src/ndp.c

   1 /**
   2  * Copyright (C) 2012-2013 Steven Barth <steven@midlink.org>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License v2 as published by
   6  * the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  */
  14
  15 #include <stdio.h>
  16 #include <stdlib.h>
  17 #include <signal.h>
  18 #include <errno.h>
  19
  20 #include <fcntl.h>
  21 #include <unistd.h>
  22 #include <arpa/inet.h>
  23 #include <sys/socket.h>
  24 #include <net/ethernet.h>
  25 #include <netinet/ip6.h>
  26 #include <netinet/icmp6.h>
  27 #include <netpacket/packet.h>
  28
  29 #include <linux/rtnetlink.h>
  30 #include <linux/filter.h>
  31 #include "router.h"
  32 #include "ndp.h"
  33
  34
  35
  36 static void handle_solicit(void *addr, void *data, size_t len,
  37                 struct interface *iface, void *dest);
  38 static void handle_rtnetlink(void *addr, void *data, size_t len,
  39                 struct interface *iface, void *dest);
  40 static ssize_t ping6(struct in6_addr *addr,
  41                 const struct interface *iface);
  42
  43 static uint32_t rtnl_seqid = 0;
  44 static int ping_socket = -1;
  45 static struct odhcpd_event rtnl_event = {{.fd = -1}, handle_rtnetlink};
  46
  47
  48 // Filter ICMPv6 messages of type neighbor soliciation
  49 static struct sock_filter bpf[] = {
  50         BPF_STMT(BPF_LD | BPF_B | BPF_ABS, offsetof(struct ip6_hdr, ip6_nxt)),
  51         BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, IPPROTO_ICMPV6, 0, 3),
  52         BPF_STMT(BPF_LD | BPF_B | BPF_ABS, sizeof(struct ip6_hdr) +
  53                         offsetof(struct icmp6_hdr, icmp6_type)),
  54         BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, ND_NEIGHBOR_SOLICIT, 0, 1),
  55         BPF_STMT(BPF_RET | BPF_K, 0xffffffff),
  56         BPF_STMT(BPF_RET | BPF_K, 0),
  57 };
  58 static const struct sock_fprog bpf_prog = {sizeof(bpf) / sizeof(*bpf), bpf};
  59
  60
  61 // Initialize NDP-proxy
  62 int init_ndp(void)
  63 {
  64         // Setup netlink socket
  65         if ((rtnl_event.uloop.fd = odhcpd_open_rtnl()) < 0)
  66                 return -1;
  67
  68         // Receive netlink neighbor and ip-address events
  69         uint32_t group = RTNLGRP_IPV6_IFADDR;
  70         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
  71                         NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
  72         group = RTNLGRP_IPV6_ROUTE;
  73         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
  74                         NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
  75
  76         odhcpd_register(&rtnl_event);
  77
  78         // Open ICMPv6 socket
  79         ping_socket = socket(AF_INET6, SOCK_RAW | SOCK_CLOEXEC, IPPROTO_ICMPV6);
  80         if (ping_socket < 0) {
  81                 syslog(LOG_ERR, "Unable to open raw socket: %s", strerror(errno));
  82                         return -1;
  83         }
  84
  85         int val = 2;
  86         setsockopt(ping_socket, IPPROTO_RAW, IPV6_CHECKSUM, &val, sizeof(val));
  87
  88         // This is required by RFC 4861
  89         val = 255;
  90         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &val, sizeof(val));
  91         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &val, sizeof(val));
  92
  93         // Filter all packages, we only want to send
  94         struct icmp6_filter filt;
  95         ICMP6_FILTER_SETBLOCKALL(&filt);
  96         setsockopt(ping_socket, IPPROTO_ICMPV6, ICMP6_FILTER, &filt, sizeof(filt));
  97
  98
  99         // Netlink socket, continued...
