zebra/rt_socket.c

   1 /*
   2  * Kernel routing table updates by routing socket.
   3  * Copyright (C) 1997, 98 Kunihiro Ishiguro
   4  *
   5  * This file is part of GNU Zebra.
   6  *
   7  * GNU Zebra is free software; you can redistribute it and/or modify it
   8  * under the terms of the GNU General Public License as published by the
   9  * Free Software Foundation; either version 2, or (at your option) any
  10  * later version.
  11  *
  12  * GNU Zebra is distributed in the hope that it will be useful, but
  13  * WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License along
  18  * with this program; see the file COPYING; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #include <zebra.h>
  23
  24 #ifndef HAVE_NETLINK
  25
  26 #ifdef __OpenBSD__
  27 #include <netmpls/mpls.h>
  28 #endif
  29
  30 #include "if.h"
  31 #include "prefix.h"
  32 #include "sockunion.h"
  33 #include "log.h"
  34 #include "privs.h"
  35 #include "vxlan.h"
  36 #include "lib_errors.h"
  37
  38 #include "zebra/debug.h"
  39 #include "zebra/rib.h"
  40 #include "zebra/rt.h"
  41 #include "zebra/kernel_socket.h"
  42 #include "zebra/zebra_mpls.h"
  43 #include "zebra/zebra_errors.h"
  44
  45 extern struct zebra_privs_t zserv_privs;
  46
  47 #ifdef __OpenBSD__
  48 static int kernel_rtm_add_labels(struct mpls_label_stack *nh_label,
  49                                  struct sockaddr_mpls *smpls)
  50 {
  51         if (nh_label->num_labels > 1) {
  52                 flog_warn(EC_ZEBRA_MAX_LABELS_PUSH,
  53                           "%s: can't push %u labels at "
  54                           "once (maximum is 1)",
  55                           __func__, nh_label->num_labels);
  56                 return -1;
  57         }
  58
  59         memset(smpls, 0, sizeof(*smpls));
  60         smpls->smpls_len = sizeof(*smpls);
  61         smpls->smpls_family = AF_MPLS;
  62         smpls->smpls_label = htonl(nh_label->label[0] << MPLS_LABEL_OFFSET);
  63
  64         return 0;
  65 }
  66 #endif
  67
  68 /* Interface between zebra message and rtm message. */
  69 static int kernel_rtm(int cmd, const struct prefix *p,
  70                       const struct nexthop_group *ng, uint32_t metric)
  71
  72 {
  73         union sockunion sin_dest, sin_mask, sin_gate;
  74 #ifdef __OpenBSD__
  75         struct sockaddr_mpls smpls;
  76 #endif
  77         union sockunion *smplsp = NULL;
  78         struct nexthop *nexthop;
  79         int nexthop_num = 0;
  80         ifindex_t ifindex = 0;
  81         bool gate = false;
  82         int error;
  83         char gate_buf[INET6_BUFSIZ];
  84         char prefix_buf[PREFIX_STRLEN];
  85         enum blackhole_type bh_type = BLACKHOLE_UNSPEC;
  86
  87         prefix2str(p, prefix_buf, sizeof(prefix_buf));
  88
  89         /*
  90          * We only have the ability to ADD or DELETE at this point
  91          * in time.
