examples/l3fwd/l3fwd_lpm.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2016 Intel Corporation
   3  */
   4
   5 #include <stdio.h>
   6 #include <stdlib.h>
   7 #include <stdint.h>
   8 #include <inttypes.h>
   9 #include <sys/types.h>
  10 #include <string.h>
  11 #include <sys/queue.h>
  12 #include <stdarg.h>
  13 #include <errno.h>
  14 #include <getopt.h>
  15 #include <stdbool.h>
  16 #include <arpa/inet.h>
  17
  18 #include <rte_debug.h>
  19 #include <rte_ether.h>
  20 #include <rte_ethdev.h>
  21 #include <rte_cycles.h>
  22 #include <rte_mbuf.h>
  23 #include <rte_ip.h>
  24 #include <rte_tcp.h>
  25 #include <rte_udp.h>
  26 #include <rte_lpm.h>
  27 #include <rte_lpm6.h>
  28
  29 #include "l3fwd.h"
  30
  31 struct ipv4_l3fwd_lpm_route {
  32         uint32_t ip;
  33         uint8_t  depth;
  34         uint8_t  if_out;
  35 };
  36
  37 struct ipv6_l3fwd_lpm_route {
  38         uint8_t ip[16];
  39         uint8_t  depth;
  40         uint8_t  if_out;
  41 };
  42
  43 /* 192.18.0.0/16 are set aside for RFC2544 benchmarking. */
  44 static struct ipv4_l3fwd_lpm_route ipv4_l3fwd_lpm_route_array[] = {
  45         {IPv4(192, 18, 0, 0), 24, 0},
  46         {IPv4(192, 18, 1, 0), 24, 1},
  47         {IPv4(192, 18, 2, 0), 24, 2},
  48         {IPv4(192, 18, 3, 0), 24, 3},
  49         {IPv4(192, 18, 4, 0), 24, 4},
  50         {IPv4(192, 18, 5, 0), 24, 5},
  51         {IPv4(192, 18, 6, 0), 24, 6},
  52         {IPv4(192, 18, 7, 0), 24, 7},
  53 };
  54
  55 /* 2001:0200::/48 is IANA reserved range for IPv6 benchmarking (RFC5180) */
  56 static struct ipv6_l3fwd_lpm_route ipv6_l3fwd_lpm_route_array[] = {
  57         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}, 48, 0},
  58         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0}, 48, 1},
  59         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0}, 48, 2},
  60         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0}, 48, 3},
  61         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0}, 48, 4},
  62         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 5, 0, 0, 0, 0, 0, 0}, 48, 5},
  63         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 6, 0, 0, 0, 0, 0, 0}, 48, 6},
  64         {{32, 1, 2, 0, 0, 0, 0, 0, 0, 7, 0, 0, 0, 0, 0, 0}, 48, 7},
  65 };
  66
  67 #define IPV4_L3FWD_LPM_NUM_ROUTES \
  68         (sizeof(ipv4_l3fwd_lpm_route_array) / sizeof(ipv4_l3fwd_lpm_route_array[0]))
  69 #define IPV6_L3FWD_LPM_NUM_ROUTES \
  70         (sizeof(ipv6_l3fwd_lpm_route_array) / sizeof(ipv6_l3fwd_lpm_route_array[0]))
  71
  72 #define IPV4_L3FWD_LPM_MAX_RULES         1024
  73 #define IPV4_L3FWD_LPM_NUMBER_TBL8S (1 << 8)
  74 #define IPV6_L3FWD_LPM_MAX_RULES         1024
  75 #define IPV6_L3FWD_LPM_NUMBER_TBL8S (1 << 16)
  76
  77 struct rte_lpm *ipv4_l3fwd_lpm_lookup_struct[NB_SOCKETS];
  78 struct rte_lpm6 *ipv6_l3fwd_lpm_lookup_struct[NB_SOCKETS];
  79
  80 static inline uint16_t
  81 lpm_get_ipv4_dst_port(void *ipv4_hdr, uint16_t portid, void *lookup_struct)
  82 {
  83         uint32_t next_hop;
  84         struct rte_lpm *ipv4_l3fwd_lookup_struct =
  85                 (struct rte_lpm *)lookup_struct;
  86
  87         return (uint16_t) ((rte_lpm_lookup(ipv4_l3fwd_lookup_struct,
  88                 rte_be_to_cpu_32(((struct ipv4_hdr *)ipv4_hdr)->dst_addr),
  89                 &next_hop) == 0) ? next_hop : portid);
  90 }
  91
  92 static inline uint16_t
  93 lpm_get_ipv6_dst_port(void *ipv6_hdr, uint16_t portid, void *lookup_struct)
  94 {
  95         uint32_t next_hop;
  96         struct rte_lpm6 *ipv6_l3fwd_lookup_struct =
  97                 (struct rte_lpm6 *)lookup_struct;
  98
  99         return (uint16_t) ((rte_lpm6_lookup(ipv6_l3fwd_lookup_struct,
 100                         ((struct ipv6_hdr *)ipv6_hdr)->dst_addr,
 101                         &next_hop) == 0) ?  next_hop : portid);
 102 }
 103
 104 static __rte_always_inline uint16_t
 105 lpm_get_dst_port(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
 106                 uint16_t portid)
 107 {
 108         struct ipv6_hdr *ipv6_hdr;
 109         struct ipv4_hdr *ipv4_hdr;
 110         struct ether_hdr *eth_hdr;
 111
 112         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
 113
 114                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 115                 ipv4_hdr = (struct ipv4_hdr *)(eth_hdr + 1);
 116
 117                 return lpm_get_ipv4_dst_port(ipv4_hdr, portid,
 118                                              qconf->ipv4_lookup_struct);
 119         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
 120
 121                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 122                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 123
 124                 return lpm_get_ipv6_dst_port(ipv6_hdr, portid,
 125                                              qconf->ipv6_lookup_struct);
 126         }
 127
 128         return portid;
 129 }
 130
 131 /*
 132  * lpm_get_dst_port optimized routine for packets where dst_ipv4 is already
 133  * precalculated. If packet is ipv6 dst_addr is taken directly from packet
 134  * header and dst_ipv4 value is not used.
