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
  41 static uint32_t rtnl_seqid = 0;
  42 static int ping_socket = -1;
  43 static struct odhcpd_event rtnl_event = {{.fd = -1}, handle_rtnetlink};
  44
  45
  46 // Filter ICMPv6 messages of type neighbor soliciation
  47 static struct sock_filter bpf[] = {
  48         BPF_STMT(BPF_LD | BPF_B | BPF_ABS, offsetof(struct ip6_hdr, ip6_nxt)),
  49         BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, IPPROTO_ICMPV6, 0, 3),
  50         BPF_STMT(BPF_LD | BPF_B | BPF_ABS, sizeof(struct ip6_hdr) +
  51                         offsetof(struct icmp6_hdr, icmp6_type)),
  52         BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, ND_NEIGHBOR_SOLICIT, 0, 1),
  53         BPF_STMT(BPF_RET | BPF_K, 0xffffffff),
  54         BPF_STMT(BPF_RET | BPF_K, 0),
  55 };
  56 static const struct sock_fprog bpf_prog = {sizeof(bpf) / sizeof(*bpf), bpf};
  57
  58
  59 // Initialize NDP-proxy
  60 int init_ndp(void)
  61 {
  62         // Setup netlink socket
  63         if ((rtnl_event.uloop.fd = odhcpd_open_rtnl()) < 0)
  64                 return -1;
  65
  66         // Receive netlink neighbor and ip-address events
  67         uint32_t group = RTNLGRP_IPV6_IFADDR;
  68         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
  69                         NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
  70         group = RTNLGRP_IPV6_ROUTE;
  71         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
  72                         NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
  73
  74         odhcpd_register(&rtnl_event);
  75
  76         // Open ICMPv6 socket
  77         ping_socket = socket(AF_INET6, SOCK_RAW | SOCK_CLOEXEC, IPPROTO_ICMPV6);
  78         if (ping_socket < 0) {
  79                 syslog(LOG_ERR, "Unable to open raw socket: %s", strerror(errno));
  80                         return -1;
  81         }
  82
  83         int val = 2;
  84         setsockopt(ping_socket, IPPROTO_RAW, IPV6_CHECKSUM, &val, sizeof(val));
  85
  86         // This is required by RFC 4861
  87         val = 255;
  88         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &val, sizeof(val));
  89         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &val, sizeof(val));
  90
  91         // Filter all packages, we only want to send
  92         struct icmp6_filter filt;
  93         ICMP6_FILTER_SETBLOCKALL(&filt);
  94         setsockopt(ping_socket, IPPROTO_ICMPV6, ICMP6_FILTER, &filt, sizeof(filt));
  95
  96
  97         // Netlink socket, continued...
  98         group = RTNLGRP_NEIGH;
  99         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
 100
 101         return 0;
 102 }
 103
 104
 105 static void dump_neigh_table(bool proxy)
 106 {
 107         struct {
 108                 struct nlmsghdr nh;
 109                 struct ndmsg ndm;
 110         } req = {
 111                 {sizeof(req), RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP,
 112                                 ++rtnl_seqid, 0},
 113                 {.ndm_family = AF_INET6, .ndm_flags = (proxy) ? NTF_PROXY : 0}
 114         };
 115         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 116         odhcpd_process(&rtnl_event);
 117 }
 118
 119
 120 int setup_ndp_interface(struct interface *iface, bool enable)
 121 {
 122         char procbuf[64];
 123         snprintf(procbuf, sizeof(procbuf), "/proc/sys/net/ipv6/conf/%s/proxy_ndp", iface->ifname);
 124         int procfd = open(procbuf, O_WRONLY);
 125         bool dump_neigh = false;
 126
 127         if (iface->ndp_event.uloop.fd > 0) {
 128                 uloop_fd_delete(&iface->ndp_event.uloop);
 129                 close(iface->ndp_event.uloop.fd);
 130                 iface->ndp_event.uloop.fd = -1;
 131
 132                 if (!enable || iface->ndp != RELAYD_RELAY)
 133                         if (write(procfd, "0\n", 2) < 0) {}
 134
 135                 dump_neigh = true;
 136         }
 137
 138         if (enable && (iface->ra == RELAYD_SERVER ||
 139                         iface->dhcpv6 == RELAYD_SERVER || iface->ndp == RELAYD_RELAY)) {
 140                 // Synthesize initial address events
 141                 struct {
 142                         struct nlmsghdr nh;
 143                         struct ifaddrmsg ifa;
 144                 } req2 = {
 145                         {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
 146                                         ++rtnl_seqid, 0},
 147                         {.ifa_family = AF_INET6, .ifa_index = iface->ifindex}
 148                 };
 149                 send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
