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;
60 LINKLOCAL_ADDR = 0xa9fe0000,
62 /* 0-1 seconds before sending 1st probe */
64 /* 1-2 seconds between probes */
67 PROBE_NUM = 3, /* total probes to send */
68 ANNOUNCE_INTERVAL = 2, /* 2 seconds between announces */
69 ANNOUNCE_NUM = 3, /* announces to send */
70 /* if probe/announce sees a conflict, multiply RANDOM(NUM_CONFLICT) by... */
71 CONFLICT_MULTIPLIER = 2,
72 /* if we monitor and see a conflict, how long is defend state? */
76 /* States during the configuration process. */
84 #define VDBG(...) do { } while (0)
92 struct sockaddr iface_sockaddr;
93 struct ether_addr eth_addr;
96 #define G (*(struct globals*)&bb_common_bufsiz1)
97 #define INIT_G() do { } while (0)
101 * Pick a random link local IP address on 169.254/16, except that
102 * the first and last 256 addresses are reserved.
104 static uint32_t pick_nip(void)
109 tmp = rand() & IN_CLASSB_HOST;
110 } while (tmp > (IN_CLASSB_HOST - 0x0200));
111 return htonl((G.localnet_ip + 0x0100) + tmp);
114 static const char *nip_to_a(uint32_t nip)
118 return inet_ntoa(in);
122 * Broadcast an ARP packet.
125 /* int op, - always ARPOP_REQUEST */
126 /* const struct ether_addr *source_eth, - always &G.eth_addr */
128 const struct ether_addr *target_eth, uint32_t target_nip)
130 enum { op = ARPOP_REQUEST };
131 #define source_eth (&G.eth_addr)
134 memset(&p, 0, sizeof(p));
137 p.eth.ether_type = htons(ETHERTYPE_ARP);
138 memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
139 memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
142 p.arp.arp_hrd = htons(ARPHRD_ETHER);
143 p.arp.arp_pro = htons(ETHERTYPE_IP);
144 p.arp.arp_hln = ETH_ALEN;
146 p.arp.arp_op = htons(op);
147 memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
148 memcpy(&p.arp.arp_spa, &source_nip, 4);
149 memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
150 memcpy(&p.arp.arp_tpa, &target_nip, 4);
153 // Even though sock_fd is already bound to G.iface_sockaddr, just send()
154 // won't work, because "socket is not connected"
155 // (and connect() won't fix that, "operation not supported").
156 // Thus we sendto() to G.iface_sockaddr. I wonder which sockaddr
157 // (from bind() or from sendto()?) kernel actually uses
158 // to determine iface to emit the packet from...
159 xsendto(sock_fd, &p, sizeof(p), &G.iface_sockaddr, sizeof(G.iface_sockaddr));
165 * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL
167 static int run(char *argv[3], const char *param, uint32_t nip)
170 const char *addr = addr; /* for gcc */
171 const char *fmt = "%s %s %s" + 3;
173 argv[2] = (char*)param;
175 VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]);
178 addr = nip_to_a(nip);
182 bb_info_msg(fmt, argv[2], argv[0], addr);
184 status = spawn_and_wait(argv + 1);
186 bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]);
190 bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff);
195 * Return milliseconds of random delay, up to "secs" seconds.
