Keep managed PD for at least 150 seconds
[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         if (ping_socket < 0) {
115                 syslog(LOG_ERR, "Unable to open raw socket: %s", strerror(errno));
116                         return -1;
117         }
118
119         int val = 2;
120         setsockopt(ping_socket, IPPROTO_RAW, IPV6_CHECKSUM, &val, sizeof(val));
121
122         // This is required by RFC 4861
123         val = 255;
124         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_MULTICAST_HOPS, &val, sizeof(val));
125         setsockopt(ping_socket, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &val, sizeof(val));
126
127         // Filter all packages, we only want to send
128         struct icmp6_filter filt;
129         ICMP6_FILTER_SETBLOCKALL(&filt);
130         setsockopt(ping_socket, IPPROTO_ICMPV6, ICMP6_FILTER, &filt, sizeof(filt));
131
132
133         // Netlink socket, continued...
134         group = RTNLGRP_NEIGH;
135         setsockopt(rtnl_event.uloop.fd, SOL_NETLINK, NETLINK_ADD_MEMBERSHIP, &group, sizeof(group));
136
137         // Synthesize initial neighbor events
138         struct {
139                 struct nlmsghdr nh;
140                 struct ndmsg ndm;
141         } req = {
142                 {sizeof(req), RTM_GETNEIGH, NLM_F_REQUEST | NLM_F_DUMP,
143                                 ++rtnl_seqid, 0},
144                 {.ndm_family = AF_INET6}
145         };
146         send(rtnl_event.uloop.fd, &req, sizeof(req), MSG_DONTWAIT);
147
148         return 0;
149 }
150
151
152 int setup_ndp_interface(struct interface *iface, bool enable)
153 {
154         struct packet_mreq mreq = {iface->ifindex, PACKET_MR_ALLMULTI, ETH_ALEN, {0}};
155         setsockopt(ndp_event.uloop.fd, SOL_PACKET, PACKET_DROP_MEMBERSHIP, &mreq, sizeof(mreq));
156
157         struct ndp_neighbor *c, *n;
158         list_for_each_entry_safe(c, n, &neighbors, head)
159                 if (c->iface == iface && (c->timeout == 0 || iface->ndp != RELAYD_RELAY || !enable))
160                         modify_neighbor(&c->addr, c->iface, false);
161
162         if (enable && iface->ndp == RELAYD_RELAY) {
163                 setsockopt(ndp_event.uloop.fd, SOL_PACKET, PACKET_ADD_MEMBERSHIP, &mreq, sizeof(mreq));
164
165                 if (iface->static_ndp_len) {
166                         char *entry = alloca(iface->static_ndp_len), *saveptr;
167                         if (!entry) {
168                                 syslog(LOG_ERR, "Alloca failed for static NDP list");
169                                 return -1;
170                         }
171                         memcpy(entry, iface->static_ndp, iface->static_ndp_len);
172
173                         for (entry = strtok_r(entry, " ", &saveptr); entry; entry = strtok_r(NULL, " ", &saveptr)) {
174                                 char *sep;
175                                 struct ndp_neighbor *n = malloc(sizeof(*n));
176                                 if (!n) {
177                                         syslog(LOG_ERR, "Malloc failed for static NDP-prefix %s", entry);
178                                         return -1;
179                                 }
180
181                                 n->iface = iface;
182                                 n->timeout = 0;
183
184                                 sep = strchr(entry, '/');
185                                 if (!sep) {
186                                         free(n);
187                                         syslog(LOG_ERR, "Invalid static NDP-prefix %s", entry);
188                                         return -1;
189                                 }
190                                 
191                                 *sep = 0;
192                                 n->len = atoi(sep + 1);
193                                 if (inet_pton(AF_INET6, entry, &n->addr) != 1 || n->len > 128) {
194                                         free(n);
195                                         syslog(LOG_ERR, "Invalid static NDP-prefix %s/%s", entry, sep + 1);
196                                         return -1;
197                                 }
198
199                                 list_add(&n->head, &neighbors);
200                         }
