Contents of /trunk/mkinitrd-magellan/busybox/networking/zcip.c
Parent Directory | Revision Log
Revision 532 -
(show annotations)
(download)
Sat Sep 1 22:45:15 2007 UTC (16 years, 8 months ago) by niro
File MIME type: text/plain
File size: 14215 byte(s)
Sat Sep 1 22:45:15 2007 UTC (16 years, 8 months ago) by niro
File MIME type: text/plain
File size: 14215 byte(s)
-import if magellan mkinitrd; it is a fork of redhats mkinitrd-5.0.8 with all magellan patches and features; deprecates magellan-src/mkinitrd
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 "busybox.h" |
27 | #include <syslog.h> |
28 | #include <poll.h> |
29 | #include <sys/wait.h> |
30 | #include <netinet/ether.h> |
31 | #include <net/ethernet.h> |
32 | #include <net/if.h> |
33 | #include <net/if_arp.h> |
34 | |
35 | #include <linux/if_packet.h> |
36 | #include <linux/sockios.h> |
37 | |
38 | |
39 | struct arp_packet { |
40 | struct ether_header hdr; |
41 | struct ether_arp arp; |
42 | } ATTRIBUTE_PACKED; |
43 | |
44 | enum { |
45 | /* 169.254.0.0 */ |
46 | LINKLOCAL_ADDR = 0xa9fe0000, |
47 | |
48 | /* protocol timeout parameters, specified in seconds */ |
49 | PROBE_WAIT = 1, |
50 | PROBE_MIN = 1, |
51 | PROBE_MAX = 2, |
52 | PROBE_NUM = 3, |
53 | MAX_CONFLICTS = 10, |
54 | RATE_LIMIT_INTERVAL = 60, |
55 | ANNOUNCE_WAIT = 2, |
56 | ANNOUNCE_NUM = 2, |
57 | ANNOUNCE_INTERVAL = 2, |
58 | DEFEND_INTERVAL = 10 |
59 | }; |
60 | |
61 | /* States during the configuration process. */ |
62 | enum { |
63 | PROBE = 0, |
64 | RATE_LIMIT_PROBE, |
65 | ANNOUNCE, |
66 | MONITOR, |
67 | DEFEND |
68 | }; |
69 | |
70 | #define VDBG(fmt,args...) \ |
71 | do { } while (0) |
72 | |
73 | static unsigned opts; |
74 | #define FOREGROUND (opts & 1) |
75 | #define QUIT (opts & 2) |
76 | |
77 | /** |
78 | * Pick a random link local IP address on 169.254/16, except that |
79 | * the first and last 256 addresses are reserved. |
80 | */ |
81 | static void pick(struct in_addr *ip) |
82 | { |
83 | unsigned tmp; |
84 | |
85 | /* use cheaper math than lrand48() mod N */ |
86 | do { |
87 | tmp = (lrand48() >> 16) & IN_CLASSB_HOST; |
88 | } while (tmp > (IN_CLASSB_HOST - 0x0200)); |
89 | ip->s_addr = htonl((LINKLOCAL_ADDR + 0x0100) + tmp); |
90 | } |
91 | |
92 | /* TODO: we need a flag to direct bb_[p]error_msg output to stderr. */ |
93 | |
94 | /** |
95 | * Broadcast an ARP packet. |
96 | */ |
97 | static void arp(int fd, struct sockaddr *saddr, int op, |
98 | const struct ether_addr *source_addr, struct in_addr source_ip, |
99 | const struct ether_addr *target_addr, struct in_addr target_ip) |
100 | { |
101 | struct arp_packet p; |
102 | memset(&p, 0, sizeof(p)); |
103 | |
104 | // ether header |
105 | p.hdr.ether_type = htons(ETHERTYPE_ARP); |
106 | memcpy(p.hdr.ether_shost, source_addr, ETH_ALEN); |
107 | memset(p.hdr.ether_dhost, 0xff, ETH_ALEN); |
108 | |
109 | // arp request |
110 | p.arp.arp_hrd = htons(ARPHRD_ETHER); |
111 | p.arp.arp_pro = htons(ETHERTYPE_IP); |
112 | p.arp.arp_hln = ETH_ALEN; |
113 | p.arp.arp_pln = 4; |
114 | p.arp.arp_op = htons(op); |
115 | memcpy(&p.arp.arp_sha, source_addr, ETH_ALEN); |
116 | memcpy(&p.arp.arp_spa, &source_ip, sizeof (p.arp.arp_spa)); |
117 | memcpy(&p.arp.arp_tha, target_addr, ETH_ALEN); |
118 | memcpy(&p.arp.arp_tpa, &target_ip, sizeof (p.arp.arp_tpa)); |
119 | |
120 | // send it |
