zcip: allow our own class B range to be used for ZeroConf
[oweals/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        -l x.x.0.0      Use this range instead of 169.254"
34 //usage:     "\n        -v              Verbose"
35 //usage:     "\n"
36 //usage:     "\nWith no -q, runs continuously monitoring for ARP conflicts,"
37 //usage:     "\nexits only on I/O errors (link down etc)"
38
39 #include "libbb.h"
40 #include <netinet/ether.h>
41 #include <net/if.h>
42 #include <net/if_arp.h>
43 #include <linux/sockios.h>
44
45 #include <syslog.h>
46
47 /* We don't need more than 32 bits of the counter */
48 #define MONOTONIC_US() ((unsigned)monotonic_us())
49
50 struct arp_packet {
51         struct ether_header eth;
52         struct ether_arp arp;
53 } PACKED;
54
55 enum {
56 /* 169.254.0.0 */
57         LINKLOCAL_ADDR = 0xa9fe0000,
58
59 /* protocol timeout parameters, specified in seconds */
60         PROBE_WAIT = 1,
61         PROBE_MIN = 1,
62         PROBE_MAX = 2,
63         PROBE_NUM = 3,
64         MAX_CONFLICTS = 10,
65         RATE_LIMIT_INTERVAL = 60,
66         ANNOUNCE_WAIT = 2,
67         ANNOUNCE_NUM = 2,
68         ANNOUNCE_INTERVAL = 2,
69         DEFEND_INTERVAL = 10
70 };
71
72 /* States during the configuration process. */
73 enum {
74         PROBE = 0,
75         RATE_LIMIT_PROBE,
76         ANNOUNCE,
77         MONITOR,
78         DEFEND
79 };
80
81 #define VDBG(...) do { } while (0)
82
83
84 enum {
85         sock_fd = 3
86 };
87
88 struct globals {
89         struct sockaddr saddr;
90         struct ether_addr eth_addr;
91         uint32_t localnet_ip;
92 } FIX_ALIASING;
93 #define G (*(struct globals*)&bb_common_bufsiz1)
94 #define saddr    (G.saddr   )
95 #define eth_addr (G.eth_addr)
96 #define INIT_G() do { } while (0)
97
98
99 /**
100  * Pick a random link local IP address on 169.254/16, except that
101  * the first and last 256 addresses are reserved.
102  */
103 static uint32_t pick_nip(void)
104 {
105         unsigned tmp;
106
107         do {
108                 tmp = rand() & IN_CLASSB_HOST;
109         } while (tmp > (IN_CLASSB_HOST - 0x0200));
110         return htonl((G.localnet_ip + 0x0100) + tmp);
111 }
112
113 /**
114  * Broadcast an ARP packet.
115  */
116 static void arp(
117         /* int op, - always ARPOP_REQUEST */
118         /* const struct ether_addr *source_eth, - always &eth_addr */
119                                         struct in_addr source_ip,
120         const struct ether_addr *target_eth, struct in_addr target_ip)
121 {
122         enum { op = ARPOP_REQUEST };
123 #define source_eth (&eth_addr)
124
125         struct arp_packet p;
126         memset(&p, 0, sizeof(p));
127
128         // ether header
129         p.eth.ether_type = htons(ETHERTYPE_ARP);
130         memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
131         memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
132
133         // arp request
134         p.arp.arp_hrd = htons(ARPHRD_ETHER);
135         p.arp.arp_pro = htons(ETHERTYPE_IP);
136         p.arp.arp_hln = ETH_ALEN;
137         p.arp.arp_pln = 4;
138         p.arp.arp_op = htons(op);
139         memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
140         memcpy(&p.arp.arp_spa, &source_ip, sizeof(p.arp.arp_spa));
141         memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
142         memcpy(&p.arp.arp_tpa, &target_ip, sizeof(p.arp.arp_tpa));
143
144         // send it
145         // Even though sock_fd is already bound to saddr, just send()
146         // won't work, because "socket is not connected"
147         // (and connect() won't fix that, "operation not supported").
