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_tailq.h>
  54 #include <rte_eal.h>
  55 #include <rte_per_lcore.h>
  56 #include <rte_launch.h>
  57 #include <rte_atomic.h>
  58 #include <rte_cycles.h>
  59 #include <rte_prefetch.h>
  60 #include <rte_lcore.h>
  61 #include <rte_per_lcore.h>
  62 #include <rte_branch_prediction.h>
  63 #include <rte_interrupts.h>
  64 #include <rte_pci.h>
  65 #include <rte_random.h>
  66 #include <rte_debug.h>
  67 #include <rte_ether.h>
  68 #include <rte_ethdev.h>
  69 #include <rte_ring.h>
  70 #include <rte_mempool.h>
  71 #include <rte_mbuf.h>
  72 #include <rte_malloc.h>
  73 #include <rte_ip.h>
  74 #include <rte_tcp.h>
  75 #include <rte_udp.h>
  76 #include <rte_string_fns.h>
  77 #include <rte_lpm.h>
  78 #include <rte_lpm6.h>
  79
  80 #include <rte_ip_frag.h>
  81
  82 #define MAX_PKT_BURST 32
  83
  84
  85 #define RTE_LOGTYPE_IP_RSMBL RTE_LOGTYPE_USER1
  86
  87 #define MAX_JUMBO_PKT_LEN  9600
  88
  89 #define BUF_SIZE        2048
  90 #define MBUF_SIZE       \
  91         (BUF_SIZE + sizeof(struct rte_mbuf) + RTE_PKTMBUF_HEADROOM)
  92
  93 #define NB_MBUF 8192
  94
  95 /* allow max jumbo frame 9.5 KB */
  96 #define JUMBO_FRAME_MAX_SIZE    0x2600
  97
  98 #define MAX_FLOW_NUM    UINT16_MAX
  99 #define MIN_FLOW_NUM    1
 100 #define DEF_FLOW_NUM    0x1000
 101
 102 /* TTL numbers are in ms. */
 103 #define MAX_FLOW_TTL    (3600 * MS_PER_S)
 104 #define MIN_FLOW_TTL    1
 105 #define DEF_FLOW_TTL    MS_PER_S
 106
 107 #define MAX_FRAG_NUM RTE_LIBRTE_IP_FRAG_MAX_FRAG
 108
 109 /* Should be power of two. */
 110 #define IP_FRAG_TBL_BUCKET_ENTRIES      16
 111
 112 static uint32_t max_flow_num = DEF_FLOW_NUM;
 113 static uint32_t max_flow_ttl = DEF_FLOW_TTL;
 114
 115 #define BURST_TX_DRAIN_US 100 /* TX drain every ~100us */
 116
 117 #define NB_SOCKETS 8
 118
 119 /* Configure how many packets ahead to prefetch, when reading packets */
 120 #define PREFETCH_OFFSET 3
 121
 122 /*
 123  * Configurable number of RX/TX ring descriptors
 124  */
 125 #define RTE_TEST_RX_DESC_DEFAULT 128
 126 #define RTE_TEST_TX_DESC_DEFAULT 512
 127
 128 static uint16_t nb_rxd = RTE_TEST_RX_DESC_DEFAULT;
 129 static uint16_t nb_txd = RTE_TEST_TX_DESC_DEFAULT;
 130
 131 /* ethernet addresses of ports */
 132 static struct ether_addr ports_eth_addr[RTE_MAX_ETHPORTS];
 133
 134 #ifndef IPv4_BYTES
 135 #define IPv4_BYTES_FMT "%" PRIu8 ".%" PRIu8 ".%" PRIu8 ".%" PRIu8
 136 #define IPv4_BYTES(addr) \
 137                 (uint8_t) (((addr) >> 24) & 0xFF),\
 138                 (uint8_t) (((addr) >> 16) & 0xFF),\
 139                 (uint8_t) (((addr) >> 8) & 0xFF),\
 140                 (uint8_t) ((addr) & 0xFF)
 141 #endif
 142
 143 #ifndef IPv6_BYTES
 144 #define IPv6_BYTES_FMT "%02x%02x:%02x%02x:%02x%02x:%02x%02x:"\
 145                        "%02x%02x:%02x%02x:%02x%02x:%02x%02x"
 146 #define IPv6_BYTES(addr) \
 147         addr[0],  addr[1], addr[2],  addr[3], \
 148         addr[4],  addr[5], addr[6],  addr[7], \
 149         addr[8],  addr[9], addr[10], addr[11],\
 150         addr[12], addr[13],addr[14], addr[15]
 151 #endif
 152
 153 #define IPV6_ADDR_LEN 16
 154
 155 /* mask of enabled ports */
 156 static uint32_t enabled_port_mask = 0;
 157
 158 static int rx_queue_per_lcore = 1;
 159
 160 struct mbuf_table {
 161         uint32_t len;
 162         uint32_t head;
 163         uint32_t tail;
 164         struct rte_mbuf *m_table[0];
 165 };
 166
 167 struct rx_queue {
 168         struct rte_ip_frag_tbl *frag_tbl;
 169         struct rte_mempool *pool;
 170         struct rte_lpm *lpm;
 171         struct rte_lpm6 *lpm6;
 172         uint8_t portid;
 173 };
 174
 175 struct tx_lcore_stat {
 176         uint64_t call;
 177         uint64_t drop;
 178         uint64_t queue;
 179         uint64_t send;
 180 };
 181
