, "Non-existent ports in portmask!\n");
1053
1054         /* initialize all ports */
1055         for (portid = 0; portid < nb_ports; portid++) {
1056                 /* skip ports that are not enabled */
1057                 if ((enabled_port_mask & (1 << portid)) == 0) {
1058                         printf("\nSkipping disabled port %d\n", portid);
1059                         continue;
1060                 }
1061
1062                 qconf = &lcore_queue_conf[rx_lcore_id];
1063
1064                 /* limit the frame size to the maximum supported by NIC */
1065                 rte_eth_dev_info_get(portid, &dev_info);
1066                 port_conf.rxmode.max_rx_pkt_len = RTE_MIN(
1067                     dev_info.max_rx_pktlen, port_conf.rxmode.max_rx_pkt_len);
1068
1069                 /* get the lcore_id for this port */
1070                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1071                            qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1072
1073                         rx_lcore_id++;
1074                         if (rx_lcore_id >= RTE_MAX_LCORE)
1075                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1076
1077                         qconf = &lcore_queue_conf[rx_lcore_id];
1078                 }
1079
1080                 socket = rte_lcore_to_socket_id(portid);
1081                 if (socket == SOCKET_ID_ANY)
1082                         socket = 0;
1083
1084                 queueid = qconf->n_rx_queue;
1085                 rxq = &qconf->rx_queue_list[queueid];
1086                 rxq->portid = portid;
1087                 rxq->lpm = socket_lpm[socket];
1088                 rxq->lpm6 = socket_lpm6[socket];
1089
1090                 ret = rte_eth_dev_adjust_nb_rx_tx_desc(portid, &nb_rxd,
1091                                                        &nb_txd);
1092                 if (ret < 0)
1093                         rte_exit(EXIT_FAILURE,
1094                                  "Cannot adjust number of descriptors: err=%d, port=%d\n",
1095                                  ret, portid);
1096
1097                 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1098                         rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1099                 qconf->n_rx_queue++;
1100
1101                 /* init port */
1102                 printf("Initializing port %d ... ", portid );
1103                 fflush(stdout);
1104
1105                 n_tx_queue = nb_lcores;
1106                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1107                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1108                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1109                                             &port_conf);
1110                 if (ret < 0) {
1111                         printf("\n");
1112                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1113                                 "err=%d, port=%d\n",
1114                                 ret, portid);
1115                 }
1116
1117                 /* init one RX queue */
1118                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1119                                              socket, NULL,
1120                                              rxq->pool);
1121                 if (ret < 0) {
1122                         printf("\n");
1123                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1124                                 "err=%d, port=%d\n",
1125                                 ret, portid);
1126                 }
1127
1128                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1129                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1130                 printf("\n");
1131
1132                 /* init one TX queue per couple (lcore,port) */
1133                 queueid = 0;
1134                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1135                         if (rte_lcore_is_enabled(lcore_id) == 0)
1136                                 continue;
1137
1138                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1139
1140                         printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1141                         fflush(stdout);
1142
1143                         txconf = &dev_info.default_txconf;
1144                         txconf->txq_flags = 0;
1145
1146                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1147                                         socket, txconf);
1148                         if (ret < 0)
1149                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1150                                         "port=%d\n", ret, portid);
1151
1152                         qconf = &lcore_queue_conf[lcore_id];
1153                         qconf->tx_queue_id[portid] = queueid;
1154                         setup_port_tbl(qconf, lcore_id, socket, portid);
1155                         queueid++;
1156                 }
1157                 printf("\n");
1158         }
1159
1160         printf("\n");
1161
1162         /* start ports */
1163         for (portid = 0; portid < nb_ports; portid++) {
1164                 if ((enabled_port_mask & (1 << portid)) == 0) {
1165                         continue;
1166                 }
1167                 /* Start device */
1168                 ret = rte_eth_dev_start(portid);
1169                 if (ret < 0)
1170                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1171                                 ret, portid);
1172
1173                 rte_eth_promiscuous_enable(portid);
1174         }
1175
1176         if (init_routing_table() < 0)
1177                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1178
1179         check_all_ports_link_status(nb_ports, enabled_port_mask);
1180
1181         signal(SIGUSR1, signal_handler);
1182         signal(SIGTERM, signal_handler);
1183         signal(SIGINT, signal_handler);
1184
1185         /* launch per-lcore init on every lcore */
1186         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1187         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1188                 if (rte_eal_wait_lcore(lcore_id) < 0)
1189                         return -1;
1190         }
1191
1192         return 0;
1193 }