 100         group = RTNLGRP_NEIGH;
 101         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
 102
 103         return 0;
 104 }
 105
 106
 107 static void dump_neigh_table(bool proxy)
 108 {
 109         struct {
 110                 struct nlmsghdr nh;
 111                 struct ndmsg ndm;
 112         } req = {
 113                 {sizeof(req), RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP,
 114                                 ++rtnl_seqid, 0},
 115                 {.ndm_family = AF_INET6, .ndm_flags = (proxy) ? NTF_PROXY : 0}
 116         };
 117         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 118         odhcpd_process(&rtnl_event);
 119 }
 120
 121
 122 int setup_ndp_interface(struct interface *iface, bool enable)
 123 {
 124         char procbuf[64];
 125         snprintf(procbuf, sizeof(procbuf), "/proc/sys/net/ipv6/conf/%s/proxy_ndp", iface->ifname);
 126         int procfd = open(procbuf, O_WRONLY);
 127         bool dump_neigh = false;
 128
 129         if (iface->ndp_event.uloop.fd > 0) {
 130                 uloop_fd_delete(&iface->ndp_event.uloop);
 131                 close(iface->ndp_event.uloop.fd);
 132                 iface->ndp_event.uloop.fd = -1;
 133
 134                 if (!enable || iface->ndp != RELAYD_RELAY)
 135                         if (write(procfd, "0\n", 2) < 0) {}
 136
 137                 dump_neigh = true;
 138         }
 139
 140         if (enable && (iface->ra == RELAYD_SERVER ||
 141                         iface->dhcpv6 == RELAYD_SERVER || iface->ndp == RELAYD_RELAY)) {
 142                 // Synthesize initial address events
 143                 struct {
 144                         struct nlmsghdr nh;
 145                         struct ifaddrmsg ifa;
 146                 } req2 = {
 147                         {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
 148                                         ++rtnl_seqid, 0},
 149                         {.ifa_family = AF_INET6, .ifa_index = iface->ifindex}
 150                 };
 151                 send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
 152         }
 153
 154         if (enable && iface->ndp == RELAYD_RELAY) {
 155                 if (write(procfd, "1\n", 2) < 0) {}
 156                 close(procfd);
 157
 158                 int sock = socket(AF_PACKET, SOCK_DGRAM | SOCK_CLOEXEC, htons(ETH_P_IPV6));
 159                 if (sock < 0) {
 160                         syslog(LOG_ERR, "Unable to open packet socket: %s",
 161                                         strerror(errno));
 162                         return -1;
 163                 }
 164
 165 #ifdef PACKET_RECV_TYPE
 166                 int pktt = 1 << PACKET_MULTICAST;
 167                 setsockopt(sock, SOL_PACKET, PACKET_RECV_TYPE, &pktt, sizeof(pktt));
 168 #endif
 169
 170                 if (setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER,
 171                                 &bpf_prog, sizeof(bpf_prog))) {
 172                         syslog(LOG_ERR, "Failed to set BPF: %s", strerror(errno));
 173                         return -1;
 174                 }
 175
 176                 struct sockaddr_ll ll = {
 177                         .sll_family = AF_PACKET,
 178                         .sll_ifindex = iface->ifindex,
 179                         .sll_protocol = htons(ETH_P_IPV6),
 180                         .sll_hatype = 0,
 181                         .sll_pkttype = 0,
 182                         .sll_halen = 0,
 183                         .sll_addr = {0},
 184                 };
 185                 bind(sock, (struct sockaddr*)&ll, sizeof(ll));
 186
 187                 struct packet_mreq mreq = {iface->ifindex, PACKET_MR_ALLMULTI, ETH_ALEN, {0}};
 188                 setsockopt(sock, SOL_PACKET, PACKET_ADD_MEMBERSHIP, &mreq, sizeof(mreq));
 189
 190                 iface->ndp_event.uloop.fd = sock;
 191                 iface->ndp_event.handle_dgram = handle_solicit;
 192                 odhcpd_register(&iface->ndp_event);
 193
 194                 // If we already were enabled dump is unnecessary, if not do dump
 195                 if (!dump_neigh)
 196                         dump_neigh_table(false);
 197                 else
 198                         dump_neigh = false;
 199         } else {
 200                 close(procfd);
 201         }
 202
 203         if (dump_neigh)
 204                 dump_neigh_table(true);
 205
 206         return 0;
 207 }
 208
 209
 210 // Send an ICMP-ECHO. This is less for actually pinging but for the
 211 // neighbor cache to be kept up-to-date.