 182 #define MAX_RX_QUEUE_PER_LCORE 16
 183 #define MAX_TX_QUEUE_PER_PORT 16
 184 #define MAX_RX_QUEUE_PER_PORT 128
 185
 186 struct lcore_queue_conf {
 187         uint16_t n_rx_queue;
 188         struct rx_queue rx_queue_list[MAX_RX_QUEUE_PER_LCORE];
 189         uint16_t tx_queue_id[RTE_MAX_ETHPORTS];
 190         struct rte_ip_frag_death_row death_row;
 191         struct mbuf_table *tx_mbufs[RTE_MAX_ETHPORTS];
 192         struct tx_lcore_stat tx_stat;
 193 } __rte_cache_aligned;
 194 static struct lcore_queue_conf lcore_queue_conf[RTE_MAX_LCORE];
 195
 196 static struct rte_eth_conf port_conf = {
 197         .rxmode = {
 198                 .mq_mode        = ETH_MQ_RX_RSS,
 199                 .max_rx_pkt_len = JUMBO_FRAME_MAX_SIZE,
 200                 .split_hdr_size = 0,
 201                 .header_split   = 0, /**< Header Split disabled */
 202                 .hw_ip_checksum = 1, /**< IP checksum offload enabled */
 203                 .hw_vlan_filter = 0, /**< VLAN filtering disabled */
 204                 .jumbo_frame    = 1, /**< Jumbo Frame Support disabled */
 205                 .hw_strip_crc   = 0, /**< CRC stripped by hardware */
 206         },
 207         .rx_adv_conf = {
 208                         .rss_conf = {
 209                                 .rss_key = NULL,
 210                                 .rss_hf = ETH_RSS_IP,
 211                 },
 212         },
 213         .txmode = {
 214                 .mq_mode = ETH_MQ_TX_NONE,
 215         },
 216 };
 217
 218 /*
 219  * IPv4 forwarding table
 220  */
 221 struct l3fwd_ipv4_route {
 222         uint32_t ip;
 223         uint8_t  depth;
 224         uint8_t  if_out;
 225 };
 226
 227 struct l3fwd_ipv4_route l3fwd_ipv4_route_array[] = {
 228                 {IPv4(100,10,0,0), 16, 0},
 229                 {IPv4(100,20,0,0), 16, 1},
 230                 {IPv4(100,30,0,0), 16, 2},
 231                 {IPv4(100,40,0,0), 16, 3},
 232                 {IPv4(100,50,0,0), 16, 4},
 233                 {IPv4(100,60,0,0), 16, 5},
 234                 {IPv4(100,70,0,0), 16, 6},
 235                 {IPv4(100,80,0,0), 16, 7},
 236 };
 237
 238 /*
 239  * IPv6 forwarding table
 240  */
 241
 242 struct l3fwd_ipv6_route {
 243         uint8_t ip[IPV6_ADDR_LEN];
 244         uint8_t depth;
 245         uint8_t if_out;
 246 };
 247
 248 static struct l3fwd_ipv6_route l3fwd_ipv6_route_array[] = {
 249         {{1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 0},
 250         {{2,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 1},
 251         {{3,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 2},
 252         {{4,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 3},
 253         {{5,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 4},
 254         {{6,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 5},
 255         {{7,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 6},
 256         {{8,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1}, 48, 7},
 257 };
 258
 259 #define LPM_MAX_RULES         1024
 260 #define LPM6_MAX_RULES         1024
 261 #define LPM6_NUMBER_TBL8S (1 << 16)
 262
 263 struct rte_lpm6_config lpm6_config = {
 264                 .max_rules = LPM6_MAX_RULES,
 265                 .number_tbl8s = LPM6_NUMBER_TBL8S,
 266                 .flags = 0
 267 };
 268
 269 static struct rte_lpm *socket_lpm[RTE_MAX_NUMA_NODES];
 270 static struct rte_lpm6 *socket_lpm6[RTE_MAX_NUMA_NODES];
 271
 272 #ifdef RTE_LIBRTE_IP_FRAG_TBL_STAT
 273 #define TX_LCORE_STAT_UPDATE(s, f, v)   ((s)->f += (v))
 274 #else
 275 #define TX_LCORE_STAT_UPDATE(s, f, v)   do {} while (0)
 276 #endif /* RTE_LIBRTE_IP_FRAG_TBL_STAT */
 277
 278 /*
 279  * If number of queued packets reached given threahold, then
 280  * send burst of packets on an output interface.
