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 GPLv2 or later, see file LICENSE in this source tree.
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 //usage:#define zcip_trivial_usage
27 //usage: "[OPTIONS] IFACE SCRIPT"
28 //usage:#define zcip_full_usage "\n\n"
29 //usage: "Manage a ZeroConf IPv4 link-local address\n"
30 //usage: "\n -f Run in foreground"
31 //usage: "\n -q Quit after obtaining address"
32 //usage: "\n -r 169.254.x.x Request this address first"
33 //usage: "\n -l x.x.0.0 Use this range instead of 169.254"
34 //usage: "\n -v Verbose"
36 //usage: "\n$LOGGING=none Suppress logging"
37 //usage: "\n$LOGGING=syslog Log to syslog"
39 //usage: "\nWith no -q, runs continuously monitoring for ARP conflicts,"
40 //usage: "\nexits only on I/O errors (link down etc)"
43 #include <netinet/ether.h>
45 #include <net/if_arp.h>
46 #include <linux/sockios.h>
50 /* We don't need more than 32 bits of the counter */
51 #define MONOTONIC_US() ((unsigned)monotonic_us())
54 struct ether_header eth;
59 /* 0-1 seconds before sending 1st probe */
61 /* 1-2 seconds between probes */
64 PROBE_NUM = 3, /* total probes to send */
65 ANNOUNCE_INTERVAL = 2, /* 2 seconds between announces */
66 ANNOUNCE_NUM = 3, /* announces to send */
67 /* if probe/announce sees a conflict, multiply RANDOM(NUM_CONFLICT) by... */
68 CONFLICT_MULTIPLIER = 2,
69 /* if we monitor and see a conflict, how long is defend state? */
73 /* States during the configuration process. */
81 #define VDBG(...) do { } while (0)
89 struct sockaddr iface_sockaddr;
90 struct ether_addr our_ethaddr;
93 #define G (*(struct globals*)&bb_common_bufsiz1)
94 #define INIT_G() do { } while (0)
98 * Pick a random link local IP address on 169.254/16, except that
99 * the first and last 256 addresses are reserved.
101 static uint32_t pick_nip(void)
106 tmp = rand() & IN_CLASSB_HOST;
107 } while (tmp > (IN_CLASSB_HOST - 0x0200));
108 return htonl((G.localnet_ip + 0x0100) + tmp);
111 static const char *nip_to_a(uint32_t nip)
115 return inet_ntoa(in);
119 * Broadcast an ARP packet.
121 static void send_arp_request(
122 /* int op, - always ARPOP_REQUEST */
123 /* const struct ether_addr *source_eth, - always &G.our_ethaddr */
125 const struct ether_addr *target_eth, uint32_t target_nip)
127 enum { op = ARPOP_REQUEST };
128 #define source_eth (&G.our_ethaddr)
131 memset(&p, 0, sizeof(p));
134 p.eth.ether_type = htons(ETHERTYPE_ARP);
135 memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
136 memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
139 p.arp.arp_hrd = htons(ARPHRD_ETHER);
140 p.arp.arp_pro = htons(ETHERTYPE_IP);
141 p.arp.arp_hln = ETH_ALEN;
143 p.arp.arp_op = htons(op);
144 memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
145 memcpy(&p.arp.arp_spa, &source_nip, 4);
146 memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
147 memcpy(&p.arp.arp_tpa, &target_nip, 4);
150 // Even though sock_fd is already bound to G.iface_sockaddr, just send()
151 // won't work, because "socket is not connected"
152 // (and connect() won't fix that, "operation not supported").
153 // Thus we sendto() to G.iface_sockaddr. I wonder which sockaddr
154 // (from bind() or from sendto()?) kernel actually uses
155 // to determine iface to emit the packet from...
156 xsendto(sock_fd, &p, sizeof(p), &G.iface_sockaddr, sizeof(G.iface_sockaddr));
162 * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL
164 static int run(char *argv[3], const char *param, uint32_t nip)
167 const char *addr = addr; /* for gcc */
168 const char *fmt = "%s %s %s" + 3;
170 argv[2] = (char*)param;
172 VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]);
175 addr = nip_to_a(nip);
179 bb_info_msg(fmt, argv[2], argv[0], addr);
181 status = spawn_and_wait(argv + 1);
183 bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]);
187 bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff);
192 * Return milliseconds of random delay, up to "secs" seconds.
