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.
20 #include <arpa/inet.h>
21 #include <sys/socket.h>
22 #include <net/ethernet.h>
23 #include <netinet/ip6.h>
24 #include <netinet/icmp6.h>
25 #include <netpacket/packet.h>
27 #include <linux/rtnetlink.h>
28 #include <linux/filter.h>
34 static void handle_solicit(void *addr, void *data, size_t len,
35 struct interface *iface);
36 static void handle_rtnetlink(void *addr, void *data, size_t len,
37 struct interface *iface);
38 static struct ndp_neighbor* find_neighbor(struct in6_addr *addr, bool strict);
39 static void modify_neighbor(struct in6_addr *addr, struct interface *iface,
41 static ssize_t ping6(struct in6_addr *addr,
42 const struct interface *iface);
44 static struct list_head neighbors = LIST_HEAD_INIT(neighbors);
45 static size_t neighbor_count = 0;
46 static uint32_t rtnl_seqid = 0;
48 static int ping_socket = -1;
49 static struct odhcpd_event ndp_event = {{.fd = -1}, handle_solicit};
50 static struct odhcpd_event rtnl_event = {{.fd = -1}, handle_rtnetlink};
53 // Filter ICMPv6 messages of type neighbor soliciation
54 static struct sock_filter bpf[] = {
55 BPF_STMT(BPF_LD | BPF_B | BPF_ABS, offsetof(struct ip6_hdr, ip6_nxt)),
56 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, IPPROTO_ICMPV6, 0, 3),
57 BPF_STMT(BPF_LD | BPF_B | BPF_ABS, sizeof(struct ip6_hdr) +
58 offsetof(struct icmp6_hdr, icmp6_type)),
59 BPF_JUMP(BPF_JMP | BPF_JEQ | BPF_K, ND_NEIGHBOR_SOLICIT, 0, 1),
60 BPF_STMT(BPF_RET | BPF_K, 0xffffffff),
61 BPF_STMT(BPF_RET | BPF_K, 0),
63 static const struct sock_fprog bpf_prog = {sizeof(bpf) / sizeof(*bpf), bpf};
66 // Initialize NDP-proxy
69 // Setup netlink socket
70 if ((rtnl_event.uloop.fd = odhcpd_open_rtnl()) < 0)
73 // Receive netlink neighbor and ip-address events
74 uint32_t group = RTNLGRP_IPV6_IFADDR;
75 setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
76 NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
77 group = RTNLGRP_IPV6_ROUTE;
78 setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
79 NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
81 // Synthesize initial address events
86 {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
88 {.ifa_family = AF_INET6}
90 send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
91 odhcpd_register(&rtnl_event);
94 // Create socket for intercepting NDP
95 int sock = socket(AF_PACKET, SOCK_DGRAM | SOCK_CLOEXEC | SOCK_NONBLOCK,
96 htons(ETH_P_ALL)); // ETH_P_ALL for ingress + egress
98 syslog(LOG_ERR, "Unable to open packet socket: %s",
103 if (setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER,
104 &bpf_prog, sizeof(bpf_prog))) {
105 syslog(LOG_ERR, "Failed to set BPF: %s", strerror(errno));
109 ndp_event.uloop.fd = sock;
110 odhcpd_register(&ndp_event);
112 // Open ICMPv6 socket
113 ping_socket = socket(AF_INET6, SOCK_RAW | SOCK_CLOEXEC, IPPROTO_ICMPV6);
116 setsockopt(ping_socket, IPPROTO_RAW, IPV6_CHECKSUM, &val, sizeof(val));
118 // This is required by RFC 4861
120 setsockopt(ping_socket, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &val, sizeof(val));
121 setsockopt(ping_socket, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &val, sizeof(val));
123 // Filter all packages, we only want to send
124 struct icmp6_filter filt;
125 ICMP6_FILTER_SETBLOCKALL(&filt);
126 setsockopt(ping_socket, IPPROTO_ICMPV6, ICMP6_FILTER, &filt, sizeof(filt));
129 // Netlink socket, continued...