 281  */
 282 static inline uint32_t
 283 send_burst(struct lcore_queue_conf *qconf, uint32_t thresh, uint8_t port)
 284 {
 285         uint32_t fill, len, k, n;
 286         struct mbuf_table *txmb;
 287
 288         txmb = qconf->tx_mbufs[port];
 289         len = txmb->len;
 290
 291         if ((int32_t)(fill = txmb->head - txmb->tail) < 0)
 292                 fill += len;
 293
 294         if (fill >= thresh) {
 295                 n = RTE_MIN(len - txmb->tail, fill);
 296
 297                 k = rte_eth_tx_burst(port, qconf->tx_queue_id[port],
 298                         txmb->m_table + txmb->tail, (uint16_t)n);
 299
 300                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, call, 1);
 301                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, send, k);
 302
 303                 fill -= k;
 304                 if ((txmb->tail += k) == len)
 305                         txmb->tail = 0;
 306         }
 307
 308         return (fill);
 309 }
 310
 311 /* Enqueue a single packet, and send burst if queue is filled */
 312 static inline int
 313 send_single_packet(struct rte_mbuf *m, uint8_t port)
 314 {
 315         uint32_t fill, lcore_id, len;
 316         struct lcore_queue_conf *qconf;
 317         struct mbuf_table *txmb;
 318
 319         lcore_id = rte_lcore_id();
 320         qconf = &lcore_queue_conf[lcore_id];
 321
 322         txmb = qconf->tx_mbufs[port];
 323         len = txmb->len;
 324
 325         fill = send_burst(qconf, MAX_PKT_BURST, port);
 326
 327         if (fill == len - 1) {
 328                 TX_LCORE_STAT_UPDATE(&qconf->tx_stat, drop, 1);
 329                 rte_pktmbuf_free(txmb->m_table[txmb->tail]);
 330                 if (++txmb->tail == len)
 331                         txmb->tail = 0;
 332         }
 333
 334         TX_LCORE_STAT_UPDATE(&qconf->tx_stat, queue, 1);
 335         txmb->m_table[txmb->head] = m;
 336         if(++txmb->head == len)
 337                 txmb->head = 0;
 338
 339         return (0);
 340 }
 341
 342 static inline void
 343 reassemble(struct rte_mbuf *m, uint8_t portid, uint32_t queue,
 344         struct lcore_queue_conf *qconf, uint64_t tms)
 345 {
 346         struct ether_hdr *eth_hdr;
 347         struct rte_ip_frag_tbl *tbl;
 348         struct rte_ip_frag_death_row *dr;
 349         struct rx_queue *rxq;
 350         void *d_addr_bytes;
 351         uint8_t next_hop, dst_port;
 352
 353         rxq = &qconf->rx_queue_list[queue];
 354
 355         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 356
 357         dst_port = portid;
 358
 359         /* if packet is IPv4 */
 360         if (m->ol_flags & (PKT_RX_IPV4_HDR)) {
 361                 struct ipv4_hdr *ip_hdr;
 362                 uint32_t ip_dst;
 363
 364                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 365
 366                  /* if it is a fragmented packet, then try to reassemble. */
 367                 if (rte_ipv4_frag_pkt_is_fragmented(ip_hdr)) {
 368                         struct rte_mbuf *mo;
 369
 370                         tbl = rxq->frag_tbl;
 371                         dr = &qconf->death_row;
 372
 373                         /* prepare mbuf: setup l2_len/l3_len. */
 374                         m->l2_len = sizeof(*eth_hdr);
 375                         m->l3_len = sizeof(*ip_hdr);
 376
 377                         /* process this fragment. */
 378                         mo = rte_ipv4_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr);
 379                         if (mo == NULL)
 380                                 /* no packet to send out. */
 381                                 return;
 382
 383                         /* we have our packet reassembled. */
 384                         if (mo != m) {
 385                                 m = mo;
 386                                 eth_hdr = rte_pktmbuf_mtod(m,
 387                                         struct ether_hdr *);
 388                                 ip_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 389                         }
 390                 }
 391                 ip_dst = rte_be_to_cpu_32(ip_hdr->dst_addr);
 392
 393                 /* Find destination port */
 394                 if (rte_lpm_lookup(rxq->lpm, ip_dst, &next_hop) == 0 &&
 395                                 (enabled_port_mask & 1 << next_hop) != 0) {
 396                         dst_port = next_hop;
 397                 }
 398
 399                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv4);
 400         }
 401         /* if packet is IPv6 */
 402         else if (m->ol_flags & (PKT_RX_IPV6_HDR | PKT_RX_IPV6_HDR_EXT)) {
 403                 struct ipv6_extension_fragment *frag_hdr;
 404                 struct ipv6_hdr *ip_hdr;
 405
 406                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 407
 408                 frag_hdr = rte_ipv6_frag_get_ipv6_fragment_header(ip_hdr);
 409
 410                 if (frag_hdr != NULL) {
 411                         struct rte_mbuf *mo;
 412
 413                         tbl = rxq->frag_tbl;
 414                         dr  = &qconf->death_row;
 415
 416                         /* prepare mbuf: setup l2_len/l3_len. */
 417                         m->l2_len = sizeof(*eth_hdr);
 418                         m->l3_len = sizeof(*ip_hdr) + sizeof(*frag_hdr);
 419
 420                         mo = rte_ipv6_frag_reassemble_packet(tbl, dr, m, tms, ip_hdr, frag_hdr);
 421                         if (mo == NULL)
 422                                 return;
 423
 424                         if (mo != m) {
 425                                 m = mo;
 426                                 eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 427                                 ip_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 428                         }
 429                 }
 430
 431                 /* Find destination port */
 432                 if (rte_lpm6_lookup(rxq->lpm6, ip_hdr->dst_addr, &next_hop) == 0 &&
 433                                 (enabled_port_mask & 1 << next_hop) != 0) {
 434                         dst_port = next_hop;
 435                 }
 436
 437                 eth_hdr->ether_type = rte_be_to_cpu_16(ETHER_TYPE_IPv6);
 438         }
