1 /* vi: set sw=4 ts=4: */
2 /*
3  * RFC3927 ZeroConf IPv4 Link-Local addressing
4  * (see <http://www.zeroconf.org/>)
5  *
6  * Copyright (C) 2003 by Arthur van Hoff (avh@strangeberry.com)
7  * Copyright (C) 2004 by David Brownell
8  *
9  * Licensed under GPLv2 or later, see file LICENSE in this source tree.
10  */
11 /*
12  * ZCIP just manages the 169.254.*.* addresses.  That network is not
13  * routed at the IP level, though various proxies or bridges can
14  * certainly be used.  Its naming is built over multicast DNS.
15  */
16 //config:config ZCIP
17 //config:	bool "zcip (8.4 kb)"
18 //config:	default y
19 //config:	select FEATURE_SYSLOG
20 //config:	help
21 //config:	ZCIP provides ZeroConf IPv4 address selection, according to RFC 3927.
22 //config:	It's a daemon that allocates and defends a dynamically assigned
23 //config:	address on the 169.254/16 network, requiring no system administrator.
24 //config:
25 //config:	See http://www.zeroconf.org for further details, and "zcip.script"
26 //config:	in the busybox examples.
27 
28 //applet:IF_ZCIP(APPLET(zcip, BB_DIR_SBIN, BB_SUID_DROP))
29 
30 //kbuild:lib-$(CONFIG_ZCIP) += zcip.o
31 
32 //#define DEBUG
33 
34 // TODO:
35 // - more real-world usage/testing, especially daemon mode
36 // - kernel packet filters to reduce scheduling noise
37 // - avoid silent script failures, especially under load...
38 // - link status monitoring (restart on link-up; stop on link-down)
39 
40 //usage:#define zcip_trivial_usage
41 //usage:       "[OPTIONS] IFACE SCRIPT"
42 //usage:#define zcip_full_usage "\n\n"
43 //usage:       "Manage a ZeroConf IPv4 link-local address\n"
44 //usage:     "\n	-f		Run in foreground"
45 //usage:     "\n	-q		Quit after obtaining address"
46 //usage:     "\n	-r 169.254.x.x	Request this address first"
47 //usage:     "\n	-l x.x.0.0	Use this range instead of 169.254"
48 //usage:     "\n	-v		Verbose"
49 //usage:     "\n"
50 //usage:     "\n$LOGGING=none		Suppress logging"
51 //usage:     "\n$LOGGING=syslog 	Log to syslog"
52 //usage:     "\n"
53 //usage:     "\nWith no -q, runs continuously monitoring for ARP conflicts,"
54 //usage:     "\nexits only on I/O errors (link down etc)"
55 
56 #include "libbb.h"
57 #include "common_bufsiz.h"
58 #include <netinet/ether.h>
59 #include <net/if.h>
60 #include <net/if_arp.h>
61 #include <linux/sockios.h>
62 
63 #include <syslog.h>
64 
65 /* We don't need more than 32 bits of the counter */
66 #define MONOTONIC_US() ((unsigned)monotonic_us())
67 
68 struct arp_packet {
69 	struct ether_header eth;
70 	struct ether_arp arp;
71 } PACKED;
72 
73 enum {
74 	/* 0-1 seconds before sending 1st probe */
75 	PROBE_WAIT = 1,
76 	/* 1-2 seconds between probes */
77 	PROBE_MIN = 1,
78 	PROBE_MAX = 2,
79 	PROBE_NUM = 3,		/* total probes to send */
80 	ANNOUNCE_INTERVAL = 2,  /* 2 seconds between announces */
81 	ANNOUNCE_NUM = 3,	/* announces to send */
82 	/* if probe/announce sees a conflict, multiply RANDOM(NUM_CONFLICT) by... */
83 	CONFLICT_MULTIPLIER = 2,
84 	/* if we monitor and see a conflict, how long is defend state? */
85 	DEFEND_INTERVAL = 10,
86 };
87 
88 /* States during the configuration process. */
89 enum {
90 	PROBE = 0,
91 	ANNOUNCE,
92 	MONITOR,
93 	DEFEND
94 };
95 
96 #define VDBG(...) do { } while (0)
97 
98 
99 enum {
100 	sock_fd = 3
101 };
102 
103 struct globals {
104 	struct sockaddr iface_sockaddr;
105 	struct ether_addr our_ethaddr;
106 	uint32_t localnet_ip;
107 } FIX_ALIASING;
108 #define G (*(struct globals*)bb_common_bufsiz1)
109 #define INIT_G() do { setup_common_bufsiz(); } while (0)
110 
111 
112 /**
113  * Pick a random link local IP address on 169.254/16, except that
114  * the first and last 256 addresses are reserved.