130 group = RTNLGRP_NEIGH;
131 setsockopt(rtnl_event.uloop.fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
133 // Synthesize initial neighbor events
138 {sizeof(req), RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP,
140 {.ndm_family = AF_INET6}
142 send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
148 int setup_ndp_interface(struct interface *iface, bool enable)
150 struct packet_mreq mreq = {iface->ifindex, PACKET_MR_ALLMULTI, ETH_ALEN, {0}};
151 setsockopt(ndp_event.uloop.fd, SOL_PACKET, PACKET_DROP_MEMBERSHIP, &mreq, sizeof(mreq));
153 struct ndp_neighbor *c, *n;
154 list_for_each_entry_safe(c, n, &neighbors, head)
155 if (c->iface == iface && (c->timeout == 0 || iface->ndp != RELAYD_RELAY || !enable))
156 modify_neighbor(&c->addr, c->iface, false);
158 if (enable && iface->ndp == RELAYD_RELAY) {
159 setsockopt(ndp_event.uloop.fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP, &mreq, sizeof(mreq));
161 if (iface->static_ndp_len) {
162 char *entry = alloca(iface->static_ndp_len), *saveptr;
164 syslog(LOG_ERR, "Alloca failed for static NDP list");
167 memcpy(entry, iface->static_ndp, iface->static_ndp_len);
169 for (entry = strtok_r(entry, " ", &saveptr); entry; entry = strtok_r(NULL, " ", &saveptr)) {
171 struct ndp_neighbor *n = malloc(sizeof(*n));
173 syslog(LOG_ERR, "Malloc failed for static NDP-prefix %s", entry);
180 sep = strchr(entry, '/');
183 syslog(LOG_ERR, "Invalid static NDP-prefix %s", entry);
188 n->len = atoi(sep + 1);
189 if (inet_pton(AF_INET6, entry, &n->addr) != 1 || n->len > 128) {
191 syslog(LOG_ERR, "Invalid static NDP-prefix %s/%s", entry, sep + 1);
195 list_add(&n->head, &neighbors);
204 // Send an ICMP-ECHO. This is less for actually pinging but for the
205 // neighbor cache to be kept up-to-date.
206 static ssize_t ping6(struct in6_addr *addr,
207 const struct interface *iface)
209 struct sockaddr_in6 dest = {AF_INET6, 0, 0, *addr, 0};
210 struct icmp6_hdr echo = {.icmp6_type = ICMP6_ECHO_REQUEST};
211 struct iovec iov = {&echo, sizeof(echo)};
213 // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
214 setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
215 iface->ifname, sizeof(iface->ifname));
216 return odhcpd_send(ping_socket, &dest, &iov, 1, iface);
220 // Handle solicitations
221 static void handle_solicit(void *addr, void *data, size_t len,
222 struct interface *iface)
224 struct ip6_hdr *ip6 = data;
225 struct nd_neighbor_solicit *req = (struct nd_neighbor_solicit*)&ip6[1];
226 struct sockaddr_ll *ll = addr;
228 // Solicitation is for duplicate address detection
229 bool ns_is_dad = IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src);
231 // Don't forward any non-DAD solicitation for external ifaces
232 // TODO: check if we should even forward DADs for them
233 if (iface->external && !ns_is_dad)
236 if (len < sizeof(*ip6) + sizeof(*req))
237 return; // Invalid reqicitation
239 if (IN6_IS_ADDR_LINKLOCAL(&req->nd_ns_target) ||
240 IN6_IS_ADDR_LOOPBACK(&req->nd_ns_target) ||
241 IN6_IS_ADDR_MULTICAST(&req->nd_ns_target))
242 return; // Invalid target
244 char ipbuf[INET6_ADDRSTRLEN];
245 inet_ntop(AF_INET6, &req->nd_ns_target, ipbuf, sizeof(ipbuf));
246 syslog(LOG_DEBUG, "Got a NS for %s", ipbuf);
249 odhcpd_get_mac(iface, mac);
250 if (!memcmp(ll->sll_addr, mac, sizeof(mac)) &&
251 ll->sll_pkttype != PACKET_OUTGOING)
252 return; // Looped back
254 time_t now = time(NULL);
256 struct ndp_neighbor *n = find_neighbor(&req->nd_ns_target, false);
257 if (n && (n->iface || abs(n->timeout - now) < 5)) {
258 syslog(LOG_DEBUG, "%s is on %s", ipbuf,
259 (n->iface) ? n->iface->ifname : "<pending>");
260 if (!n->iface || n->iface == iface)
263 // Found on other interface, answer with advertisement
265 struct nd_neighbor_advert body;
266 struct nd_opt_hdr opt_ll_hdr;
270 .nd_na_hdr = {ND_NEIGHBOR_ADVERT,
272 .nd_na_target = req->nd_ns_target,
274 .opt_ll_hdr = {ND_OPT_TARGET_LINKADDR, 1},
277 memcpy(advert.mac, mac, sizeof(advert.mac));
278 advert.body.nd_na_flags_reserved = ND_NA_FLAG_ROUTER |
279 ND_NA_FLAG_SOLICITED;
281 struct sockaddr_in6 dest = {AF_INET6, 0, 0, ALL_IPV6_NODES, 0};
282 if (!ns_is_dad) // If not DAD, then unicast to source
283 dest.sin6_addr = ip6->ip6_src;
285 // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
286 setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
287 iface->ifname, sizeof(iface->ifname));
288 struct iovec iov = {&advert, sizeof(advert)};
289 odhcpd_send(ping_socket, &dest, &iov, 1, iface);
291 // Send echo to all other interfaces to see where target is on
292 // This will trigger neighbor discovery which is what we want.
