source: MondoRescue/trunk/mindi-busybox/networking/zcip.c@ 904

Last change on this file since 904 was 904, checked in by Bruno Cornec, 17 years ago

merge -r890:902 $SVN_M/branches/stable
File size: 13.7 KB
Line 
1/*
2 * RFC3927 ZeroConf IPv4 Link-Local addressing
3 * (see <http://www.zeroconf.org/>)
4 *
5 * Copyright (C) 2003 by Arthur van Hoff (avh@strangeberry.com)
6 * Copyright (C) 2004 by David Brownell
7 *
8 * Licensed under the GPL v2 or later, see the file LICENSE in this tarball.
9 */
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
17//#define DEBUG
18
19// TODO:
20// - more real-world usage/testing, especially daemon mode
21// - kernel packet filters to reduce scheduling noise
22// - avoid silent script failures, especially under load...
23// - link status monitoring (restart on link-up; stop on link-down)
24
25#include "busybox.h"
26#include <errno.h>
27#include <string.h>
28#include <syslog.h>
29#include <poll.h>
30#include <time.h>
31
32#include <sys/wait.h>
33
34#include <netinet/ether.h>
35#include <net/ethernet.h>
36#include <net/if.h>
37#include <net/if_arp.h>
38
39#include <linux/if_packet.h>
40#include <linux/sockios.h>
41
42
43struct arp_packet {
44 struct ether_header hdr;
45 struct ether_arp arp;
46} ATTRIBUTE_PACKED;
47
48enum {
49/* 169.254.0.0 */
50 LINKLOCAL_ADDR = 0xa9fe0000,
51
52/* protocol timeout parameters, specified in seconds */
53 PROBE_WAIT = 1,
54 PROBE_MIN = 1,
55 PROBE_MAX = 2,
56 PROBE_NUM = 3,
57 MAX_CONFLICTS = 10,
58 RATE_LIMIT_INTERVAL = 60,
59 ANNOUNCE_WAIT = 2,
60 ANNOUNCE_NUM = 2,
61 ANNOUNCE_INTERVAL = 2,
62 DEFEND_INTERVAL = 10
63};
64
65/* States during the configuration process. */
66enum {
67 PROBE = 0,
68 RATE_LIMIT_PROBE,
69 ANNOUNCE,
70 MONITOR,
71 DEFEND
72};
73
74/* Implicitly zero-initialized */
75static const struct in_addr null_ip;
76static const struct ether_addr null_addr;
77static int verbose;
78
79#define DBG(fmt,args...) \
80 do { } while (0)
81#define VDBG DBG
82
83/**
84 * Pick a random link local IP address on 169.254/16, except that
85 * the first and last 256 addresses are reserved.
86 */
87static void pick(struct in_addr *ip)
88{
89 unsigned tmp;
90
91 /* use cheaper math than lrand48() mod N */
92 do {
93 tmp = (lrand48() >> 16) & IN_CLASSB_HOST;
94 } while (tmp > (IN_CLASSB_HOST - 0x0200));
95 ip->s_addr = htonl((LINKLOCAL_ADDR + 0x0100) + tmp);
96}
97
98/**
99 * Broadcast an ARP packet.
100 */
101static int arp(int fd, struct sockaddr *saddr, int op,
102 const struct ether_addr *source_addr, struct in_addr source_ip,
103 const struct ether_addr *target_addr, struct in_addr target_ip)
104{
105 struct arp_packet p;
106 memset(&p, 0, sizeof(p));
107
108 // ether header
109 p.hdr.ether_type = htons(ETHERTYPE_ARP);
110 memcpy(p.hdr.ether_shost, source_addr, ETH_ALEN);
111 memset(p.hdr.ether_dhost, 0xff, ETH_ALEN);
112
113 // arp request
114 p.arp.arp_hrd = htons(ARPHRD_ETHER);
115 p.arp.arp_pro = htons(ETHERTYPE_IP);
116 p.arp.arp_hln = ETH_ALEN;
117 p.arp.arp_pln = 4;
118 p.arp.arp_op = htons(op);
119 memcpy(&p.arp.arp_sha, source_addr, ETH_ALEN);
120 memcpy(&p.arp.arp_spa, &source_ip, sizeof (p.arp.arp_spa));
121 memcpy(&p.arp.arp_tha, target_addr, ETH_ALEN);
122 memcpy(&p.arp.arp_tpa, &target_ip, sizeof (p.arp.arp_tpa));
123
124 // send it
125 if (sendto(fd, &p, sizeof (p), 0, saddr, sizeof (*saddr)) < 0) {
126 perror("sendto");
127 return -errno;
128 }
129 return 0;
130}
131
132/**
133 * Run a script.
