2 * Copyright (C) 2012-2013 Steven Barth <steven@midlink.org>
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.
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.
21 #include <arpa/inet.h>
22 #include <sys/socket.h>
23 #include <net/ethernet.h>
24 #include <netinet/ip6.h>
25 #include <netinet/icmp6.h>
26 #include <netpacket/packet.h>
28 #include <linux/rtnetlink.h>
29 #include <linux/filter.h>
35 static void handle_solicit(void *addr, void *data, size_t len,
36 struct interface *iface, void *dest);
37 static void handle_rtnetlink(void *addr, void *data, size_t len,
38 struct interface *iface, void *dest);
39 static struct ndp_neighbor* find_neighbor(struct in6_addr *addr, bool strict);
40 static void modify_neighbor(struct in6_addr *addr, struct interface *iface,
42 static ssize_t ping6(struct in6_addr *addr,
43 const struct interface *iface);
45 static struct list_head neighbors = LIST_HEAD_INIT(neighbors);
46 static size_t neighbor_count = 0;
47 static uint32_t rtnl_seqid = 0;
49 static int ping_socket = -1;
50 static struct odhcpd_event ndp_event = {{.fd = -1}, handle_solicit};
51 static struct odhcpd_event rtnl_event = {{.fd = -1}, handle_rtnetlink};
54 // Filter ICMPv6 messages of type neighbor soliciation
55 static struct sock_filter bpf[] = {
56 BPF_STMT(BPF_LD | BPF_B | BPF_ABS, offsetof(struct ip6_hdr, ip6_nxt)),
57 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, IPPROTO_ICMPV6, 0, 3),
58 BPF_STMT(BPF_LD | BPF_B | BPF_ABS, sizeof(struct ip6_hdr) +
59 offsetof(struct icmp6_hdr, icmp6_type)),
60 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, ND_NEIGHBOR_SOLICIT, 0, 1),
61 BPF_STMT(BPF_RET | BPF_K, 0xffffffff),
62 BPF_STMT(BPF_RET | BPF_K, 0),
64 static const struct sock_fprog bpf_prog = {sizeof(bpf) / sizeof(*bpf), bpf};
67 // Initialize NDP-proxy
70 // Setup netlink socket
71 if ((rtnl_event.uloop.fd = odhcpd_open_rtnl()) < 0)
74 // Receive netlink neighbor and ip-address events
75 uint32_t group = RTNLGRP_IPV6_IFADDR;
76 setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
77 NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
78 group = RTNLGRP_IPV6_ROUTE;
79 setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
80 NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
82 // Synthesize initial address events
87 {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
89 {.ifa_family = AF_INET6}
91 send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
92 odhcpd_register(&rtnl_event);
95 ping_socket = socket(AF_INET6, SOCK_RAW | SOCK_CLOEXEC, IPPROTO_ICMPV6);
96 if (ping_socket < 0) {
97 syslog(LOG_ERR, "Unable to open raw socket: %s", strerror(errno));
102 setsockopt(ping_socket, IPPROTO_RAW, IPV6_CHECKSUM, &val, sizeof(val));
104 // This is required by RFC 4861
106 setsockopt(ping_socket, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &val, sizeof(val));
107 setsockopt(ping_socket, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &val, sizeof(val));
109 // Filter all packages, we only want to send
110 struct icmp6_filter filt;
111 ICMP6_FILTER_SETBLOCKALL(&filt);
112 setsockopt(ping_socket, IPPROTO_ICMPV6, ICMP6_FILTER, &filt, sizeof(filt));
115 // Netlink socket, continued...
