source: MondoRescue/branches/2.2.5/mindi-busybox/networking/zcip.c@ 1765

Last change on this file since 1765 was 1765, checked in by Bruno Cornec, 16 years ago

Update to busybox 1.7.2
File size: 13.7 KB
Line 
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 the GPL v2 or later, see the file LICENSE in this tarball.
10 */
11
12/*
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.
16 */
17
18//#define DEBUG
19
20// TODO:
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)
25
26#include <syslog.h>
27#include <poll.h>
28#include <sys/wait.h>
29#include <netinet/ether.h>
30#include <net/ethernet.h>
31#include <net/if.h>
32#include <net/if_arp.h>
33#include <linux/if_packet.h>
34#include <linux/sockios.h>
35
36#include "libbb.h"
37
38/* We don't need more than 32 bits of the counter */
39#define MONOTONIC_US() ((unsigned)monotonic_us())
40
41struct arp_packet {
42 struct ether_header hdr;
43 struct ether_arp arp;
44} ATTRIBUTE_PACKED;
45
46enum {
47/* 169.254.0.0 */
48 LINKLOCAL_ADDR = 0xa9fe0000,
49
50/* protocol timeout parameters, specified in seconds */
51 PROBE_WAIT = 1,
52 PROBE_MIN = 1,
53 PROBE_MAX = 2,
54 PROBE_NUM = 3,
55 MAX_CONFLICTS = 10,
56 RATE_LIMIT_INTERVAL = 60,
57 ANNOUNCE_WAIT = 2,
58 ANNOUNCE_NUM = 2,
59 ANNOUNCE_INTERVAL = 2,
60 DEFEND_INTERVAL = 10
61};
62
63/* States during the configuration process. */
64enum {
65 PROBE = 0,
66 RATE_LIMIT_PROBE,
67 ANNOUNCE,
68 MONITOR,
69 DEFEND
70};
71
72#define VDBG(fmt,args...) \
73 do { } while (0)
74
75/**
76 * Pick a random link local IP address on 169.254/16, except that
77 * the first and last 256 addresses are reserved.
78 */
79static void pick(struct in_addr *ip)
80{
81 unsigned tmp;
82
83 do {
84 tmp = rand() & IN_CLASSB_HOST;
85 } while (tmp > (IN_CLASSB_HOST - 0x0200));
86 ip->s_addr = htonl((LINKLOCAL_ADDR + 0x0100) + tmp);
87}
88
89/**
90 * Broadcast an ARP packet.
91 */
92static void arp(int fd, struct sockaddr *saddr, int op,
93 const struct ether_addr *source_addr, struct in_addr source_ip,
94 const struct ether_addr *target_addr, struct in_addr target_ip)
95{
96 struct arp_packet p;
97 memset(&p, 0, sizeof(p));
98
99 // ether header
100 p.hdr.ether_type = htons(ETHERTYPE_ARP);
101 memcpy(p.hdr.ether_shost, source_addr, ETH_ALEN);
102 memset(p.hdr.ether_dhost, 0xff, ETH_ALEN);
103
104 // arp request
105 p.arp.arp_hrd = htons(ARPHRD_ETHER);
106 p.arp.arp_pro = htons(ETHERTYPE_IP);
107 p.arp.arp_hln = ETH_ALEN;
108 p.arp.arp_pln = 4;
109 p.arp.arp_op = htons(op);
110 memcpy(&p.arp.arp_sha, source_addr, ETH_ALEN);
111 memcpy(&p.arp.arp_spa, &source_ip, sizeof(p.arp.arp_spa));
112 memcpy(&p.arp.arp_tha, target_addr, ETH_ALEN);
113 memcpy(&p.arp.arp_tpa, &target_ip, sizeof(p.arp.arp_tpa));
114
115 // send it
116 xsendto(fd, &p, sizeof(p), saddr, sizeof(*saddr));
117
118 // Currently all callers ignore errors, that's why returns are
119 // commented out...
120 //return 0;
121}
122
123/**
124 * Run a script. argv[2] is already NULL.
125 */
126static int run(char *argv[3], const char *intf, struct in_addr *ip)
127{
128 int status;
129
130 VDBG("%s run %s %s\n", intf, argv[0], argv[1]);
131
132 if (ip) {
133 char *addr = inet_ntoa(*ip);
134 setenv("ip", addr, 1);
135 bb_info_msg("%s %s %s", argv[1], intf, addr);
136 }
137
138 status = wait4pid(spawn(argv));
139 if (status < 0) {
140 bb_perror_msg("%s %s", argv[1], intf);
141 return -errno;
142 }
143 if (status != 0)
144 bb_error_msg("script %s %s failed, exitcode=%d", argv[0], argv[1], status);
145 return status;
146}
147
148/**
149 * Return milliseconds of random delay, up to "secs" seconds.