 439         /* if packet wasn't IPv4 or IPv6, it's forwarded to the port it came from */
 440
 441         /* 02:00:00:00:00:xx */
 442         d_addr_bytes = &eth_hdr->d_addr.addr_bytes[0];
 443         *((uint64_t *)d_addr_bytes) = 0x000000000002 + ((uint64_t)dst_port << 40);
 444
 445         /* src addr */
 446         ether_addr_copy(&ports_eth_addr[dst_port], &eth_hdr->s_addr);
 447
 448         send_single_packet(m, dst_port);
 449 }
 450
 451 /* main processing loop */
 452 static int
 453 main_loop(__attribute__((unused)) void *dummy)
 454 {
 455         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 456         unsigned lcore_id;
 457         uint64_t diff_tsc, cur_tsc, prev_tsc;
 458         int i, j, nb_rx;
 459         uint8_t portid;
 460         struct lcore_queue_conf *qconf;
 461         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) / US_PER_S * BURST_TX_DRAIN_US;
 462
 463         prev_tsc = 0;
 464
 465         lcore_id = rte_lcore_id();
 466         qconf = &lcore_queue_conf[lcore_id];
 467
 468         if (qconf->n_rx_queue == 0) {
 469                 RTE_LOG(INFO, IP_RSMBL, "lcore %u has nothing to do\n", lcore_id);
 470                 return 0;
 471         }
 472
 473         RTE_LOG(INFO, IP_RSMBL, "entering main loop on lcore %u\n", lcore_id);
 474
 475         for (i = 0; i < qconf->n_rx_queue; i++) {
 476
 477                 portid = qconf->rx_queue_list[i].portid;
 478                 RTE_LOG(INFO, IP_RSMBL, " -- lcoreid=%u portid=%hhu\n", lcore_id,
 479                         portid);
 480         }
 481
 482         while (1) {
 483
 484                 cur_tsc = rte_rdtsc();
 485
 486                 /*
 487                  * TX burst queue drain
 488                  */
 489                 diff_tsc = cur_tsc - prev_tsc;
 490                 if (unlikely(diff_tsc > drain_tsc)) {
 491
 492                         /*
 493                          * This could be optimized (use queueid instead of
 494                          * portid), but it is not called so often
 495                          */
 496                         for (portid = 0; portid < RTE_MAX_ETHPORTS; portid++) {
 497                                 if ((enabled_port_mask & (1 << portid)) != 0)
 498                                         send_burst(qconf, 1, portid);
 499                         }
 500
 501                         prev_tsc = cur_tsc;
 502                 }
 503
 504                 /*
 505                  * Read packet from RX queues
 506                  */
 507                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 508
 509                         portid = qconf->rx_queue_list[i].portid;
 510
 511                         nb_rx = rte_eth_rx_burst(portid, 0, pkts_burst,
 512                                 MAX_PKT_BURST);
 513
 514                         /* Prefetch first packets */
 515                         for (j = 0; j < PREFETCH_OFFSET && j < nb_rx; j++) {
 516                                 rte_prefetch0(rte_pktmbuf_mtod(
 517                                                 pkts_burst[j], void *));
 518                         }
 519
 520                         /* Prefetch and forward already prefetched packets */
 521                         for (j = 0; j < (nb_rx - PREFETCH_OFFSET); j++) {
 522                                 rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[
 523                                         j + PREFETCH_OFFSET], void *));
 524                                 reassemble(pkts_burst[j], portid,
 525                                         i, qconf, cur_tsc);
 526                         }
 527
 528                         /* Forward remaining prefetched packets */
 529                         for (; j < nb_rx; j++) {
 530                                 reassemble(pkts_burst[j], portid,
 531                                         i, qconf, cur_tsc);
 532                         }
 533
 534                         rte_ip_frag_free_death_row(&qconf->death_row,
 535                                 PREFETCH_OFFSET);
 536                 }
 537         }
 538 }
 539
 540 /* display usage */
 541 static void
 542 print_usage(const char *prgname)
 543 {
 544         printf("%s [EAL options] -- -p PORTMASK [-q NQ]"
 545                 "  [--max-pkt-len PKTLEN]"
 546                 "  [--maxflows=<flows>]  [--flowttl=<ttl>[(s|ms)]]\n"
 547                 "  -p PORTMASK: hexadecimal bitmask of ports to configure\n"
 548                 "  -q NQ: number of RX queues per lcore\n"
 549                 "  --maxflows=<flows>: optional, maximum number of flows "
 550                 "supported\n"
 551                 "  --flowttl=<ttl>[(s|ms)]: optional, maximum TTL for each "
 552                 "flow\n",
 553                 prgname);
 554 }
 555
 556 static uint32_t
 557 parse_flow_num(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 558 {
 559         char *end;
 560         uint64_t v;
 561
 562         /* parse decimal string */
 563         errno = 0;
 564         v = strtoul(str, &end, 10);
 565         if (errno != 0 || *end != '\0')
 566                 return (-EINVAL);
 567
 568         if (v < min || v > max)
 569                 return (-EINVAL);
 570
 571         *val = (uint32_t)v;
 572         return (0);
 573 }
 574
 575 static int
 576 parse_flow_ttl(const char *str, uint32_t min, uint32_t max, uint32_t *val)
 577 {
 578         char *end;
 579         uint64_t v;
 580
 581         static const char frmt_sec[] = "s";
 582         static const char frmt_msec[] = "ms";
 583
 584         /* parse decimal string */
 585         errno = 0;
 586         v = strtoul(str, &end, 10);
 587         if (errno != 0)
 588                 return (-EINVAL);
 589
 590         if (*end != '\0') {
 591                 if (strncmp(frmt_sec, end, sizeof(frmt_sec)) == 0)
 592                         v *= MS_PER_S;
 593                 else if (strncmp(frmt_msec, end, sizeof (frmt_msec)) != 0)
 594                         return (-EINVAL);
 595         }
 596
 597         if (v < min || v > max)
 598                 return (-EINVAL);
 599
 600         *val = (uint32_t)v;
 601         return (0);
 602 }
 603
 604 static int
 605 parse_portmask(const char *portmask)
 606 {