121 | if (sendto(fd, &p, sizeof (p), 0, saddr, sizeof (*saddr)) < 0) { |
122 | bb_perror_msg("sendto"); |
123 | //return -errno; |
124 | } |
125 | // Currently all callers ignore errors, that's why returns are |
126 | // commented out... |
127 | //return 0; |
128 | } |
129 | |
130 | /** |
131 | * Run a script. |
132 | */ |
133 | static int run(char *script, char *arg, char *intf, struct in_addr *ip) |
134 | { |
135 | int pid, status; |
136 | char *why; |
137 | |
138 | if(1) { //always true: if (script != NULL) |
139 | VDBG("%s run %s %s\n", intf, script, arg); |
140 | if (ip != NULL) { |
141 | char *addr = inet_ntoa(*ip); |
142 | setenv("ip", addr, 1); |
143 | bb_info_msg("%s %s %s", arg, intf, addr); |
144 | } |
145 | |
146 | pid = vfork(); |
147 | if (pid < 0) { // error |
148 | why = "vfork"; |
149 | goto bad; |
150 | } else if (pid == 0) { // child |
151 | execl(script, script, arg, NULL); |
152 | bb_perror_msg("execl"); |
153 | _exit(EXIT_FAILURE); |
154 | } |
155 | |
156 | if (waitpid(pid, &status, 0) <= 0) { |
157 | why = "waitpid"; |
158 | goto bad; |
159 | } |
160 | if (WEXITSTATUS(status) != 0) { |
161 | bb_error_msg("script %s failed, exit=%d", |
162 | script, WEXITSTATUS(status)); |
163 | return -errno; |
164 | } |
165 | } |
166 | return 0; |
167 | bad: |
168 | status = -errno; |
169 | bb_perror_msg("%s %s, %s", arg, intf, why); |
170 | return status; |
171 | } |
172 | |
173 | |
174 | /** |
175 | * Return milliseconds of random delay, up to "secs" seconds. |
176 | */ |
177 | static unsigned ATTRIBUTE_ALWAYS_INLINE ms_rdelay(unsigned secs) |
178 | { |
179 | return lrand48() % (secs * 1000); |
180 | } |
181 | |
182 | /** |
183 | * main program |
184 | */ |
185 | |
186 | /* Used to be auto variables on main() stack, but |
187 | * most of them were zero-inited. Moving them to bss |
188 | * is more space-efficient. |
189 | */ |
190 | static const struct in_addr null_ip; // = { 0 }; |
191 | static const struct ether_addr null_addr; // = { {0, 0, 0, 0, 0, 0} }; |
192 | |
193 | static struct sockaddr saddr; // memset(0); |
194 | static struct in_addr ip; // = { 0 }; |
195 | static struct ifreq ifr; //memset(0); |
196 | |
197 | static char *intf; // = NULL; |
198 | static char *script; // = NULL; |
199 | static suseconds_t timeout; // = 0; // milliseconds |
200 | static unsigned conflicts; // = 0; |
201 | static unsigned nprobes; // = 0; |
202 | static unsigned nclaims; // = 0; |
203 | static int ready; // = 0; |
204 | static int verbose; // = 0; |
205 | static int state = PROBE; |
206 | |
207 | int zcip_main(int argc, char *argv[]) |
208 | { |
209 | struct ether_addr eth_addr; |
210 | char *why; |
211 | int fd; |
212 | |
213 | // parse commandline: prog [options] ifname script |
214 | char *r_opt; |
215 | opt_complementary = "vv:vf"; // -v accumulates and implies -f |
216 | opts = getopt32(argc, argv, "fqr:v", &r_opt, &verbose); |
217 | if (!FOREGROUND) { |
218 | /* Do it early, before all bb_xx_msg calls */ |
219 | logmode = LOGMODE_SYSLOG; |
220 | openlog(applet_name, 0, LOG_DAEMON); |
221 | } |
222 | if (opts & 4) { // -r n.n.n.n |
223 | if (inet_aton(r_opt, &ip) == 0 |
224 | || (ntohl(ip.s_addr) & IN_CLASSB_NET) != LINKLOCAL_ADDR |
225 | ) { |
226 | bb_error_msg_and_die("invalid link address"); |
227 | } |
228 | } |
229 | argc -= optind; |
230 | argv += optind; |
231 | if (argc != 2) |
232 | bb_show_usage(); |