  92          */
  93         if (cmd != RTM_ADD && cmd != RTM_DELETE) {
  94                 if (IS_ZEBRA_DEBUG_KERNEL)
  95                         zlog_debug("%s: %s odd command %s",
  96                                    __func__, prefix_buf,
  97                                    lookup_msg(rtm_type_str, cmd, NULL));
  98                 return 0;
  99         }
 100
 101         memset(&sin_dest, 0, sizeof(sin_dest));
 102         memset(&sin_gate, 0, sizeof(sin_gate));
 103         memset(&sin_mask, 0, sizeof(sin_mask));
 104
 105         switch (p->family) {
 106         case AF_INET:
 107                 sin_dest.sin.sin_family = AF_INET;
 108                 sin_dest.sin.sin_addr = p->u.prefix4;
 109                 sin_gate.sin.sin_family = AF_INET;
 110 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
 111                 sin_dest.sin.sin_len = sizeof(struct sockaddr_in);
 112                 sin_gate.sin.sin_len = sizeof(struct sockaddr_in);
 113 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
 114                 break;
 115         case AF_INET6:
 116                 sin_dest.sin6.sin6_family = AF_INET6;
 117                 sin_dest.sin6.sin6_addr = p->u.prefix6;
 118                 sin_gate.sin6.sin6_family = AF_INET6;
 119 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
 120                 sin_dest.sin6.sin6_len = sizeof(struct sockaddr_in6);
 121                 sin_gate.sin6.sin6_len = sizeof(struct sockaddr_in6);
 122 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
 123                 break;
 124         }
 125
 126         /* Make gateway. */
 127         for (ALL_NEXTHOPS_PTR(ng, nexthop)) {
 128                 /*
 129                  * We only want to use the actual good nexthops
 130                  */
 131                 if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE) ||
 132                     !CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_ACTIVE))
 133                         continue;
 134
 135                 smplsp = NULL;
 136                 gate = false;
 137                 snprintf(gate_buf, sizeof(gate_buf), "NULL");
 138
 139                 switch (nexthop->type) {
 140                 case NEXTHOP_TYPE_IPV4:
 141                 case NEXTHOP_TYPE_IPV4_IFINDEX:
 142                         sin_gate.sin.sin_addr = nexthop->gate.ipv4;
 143                         sin_gate.sin.sin_family = AF_INET;
 144 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
 145                         sin_gate.sin.sin_len = sizeof(struct sockaddr_in);
 146 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
 147                         ifindex = nexthop->ifindex;
 148                         gate = true;
 149                         break;
 150                 case NEXTHOP_TYPE_IPV6:
 151                 case NEXTHOP_TYPE_IPV6_IFINDEX:
 152                         sin_gate.sin6.sin6_addr = nexthop->gate.ipv6;
 153                         sin_gate.sin6.sin6_family = AF_INET6;
 154 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
 155                         sin_gate.sin6.sin6_len = sizeof(struct sockaddr_in6);
 156 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
 157                         ifindex = nexthop->ifindex;
 158 /* Under kame set interface index to link local address */
 159 #ifdef KAME
 160
 161 #define SET_IN6_LINKLOCAL_IFINDEX(a, i)                                        \
 162         do {                                                                   \
 163                 (a).s6_addr[2] = ((i) >> 8) & 0xff;                            \
 164                 (a).s6_addr[3] = (i)&0xff;                                     \
 165         } while (0)
 166
 167                         if (IN6_IS_ADDR_LINKLOCAL(&sin_gate.sin6.sin6_addr))
 168                                 SET_IN6_LINKLOCAL_IFINDEX(
 169                                         sin_gate.sin6.sin6_addr,
 170                                         ifindex);
 171 #endif /* KAME */
 172
 173                         gate = true;
 174                         break;
 175                 case NEXTHOP_TYPE_IFINDEX:
 176                         ifindex = nexthop->ifindex;
 177                         break;
 178                 case NEXTHOP_TYPE_BLACKHOLE:
 179                         bh_type = nexthop->bh_type;
 180                         switch (p->family) {
 181                         case AF_INET: {
 182                                 struct in_addr loopback;
 183
 184                                 loopback.s_addr = htonl(INADDR_LOOPBACK);
 185                                 sin_gate.sin.sin_addr = loopback;
 186 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
 187                                 sin_gate.sin.sin_len =
 188                                         sizeof(struct sockaddr_in);
 189 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
 190                                 gate = true;
 191                         } break;
 192                         case AF_INET6: {
 193                                 struct in6_addr loopback;
 194
 195                                 inet_pton(AF_INET6, "::1", &loopback);
 196
 197                                 sin_gate.sin6.sin6_addr = loopback;
 198                                 sin_gate.sin6.sin6_family = AF_INET6;
 199
 200 #ifdef HAVE_STRUCTSOCKADDR_SA_LEN
 201                                 sin_gate.sin6.sin6_len =
 202                                         sizeof(struct sockaddr_in6);
