Merge pull request #3 from dedeckeh/master
[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 <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>
26
27 #include <linux/rtnetlink.h>
28 #include <linux/filter.h>
29 #include "router.h"
30 #include "ndp.h"
31
32
33
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,
40                 bool add);
41 static ssize_t ping6(struct in6_addr *addr,
42                 const struct interface *iface);
43
44 static struct list_head neighbors = LIST_HEAD_INIT(neighbors);
45 static size_t neighbor_count = 0;
46 static uint32_t rtnl_seqid = 0;
47
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};
51
52
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),
62 };
63 static const struct sock_fprog bpf_prog = {sizeof(bpf) / sizeof(*bpf), bpf};
64
65
66 // Initialize NDP-proxy
67 int init_ndp(void)
68 {
69         // Setup netlink socket
70         if ((rtnl_event.uloop.fd = odhcpd_open_rtnl()) < 0)
71                 return -1;
72
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));
80
81         // Synthesize initial address events
82         struct {
83                 struct nlmsghdr nh;
84                 struct ifaddrmsg ifa;
85         } req2 = {
86                 {sizeof(req2), RTM_GETADDR, NLM_F_REQUEST | NLM_F_DUMP,
87                                 ++rtnl_seqid, 0},
88                 {.ifa_family = AF_INET6}
89         };
90         send(rtnl_event.uloop.fd, &req2, sizeof(req2), MSG_DONTWAIT);
91         odhcpd_register(&rtnl_event);
92
93
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
97         if (sock < 0) {
98                 syslog(LOG_ERR, "Unable to open packet socket: %s",
99                                 strerror(errno));
100                 return -1;
101         }
102
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));
106                 return -1;
107         }
108
109         ndp_event.uloop.fd = sock;
110         odhcpd_register(&ndp_event);
111
112         // Open ICMPv6 socket
113         ping_socket = socket(AF_INET6, SOCK_RAW | SOCK_CLOEXEC, IPPROTO_ICMPV6);
114
115         int val = 2;
116         setsockopt(ping_socket, IPPROTO_RAW, IPV6_CHECKSUM, &val, sizeof(val));
117
118         // This is required by RFC 4861
119         val = 255;
120         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &val, sizeof(val));
121         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &val, sizeof(val));
122
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));
127
128
129         // Netlink socket, continued...
130         group = RTNLGRP_NEIGH;
131         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
132
133         // Synthesize initial neighbor events
134         struct {
135                 struct nlmsghdr nh;
136                 struct ndmsg ndm;
137         } req = {
138                 {sizeof(req), RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP,
139                                 ++rtnl_seqid, 0},
140                 {.ndm_family = AF_INET6}
141         };
142         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
143
144         return 0;
145 }
146
147
148 int setup_ndp_interface(struct interface *iface, bool enable)
149 {
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));
152
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);
157
158         if (enable && iface->ndp == RELAYD_RELAY) {
159                 setsockopt(ndp_event.uloop.fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP, &mreq, sizeof(mreq));
160
161                 if (iface->static_ndp_len) {
162                         char *entry = alloca(iface->static_ndp_len), *saveptr;
163                         if (!entry) {
164                                 syslog(LOG_ERR, "Alloca failed for static NDP list");
165                                 return -1;
166                         }
167                         memcpy(entry, iface->static_ndp, iface->static_ndp_len);
168
169                         for (entry = strtok_r(entry, " ", &saveptr); entry; entry = strtok_r(NULL, " ", &saveptr)) {
170                                 struct ndp_neighbor *n = malloc(sizeof(*n));
171                                 if (!n) {
172                                         syslog(LOG_ERR, "Malloc failed for static NDP-prefix %s", entry);
173                                         return -1;
174                                 }
175
176                                 n->iface = iface;
177                                 n->timeout = 0;
178
179                                 char ipbuf[INET6_ADDRSTRLEN];
180                                 if (sscanf(entry, "%45s/%hhu", ipbuf, &n->len) < 2
181                                                 || n->len > 128 || inet_pton(AF_INET6, ipbuf, &n->addr) != 1) {
182                                         syslog(LOG_ERR, "Invalid static NDP-prefix %s", entry);
183                                         return -1;
184                                 }
185
186                                 list_add(&n->head, &neighbors);
187                         }
188                 }
189         }
190
191         return 0;
192 }
193
194
195 // Send an ICMP-ECHO. This is less for actually pinging but for the
196 // neighbor cache to be kept up-to-date.