 607         char *end = NULL;
 608         unsigned long pm;
 609
 610         /* parse hexadecimal string */
 611         pm = strtoul(portmask, &end, 16);
 612         if ((portmask[0] == '\0') || (end == NULL) || (*end != '\0'))
 613                 return -1;
 614
 615         if (pm == 0)
 616                 return -1;
 617
 618         return pm;
 619 }
 620
 621 static int
 622 parse_nqueue(const char *q_arg)
 623 {
 624         char *end = NULL;
 625         unsigned long n;
 626
 627         printf("%p\n", q_arg);
 628
 629         /* parse hexadecimal string */
 630         n = strtoul(q_arg, &end, 10);
 631         if ((q_arg[0] == '\0') || (end == NULL) || (*end != '\0'))
 632                 return -1;
 633         if (n == 0)
 634                 return -1;
 635         if (n >= MAX_RX_QUEUE_PER_LCORE)
 636                 return -1;
 637
 638         return n;
 639 }
 640
 641 /* Parse the argument given in the command line of the application */
 642 static int
 643 parse_args(int argc, char **argv)
 644 {
 645         int opt, ret;
 646         char **argvopt;
 647         int option_index;
 648         char *prgname = argv[0];
 649         static struct option lgopts[] = {
 650                 {"max-pkt-len", 1, 0, 0},
 651                 {"maxflows", 1, 0, 0},
 652                 {"flowttl", 1, 0, 0},
 653                 {NULL, 0, 0, 0}
 654         };
 655
 656         argvopt = argv;
 657
 658         while ((opt = getopt_long(argc, argvopt, "p:q:",
 659                                 lgopts, &option_index)) != EOF) {
 660
 661                 switch (opt) {
 662                 /* portmask */
 663                 case 'p':
 664                         enabled_port_mask = parse_portmask(optarg);
 665                         if (enabled_port_mask == 0) {
 666                                 printf("invalid portmask\n");
 667                                 print_usage(prgname);
 668                                 return -1;
 669                         }
 670                         break;
 671
 672                 /* nqueue */
 673                 case 'q':
 674                         rx_queue_per_lcore = parse_nqueue(optarg);
 675                         if (rx_queue_per_lcore < 0) {
 676                                 printf("invalid queue number\n");
 677                                 print_usage(prgname);
 678                                 return -1;
 679                         }
 680                         break;
 681
 682                 /* long options */
 683                 case 0:
 684                         if (!strncmp(lgopts[option_index].name,
 685                                         "maxflows", 8)) {
 686                                 if ((ret = parse_flow_num(optarg, MIN_FLOW_NUM,
 687                                                 MAX_FLOW_NUM,
 688                                                 &max_flow_num)) != 0) {
 689                                         printf("invalid value: \"%s\" for "
 690                                                 "parameter %s\n",
 691                                                 optarg,
 692                                                 lgopts[option_index].name);
 693                                         print_usage(prgname);
 694                                         return (ret);
 695                                 }
 696                         }
 697
 698                         if (!strncmp(lgopts[option_index].name, "flowttl", 7)) {
 699                                 if ((ret = parse_flow_ttl(optarg, MIN_FLOW_TTL,
 700                                                 MAX_FLOW_TTL,
 701                                                 &max_flow_ttl)) != 0) {
 702                                         printf("invalid value: \"%s\" for "
 703                                                 "parameter %s\n",
 704                                                 optarg,
 705                                                 lgopts[option_index].name);
 706                                         print_usage(prgname);
 707                                         return (ret);
 708                                 }
 709                         }
 710
 711                         break;
 712
 713                 default:
 714                         print_usage(prgname);
 715                         return -1;
 716                 }
 717         }
 718
 719         if (optind >= 0)
 720                 argv[optind-1] = prgname;
 721
 722         ret = optind-1;
 723         optind = 0; /* reset getopt lib */
 724         return ret;
 725 }
 726
 727 static void
 728 print_ethaddr(const char *name, const struct ether_addr *eth_addr)
 729 {
 730         char buf[ETHER_ADDR_FMT_SIZE];
 731         ether_format_addr(buf, ETHER_ADDR_FMT_SIZE, eth_addr);
 732         printf("%s%s", name, buf);
 733 }
 734
 735 /* Check the link status of all ports in up to 9s, and print them finally */
 736 static void
 737 check_all_ports_link_status(uint8_t port_num, uint32_t port_mask)
 738 {
 739 #define CHECK_INTERVAL 100 /* 100ms */
 740 #define MAX_CHECK_TIME 90 /* 9s (90 * 100ms) in total */
 741         uint8_t portid, count, all_ports_up, print_flag = 0;
 742         struct rte_eth_link link;
 743
 744         printf("\nChecking link status");
 745         fflush(stdout);
 746         for (count = 0; count <= MAX_CHECK_TIME; count++) {
 747                 all_ports_up = 1;
 748                 for (portid = 0; portid < port_num; portid++) {
 749                         if ((port_mask & (1 << portid)) == 0)
 750                                 continue;
 751                         memset(&link, 0, sizeof(link));
 752                         rte_eth_link_get_nowait(portid, &link);
 753                         /* print link status if flag set */
 754                         if (print_flag == 1) {
 755                                 if (link.link_status)
 756                                         printf("Port %d Link Up - speed %u "
 757                                                 "Mbps - %s\n", (uint8_t)portid,
 758                                                 (unsigned)link.link_speed,
 759                                 (link.link_duplex == ETH_LINK_FULL_DUPLEX) ?