 203 #endif /* HAVE_STRUCTSOCKADDR_SA_LEN */
 204                                 gate = true;
 205                         } break;
 206                         }
 207                 }
 208
 209                 switch (p->family) {
 210                 case AF_INET:
 211                         masklen2ip(p->prefixlen, &sin_mask.sin.sin_addr);
 212                         sin_mask.sin.sin_family = AF_INET;
 213 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
 214                         sin_mask.sin.sin_len = sizeof(struct sockaddr_in);
 215 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
 216                         break;
 217                 case AF_INET6:
 218                         masklen2ip6(p->prefixlen, &sin_mask.sin6.sin6_addr);
 219                         sin_mask.sin6.sin6_family = AF_INET6;
 220 #ifdef HAVE_STRUCT_SOCKADDR_SA_LEN
 221                         sin_mask.sin6.sin6_len = sizeof(struct sockaddr_in6);
 222 #endif /* HAVE_STRUCT_SOCKADDR_SA_LEN */
 223                         break;
 224                 }
 225
 226 #ifdef __OpenBSD__
 227                 if (nexthop->nh_label) {
 228                         if (kernel_rtm_add_labels(nexthop->nh_label,
 229                                                   &smpls) != 0)
 230                                 continue;
 231                         smplsp = (union sockunion *)&smpls;
 232                 }
 233 #endif
 234                 error = rtm_write(cmd, &sin_dest, &sin_mask,
 235                                   gate ? &sin_gate : NULL, smplsp,
 236                                   ifindex, bh_type, metric);
 237
 238                 if (IS_ZEBRA_DEBUG_KERNEL) {
 239                         if (!gate) {
 240                                 zlog_debug(
 241                                         "%s: %s: attention! gate not found for re",
 242                                         __func__, prefix_buf);
 243                         } else {
 244                                 switch (p->family) {
 245                                 case AF_INET:
 246                                         inet_ntop(AF_INET,
 247                                                   &sin_gate.sin.sin_addr,
 248                                                   gate_buf, sizeof(gate_buf));
 249                                         break;
 250
 251                                 case AF_INET6:
 252                                         inet_ntop(AF_INET6,
 253                                                   &sin_gate.sin6.sin6_addr,
 254                                                   gate_buf, sizeof(gate_buf));
 255                                         break;
 256
 257                                 default:
 258                                         snprintf(gate_buf, sizeof(gate_buf),
 259                                                  "(invalid-af)");
 260                                         break;
 261                                 }
 262                         }
 263                 }
 264                 switch (error) {
 265                         /* We only flag nexthops as being in FIB if
 266                          * rtm_write() did its work. */
 267                 case ZEBRA_ERR_NOERROR:
 268                         nexthop_num++;
 269                         if (IS_ZEBRA_DEBUG_KERNEL)
 270                                 zlog_debug("%s: %s: successfully did NH %s",
 271                                            __func__, prefix_buf, gate_buf);
 272                         if (cmd == RTM_ADD)
 273                                 SET_FLAG(nexthop->flags, NEXTHOP_FLAG_FIB);
 274                         break;
 275
 276                         /* The only valid case for this error is
 277                          * kernel's failure to install a multipath
 278                          * route, which is common for FreeBSD. This
 279                          * should be ignored silently, but logged as an error
 280                          * otherwise.
 281                          */
 282                 case ZEBRA_ERR_RTEXIST:
 283                         if (cmd != RTM_ADD)
 284                                 flog_err(EC_LIB_SYSTEM_CALL,
 285                                          "%s: rtm_write() returned %d for command %d",
 286                                          __func__, error, cmd);
 287                         continue;
 288
 289                         /* Note any unexpected status returns */
 290                 default:
 291                         flog_err(
 292                                 EC_LIB_SYSTEM_CALL,
 293                                 "%s: %s: rtm_write() unexpectedly returned %d for command %s",
 294                                 __func__, prefix_buf, error,
 295                                 lookup_msg(rtm_type_str, cmd, NULL));
 296                         break;
 297                 }
 298         } /* for (ALL_NEXTHOPS(...))*/
 299
 300         /* If there was no useful nexthop, then complain. */
 301         if (nexthop_num == 0) {
 302                 if (IS_ZEBRA_DEBUG_KERNEL)
 303                         zlog_debug(
 304                                 "%s: No useful nexthops were found in RIB prefix %s",
 305                                 __func__, prefix_buf);
 306                 return 1;
 307         }
 308
 309         return 0; /*XXX*/
 310 }
 311
 312 /*
 313  * Update or delete a prefix from the kernel,
 314  * using info from a dataplane context struct.