197 static ssize_t ping6(struct in6_addr *addr,
198                 const struct interface *iface)
199 {
200         struct sockaddr_in6 dest = {AF_INET6, 0, 0, *addr, 0};
201         struct icmp6_hdr echo = {.icmp6_type = ICMP6_ECHO_REQUEST};
202         struct iovec iov = {&echo, sizeof(echo)};
203
204         // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
205         setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
206                         iface->ifname, sizeof(iface->ifname));
207         return odhcpd_send(ping_socket, &dest, &iov, 1, iface);
208 }
209
210
211 // Handle solicitations
212 static void handle_solicit(void *addr, void *data, size_t len,
213                 struct interface *iface)
214 {
215         struct ip6_hdr *ip6 = data;
216         struct nd_neighbor_solicit *req = (struct nd_neighbor_solicit*)&ip6[1];
217         struct sockaddr_ll *ll = addr;
218
219         // Solicitation is for duplicate address detection
220         bool ns_is_dad = IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src);
221
222         // Don't forward any non-DAD solicitation for external ifaces
223         // TODO: check if we should even forward DADs for them
224         if (iface->external && !ns_is_dad)
225                 return;
226
227         if (len < sizeof(*ip6) + sizeof(*req))
228                 return; // Invalid reqicitation
229
230         if (IN6_IS_ADDR_LINKLOCAL(&req->nd_ns_target) ||
231                         IN6_IS_ADDR_LOOPBACK(&req->nd_ns_target) ||
232                         IN6_IS_ADDR_MULTICAST(&req->nd_ns_target))
233                 return; // Invalid target
234
235         char ipbuf[INET6_ADDRSTRLEN];
236         inet_ntop(AF_INET6, &req->nd_ns_target, ipbuf, sizeof(ipbuf));
237         syslog(LOG_NOTICE, "Got a NS for %s", ipbuf);
238
239         uint8_t mac[6];
240         odhcpd_get_mac(iface, mac);
241         if (!memcmp(ll->sll_addr, mac, sizeof(mac)) &&
242                         ll->sll_pkttype != PACKET_OUTGOING)
243                 return; // Looped back
244
245         time_t now = time(NULL);
246
247         struct ndp_neighbor *n = find_neighbor(&req->nd_ns_target, false);
248         if (n && (n->iface || abs(n->timeout - now) < 5)) {
249                 syslog(LOG_NOTICE, "%s is on %s", ipbuf,
250                                 (n->iface) ? n->iface->ifname : "<pending>");
251                 if (!n->iface || n->iface == iface)
252                         return;
253
254                 // Found on other interface, answer with advertisement
255                 struct {
256                         struct nd_neighbor_advert body;
257                         struct nd_opt_hdr opt_ll_hdr;
258                         uint8_t mac[6];
259                 } advert = {
260                         .body = {
261                                 .nd_na_hdr = {ND_NEIGHBOR_ADVERT,
262                                         0, 0, {{0}}},
263                                 .nd_na_target = req->nd_ns_target,
264                         },
265                         .opt_ll_hdr = {ND_OPT_TARGET_LINKADDR, 1},
266                 };
267
268                 memcpy(advert.mac, mac, sizeof(advert.mac));
269                 advert.body.nd_na_flags_reserved = ND_NA_FLAG_ROUTER |
270                                 ND_NA_FLAG_SOLICITED;
271
272                 struct sockaddr_in6 dest = {AF_INET6, 0, 0, ALL_IPV6_NODES, 0};
273                 if (!ns_is_dad) // If not DAD, then unicast to source
274                         dest.sin6_addr = ip6->ip6_src;
275
276                 // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
277                 setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
278                                         iface->ifname, sizeof(iface->ifname));
279                 struct iovec iov = {&advert, sizeof(advert)};
280                 odhcpd_send(ping_socket, &dest, &iov, 1, iface);
281         } else {
282                 // Send echo to all other interfaces to see where target is on
283                 // This will trigger neighbor discovery which is what we want.
