config: restore interface defaults when cleaning interface
[project/odhcpd.git] / 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 "dhcpv6.h"
33 #include "ndp.h"
34
35
36
37 static void handle_solicit(void *addr, void *data, size_t len,
38                 struct interface *iface, void *dest);
39 static void handle_rtnetlink(void *addr, void *data, size_t len,
40                 struct interface *iface, void *dest);
41 static void catch_rtnetlink(int error);
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, catch_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         int val = 256 * 1024;
65
66         // Setup netlink socket
67         if ((rtnl_event.uloop.fd = socket(AF_NETLINK, SOCK_RAW | SOCK_CLOEXEC, NETLINK_ROUTE)) < 0)
68                 return -1;
69
70         struct sockaddr_nl nl = {.nl_family = AF_NETLINK};
71         if (connect(rtnl_event.uloop.fd, (struct sockaddr*)&nl, sizeof(nl)) < 0)
72                 return -1;
73
74         if (setsockopt(rtnl_event.uloop.fd, SOL_SOCKET, SO_RCVBUF, &val, sizeof(val)))
75                 setsockopt(rtnl_event.uloop.fd, SOL_SOCKET, SO_RCVBUFFORCE, &val, sizeof(val));
76
77         // Receive netlink neighbor and ip-address events
78         uint32_t group = RTNLGRP_IPV6_IFADDR;
79         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
80                         NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
81         group = RTNLGRP_IPV6_ROUTE;
82         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK,
83                         NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
84
85         odhcpd_register(&rtnl_event);
86
87         // Open ICMPv6 socket
88         ping_socket = socket(AF_INET6, SOCK_RAW | SOCK_CLOEXEC, IPPROTO_ICMPV6);
89         if (ping_socket < 0) {
90                 syslog(LOG_ERR, "Unable to open raw socket: %s", strerror(errno));
91                         return -1;
92         }
93
94         val = 2;
95         setsockopt(ping_socket, IPPROTO_RAW, IPV6_CHECKSUM, &val, sizeof(val));
96
97         // This is required by RFC 4861
98         val = 255;
99         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &val, sizeof(val));
100         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &val, sizeof(val));
101
102         // Filter all packages, we only want to send
103         struct icmp6_filter filt;
104         ICMP6_FILTER_SETBLOCKALL(&filt);
105         setsockopt(ping_socket, IPPROTO_ICMPV6, ICMP6_FILTER, &filt, sizeof(filt));
106
107
108         // Netlink socket, continued...
109         group = RTNLGRP_NEIGH;
110         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
111
112         return 0;
113 }
114
115
116 static void dump_neigh_table(bool proxy)
117 {
118         struct {
119                 struct nlmsghdr nh;
120                 struct ndmsg ndm;
121         } req = {
122                 {sizeof(req), RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP,
123                                 ++rtnl_seqid, 0},
124                 {.ndm_family = AF_INET6, .ndm_flags = (proxy) ? NTF_PROXY : 0}
125         };
126         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
127         odhcpd_process(&rtnl_event);
128 }
129
130
131 int setup_ndp_interface(struct interface *iface, bool enable)
132 {
133         char procbuf[64];
134         snprintf(procbuf, sizeof(procbuf), "/proc/sys/net/ipv6/conf/%s/proxy_ndp", iface->ifname);
135         int procfd = open(procbuf, O_WRONLY);
136         bool dump_neigh = false;
