1 /* vi: set sw=4 ts=4: */
3 * RFC3927 ZeroConf IPv4 Link-Local addressing
4 * (see <http://www.zeroconf.org/>)
6 * Copyright (C) 2003 by Arthur van Hoff (avh@strangeberry.com)
7 * Copyright (C) 2004 by David Brownell
9 * Licensed under the GPL v2 or later, see the file LICENSE in this tarball.
13 * ZCIP just manages the 169.254.*.* addresses. That network is not
14 * routed at the IP level, though various proxies or bridges can
15 * certainly be used. Its naming is built over multicast DNS.
21 // - more real-world usage/testing, especially daemon mode
22 // - kernel packet filters to reduce scheduling noise
23 // - avoid silent script failures, especially under load...
24 // - link status monitoring (restart on link-up; stop on link-down)
26 #include <netinet/ether.h>
27 #include <net/ethernet.h>
29 #include <net/if_arp.h>
30 #include <linux/if_packet.h>
31 #include <linux/sockios.h>
36 /* We don't need more than 32 bits of the counter */
37 #define MONOTONIC_US() ((unsigned)monotonic_us())
40 struct ether_header eth;
46 LINKLOCAL_ADDR = 0xa9fe0000,
48 /* protocol timeout parameters, specified in seconds */
54 RATE_LIMIT_INTERVAL = 60,
57 ANNOUNCE_INTERVAL = 2,
61 /* States during the configuration process. */
70 #define VDBG(...) do { } while (0)
79 struct sockaddr saddr;
81 #define G (*(struct globals*)&bb_common_bufsiz1)
82 #define intf (G.intf )
83 #define saddr (G.saddr)
87 * Pick a random link local IP address on 169.254/16, except that
88 * the first and last 256 addresses are reserved.
90 static void pick(struct in_addr *ip)
95 tmp = rand() & IN_CLASSB_HOST;
96 } while (tmp > (IN_CLASSB_HOST - 0x0200));
97 ip->s_addr = htonl((LINKLOCAL_ADDR + 0x0100) + tmp);
101 * Broadcast an ARP packet.
103 static void arp(int op,
104 const struct ether_addr *source_eth, struct in_addr source_ip,
105 const struct ether_addr *target_eth, struct in_addr target_ip)
108 memset(&p, 0, sizeof(p));
111 p.eth.ether_type = htons(ETHERTYPE_ARP);
112 memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
113 memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
116 p.arp.arp_hrd = htons(ARPHRD_ETHER);
117 p.arp.arp_pro = htons(ETHERTYPE_IP);
118 p.arp.arp_hln = ETH_ALEN;
120 p.arp.arp_op = htons(op);
121 memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
122 memcpy(&p.arp.arp_spa, &source_ip, sizeof(p.arp.arp_spa));
123 memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
124 memcpy(&p.arp.arp_tpa, &target_ip, sizeof(p.arp.arp_tpa));
127 xsendto(sock_fd, &p, sizeof(p), &saddr, sizeof(saddr));
129 // Currently all callers ignore errors, that's why returns are
135 * Run a script. argv[2] is already NULL.
137 static int run(char *argv[3], struct in_addr *ip)
140 char *addr = addr; /* for gcc */
141 const char *fmt = "%s %s %s" + 3;
143 VDBG("%s run %s %s\n", intf, argv[0], argv[1]);
146 addr = inet_ntoa(*ip);
147 setenv("ip", addr, 1);
150 bb_info_msg(fmt, argv[1], intf, addr);
152 status = wait4pid(spawn(argv));
154 bb_perror_msg("%s %s %s" + 3, argv[1], intf);
158 bb_error_msg("script %s %s failed, exitcode=%d", argv[0], argv[1], status);
163 * Return milliseconds of random delay, up to "secs" seconds.