293 // We will observe the neighbor cache to see results.
297 list_for_each_entry(c, &interfaces, head)
298 if (iface->ndp == RELAYD_RELAY && iface != c &&
299 (!ns_is_dad || !c->external == false))
300 sent += ping6(&req->nd_ns_target, c);
302 if (sent > 0) // Sent a ping, add pending neighbor entry
303 modify_neighbor(&req->nd_ns_target, NULL, true);
308 void odhcpd_setup_route(const struct in6_addr *addr, int prefixlen,
309 const struct interface *iface, const struct in6_addr *gw, bool add)
314 struct rtattr rta_dst;
315 struct in6_addr dst_addr;
316 struct rtattr rta_oif;
318 struct rtattr rta_table;
320 struct rtattr rta_gw;
323 {sizeof(req), 0, NLM_F_REQUEST, ++rtnl_seqid, 0},
324 {AF_INET6, prefixlen, 0, 0, 0, 0, 0, 0, 0},
325 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_DST},
327 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_OIF},
329 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_TABLE},
331 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_GATEWAY},
339 req.nh.nlmsg_type = RTM_NEWROUTE;
340 req.nh.nlmsg_flags |= (NLM_F_CREATE | NLM_F_REPLACE);
341 req.rtm.rtm_protocol = RTPROT_BOOT;
342 req.rtm.rtm_scope = (gw) ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK;
343 req.rtm.rtm_type = RTN_UNICAST;
345 req.nh.nlmsg_type = RTM_DELROUTE;
346 req.rtm.rtm_scope = RT_SCOPE_NOWHERE;
349 size_t reqlen = (gw) ? sizeof(req) : offsetof(struct req, rta_gw);
350 send(rtnl_event.uloop.fd, &req, reqlen, MSG_DONTWAIT);
353 // Use rtnetlink to modify kernel routes
354 static void setup_route(struct in6_addr *addr, struct interface *iface,
357 char namebuf[INET6_ADDRSTRLEN];
358 inet_ntop(AF_INET6, addr, namebuf, sizeof(namebuf));
359 syslog(LOG_NOTICE, "%s about %s on %s", (add) ? "Learned" : "Forgot",
360 namebuf, (iface) ? iface->ifname : "<pending>");
362 if (!iface || !iface->learn_routes)
365 odhcpd_setup_route(addr, 128, iface, NULL, add);
368 static void free_neighbor(struct ndp_neighbor *n)
370 setup_route(&n->addr, n->iface, false);
376 static struct ndp_neighbor* find_neighbor(struct in6_addr *addr, bool strict)
378 time_t now = time(NULL);
379 struct ndp_neighbor *n, *e;
380 list_for_each_entry_safe(n, e, &neighbors, head) {
381 if ((!strict && !odhcpd_bmemcmp(&n->addr, addr, n->len)) ||
382 (n->len == 128 && IN6_ARE_ADDR_EQUAL(&n->addr, addr)))
385 if (!n->iface && abs(n->timeout - now) >= 5)
392 // Modified our own neighbor-entries
393 static void modify_neighbor(struct in6_addr *addr,
394 struct interface *iface, bool add)
396 if (!addr || (void*)addr == (void*)iface)
399 struct ndp_neighbor *n = find_neighbor(addr, true);
400 if (!add) { // Delete action
401 if (n && (!n->iface || n->iface == iface))
403 } else if (!n) { // No entry yet, add one if possible
404 if (neighbor_count >= NDP_MAX_NEIGHBORS ||
405 !(n = malloc(sizeof(*n))))
413 list_add(&n->head, &neighbors);
415 setup_route(addr, n->iface, add);
416 } else if (n->iface == iface) {
419 } else if (iface && (!n->iface ||
420 (!iface->external && n->iface->external))) {
421 setup_route(addr, n->iface, false);
423 setup_route(addr, n->iface, add);
425 // TODO: In case a host switches interfaces we might want
426 // to set its old neighbor entry to NUD_STALE and ping it
427 // on the old interface to confirm if the MACs match.