 760                                         ("full-duplex") : ("half-duplex\n"));
 761                                 else
 762                                         printf("Port %d Link Down\n",
 763                                                 (uint8_t)portid);
 764                                 continue;
 765                         }
 766                         /* clear all_ports_up flag if any link down */
 767                         if (link.link_status == 0) {
 768                                 all_ports_up = 0;
 769                                 break;
 770                         }
 771                 }
 772                 /* after finally printing all link status, get out */
 773                 if (print_flag == 1)
 774                         break;
 775
 776                 if (all_ports_up == 0) {
 777                         printf(".");
 778                         fflush(stdout);
 779                         rte_delay_ms(CHECK_INTERVAL);
 780                 }
 781
 782                 /* set the print_flag if all ports up or timeout */
 783                 if (all_ports_up == 1 || count == (MAX_CHECK_TIME - 1)) {
 784                         print_flag = 1;
 785                         printf("\ndone\n");
 786                 }
 787         }
 788 }
 789
 790 static int
 791 init_routing_table(void)
 792 {
 793         struct rte_lpm *lpm;
 794         struct rte_lpm6 *lpm6;
 795         int socket, ret;
 796         unsigned i;
 797
 798         for (socket = 0; socket < RTE_MAX_NUMA_NODES; socket++) {
 799                 if (socket_lpm[socket]) {
 800                         lpm = socket_lpm[socket];
 801                         /* populate the LPM table */
 802                         for (i = 0; i < RTE_DIM(l3fwd_ipv4_route_array); i++) {
 803                                 ret = rte_lpm_add(lpm,
 804                                         l3fwd_ipv4_route_array[i].ip,
 805                                         l3fwd_ipv4_route_array[i].depth,
 806                                         l3fwd_ipv4_route_array[i].if_out);
 807
 808                                 if (ret < 0) {
 809                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
 810                                                 "LPM table\n", i);
 811                                         return -1;
 812                                 }
 813
 814                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv4_BYTES_FMT
 815                                                 "/%d (port %d)\n",
 816                                         socket,
 817                                         IPv4_BYTES(l3fwd_ipv4_route_array[i].ip),
 818                                         l3fwd_ipv4_route_array[i].depth,
 819                                         l3fwd_ipv4_route_array[i].if_out);
 820                         }
 821                 }
 822
 823                 if (socket_lpm6[socket]) {
 824                         lpm6 = socket_lpm6[socket];
 825                         /* populate the LPM6 table */
 826                         for (i = 0; i < RTE_DIM(l3fwd_ipv6_route_array); i++) {
 827                                 ret = rte_lpm6_add(lpm6,
 828                                         l3fwd_ipv6_route_array[i].ip,
 829                                         l3fwd_ipv6_route_array[i].depth,
 830                                         l3fwd_ipv6_route_array[i].if_out);
 831
 832                                 if (ret < 0) {
 833                                         RTE_LOG(ERR, IP_RSMBL, "Unable to add entry %i to the l3fwd "
 834                                                 "LPM6 table\n", i);
 835                                         return -1;
 836                                 }
 837
 838                                 RTE_LOG(INFO, IP_RSMBL, "Socket %i: adding route " IPv6_BYTES_FMT
 839                                                 "/%d (port %d)\n",
 840                                         socket,
 841                                         IPv6_BYTES(l3fwd_ipv6_route_array[i].ip),
 842                                         l3fwd_ipv6_route_array[i].depth,
 843                                         l3fwd_ipv6_route_array[i].if_out);
 844                         }
 845                 }
 846         }
 847         return 0;
 848 }
 849
 850 static int
 851 setup_port_tbl(struct lcore_queue_conf *qconf, uint32_t lcore, int socket,
 852         uint32_t port)
 853 {
 854         struct mbuf_table *mtb;
 855         uint32_t n;
 856         size_t sz;
 857
 858         n = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST);
 859         sz = sizeof (*mtb) + sizeof (mtb->m_table[0]) *  n;
 860
 861         if ((mtb = rte_zmalloc_socket(__func__, sz, RTE_CACHE_LINE_SIZE,
 862                         socket)) == NULL) {
 863                 RTE_LOG(ERR, IP_RSMBL, "%s() for lcore: %u, port: %u "
 864                         "failed to allocate %zu bytes\n",
 865                         __func__, lcore, port, sz);
 866                 return -1;
 867         }
 868
 869         mtb->len = n;
 870         qconf->tx_mbufs[port] = mtb;
 871
 872         return 0;
 873 }
 874
 875 static int
 876 setup_queue_tbl(struct rx_queue *rxq, uint32_t lcore, uint32_t queue)
 877 {
 878         int socket;
 879         uint32_t nb_mbuf;
 880         uint64_t frag_cycles;
 881         char buf[RTE_MEMPOOL_NAMESIZE];
 882
 883         socket = rte_lcore_to_socket_id(lcore);
 884         if (socket == SOCKET_ID_ANY)
 885                 socket = 0;
 886
 887         frag_cycles = (rte_get_tsc_hz() + MS_PER_S - 1) / MS_PER_S *
 888                 max_flow_ttl;
 889
 890         if ((rxq->frag_tbl = rte_ip_frag_table_create(max_flow_num,
 891                         IP_FRAG_TBL_BUCKET_ENTRIES, max_flow_num, frag_cycles,
 892                         socket)) == NULL) {
 893                 RTE_LOG(ERR, IP_RSMBL, "ip_frag_tbl_create(%u) on "
 894                         "lcore: %u for queue: %u failed\n",
 895                         max_flow_num, lcore, queue);
 896                 return -1;
 897         }
 898
 899         /*
 900          * At any given moment up to <max_flow_num * (MAX_FRAG_NUM)>
 901          * mbufs could be stored int the fragment table.