233 | intf = argv[0]; |
234 | script = argv[1]; |
235 | setenv("interface", intf, 1); |
236 | |
237 | // initialize the interface (modprobe, ifup, etc) |
238 | if (run(script, "init", intf, NULL) < 0) |
239 | return EXIT_FAILURE; |
240 | |
241 | // initialize saddr |
242 | //memset(&saddr, 0, sizeof (saddr)); |
243 | safe_strncpy(saddr.sa_data, intf, sizeof (saddr.sa_data)); |
244 | |
245 | // open an ARP socket |
246 | fd = xsocket(PF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)); |
247 | // bind to the interface's ARP socket |
248 | xbind(fd, &saddr, sizeof (saddr)); |
249 | |
250 | // get the interface's ethernet address |
251 | //memset(&ifr, 0, sizeof (ifr)); |
252 | strncpy(ifr.ifr_name, intf, sizeof (ifr.ifr_name)); |
253 | if (ioctl(fd, SIOCGIFHWADDR, &ifr) < 0) { |
254 | bb_perror_msg_and_die("get ethernet address"); |
255 | } |
256 | memcpy(ð_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN); |
257 | |
258 | // start with some stable ip address, either a function of |
259 | // the hardware address or else the last address we used. |
260 | // NOTE: the sequence of addresses we try changes only |
261 | // depending on when we detect conflicts. |
262 | // (SVID 3 bogon: who says that "short" is always 16 bits?) |
263 | seed48( (unsigned short*)&ifr.ifr_hwaddr.sa_data ); |
264 | if (ip.s_addr == 0) |
265 | pick(&ip); |
266 | |
267 | // FIXME cases to handle: |
268 | // - zcip already running! |
269 | // - link already has local address... just defend/update |
270 | |
271 | // daemonize now; don't delay system startup |
272 | if (!FOREGROUND) { |
273 | setsid(); |
274 | bb_daemonize(); |
275 | bb_info_msg("start, interface %s", intf); |
276 | } |
277 | |
278 | // run the dynamic address negotiation protocol, |
279 | // restarting after address conflicts: |
280 | // - start with some address we want to try |
281 | // - short random delay |
282 | // - arp probes to see if another host else uses it |
283 | // - arp announcements that we're claiming it |
284 | // - use it |
285 | // - defend it, within limits |
286 | while (1) { |
287 | struct pollfd fds[1]; |
288 | struct timeval tv1; |
289 | struct arp_packet p; |
290 | |
291 | int source_ip_conflict = 0; |
292 | int target_ip_conflict = 0; |
293 | |
294 | fds[0].fd = fd; |
295 | fds[0].events = POLLIN; |
296 | fds[0].revents = 0; |
297 | |
298 | // poll, being ready to adjust current timeout |
299 | if (!timeout) { |
300 | timeout = ms_rdelay(PROBE_WAIT); |
301 | // FIXME setsockopt(fd, SO_ATTACH_FILTER, ...) to |
302 | // make the kernel filter out all packets except |
303 | // ones we'd care about. |
304 | } |
305 | // set tv1 to the point in time when we timeout |
306 | gettimeofday(&tv1, NULL); |
307 | tv1.tv_usec += (timeout % 1000) * 1000; |
308 | while (tv1.tv_usec > 1000000) { |
309 | tv1.tv_usec -= 1000000; |
310 | tv1.tv_sec++; |
311 | } |
312 | tv1.tv_sec += timeout / 1000; |
313 | |
314 | VDBG("...wait %ld %s nprobes=%d, nclaims=%d\n", |
315 | timeout, intf, nprobes, nclaims); |
316 | switch (poll(fds, 1, timeout)) { |
317 | |
318 | // timeout |
319 | case 0: |
320 | VDBG("state = %d\n", state); |
321 | switch (state) { |
322 | case PROBE: |
323 | // timeouts in the PROBE state mean no conflicting ARP packets |
324 | // have been received, so we can progress through the states |
325 | if (nprobes < PROBE_NUM) { |