194 static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
196 return (unsigned)rand() % (secs * 1000);
202 int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
203 int zcip_main(int argc UNUSED_PARAM, char **argv)
206 const char *l_opt = "169.254.0.0";
211 // Ugly trick, but I want these zeroed in one go
213 const struct ether_addr null_ethaddr;
217 int timeout_ms; // must be signed
220 #define null_ethaddr (L.null_ethaddr)
222 #define chosen_nip (L.chosen_nip )
223 #define conflicts (L.conflicts )
224 #define timeout_ms (L.timeout_ms )
225 #define verbose (L.verbose )
227 memset(&L, 0, sizeof(L));
230 #define FOREGROUND (opts & 1)
231 #define QUIT (opts & 2)
232 // Parse commandline: prog [options] ifname script
233 // exactly 2 args; -v accumulates and implies -f
234 opt_complementary = "=2:vv:vf";
235 opts = getopt32(argv, "fqr:l:v", &r_opt, &l_opt, &verbose);
237 // on NOMMU reexec early (or else we will rerun things twice)
239 bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
241 // Open an ARP socket
242 // (need to do it before openlog to prevent openlog from taking
243 // fd 3 (sock_fd==3))
244 xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
246 // do it before all bb_xx_msg calls
247 openlog(applet_name, 0, LOG_DAEMON);
248 logmode |= LOGMODE_SYSLOG;
250 bb_logenv_override();
254 if (inet_aton(l_opt, &net) == 0
255 || (net.s_addr & htonl(IN_CLASSB_NET)) != net.s_addr
257 bb_error_msg_and_die("invalid network address");
259 G.localnet_ip = ntohl(net.s_addr);
261 if (opts & 4) { // -r n.n.n.n
263 if (inet_aton(r_opt, &ip) == 0
264 || (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
266 bb_error_msg_and_die("invalid link address");
268 chosen_nip = ip.s_addr;
272 /* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
273 /* We need to make space for script argument: */
276 /* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
277 #define argv_intf (argv[0])
279 xsetenv("interface", argv_intf);
281 // Initialize the interface (modprobe, ifup, etc)
282 if (run(argv, "init", 0))
285 // Initialize G.iface_sockaddr
286 // G.iface_sockaddr is: { u16 sa_family; u8 sa_data[14]; }
287 //memset(&G.iface_sockaddr, 0, sizeof(G.iface_sockaddr));
288 //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
289 safe_strncpy(G.iface_sockaddr.sa_data, argv_intf, sizeof(G.iface_sockaddr.sa_data));
291 // Bind to the interface's ARP socket
292 xbind(sock_fd, &G.iface_sockaddr, sizeof(G.iface_sockaddr));
294 // Get the interface's ethernet address
295 //memset(&ifr, 0, sizeof(ifr));
296 strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
297 xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
298 memcpy(&G.our_ethaddr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
300 // Start with some stable ip address, either a function of
301 // the hardware address or else the last address we used.
302 // we are taking low-order four bytes, as top-order ones
303 // aren't random enough.
304 // NOTE: the sequence of addresses we try changes only
305 // depending on when we detect conflicts.
308 move_from_unaligned32(t, ((char *)&G.our_ethaddr + 2));
311 // FIXME cases to handle:
312 // - zcip already running!
313 // - link already has local address... just defend/update
315 // Daemonize now; don't delay system startup
318 bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
320 bb_info_msg("start, interface %s", argv_intf);
323 // Run the dynamic address negotiation protocol,
324 // restarting after address conflicts:
325 // - start with some address we want to try
326 // - short random delay
327 // - arp probes to see if another host uses it
328 // 00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff arp who-has 169.254.194.171 tell 0.0.0.0
329 // - arp announcements that we're claiming it
330 // 00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff arp who-has 169.254.194.171 (00:04:e2:64:23:c2) tell 169.254.194.171
332 // - defend it, within limits
334 // - address is successfully obtained and -q was given:
335 // run "<script> config", then exit with exitcode 0
336 // - poll error (when does this happen?)
337 // - read error (when does this happen?)
338 // - sendto error (in send_arp_request()) (when does this happen?)
339 // - revents & POLLERR (link down). run "<script> deconfig" first
340 if (chosen_nip == 0) {
342 chosen_nip = pick_nip();
347 struct pollfd fds[1];
348 unsigned deadline_us;
354 fds[0].events = POLLIN;
357 // Poll, being ready to adjust current timeout
359 timeout_ms = random_delay_ms(PROBE_WAIT);
360 // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
361 // make the kernel filter out all packets except
362 // ones we'd care about.
364 // Set deadline_us to the point in time when we timeout
365 deadline_us = MONOTONIC_US() + timeout_ms * 1000;
367 VDBG("...wait %d %s nsent=%u\n",
368 timeout_ms, argv_intf, nsent);
370 n = safe_poll(fds, 1, timeout_ms);
372 //bb_perror_msg("poll"); - done in safe_poll
375 if (n == 0) { // timed out?