 150         }
 151
 152         if (enable && iface->ndp == RELAYD_RELAY) {
 153                 if (write(procfd, "1\n", 2) < 0) {}
 154                 close(procfd);
 155
 156                 int sock = socket(AF_PACKET, SOCK_DGRAM | SOCK_CLOEXEC, htons(ETH_P_IPV6));
 157                 if (sock < 0) {
 158                         syslog(LOG_ERR, "Unable to open packet socket: %s",
 159                                         strerror(errno));
 160                         return -1;
 161                 }
 162
 163 #ifdef PACKET_RECV_TYPE
 164                 int pktt = 1 << PACKET_MULTICAST;
 165                 setsockopt(sock, SOL_PACKET, PACKET_RECV_TYPE, &pktt, sizeof(pktt));
 166 #endif
 167
 168                 if (setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER,
 169                                 &bpf_prog, sizeof(bpf_prog))) {
 170                         syslog(LOG_ERR, "Failed to set BPF: %s", strerror(errno));
 171                         return -1;
 172                 }
 173
 174                 struct sockaddr_ll ll = {
 175                         .sll_family = AF_PACKET,
 176                         .sll_ifindex = iface->ifindex,
 177                         .sll_protocol = htons(ETH_P_IPV6),
 178                         .sll_hatype = 0,
 179                         .sll_pkttype = 0,
 180                         .sll_halen = 0,
 181                         .sll_addr = {0},
 182                 };
 183                 bind(sock, (struct sockaddr*)&ll, sizeof(ll));
 184
 185                 struct packet_mreq mreq = {iface->ifindex, PACKET_MR_ALLMULTI, ETH_ALEN, {0}};
 186                 setsockopt(sock, SOL_PACKET, PACKET_ADD_MEMBERSHIP, &mreq, sizeof(mreq));
 187
 188                 iface->ndp_event.uloop.fd = sock;
 189                 iface->ndp_event.handle_dgram = handle_solicit;
 190                 odhcpd_register(&iface->ndp_event);
 191
 192                 // If we already were enabled dump is unnecessary, if not do dump
 193                 if (!dump_neigh)
 194                         dump_neigh_table(false);
 195                 else
 196                         dump_neigh = false;
 197         } else {
 198                 close(procfd);
 199         }
 200
 201         if (dump_neigh)
 202                 dump_neigh_table(true);
 203
 204         return 0;
 205 }
 206
 207
 208 // Send an ICMP-ECHO. This is less for actually pinging but for the
 209 // neighbor cache to be kept up-to-date.
 210 static void ping6(struct in6_addr *addr,
 211                 const struct interface *iface)
 212 {
 213         struct sockaddr_in6 dest = {AF_INET6, 0, 0, *addr, iface->ifindex};
 214         struct icmp6_hdr echo = {.icmp6_type = ICMP6_ECHO_REQUEST};
 215         struct iovec iov = {&echo, sizeof(echo)};
 216
 217         odhcpd_setup_route(addr, 128, iface, NULL, 128, true);
 218         odhcpd_send(ping_socket, &dest, &iov, 1, iface);
 219         odhcpd_setup_route(addr, 128, iface, NULL, 128, false);
 220 }
 221
 222
 223 // Handle solicitations
 224 static void handle_solicit(void *addr, void *data, size_t len,
 225                 struct interface *iface, _unused void *dest)
 226 {
 227         struct ip6_hdr *ip6 = data;
 228         struct nd_neighbor_solicit *req = (struct nd_neighbor_solicit*)&ip6[1];
 229         struct sockaddr_ll *ll = addr;
 230
 231         // Solicitation is for duplicate address detection
 232         bool ns_is_dad = IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src);
 233
 234         // Don't forward any non-DAD solicitation for external ifaces
 235         // TODO: check if we should even forward DADs for them
 236         if (iface->external && !ns_is_dad)
 237                 return;
 238
 239         if (len < sizeof(*ip6) + sizeof(*req))
 240                 return; // Invalid reqicitation
 241
 242         if (IN6_IS_ADDR_LINKLOCAL(&req->nd_ns_target) ||
 243                         IN6_IS_ADDR_LOOPBACK(&req->nd_ns_target) ||
 244                         IN6_IS_ADDR_MULTICAST(&req->nd_ns_target))
 245                 return; // Invalid target
 246
 247         char ipbuf[INET6_ADDRSTRLEN];
 248         inet_ntop(AF_INET6, &req->nd_ns_target, ipbuf, sizeof(ipbuf));
 249         syslog(LOG_DEBUG, "Got a NS for %s", ipbuf);
 250
 251         uint8_t mac[6];
 252         odhcpd_get_mac(iface, mac);
 253         if (!memcmp(ll->sll_addr, mac, sizeof(mac)))
 254                 return; // Looped back
 255
 256         struct interface *c;
 257         list_for_each_entry(c, &interfaces, head)
 258                 if (iface->ndp == RELAYD_RELAY && iface != c &&
 259                                 (ns_is_dad || !c->external))