 212 static ssize_t ping6(struct in6_addr *addr,
 213                 const struct interface *iface)
 214 {
 215         struct sockaddr_in6 dest = {AF_INET6, 0, 0, *addr, iface->ifindex};
 216         struct icmp6_hdr echo = {.icmp6_type = ICMP6_ECHO_REQUEST};
 217         struct iovec iov = {&echo, sizeof(echo)};
 218
 219         // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
 220         setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
 221                         iface->ifname, sizeof(iface->ifname));
 222         return odhcpd_send(ping_socket, &dest, &iov, 1, iface);
 223 }
 224
 225
 226 // Handle solicitations
 227 static void handle_solicit(void *addr, void *data, size_t len,
 228                 struct interface *iface, _unused void *dest)
 229 {
 230         struct ip6_hdr *ip6 = data;
 231         struct nd_neighbor_solicit *req = (struct nd_neighbor_solicit*)&ip6[1];
 232         struct sockaddr_ll *ll = addr;
 233
 234         // Solicitation is for duplicate address detection
 235         bool ns_is_dad = IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src);
 236
 237         // Don't forward any non-DAD solicitation for external ifaces
 238         // TODO: check if we should even forward DADs for them
 239         if (iface->external && !ns_is_dad)
 240                 return;
 241
 242         if (len < sizeof(*ip6) + sizeof(*req))
 243                 return; // Invalid reqicitation
 244
 245         if (IN6_IS_ADDR_LINKLOCAL(&req->nd_ns_target) ||
 246                         IN6_IS_ADDR_LOOPBACK(&req->nd_ns_target) ||
 247                         IN6_IS_ADDR_MULTICAST(&req->nd_ns_target))
 248                 return; // Invalid target
 249
 250         char ipbuf[INET6_ADDRSTRLEN];
 251         inet_ntop(AF_INET6, &req->nd_ns_target, ipbuf, sizeof(ipbuf));
 252         syslog(LOG_DEBUG, "Got a NS for %s", ipbuf);
 253
 254         uint8_t mac[6];
 255         odhcpd_get_mac(iface, mac);
 256         if (!memcmp(ll->sll_addr, mac, sizeof(mac)))
 257                 return; // Looped back
 258
 259         struct interface *c;
 260         list_for_each_entry(c, &interfaces, head)
 261                 if (iface->ndp == RELAYD_RELAY && iface != c &&
 262                                 (ns_is_dad || !c->external))
 263                         ping6(&req->nd_ns_target, c);
 264 }
 265
 266
 267 void odhcpd_setup_route(const struct in6_addr *addr, int prefixlen,
 268                 const struct interface *iface, const struct in6_addr *gw, bool add)
 269 {
 270         struct req {
 271                 struct nlmsghdr nh;
 272                 struct rtmsg rtm;
 273                 struct rtattr rta_dst;
 274                 struct in6_addr dst_addr;
 275                 struct rtattr rta_oif;
 276                 uint32_t ifindex;
 277                 struct rtattr rta_table;
 278                 uint32_t table;
 279                 struct rtattr rta_gw;
 280                 struct in6_addr gw;
 281         } req = {
 282                 {sizeof(req), 0, NLM_F_REQUEST, ++rtnl_seqid, 0},
 283                 {AF_INET6, prefixlen, 0, 0, 0, 0, 0, 0, 0},
 284                 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_DST},
 285                 *addr,
 286                 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_OIF},
 287                 iface->ifindex,
 288                 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_TABLE},
 289                 RT_TABLE_MAIN,
 290                 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_GATEWAY},
 291                 IN6ADDR_ANY_INIT,
 292         };
 293
 294         if (gw)
 295                 req.gw = *gw;
 296
 297         if (add) {
 298                 req.nh.nlmsg_type = RTM_NEWROUTE;
 299                 req.nh.nlmsg_flags |= (NLM_F_CREATE | NLM_F_REPLACE);
 300                 req.rtm.rtm_protocol = RTPROT_STATIC;
 301                 req.rtm.rtm_scope = (gw) ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK;
 302                 req.rtm.rtm_type = RTN_UNICAST;
 303         } else {
 304                 req.nh.nlmsg_type = RTM_DELROUTE;
 305                 req.rtm.rtm_scope = RT_SCOPE_NOWHERE;
 306         }
 307
 308         req.nh.nlmsg_len = (gw) ? sizeof(req) : offsetof(struct req, rta_gw);
 309         send(rtnl_event.uloop.fd, &req, req.nh.nlmsg_len, MSG_DONTWAIT);
 310 }
 311
 312 // Use rtnetlink to modify kernel routes
 313 static void setup_route(struct in6_addr *addr, struct interface *iface, bool add)
 314 {
 315         char namebuf[INET6_ADDRSTRLEN];
 316         inet_ntop(AF_INET6, addr, namebuf, sizeof(namebuf));
 317         syslog(LOG_NOTICE, "%s about %s on %s",
 318                         (add) ? "Learned" : "Forgot", namebuf, iface->ifname);
 319
 320         if (iface->learn_routes)
 321                 odhcpd_setup_route(addr, 128, iface, NULL, add);
 322 }
 323
 324
 325 // Handler for neighbor cache entries from the kernel. This is our source
 326 // to learn and unlearn hosts on interfaces.