148         // Thus we sendto() to saddr. I wonder which sockaddr
149         // (from bind() or from sendto()?) kernel actually uses
150         // to determine iface to emit the packet from...
151         xsendto(sock_fd, &p, sizeof(p), &saddr, sizeof(saddr));
152 #undef source_eth
153 }
154
155 /**
156  * Run a script.
157  * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL
158  */
159 static int run(char *argv[3], const char *param, struct in_addr *ip)
160 {
161         int status;
162         char *addr = addr; /* for gcc */
163         const char *fmt = "%s %s %s" + 3;
164
165         argv[2] = (char*)param;
166
167         VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]);
168
169         if (ip) {
170                 addr = inet_ntoa(*ip);
171                 xsetenv("ip", addr);
172                 fmt -= 3;
173         }
174         bb_info_msg(fmt, argv[2], argv[0], addr);
175
176         status = spawn_and_wait(argv + 1);
177         if (status < 0) {
178                 bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]);
179                 return -errno;
180         }
181         if (status != 0)
182                 bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff);
183         return status;
184 }
185
186 /**
187  * Return milliseconds of random delay, up to "secs" seconds.
188  */
189 static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
190 {
191         return rand() % (secs * 1000);
192 }
193
194 /**
195  * main program
196  */
197 int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
198 int zcip_main(int argc UNUSED_PARAM, char **argv)
199 {
200         int state;
201         char *r_opt;
202         const char *l_opt = "169.254.0.0";
203         unsigned opts;
204
205         // ugly trick, but I want these zeroed in one go
206         struct {
207                 const struct in_addr null_ip;
208                 const struct ether_addr null_addr;
209                 struct in_addr ip;
210                 struct ifreq ifr;
211                 int timeout_ms; /* must be signed */
212                 unsigned conflicts;
213                 unsigned nprobes;
214                 unsigned nclaims;
215                 int ready;
216                 int verbose;
217         } L;
218 #define null_ip    (L.null_ip   )
219 #define null_addr  (L.null_addr )
220 #define ip         (L.ip        )
221 #define ifr        (L.ifr       )
222 #define timeout_ms (L.timeout_ms)
223 #define conflicts  (L.conflicts )
224 #define nprobes    (L.nprobes   )
225 #define nclaims    (L.nclaims   )
226 #define ready      (L.ready     )
227 #define verbose    (L.verbose   )
228
229         memset(&L, 0, sizeof(L));
230         INIT_G();
231
232 #define FOREGROUND (opts & 1)
233 #define QUIT       (opts & 2)
234         // parse commandline: prog [options] ifname script
235         // exactly 2 args; -v accumulates and implies -f
236         opt_complementary = "=2:vv:vf";
237         opts = getopt32(argv, "fqr:l:v", &r_opt, &l_opt, &verbose);
238 #if !BB_MMU
239         // on NOMMU reexec early (or else we will rerun things twice)
240         if (!FOREGROUND)
241                 bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
242 #endif
243         // open an ARP socket
244         // (need to do it before openlog to prevent openlog from taking
245         // fd 3 (sock_fd==3))
246         xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
247         if (!FOREGROUND) {
248                 // do it before all bb_xx_msg calls
249                 openlog(applet_name, 0, LOG_DAEMON);
250                 logmode |= LOGMODE_SYSLOG;
251         }
252         { // -l n.n.n.n
253                 struct in_addr net;
254                 if (inet_aton(l_opt, &net) == 0
255                  || (net.s_addr & htonl(IN_CLASSB_NET)) != net.s_addr
256                 ) {
257                         bb_error_msg_and_die("invalid network address");
258                 }
259                 G.localnet_ip = ntohl(net.s_addr);
260         }
261         if (opts & 4) { // -r n.n.n.n
262                 if (inet_aton(r_opt, &ip) == 0
263                  || (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
264                 ) {
265                         bb_error_msg_and_die("invalid link address");
266                 }
267         }
268         argv += optind - 1;
269
270         /* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
271         /* We need to make space for script argument: */
272         argv[0] = argv[1];
273         argv[1] = argv[2];
274         /* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
275 #define argv_intf (argv[0])
276
277         xsetenv("interface", argv_intf);
278
279         // initialize the interface (modprobe, ifup, etc)
280         if (run(argv, "init", NULL))
281                 return EXIT_FAILURE;
282
283         // initialize saddr
284         // saddr is: { u16 sa_family; u8 sa_data[14]; }
285         //memset(&saddr, 0, sizeof(saddr));
286         //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
287         safe_strncpy(saddr.sa_data, argv_intf, sizeof(saddr.sa_data));
288
289         // bind to the interface's ARP socket
290         xbind(sock_fd, &saddr, sizeof(saddr));
291
292         // get the interface's ethernet address
293         //memset(&ifr, 0, sizeof(ifr));
294         strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
295         xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
296         memcpy(&eth_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
297
298         // start with some stable ip address, either a function of
299         // the hardware address or else the last address we used.