197 static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
199 return rand() % (secs * 1000);
205 int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
206 int zcip_main(int argc UNUSED_PARAM, char **argv)
210 const char *l_opt = "169.254.0.0";
213 // ugly trick, but I want these zeroed in one go
215 const struct ether_addr null_addr;
218 int timeout_ms; // must be signed
223 #define null_addr (L.null_addr )
225 #define chosen_nip (L.chosen_nip)
226 #define timeout_ms (L.timeout_ms)
227 #define conflicts (L.conflicts )
228 #define nsent (L.nsent )
229 #define verbose (L.verbose )
231 memset(&L, 0, sizeof(L));
234 #define FOREGROUND (opts & 1)
235 #define QUIT (opts & 2)
236 // parse commandline: prog [options] ifname script
237 // exactly 2 args; -v accumulates and implies -f
238 opt_complementary = "=2:vv:vf";
239 opts = getopt32(argv, "fqr:l:v", &r_opt, &l_opt, &verbose);
241 // on NOMMU reexec early (or else we will rerun things twice)
243 bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
245 // open an ARP socket
246 // (need to do it before openlog to prevent openlog from taking
247 // fd 3 (sock_fd==3))
248 xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
250 // do it before all bb_xx_msg calls
251 openlog(applet_name, 0, LOG_DAEMON);
252 logmode |= LOGMODE_SYSLOG;
254 bb_logenv_override();
258 if (inet_aton(l_opt, &net) == 0
259 || (net.s_addr & htonl(IN_CLASSB_NET)) != net.s_addr
261 bb_error_msg_and_die("invalid network address");
263 G.localnet_ip = ntohl(net.s_addr);
265 if (opts & 4) { // -r n.n.n.n
267 if (inet_aton(r_opt, &ip) == 0
268 || (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
270 bb_error_msg_and_die("invalid link address");
272 chosen_nip = ip.s_addr;
276 /* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
277 /* We need to make space for script argument: */
280 /* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
281 #define argv_intf (argv[0])
283 xsetenv("interface", argv_intf);
285 // initialize the interface (modprobe, ifup, etc)
286 if (run(argv, "init", 0))
289 // initialize G.iface_sockaddr
290 // G.iface_sockaddr is: { u16 sa_family; u8 sa_data[14]; }
291 //memset(&G.iface_sockaddr, 0, sizeof(G.iface_sockaddr));
292 //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
293 safe_strncpy(G.iface_sockaddr.sa_data, argv_intf, sizeof(G.iface_sockaddr.sa_data));
295 // bind to the interface's ARP socket
296 xbind(sock_fd, &G.iface_sockaddr, sizeof(G.iface_sockaddr));
298 // get the interface's ethernet address
299 //memset(&ifr, 0, sizeof(ifr));
300 strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
301 xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
302 memcpy(&G.eth_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
304 // start with some stable ip address, either a function of
305 // the hardware address or else the last address we used.
306 // we are taking low-order four bytes, as top-order ones
307 // aren't random enough.
308 // NOTE: the sequence of addresses we try changes only
309 // depending on when we detect conflicts.
312 move_from_unaligned32(t, ((char *)&G.eth_addr + 2));
316 chosen_nip = pick_nip();
318 // FIXME cases to handle:
319 // - zcip already running!
320 // - link already has local address... just defend/update
322 // daemonize now; don't delay system startup
325 bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
327 bb_info_msg("start, interface %s", argv_intf);
330 // run the dynamic address negotiation protocol,
331 // restarting after address conflicts:
332 // - start with some address we want to try
333 // - short random delay
334 // - arp probes to see if another host uses it
335 // 00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff, ARP (0x0806): arp who-has 169.254.194.171 tell 0.0.0.0
336 // - arp announcements that we're claiming it
337 // 00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff, ARP (0x0806): arp who-has 169.254.194.171 (00:04:e2:64:23:c2) tell 169.254.194.171
339 // - defend it, within limits
341 // - address is successfully obtained and -q was given:
342 // run "<script> config", then exit with exitcode 0
343 // - poll error (when does this happen?)
344 // - read error (when does this happen?)
345 // - sendto error (in arp()) (when does this happen?)
346 // - revents & POLLERR (link down). run "<script> deconfig" first
349 struct pollfd fds[1];
350 unsigned deadline_us;
355 fds[0].events = POLLIN;
358 // poll, being ready to adjust current timeout
360 timeout_ms = random_delay_ms(PROBE_WAIT);
361 // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
362 // make the kernel filter out all packets except
363 // ones we'd care about.