116 group = RTNLGRP_NEIGH;
117 setsockopt(rtnl_event.uloop.fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
119 // Synthesize initial neighbor events
124 {sizeof(req), RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP,
126 {.ndm_family = AF_INET6}
128 send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
134 int setup_ndp_interface(struct interface *iface, bool enable)
136 struct packet_mreq mreq = {iface->ifindex, PACKET_MR_ALLMULTI, ETH_ALEN, {0}};
137 setsockopt(ndp_event.uloop.fd, SOL_PACKET, PACKET_DROP_MEMBERSHIP, &mreq, sizeof(mreq));
139 struct ndp_neighbor *c, *n;
140 list_for_each_entry_safe(c, n, &neighbors, head)
141 if (c->iface == iface && (c->timeout == 0 || iface->ndp != RELAYD_RELAY || !enable))
142 modify_neighbor(&c->addr, c->iface, false);
144 if (enable && iface->ndp == RELAYD_RELAY) {
145 setsockopt(ndp_event.uloop.fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP, &mreq, sizeof(mreq));
147 if (iface->static_ndp_len) {
148 char *entry = alloca(iface->static_ndp_len), *saveptr;
150 syslog(LOG_ERR, "Alloca failed for static NDP list");
153 memcpy(entry, iface->static_ndp, iface->static_ndp_len);
155 for (entry = strtok_r(entry, " ", &saveptr); entry; entry = strtok_r(NULL, " ", &saveptr)) {
157 struct ndp_neighbor *n = malloc(sizeof(*n));
159 syslog(LOG_ERR, "Malloc failed for static NDP-prefix %s", entry);
166 sep = strchr(entry, '/');
169 syslog(LOG_ERR, "Invalid static NDP-prefix %s", entry);
174 n->len = atoi(sep + 1);
175 if (inet_pton(AF_INET6, entry, &n->addr) != 1 || n->len > 128) {
177 syslog(LOG_ERR, "Invalid static NDP-prefix %s/%s", entry, sep + 1);
181 list_add(&n->head, &neighbors);
186 bool enable_packet = false;
188 list_for_each_entry(i, &interfaces, head) {
189 if (i == iface && !enable)
192 if (i->ndp == RELAYD_RELAY)
193 enable_packet = true;
196 if (enable_packet && ndp_event.uloop.fd < 0) {
197 // Create socket for intercepting NDP
198 int sock = socket(AF_PACKET, SOCK_DGRAM | SOCK_CLOEXEC | SOCK_NONBLOCK,
199 htons(ETH_P_ALL)); // ETH_P_ALL for ingress + egress
201 syslog(LOG_ERR, "Unable to open packet socket: %s",
206 if (setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER,
207 &bpf_prog, sizeof(bpf_prog))) {
208 syslog(LOG_ERR, "Failed to set BPF: %s", strerror(errno));
212 ndp_event.uloop.fd = sock;
213 odhcpd_register(&ndp_event);
214 } else if (!enable_packet && ndp_event.uloop.fd >= 0) {
215 close(ndp_event.uloop.fd);
216 ndp_event.uloop.fd = -1;
223 // Send an ICMP-ECHO. This is less for actually pinging but for the
224 // neighbor cache to be kept up-to-date.
225 static ssize_t ping6(struct in6_addr *addr,
226 const struct interface *iface)
228 struct sockaddr_in6 dest = {AF_INET6, 0, 0, *addr, 0};
229 struct icmp6_hdr echo = {.icmp6_type = ICMP6_ECHO_REQUEST};
230 struct iovec iov = {&echo, sizeof(echo)};
232 // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
233 setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
234 iface->ifname, sizeof(iface->ifname));
235 return odhcpd_send(ping_socket, &dest, &iov, 1, iface);
239 // Handle solicitations
240 static void handle_solicit(void *addr, void *data, size_t len,
241 struct interface *iface, _unused void *dest)
243 struct ip6_hdr *ip6 = data;
244 struct nd_neighbor_solicit *req = (struct nd_neighbor_solicit*)&ip6[1];
245 struct sockaddr_ll *ll = addr;
247 // Solicitation is for duplicate address detection
248 bool ns_is_dad = IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src);
250 // Don't forward any non-DAD solicitation for external ifaces
251 // TODO: check if we should even forward DADs for them
252 if (iface->external && !ns_is_dad)
255 if (len < sizeof(*ip6) + sizeof(*req))
256 return; // Invalid reqicitation
258 if (IN6_IS_ADDR_LINKLOCAL(&req->nd_ns_target) ||
259 IN6_IS_ADDR_LOOPBACK(&req->nd_ns_target) ||
260 IN6_IS_ADDR_MULTICAST(&req->nd_ns_target))
261 return; // Invalid target
263 char ipbuf[INET6_ADDRSTRLEN];
264 inet_ntop(AF_INET6, &req->nd_ns_target, ipbuf, sizeof(ipbuf));
265 syslog(LOG_DEBUG, "Got a NS for %s", ipbuf);
268 odhcpd_get_mac(iface, mac);
269 if (!memcmp(ll->sll_addr, mac, sizeof(mac)) &&
270 ll->sll_pkttype != PACKET_OUTGOING)
271 return; // Looped back
273 time_t now = time(NULL);
275 struct ndp_neighbor *n = find_neighbor(&req->nd_ns_target, false);
276 if (n && (n->iface || abs(n->timeout - now) < 5)) {
277 syslog(LOG_DEBUG, "%s is on %s", ipbuf,
278 (n->iface) ? n->iface->ifname : "<pending>");
279 if (!n->iface || n->iface == iface)
282 // Found on other interface, answer with advertisement
284 struct nd_neighbor_advert body;
285 struct nd_opt_hdr opt_ll_hdr;
289 .nd_na_hdr = {ND_NEIGHBOR_ADVERT,
291 .nd_na_target = req->nd_ns_target,
293 .opt_ll_hdr = {ND_OPT_TARGET_LINKADDR, 1},
296 memcpy(advert.mac, mac, sizeof(advert.mac));
297 advert.body.nd_na_flags_reserved = ND_NA_FLAG_ROUTER |
298 ND_NA_FLAG_SOLICITED;
300 struct sockaddr_in6 dest = {AF_INET6, 0, 0, ALL_IPV6_NODES, 0};
301 if (!ns_is_dad) // If not DAD, then unicast to source
302 dest.sin6_addr = ip6->ip6_src;
304 // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
305 setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
306 iface->ifname, sizeof(iface->ifname));
307 struct iovec iov = {&advert, sizeof(advert)};
308 odhcpd_send(ping_socket, &dest, &iov, 1, iface);
310 // Send echo to all other interfaces to see where target is on
311 // This will trigger neighbor discovery which is what we want.