150 */
151static unsigned ALWAYS_INLINE ms_rdelay(unsigned secs)
152{
153 return rand() % (secs * 1000);
154}
155
156/**
157 * main program
158 */
159int zcip_main(int argc, char **argv);
160int zcip_main(int argc, char **argv)
161{
162 int state = PROBE;
163 struct ether_addr eth_addr;
164 const char *why;
165 int fd;
166 char *r_opt;
167 unsigned opts;
168
169 /* Ugly trick, but I want these zeroed in one go */
170 struct {
171 const struct in_addr null_ip;
172 const struct ether_addr null_addr;
173 struct sockaddr saddr;
174 struct in_addr ip;
175 struct ifreq ifr;
176 char *intf;
177 char *script_av[3];
178 int timeout_ms; /* must be signed */
179 unsigned conflicts;
180 unsigned nprobes;
181 unsigned nclaims;
182 int ready;
183 int verbose;
184 } L;
185#define null_ip (L.null_ip )
186#define null_addr (L.null_addr )
187#define saddr (L.saddr )
188#define ip (L.ip )
189#define ifr (L.ifr )
190#define intf (L.intf )
191#define script_av (L.script_av )
192#define timeout_ms (L.timeout_ms)
193#define conflicts (L.conflicts )
194#define nprobes (L.nprobes )
195#define nclaims (L.nclaims )
196#define ready (L.ready )
197#define verbose (L.verbose )
198
199 memset(&L, 0, sizeof(L));
200
201#define FOREGROUND (opts & 1)
202#define QUIT (opts & 2)
203 // parse commandline: prog [options] ifname script
204 // exactly 2 args; -v accumulates and implies -f
205 opt_complementary = "=2:vv:vf";
206 opts = getopt32(argv, "fqr:v", &r_opt, &verbose);
207 if (!FOREGROUND) {
208 /* Do it early, before all bb_xx_msg calls */
209 openlog(applet_name, 0, LOG_DAEMON);
210 logmode |= LOGMODE_SYSLOG;
211 }
212 if (opts & 4) { // -r n.n.n.n
213 if (inet_aton(r_opt, &ip) == 0
214 || (ntohl(ip.s_addr) & IN_CLASSB_NET) != LINKLOCAL_ADDR
215 ) {
216 bb_error_msg_and_die("invalid link address");
217 }
218 }
219 // On NOMMU reexec early (or else we will rerun things twice)
220#if !BB_MMU
221 if (!FOREGROUND)
222 bb_daemonize_or_rexec(DAEMON_CHDIR_ROOT, argv);
223#endif
224 argc -= optind;
225 argv += optind;
226
227 intf = argv[0];
228 script_av[0] = argv[1];
229 setenv("interface", intf, 1);
230
231 // initialize the interface (modprobe, ifup, etc)
232 script_av[1] = (char*)"init";
233 if (run(script_av, intf, NULL))
234 return EXIT_FAILURE;
235
236 // initialize saddr
237 //memset(&saddr, 0, sizeof(saddr));
238 safe_strncpy(saddr.sa_data, intf, sizeof(saddr.sa_data));
239
240 // open an ARP socket
241 fd = xsocket(PF_PACKET, SOCK_PACKET, htons(ETH_P_ARP));
242 // bind to the interface's ARP socket
243 xbind(fd, &saddr, sizeof(saddr));
244
245 // get the interface's ethernet address
246 //memset(&ifr, 0, sizeof(ifr));
247 strncpy(ifr.ifr_name, intf, sizeof(ifr.ifr_name));
248 xioctl(fd, SIOCGIFHWADDR, &ifr);
249 memcpy(&eth_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
250
251 // start with some stable ip address, either a function of
252 // the hardware address or else the last address we used.
253 // NOTE: the sequence of addresses we try changes only
254 // depending on when we detect conflicts.
255 srand(*(unsigned*)&ifr.ifr_hwaddr.sa_data);
256 if (ip.s_addr == 0)
257 pick(&ip);
258
259 // FIXME cases to handle:
260 // - zcip already running!