431 // Handler for neighbor cache entries from the kernel. This is our source
432 // to learn and unlearn hosts on interfaces.
433 static void handle_rtnetlink(_unused void *addr, void *data, size_t len,
434 _unused struct interface *iface)
436 for (struct nlmsghdr *nh = data; NLMSG_OK(nh, len);
437 nh = NLMSG_NEXT(nh, len)) {
438 struct rtmsg *rtm = NLMSG_DATA(nh);
439 if ((nh->nlmsg_type == RTM_NEWROUTE ||
440 nh->nlmsg_type == RTM_DELROUTE) &&
441 rtm->rtm_dst_len == 0)
442 raise(SIGUSR1); // Inform about a change in default route
444 struct ndmsg *ndm = NLMSG_DATA(nh);
445 struct ifaddrmsg *ifa = NLMSG_DATA(nh);
446 if (nh->nlmsg_type != RTM_NEWNEIGH
447 && nh->nlmsg_type != RTM_DELNEIGH
448 && nh->nlmsg_type != RTM_NEWADDR
449 && nh->nlmsg_type != RTM_DELADDR)
450 continue; // Unrelated message type
451 bool is_addr = (nh->nlmsg_type == RTM_NEWADDR
452 || nh->nlmsg_type == RTM_DELADDR);
454 // Family and ifindex are on the same offset for NEIGH and ADDR
455 if (NLMSG_PAYLOAD(nh, 0) < sizeof(*ndm)
456 || ndm->ndm_family != AF_INET6)
460 struct interface *iface;
461 if (!(iface = odhcpd_get_interface_by_index(ndm->ndm_ifindex)))
465 size_t rta_offset = (is_addr) ? sizeof(*ifa) : sizeof(*ndm);
466 uint16_t atype = (is_addr) ? IFA_ADDRESS : NDA_DST;
467 ssize_t alen = NLMSG_PAYLOAD(nh, rta_offset);
468 struct in6_addr *addr = NULL;
470 for (struct rtattr *rta = (void*)(((uint8_t*)ndm) + rta_offset);
471 RTA_OK(rta, alen); rta = RTA_NEXT(rta, alen))
472 if (rta->rta_type == atype &&
473 RTA_PAYLOAD(rta) >= sizeof(*addr))
474 addr = RTA_DATA(rta);
476 // Address not specified or unrelated
477 if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
478 IN6_IS_ADDR_MULTICAST(addr))
484 add = (nh->nlmsg_type == RTM_NEWADDR);
486 add = (nh->nlmsg_type == RTM_NEWNEIGH && (ndm->ndm_state &
487 (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE
488 | NUD_PERMANENT | NUD_NOARP)));
490 if (iface->ndp == RELAYD_RELAY)
491 modify_neighbor(addr, iface, add);
493 if (is_addr && iface->ra == RELAYD_SERVER)
494 raise(SIGUSR1); // Inform about a change in addresses
496 if (is_addr && iface->dhcpv6 == RELAYD_SERVER)
497 iface->ia_reconf = true;
499 if (iface->ndp == RELAYD_RELAY && is_addr && iface->master) {
500 // Replay address changes on all slave interfaces
501 nh->nlmsg_flags = NLM_F_REQUEST;
503 if (nh->nlmsg_type == RTM_NEWADDR)
504 nh->nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
507 list_for_each_entry(c, &interfaces, head) {
508 if (c->ndp == RELAYD_RELAY && !c->master) {
509 ifa->ifa_index = c->ifindex;
510 send(rtnl_event.uloop.fd, nh, nh->nlmsg_len, MSG_DONTWAIT);
515 /* TODO: See if this is required for optimal operation
516 // Keep neighbor entries alive so we don't loose routes
518 if (add && (ndm->ndm_state & NUD_STALE))