134 */
135static int run(char *script, char *arg, char *intf, struct in_addr *ip)
136{
137 int pid, status;
138 char *why;
139
140 if (script != NULL) {
141 VDBG("%s run %s %s\n", intf, script, arg);
142 if (ip != NULL) {
143 char *addr = inet_ntoa(*ip);
144 setenv("ip", addr, 1);
145 syslog(LOG_INFO, "%s %s %s", arg, intf, addr);
146 }
147
148 pid = vfork();
149 if (pid < 0) { // error
150 why = "vfork";
151 goto bad;
152 } else if (pid == 0) { // child
153 execl(script, script, arg, NULL);
154 perror("execl");
155 _exit(EXIT_FAILURE);
156 }
157
158 if (waitpid(pid, &status, 0) <= 0) {
159 why = "waitpid";
160 goto bad;
161 }
162 if (WEXITSTATUS(status) != 0) {
163 bb_error_msg("script %s failed, exit=%d\n",
164 script, WEXITSTATUS(status));
165 return -errno;
166 }
167 }
168 return 0;
169bad:
170 status = -errno;
171 syslog(LOG_ERR, "%s %s, %s error: %s",
172 arg, intf, why, strerror(errno));
173 return status;
174}
175
176
177/**
178 * Return milliseconds of random delay, up to "secs" seconds.
179 */
180static inline unsigned ms_rdelay(unsigned secs)
181{
182 return lrand48() % (secs * 1000);
183}
184
185/**
186 * main program
187 */
188
189int zcip_main(int argc, char *argv[])
190{
191 char *intf = NULL;
192 char *script = NULL;
193 int quit = 0;
194 int foreground = 0;
195
196 char *why;
197 struct sockaddr saddr;
198 struct ether_addr addr;
199 struct in_addr ip = { 0 };
200 int fd;
201 int ready = 0;
202 suseconds_t timeout = 0; // milliseconds
203 unsigned conflicts = 0;
204 unsigned nprobes = 0;
205 unsigned nclaims = 0;
206 int t;
207 int state = PROBE;
208
209 // parse commandline: prog [options] ifname script
210 while ((t = getopt(argc, argv, "fqr:v")) != EOF) {
211 switch (t) {
212 case 'f':
213 foreground = 1;
214 continue;
215 case 'q':
216 quit = 1;
217 continue;
218 case 'r':
219 if (inet_aton(optarg, &ip) == 0
220 || (ntohl(ip.s_addr) & IN_CLASSB_NET)
221 != LINKLOCAL_ADDR) {
222 bb_error_msg_and_die("invalid link address");
223 }
224 continue;
225 case 'v':
226 verbose++;
227 foreground = 1;
228 continue;
229 default:
230 bb_error_msg_and_die("bad option");
231 }
232 }
233 if (optind < argc - 1) {
234 intf = argv[optind++];
235 setenv("interface", intf, 1);
236 script = argv[optind++];
237 }
238 if (optind != argc || !intf)
239 bb_show_usage();
240 openlog(bb_applet_name, 0, LOG_DAEMON);
241
242 // initialize the interface (modprobe, ifup, etc)
243 if (run(script, "init", intf, NULL) < 0)
244 return EXIT_FAILURE;
245
246 // initialize saddr
247 memset(&saddr, 0, sizeof (saddr));
248 safe_strncpy(saddr.sa_data, intf, sizeof (saddr.sa_data));
249
250 // open an ARP socket
251 if ((fd = socket(PF_PACKET, SOCK_PACKET, htons(ETH_P_ARP))) < 0) {
252 why = "open";
253fail:
254 foreground = 1;
255 goto bad;
256 }
257 // bind to the interface's ARP socket
258 if (bind(fd, &saddr, sizeof (saddr)) < 0) {
259 why = "bind";
260 goto fail;
261 } else {
262 struct ifreq ifr;
263 unsigned short seed[3];
264
265 // get the interface's ethernet address
266 memset(&ifr, 0, sizeof (ifr));
267 strncpy(ifr.ifr_name, intf, sizeof (ifr.ifr_name));
268 if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) {
269 why = "get ethernet address";
270 goto fail;
271 }
272 memcpy(&addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
273
274 // start with some stable ip address, either a function of
275 // the hardware address or else the last address we used.
276 // NOTE: the sequence of addresses we try changes only
277 // depending on when we detect conflicts.
278 memcpy(seed, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
279 seed48(seed);
280 if (ip.s_addr == 0)
281 pick(&ip);
282 }
283
284 // FIXME cases to handle:
285 // - zcip already running!