261 // - link already has local address... just defend/update
262
263 // daemonize now; don't delay system startup
264 if (!FOREGROUND) {
265#if BB_MMU
266 bb_daemonize(DAEMON_CHDIR_ROOT);
267#endif
268 bb_info_msg("start, interface %s", intf);
269 }
270
271 // run the dynamic address negotiation protocol,
272 // restarting after address conflicts:
273 // - start with some address we want to try
274 // - short random delay
275 // - arp probes to see if another host else uses it
276 // - arp announcements that we're claiming it
277 // - use it
278 // - defend it, within limits
279 while (1) {
280 struct pollfd fds[1];
281 unsigned deadline_us;
282 struct arp_packet p;
283
284 int source_ip_conflict = 0;
285 int target_ip_conflict = 0;
286
287 fds[0].fd = fd;
288 fds[0].events = POLLIN;
289 fds[0].revents = 0;
290
291 // poll, being ready to adjust current timeout
292 if (!timeout_ms) {
293 timeout_ms = ms_rdelay(PROBE_WAIT);
294 // FIXME setsockopt(fd, SO_ATTACH_FILTER, ...) to
295 // make the kernel filter out all packets except
296 // ones we'd care about.
297 }
298 // set deadline_us to the point in time when we timeout
299 deadline_us = MONOTONIC_US() + timeout_ms * 1000;
300
301 VDBG("...wait %d %s nprobes=%u, nclaims=%u\n",
302 timeout_ms, intf, nprobes, nclaims);
303 switch (poll(fds, 1, timeout_ms)) {
304
305 // timeout
306 case 0:
307 VDBG("state = %d\n", state);
308 switch (state) {
309 case PROBE:
310 // timeouts in the PROBE state mean no conflicting ARP packets
311 // have been received, so we can progress through the states
312 if (nprobes < PROBE_NUM) {
313 nprobes++;
314 VDBG("probe/%u %s@%s\n",
315 nprobes, intf, inet_ntoa(ip));
316 arp(fd, &saddr, ARPOP_REQUEST,
317 &eth_addr, null_ip,
318 &null_addr, ip);
319 timeout_ms = PROBE_MIN * 1000;
320 timeout_ms += ms_rdelay(PROBE_MAX - PROBE_MIN);
321 }
322 else {
323 // Switch to announce state.
324 state = ANNOUNCE;
325 nclaims = 0;
326 VDBG("announce/%u %s@%s\n",
327 nclaims, intf, inet_ntoa(ip));
328 arp(fd, &saddr, ARPOP_REQUEST,
329 &eth_addr, ip,
330 &eth_addr, ip);
331 timeout_ms = ANNOUNCE_INTERVAL * 1000;
332 }
333 break;
334 case RATE_LIMIT_PROBE:
335 // timeouts in the RATE_LIMIT_PROBE state mean no conflicting ARP packets
336 // have been received, so we can move immediately to the announce state
337 state = ANNOUNCE;
338 nclaims = 0;
339 VDBG("announce/%u %s@%s\n",
340 nclaims, intf, inet_ntoa(ip));
341 arp(fd, &saddr, ARPOP_REQUEST,
342 &eth_addr, ip,
343 &eth_addr, ip);
344 timeout_ms = ANNOUNCE_INTERVAL * 1000;
345 break;
346 case ANNOUNCE:
347 // timeouts in the ANNOUNCE state mean no conflicting ARP packets
348 // have been received, so we can progress through the states
349 if (nclaims < ANNOUNCE_NUM) {
350 nclaims++;
351 VDBG("announce/%u %s@%s\n",
352 nclaims, intf, inet_ntoa(ip));
353 arp(fd, &saddr, ARPOP_REQUEST,
354 &eth_addr, ip,
355 &eth_addr, ip);
356 timeout_ms = ANNOUNCE_INTERVAL * 1000;
357 }
358 else {
359 // Switch to monitor state.
360 state = MONITOR;
361 // link is ok to use earlier
362 // FIXME update filters
363 script_av[1] = (char*)"config";
364 run(script_av, intf, &ip);
365 ready = 1;
366 conflicts = 0;
367 timeout_ms = -1; // Never timeout in the monitor state.
368
369 // NOTE: all other exit paths
370 // should deconfig ...
371 if (QUIT)
372 return EXIT_SUCCESS;
373 }
374 break;
375 case DEFEND:
376 // We won! No ARP replies, so just go back to monitor.
377 state = MONITOR;
378 timeout_ms = -1;
379 conflicts = 0;
380 break;
381 default:
382 // Invalid, should never happen. Restart the whole protocol.
383 state = PROBE;
384 pick(&ip);
385 timeout_ms = 0;
386 nprobes = 0;
387 nclaims = 0;
388 break;
389 } // switch (state)
390 break; // case 0 (timeout)
391 // packets arriving
392 case 1:
393 // We need to adjust the timeout in case we didn't receive
394 // a conflicting packet.
395 if (timeout_ms > 0) {
396 unsigned diff = deadline_us - MONOTONIC_US();
397 if ((int)(diff) < 0) {
398 // Current time is greater than the expected timeout time.
399 // Should never happen.