137
138         if (iface->ndp_event.uloop.fd > 0) {
139                 uloop_fd_delete(&iface->ndp_event.uloop);
140                 close(iface->ndp_event.uloop.fd);
141                 iface->ndp_event.uloop.fd = -1;
142
143                 if (!enable || iface->ndp != RELAYD_RELAY)
144                         if (write(procfd, "0\n", 2) < 0) {}
145
146                 dump_neigh = true;
147         }
148
149         if (enable && (iface->ra == RELAYD_SERVER ||
150                         iface->dhcpv6 == RELAYD_SERVER || iface->ndp == RELAYD_RELAY)) {
151                 // Synthesize initial address events
152                 struct {
153                         struct nlmsghdr nh;
154                         struct ifaddrmsg ifa;
155                 } req2 = {
156                         {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
157                                         ++rtnl_seqid, 0},
158                         {.ifa_family = AF_INET6, .ifa_index = iface->ifindex}
159                 };
160                 send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
161         }
162
163         if (enable && iface->ndp == RELAYD_RELAY) {
164                 if (write(procfd, "1\n", 2) < 0) {}
165                 close(procfd);
166
167                 int sock = socket(AF_PACKET, SOCK_DGRAM | SOCK_CLOEXEC, htons(ETH_P_IPV6));
168                 if (sock < 0) {
169                         syslog(LOG_ERR, "Unable to open packet socket: %s",
170                                         strerror(errno));
171                         return -1;
172                 }
173
174 #ifdef PACKET_RECV_TYPE
175                 int pktt = 1 << PACKET_MULTICAST;
176                 setsockopt(sock, SOL_PACKET, PACKET_RECV_TYPE, &pktt, sizeof(pktt));
177 #endif
178
179                 if (setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER,
180                                 &bpf_prog, sizeof(bpf_prog))) {
181                         syslog(LOG_ERR, "Failed to set BPF: %s", strerror(errno));
182                         return -1;
183                 }
184
185                 struct sockaddr_ll ll = {
186                         .sll_family = AF_PACKET,
187                         .sll_ifindex = iface->ifindex,
188                         .sll_protocol = htons(ETH_P_IPV6),
189                         .sll_hatype = 0,
190                         .sll_pkttype = 0,
191                         .sll_halen = 0,
192                         .sll_addr = {0},
193                 };
194                 bind(sock, (struct sockaddr*)&ll, sizeof(ll));
195
196                 struct packet_mreq mreq = {iface->ifindex, PACKET_MR_ALLMULTI, ETH_ALEN, {0}};
197                 setsockopt(sock, SOL_PACKET, PACKET_ADD_MEMBERSHIP, &mreq, sizeof(mreq));
198
199                 iface->ndp_event.uloop.fd = sock;
200                 iface->ndp_event.handle_dgram = handle_solicit;
201                 odhcpd_register(&iface->ndp_event);
202
203                 // If we already were enabled dump is unnecessary, if not do dump
204                 if (!dump_neigh)
205                         dump_neigh_table(false);
206                 else
207                         dump_neigh = false;
208         } else {
209                 close(procfd);
210         }
211
212         if (dump_neigh)
213                 dump_neigh_table(true);
214
215         return 0;
216 }
217
218
219 // Send an ICMP-ECHO. This is less for actually pinging but for the
220 // neighbor cache to be kept up-to-date.