 327 static void handle_rtnetlink(_unused void *addr, void *data, size_t len,
 328                 _unused struct interface *iface, _unused void *dest)
 329 {
 330         bool dump_neigh = false;
 331         struct in6_addr last_solicited = IN6ADDR_ANY_INIT;
 332
 333         for (struct nlmsghdr *nh = data; NLMSG_OK(nh, len);
 334                         nh = NLMSG_NEXT(nh, len)) {
 335                 struct ndmsg *ndm = NLMSG_DATA(nh);
 336                 struct rtmsg *rtm = NLMSG_DATA(nh);
 337
 338                 bool is_addr = (nh->nlmsg_type == RTM_NEWADDR
 339                                 || nh->nlmsg_type == RTM_DELADDR);
 340                 bool is_route = (nh->nlmsg_type == RTM_NEWROUTE
 341                                 || nh->nlmsg_type == RTM_DELROUTE);
 342                 bool is_neigh = (nh->nlmsg_type == RTM_NEWNEIGH
 343                                 || nh->nlmsg_type == RTM_DELNEIGH);
 344
 345                 // Family and ifindex are on the same offset for NEIGH and ADDR
 346                 if ((!is_addr && !is_route && !is_neigh)
 347                                 || NLMSG_PAYLOAD(nh, 0) < sizeof(*ndm)
 348                                 || ndm->ndm_family != AF_INET6)
 349                         continue;
 350
 351                 // Inform about a change in default route
 352                 if (is_route && rtm->rtm_dst_len == 0)
 353                         raise(SIGUSR1);
 354                 else if (is_route && rtm->rtm_dst_len == 128)
 355                         continue;
 356
 357                 // Data to retrieve
 358                 size_t rta_offset = (is_route) ? sizeof(*rtm) : (is_addr) ?
 359                                 sizeof(struct ifaddrmsg) : sizeof(*ndm);
 360                 uint16_t atype = (is_route) ? RTA_DST : (is_addr) ? IFA_ADDRESS : NDA_DST;
 361                 ssize_t alen = NLMSG_PAYLOAD(nh, rta_offset);
 362                 struct in6_addr *addr = NULL;
 363                 int *ifindex = (!is_route) ? &ndm->ndm_ifindex : NULL;
 364                 int *metric = NULL;
 365
 366                 for (struct rtattr *rta = (void*)(((uint8_t*)ndm) + rta_offset);
 367                                 RTA_OK(rta, alen); rta = RTA_NEXT(rta, alen)) {
 368                         if (rta->rta_type == atype &&
 369                                         RTA_PAYLOAD(rta) >= sizeof(*addr)) {
 370                                 addr = RTA_DATA(rta);
 371                         } else if (is_route && rta->rta_type == RTA_OIF &&
 372                                         RTA_PAYLOAD(rta) == sizeof(int)) {
 373                                 ifindex = (int*)RTA_DATA(rta);
 374                         } else if (is_route && rta->rta_type == RTA_GATEWAY) {
 375                                 ifindex = NULL;
 376                                 break;
 377                         } else if (is_route && rta->rta_type == RTA_PRIORITY) {
 378                                 metric = (int*)RTA_DATA(rta);
 379                         }
 380                 }
 381
 382                 // Lookup interface
 383                 struct interface *iface = ifindex ? odhcpd_get_interface_by_index(*ifindex) : NULL;
 384                 if (!iface)
 385                         continue;
 386
 387                 // Keep-alive neighbor entries for RA sending
 388                 if (nh->nlmsg_type == RTM_DELNEIGH && !(ndm->ndm_state & NUD_FAILED) &&
 389                                 addr && IN6_IS_ADDR_LINKLOCAL(addr) && iface->ra == RELAYD_SERVER)
 390                         ping6(addr, iface);
 391
 392                 // Address not specified or unrelated
 393                 if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
 394                                 IN6_IS_ADDR_MULTICAST(addr))
 395                         continue;
 396
 397                 // Check for states
 398                 bool add;
 399                 if (is_addr)
 400                         add = (nh->nlmsg_type == RTM_NEWADDR);
 401                 else
 402                         add = (nh->nlmsg_type == RTM_NEWNEIGH && (ndm->ndm_state &
 403                                 (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE
 404                                                 | NUD_PERMANENT | NUD_NOARP)));
 405
 406                 if (iface->ndp == RELAYD_RELAY && !is_route) {
 407                         // Replay change to all neighbor cache
 408                         struct {
 409                                 struct nlmsghdr nh;
 410                                 struct ndmsg ndm;
 411                                 struct nlattr nla_dst;
 412                                 struct in6_addr dst;
 413                         } req = {
 414                                 {sizeof(req), RTM_DELNEIGH, NLM_F_REQUEST,
 415                                                 ++rtnl_seqid, 0},
 416                                 {.ndm_family = AF_INET6, .ndm_flags = NTF_PROXY},
 417                                 {sizeof(struct nlattr) + sizeof(struct in6_addr), NDA_DST},
 418                                 *addr
 419                         };
 420
 421                         if (ndm->ndm_flags & NTF_PROXY) {
 422                                 // Dump & flush proxy entries