312 // We will observe the neighbor cache to see results.
316 list_for_each_entry(c, &interfaces, head)
317 if (iface->ndp == RELAYD_RELAY && iface != c &&
318 (!ns_is_dad || !c->external == false))
319 sent += ping6(&req->nd_ns_target, c);
321 if (sent > 0) // Sent a ping, add pending neighbor entry
322 modify_neighbor(&req->nd_ns_target, NULL, true);
327 void odhcpd_setup_route(const struct in6_addr *addr, int prefixlen,
328 const struct interface *iface, const struct in6_addr *gw, bool add)
333 struct rtattr rta_dst;
334 struct in6_addr dst_addr;
335 struct rtattr rta_oif;
337 struct rtattr rta_table;
339 struct rtattr rta_gw;
342 {sizeof(req), 0, NLM_F_REQUEST, ++rtnl_seqid, 0},
343 {AF_INET6, prefixlen, 0, 0, 0, 0, 0, 0, 0},
344 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_DST},
346 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_OIF},
348 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_TABLE},
350 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_GATEWAY},
358 req.nh.nlmsg_type = RTM_NEWROUTE;
359 req.nh.nlmsg_flags |= (NLM_F_CREATE | NLM_F_REPLACE);
360 req.rtm.rtm_protocol = RTPROT_BOOT;
361 req.rtm.rtm_scope = (gw) ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK;
362 req.rtm.rtm_type = RTN_UNICAST;
364 req.nh.nlmsg_type = RTM_DELROUTE;
365 req.rtm.rtm_scope = RT_SCOPE_NOWHERE;
368 size_t reqlen = (gw) ? sizeof(req) : offsetof(struct req, rta_gw);
369 send(rtnl_event.uloop.fd, &req, reqlen, MSG_DONTWAIT);
372 // Use rtnetlink to modify kernel routes
373 static void setup_route(struct in6_addr *addr, struct interface *iface,
376 char namebuf[INET6_ADDRSTRLEN];
377 inet_ntop(AF_INET6, addr, namebuf, sizeof(namebuf));
378 syslog(LOG_NOTICE, "%s about %s on %s", (add) ? "Learned" : "Forgot",
379 namebuf, (iface) ? iface->ifname : "<pending>");
381 if (!iface || !iface->learn_routes)
384 odhcpd_setup_route(addr, 128, iface, NULL, add);
387 static void free_neighbor(struct ndp_neighbor *n)
389 setup_route(&n->addr, n->iface, false);
395 static struct ndp_neighbor* find_neighbor(struct in6_addr *addr, bool strict)
397 time_t now = time(NULL);
398 struct ndp_neighbor *n, *e;
399 list_for_each_entry_safe(n, e, &neighbors, head) {
400 if ((!strict && !odhcpd_bmemcmp(&n->addr, addr, n->len)) ||
401 (n->len == 128 && IN6_ARE_ADDR_EQUAL(&n->addr, addr)))
404 if (!n->iface && abs(n->timeout - now) >= 5)
411 // Modified our own neighbor-entries
412 static void modify_neighbor(struct in6_addr *addr,
413 struct interface *iface, bool add)
415 if (!addr || (void*)addr == (void*)iface)
418 struct ndp_neighbor *n = find_neighbor(addr, true);
419 if (!add) { // Delete action
420 if (n && (!n->iface || n->iface == iface))
422 } else if (!n) { // No entry yet, add one if possible
423 if (neighbor_count >= NDP_MAX_NEIGHBORS ||
424 !(n = malloc(sizeof(*n))))
432 list_add(&n->head, &neighbors);
434 setup_route(addr, n->iface, add);
435 } else if (n->iface == iface) {
438 } else if (iface && (!n->iface ||
439 (!iface->external && n->iface->external))) {
440 setup_route(addr, n->iface, false);
442 setup_route(addr, n->iface, add);
444 // TODO: In case a host switches interfaces we might want
445 // to set its old neighbor entry to NUD_STALE and ping it
446 // on the old interface to confirm if the MACs match.