221 static void ping6(struct in6_addr *addr,
222                 const struct interface *iface)
223 {
224         struct sockaddr_in6 dest = {AF_INET6, 0, 0, *addr, iface->ifindex};
225         struct icmp6_hdr echo = {.icmp6_type = ICMP6_ECHO_REQUEST};
226         struct iovec iov = {&echo, sizeof(echo)};
227
228         odhcpd_setup_route(addr, 128, iface, NULL, 128, true);
229         odhcpd_send(ping_socket, &dest, &iov, 1, iface);
230         odhcpd_setup_route(addr, 128, iface, NULL, 128, false);
231 }
232
233
234 // Handle solicitations
235 static void handle_solicit(void *addr, void *data, size_t len,
236                 struct interface *iface, _unused void *dest)
237 {
238         struct ip6_hdr *ip6 = data;
239         struct nd_neighbor_solicit *req = (struct nd_neighbor_solicit*)&ip6[1];
240         struct sockaddr_ll *ll = addr;
241
242         // Solicitation is for duplicate address detection
243         bool ns_is_dad = IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src);
244
245         // Don't forward any non-DAD solicitation for external ifaces
246         // TODO: check if we should even forward DADs for them
247         if (iface->external && !ns_is_dad)
248                 return;
249
250         if (len < sizeof(*ip6) + sizeof(*req))
251                 return; // Invalid reqicitation
252
253         if (IN6_IS_ADDR_LINKLOCAL(&req->nd_ns_target) ||
254                         IN6_IS_ADDR_LOOPBACK(&req->nd_ns_target) ||
255                         IN6_IS_ADDR_MULTICAST(&req->nd_ns_target))
256                 return; // Invalid target
257
258         char ipbuf[INET6_ADDRSTRLEN];
259         inet_ntop(AF_INET6, &req->nd_ns_target, ipbuf, sizeof(ipbuf));
260         syslog(LOG_DEBUG, "Got a NS for %s", ipbuf);
261
262         uint8_t mac[6];
263         odhcpd_get_mac(iface, mac);
264         if (!memcmp(ll->sll_addr, mac, sizeof(mac)))
265                 return; // Looped back
266
267         struct interface *c;
268         list_for_each_entry(c, &interfaces, head)
269                 if (iface->ndp == RELAYD_RELAY && iface != c &&
270                                 (ns_is_dad || !c->external))
271                         ping6(&req->nd_ns_target, c);
272 }
273
274 // Use rtnetlink to modify kernel routes
275 static void setup_route(struct in6_addr *addr, struct interface *iface, bool add)
276 {
277         char namebuf[INET6_ADDRSTRLEN];
278         inet_ntop(AF_INET6, addr, namebuf, sizeof(namebuf));
279         syslog(LOG_NOTICE, "%s about %s on %s",
280                         (add) ? "Learned" : "Forgot", namebuf, iface->ifname);
281
282         if (iface->learn_routes)
283                 odhcpd_setup_route(addr, 128, iface, NULL, 1024, add);
284 }
285
286 // compare prefixes
287 static int prefixcmp(const void *va, const void *vb)
288 {
289         const struct odhcpd_ipaddr *a = va, *b = vb;
290         uint32_t a_pref = ((a->addr.s6_addr[0] & 0xfe) != 0xfc) ? a->preferred : 1;
291         uint32_t b_pref = ((b->addr.s6_addr[0] & 0xfe) != 0xfc) ? b->preferred : 1;
292         return (a_pref < b_pref) ? 1 : (a_pref > b_pref) ? -1 : 0;
293 }
294
295 // Check address update
296 static void check_updates(struct interface *iface)
297 {
298         struct odhcpd_ipaddr addr[RELAYD_MAX_ADDRS] = {{IN6ADDR_ANY_INIT, 0, 0, 0, 0}};
299         time_t now = odhcpd_time();
300         ssize_t len = odhcpd_get_interface_addresses(iface->ifindex, addr, ARRAY_SIZE(addr));
301
302         if (len < 0)
303                 return;
304
305         qsort(addr, len, sizeof(*addr), prefixcmp);
306
307         for (int i = 0; i < len; ++i) {
308                 addr[i].addr.s6_addr32[3] = 0;
309
310                 if (addr[i].preferred < UINT32_MAX - now)
311                         addr[i].preferred += now;
312
313                 if (addr[i].valid < UINT32_MAX - now)
314                         addr[i].valid += now;
315         }
316
317         bool change = len != (ssize_t)iface->ia_addr_len;
318         for (ssize_t i = 0; !change && i < len; ++i)
319                 if (!IN6_ARE_ADDR_EQUAL(&addr[i].addr, &iface->ia_addr[i].addr) ||
320                                 (addr[i].preferred > 0) != (iface->ia_addr[i].preferred > 0) ||
321                                 addr[i].valid < iface->ia_addr[i].valid ||
322                                 addr[i].preferred < iface->ia_addr[i].preferred)
323                         change = true;
324
325         if (change)
326                 dhcpv6_ia_preupdate(iface);
327
328         memcpy(iface->ia_addr, addr, len * sizeof(*addr));
329         iface->ia_addr_len = len;
330
331         if (change)
332                 dhcpv6_ia_postupdate(iface, now);
333
334         if (change) {
335                 syslog(LOG_INFO, "Raising SIGUSR1 due to address change on %s", iface->ifname);
336                 raise(SIGUSR1);
337         }
338 }
339
340
341 // Handler for neighbor cache entries from the kernel. This is our source
342 // to learn and unlearn hosts on interfaces.