 260                         ping6(&req->nd_ns_target, c);
 261 }
 262
 263 // Use rtnetlink to modify kernel routes
 264 static void setup_route(struct in6_addr *addr, struct interface *iface, bool add)
 265 {
 266         char namebuf[INET6_ADDRSTRLEN];
 267         inet_ntop(AF_INET6, addr, namebuf, sizeof(namebuf));
 268         syslog(LOG_NOTICE, "%s about %s on %s",
 269                         (add) ? "Learned" : "Forgot", namebuf, iface->ifname);
 270
 271         if (iface->learn_routes)
 272                 odhcpd_setup_route(addr, 128, iface, NULL, 1024, add);
 273 }
 274
 275
 276 // Handler for neighbor cache entries from the kernel. This is our source
 277 // to learn and unlearn hosts on interfaces.
 278 static void handle_rtnetlink(_unused void *addr, void *data, size_t len,
 279                 _unused struct interface *iface, _unused void *dest)
 280 {
 281         bool dump_neigh = false;
 282         struct in6_addr last_solicited = IN6ADDR_ANY_INIT;
 283
 284         for (struct nlmsghdr *nh = data; NLMSG_OK(nh, len);
 285                         nh = NLMSG_NEXT(nh, len)) {
 286                 struct ndmsg *ndm = NLMSG_DATA(nh);
 287                 struct rtmsg *rtm = NLMSG_DATA(nh);
 288
 289                 bool is_addr = (nh->nlmsg_type == RTM_NEWADDR
 290                                 || nh->nlmsg_type == RTM_DELADDR);
 291                 bool is_route = (nh->nlmsg_type == RTM_NEWROUTE
 292                                 || nh->nlmsg_type == RTM_DELROUTE);
 293                 bool is_neigh = (nh->nlmsg_type == RTM_NEWNEIGH
 294                                 || nh->nlmsg_type == RTM_DELNEIGH);
 295
 296                 // Family and ifindex are on the same offset for NEIGH and ADDR
 297                 if ((!is_addr && !is_route && !is_neigh)
 298                                 || NLMSG_PAYLOAD(nh, 0) < sizeof(*ndm)
 299                                 || ndm->ndm_family != AF_INET6)
 300                         continue;
 301
 302                 // Inform about a change in default route
 303                 if (is_route && rtm->rtm_dst_len == 0)
 304                         raise(SIGUSR1);
 305                 else if (is_route)
 306                         continue;
 307
 308                 // Data to retrieve
 309                 size_t rta_offset = (is_addr) ? sizeof(struct ifaddrmsg) : sizeof(*ndm);
 310                 uint16_t atype = (is_addr) ? IFA_ADDRESS : NDA_DST;
 311                 ssize_t alen = NLMSG_PAYLOAD(nh, rta_offset);
 312                 struct in6_addr *addr = NULL;
 313
 314                 for (struct rtattr *rta = (void*)(((uint8_t*)ndm) + rta_offset);
 315                                 RTA_OK(rta, alen); rta = RTA_NEXT(rta, alen)) {
 316                         if (rta->rta_type == atype &&
 317                                         RTA_PAYLOAD(rta) >= sizeof(*addr)) {
 318                                 addr = RTA_DATA(rta);
 319                         }
 320                 }
 321
 322                 // Lookup interface
 323                 struct interface *iface = odhcpd_get_interface_by_index(ndm->ndm_ifindex);
 324                 if (!iface)
 325                         continue;
 326
 327                 // Keep-alive neighbor entries for RA sending
 328                 if (nh->nlmsg_type == RTM_DELNEIGH && !(ndm->ndm_state & NUD_FAILED) &&
 329                                 addr && IN6_IS_ADDR_LINKLOCAL(addr) && iface->ra == RELAYD_SERVER)
 330                         ping6(addr, iface);
 331
 332                 // Address not specified or unrelated
 333                 if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
 334                                 IN6_IS_ADDR_MULTICAST(addr))
 335                         continue;
 336
 337                 // Check for states
 338                 bool add;
 339                 if (is_addr)
 340                         add = (nh->nlmsg_type == RTM_NEWADDR);
 341                 else
 342                         add = (nh->nlmsg_type == RTM_NEWNEIGH && (ndm->ndm_state &
 343                                 (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE
 344                                                 | NUD_PERMANENT | NUD_NOARP)));
 345
 346                 if (iface->ndp == RELAYD_RELAY) {
 347                         // Replay change to all neighbor cache
 348                         struct {
 349                                 struct nlmsghdr nh;
 350                                 struct ndmsg ndm;
 351                                 struct nlattr nla_dst;
 352                                 struct in6_addr dst;
 353                         } req = {