300         // we are taking low-order four bytes, as top-order ones
301         // aren't random enough.
302         // NOTE: the sequence of addresses we try changes only
303         // depending on when we detect conflicts.
304         {
305                 uint32_t t;
306                 move_from_unaligned32(t, ((char *)&eth_addr + 2));
307                 srand(t);
308         }
309         if (ip.s_addr == 0)
310                 ip.s_addr = pick_nip();
311
312         // FIXME cases to handle:
313         //  - zcip already running!
314         //  - link already has local address... just defend/update
315
316         // daemonize now; don't delay system startup
317         if (!FOREGROUND) {
318 #if BB_MMU
319                 bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
320 #endif
321                 bb_info_msg("start, interface %s", argv_intf);
322         }
323
324         // run the dynamic address negotiation protocol,
325         // restarting after address conflicts:
326         //  - start with some address we want to try
327         //  - short random delay
328         //  - arp probes to see if another host uses it
329         //  - arp announcements that we're claiming it
330         //  - use it
331         //  - defend it, within limits
332         // exit if:
333         // - address is successfully obtained and -q was given:
334         //   run "<script> config", then exit with exitcode 0
335         // - poll error (when does this happen?)
336         // - read error (when does this happen?)
337         // - sendto error (in arp()) (when does this happen?)
338         // - revents & POLLERR (link down). run "<script> deconfig" first
339         state = PROBE;
340         while (1) {
341                 struct pollfd fds[1];
342                 unsigned deadline_us;
343                 struct arp_packet p;
344                 int source_ip_conflict;
345                 int target_ip_conflict;
346
347                 fds[0].fd = sock_fd;
348                 fds[0].events = POLLIN;
349                 fds[0].revents = 0;
350
351                 // poll, being ready to adjust current timeout
352                 if (!timeout_ms) {
353                         timeout_ms = random_delay_ms(PROBE_WAIT);
354                         // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
355                         // make the kernel filter out all packets except
356                         // ones we'd care about.
357                 }
358                 // set deadline_us to the point in time when we timeout
359                 deadline_us = MONOTONIC_US() + timeout_ms * 1000;
360
361                 VDBG("...wait %d %s nprobes=%u, nclaims=%u\n",
362                                 timeout_ms, argv_intf, nprobes, nclaims);
363
364                 switch (safe_poll(fds, 1, timeout_ms)) {
365
366                 default:
367                         //bb_perror_msg("poll"); - done in safe_poll
368                         return EXIT_FAILURE;
369
370                 // timeout
371                 case 0:
372                         VDBG("state = %d\n", state);
373                         switch (state) {
374                         case PROBE:
375                                 // timeouts in the PROBE state mean no conflicting ARP packets
376                                 // have been received, so we can progress through the states
377                                 if (nprobes < PROBE_NUM) {
378                                         nprobes++;
379                                         VDBG("probe/%u %s@%s\n",
380                                                         nprobes, argv_intf, inet_ntoa(ip));
381                                         timeout_ms = PROBE_MIN * 1000;
382                                         timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
383                                         arp(/* ARPOP_REQUEST, */
384                                                         /* &eth_addr, */ null_ip,
385                                                         &null_addr, ip);
386                                 }
387                                 else {
388                                         // Switch to announce state.