 423                                 if (nh->nlmsg_type == RTM_NEWNEIGH) {
 424                                         req.ndm.ndm_ifindex = iface->ifindex;
 425                                         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 426                                         setup_route(addr, iface, false);
 427                                         dump_neigh = true;
 428                                 }
 429                         } else if (add) {
 430                                 struct interface *c;
 431                                 list_for_each_entry(c, &interfaces, head) {
 432                                         if (iface == c)
 433                                                 continue;
 434
 435                                         if (c->ndp == RELAYD_RELAY) {
 436                                                 req.nh.nlmsg_type = RTM_NEWNEIGH;
 437                                                 req.nh.nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
 438
 439                                                 req.ndm.ndm_ifindex = c->ifindex;
 440                                                 send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 441                                         } else { // Delete NDP cache from interfaces without relay
 442                                                 req.nh.nlmsg_type = RTM_DELNEIGH;
 443                                                 req.nh.nlmsg_flags &= ~(NLM_F_CREATE | NLM_F_REPLACE);
 444
 445                                                 req.ndm.ndm_ifindex = c->ifindex;
 446                                                 send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 447                                         }
 448                                 }
 449
 450                                 setup_route(addr, iface, true);
 451                         } else {
 452                                 if (nh->nlmsg_type == RTM_NEWNEIGH) {
 453                                         // might be locally originating
 454                                         if (!IN6_ARE_ADDR_EQUAL(&last_solicited, addr)) {
 455                                                 last_solicited = *addr;
 456
 457                                                 struct interface *c;
 458                                                 list_for_each_entry(c, &interfaces, head)
 459                                                         if (iface->ndp == RELAYD_RELAY && iface != c &&
 460                                                                         !c->external == false)
 461                                                                 ping6(addr, c);
 462                                         }
 463                                 } else {
 464                                         struct interface *c;
 465                                         list_for_each_entry(c, &interfaces, head) {
 466                                                 if (c->ndp == RELAYD_RELAY && iface != c) {
 467                                                         req.ndm.ndm_ifindex = c->ifindex;
 468                                                         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 469                                                 }
 470                                         }
 471                                         setup_route(addr, iface, false);
 472
 473                                         // also: dump to add proxies back in case it moved elsewhere
 474                                         dump_neigh = true;
 475                                 }
 476                         }
 477                 }
 478
 479                 if (is_addr) {
 480                         if (iface->ra == RELAYD_SERVER)
 481                                 raise(SIGUSR1); // Inform about a change in addresses
 482
 483                         if (iface->dhcpv6 == RELAYD_SERVER)
 484                                 iface->ia_reconf = true;
 485                 } else if (is_route) {
 486                         if (iface->ndp == RELAYD_RELAY && iface->master) {
 487                                 // Replay on-link route changes on all slave interfaces
 488                                 nh->nlmsg_flags = NLM_F_REQUEST;
 489
 490                                 if (nh->nlmsg_type == RTM_NEWROUTE)
 491                                         nh->nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
 492
 493                                 struct interface *c;
 494                                 list_for_each_entry(c, &interfaces, head) {
 495                                         if (c->ndp == RELAYD_RELAY && !c->master) {
 496                                                 *ifindex = c->ifindex;
 497                                                 *metric = (*metric & 0xffff) | (c->ifindex << 16);
 498                                                 send(rtnl_event.uloop.fd, nh, nh->nlmsg_len, MSG_DONTWAIT);
 499                                         }
 500                                 }
 501                         }
 502                 }
 503         }
 504
 505         if (dump_neigh)
 506                 dump_neigh_table(false);
 507 }