Contents of /tags/mkinitrd-6_3_2/busybox/networking/zcip.c
Parent Directory | Revision Log
Revision 1164 -
(show annotations)
(download)
Tue Sep 14 20:33:28 2010 UTC (14 years ago) by niro
File MIME type: text/plain
File size: 15413 byte(s)
Tue Sep 14 20:33:28 2010 UTC (14 years ago) by niro
File MIME type: text/plain
File size: 15413 byte(s)
tagged 'mkinitrd-6_3_2'
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 <netinet/ether.h> |
27 | #include <net/ethernet.h> |
28 | #include <net/if.h> |
29 | #include <net/if_arp.h> |
30 | #include <linux/if_packet.h> |
31 | #include <linux/sockios.h> |
32 | |
33 | #include "libbb.h" |
34 | #include <syslog.h> |
35 | |
36 | /* We don't need more than 32 bits of the counter */ |
37 | #define MONOTONIC_US() ((unsigned)monotonic_us()) |
38 | |
39 | struct arp_packet { |
40 | struct ether_header eth; |
41 | struct ether_arp arp; |
42 | } 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(...) do { } while (0) |
71 | |
72 | |
73 | enum { |
74 | sock_fd = 3 |
75 | }; |
76 | |
77 | struct globals { |
78 | struct sockaddr saddr; |
79 | struct ether_addr eth_addr; |
80 | } FIX_ALIASING; |
81 | #define G (*(struct globals*)&bb_common_bufsiz1) |
82 | #define saddr (G.saddr ) |
83 | #define eth_addr (G.eth_addr) |
84 | |
85 | |
86 | /** |
87 | * Pick a random link local IP address on 169.254/16, except that |
88 | * the first and last 256 addresses are reserved. |
89 | */ |
90 | static uint32_t pick(void) |
91 | { |
92 | unsigned tmp; |
93 | |
94 | do { |
95 | tmp = rand() & IN_CLASSB_HOST; |
96 | } while (tmp > (IN_CLASSB_HOST - 0x0200)); |
97 | return htonl((LINKLOCAL_ADDR + 0x0100) + tmp); |
98 | } |
99 | |
100 | /** |
101 | * Broadcast an ARP packet. |
102 | */ |
103 | static void arp( |
104 | /* int op, - always ARPOP_REQUEST */ |
105 | /* const struct ether_addr *source_eth, - always ð_addr */ |
106 | struct in_addr source_ip, |
107 | const struct ether_addr *target_eth, struct in_addr target_ip) |
108 | { |
109 | enum { op = ARPOP_REQUEST }; |
110 | #define source_eth (ð_addr) |
111 | |
112 | struct arp_packet p; |
113 | memset(&p, 0, sizeof(p)); |
114 | |
115 | // ether header |
116 | p.eth.ether_type = htons(ETHERTYPE_ARP); |
117 | memcpy(p.eth.ether_shost, source_eth, ETH_ALEN); |
118 | memset(p.eth.ether_dhost, 0xff, ETH_ALEN); |
119 | |
120 | // arp request |
121 | p.arp.arp_hrd = htons(ARPHRD_ETHER); |
122 | p.arp.arp_pro = htons(ETHERTYPE_IP); |
123 | p.arp.arp_hln = ETH_ALEN; |
124 | p.arp.arp_pln = 4; |
125 | p.arp.arp_op = htons(op); |
126 | memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN); |
127 | memcpy(&p.arp.arp_spa, &source_ip, sizeof(p.arp.arp_spa)); |
128 | memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN); |
129 | memcpy(&p.arp.arp_tpa, &target_ip, sizeof(p.arp.arp_tpa)); |
130 | |
131 | // send it |
132 | // Even though sock_fd is already bound to saddr, just send() |
133 | // won't work, because "socket is not connected" |
134 | // (and connect() won't fix that, "operation not supported"). |
135 | // Thus we sendto() to saddr. I wonder which sockaddr |
136 | // (from bind() or from sendto()?) kernel actually uses |
137 | // to determine iface to emit the packet from... |
138 | xsendto(sock_fd, &p, sizeof(p), &saddr, sizeof(saddr)); |
139 | #undef source_eth |
140 | } |
141 | |
142 | /** |
143 | * Run a script. |
144 | * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL |
145 | */ |
146 | static int run(char *argv[3], const char *param, struct in_addr *ip) |
147 | { |
148 | int status; |
149 | char *addr = addr; /* for gcc */ |
150 | const char *fmt = "%s %s %s" + 3; |
151 | |
152 | argv[2] = (char*)param; |
153 | |
154 | VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]); |
155 | |
156 | if (ip) { |
157 | addr = inet_ntoa(*ip); |
158 | xsetenv("ip", addr); |
159 | fmt -= 3; |
160 | } |
161 | bb_info_msg(fmt, argv[2], argv[0], addr); |
162 | |
163 | status = spawn_and_wait(argv + 1); |
164 | if (status < 0) { |
165 | bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]); |
166 | return -errno; |
167 | } |
168 | if (status != 0) |
169 | bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff); |
170 | return status; |
171 | } |
172 | |
173 | /** |
174 | * Return milliseconds of random delay, up to "secs" seconds. |
175 | */ |
176 | static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs) |
177 | { |
178 | return rand() % (secs * 1000); |
179 | } |
180 | |
181 | /** |
182 | * main program |
183 | */ |
184 | int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE; |
185 | int zcip_main(int argc UNUSED_PARAM, char **argv) |
186 | { |
187 | int state; |
188 | char *r_opt; |
189 | unsigned opts; |
190 | |
191 | // ugly trick, but I want these zeroed in one go |
192 | struct { |
193 | const struct in_addr null_ip; |
194 | const struct ether_addr null_addr; |
195 | struct in_addr ip; |
196 | struct ifreq ifr; |
197 | int timeout_ms; /* must be signed */ |
198 | unsigned conflicts; |
199 | unsigned nprobes; |
200 | unsigned nclaims; |
201 | int ready; |
202 | int verbose; |
203 | } L; |
204 | #define null_ip (L.null_ip ) |
205 | #define null_addr (L.null_addr ) |
206 | #define ip (L.ip ) |
207 | #define ifr (L.ifr ) |
208 | #define timeout_ms (L.timeout_ms) |
209 | #define conflicts (L.conflicts ) |
210 | #define nprobes (L.nprobes ) |
211 | #define nclaims (L.nclaims ) |
212 | #define ready (L.ready ) |
213 | #define verbose (L.verbose ) |
214 | |
215 | memset(&L, 0, sizeof(L)); |
216 | |
217 | #define FOREGROUND (opts & 1) |
218 | #define QUIT (opts & 2) |
219 | // parse commandline: prog [options] ifname script |
220 | // exactly 2 args; -v accumulates and implies -f |
221 | opt_complementary = "=2:vv:vf"; |
222 | opts = getopt32(argv, "fqr:v", &r_opt, &verbose); |
223 | #if !BB_MMU |
224 | // on NOMMU reexec early (or else we will rerun things twice) |
225 | if (!FOREGROUND) |
226 | bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv); |
227 | #endif |
228 | // open an ARP socket |
229 | // (need to do it before openlog to prevent openlog from taking |
230 | // fd 3 (sock_fd==3)) |
231 | xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd); |
232 | if (!FOREGROUND) { |
233 | // do it before all bb_xx_msg calls |
234 | openlog(applet_name, 0, LOG_DAEMON); |
235 | logmode |= LOGMODE_SYSLOG; |
236 | } |
237 | if (opts & 4) { // -r n.n.n.n |
238 | if (inet_aton(r_opt, &ip) == 0 |
239 | || (ntohl(ip.s_addr) & IN_CLASSB_NET) != LINKLOCAL_ADDR |
240 | ) { |
241 | bb_error_msg_and_die("invalid link address"); |
242 | } |
243 | } |
244 | argv += optind - 1; |
245 | |
246 | /* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */ |
247 | /* We need to make space for script argument: */ |
248 | argv[0] = argv[1]; |
249 | argv[1] = argv[2]; |
250 | /* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */ |
251 | #define argv_intf (argv[0]) |
252 | |
253 | xsetenv("interface", argv_intf); |
254 | |
255 | // initialize the interface (modprobe, ifup, etc) |
256 | if (run(argv, "init", NULL)) |
257 | return EXIT_FAILURE; |
258 | |
259 | // initialize saddr |
260 | // saddr is: { u16 sa_family; u8 sa_data[14]; } |
261 | //memset(&saddr, 0, sizeof(saddr)); |
262 | //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?! |
263 | safe_strncpy(saddr.sa_data, argv_intf, sizeof(saddr.sa_data)); |
264 | |
265 | // bind to the interface's ARP socket |
266 | xbind(sock_fd, &saddr, sizeof(saddr)); |
267 | |
268 | // get the interface's ethernet address |
269 | //memset(&ifr, 0, sizeof(ifr)); |
270 | strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf); |
271 | xioctl(sock_fd, SIOCGIFHWADDR, &ifr); |
272 | memcpy(ð_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 | // we are taking low-order four bytes, as top-order ones |
277 | // aren't random enough. |
278 | // NOTE: the sequence of addresses we try changes only |
279 | // depending on when we detect conflicts. |
280 | { |
281 | uint32_t t; |
282 | move_from_unaligned32(t, ((char *)ð_addr + 2)); |
283 | srand(t); |
284 | } |
285 | if (ip.s_addr == 0) |
286 | ip.s_addr = pick(); |
287 | |
288 | // FIXME cases to handle: |
289 | // - zcip already running! |
290 | // - link already has local address... just defend/update |
291 | |
292 | // daemonize now; don't delay system startup |
293 | if (!FOREGROUND) { |
294 | #if BB_MMU |
295 | bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/); |
296 | #endif |
297 | bb_info_msg("start, interface %s", argv_intf); |
298 | } |
299 | |
300 | // run the dynamic address negotiation protocol, |
301 | // restarting after address conflicts: |
302 | // - start with some address we want to try |
303 | // - short random delay |
304 | // - arp probes to see if another host uses it |
305 | // - arp announcements that we're claiming it |
306 | // - use it |
307 | // - defend it, within limits |
308 | // exit if: |
309 | // - address is successfully obtained and -q was given: |
310 | // run "<script> config", then exit with exitcode 0 |
311 | // - poll error (when does this happen?) |
312 | // - read error (when does this happen?) |
313 | // - sendto error (in arp()) (when does this happen?) |
314 | // - revents & POLLERR (link down). run "<script> deconfig" first |
315 | state = PROBE; |
316 | while (1) { |
317 | struct pollfd fds[1]; |
318 | unsigned deadline_us; |
319 | struct arp_packet p; |
320 | int source_ip_conflict; |
321 | int target_ip_conflict; |
322 | |
323 | fds[0].fd = sock_fd; |
324 | fds[0].events = POLLIN; |
325 | fds[0].revents = 0; |
326 | |
327 | // poll, being ready to adjust current timeout |
328 | if (!timeout_ms) { |
329 | timeout_ms = random_delay_ms(PROBE_WAIT); |
330 | // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to |
331 | // make the kernel filter out all packets except |
332 | // ones we'd care about. |
333 | } |
334 | // set deadline_us to the point in time when we timeout |
335 | deadline_us = MONOTONIC_US() + timeout_ms * 1000; |
336 | |
337 | VDBG("...wait %d %s nprobes=%u, nclaims=%u\n", |
338 | timeout_ms, argv_intf, nprobes, nclaims); |
339 | |
340 | switch (safe_poll(fds, 1, timeout_ms)) { |
341 | |
342 | default: |
343 | //bb_perror_msg("poll"); - done in safe_poll |
344 | return EXIT_FAILURE; |