 315  */
 316 enum zebra_dplane_result kernel_route_update(struct zebra_dplane_ctx *ctx)
 317 {
 318         enum zebra_dplane_result res = ZEBRA_DPLANE_REQUEST_SUCCESS;
 319         uint32_t type, old_type;
 320
 321         if (dplane_ctx_get_src(ctx) != NULL) {
 322                 zlog_err("route add: IPv6 sourcedest routes unsupported!");
 323                 return ZEBRA_DPLANE_REQUEST_FAILURE;
 324         }
 325
 326         type = dplane_ctx_get_type(ctx);
 327         old_type = dplane_ctx_get_old_type(ctx);
 328
 329         frr_with_privs(&zserv_privs) {
 330
 331                 if (dplane_ctx_get_op(ctx) == DPLANE_OP_ROUTE_DELETE) {
 332                         if (!RSYSTEM_ROUTE(type))
 333                                 kernel_rtm(RTM_DELETE, dplane_ctx_get_dest(ctx),
 334                                            dplane_ctx_get_ng(ctx),
 335                                            dplane_ctx_get_metric(ctx));
 336                 } else if (dplane_ctx_get_op(ctx) == DPLANE_OP_ROUTE_INSTALL) {
 337                         if (!RSYSTEM_ROUTE(type))
 338                                 kernel_rtm(RTM_ADD, dplane_ctx_get_dest(ctx),
 339                                            dplane_ctx_get_ng(ctx),
 340                                            dplane_ctx_get_metric(ctx));
 341                 } else if (dplane_ctx_get_op(ctx) == DPLANE_OP_ROUTE_UPDATE) {
 342                         /* Must do delete and add separately -
 343                          * no update available
 344                          */
 345                         if (!RSYSTEM_ROUTE(old_type))
 346                                 kernel_rtm(RTM_DELETE, dplane_ctx_get_dest(ctx),
 347                                            dplane_ctx_get_old_ng(ctx),
 348                                            dplane_ctx_get_old_metric(ctx));
 349
 350                         if (!RSYSTEM_ROUTE(type))
 351                                 kernel_rtm(RTM_ADD, dplane_ctx_get_dest(ctx),
 352                                            dplane_ctx_get_ng(ctx),
 353                                            dplane_ctx_get_metric(ctx));
 354                 } else {
 355                         zlog_err("Invalid routing socket update op %s (%u)",
 356                                  dplane_op2str(dplane_ctx_get_op(ctx)),
 357                                  dplane_ctx_get_op(ctx));
 358                         res = ZEBRA_DPLANE_REQUEST_FAILURE;
 359                 }
 360         } /* Elevated privs */
 361
 362         if (RSYSTEM_ROUTE(type)
 363             && dplane_ctx_get_op(ctx) != DPLANE_OP_ROUTE_DELETE) {
 364                 struct nexthop *nexthop;
 365
 366                 for (ALL_NEXTHOPS_PTR(dplane_ctx_get_ng(ctx), nexthop)) {
 367                         if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_RECURSIVE))
 368                                 continue;
 369
 370                         if (CHECK_FLAG(nexthop->flags, NEXTHOP_FLAG_ACTIVE)) {
 371                                 SET_FLAG(nexthop->flags, NEXTHOP_FLAG_FIB);
 372                         }
 373                 }
 374         }
 375
 376         return res;
 377 }
 378
 379 enum zebra_dplane_result kernel_nexthop_update(struct zebra_dplane_ctx *ctx)
 380 {
 381         return ZEBRA_DPLANE_REQUEST_SUCCESS;
 382 }
 383
 384 int kernel_neigh_update(int add, int ifindex, uint32_t addr, char *lla,
 385                         int llalen, ns_id_t ns_id)
 386 {
 387         /* TODO */
 388         return 0;
 389 }
 390
 391 /* NYI on routing-socket platforms, but we've always returned 'success'... */
 392 enum zebra_dplane_result kernel_neigh_update_ctx(struct zebra_dplane_ctx *ctx)
 393 {
 394         return ZEBRA_DPLANE_REQUEST_SUCCESS;
 395 }
 396
 397 extern int kernel_get_ipmr_sg_stats(struct zebra_vrf *zvrf, void *mroute)
 398 {
 399         return 0;
 400 }
 401
 402 /*
 403  * Update MAC, using dataplane context object. No-op here for now.
 404  */
 405 enum zebra_dplane_result kernel_mac_update_ctx(struct zebra_dplane_ctx *ctx)
 406 {
 407         return ZEBRA_DPLANE_REQUEST_SUCCESS;
 408 }
 409
 410 extern int kernel_interface_set_master(struct interface *master,
 411                                        struct interface *slave)
 412 {
 413         return 0;
 414 }
 415
 416 uint32_t kernel_get_speed(struct interface *ifp, int *error)
 417 {
 418         return ifp->speed;
 419 }
 420
 421 #endif /* !HAVE_NETLINK */