284                 // We will observe the neighbor cache to see results.
285
286                 ssize_t sent = 0;
287                 struct interface *c;
288                 list_for_each_entry(c, &interfaces, head)
289                         if (iface->ndp == RELAYD_RELAY && iface != c &&
290                                         (!ns_is_dad || !c->external == false))
291                                 sent += ping6(&req->nd_ns_target, c);
292
293                 if (sent > 0) // Sent a ping, add pending neighbor entry
294                         modify_neighbor(&req->nd_ns_target, NULL, true);
295         }
296 }
297
298
299 void odhcpd_setup_route(const struct in6_addr *addr, int prefixlen,
300                 const struct interface *iface, const struct in6_addr *gw, bool add)
301 {
302         struct req {
303                 struct nlmsghdr nh;
304                 struct rtmsg rtm;
305                 struct rtattr rta_dst;
306                 struct in6_addr dst_addr;
307                 struct rtattr rta_oif;
308                 uint32_t ifindex;
309                 struct rtattr rta_table;
310                 uint32_t table;
311                 struct rtattr rta_gw;
312                 struct in6_addr gw;
313         } req = {
314                 {sizeof(req), 0, NLM_F_REQUEST, ++rtnl_seqid, 0},
315                 {AF_INET6, prefixlen, 0, 0, 0, 0, 0, 0, 0},
316                 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_DST},
317                 *addr,
318                 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_OIF},
319                 iface->ifindex,
320                 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_TABLE},
321                 RT_TABLE_MAIN,
322                 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_GATEWAY},
323                 IN6ADDR_ANY_INIT,
324         };
325
326         if (gw)
327                 req.gw = *gw;
328
329         if (add) {
330                 req.nh.nlmsg_type = RTM_NEWROUTE;
331                 req.nh.nlmsg_flags |= (NLM_F_CREATE | NLM_F_REPLACE);
332                 req.rtm.rtm_protocol = RTPROT_BOOT;
333                 req.rtm.rtm_scope = (gw) ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK;
334                 req.rtm.rtm_type = RTN_UNICAST;
335         } else {
336                 req.nh.nlmsg_type = RTM_DELROUTE;
337                 req.rtm.rtm_scope = RT_SCOPE_NOWHERE;
338         }
339
340         size_t reqlen = (gw) ? sizeof(req) : offsetof(struct req, rta_gw);
341         send(rtnl_event.uloop.fd, &req, reqlen, MSG_DONTWAIT);
342 }
343
344 // Use rtnetlink to modify kernel routes
345 static void setup_route(struct in6_addr *addr, struct interface *iface,
346                 bool add)
347 {
348         char namebuf[INET6_ADDRSTRLEN];
349         inet_ntop(AF_INET6, addr, namebuf, sizeof(namebuf));
350         syslog(LOG_NOTICE, "%s about %s on %s", (add) ? "Learned" : "Forgot",
351                         namebuf, (iface) ? iface->ifname : "<pending>");
352
353         if (!iface || !iface->learn_routes)
354                 return;
355
356         odhcpd_setup_route(addr, 128, iface, NULL, add);
357 }
358
359 static void free_neighbor(struct ndp_neighbor *n)
360 {
361         setup_route(&n->addr, n->iface, false);
362         list_del(&n->head);
363         free(n);
364         --neighbor_count;
365 }
366
367