165 static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
167 return rand() % (secs * 1000);
173 int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
174 int zcip_main(int argc, char **argv)
177 struct ether_addr eth_addr;
181 // ugly trick, but I want these zeroed in one go
183 const struct in_addr null_ip;
184 const struct ether_addr null_addr;
188 int timeout_ms; /* must be signed */
195 #define null_ip (L.null_ip )
196 #define null_addr (L.null_addr )
199 #define script_av (L.script_av )
200 #define timeout_ms (L.timeout_ms)
201 #define conflicts (L.conflicts )
202 #define nprobes (L.nprobes )
203 #define nclaims (L.nclaims )
204 #define ready (L.ready )
205 #define verbose (L.verbose )
207 memset(&L, 0, sizeof(L));
209 #define FOREGROUND (opts & 1)
210 #define QUIT (opts & 2)
211 // parse commandline: prog [options] ifname script
212 // exactly 2 args; -v accumulates and implies -f
213 opt_complementary = "=2:vv:vf";
214 opts = getopt32(argv, "fqr:v", &r_opt, &verbose);
216 // on NOMMU reexec early (or else we will rerun things twice)
218 bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
220 // open an ARP socket
221 // (need to do it before openlog to prevent openlog from taking
222 // fd 3 (sock_fd==3))
223 xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
225 // do it before all bb_xx_msg calls
226 openlog(applet_name, 0, LOG_DAEMON);
227 logmode |= LOGMODE_SYSLOG;
229 if (opts & 4) { // -r n.n.n.n
230 if (inet_aton(r_opt, &ip) == 0
231 || (ntohl(ip.s_addr) & IN_CLASSB_NET) != LINKLOCAL_ADDR
233 bb_error_msg_and_die("invalid link address");
240 script_av[0] = argv[1];
241 setenv("interface", intf, 1);
243 // initialize the interface (modprobe, ifup, etc)
244 script_av[1] = (char*)"init";
245 if (run(script_av, NULL))
249 // saddr is: { u16 sa_family; u8 sa_data[14]; }
250 //memset(&saddr, 0, sizeof(saddr));
251 //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
252 safe_strncpy(saddr.sa_data, intf, sizeof(saddr.sa_data));
254 // bind to the interface's ARP socket
255 xbind(sock_fd, &saddr, sizeof(saddr));
257 // get the interface's ethernet address
258 //memset(&ifr, 0, sizeof(ifr));
259 strncpy(ifr.ifr_name, intf, sizeof(ifr.ifr_name));
260 xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
261 memcpy(ð_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
263 // start with some stable ip address, either a function of
264 // the hardware address or else the last address we used.
265 // we are taking low-order four bytes, as top-order ones
266 // aren't random enough.
267 // NOTE: the sequence of addresses we try changes only
268 // depending on when we detect conflicts.
271 memcpy(&t, (char*)ð_addr + 2, 4);
277 // FIXME cases to handle:
278 // - zcip already running!
279 // - link already has local address... just defend/update
281 // daemonize now; don't delay system startup
284 bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
286 bb_info_msg("start, interface %s", intf);
289 // run the dynamic address negotiation protocol,
290 // restarting after address conflicts:
291 // - start with some address we want to try
292 // - short random delay
293 // - arp probes to see if another host uses it
294 // - arp announcements that we're claiming it
296 // - defend it, within limits
298 struct pollfd fds[1];
299 unsigned deadline_us;
301 int source_ip_conflict;
302 int target_ip_conflict;
305 fds[0].events = POLLIN;
308 // poll, being ready to adjust current timeout
310 timeout_ms = random_delay_ms(PROBE_WAIT);
311 // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
312 // make the kernel filter out all packets except
313 // ones we'd care about.
315 // set deadline_us to the point in time when we timeout
316 deadline_us = MONOTONIC_US() + timeout_ms * 1000;
318 VDBG("...wait %d %s nprobes=%u, nclaims=%u\n",
319 timeout_ms, intf, nprobes, nclaims);
321 switch (safe_poll(fds, 1, timeout_ms)) {
324 //bb_perror_msg("poll"); - done in safe_poll
329 VDBG("state = %d\n", state);
332 // timeouts in the PROBE state mean no conflicting ARP packets
333 // have been received, so we can progress through the states
334 if (nprobes < PROBE_NUM) {
336 VDBG("probe/%u %s@%s\n",
337 nprobes, intf, inet_ntoa(ip));
341 timeout_ms = PROBE_MIN * 1000;
342 timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
345 // Switch to announce state.