 902          * Plus, each TX queue can hold up to <max_flow_num> packets.
 903          */
 904
 905         nb_mbuf = RTE_MAX(max_flow_num, 2UL * MAX_PKT_BURST) * MAX_FRAG_NUM;
 906         nb_mbuf *= (port_conf.rxmode.max_rx_pkt_len + BUF_SIZE - 1) / BUF_SIZE;
 907         nb_mbuf *= 2; /* ipv4 and ipv6 */
 908         nb_mbuf += RTE_TEST_RX_DESC_DEFAULT + RTE_TEST_TX_DESC_DEFAULT;
 909
 910         nb_mbuf = RTE_MAX(nb_mbuf, (uint32_t)NB_MBUF);
 911
 912         snprintf(buf, sizeof(buf), "mbuf_pool_%u_%u", lcore, queue);
 913
 914         if ((rxq->pool = rte_mempool_create(buf, nb_mbuf, MBUF_SIZE, 0,
 915                         sizeof(struct rte_pktmbuf_pool_private),
 916                         rte_pktmbuf_pool_init, NULL, rte_pktmbuf_init, NULL,
 917                         socket, MEMPOOL_F_SP_PUT | MEMPOOL_F_SC_GET)) == NULL) {
 918                 RTE_LOG(ERR, IP_RSMBL, "mempool_create(%s) failed", buf);
 919                 return -1;
 920         }
 921
 922         return 0;
 923 }
 924
 925 static int
 926 init_mem(void)
 927 {
 928         char buf[PATH_MAX];
 929         struct rte_lpm *lpm;
 930         struct rte_lpm6 *lpm6;
 931         int socket;
 932         unsigned lcore_id;
 933
 934         /* traverse through lcores and initialize structures on each socket */
 935
 936         for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
 937
 938                 if (rte_lcore_is_enabled(lcore_id) == 0)
 939                         continue;
 940
 941                 socket = rte_lcore_to_socket_id(lcore_id);
 942
 943                 if (socket == SOCKET_ID_ANY)
 944                         socket = 0;
 945
 946                 if (socket_lpm[socket] == NULL) {
 947                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM table on socket %i\n", socket);
 948                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
 949
 950                         lpm = rte_lpm_create(buf, socket, LPM_MAX_RULES, 0);
 951                         if (lpm == NULL) {
 952                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
 953                                 return -1;
 954                         }
 955                         socket_lpm[socket] = lpm;
 956                 }
 957
 958                 if (socket_lpm6[socket] == NULL) {
 959                         RTE_LOG(INFO, IP_RSMBL, "Creating LPM6 table on socket %i\n", socket);
 960                         snprintf(buf, sizeof(buf), "IP_RSMBL_LPM_%i", socket);
 961
 962                         lpm6 = rte_lpm6_create("IP_RSMBL_LPM6", socket, &lpm6_config);
 963                         if (lpm6 == NULL) {
 964                                 RTE_LOG(ERR, IP_RSMBL, "Cannot create LPM table\n");
 965                                 return -1;
 966                         }
 967                         socket_lpm6[socket] = lpm6;
 968                 }
 969         }
 970
 971         return 0;
 972 }
 973
 974 static void
 975 queue_dump_stat(void)
 976 {
 977         uint32_t i, lcore;
 978         const struct lcore_queue_conf *qconf;
 979
 980         for (lcore = 0; lcore < RTE_MAX_LCORE; lcore++) {
 981                 if (rte_lcore_is_enabled(lcore) == 0)
 982                         continue;
 983
 984                 qconf = &lcore_queue_conf[lcore];
 985                 for (i = 0; i < qconf->n_rx_queue; i++) {
 986
 987                         fprintf(stdout, " -- lcoreid=%u portid=%hhu "
 988                                 "frag tbl stat:\n",
 989                                 lcore,  qconf->rx_queue_list[i].portid);
 990                         rte_ip_frag_table_statistics_dump(stdout,
 991                                         qconf->rx_queue_list[i].frag_tbl);
 992                         fprintf(stdout, "TX bursts:\t%" PRIu64 "\n"
 993                                 "TX packets _queued:\t%" PRIu64 "\n"
 994                                 "TX packets dropped:\t%" PRIu64 "\n"
 995                                 "TX packets send:\t%" PRIu64 "\n",
 996                                 qconf->tx_stat.call,
 997                                 qconf->tx_stat.queue,
 998                                 qconf->tx_stat.drop,
 999                                 qconf->tx_stat.send);
1000                 }
1001         }
1002 }
1003
1004 static void
1005 signal_handler(int signum)
1006 {
1007         queue_dump_stat();
1008         if (signum != SIGUSR1)
1009                 rte_exit(0, "received signal: %d, exiting\n", signum);
1010 }
1011
1012 int
1013 main(int argc, char **argv)
1014 {
1015         struct lcore_queue_conf *qconf;
1016         struct rte_eth_dev_info dev_info;
1017         struct rte_eth_txconf *txconf;
1018         struct rx_queue *rxq;
1019         int ret, socket;
1020         unsigned nb_ports;
1021         uint16_t queueid;
1022         unsigned lcore_id = 0, rx_lcore_id = 0;
1023         uint32_t n_tx_queue, nb_lcores;
1024         uint8_t portid;
1025
1026         /* init EAL */
1027         ret = rte_eal_init(argc, argv);
1028         if (ret < 0)
1029                 rte_exit(EXIT_FAILURE, "Invalid EAL parameters\n");
1030         argc -= ret;
1031         argv += ret;
1032
1033         /* parse application arguments (after the EAL ones) */
1034         ret = parse_args(argc, argv);
1035         if (ret < 0)
1036                 rte_exit(EXIT_FAILURE, "Invalid IP reassembly parameters\n");
1037
1038         nb_ports = rte_eth_dev_count();