368 static bool match_neighbor(struct ndp_neighbor *n, struct in6_addr *addr)
369 {
370         if (n->len <= 32)
371                 return ntohl(n->addr.s6_addr32[0]) >> (32 - n->len) ==
372                                 ntohl(addr->s6_addr32[0]) >> (32 - n->len);
373
374         if (n->addr.s6_addr32[0] != addr->s6_addr32[0])
375                 return false;
376
377         if (n->len <= 64)
378                 return ntohl(n->addr.s6_addr32[1]) >> (64 - n->len) ==
379                                 ntohl(addr->s6_addr32[1]) >> (64 - n->len);
380
381         if (n->addr.s6_addr32[1] != addr->s6_addr32[1])
382                 return false;
383
384         if (n->len <= 96)
385                 return ntohl(n->addr.s6_addr32[2]) >> (96 - n->len) ==
386                                 ntohl(addr->s6_addr32[2]) >> (96 - n->len);
387
388         if (n->addr.s6_addr32[2] != addr->s6_addr32[2])
389                 return false;
390
391         return ntohl(n->addr.s6_addr32[3]) >> (128 - n->len) ==
392                         ntohl(addr->s6_addr32[3]) >> (128 - n->len);
393 }
394
395
396 static struct ndp_neighbor* find_neighbor(struct in6_addr *addr, bool strict)
397 {
398         time_t now = time(NULL);
399         struct ndp_neighbor *n, *e;
400         list_for_each_entry_safe(n, e, &neighbors, head) {
401                 if ((!strict && match_neighbor(n, addr)) ||
402                                 (n->len == 128 && IN6_ARE_ADDR_EQUAL(&n->addr, addr)))
403                         return n;
404
405                 if (!n->iface && abs(n->timeout - now) >= 5)
406                         free_neighbor(n);
407         }
408         return NULL;
409 }
410
411
412 // Modified our own neighbor-entries
413 static void modify_neighbor(struct in6_addr *addr,
414                 struct interface *iface, bool add)
415 {
416         if (!addr || (void*)addr == (void*)iface)
417                 return;
418
419         struct ndp_neighbor *n = find_neighbor(addr, true);
420         if (!add) { // Delete action
421                 if (n && (!n->iface || n->iface == iface))
422                         free_neighbor(n);
423         } else if (!n) { // No entry yet, add one if possible
424                 if (neighbor_count >= NDP_MAX_NEIGHBORS ||
425                                 !(n = malloc(sizeof(*n))))
426                         return;
427
428                 n->len = 128;
429                 n->addr = *addr;
430                 n->iface = iface;
431                 if (!n->iface)
432                         time(&n->timeout);
433                 list_add(&n->head, &neighbors);
434                 ++neighbor_count;
435                 setup_route(addr, n->iface, add);
436         } else if (n->iface == iface) {
437                 if (!n->iface)
438                         time(&n->timeout);
439         } else if (iface && (!n->iface ||
440                         (!iface->external && n->iface->external))) {
441                 setup_route(addr, n->iface, false);
442                 n->iface = iface;
443                 setup_route(addr, n->iface, add);
444         }
445         // TODO: In case a host switches interfaces we might want
446         // to set its old neighbor entry to NUD_STALE and ping it
447         // on the old interface to confirm if the MACs match.
448 }
449
450
451 // Handler for neighbor cache entries from the kernel. This is our source
452 // to learn and unlearn hosts on interfaces.