286 // - link already has local address... just defend/update
287
288 // daemonize now; don't delay system startup
289 if (!foreground) {
290 if (daemon(0, verbose) < 0) {
291 why = "daemon";
292 goto bad;
293 }
294 syslog(LOG_INFO, "start, interface %s", intf);
295 }
296
297 // run the dynamic address negotiation protocol,
298 // restarting after address conflicts:
299 // - start with some address we want to try
300 // - short random delay
301 // - arp probes to see if another host else uses it
302 // - arp announcements that we're claiming it
303 // - use it
304 // - defend it, within limits
305 while (1) {
306 struct pollfd fds[1];
307 struct timeval tv1;
308 struct arp_packet p;
309
310 fds[0].fd = fd;
311 fds[0].events = POLLIN;
312 fds[0].revents = 0;
313
314 int source_ip_conflict = 0;
315 int target_ip_conflict = 0;
316
317 // poll, being ready to adjust current timeout
318 if (!timeout) {
319 timeout = ms_rdelay(PROBE_WAIT);
320 // FIXME setsockopt(fd, SO_ATTACH_FILTER, ...) to
321 // make the kernel filter out all packets except
322 // ones we'd care about.
323 }
324 // set tv1 to the point in time when we timeout
325 gettimeofday(&tv1, NULL);
326 tv1.tv_usec += (timeout % 1000) * 1000;
327 while (tv1.tv_usec > 1000000) {
328 tv1.tv_usec -= 1000000;
329 tv1.tv_sec++;
330 }
331 tv1.tv_sec += timeout / 1000;
332
333 VDBG("...wait %ld %s nprobes=%d, nclaims=%d\n",
334 timeout, intf, nprobes, nclaims);
335 switch (poll(fds, 1, timeout)) {
336
337 // timeout
338 case 0:
339 VDBG("state = %d\n", state);
340 switch (state) {
341 case PROBE:
342 // timeouts in the PROBE state means no conflicting ARP packets
343 // have been received, so we can progress through the states
344 if (nprobes < PROBE_NUM) {
345 nprobes++;
346 VDBG("probe/%d %s@%s\n",
347 nprobes, intf, inet_ntoa(ip));
348 (void)arp(fd, &saddr, ARPOP_REQUEST,
349 &addr, null_ip,
350 &null_addr, ip);
351 timeout = PROBE_MIN * 1000;
352 timeout += ms_rdelay(PROBE_MAX
353 - PROBE_MIN);
354 }
355 else {
356 // Switch to announce state.
357 state = ANNOUNCE;
358 nclaims = 0;
359 VDBG("announce/%d %s@%s\n",
360 nclaims, intf, inet_ntoa(ip));
361 (void)arp(fd, &saddr, ARPOP_REQUEST,
362 &addr, ip,
363 &addr, ip);
364 timeout = ANNOUNCE_INTERVAL * 1000;
365 }
366 break;
367 case RATE_LIMIT_PROBE:
368 // timeouts in the RATE_LIMIT_PROBE state means no conflicting ARP packets
369 // have been received, so we can move immediately to the announce state
370 state = ANNOUNCE;
371 nclaims = 0;
372 VDBG("announce/%d %s@%s\n",
373 nclaims, intf, inet_ntoa(ip));
374 (void)arp(fd, &saddr, ARPOP_REQUEST,
375 &addr, ip,
376 &addr, ip);
377 timeout = ANNOUNCE_INTERVAL * 1000;
378 break;
379 case ANNOUNCE:
380 // timeouts in the ANNOUNCE state means no conflicting ARP packets
381 // have been received, so we can progress through the states
382 if (nclaims < ANNOUNCE_NUM) {
383 nclaims++;
384 VDBG("announce/%d %s@%s\n",
385 nclaims, intf, inet_ntoa(ip));
386 (void)arp(fd, &saddr, ARPOP_REQUEST,
387 &addr, ip,
388 &addr, ip);
389 timeout = ANNOUNCE_INTERVAL * 1000;
390 }
391 else {
392 // Switch to monitor state.
393 state = MONITOR;
394 // link is ok to use earlier
395 // FIXME update filters
396 run(script, "config", intf, &ip);
397 ready = 1;
398 conflicts = 0;
399 timeout = -1; // Never timeout in the monitor state.
400
401 // NOTE: all other exit paths
402 // should deconfig ...
403 if (quit)
404 return EXIT_SUCCESS;
405 }
406 break;
407 case DEFEND:
408 // We won! No ARP replies, so just go back to monitor.
409 state = MONITOR;
410 timeout = -1;
411 conflicts = 0;
412 break;
413 default:
414 // Invalid, should never happen. Restart the whole protocol.
415 state = PROBE;
416 pick(&ip);
417 timeout = 0;
418 nprobes = 0;
419 nclaims = 0;
420 break;
421 } // switch (state)
422 break; // case 0 (timeout)
423 // packets arriving
424 case 1:
425 // We need to adjust the timeout in case we didn't receive
426 // a conflicting packet.
427 if (timeout > 0) {
428 struct timeval tv2;
429
430 gettimeofday(&tv2, NULL);
431 if (timercmp(&tv1, &tv2, <)) {
432 // Current time is greater than the expected timeout time.
