nslookup: set default DNS server again. Hopefully helps with 675
[platform/upstream/busybox.git] / networking / zcip.c
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 /*
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 //usage:#define zcip_trivial_usage
27 //usage:       "[OPTIONS] IFACE SCRIPT"
28 //usage:#define zcip_full_usage "\n\n"
29 //usage:       "Manage a ZeroConf IPv4 link-local address\n"
30 //usage:     "\n        -f              Run in foreground"
31 //usage:     "\n        -q              Quit after obtaining address"
32 //usage:     "\n        -r 169.254.x.x  Request this address first"
33 //usage:     "\n        -v              Verbose"
34 //usage:     "\n"
35 //usage:     "\nWith no -q, runs continuously monitoring for ARP conflicts,"
36 //usage:     "\nexits only on I/O errors (link down etc)"
37
38 #include "libbb.h"
39 #include <netinet/ether.h>
40 #include <net/if.h>
41 #include <net/if_arp.h>
42 #include <linux/sockios.h>
43
44 #include <syslog.h>
45
46 /* We don't need more than 32 bits of the counter */
47 #define MONOTONIC_US() ((unsigned)monotonic_us())
48
49 struct arp_packet {
50         struct ether_header eth;
51         struct ether_arp arp;
52 } PACKED;
53
54 enum {
55 /* 169.254.0.0 */
56         LINKLOCAL_ADDR = 0xa9fe0000,
57
58 /* protocol timeout parameters, specified in seconds */
59         PROBE_WAIT = 1,
60         PROBE_MIN = 1,
61         PROBE_MAX = 2,
62         PROBE_NUM = 3,
63         MAX_CONFLICTS = 10,
64         RATE_LIMIT_INTERVAL = 60,
65         ANNOUNCE_WAIT = 2,
66         ANNOUNCE_NUM = 2,
67         ANNOUNCE_INTERVAL = 2,
68         DEFEND_INTERVAL = 10
69 };
70
71 /* States during the configuration process. */
72 enum {
73         PROBE = 0,
74         RATE_LIMIT_PROBE,
75         ANNOUNCE,
76         MONITOR,
77         DEFEND
78 };
79
80 #define VDBG(...) do { } while (0)
81
82
83 enum {
84         sock_fd = 3
85 };
86
87 struct globals {
88         struct sockaddr saddr;
89         struct ether_addr eth_addr;
90 } FIX_ALIASING;
91 #define G (*(struct globals*)&bb_common_bufsiz1)
92 #define saddr    (G.saddr   )
93 #define eth_addr (G.eth_addr)
94 #define INIT_G() do { } while (0)
95
96
97 /**
98  * Pick a random link local IP address on 169.254/16, except that
99  * the first and last 256 addresses are reserved.
100  */
101 static uint32_t pick(void)
102 {
103         unsigned tmp;
104
105         do {
106                 tmp = rand() & IN_CLASSB_HOST;
107         } while (tmp > (IN_CLASSB_HOST - 0x0200));
108         return htonl((LINKLOCAL_ADDR + 0x0100) + tmp);
109 }
110
111 /**
112  * Broadcast an ARP packet.
113  */
114 static void arp(
115         /* int op, - always ARPOP_REQUEST */
116         /* const struct ether_addr *source_eth, - always &eth_addr */
117                                         struct in_addr source_ip,
118         const struct ether_addr *target_eth, struct in_addr target_ip)
119 {
120         enum { op = ARPOP_REQUEST };
121 #define source_eth (&eth_addr)
122
123         struct arp_packet p;
124         memset(&p, 0, sizeof(p));
125
126         // ether header
127         p.eth.ether_type = htons(ETHERTYPE_ARP);
128         memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
129         memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
130
131         // arp request
132         p.arp.arp_hrd = htons(ARPHRD_ETHER);
133         p.arp.arp_pro = htons(ETHERTYPE_IP);
134         p.arp.arp_hln = ETH_ALEN;
135         p.arp.arp_pln = 4;
136         p.arp.arp_op = htons(op);
137         memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
138         memcpy(&p.arp.arp_spa, &source_ip, sizeof(p.arp.arp_spa));
139         memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
140         memcpy(&p.arp.arp_tpa, &target_ip, sizeof(p.arp.arp_tpa));
141
142         // send it
143         // Even though sock_fd is already bound to saddr, just send()
144         // won't work, because "socket is not connected"
145         // (and connect() won't fix that, "operation not supported").