 135  */
 136 static __rte_always_inline uint16_t
 137 lpm_get_dst_port_with_ipv4(const struct lcore_conf *qconf, struct rte_mbuf *pkt,
 138         uint32_t dst_ipv4, uint16_t portid)
 139 {
 140         uint32_t next_hop;
 141         struct ipv6_hdr *ipv6_hdr;
 142         struct ether_hdr *eth_hdr;
 143
 144         if (RTE_ETH_IS_IPV4_HDR(pkt->packet_type)) {
 145                 return (uint16_t) ((rte_lpm_lookup(qconf->ipv4_lookup_struct,
 146                                                    dst_ipv4, &next_hop) == 0)
 147                                    ? next_hop : portid);
 148
 149         } else if (RTE_ETH_IS_IPV6_HDR(pkt->packet_type)) {
 150
 151                 eth_hdr = rte_pktmbuf_mtod(pkt, struct ether_hdr *);
 152                 ipv6_hdr = (struct ipv6_hdr *)(eth_hdr + 1);
 153
 154                 return (uint16_t) ((rte_lpm6_lookup(qconf->ipv6_lookup_struct,
 155                                 ipv6_hdr->dst_addr, &next_hop) == 0)
 156                                 ? next_hop : portid);
 157
 158         }
 159
 160         return portid;
 161 }
 162
 163 #if defined(RTE_ARCH_X86)
 164 #include "l3fwd_lpm_sse.h"
 165 #elif defined RTE_MACHINE_CPUFLAG_NEON
 166 #include "l3fwd_lpm_neon.h"
 167 #elif defined(RTE_ARCH_PPC_64)
 168 #include "l3fwd_lpm_altivec.h"
 169 #else
 170 #include "l3fwd_lpm.h"
 171 #endif
 172
 173 /* main processing loop */
 174 int
 175 lpm_main_loop(__attribute__((unused)) void *dummy)
 176 {
 177         struct rte_mbuf *pkts_burst[MAX_PKT_BURST];
 178         unsigned lcore_id;
 179         uint64_t prev_tsc, diff_tsc, cur_tsc;
 180         int i, nb_rx;
 181         uint16_t portid;
 182         uint8_t queueid;
 183         struct lcore_conf *qconf;
 184         const uint64_t drain_tsc = (rte_get_tsc_hz() + US_PER_S - 1) /
 185                 US_PER_S * BURST_TX_DRAIN_US;
 186
 187         prev_tsc = 0;
 188
 189         lcore_id = rte_lcore_id();
 190         qconf = &lcore_conf[lcore_id];
 191
 192         if (qconf->n_rx_queue == 0) {
 193                 RTE_LOG(INFO, L3FWD, "lcore %u has nothing to do\n", lcore_id);
 194                 return 0;
 195         }
 196
 197         RTE_LOG(INFO, L3FWD, "entering main loop on lcore %u\n", lcore_id);
 198
 199         for (i = 0; i < qconf->n_rx_queue; i++) {
 200
 201                 portid = qconf->rx_queue_list[i].port_id;
 202                 queueid = qconf->rx_queue_list[i].queue_id;
 203                 RTE_LOG(INFO, L3FWD,
 204                         " -- lcoreid=%u portid=%u rxqueueid=%hhu\n",
 205                         lcore_id, portid, queueid);
 206         }
 207
 208         while (!force_quit) {
 209
 210                 cur_tsc = rte_rdtsc();
 211
 212                 /*
 213                  * TX burst queue drain
 214                  */
 215                 diff_tsc = cur_tsc - prev_tsc;
 216                 if (unlikely(diff_tsc > drain_tsc)) {
 217
 218                         for (i = 0; i < qconf->n_tx_port; ++i) {
 219                                 portid = qconf->tx_port_id[i];
 220                                 if (qconf->tx_mbufs[portid].len == 0)
 221                                         continue;
 222                                 send_burst(qconf,
 223                                         qconf->tx_mbufs[portid].len,
 224                                         portid);
 225                                 qconf->tx_mbufs[portid].len = 0;
 226                         }
 227
 228                         prev_tsc = cur_tsc;
 229                 }
 230
 231                 /*
 232                  * Read packet from RX queues
 233                  */
 234                 for (i = 0; i < qconf->n_rx_queue; ++i) {