 354                                 {sizeof(req), RTM_DELNEIGH, NLM_F_REQUEST,
 355                                                 ++rtnl_seqid, 0},
 356                                 {.ndm_family = AF_INET6, .ndm_flags = NTF_PROXY},
 357                                 {sizeof(struct nlattr) + sizeof(struct in6_addr), NDA_DST},
 358                                 *addr
 359                         };
 360
 361                         if (ndm->ndm_flags & NTF_PROXY) {
 362                                 // Dump & flush proxy entries
 363                                 if (nh->nlmsg_type == RTM_NEWNEIGH) {
 364                                         req.ndm.ndm_ifindex = iface->ifindex;
 365                                         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 366                                         setup_route(addr, iface, false);
 367                                         dump_neigh = true;
 368                                 }
 369                         } else if (add) {
 370                                 struct interface *c;
 371                                 list_for_each_entry(c, &interfaces, head) {
 372                                         if (iface == c)
 373                                                 continue;
 374
 375                                         if (c->ndp == RELAYD_RELAY) {
 376                                                 req.nh.nlmsg_type = RTM_NEWNEIGH;
 377                                                 req.nh.nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
 378
 379                                                 req.ndm.ndm_ifindex = c->ifindex;
 380                                                 send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 381                                         } else { // Delete NDP cache from interfaces without relay
 382                                                 req.nh.nlmsg_type = RTM_DELNEIGH;
 383                                                 req.nh.nlmsg_flags &= ~(NLM_F_CREATE | NLM_F_REPLACE);
 384
 385                                                 req.ndm.ndm_ifindex = c->ifindex;
 386                                                 send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 387                                         }
 388                                 }
 389
 390                                 setup_route(addr, iface, true);
 391                         } else {
 392                                 if (nh->nlmsg_type == RTM_NEWNEIGH) {
 393                                         // might be locally originating
 394                                         if (!IN6_ARE_ADDR_EQUAL(&last_solicited, addr)) {
 395                                                 last_solicited = *addr;
 396
 397                                                 struct interface *c;
 398                                                 list_for_each_entry(c, &interfaces, head)
 399                                                         if (iface->ndp == RELAYD_RELAY && iface != c &&
 400                                                                         !c->external == false)
 401                                                                 ping6(addr, c);
 402                                         }
 403                                 } else {
 404                                         struct interface *c;
 405                                         list_for_each_entry(c, &interfaces, head) {
 406                                                 if (c->ndp == RELAYD_RELAY && iface != c) {
 407                                                         req.ndm.ndm_ifindex = c->ifindex;
 408                                                         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
 409                                                 }
 410                                         }
 411                                         setup_route(addr, iface, false);
 412
 413                                         // also: dump to add proxies back in case it moved elsewhere
 414                                         dump_neigh = true;
 415                                 }
 416                         }
 417                 }
 418
 419                 if (is_addr) {
 420                         if (iface->ra == RELAYD_SERVER)
 421                                 raise(SIGUSR1); // Inform about a change in addresses
 422
 423                         if (iface->dhcpv6 == RELAYD_SERVER)
 424                                 iface->ia_reconf = true;
 425                 }
 426         }
 427
 428         if (dump_neigh)
 429                 dump_neigh_table(false);
 430 }