146         // Thus we sendto() to saddr. I wonder which sockaddr
147         // (from bind() or from sendto()?) kernel actually uses
148         // to determine iface to emit the packet from...
149         xsendto(sock_fd, &p, sizeof(p), &saddr, sizeof(saddr));
150 #undef source_eth
151 }
152
153 /**
154  * Run a script.
155  * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL
156  */
157 static int run(char *argv[3], const char *param, struct in_addr *ip)
158 {
159         int status;
160         char *addr = addr; /* for gcc */
161         const char *fmt = "%s %s %s" + 3;
162
163         argv[2] = (char*)param;
164
165         VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]);
166
167         if (ip) {
168                 addr = inet_ntoa(*ip);
169                 xsetenv("ip", addr);
170                 fmt -= 3;
171         }
172         bb_info_msg(fmt, argv[2], argv[0], addr);
173
174         status = spawn_and_wait(argv + 1);
175         if (status < 0) {
176                 bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]);
177                 return -errno;
178         }
179         if (status != 0)
180                 bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff);
181         return status;
182 }
183
184 /**
185  * Return milliseconds of random delay, up to "secs" seconds.
186  */
187 static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
188 {
189         return rand() % (secs * 1000);
190 }
191
192 /**
193  * main program
194  */
195 int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
196 int zcip_main(int argc UNUSED_PARAM, char **argv)
197 {
198         int state;
199         char *r_opt;
200         unsigned opts;
201
202         // ugly trick, but I want these zeroed in one go
203         struct {
204                 const struct in_addr null_ip;
205                 const struct ether_addr null_addr;
206                 struct in_addr ip;
207                 struct ifreq ifr;
208                 int timeout_ms; /* must be signed */
209                 unsigned conflicts;
210                 unsigned nprobes;
211                 unsigned nclaims;
212                 int ready;
213                 int verbose;
214         } L;
215 #define null_ip    (L.null_ip   )
216 #define null_addr  (L.null_addr )
217 #define ip         (L.ip        )
218 #define ifr        (L.ifr       )
219 #define timeout_ms (L.timeout_ms)
220 #define conflicts  (L.conflicts )
221 #define nprobes    (L.nprobes   )
222 #define nclaims    (L.nclaims   )
223 #define ready      (L.ready     )
224 #define verbose    (L.verbose   )
225
226         memset(&L, 0, sizeof(L));
227         INIT_G();
228
229 #define FOREGROUND (opts & 1)
230 #define QUIT       (opts & 2)
231         // parse commandline: prog [options] ifname script
232         // exactly 2 args; -v accumulates and implies -f
233         opt_complementary = "=2:vv:vf";
234         opts = getopt32(argv, "fqr:v", &r_opt, &verbose);
235 #if !BB_MMU
236         // on NOMMU reexec early (or else we will rerun things twice)
237         if (!FOREGROUND)
238                 bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
239 #endif
240         // open an ARP socket
241         // (need to do it before openlog to prevent openlog from taking
242         // fd 3 (sock_fd==3))
243         xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
244         if (!FOREGROUND) {
245                 // do it before all bb_xx_msg calls
246                 openlog(applet_name, 0, LOG_DAEMON);
247                 logmode |= LOGMODE_SYSLOG;
248         }
249         if (opts & 4) { // -r n.n.n.n
250                 if (inet_aton(r_opt, &ip) == 0
251                  || (ntohl(ip.s_addr) & IN_CLASSB_NET) != LINKLOCAL_ADDR
252                 ) {
253                         bb_error_msg_and_die("invalid link address");
254                 }
255         }
256         argv += optind - 1;
257
258         /* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
259         /* We need to make space for script argument: */
260         argv[0] = argv[1];
261         argv[1] = argv[2];
262         /* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
263 #define argv_intf (argv[0])
264
265         xsetenv("interface", argv_intf);
266
267         // initialize the interface (modprobe, ifup, etc)
268         if (run(argv, "init", NULL))
269                 return EXIT_FAILURE;
270
271         // initialize saddr
272         // saddr is: { u16 sa_family; u8 sa_data[14]; }
273         //memset(&saddr, 0, sizeof(saddr));
274         //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
275         safe_strncpy(saddr.sa_data, argv_intf, sizeof(saddr.sa_data));
276
277         // bind to the interface's ARP socket
278         xbind(sock_fd, &saddr, sizeof(saddr));
279
280         // get the interface's ethernet address
281         //memset(&ifr, 0, sizeof(ifr));
282         strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
283         xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
284         memcpy(&eth_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
285
286         // start with some stable ip address, either a function of
287         // the hardware address or else the last address we used.