450 // Handler for neighbor cache entries from the kernel. This is our source
451 // to learn and unlearn hosts on interfaces.
452 static void handle_rtnetlink(_unused void *addr, void *data, size_t len,
453 _unused struct interface *iface, _unused void *dest)
455 for (struct nlmsghdr *nh = data; NLMSG_OK(nh, len);
456 nh = NLMSG_NEXT(nh, len)) {
457 struct rtmsg *rtm = NLMSG_DATA(nh);
458 if ((nh->nlmsg_type == RTM_NEWROUTE ||
459 nh->nlmsg_type == RTM_DELROUTE) &&
460 rtm->rtm_dst_len == 0)
461 raise(SIGUSR1); // Inform about a change in default route
463 struct ndmsg *ndm = NLMSG_DATA(nh);
464 struct ifaddrmsg *ifa = NLMSG_DATA(nh);
465 if (nh->nlmsg_type != RTM_NEWNEIGH
466 && nh->nlmsg_type != RTM_DELNEIGH
467 && nh->nlmsg_type != RTM_NEWADDR
468 && nh->nlmsg_type != RTM_DELADDR)
469 continue; // Unrelated message type
470 bool is_addr = (nh->nlmsg_type == RTM_NEWADDR
471 || nh->nlmsg_type == RTM_DELADDR);
473 // Family and ifindex are on the same offset for NEIGH and ADDR
474 if (NLMSG_PAYLOAD(nh, 0) < sizeof(*ndm)
475 || ndm->ndm_family != AF_INET6)
479 struct interface *iface;
480 if (!(iface = odhcpd_get_interface_by_index(ndm->ndm_ifindex)))
484 size_t rta_offset = (is_addr) ? sizeof(*ifa) : sizeof(*ndm);
485 uint16_t atype = (is_addr) ? IFA_ADDRESS : NDA_DST;
486 ssize_t alen = NLMSG_PAYLOAD(nh, rta_offset);
487 struct in6_addr *addr = NULL;
489 for (struct rtattr *rta = (void*)(((uint8_t*)ndm) + rta_offset);
490 RTA_OK(rta, alen); rta = RTA_NEXT(rta, alen))
491 if (rta->rta_type == atype &&
492 RTA_PAYLOAD(rta) >= sizeof(*addr))
493 addr = RTA_DATA(rta);
495 // Address not specified or unrelated
496 if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
497 IN6_IS_ADDR_MULTICAST(addr))
503 add = (nh->nlmsg_type == RTM_NEWADDR);
505 add = (nh->nlmsg_type == RTM_NEWNEIGH && (ndm->ndm_state &
506 (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE
507 | NUD_PERMANENT | NUD_NOARP)));
509 if (iface->ndp == RELAYD_RELAY)
510 modify_neighbor(addr, iface, add);
512 if (is_addr && iface->ra == RELAYD_SERVER)
513 raise(SIGUSR1); // Inform about a change in addresses
515 if (is_addr && iface->dhcpv6 == RELAYD_SERVER)
516 iface->ia_reconf = true;
518 if (iface->ndp == RELAYD_RELAY && is_addr && iface->master) {
519 // Replay address changes on all slave interfaces
520 nh->nlmsg_flags = NLM_F_REQUEST;
522 if (nh->nlmsg_type == RTM_NEWADDR)
523 nh->nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
526 list_for_each_entry(c, &interfaces, head) {
527 if (c->ndp == RELAYD_RELAY && !c->master) {
528 ifa->ifa_index = c->ifindex;
529 send(rtnl_event.uloop.fd, nh, nh->nlmsg_len, MSG_DONTWAIT);
534 /* TODO: See if this is required for optimal operation
535 // Keep neighbor entries alive so we don't loose routes
537 if (add && (ndm->ndm_state & NUD_STALE))