343 static void handle_rtnetlink(_unused void *addr, void *data, size_t len,
344                 _unused struct interface *iface, _unused void *dest)
345 {
346         bool dump_neigh = false;
347         struct in6_addr last_solicited = IN6ADDR_ANY_INIT;
348
349         for (struct nlmsghdr *nh = data; NLMSG_OK(nh, len);
350                         nh = NLMSG_NEXT(nh, len)) {
351                 struct ndmsg *ndm = NLMSG_DATA(nh);
352                 struct rtmsg *rtm = NLMSG_DATA(nh);
353
354                 bool is_addr = (nh->nlmsg_type == RTM_NEWADDR
355                                 || nh->nlmsg_type == RTM_DELADDR);
356                 bool is_route = (nh->nlmsg_type == RTM_NEWROUTE
357                                 || nh->nlmsg_type == RTM_DELROUTE);
358                 bool is_neigh = (nh->nlmsg_type == RTM_NEWNEIGH
359                                 || nh->nlmsg_type == RTM_DELNEIGH);
360
361                 // Family and ifindex are on the same offset for NEIGH and ADDR
362                 if ((!is_addr && !is_route && !is_neigh)
363                                 || NLMSG_PAYLOAD(nh, 0) < sizeof(*ndm)
364                                 || ndm->ndm_family != AF_INET6)
365                         continue;
366
367                 if (is_route) {
368                         // Inform about a change in default route
369                         if (rtm->rtm_dst_len == 0) {
370                                 syslog(LOG_INFO, "Raising SIGUSR1 due to default route change");
371                                 raise(SIGUSR1);
372                         }
373
374                         continue;
375                 }
376
377                 // Data to retrieve
378                 size_t rta_offset = (is_addr) ? sizeof(struct ifaddrmsg) : sizeof(*ndm);
379                 uint16_t atype = (is_addr) ? IFA_ADDRESS : NDA_DST;
380                 ssize_t alen = NLMSG_PAYLOAD(nh, rta_offset);
381                 struct in6_addr *addr = NULL;
382
383                 for (struct rtattr *rta = (void*)(((uint8_t*)ndm) + rta_offset);
384                                 RTA_OK(rta, alen); rta = RTA_NEXT(rta, alen)) {
385                         if (rta->rta_type == atype &&
386                                         RTA_PAYLOAD(rta) >= sizeof(*addr)) {
387                                 addr = RTA_DATA(rta);
388                         }
389                 }
390
391                 // Lookup interface
392                 struct interface *iface = odhcpd_get_interface_by_index(ndm->ndm_ifindex);
393                 if (!iface)
394                         continue;
395
396                 // Address not specified or unrelated
397                 if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
398                                 IN6_IS_ADDR_MULTICAST(addr))
399                         continue;
400
401                 // Check for states
402                 bool add;
403                 if (is_addr)
404                         add = (nh->nlmsg_type == RTM_NEWADDR);
405                 else
406                         add = (nh->nlmsg_type == RTM_NEWNEIGH && (ndm->ndm_state &
407                                 (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE
408                                                 | NUD_PERMANENT | NUD_NOARP)));
409
410                 if (iface->ndp == RELAYD_RELAY) {
411                         // Replay change to all neighbor cache
412                         struct {
413                                 struct nlmsghdr nh;
414                                 struct ndmsg ndm;
415                                 struct nlattr nla_dst;
416                                 struct in6_addr dst;
417                         } req = {
418                                 {sizeof(req), RTM_DELNEIGH, NLM_F_REQUEST,
419                                                 ++rtnl_seqid, 0},
420                                 {.ndm_family = AF_INET6, .ndm_flags = NTF_PROXY},
421                                 {sizeof(struct nlattr) + sizeof(struct in6_addr), NDA_DST},
422                                 *addr
423                         };
424
425                         if (ndm->ndm_flags & NTF_PROXY) {
426                                 // Dump & flush proxy entries
427                                 if (nh->nlmsg_type == RTM_NEWNEIGH) {
428                                         req.ndm.ndm_ifindex = iface->ifindex;
429                                         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
430                                         setup_route(addr, iface, false);
431                                         dump_neigh = true;
432                                 }