288         // we are taking low-order four bytes, as top-order ones
289         // aren't random enough.
290         // NOTE: the sequence of addresses we try changes only
291         // depending on when we detect conflicts.
292         {
293                 uint32_t t;
294                 move_from_unaligned32(t, ((char *)&eth_addr + 2));
295                 srand(t);
296         }
297         if (ip.s_addr == 0)
298                 ip.s_addr = pick();
299
300         // FIXME cases to handle:
301         //  - zcip already running!
302         //  - link already has local address... just defend/update
303
304         // daemonize now; don't delay system startup
305         if (!FOREGROUND) {
306 #if BB_MMU
307                 bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
308 #endif
309                 bb_info_msg("start, interface %s", argv_intf);
310         }
311
312         // run the dynamic address negotiation protocol,
313         // restarting after address conflicts:
314         //  - start with some address we want to try
315         //  - short random delay
316         //  - arp probes to see if another host uses it
317         //  - arp announcements that we're claiming it
318         //  - use it
319         //  - defend it, within limits
320         // exit if:
321         // - address is successfully obtained and -q was given:
322         //   run "<script> config", then exit with exitcode 0
323         // - poll error (when does this happen?)
324         // - read error (when does this happen?)
325         // - sendto error (in arp()) (when does this happen?)
326         // - revents & POLLERR (link down). run "<script> deconfig" first
327         state = PROBE;
328         while (1) {
329                 struct pollfd fds[1];
330                 unsigned deadline_us;
331                 struct arp_packet p;
332                 int source_ip_conflict;
333                 int target_ip_conflict;
334
335                 fds[0].fd = sock_fd;
336                 fds[0].events = POLLIN;
337                 fds[0].revents = 0;
338
339                 // poll, being ready to adjust current timeout
340                 if (!timeout_ms) {
341                         timeout_ms = random_delay_ms(PROBE_WAIT);
342                         // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
343                         // make the kernel filter out all packets except
344                         // ones we'd care about.
345                 }
346                 // set deadline_us to the point in time when we timeout
347                 deadline_us = MONOTONIC_US() + timeout_ms * 1000;
348
349                 VDBG("...wait %d %s nprobes=%u, nclaims=%u\n",
350                                 timeout_ms, argv_intf, nprobes, nclaims);
351
352                 switch (safe_poll(fds, 1, timeout_ms)) {
353
354                 default:
355                         //bb_perror_msg("poll"); - done in safe_poll
356                         return EXIT_FAILURE;
357
358                 // timeout
359                 case 0:
360                         VDBG("state = %d\n", state);
361                         switch (state) {
362                         case PROBE:
363                                 // timeouts in the PROBE state mean no conflicting ARP packets
364                                 // have been received, so we can progress through the states
365                                 if (nprobes < PROBE_NUM) {
366                                         nprobes++;
367                                         VDBG("probe/%u %s@%s\n",
368                                                         nprobes, argv_intf, inet_ntoa(ip));
369                                         arp(/* ARPOP_REQUEST, */
370                                                         /* &eth_addr, */ null_ip,
371                                                         &null_addr, ip);
372                                         timeout_ms = PROBE_MIN * 1000;
373                                         timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
374                                 }
375                                 else {
376                                         // Switch to announce state.