201                 }
202         }
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 ssize_t ping6(struct in6_addr *addr,
211                 const struct interface *iface)
212 {
213         struct sockaddr_in6 dest = {AF_INET6, 0, 0, *addr, 0};
214         struct icmp6_hdr echo = {.icmp6_type = ICMP6_ECHO_REQUEST};
215         struct iovec iov = {&echo, sizeof(echo)};
216
217         // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
218         setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
219                         iface->ifname, sizeof(iface->ifname));
220         return odhcpd_send(ping_socket, &dest, &iov, 1, iface);
221 }
222
223
224 // Handle solicitations
225 static void handle_solicit(void *addr, void *data, size_t len,
226                 struct interface *iface)
227 {
228         struct ip6_hdr *ip6 = data;
229         struct nd_neighbor_solicit *req = (struct nd_neighbor_solicit*)&ip6[1];
230         struct sockaddr_ll *ll = addr;
231
232         // Solicitation is for duplicate address detection
233         bool ns_is_dad = IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src);
234
235         // Don't forward any non-DAD solicitation for external ifaces
236         // TODO: check if we should even forward DADs for them
237         if (iface->external && !ns_is_dad)
238                 return;
239
240         if (len < sizeof(*ip6) + sizeof(*req))
241                 return; // Invalid reqicitation
242
243         if (IN6_IS_ADDR_LINKLOCAL(&req->nd_ns_target) ||
244                         IN6_IS_ADDR_LOOPBACK(&req->nd_ns_target) ||
245                         IN6_IS_ADDR_MULTICAST(&req->nd_ns_target))
246                 return; // Invalid target
247
248         char ipbuf[INET6_ADDRSTRLEN];
249         inet_ntop(AF_INET6, &req->nd_ns_target, ipbuf, sizeof(ipbuf));
250         syslog(LOG_DEBUG, "Got a NS for %s", ipbuf);
251
252         uint8_t mac[6];
253         odhcpd_get_mac(iface, mac);
254         if (!memcmp(ll->sll_addr, mac, sizeof(mac)) &&
255                         ll->sll_pkttype != PACKET_OUTGOING)
256                 return; // Looped back
257
258         time_t now = time(NULL);
259
260         struct ndp_neighbor *n = find_neighbor(&req->nd_ns_target, false);
261         if (n && (n->iface || abs(n->timeout - now) < 5)) {
262                 syslog(LOG_DEBUG, "%s is on %s", ipbuf,
263                                 (n->iface) ? n->iface->ifname : "<pending>");
264                 if (!n->iface || n->iface == iface)
265                         return;
266
267                 // Found on other interface, answer with advertisement
268                 struct {
269                         struct nd_neighbor_advert body;
270                         struct nd_opt_hdr opt_ll_hdr;
271                         uint8_t mac[6];
272                 } advert = {
273                         .body = {
274                                 .nd_na_hdr = {ND_NEIGHBOR_ADVERT,
275                                         0, 0, {{0}}},
276                                 .nd_na_target = req->nd_ns_target,
277                         },
278                         .opt_ll_hdr = {ND_OPT_TARGET_LINKADDR, 1},
279                 };
280
281                 memcpy(advert.mac, mac, sizeof(advert.mac));
282                 advert.body.nd_na_flags_reserved = ND_NA_FLAG_ROUTER |
283                                 ND_NA_FLAG_SOLICITED;
284
285                 struct sockaddr_in6 dest = {AF_INET6, 0, 0, ALL_IPV6_NODES, 0};
286                 if (!ns_is_dad) // If not DAD, then unicast to source
287                         dest.sin6_addr = ip6->ip6_src;
288
289                 // Linux seems to not honor IPV6_PKTINFO on raw-sockets, so work around
290                 setsockopt(ping_socket, SOL_SOCKET, SO_BINDTODEVICE,
291                                         iface->ifname, sizeof(iface->ifname));
292                 struct iovec iov = {&advert, sizeof(advert)};
293                 odhcpd_send(ping_socket, &dest, &iov, 1, iface);
294         } else {
295                 // Send echo to all other interfaces to see where target is on
296                 // This will trigger neighbor discovery which is what we want.
297                 // We will observe the neighbor cache to see results.