345 | |
346 | // timeout |
347 | case 0: |
348 | VDBG("state = %d\n", state); |
349 | switch (state) { |
350 | case PROBE: |
351 | // timeouts in the PROBE state mean no conflicting ARP packets |
352 | // have been received, so we can progress through the states |
353 | if (nprobes < PROBE_NUM) { |
354 | nprobes++; |
355 | VDBG("probe/%u %s@%s\n", |
356 | nprobes, argv_intf, inet_ntoa(ip)); |
357 | arp(/* ARPOP_REQUEST, */ |
358 | /* ð_addr, */ null_ip, |
359 | &null_addr, ip); |
360 | timeout_ms = PROBE_MIN * 1000; |
361 | timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN); |
362 | } |
363 | else { |
364 | // Switch to announce state. |
365 | state = ANNOUNCE; |
366 | nclaims = 0; |
367 | VDBG("announce/%u %s@%s\n", |
368 | nclaims, argv_intf, inet_ntoa(ip)); |
369 | arp(/* ARPOP_REQUEST, */ |
370 | /* ð_addr, */ ip, |
371 | ð_addr, ip); |
372 | timeout_ms = ANNOUNCE_INTERVAL * 1000; |
373 | } |
374 | break; |
375 | case RATE_LIMIT_PROBE: |
376 | // timeouts in the RATE_LIMIT_PROBE state mean no conflicting ARP packets |
377 | // have been received, so we can move immediately to the announce state |
378 | state = ANNOUNCE; |
379 | nclaims = 0; |
380 | VDBG("announce/%u %s@%s\n", |
381 | nclaims, argv_intf, inet_ntoa(ip)); |
382 | arp(/* ARPOP_REQUEST, */ |
383 | /* ð_addr, */ ip, |
384 | ð_addr, ip); |
385 | timeout_ms = ANNOUNCE_INTERVAL * 1000; |
386 | break; |
387 | case ANNOUNCE: |
388 | // timeouts in the ANNOUNCE state mean no conflicting ARP packets |
389 | // have been received, so we can progress through the states |
390 | if (nclaims < ANNOUNCE_NUM) { |
391 | nclaims++; |
392 | VDBG("announce/%u %s@%s\n", |
393 | nclaims, argv_intf, inet_ntoa(ip)); |
394 | arp(/* ARPOP_REQUEST, */ |
395 | /* ð_addr, */ ip, |
396 | ð_addr, ip); |
397 | timeout_ms = ANNOUNCE_INTERVAL * 1000; |
398 | } |
399 | else { |
400 | // Switch to monitor state. |
401 | state = MONITOR; |
402 | // link is ok to use earlier |
403 | // FIXME update filters |
404 | run(argv, "config", &ip); |
405 | ready = 1; |
406 | conflicts = 0; |
407 | timeout_ms = -1; // Never timeout in the monitor state. |
408 | |
409 | // NOTE: all other exit paths |
410 | // should deconfig ... |
411 | if (QUIT) |
412 | return EXIT_SUCCESS; |
413 | } |
414 | break; |
415 | case DEFEND: |
416 | // We won! No ARP replies, so just go back to monitor. |
417 | state = MONITOR; |
418 | timeout_ms = -1; |
419 | conflicts = 0; |
420 | break; |
421 | default: |
422 | // Invalid, should never happen. Restart the whole protocol. |
423 | state = PROBE; |
424 | ip.s_addr = pick(); |
425 | timeout_ms = 0; |
426 | nprobes = 0; |
427 | nclaims = 0; |
428 | break; |
429 | } // switch (state) |
430 | break; // case 0 (timeout) |
431 | |
432 | // packets arriving, or link went down |
433 | case 1: |
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. |
440 | // Should never happen. |
441 | VDBG("missed an expected timeout\n"); |
442 | timeout_ms = 0; |
443 | } else { |
444 | VDBG("adjusting timeout\n"); |
445 | timeout_ms = (diff / 1000) | 1; /* never 0 */ |
446 | } |
447 | } |
448 | |
449 | if ((fds[0].revents & POLLIN) == 0) { |
450 | if (fds[0].revents & POLLERR) { |
451 | // FIXME: links routinely go down; |
452 | // this shouldn't necessarily exit. |
453 | bb_error_msg("iface %s is down", argv_intf); |
454 | if (ready) { |
455 | run(argv, "deconfig", &ip); |
456 | } |
457 | return EXIT_FAILURE; |