433 // Should never happen.
434 VDBG("missed an expected timeout\n");
435 timeout = 0;
436 } else {
437 VDBG("adjusting timeout\n");
438 timersub(&tv1, &tv2, &tv1);
439 timeout = 1000 * tv1.tv_sec
440 + tv1.tv_usec / 1000;
441 }
442 }
443
444 if ((fds[0].revents & POLLIN) == 0) {
445 if (fds[0].revents & POLLERR) {
446 // FIXME: links routinely go down;
447 // this shouldn't necessarily exit.
448 bb_error_msg("%s: poll error\n", intf);
449 if (ready) {
450 run(script, "deconfig",
451 intf, &ip);
452 }
453 return EXIT_FAILURE;
454 }
455 continue;
456 }
457
458 // read ARP packet
459 if (recv(fd, &p, sizeof (p), 0) < 0) {
460 why = "recv";
461 goto bad;
462 }
463 if (p.hdr.ether_type != htons(ETHERTYPE_ARP))
464 continue;
465
466#ifdef DEBUG
467 {
468 struct ether_addr * sha = (struct ether_addr *) p.arp.arp_sha;
469 struct ether_addr * tha = (struct ether_addr *) p.arp.arp_tha;
470 struct in_addr * spa = (struct in_addr *) p.arp.arp_spa;
471 struct in_addr * tpa = (struct in_addr *) p.arp.arp_tpa;
472 VDBG("%s recv arp type=%d, op=%d,\n",
473 intf, ntohs(p.hdr.ether_type),
474 ntohs(p.arp.arp_op));
475 VDBG("\tsource=%s %s\n",
476 ether_ntoa(sha),
477 inet_ntoa(*spa));
478 VDBG("\ttarget=%s %s\n",
479 ether_ntoa(tha),
480 inet_ntoa(*tpa));
481 }
482#endif
483 if (p.arp.arp_op != htons(ARPOP_REQUEST)
484 && p.arp.arp_op != htons(ARPOP_REPLY))
485 continue;
486
487 if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0 &&
488 memcmp(&addr, &p.arp.arp_sha, ETH_ALEN) != 0) {
489 source_ip_conflict = 1;
490 }
491 if (memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0 &&
492 p.arp.arp_op == htons(ARPOP_REQUEST) &&
493 memcmp(&addr, &p.arp.arp_tha, ETH_ALEN) != 0) {
494 target_ip_conflict = 1;
495 }
496
497 VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n",
498 state, source_ip_conflict, target_ip_conflict);
499 switch (state) {
500 case PROBE:
501 case ANNOUNCE:
502 // When probing or announcing, check for source IP conflicts
503 // and other hosts doing ARP probes (target IP conflicts).
504 if (source_ip_conflict || target_ip_conflict) {
505 conflicts++;
506 if (conflicts >= MAX_CONFLICTS) {
507 VDBG("%s ratelimit\n", intf);
508 timeout = RATE_LIMIT_INTERVAL * 1000;
509 state = RATE_LIMIT_PROBE;
510 }
511
512 // restart the whole protocol
513 pick(&ip);
514 timeout = 0;
515 nprobes = 0;
516 nclaims = 0;
517 }
518 break;
519 case MONITOR:
520 // If a conflict, we try to defend with a single ARP probe.
521 if (source_ip_conflict) {
522 VDBG("monitor conflict -- defending\n");
523 state = DEFEND;
524 timeout = DEFEND_INTERVAL * 1000;
525 (void)arp(fd, &saddr,
526 ARPOP_REQUEST,
527 &addr, ip,
528 &addr, ip);
529 }
530 break;
531 case DEFEND:
532 // Well, we tried. Start over (on conflict).
533 if (source_ip_conflict) {
534 state = PROBE;
535 VDBG("defend conflict -- starting over\n");
536 ready = 0;
537 run(script, "deconfig", intf, &ip);
538
539 // restart the whole protocol
540 pick(&ip);
541 timeout = 0;
542 nprobes = 0;
543 nclaims = 0;
544 }
545 break;
546 default:
547 // Invalid, should never happen. Restart the whole protocol.
548 VDBG("invalid state -- starting over\n");
549 state = PROBE;
550 pick(&ip);
551 timeout = 0;
552 nprobes = 0;
553 nclaims = 0;
554 break;
555 } // switch state
556
557 break; // case 1 (packets arriving)
558 default:
559 why = "poll";
560 goto bad;
561 } // switch poll
562 }
563bad:
564 if (foreground)
565 perror(why);
566 else
567 syslog(LOG_ERR, "%s %s, %s error: %s",
568 bb_applet_name, intf, why, strerror(errno));
569 return EXIT_FAILURE;
570}
Note: See TracBrowser for help on using the repository browser.