326 | nprobes++; |
327 | VDBG("probe/%d %s@%s\n", |
328 | nprobes, intf, inet_ntoa(ip)); |
329 | arp(fd, &saddr, ARPOP_REQUEST, |
330 | ð_addr, null_ip, |
331 | &null_addr, ip); |
332 | timeout = PROBE_MIN * 1000; |
333 | timeout += ms_rdelay(PROBE_MAX |
334 | - PROBE_MIN); |
335 | } |
336 | else { |
337 | // Switch to announce state. |
338 | state = ANNOUNCE; |
339 | nclaims = 0; |
340 | VDBG("announce/%d %s@%s\n", |
341 | nclaims, intf, inet_ntoa(ip)); |
342 | arp(fd, &saddr, ARPOP_REQUEST, |
343 | ð_addr, ip, |
344 | ð_addr, ip); |
345 | timeout = ANNOUNCE_INTERVAL * 1000; |
346 | } |
347 | break; |
348 | case RATE_LIMIT_PROBE: |
349 | // timeouts in the RATE_LIMIT_PROBE state mean no conflicting ARP packets |
350 | // have been received, so we can move immediately to the announce state |
351 | state = ANNOUNCE; |
352 | nclaims = 0; |
353 | VDBG("announce/%d %s@%s\n", |
354 | nclaims, intf, inet_ntoa(ip)); |
355 | arp(fd, &saddr, ARPOP_REQUEST, |
356 | ð_addr, ip, |
357 | ð_addr, ip); |
358 | timeout = ANNOUNCE_INTERVAL * 1000; |
359 | break; |
360 | case ANNOUNCE: |
361 | // timeouts in the ANNOUNCE state mean no conflicting ARP packets |
362 | // have been received, so we can progress through the states |
363 | if (nclaims < ANNOUNCE_NUM) { |
364 | nclaims++; |
365 | VDBG("announce/%d %s@%s\n", |
366 | nclaims, intf, inet_ntoa(ip)); |
367 | arp(fd, &saddr, ARPOP_REQUEST, |
368 | ð_addr, ip, |
369 | ð_addr, ip); |
370 | timeout = ANNOUNCE_INTERVAL * 1000; |
371 | } |
372 | else { |
373 | // Switch to monitor state. |
374 | state = MONITOR; |
375 | // link is ok to use earlier |
376 | // FIXME update filters |
377 | run(script, "config", intf, &ip); |
378 | ready = 1; |
379 | conflicts = 0; |
380 | timeout = -1; // Never timeout in the monitor state. |
381 | |
382 | // NOTE: all other exit paths |
383 | // should deconfig ... |
384 | if (QUIT) |
385 | return EXIT_SUCCESS; |
386 | } |
387 | break; |
388 | case DEFEND: |
389 | // We won! No ARP replies, so just go back to monitor. |
390 | state = MONITOR; |
391 | timeout = -1; |
392 | conflicts = 0; |
393 | break; |
394 | default: |
395 | // Invalid, should never happen. Restart the whole protocol. |
396 | state = PROBE; |
397 | pick(&ip); |
398 | timeout = 0; |
399 | nprobes = 0; |
400 | nclaims = 0; |
401 | break; |
402 | } // switch (state) |
403 | break; // case 0 (timeout) |
404 | // packets arriving |
405 | case 1: |
406 | // We need to adjust the timeout in case we didn't receive |
407 | // a conflicting packet. |
408 | if (timeout > 0) { |
409 | struct timeval tv2; |
410 | |
411 | gettimeofday(&tv2, NULL); |
412 | if (timercmp(&tv1, &tv2, <)) { |
413 | // Current time is greater than the expected timeout time. |
414 | // Should never happen. |
415 | VDBG("missed an expected timeout\n"); |
416 | timeout = 0; |
417 | } else { |
418 | VDBG("adjusting timeout\n"); |
419 | timersub(&tv1, &tv2, &tv1); |
420 | timeout = 1000 * tv1.tv_sec |
421 | + tv1.tv_usec / 1000; |
422 | } |
423 | } |
424 | |
425 | if ((fds[0].revents & POLLIN) == 0) { |
426 | if (fds[0].revents & POLLERR) { |
427 | // FIXME: links routinely go down; |
428 | // this shouldn't necessarily exit. |
429 | bb_error_msg("%s: poll error", intf); |
430 | if (ready) { |
431 | run(script, "deconfig", |
432 | intf, &ip); |
433 | } |
434 | return EXIT_FAILURE; |
435 | } |
436 | continue; |