458 | } |
459 | continue; |
460 | } |
461 | |
462 | // read ARP packet |
463 | if (safe_read(sock_fd, &p, sizeof(p)) < 0) { |
464 | bb_perror_msg_and_die(bb_msg_read_error); |
465 | } |
466 | if (p.eth.ether_type != htons(ETHERTYPE_ARP)) |
467 | continue; |
468 | #ifdef DEBUG |
469 | { |
470 | struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha; |
471 | struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha; |
472 | struct in_addr *spa = (struct in_addr *) p.arp.arp_spa; |
473 | struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa; |
474 | VDBG("%s recv arp type=%d, op=%d,\n", |
475 | argv_intf, ntohs(p.eth.ether_type), |
476 | ntohs(p.arp.arp_op)); |
477 | VDBG("\tsource=%s %s\n", |
478 | ether_ntoa(sha), |
479 | inet_ntoa(*spa)); |
480 | VDBG("\ttarget=%s %s\n", |
481 | ether_ntoa(tha), |
482 | inet_ntoa(*tpa)); |
483 | } |
484 | #endif |
485 | if (p.arp.arp_op != htons(ARPOP_REQUEST) |
486 | && p.arp.arp_op != htons(ARPOP_REPLY)) |
487 | continue; |
488 | |
489 | source_ip_conflict = 0; |
490 | target_ip_conflict = 0; |
491 | |
492 | if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0 |
493 | && memcmp(&p.arp.arp_sha, ð_addr, ETH_ALEN) != 0 |
494 | ) { |
495 | source_ip_conflict = 1; |
496 | } |
497 | if (p.arp.arp_op == htons(ARPOP_REQUEST) |
498 | && memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0 |
499 | && memcmp(&p.arp.arp_tha, ð_addr, ETH_ALEN) != 0 |
500 | ) { |
501 | target_ip_conflict = 1; |
502 | } |
503 | |
504 | VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n", |
505 | state, source_ip_conflict, target_ip_conflict); |
506 | switch (state) { |
507 | case PROBE: |
508 | case ANNOUNCE: |
509 | // When probing or announcing, check for source IP conflicts |
510 | // and other hosts doing ARP probes (target IP conflicts). |
511 | if (source_ip_conflict || target_ip_conflict) { |
512 | conflicts++; |
513 | if (conflicts >= MAX_CONFLICTS) { |
514 | VDBG("%s ratelimit\n", argv_intf); |
515 | timeout_ms = RATE_LIMIT_INTERVAL * 1000; |
516 | state = RATE_LIMIT_PROBE; |
517 | } |
518 | |
519 | // restart the whole protocol |
520 | ip.s_addr = pick(); |
521 | timeout_ms = 0; |
522 | nprobes = 0; |
523 | nclaims = 0; |
524 | } |
525 | break; |
526 | case MONITOR: |
527 | // If a conflict, we try to defend with a single ARP probe. |
528 | if (source_ip_conflict) { |
529 | VDBG("monitor conflict -- defending\n"); |
530 | state = DEFEND; |
531 | timeout_ms = DEFEND_INTERVAL * 1000; |
532 | arp(/* ARPOP_REQUEST, */ |
533 | /* ð_addr, */ ip, |
534 | ð_addr, ip); |
535 | } |
536 | break; |
537 | case DEFEND: |
538 | // Well, we tried. Start over (on conflict). |
539 | if (source_ip_conflict) { |
540 | state = PROBE; |
541 | VDBG("defend conflict -- starting over\n"); |
542 | ready = 0; |
543 | run(argv, "deconfig", &ip); |
544 | |
545 | // restart the whole protocol |
546 | ip.s_addr = pick(); |
547 | timeout_ms = 0; |
548 | nprobes = 0; |
549 | nclaims = 0; |
550 | } |
551 | break; |
552 | default: |
553 | // Invalid, should never happen. Restart the whole protocol. |
554 | VDBG("invalid state -- starting over\n"); |
555 | state = PROBE; |
556 | ip.s_addr = pick(); |
557 | timeout_ms = 0; |
558 | nprobes = 0; |
559 | nclaims = 0; |
560 | break; |
561 | } // switch state |
562 | break; // case 1 (packets arriving) |
563 | } // switch poll |
564 | } // while (1) |
565 | #undef argv_intf |
566 | } |