115  */
pick_nip(void)116 static uint32_t pick_nip(void)
117 {
118 	unsigned tmp;
119 
120 	do {
121 		tmp = rand() & IN_CLASSB_HOST;
122 	} while (tmp > (IN_CLASSB_HOST - 0x0200));
123 	return htonl((G.localnet_ip + 0x0100) + tmp);
124 }
125 
nip_to_a(uint32_t nip)126 static const char *nip_to_a(uint32_t nip)
127 {
128 	struct in_addr in;
129 	in.s_addr = nip;
130 	return inet_ntoa(in);
131 }
132 
133 /**
134  * Broadcast an ARP packet.
135  */
send_arp_request(uint32_t source_nip,const struct ether_addr * target_eth,uint32_t target_nip)136 static void send_arp_request(
137 	/* int op, - always ARPOP_REQUEST */
138 	/* const struct ether_addr *source_eth, - always &G.our_ethaddr */
139 					uint32_t source_nip,
140 	const struct ether_addr *target_eth, uint32_t target_nip)
141 {
142 	enum { op = ARPOP_REQUEST };
143 #define source_eth (&G.our_ethaddr)
144 
145 	struct arp_packet p;
146 	memset(&p, 0, sizeof(p));
147 
148 	// ether header
149 	p.eth.ether_type = htons(ETHERTYPE_ARP);
150 	memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
151 	memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
152 
153 	// arp request
154 	p.arp.arp_hrd = htons(ARPHRD_ETHER);
155 	p.arp.arp_pro = htons(ETHERTYPE_IP);
156 	p.arp.arp_hln = ETH_ALEN;
157 	p.arp.arp_pln = 4;
158 	p.arp.arp_op = htons(op);
159 	memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
160 	memcpy(&p.arp.arp_spa, &source_nip, 4);
161 	memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
162 	memcpy(&p.arp.arp_tpa, &target_nip, 4);
163 
164 	// send it
165 	// Even though sock_fd is already bound to G.iface_sockaddr, just send()
166 	// won't work, because "socket is not connected"
167 	// (and connect() won't fix that, "operation not supported").
168 	// Thus we sendto() to G.iface_sockaddr. I wonder which sockaddr
169 	// (from bind() or from sendto()?) kernel actually uses
170 	// to determine iface to emit the packet from...
171 	xsendto(sock_fd, &p, sizeof(p), &G.iface_sockaddr, sizeof(G.iface_sockaddr));
172 #undef source_eth
173 }
174 
175 /**
176  * Run a script.
177  * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL
178  */
run(char * argv[3],const char * param,uint32_t nip)179 static int run(char *argv[3], const char *param, uint32_t nip)
180 {
181 	int status;
182 	const char *addr = addr; /* for gcc */
183 	const char *fmt = "%s %s %s" + 3;
184 	char *env_ip = env_ip;
185 
186 	argv[2] = (char*)param;
187 
188 	VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]);
189 
190 	if (nip != 0) {
191 		addr = nip_to_a(nip);
192 		/* Must not use setenv() repeatedly, it leaks memory. Use putenv() */
193 		env_ip = xasprintf("ip=%s", addr);
194 		putenv(env_ip);
195 		fmt -= 3;
196 	}
197 	bb_info_msg(fmt, argv[2], argv[0], addr);
198 	status = spawn_and_wait(argv + 1);
199 	if (nip != 0)
200 		bb_unsetenv_and_free(env_ip);
201 
202 	if (status < 0) {
203 		bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]);
204 		return -errno;
205 	}
206 	if (status != 0)
207 		bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff);
208 	return status;
209 }
210 
211 /**
212  * Return milliseconds of random delay, up to "secs" seconds.