1039         if (nb_ports > RTE_MAX_ETHPORTS)
1040                 nb_ports = RTE_MAX_ETHPORTS;
1041         else 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                 /* get the lcore_id for this port */
1065                 while (rte_lcore_is_enabled(rx_lcore_id) == 0 ||
1066                            qconf->n_rx_queue == (unsigned)rx_queue_per_lcore) {
1067
1068                         rx_lcore_id++;
1069                         if (rx_lcore_id >= RTE_MAX_LCORE)
1070                                 rte_exit(EXIT_FAILURE, "Not enough cores\n");
1071
1072                         qconf = &lcore_queue_conf[rx_lcore_id];
1073                 }
1074
1075                 socket = rte_lcore_to_socket_id(portid);
1076                 if (socket == SOCKET_ID_ANY)
1077                         socket = 0;
1078
1079                 queueid = qconf->n_rx_queue;
1080                 rxq = &qconf->rx_queue_list[queueid];
1081                 rxq->portid = portid;
1082                 rxq->lpm = socket_lpm[socket];
1083                 rxq->lpm6 = socket_lpm6[socket];
1084                 if (setup_queue_tbl(rxq, rx_lcore_id, queueid) < 0)
1085                         rte_exit(EXIT_FAILURE, "Failed to set up queue table\n");
1086                 qconf->n_rx_queue++;
1087
1088                 /* init port */
1089                 printf("Initializing port %d ... ", portid );
1090                 fflush(stdout);
1091
1092                 n_tx_queue = nb_lcores;
1093                 if (n_tx_queue > MAX_TX_QUEUE_PER_PORT)
1094                         n_tx_queue = MAX_TX_QUEUE_PER_PORT;
1095                 ret = rte_eth_dev_configure(portid, 1, (uint16_t)n_tx_queue,
1096                                             &port_conf);
1097                 if (ret < 0) {
1098                         printf("\n");
1099                         rte_exit(EXIT_FAILURE, "Cannot configure device: "
1100                                 "err=%d, port=%d\n",
1101                                 ret, portid);
1102                 }
1103
1104                 /* init one RX queue */
1105                 ret = rte_eth_rx_queue_setup(portid, 0, nb_rxd,
1106                                              socket, NULL,
1107                                              rxq->pool);
1108                 if (ret < 0) {
1109                         printf("\n");
1110                         rte_exit(EXIT_FAILURE, "rte_eth_rx_queue_setup: "
1111                                 "err=%d, port=%d\n",
1112                                 ret, portid);
1113                 }
1114
1115                 rte_eth_macaddr_get(portid, &ports_eth_addr[portid]);
1116                 print_ethaddr(" Address:", &ports_eth_addr[portid]);
1117                 printf("\n");
1118
1119                 /* init one TX queue per couple (lcore,port) */
1120                 queueid = 0;
1121                 for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
1122                         if (rte_lcore_is_enabled(lcore_id) == 0)
1123                                 continue;
1124
1125                         socket = (int) rte_lcore_to_socket_id(lcore_id);
1126
1127                         printf("txq=%u,%d,%d ", lcore_id, queueid, socket);
1128                         fflush(stdout);
1129
1130                         rte_eth_dev_info_get(portid, &dev_info);
1131                         txconf = &dev_info.default_txconf;
1132                         txconf->txq_flags = 0;
1133
1134                         ret = rte_eth_tx_queue_setup(portid, queueid, nb_txd,
1135                                         socket, txconf);
1136                         if (ret < 0)
1137                                 rte_exit(EXIT_FAILURE, "rte_eth_tx_queue_setup: err=%d, "
1138                                         "port=%d\n", ret, portid);
1139
1140                         qconf = &lcore_queue_conf[lcore_id];
1141                         qconf->tx_queue_id[portid] = queueid;
1142                         setup_port_tbl(qconf, lcore_id, socket, portid);
1143                         queueid++;
1144                 }
1145                 printf("\n");
1146         }
1147
1148         printf("\n");
1149
1150         /* start ports */
1151         for (portid = 0; portid < nb_ports; portid++) {
1152                 if ((enabled_port_mask & (1 << portid)) == 0) {
1153                         continue;
1154                 }
1155                 /* Start device */
1156                 ret = rte_eth_dev_start(portid);
1157                 if (ret < 0)
1158                         rte_exit(EXIT_FAILURE, "rte_eth_dev_start: err=%d, port=%d\n",
1159                                 ret, portid);
1160
1161                 rte_eth_promiscuous_enable(portid);
1162         }
1163
1164         if (init_routing_table() < 0)
1165                 rte_exit(EXIT_FAILURE, "Cannot init routing table\n");
1166
1167         check_all_ports_link_status((uint8_t)nb_ports, enabled_port_mask);
1168
1169         signal(SIGUSR1, signal_handler);
1170         signal(SIGTERM, signal_handler);
1171         signal(SIGINT, signal_handler);
1172
1173         /* launch per-lcore init on every lcore */
1174         rte_eal_mp_remote_launch(main_loop, NULL, CALL_MASTER);
1175         RTE_LCORE_FOREACH_SLAVE(lcore_id) {
1176                 if (rte_eal_wait_lcore(lcore_id) < 0)
1177                         return -1;
1178         }
1179
1180         return 0;
1181 }