376 VDBG("state:%d\n", state);
379 // No conflicting ARP packets were seen:
380 // we can progress through the states
381 if (nsent < PROBE_NUM) {
383 VDBG("probe/%u %s@%s\n",
384 nsent, argv_intf, nip_to_a(chosen_nip));
385 timeout_ms = PROBE_MIN * 1000;
386 timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
387 send_arp_request(0, &null_ethaddr, chosen_nip);
390 // Switch to announce state
395 // No conflicting ARP packets were seen:
396 // we can progress through the states
397 if (nsent < ANNOUNCE_NUM) {
400 VDBG("announce/%u %s@%s\n",
401 nsent, argv_intf, nip_to_a(chosen_nip));
402 timeout_ms = ANNOUNCE_INTERVAL * 1000;
403 send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
406 // Switch to monitor state
407 // FIXME update filters
408 run(argv, "config", chosen_nip);
409 // NOTE: all other exit paths should deconfig...
412 // fall through: switch to MONITOR
415 // case MONITOR: (shouldn't happen, MONITOR timeout is infinite)
416 // Defend period ended with no ARP replies - we won
417 timeout_ms = -1; // never timeout in monitor state
423 // Packet arrived, or link went down.
424 // We need to adjust the timeout in case we didn't receive
425 // a conflicting packet.
426 if (timeout_ms > 0) {
427 unsigned diff = deadline_us - MONOTONIC_US();
428 if ((int)(diff) < 0) {
429 // Current time is greater than the expected timeout time.
432 VDBG("adjusting timeout\n");
433 timeout_ms = (diff / 1000) | 1; // never 0
436 if ((fds[0].revents & POLLIN) == 0) {
437 if (fds[0].revents & POLLERR) {
438 // FIXME: links routinely go down;
439 // this shouldn't necessarily exit.
440 bb_error_msg("iface %s is down", argv_intf);
441 if (state >= MONITOR) {
442 // Only if we are in MONITOR or DEFEND
443 run(argv, "deconfig", chosen_nip);
451 if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
452 bb_perror_msg_and_die(bb_msg_read_error);
455 if (p.eth.ether_type != htons(ETHERTYPE_ARP))
457 if (p.arp.arp_op != htons(ARPOP_REQUEST)
458 && p.arp.arp_op != htons(ARPOP_REPLY)
464 struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
465 struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
466 struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
467 struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
468 VDBG("source=%s %s\n", ether_ntoa(sha), inet_ntoa(*spa));
469 VDBG("target=%s %s\n", ether_ntoa(tha), inet_ntoa(*tpa));
473 if (memcmp(&p.arp.arp_sha, &G.our_ethaddr, ETH_ALEN) != 0) {
474 if (memcmp(p.arp.arp_spa, &chosen_nip, 4) == 0) {
475 // A probe or reply with source_ip == chosen ip
478 if (p.arp.arp_op == htons(ARPOP_REQUEST)
479 && memcmp(p.arp.arp_spa, &const_int_0, 4) == 0
480 && memcmp(p.arp.arp_tpa, &chosen_nip, 4) == 0
482 // A probe with source_ip == 0.0.0.0, target_ip == chosen ip:
483 // another host trying to claim this ip!
487 VDBG("state:%d ip_conflict:%d\n", state, ip_conflict);
491 // Either src or target IP conflict exists
492 if (state <= ANNOUNCE) {
495 timeout_ms = PROBE_MIN * 1000
496 + CONFLICT_MULTIPLIER * random_delay_ms(conflicts);
497 goto new_nip_and_PROBE;
500 // MONITOR or DEFEND: only src IP conflict is a problem
501 if (ip_conflict & 1) {
502 if (state == MONITOR) {
503 // Src IP conflict, defend with a single ARP probe
504 VDBG("monitor conflict - defending\n");
505 timeout_ms = DEFEND_INTERVAL * 1000;
507 send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
511 // Another src IP conflict, start over
512 VDBG("defend conflict - starting over\n");
513 run(argv, "deconfig", chosen_nip);
516 goto new_nip_and_PROBE;
518 // Note: if we only have a target IP conflict here (ip_conflict & 2),
519 // IOW: if we just saw this sort of ARP packet:
520 // aa:bb:cc:dd:ee:ff > xx:xx:xx:xx:xx:xx arp who-has <chosen_nip> tell 0.0.0.0
521 // we expect _kernel_ to respond to that, because <chosen_nip>
522 // is (expected to be) configured on this iface.