453 static void handle_rtnetlink(_unused void *addr, void *data, size_t len,
454                 _unused struct interface *iface)
455 {
456         for (struct nlmsghdr *nh = data; NLMSG_OK(nh, len);
457                         nh = NLMSG_NEXT(nh, len)) {
458                 struct rtmsg *rtm = NLMSG_DATA(nh);
459                 if ((nh->nlmsg_type == RTM_NEWROUTE ||
460                                 nh->nlmsg_type == RTM_DELROUTE) &&
461                                 rtm->rtm_dst_len == 0)
462                         raise(SIGUSR1); // Inform about a change in default route
463
464                 struct ndmsg *ndm = NLMSG_DATA(nh);
465                 struct ifaddrmsg *ifa = NLMSG_DATA(nh);
466                 if (nh->nlmsg_type != RTM_NEWNEIGH
467                                 && nh->nlmsg_type != RTM_DELNEIGH
468                                 && nh->nlmsg_type != RTM_NEWADDR
469                                 && nh->nlmsg_type != RTM_DELADDR)
470                         continue; // Unrelated message type
471                 bool is_addr = (nh->nlmsg_type == RTM_NEWADDR
472                                 || nh->nlmsg_type == RTM_DELADDR);
473
474                 // Family and ifindex are on the same offset for NEIGH and ADDR
475                 if (NLMSG_PAYLOAD(nh, 0) < sizeof(*ndm)
476                                 || ndm->ndm_family != AF_INET6)
477                         continue; //
478
479                 // Lookup interface
480                 struct interface *iface;
481                 if (!(iface = odhcpd_get_interface_by_index(ndm->ndm_ifindex)))
482                         continue;
483
484                 // Data to retrieve
485                 size_t rta_offset = (is_addr) ? sizeof(*ifa) : sizeof(*ndm);
486                 uint16_t atype = (is_addr) ? IFA_ADDRESS : NDA_DST;
487                 ssize_t alen = NLMSG_PAYLOAD(nh, rta_offset);
488                 struct in6_addr *addr = NULL;
489
490                 for (struct rtattr *rta = (void*)(((uint8_t*)ndm) + rta_offset);
491                                 RTA_OK(rta, alen); rta = RTA_NEXT(rta, alen))
492                         if (rta->rta_type == atype &&
493                                         RTA_PAYLOAD(rta) >= sizeof(*addr))
494                                 addr = RTA_DATA(rta);
495
496                 // Address not specified or unrelated
497                 if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
498                                 IN6_IS_ADDR_MULTICAST(addr))
499                         continue;
500
501                 // Check for states
502                 bool add;
503                 if (is_addr)
504                         add = (nh->nlmsg_type == RTM_NEWADDR);
505                 else
506                         add = (nh->nlmsg_type == RTM_NEWNEIGH && (ndm->ndm_state &
507                                 (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE
508                                                 | NUD_PERMANENT | NUD_NOARP)));
509
510                 if (iface->ndp == RELAYD_RELAY)
511                         modify_neighbor(addr, iface, add);
512
513                 if (is_addr && iface->ra == RELAYD_SERVER)
514                         raise(SIGUSR1); // Inform about a change in addresses
515
516                 if (is_addr && iface->dhcpv6 == RELAYD_SERVER)
517                         iface->ia_reconf = true;
518
519                 if (iface->ndp == RELAYD_RELAY && is_addr && iface->master) {
520                         // Replay address changes on all slave interfaces
521                         nh->nlmsg_flags = NLM_F_REQUEST;
522
523                         if (nh->nlmsg_type == RTM_NEWADDR)
524                                 nh->nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
525
526                         struct interface *c;
527                         list_for_each_entry(c, &interfaces, head) {
528                                 if (c->ndp == RELAYD_RELAY && !c->master) {
529                                         ifa->ifa_index = c->ifindex;
530                                         send(rtnl_event.uloop.fd, nh, nh->nlmsg_len, MSG_DONTWAIT);
531                                 }
532                         }
533                 }
534
535                 /* TODO: See if this is required for optimal operation
536                 // Keep neighbor entries alive so we don't loose routes
537                 if (add && (ndm->ndm_state & NUD_STALE))
538                         ping6(addr, iface);
539                 */
540         }
541 }