348 VDBG("announce/%u %s@%s\n",
349 nclaims, intf, inet_ntoa(ip));
353 timeout_ms = ANNOUNCE_INTERVAL * 1000;
356 case RATE_LIMIT_PROBE:
357 // timeouts in the RATE_LIMIT_PROBE state mean no conflicting ARP packets
358 // have been received, so we can move immediately to the announce state
361 VDBG("announce/%u %s@%s\n",
362 nclaims, intf, inet_ntoa(ip));
366 timeout_ms = ANNOUNCE_INTERVAL * 1000;
369 // timeouts in the ANNOUNCE state mean no conflicting ARP packets
370 // have been received, so we can progress through the states
371 if (nclaims < ANNOUNCE_NUM) {
373 VDBG("announce/%u %s@%s\n",
374 nclaims, intf, inet_ntoa(ip));
378 timeout_ms = ANNOUNCE_INTERVAL * 1000;
381 // Switch to monitor state.
383 // link is ok to use earlier
384 // FIXME update filters
385 script_av[1] = (char*)"config";
389 timeout_ms = -1; // Never timeout in the monitor state.
391 // NOTE: all other exit paths
392 // should deconfig ...
398 // We won! No ARP replies, so just go back to monitor.
404 // Invalid, should never happen. Restart the whole protocol.
412 break; // case 0 (timeout)
414 // packets arriving, or link went down
416 // We need to adjust the timeout in case we didn't receive
417 // a conflicting packet.
418 if (timeout_ms > 0) {
419 unsigned diff = deadline_us - MONOTONIC_US();
420 if ((int)(diff) < 0) {
421 // Current time is greater than the expected timeout time.
422 // Should never happen.
423 VDBG("missed an expected timeout\n");
426 VDBG("adjusting timeout\n");
427 timeout_ms = (diff / 1000) | 1; /* never 0 */
431 if ((fds[0].revents & POLLIN) == 0) {
432 if (fds[0].revents & POLLERR) {
433 // FIXME: links routinely go down;
434 // this shouldn't necessarily exit.
435 bb_error_msg("iface %s is down", intf);
437 script_av[1] = (char*)"deconfig";
446 if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
447 bb_perror_msg_and_die(bb_msg_read_error);
449 if (p.eth.ether_type != htons(ETHERTYPE_ARP))
453 struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
454 struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
455 struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
456 struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
457 VDBG("%s recv arp type=%d, op=%d,\n",
458 intf, ntohs(p.eth.ether_type),
459 ntohs(p.arp.arp_op));
460 VDBG("\tsource=%s %s\n",
463 VDBG("\ttarget=%s %s\n",
468 if (p.arp.arp_op != htons(ARPOP_REQUEST)
469 && p.arp.arp_op != htons(ARPOP_REPLY))
472 source_ip_conflict = 0;
473 target_ip_conflict = 0;
475 if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0
476 && memcmp(&p.arp.arp_sha, ð_addr, ETH_ALEN) != 0
478 source_ip_conflict = 1;
480 if (p.arp.arp_op == htons(ARPOP_REQUEST)
481 && memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0
482 && memcmp(&p.arp.arp_tha, ð_addr, ETH_ALEN) != 0
484 target_ip_conflict = 1;
487 VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n",
488 state, source_ip_conflict, target_ip_conflict);
492 // When probing or announcing, check for source IP conflicts
493 // and other hosts doing ARP probes (target IP conflicts).
494 if (source_ip_conflict || target_ip_conflict) {
496 if (conflicts >= MAX_CONFLICTS) {
497 VDBG("%s ratelimit\n", intf);
498 timeout_ms = RATE_LIMIT_INTERVAL * 1000;
499 state = RATE_LIMIT_PROBE;
502 // restart the whole protocol
510 // If a conflict, we try to defend with a single ARP probe.
511 if (source_ip_conflict) {
512 VDBG("monitor conflict -- defending\n");
514 timeout_ms = DEFEND_INTERVAL * 1000;
521 // Well, we tried. Start over (on conflict).
522 if (source_ip_conflict) {
524 VDBG("defend conflict -- starting over\n");
526 script_av[1] = (char*)"deconfig";
529 // restart the whole protocol
537 // Invalid, should never happen. Restart the whole protocol.
538 VDBG("invalid state -- starting over\n");
546 break; // case 1 (packets arriving)