213  */
random_delay_ms(unsigned secs)214 static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
215 {
216 	return (unsigned)rand() % (secs * 1000);
217 }
218 
219 /**
220  * main program
221  */
222 int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
zcip_main(int argc UNUSED_PARAM,char ** argv)223 int zcip_main(int argc UNUSED_PARAM, char **argv)
224 {
225 	char *r_opt;
226 	const char *l_opt = "169.254.0.0";
227 	int state;
228 	int nsent;
229 	unsigned opts;
230 
231 	// Ugly trick, but I want these zeroed in one go
232 	struct {
233 		const struct ether_addr null_ethaddr;
234 		struct ifreq ifr;
235 		uint32_t chosen_nip;
236 		int conflicts;
237 		int timeout_ms; // must be signed
238 		int verbose;
239 	} L;
240 #define null_ethaddr (L.null_ethaddr)
241 #define ifr          (L.ifr         )
242 #define chosen_nip   (L.chosen_nip  )
243 #define conflicts    (L.conflicts   )
244 #define timeout_ms   (L.timeout_ms  )
245 #define verbose      (L.verbose     )
246 
247 	memset(&L, 0, sizeof(L));
248 	INIT_G();
249 
250 #define FOREGROUND (opts & 1)
251 #define QUIT       (opts & 2)
252 	// Parse commandline: prog [options] ifname script
253 	// exactly 2 args; -v accumulates and implies -f
254 	opts = getopt32(argv, "^" "fqr:l:v" "\0" "=2:vv:vf",
255 				&r_opt, &l_opt, &verbose
256 	);
257 #if !BB_MMU
258 	// on NOMMU reexec early (or else we will rerun things twice)
259 	if (!FOREGROUND)
260 		bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
261 #endif
262 	// Open an ARP socket
263 	// (need to do it before openlog to prevent openlog from taking
264 	// fd 3 (sock_fd==3))
265 	xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
266 	if (!FOREGROUND) {
267 		// do it before all bb_xx_msg calls
268 		openlog(applet_name, 0, LOG_DAEMON);
269 		logmode |= LOGMODE_SYSLOG;
270 	}
271 	bb_logenv_override();
272 
273 	{ // -l n.n.n.n
274 		struct in_addr net;
275 		if (inet_aton(l_opt, &net) == 0
276 		 || (net.s_addr & htonl(IN_CLASSB_NET)) != net.s_addr
277 		) {
278 			bb_simple_error_msg_and_die("invalid network address");
279 		}
280 		G.localnet_ip = ntohl(net.s_addr);
281 	}
282 	if (opts & 4) { // -r n.n.n.n
283 		struct in_addr ip;
284 		if (inet_aton(r_opt, &ip) == 0
285 		 || (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
286 		) {
287 			bb_simple_error_msg_and_die("invalid link address");
288 		}
289 		chosen_nip = ip.s_addr;
290 	}
291 	argv += optind - 1;
292 
293 	/* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
294 	/* We need to make space for script argument: */
295 	argv[0] = argv[1];
296 	argv[1] = argv[2];
297 	/* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
298 #define argv_intf (argv[0])
299 
300 	xsetenv("interface", argv_intf);
301 
302 	// Initialize the interface (modprobe, ifup, etc)
303 	if (run(argv, "init", 0))
304 		return EXIT_FAILURE;
305 
306 	// Initialize G.iface_sockaddr
307 	// G.iface_sockaddr is: { u16 sa_family; u8 sa_data[14]; }
308 	//memset(&G.iface_sockaddr, 0, sizeof(G.iface_sockaddr));
309 	//TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
310 	safe_strncpy(G.iface_sockaddr.sa_data, argv_intf, sizeof(G.iface_sockaddr.sa_data));
311 
312 	// Bind to the interface's ARP socket
313 	xbind(sock_fd, &G.iface_sockaddr, sizeof(G.iface_sockaddr));
314 
315 	// Get the interface's ethernet address
316 	//memset(&ifr, 0, sizeof(ifr));
317 	strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
318 	xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
319 	memcpy(&G.our_ethaddr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
320 
321 	// Start with some stable ip address, either a function of
322 	// the hardware address or else the last address we used.
323 	// we are taking low-order four bytes, as top-order ones
324 	// aren't random enough.
325 	// NOTE: the sequence of addresses we try changes only
326 	// depending on when we detect conflicts.
327 	{
328 		uint32_t t;
329 		move_from_unaligned32(t, ((char *)&G.our_ethaddr + 2));
330 		srand(t);
331 	}
332 	// FIXME cases to handle:
333 	//  - zcip already running!