400 VDBG("missed an expected timeout\n");
401 timeout_ms = 0;
402 } else {
403 VDBG("adjusting timeout\n");
404 timeout_ms = diff / 1000;
405 if (!timeout_ms) timeout_ms = 1;
406 }
407 }
408
409 if ((fds[0].revents & POLLIN) == 0) {
410 if (fds[0].revents & POLLERR) {
411 // FIXME: links routinely go down;
412 // this shouldn't necessarily exit.
413 bb_error_msg("%s: poll error", intf);
414 if (ready) {
415 script_av[1] = (char*)"deconfig";
416 run(script_av, intf, &ip);
417 }
418 return EXIT_FAILURE;
419 }
420 continue;
421 }
422
423 // read ARP packet
424 if (recv(fd, &p, sizeof(p), 0) < 0) {
425 why = "recv";
426 goto bad;
427 }
428 if (p.hdr.ether_type != htons(ETHERTYPE_ARP))
429 continue;
430
431#ifdef DEBUG
432 {
433 struct ether_addr * sha = (struct ether_addr *) p.arp.arp_sha;
434 struct ether_addr * tha = (struct ether_addr *) p.arp.arp_tha;
435 struct in_addr * spa = (struct in_addr *) p.arp.arp_spa;
436 struct in_addr * tpa = (struct in_addr *) p.arp.arp_tpa;
437 VDBG("%s recv arp type=%d, op=%d,\n",
438 intf, ntohs(p.hdr.ether_type),
439 ntohs(p.arp.arp_op));
440 VDBG("\tsource=%s %s\n",
441 ether_ntoa(sha),
442 inet_ntoa(*spa));
443 VDBG("\ttarget=%s %s\n",
444 ether_ntoa(tha),
445 inet_ntoa(*tpa));
446 }
447#endif
448 if (p.arp.arp_op != htons(ARPOP_REQUEST)
449 && p.arp.arp_op != htons(ARPOP_REPLY))
450 continue;
451
452 if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0 &&
453 memcmp(&eth_addr, &p.arp.arp_sha, ETH_ALEN) != 0) {
454 source_ip_conflict = 1;
455 }
456 if (memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0 &&
457 p.arp.arp_op == htons(ARPOP_REQUEST) &&
458 memcmp(&eth_addr, &p.arp.arp_tha, ETH_ALEN) != 0) {
459 target_ip_conflict = 1;
460 }
461
462 VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n",
463 state, source_ip_conflict, target_ip_conflict);
464 switch (state) {
465 case PROBE:
466 case ANNOUNCE:
467 // When probing or announcing, check for source IP conflicts
468 // and other hosts doing ARP probes (target IP conflicts).
469 if (source_ip_conflict || target_ip_conflict) {
470 conflicts++;
471 if (conflicts >= MAX_CONFLICTS) {
472 VDBG("%s ratelimit\n", intf);
473 timeout_ms = RATE_LIMIT_INTERVAL * 1000;
474 state = RATE_LIMIT_PROBE;
475 }
476
477 // restart the whole protocol
478 pick(&ip);
479 timeout_ms = 0;
480 nprobes = 0;
481 nclaims = 0;
482 }
483 break;
484 case MONITOR:
485 // If a conflict, we try to defend with a single ARP probe.
486 if (source_ip_conflict) {
487 VDBG("monitor conflict -- defending\n");
488 state = DEFEND;
489 timeout_ms = DEFEND_INTERVAL * 1000;
490 arp(fd, &saddr,
491 ARPOP_REQUEST,
492 &eth_addr, ip,
493 &eth_addr, ip);
494 }
495 break;
496 case DEFEND:
497 // Well, we tried. Start over (on conflict).
498 if (source_ip_conflict) {
499 state = PROBE;
500 VDBG("defend conflict -- starting over\n");
501 ready = 0;
502 script_av[1] = (char*)"deconfig";
503 run(script_av, intf, &ip);
504
505 // restart the whole protocol
506 pick(&ip);
507 timeout_ms = 0;
508 nprobes = 0;
509 nclaims = 0;
510 }
511 break;
512 default:
513 // Invalid, should never happen. Restart the whole protocol.
514 VDBG("invalid state -- starting over\n");
515 state = PROBE;
516 pick(&ip);
517 timeout_ms = 0;
518 nprobes = 0;
519 nclaims = 0;
520 break;
521 } // switch state
522
523 break; // case 1 (packets arriving)
524 default:
525 why = "poll";
526 goto bad;
527 } // switch poll
528 }
529 bad:
530 bb_perror_msg("%s, %s", intf, why);
531 return EXIT_FAILURE;
532}
Note: See TracBrowser for help on using the repository browser.