377                                         state = ANNOUNCE;
378                                         nclaims = 0;
379                                         VDBG("announce/%u %s@%s\n",
380                                                         nclaims, argv_intf, inet_ntoa(ip));
381                                         arp(/* ARPOP_REQUEST, */
382                                                         /* &eth_addr, */ ip,
383                                                         &eth_addr, ip);
384                                         timeout_ms = ANNOUNCE_INTERVAL * 1000;
385                                 }
386                                 break;
387                         case RATE_LIMIT_PROBE:
388                                 // timeouts in the RATE_LIMIT_PROBE state mean no conflicting ARP packets
389                                 // have been received, so we can move immediately to the announce state
390                                 state = ANNOUNCE;
391                                 nclaims = 0;
392                                 VDBG("announce/%u %s@%s\n",
393                                                 nclaims, argv_intf, inet_ntoa(ip));
394                                 arp(/* ARPOP_REQUEST, */
395                                                 /* &eth_addr, */ ip,
396                                                 &eth_addr, ip);
397                                 timeout_ms = ANNOUNCE_INTERVAL * 1000;
398                                 break;
399                         case ANNOUNCE:
400                                 // timeouts in the ANNOUNCE state mean no conflicting ARP packets
401                                 // have been received, so we can progress through the states
402                                 if (nclaims < ANNOUNCE_NUM) {
403                                         nclaims++;
404                                         VDBG("announce/%u %s@%s\n",
405                                                         nclaims, argv_intf, inet_ntoa(ip));
406                                         arp(/* ARPOP_REQUEST, */
407                                                         /* &eth_addr, */ ip,
408                                                         &eth_addr, ip);
409                                         timeout_ms = ANNOUNCE_INTERVAL * 1000;
410                                 }
411                                 else {
412                                         // Switch to monitor state.
413                                         state = MONITOR;
414                                         // link is ok to use earlier
415                                         // FIXME update filters
416                                         run(argv, "config", &ip);
417                                         ready = 1;
418                                         conflicts = 0;
419                                         timeout_ms = -1; // Never timeout in the monitor state.
420
421                                         // NOTE: all other exit paths
422                                         // should deconfig ...
423                                         if (QUIT)
424                                                 return EXIT_SUCCESS;
425                                 }
426                                 break;
427                         case DEFEND:
428                                 // We won!  No ARP replies, so just go back to monitor.
429                                 state = MONITOR;
430                                 timeout_ms = -1;
431                                 conflicts = 0;
432                                 break;
433                         default:
434                                 // Invalid, should never happen.  Restart the whole protocol.
435                                 state = PROBE;
436                                 ip.s_addr = pick();
437                                 timeout_ms = 0;
438                                 nprobes = 0;
439                                 nclaims = 0;
440                                 break;
441                         } // switch (state)
442                         break; // case 0 (timeout)
443
444                 // packets arriving, or link went down
445                 case 1:
446                         // We need to adjust the timeout in case we didn't receive
447                         // a conflicting packet.
448                         if (timeout_ms > 0) {
449                                 unsigned diff = deadline_us - MONOTONIC_US();
450                                 if ((int)(diff) < 0) {
451                                         // Current time is greater than the expected timeout time.
452                                         // Should never happen.
453                                         VDBG("missed an expected timeout\n");
454                                         timeout_ms = 0;
455                                 } else {
456                                         VDBG("adjusting timeout\n");
457                                         timeout_ms = (diff / 1000) | 1; /* never 0 */
458                                 }
459                         }
460
461                         if ((fds[0].revents & POLLIN) == 0) {
462                                 if (fds[0].revents & POLLERR) {
463                                         // FIXME: links routinely go down;
464                                         // this shouldn't necessarily exit.