334 	//  - link already has local address... just defend/update
335 
336 	// Daemonize now; don't delay system startup
337 	if (!FOREGROUND) {
338 #if BB_MMU
339 		bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
340 #endif
341 		bb_info_msg("start, interface %s", argv_intf);
342 	}
343 
344 	// Run the dynamic address negotiation protocol,
345 	// restarting after address conflicts:
346 	//  - start with some address we want to try
347 	//  - short random delay
348 	//  - arp probes to see if another host uses it
349 	//    00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff arp who-has 169.254.194.171 tell 0.0.0.0
350 	//  - arp announcements that we're claiming it
351 	//    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
352 	//  - use it
353 	//  - defend it, within limits
354 	// exit if:
355 	// - address is successfully obtained and -q was given:
356 	//   run "<script> config", then exit with exitcode 0
357 	// - poll error (when does this happen?)
358 	// - read error (when does this happen?)
359 	// - sendto error (in send_arp_request()) (when does this happen?)
360 	// - revents & POLLERR (link down). run "<script> deconfig" first
361 	if (chosen_nip == 0) {
362  new_nip_and_PROBE:
363 		chosen_nip = pick_nip();
364 	}
365 	nsent = 0;
366 	state = PROBE;
367 	while (1) {
368 		struct pollfd fds[1];
369 		unsigned deadline_us = deadline_us;
370 		struct arp_packet p;
371 		int ip_conflict;
372 		int n;
373 
374 		fds[0].fd = sock_fd;
375 		fds[0].events = POLLIN;
376 		fds[0].revents = 0;
377 
378 		// Poll, being ready to adjust current timeout
379 		if (!timeout_ms) {
380 			timeout_ms = random_delay_ms(PROBE_WAIT);
381 			// FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
382 			// make the kernel filter out all packets except
383 			// ones we'd care about.
384 		}
385 		if (timeout_ms >= 0) {
386 			// Set deadline_us to the point in time when we timeout
387 			deadline_us = MONOTONIC_US() + timeout_ms * 1000;
388 		}
389 
390 		VDBG("...wait %d %s nsent=%u\n",
391 				timeout_ms, argv_intf, nsent);
392 
393 		n = safe_poll(fds, 1, timeout_ms);
394 		if (n < 0) {
395 			//bb_perror_msg("poll"); - done in safe_poll
396 			return EXIT_FAILURE;
397 		}
398 		if (n == 0) { // timed out?
399 			VDBG("state:%d\n", state);
400 			switch (state) {
401 			case PROBE:
402 				// No conflicting ARP packets were seen:
403 				// we can progress through the states
404 				if (nsent < PROBE_NUM) {
405 					nsent++;
406 					VDBG("probe/%u %s@%s\n",
407 							nsent, argv_intf, nip_to_a(chosen_nip));
408 					timeout_ms = PROBE_MIN * 1000;
409 					timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
410 					send_arp_request(0, &null_ethaddr, chosen_nip);
411 					continue;
412 				}
413 				// Switch to announce state
414 				nsent = 0;
415 				state = ANNOUNCE;
416 				goto send_announce;
417 			case ANNOUNCE:
418 				// No conflicting ARP packets were seen:
419 				// we can progress through the states
420 				if (nsent < ANNOUNCE_NUM) {
421  send_announce:
422 					nsent++;
423 					VDBG("announce/%u %s@%s\n",
424 							nsent, argv_intf, nip_to_a(chosen_nip));
425 					timeout_ms = ANNOUNCE_INTERVAL * 1000;
426 					send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
427 					continue;
428 				}
429 				// Switch to monitor state
430 				// FIXME update filters
431 				run(argv, "config", chosen_nip);
432 				// NOTE: all other exit paths should deconfig...
433 				if (QUIT)
434 					return EXIT_SUCCESS;
435 				// fall through: switch to MONITOR
436 			default:
437 			// case DEFEND:
438 			// case MONITOR: (shouldn't happen, MONITOR timeout is infinite)
439 				// Defend period ended with no ARP replies - we won
440 				timeout_ms = -1; // never timeout in monitor state
441 				state = MONITOR;
442 				continue;
443 			}
444 		}
445 
446 		// Packet arrived, or link went down.
447 		// We need to adjust the timeout in case we didn't receive
448 		// a conflicting packet.