298
299                 ssize_t sent = 0;
300                 struct interface *c;
301                 list_for_each_entry(c, &interfaces, head)
302                         if (iface->ndp == RELAYD_RELAY && iface != c &&
303                                         (!ns_is_dad || !c->external == false))
304                                 sent += ping6(&req->nd_ns_target, c);
305
306                 if (sent > 0) // Sent a ping, add pending neighbor entry
307                         modify_neighbor(&req->nd_ns_target, NULL, true);
308         }
309 }
310
311
312 void odhcpd_setup_route(const struct in6_addr *addr, int prefixlen,
313                 const struct interface *iface, const struct in6_addr *gw, bool add)
314 {
315         struct req {
316                 struct nlmsghdr nh;
317                 struct rtmsg rtm;
318                 struct rtattr rta_dst;
319                 struct in6_addr dst_addr;
320                 struct rtattr rta_oif;
321                 uint32_t ifindex;
322                 struct rtattr rta_table;
323                 uint32_t table;
324                 struct rtattr rta_gw;
325                 struct in6_addr gw;
326         } req = {
327                 {sizeof(req), 0, NLM_F_REQUEST, ++rtnl_seqid, 0},
328                 {AF_INET6, prefixlen, 0, 0, 0, 0, 0, 0, 0},
329                 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_DST},
330                 *addr,
331                 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_OIF},
332                 iface->ifindex,
333                 {sizeof(struct rtattr) + sizeof(uint32_t), RTA_TABLE},
334                 RT_TABLE_MAIN,
335                 {sizeof(struct rtattr) + sizeof(struct in6_addr), RTA_GATEWAY},
336                 IN6ADDR_ANY_INIT,
337         };
338
339         if (gw)
340                 req.gw = *gw;
341
342         if (add) {
343                 req.nh.nlmsg_type = RTM_NEWROUTE;
344                 req.nh.nlmsg_flags |= (NLM_F_CREATE | NLM_F_REPLACE);
345                 req.rtm.rtm_protocol = RTPROT_BOOT;
346                 req.rtm.rtm_scope = (gw) ? RT_SCOPE_UNIVERSE : RT_SCOPE_LINK;
347                 req.rtm.rtm_type = RTN_UNICAST;
348         } else {
349                 req.nh.nlmsg_type = RTM_DELROUTE;
350                 req.rtm.rtm_scope = RT_SCOPE_NOWHERE;
351         }
352
353         size_t reqlen = (gw) ? sizeof(req) : offsetof(struct req, rta_gw);
354         send(rtnl_event.uloop.fd, &req, reqlen, MSG_DONTWAIT);
355 }
356
357 // Use rtnetlink to modify kernel routes
358 static void setup_route(struct in6_addr *addr, struct interface *iface,
359                 bool add)
360 {
361         char namebuf[INET6_ADDRSTRLEN];
362         inet_ntop(AF_INET6, addr, namebuf, sizeof(namebuf));
363         syslog(LOG_NOTICE, "%s about %s on %s", (add) ? "Learned" : "Forgot",
364                         namebuf, (iface) ? iface->ifname : "<pending>");
365
366         if (!iface || !iface->learn_routes)
367                 return;
368
369         odhcpd_setup_route(addr, 128, iface, NULL, add);
370 }
371
372 static void free_neighbor(struct ndp_neighbor *n)
373 {
374         setup_route(&n->addr, n->iface, false);
375         list_del(&n->head);
376         free(n);
377         --neighbor_count;
378 }
379
380 static struct ndp_neighbor* find_neighbor(struct in6_addr *addr, bool strict)
381 {
382         time_t now = time(NULL);
383         struct ndp_neighbor *n, *e;
384         list_for_each_entry_safe(n, e, &neighbors, head) {
385                 if ((!strict && !odhcpd_bmemcmp(&n->addr, addr, n->len)) ||
386                                 (n->len == 128 && IN6_ARE_ADDR_EQUAL(&n->addr, addr)))
387                         return n;
388
389                 if (!n->iface && abs(n->timeout - now) >= 5)
390                         free_neighbor(n);
391         }
392         return NULL;
393 }
394
395
396 // Modified our own neighbor-entries
397 static void modify_neighbor(struct in6_addr *addr,
398                 struct interface *iface, bool add)
399 {
400         if (!addr || (void*)addr == (void*)iface)
401                 return;
402
403         struct ndp_neighbor *n = find_neighbor(addr, true);
404         if (!add) { // Delete action
405                 if (n && (!n->iface || n->iface == iface))
406                         free_neighbor(n);
407         } else if (!n) { // No entry yet, add one if possible
408                 if (neighbor_count >= NDP_MAX_NEIGHBORS ||
409                                 !(n = malloc(sizeof(*n))))
410                         return;
411
412                 n->len = 128;
413                 n->addr = *addr;
414                 n->iface = iface;
415                 if (!n->iface)
416                         time(&n->timeout);
417                 list_add(&n->head, &neighbors);
418                 ++neighbor_count;
419                 setup_route(addr, n->iface, add);
420         } else if (n->iface == iface) {
421                 if (!n->iface)
422                         time(&n->timeout);
423         } else if (iface && (!n->iface ||
424                         (!iface->external && n->iface->external))) {
425                 setup_route(addr, n->iface, false);
426                 n->iface = iface;
427                 setup_route(addr, n->iface, add);
428         }
429         // TODO: In case a host switches interfaces we might want
430         // to set its old neighbor entry to NUD_STALE and ping it
431         // on the old interface to confirm if the MACs match.