437 | } |
438 | |
439 | // read ARP packet |
440 | if (recv(fd, &p, sizeof (p), 0) < 0) { |
441 | why = "recv"; |
442 | goto bad; |
443 | } |
444 | if (p.hdr.ether_type != htons(ETHERTYPE_ARP)) |
445 | continue; |
446 | |
447 | #ifdef DEBUG |
448 | { |
449 | struct ether_addr * sha = (struct ether_addr *) p.arp.arp_sha; |
450 | struct ether_addr * tha = (struct ether_addr *) p.arp.arp_tha; |
451 | struct in_addr * spa = (struct in_addr *) p.arp.arp_spa; |
452 | struct in_addr * tpa = (struct in_addr *) p.arp.arp_tpa; |
453 | VDBG("%s recv arp type=%d, op=%d,\n", |
454 | intf, ntohs(p.hdr.ether_type), |
455 | ntohs(p.arp.arp_op)); |
456 | VDBG("\tsource=%s %s\n", |
457 | ether_ntoa(sha), |
458 | inet_ntoa(*spa)); |
459 | VDBG("\ttarget=%s %s\n", |
460 | ether_ntoa(tha), |
461 | inet_ntoa(*tpa)); |
462 | } |
463 | #endif |
464 | if (p.arp.arp_op != htons(ARPOP_REQUEST) |
465 | && p.arp.arp_op != htons(ARPOP_REPLY)) |
466 | continue; |
467 | |
468 | if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0 && |
469 | memcmp(ð_addr, &p.arp.arp_sha, ETH_ALEN) != 0) { |
470 | source_ip_conflict = 1; |
471 | } |
472 | if (memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0 && |
473 | p.arp.arp_op == htons(ARPOP_REQUEST) && |
474 | memcmp(ð_addr, &p.arp.arp_tha, ETH_ALEN) != 0) { |
475 | target_ip_conflict = 1; |
476 | } |
477 | |
478 | VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n", |
479 | state, source_ip_conflict, target_ip_conflict); |
480 | switch (state) { |
481 | case PROBE: |
482 | case ANNOUNCE: |
483 | // When probing or announcing, check for source IP conflicts |
484 | // and other hosts doing ARP probes (target IP conflicts). |
485 | if (source_ip_conflict || target_ip_conflict) { |
486 | conflicts++; |
487 | if (conflicts >= MAX_CONFLICTS) { |
488 | VDBG("%s ratelimit\n", intf); |
489 | timeout = RATE_LIMIT_INTERVAL * 1000; |
490 | state = RATE_LIMIT_PROBE; |
491 | } |
492 | |
493 | // restart the whole protocol |
494 | pick(&ip); |
495 | timeout = 0; |
496 | nprobes = 0; |
497 | nclaims = 0; |
498 | } |
499 | break; |
500 | case MONITOR: |
501 | // If a conflict, we try to defend with a single ARP probe. |
502 | if (source_ip_conflict) { |
503 | VDBG("monitor conflict -- defending\n"); |
504 | state = DEFEND; |
505 | timeout = DEFEND_INTERVAL * 1000; |
506 | arp(fd, &saddr, |
507 | ARPOP_REQUEST, |
508 | ð_addr, ip, |
509 | ð_addr, ip); |
510 | } |
511 | break; |
512 | case DEFEND: |
513 | // Well, we tried. Start over (on conflict). |
514 | if (source_ip_conflict) { |
515 | state = PROBE; |
516 | VDBG("defend conflict -- starting over\n"); |
517 | ready = 0; |
518 | run(script, "deconfig", intf, &ip); |
519 | |
520 | // restart the whole protocol |
521 | pick(&ip); |
522 | timeout = 0; |
523 | nprobes = 0; |
524 | nclaims = 0; |
525 | } |
526 | break; |
527 | default: |
528 | // Invalid, should never happen. Restart the whole protocol. |
529 | VDBG("invalid state -- starting over\n"); |
530 | state = PROBE; |
531 | pick(&ip); |
532 | timeout = 0; |
533 | nprobes = 0; |
534 | nclaims = 0; |
535 | break; |
536 | } // switch state |
537 | |
538 | break; // case 1 (packets arriving) |
539 | default: |
540 | why = "poll"; |
541 | goto bad; |
542 | } // switch poll |
543 | } |
544 | bad: |
545 | bb_perror_msg("%s, %s", intf, why); |
546 | return EXIT_FAILURE; |
547 | } |