433                         } else if (add) {
434                                 struct interface *c;
435                                 list_for_each_entry(c, &interfaces, head) {
436                                         if (iface == c)
437                                                 continue;
438
439                                         if (c->ndp == RELAYD_RELAY) {
440                                                 req.nh.nlmsg_type = RTM_NEWNEIGH;
441                                                 req.nh.nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
442
443                                                 req.ndm.ndm_ifindex = c->ifindex;
444                                                 send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
445                                         } else { // Delete NDP cache from interfaces without relay
446                                                 req.nh.nlmsg_type = RTM_DELNEIGH;
447                                                 req.nh.nlmsg_flags &= ~(NLM_F_CREATE | NLM_F_REPLACE);
448
449                                                 req.ndm.ndm_ifindex = c->ifindex;
450                                                 send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
451                                         }
452                                 }
453
454                                 setup_route(addr, iface, true);
455                         } else {
456                                 if (nh->nlmsg_type == RTM_NEWNEIGH) {
457                                         // might be locally originating
458                                         if (!IN6_ARE_ADDR_EQUAL(&last_solicited, addr)) {
459                                                 last_solicited = *addr;
460
461                                                 struct interface *c;
462                                                 list_for_each_entry(c, &interfaces, head)
463                                                         if (iface->ndp == RELAYD_RELAY && iface != c &&
464                                                                         !c->external == false)
465                                                                 ping6(addr, c);
466                                         }
467                                 } else {
468                                         struct interface *c;
469                                         list_for_each_entry(c, &interfaces, head) {
470                                                 if (c->ndp == RELAYD_RELAY && iface != c) {
471                                                         req.ndm.ndm_ifindex = c->ifindex;
472                                                         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
473                                                 }
474                                         }
475                                         setup_route(addr, iface, false);
476
477                                         // also: dump to add proxies back in case it moved elsewhere
478                                         dump_neigh = true;
479                                 }
480                         }
481                 }
482
483                 if (is_addr) {
484                         check_updates(iface);
485
486                         if (iface->ndp == RELAYD_RELAY && iface->master) {
487                                 // Replay address changes on all slave interfaces
488                                 nh->nlmsg_flags = NLM_F_REQUEST;
489
490                                 if (nh->nlmsg_type == RTM_NEWADDR)
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                                                 ndm->ndm_ifindex = c->ifindex;
497                                                 send(rtnl_event.uloop.fd, nh, nh->nlmsg_len, MSG_DONTWAIT);
498                                         }
499                                 }
500                         }
501                 }
502         }
503
504         if (dump_neigh)
505                 dump_neigh_table(false);
506 }
507
508 static void catch_rtnetlink(int error)
509 {
510         if (error == ENOBUFS) {
511                 struct {
512                         struct nlmsghdr nh;
513                         struct ifaddrmsg ifa;
514                 } req2 = {
515                         {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
516                                         ++rtnl_seqid, 0},
517                         {.ifa_family = AF_INET6}
518                 };
519                 send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
520         }
521 }