449 		if (timeout_ms > 0) {
450 			unsigned diff = deadline_us - MONOTONIC_US();
451 			if ((int)(diff) < 0) {
452 				// Current time is greater than the expected timeout time.
453 				diff = 0;
454 			}
455 			VDBG("adjusting timeout\n");
456 			timeout_ms = (diff / 1000) | 1; // never 0
457 		}
458 
459 		if ((fds[0].revents & POLLIN) == 0) {
460 			if (fds[0].revents & POLLERR) {
461 				// FIXME: links routinely go down;
462 				// this shouldn't necessarily exit.
463 				bb_error_msg("iface %s is down", argv_intf);
464 				if (state >= MONITOR) {
465 					// Only if we are in MONITOR or DEFEND
466 					run(argv, "deconfig", chosen_nip);
467 				}
468 				return EXIT_FAILURE;
469 			}
470 			continue;
471 		}
472 
473 		// Read ARP packet
474 		if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
475 			bb_simple_perror_msg_and_die(bb_msg_read_error);
476 		}
477 
478 		if (p.eth.ether_type != htons(ETHERTYPE_ARP))
479 			continue;
480 		if (p.arp.arp_op != htons(ARPOP_REQUEST)
481 		 && p.arp.arp_op != htons(ARPOP_REPLY)
482 		) {
483 			continue;
484 		}
485 #ifdef DEBUG
486 		{
487 			struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
488 			struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
489 			struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
490 			struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
491 			VDBG("source=%s %s\n", ether_ntoa(sha),	inet_ntoa(*spa));
492 			VDBG("target=%s %s\n", ether_ntoa(tha),	inet_ntoa(*tpa));
493 		}
494 #endif
495 		ip_conflict = 0;
496 		if (memcmp(&p.arp.arp_sha, &G.our_ethaddr, ETH_ALEN) != 0) {
497 			if (memcmp(p.arp.arp_spa, &chosen_nip, 4) == 0) {
498 				// A probe or reply with source_ip == chosen ip
499 				ip_conflict = 1;
500 			}
501 			if (p.arp.arp_op == htons(ARPOP_REQUEST)
502 			 && memcmp(p.arp.arp_spa, &const_int_0, 4) == 0
503 			 && memcmp(p.arp.arp_tpa, &chosen_nip, 4) == 0
504 			) {
505 				// A probe with source_ip == 0.0.0.0, target_ip == chosen ip:
506 				// another host trying to claim this ip!
507 				ip_conflict |= 2;
508 			}
509 		}
510 		VDBG("state:%d ip_conflict:%d\n", state, ip_conflict);
511 		if (!ip_conflict)
512 			continue;
513 
514 		// Either src or target IP conflict exists
515 		if (state <= ANNOUNCE) {
516 			// PROBE or ANNOUNCE
517 			conflicts++;
518 			timeout_ms = PROBE_MIN * 1000
519 				+ CONFLICT_MULTIPLIER * random_delay_ms(conflicts);
520 			goto new_nip_and_PROBE;
521 		}
522 
523 		// MONITOR or DEFEND: only src IP conflict is a problem
524 		if (ip_conflict & 1) {
525 			if (state == MONITOR) {
526 				// Src IP conflict, defend with a single ARP probe
527 				VDBG("monitor conflict - defending\n");
528 				timeout_ms = DEFEND_INTERVAL * 1000;
529 				state = DEFEND;
530 				send_arp_request(chosen_nip, &G.our_ethaddr, chosen_nip);
531 				continue;
532 			}
533 			// state == DEFEND
534 			// Another src IP conflict, start over
535 			VDBG("defend conflict - starting over\n");
536 			run(argv, "deconfig", chosen_nip);
537 			conflicts = 0;
538 			timeout_ms = 0;
539 			goto new_nip_and_PROBE;
540 		}
541 		// Note: if we only have a target IP conflict here (ip_conflict & 2),
542 		// IOW: if we just saw this sort of ARP packet:
543 		//  aa:bb:cc:dd:ee:ff > xx:xx:xx:xx:xx:xx arp who-has <chosen_nip> tell 0.0.0.0
544 		// we expect _kernel_ to respond to that, because <chosen_nip>
545 		// is (expected to be) configured on this iface.
546 	} // while (1)
547 #undef argv_intf
548 }
549