432 }
433
434
435 // Handler for neighbor cache entries from the kernel. This is our source
436 // to learn and unlearn hosts on interfaces.
437 static void handle_rtnetlink(_unused void *addr, void *data, size_t len,
438                 _unused struct interface *iface)
439 {
440         for (struct nlmsghdr *nh = data; NLMSG_OK(nh, len);
441                         nh = NLMSG_NEXT(nh, len)) {
442                 struct rtmsg *rtm = NLMSG_DATA(nh);
443                 if ((nh->nlmsg_type == RTM_NEWROUTE ||
444                                 nh->nlmsg_type == RTM_DELROUTE) &&
445                                 rtm->rtm_dst_len == 0)
446                         raise(SIGUSR1); // Inform about a change in default route
447
448                 struct ndmsg *ndm = NLMSG_DATA(nh);
449                 struct ifaddrmsg *ifa = NLMSG_DATA(nh);
450                 if (nh->nlmsg_type != RTM_NEWNEIGH
451                                 && nh->nlmsg_type != RTM_DELNEIGH
452                                 && nh->nlmsg_type != RTM_NEWADDR
453                                 && nh->nlmsg_type != RTM_DELADDR)
454                         continue; // Unrelated message type
455                 bool is_addr = (nh->nlmsg_type == RTM_NEWADDR
456                                 || nh->nlmsg_type == RTM_DELADDR);
457
458                 // Family and ifindex are on the same offset for NEIGH and ADDR
459                 if (NLMSG_PAYLOAD(nh, 0) < sizeof(*ndm)
460                                 || ndm->ndm_family != AF_INET6)
461                         continue; //
462
463                 // Lookup interface
464                 struct interface *iface;
465                 if (!(iface = odhcpd_get_interface_by_index(ndm->ndm_ifindex)))
466                         continue;
467
468                 // Data to retrieve
469                 size_t rta_offset = (is_addr) ? sizeof(*ifa) : sizeof(*ndm);
470                 uint16_t atype = (is_addr) ? IFA_ADDRESS : NDA_DST;
471                 ssize_t alen = NLMSG_PAYLOAD(nh, rta_offset);
472                 struct in6_addr *addr = NULL;
473
474                 for (struct rtattr *rta = (void*)(((uint8_t*)ndm) + rta_offset);
475                                 RTA_OK(rta, alen); rta = RTA_NEXT(rta, alen))
476                         if (rta->rta_type == atype &&
477                                         RTA_PAYLOAD(rta) >= sizeof(*addr))
478                                 addr = RTA_DATA(rta);
479
480                 // Address not specified or unrelated
481                 if (!addr || IN6_IS_ADDR_LINKLOCAL(addr) ||
482                                 IN6_IS_ADDR_MULTICAST(addr))
483                         continue;
484
485                 // Check for states
486                 bool add;
487                 if (is_addr)
488                         add = (nh->nlmsg_type == RTM_NEWADDR);
489                 else
490                         add = (nh->nlmsg_type == RTM_NEWNEIGH && (ndm->ndm_state &
491                                 (NUD_REACHABLE | NUD_STALE | NUD_DELAY | NUD_PROBE
492                                                 | NUD_PERMANENT | NUD_NOARP)));
493
494                 if (iface->ndp == RELAYD_RELAY)
495                         modify_neighbor(addr, iface, add);
496
497                 if (is_addr && iface->ra == RELAYD_SERVER)
498                         raise(SIGUSR1); // Inform about a change in addresses
499
500                 if (is_addr && iface->dhcpv6 == RELAYD_SERVER)
501                         iface->ia_reconf = true;
502
503                 if (iface->ndp == RELAYD_RELAY && is_addr && iface->master) {
504                         // Replay address changes on all slave interfaces
505                         nh->nlmsg_flags = NLM_F_REQUEST;
506
507                         if (nh->nlmsg_type == RTM_NEWADDR)
508                                 nh->nlmsg_flags |= NLM_F_CREATE | NLM_F_REPLACE;
509
510                         struct interface *c;
511                         list_for_each_entry(c, &interfaces, head) {
512                                 if (c->ndp == RELAYD_RELAY && !c->master) {
513                                         ifa->ifa_index = c->ifindex;
514                                         send(rtnl_event.uloop.fd, nh, nh->nlmsg_len, MSG_DONTWAIT);
515                                 }
516                         }
517                 }
518
519                 /* TODO: See if this is required for optimal operation
520                 // Keep neighbor entries alive so we don't loose routes
521                  */
522                 if (add && (ndm->ndm_state & NUD_STALE))
523                         ping6(addr, iface);
524         }
525 }