 235                         portid = qconf->rx_queue_list[i].port_id;
 236                         queueid = qconf->rx_queue_list[i].queue_id;
 237                         nb_rx = rte_eth_rx_burst(portid, queueid, pkts_burst,
 238                                 MAX_PKT_BURST);
 239                         if (nb_rx == 0)
 240                                 continue;
 241
 242 #if defined RTE_ARCH_X86 || defined RTE_MACHINE_CPUFLAG_NEON \
 243                          || defined RTE_ARCH_PPC_64
 244                         l3fwd_lpm_send_packets(nb_rx, pkts_burst,
 245                                                 portid, qconf);
 246 #else
 247                         l3fwd_lpm_no_opt_send_packets(nb_rx, pkts_burst,
 248                                                         portid, qconf);
 249 #endif /* X86 */
 250                 }
 251         }
 252
 253         return 0;
 254 }
 255
 256 void
 257 setup_lpm(const int socketid)
 258 {
 259         struct rte_lpm6_config config;
 260         struct rte_lpm_config config_ipv4;
 261         unsigned i;
 262         int ret;
 263         char s[64];
 264         char abuf[INET6_ADDRSTRLEN];
 265
 266         /* create the LPM table */
 267         config_ipv4.max_rules = IPV4_L3FWD_LPM_MAX_RULES;
 268         config_ipv4.number_tbl8s = IPV4_L3FWD_LPM_NUMBER_TBL8S;
 269         config_ipv4.flags = 0;
 270         snprintf(s, sizeof(s), "IPV4_L3FWD_LPM_%d", socketid);
 271         ipv4_l3fwd_lpm_lookup_struct[socketid] =
 272                         rte_lpm_create(s, socketid, &config_ipv4);
 273         if (ipv4_l3fwd_lpm_lookup_struct[socketid] == NULL)
 274                 rte_exit(EXIT_FAILURE,
 275                         "Unable to create the l3fwd LPM table on socket %d\n",
 276                         socketid);
 277
 278         /* populate the LPM table */
 279         for (i = 0; i < IPV4_L3FWD_LPM_NUM_ROUTES; i++) {
 280                 struct in_addr in;
 281
 282                 /* skip unused ports */
 283                 if ((1 << ipv4_l3fwd_lpm_route_array[i].if_out &
 284                                 enabled_port_mask) == 0)
 285                         continue;
 286
 287                 ret = rte_lpm_add(ipv4_l3fwd_lpm_lookup_struct[socketid],
 288                         ipv4_l3fwd_lpm_route_array[i].ip,
 289                         ipv4_l3fwd_lpm_route_array[i].depth,
 290                         ipv4_l3fwd_lpm_route_array[i].if_out);
 291
 292                 if (ret < 0) {
 293                         rte_exit(EXIT_FAILURE,
 294                                 "Unable to add entry %u to the l3fwd LPM table on socket %d\n",
 295                                 i, socketid);
 296                 }
 297
 298                 in.s_addr = htonl(ipv4_l3fwd_lpm_route_array[i].ip);
 299                 printf("LPM: Adding route %s / %d (%d)\n",
 300                        inet_ntop(AF_INET, &in, abuf, sizeof(abuf)),
 301                         ipv4_l3fwd_lpm_route_array[i].depth,
 302                         ipv4_l3fwd_lpm_route_array[i].if_out);
 303         }
 304
 305         /* create the LPM6 table */
 306         snprintf(s, sizeof(s), "IPV6_L3FWD_LPM_%d", socketid);
 307
 308         config.max_rules = IPV6_L3FWD_LPM_MAX_RULES;
 309         config.number_tbl8s = IPV6_L3FWD_LPM_NUMBER_TBL8S;
 310         config.flags = 0;
 311         ipv6_l3fwd_lpm_lookup_struct[socketid] = rte_lpm6_create(s, socketid,
 312                                 &config);
 313         if (ipv6_l3fwd_lpm_lookup_struct[socketid] == NULL)
 314                 rte_exit(EXIT_FAILURE,
 315                         "Unable to create the l3fwd LPM table on socket %d\n",
 316                         socketid);
 317
 318         /* populate the LPM table */