365 // set deadline_us to the point in time when we timeout
366 deadline_us = MONOTONIC_US() + timeout_ms * 1000;
368 VDBG("...wait %d %s nsent=%u\n",
369 timeout_ms, argv_intf, nsent);
371 switch (safe_poll(fds, 1, timeout_ms)) {
374 //bb_perror_msg("poll"); - done in safe_poll
379 VDBG("state = %d\n", state);
382 // timeouts in the PROBE state mean no conflicting ARP packets
383 // have been received, so we can progress through the states
384 if (nsent < PROBE_NUM) {
386 VDBG("probe/%u %s@%s\n",
387 nsent, argv_intf, nip_to_a(chosen_nip));
388 timeout_ms = PROBE_MIN * 1000;
389 timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
390 arp(/* ARPOP_REQUEST, */
391 /* &G.eth_addr, */ 0,
392 &null_addr, chosen_nip);
395 // Switch to announce state
400 // timeouts in the ANNOUNCE state mean no conflicting ARP packets
401 // have been received, so we can progress through the states
402 if (nsent < ANNOUNCE_NUM) {
405 VDBG("announce/%u %s@%s\n",
406 nsent, argv_intf, nip_to_a(chosen_nip));
407 timeout_ms = ANNOUNCE_INTERVAL * 1000;
408 arp(/* ARPOP_REQUEST, */
409 /* &G.eth_addr, */ chosen_nip,
410 &G.eth_addr, chosen_nip);
413 // Switch to monitor state
414 // FIXME update filters
415 run(argv, "config", chosen_nip);
416 // NOTE: all other exit paths should deconfig...
420 timeout_ms = -1; // Never timeout in the monitor state.
424 // Defend period ended with no ARP replies - we won
430 break; // case 0 (timeout)
432 // packets arriving, or link went down
434 // We need to adjust the timeout in case we didn't receive
435 // a conflicting packet.
436 if (timeout_ms > 0) {
437 unsigned diff = deadline_us - MONOTONIC_US();
438 if ((int)(diff) < 0) {
439 // Current time is greater than the expected timeout time.
442 VDBG("adjusting timeout\n");
443 timeout_ms = (diff / 1000) | 1; // never 0
446 if ((fds[0].revents & POLLIN) == 0) {
447 if (fds[0].revents & POLLERR) {
448 // FIXME: links routinely go down;
449 // this shouldn't necessarily exit.
450 bb_error_msg("iface %s is down", argv_intf);
451 if (state >= MONITOR) {
452 // only if we are in MONITOR or DEFEND
453 run(argv, "deconfig", chosen_nip);
461 if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
462 bb_perror_msg_and_die(bb_msg_read_error);
465 if (p.eth.ether_type != htons(ETHERTYPE_ARP))
467 if (p.arp.arp_op != htons(ARPOP_REQUEST)
468 && p.arp.arp_op != htons(ARPOP_REPLY)
474 struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
475 struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
476 struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
477 struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
478 VDBG("source=%s %s\n", ether_ntoa(sha), inet_ntoa(*spa));
479 VDBG("target=%s %s\n", ether_ntoa(tha), inet_ntoa(*tpa));
483 if (memcmp(&p.arp.arp_sha, &G.eth_addr, ETH_ALEN) != 0) {
484 if (memcmp(p.arp.arp_spa, &chosen_nip, 4) == 0) {
485 // A probe or reply with source_ip == chosen ip
488 if (p.arp.arp_op == htons(ARPOP_REQUEST)
489 && memcmp(p.arp.arp_spa, &const_int_0, 4) == 0
490 && memcmp(p.arp.arp_tpa, &chosen_nip, 4) == 0
492 // A probe with source_ip == 0.0.0.0, target_ip == chosen ip:
493 // another host trying to claim this ip!
497 VDBG("state:%d ip_conflict:%d\n", state, ip_conflict);
501 // Either src or target IP conflict exists
502 if (state <= ANNOUNCE) {
505 timeout_ms = PROBE_MIN * 1000
506 + CONFLICT_MULTIPLIER * random_delay_ms(conflicts);
507 chosen_nip = pick_nip();
512 // MONITOR or DEFEND: only src IP conflict is a problem
513 if (ip_conflict & 1) {
514 if (state == MONITOR) {
515 // Src IP conflict, defend with a single ARP probe
516 VDBG("monitor conflict - defending\n");
517 timeout_ms = DEFEND_INTERVAL * 1000;
519 arp(/* ARPOP_REQUEST, */
520 /* &G.eth_addr, */ chosen_nip,
521 &G.eth_addr, chosen_nip);
525 // Another src IP conflict, start over
526 VDBG("defend conflict - starting over\n");
527 run(argv, "deconfig", chosen_nip);
529 // restart the whole protocol
531 chosen_nip = pick_nip();
535 break; // case 1 (packet arrived)