465                                         bb_error_msg("iface %s is down", argv_intf);
466                                         if (ready) {
467                                                 run(argv, "deconfig", &ip);
468                                         }
469                                         return EXIT_FAILURE;
470                                 }
471                                 continue;
472                         }
473
474                         // read ARP packet
475                         if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
476                                 bb_perror_msg_and_die(bb_msg_read_error);
477                         }
478                         if (p.eth.ether_type != htons(ETHERTYPE_ARP))
479                                 continue;
480 #ifdef DEBUG
481                         {
482                                 struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
483                                 struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
484                                 struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
485                                 struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
486                                 VDBG("%s recv arp type=%d, op=%d,\n",
487                                         argv_intf, ntohs(p.eth.ether_type),
488                                         ntohs(p.arp.arp_op));
489                                 VDBG("\tsource=%s %s\n",
490                                         ether_ntoa(sha),
491                                         inet_ntoa(*spa));
492                                 VDBG("\ttarget=%s %s\n",
493                                         ether_ntoa(tha),
494                                         inet_ntoa(*tpa));
495                         }
496 #endif
497                         if (p.arp.arp_op != htons(ARPOP_REQUEST)
498                          && p.arp.arp_op != htons(ARPOP_REPLY))
499                                 continue;
500
501                         source_ip_conflict = 0;
502                         target_ip_conflict = 0;
503
504                         if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0
505                          && memcmp(&p.arp.arp_sha, &eth_addr, ETH_ALEN) != 0
506                         ) {
507                                 source_ip_conflict = 1;
508                         }
509                         if (p.arp.arp_op == htons(ARPOP_REQUEST)
510                          && memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0
511                          && memcmp(&p.arp.arp_tha, &eth_addr, ETH_ALEN) != 0
512                         ) {
513                                 target_ip_conflict = 1;
514                         }
515
516                         VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n",
517                                 state, source_ip_conflict, target_ip_conflict);
518                         switch (state) {
519                         case PROBE:
520                         case ANNOUNCE:
521                                 // When probing or announcing, check for source IP conflicts
522                                 // and other hosts doing ARP probes (target IP conflicts).
523                                 if (source_ip_conflict || target_ip_conflict) {
524                                         conflicts++;
525                                         if (conflicts >= MAX_CONFLICTS) {
526                                                 VDBG("%s ratelimit\n", argv_intf);
527                                                 timeout_ms = RATE_LIMIT_INTERVAL * 1000;
528                                                 state = RATE_LIMIT_PROBE;
529                                         }
530
531                                         // restart the whole protocol
532                                         ip.s_addr = pick();
533                                         timeout_ms = 0;
534                                         nprobes = 0;
535                                         nclaims = 0;
536                                 }
537                                 break;
538                         case MONITOR:
539                                 // If a conflict, we try to defend with a single ARP probe.
540                                 if (source_ip_conflict) {
541                                         VDBG("monitor conflict -- defending\n");
542                                         state = DEFEND;
543                                         timeout_ms = DEFEND_INTERVAL * 1000;
544                                         arp(/* ARPOP_REQUEST, */
545                                                 /* &eth_addr, */ ip,
546                                                 &eth_addr, ip);
547                                 }
548                                 break;
549                         case DEFEND:
550                                 // Well, we tried.  Start over (on conflict).
551                                 if (source_ip_conflict) {
552                                         state = PROBE;
553                                         VDBG("defend conflict -- starting over\n");
554                                         ready = 0;
555                                         run(argv, "deconfig", &ip);
556
557                                         // restart the whole protocol
558                                         ip.s_addr = pick();
559                                         timeout_ms = 0;
560                                         nprobes = 0;
561                                         nclaims = 0;
562                                 }
563                                 break;
564                         default:
565                                 // Invalid, should never happen.  Restart the whole protocol.
566                                 VDBG("invalid state -- starting over\n");
567                                 state = PROBE;
568                                 ip.s_addr = pick();
569                                 timeout_ms = 0;
570                                 nprobes = 0;
571                                 nclaims = 0;
572                                 break;
573                         } // switch state
574                         break; // case 1 (packets arriving)
575                 } // switch poll
576         } // while (1)
577 #undef argv_intf
578 }