 319         for (i = 0; i < IPV6_L3FWD_LPM_NUM_ROUTES; i++) {
 320
 321                 /* skip unused ports */
 322                 if ((1 << ipv6_l3fwd_lpm_route_array[i].if_out &
 323                                 enabled_port_mask) == 0)
 324                         continue;
 325
 326                 ret = rte_lpm6_add(ipv6_l3fwd_lpm_lookup_struct[socketid],
 327                         ipv6_l3fwd_lpm_route_array[i].ip,
 328                         ipv6_l3fwd_lpm_route_array[i].depth,
 329                         ipv6_l3fwd_lpm_route_array[i].if_out);
 330
 331                 if (ret < 0) {
 332                         rte_exit(EXIT_FAILURE,
 333                                 "Unable to add entry %u to the l3fwd LPM table on socket %d\n",
 334                                 i, socketid);
 335                 }
 336
 337                 printf("LPM: Adding route %s / %d (%d)\n",
 338                        inet_ntop(AF_INET6, ipv6_l3fwd_lpm_route_array[i].ip,
 339                                  abuf, sizeof(abuf)),
 340                        ipv6_l3fwd_lpm_route_array[i].depth,
 341                        ipv6_l3fwd_lpm_route_array[i].if_out);
 342         }
 343 }
 344
 345 int
 346 lpm_check_ptype(int portid)
 347 {
 348         int i, ret;
 349         int ptype_l3_ipv4 = 0, ptype_l3_ipv6 = 0;
 350         uint32_t ptype_mask = RTE_PTYPE_L3_MASK;
 351
 352         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, NULL, 0);
 353         if (ret <= 0)
 354                 return 0;
 355
 356         uint32_t ptypes[ret];
 357
 358         ret = rte_eth_dev_get_supported_ptypes(portid, ptype_mask, ptypes, ret);
 359         for (i = 0; i < ret; ++i) {
 360                 if (ptypes[i] & RTE_PTYPE_L3_IPV4)
 361                         ptype_l3_ipv4 = 1;
 362                 if (ptypes[i] & RTE_PTYPE_L3_IPV6)
 363                         ptype_l3_ipv6 = 1;
 364         }
 365
 366         if (ptype_l3_ipv4 == 0)
 367                 printf("port %d cannot parse RTE_PTYPE_L3_IPV4\n", portid);
 368
 369         if (ptype_l3_ipv6 == 0)
 370                 printf("port %d cannot parse RTE_PTYPE_L3_IPV6\n", portid);
 371
 372         if (ptype_l3_ipv4 && ptype_l3_ipv6)
 373                 return 1;
 374
 375         return 0;
 376
 377 }
 378
 379 static inline void
 380 lpm_parse_ptype(struct rte_mbuf *m)
 381 {
 382         struct ether_hdr *eth_hdr;
 383         uint32_t packet_type = RTE_PTYPE_UNKNOWN;
 384         uint16_t ether_type;
 385
 386         eth_hdr = rte_pktmbuf_mtod(m, struct ether_hdr *);
 387         ether_type = eth_hdr->ether_type;
 388         if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv4))
 389                 packet_type |= RTE_PTYPE_L3_IPV4_EXT_UNKNOWN;
 390         else if (ether_type == rte_cpu_to_be_16(ETHER_TYPE_IPv6))
 391                 packet_type |= RTE_PTYPE_L3_IPV6_EXT_UNKNOWN;
 392
 393         m->packet_type = packet_type;
 394 }
 395
 396 uint16_t
 397 lpm_cb_parse_ptype(uint16_t port __rte_unused, uint16_t queue __rte_unused,
 398                    struct rte_mbuf *pkts[], uint16_t nb_pkts,
 399                    uint16_t max_pkts __rte_unused,
 400                    void *user_param __rte_unused)
 401 {
 402         unsigned i;
 403
 404         for (i = 0; i < nb_pkts; ++i)
 405                 lpm_parse_ptype(pkts[i]);
 406
 407         return nb_pkts;
 408 }
 409
 410 /* Return ipv4/ipv6 lpm fwd lookup struct. */
 411 void *
 412 lpm_get_ipv4_l3fwd_lookup_struct(const int socketid)
 413 {
 414         return ipv4_l3fwd_lpm_lookup_struct[socketid];
 415 }
 416
 417 void *
 418 lpm_get_ipv6_l3fwd_lookup_struct(const int socketid)
 419 {
 420         return ipv6_l3fwd_lpm_lookup_struct[socketid];
 421 }