389                                         state = ANNOUNCE;
390                                         nclaims = 0;
391                                         VDBG("announce/%u %s@%s\n",
392                                                         nclaims, argv_intf, inet_ntoa(ip));
393                                         timeout_ms = ANNOUNCE_INTERVAL * 1000;
394                                         arp(/* ARPOP_REQUEST, */
395                                                         /* &eth_addr, */ ip,
396                                                         &eth_addr, ip);
397                                 }
398                                 break;
399                         case RATE_LIMIT_PROBE:
400                                 // timeouts in the RATE_LIMIT_PROBE state mean no conflicting ARP packets
401                                 // have been received, so we can move immediately to the announce state
402                                 state = ANNOUNCE;
403                                 nclaims = 0;
404                                 VDBG("announce/%u %s@%s\n",
405                                                 nclaims, argv_intf, inet_ntoa(ip));
406                                 timeout_ms = ANNOUNCE_INTERVAL * 1000;
407                                 arp(/* ARPOP_REQUEST, */
408                                                 /* &eth_addr, */ ip,
409                                                 &eth_addr, ip);
410                                 break;
411                         case ANNOUNCE:
412                                 // timeouts in the ANNOUNCE state mean no conflicting ARP packets
413                                 // have been received, so we can progress through the states
414                                 if (nclaims < ANNOUNCE_NUM) {
415                                         nclaims++;
416                                         VDBG("announce/%u %s@%s\n",
417                                                         nclaims, argv_intf, inet_ntoa(ip));
418                                         timeout_ms = ANNOUNCE_INTERVAL * 1000;
419                                         arp(/* ARPOP_REQUEST, */
420                                                         /* &eth_addr, */ ip,
421                                                         &eth_addr, ip);
422                                 }
423                                 else {
424                                         // Switch to monitor state.
425                                         state = MONITOR;
426                                         // link is ok to use earlier
427                                         // FIXME update filters
428                                         run(argv, "config", &ip);
429                                         ready = 1;
430                                         conflicts = 0;
431                                         timeout_ms = -1; // Never timeout in the monitor state.
432
433                                         // NOTE: all other exit paths
434                                         // should deconfig ...
435                                         if (QUIT)
436                                                 return EXIT_SUCCESS;
437                                 }
438                                 break;
439                         case DEFEND:
440                                 // We won!  No ARP replies, so just go back to monitor.
441                                 state = MONITOR;
442                                 timeout_ms = -1;
443                                 conflicts = 0;
444                                 break;
445                         default:
446                                 // Invalid, should never happen.  Restart the whole protocol.
447                                 state = PROBE;
448                                 ip.s_addr = pick_nip();
449                                 timeout_ms = 0;
450                                 nprobes = 0;
451                                 nclaims = 0;
452                                 break;
453                         } // switch (state)
454                         break; // case 0 (timeout)
455
456                 // packets arriving, or link went down
457                 case 1:
458                         // We need to adjust the timeout in case we didn't receive
459                         // a conflicting packet.
460                         if (timeout_ms > 0) {
461                                 unsigned diff = deadline_us - MONOTONIC_US();
462                                 if ((int)(diff) < 0) {
463                                         // Current time is greater than the expected timeout time.
464                                         // Should never happen.
465                                         VDBG("missed an expected timeout\n");
466                                         timeout_ms = 0;
467                                 } else {
468                                         VDBG("adjusting timeout\n");
469                                         timeout_ms = (diff / 1000) | 1; /* never 0 */
470                                 }
471                         }
472
473                         if ((fds[0].revents & POLLIN) == 0) {
474                                 if (fds[0].revents & POLLERR) {
475                                         // FIXME: links routinely go down;
476                                         // this shouldn't necessarily exit.
477                                         bb_error_msg("iface %s is down", argv_intf);
478                                         if (ready) {
479                                                 run(argv, "deconfig", &ip);
480                                         }
481                                         return EXIT_FAILURE;
482                                 }
483                                 continue;
484                         }
485
486                         // read ARP packet
487                         if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
488                                 bb_perror_msg_and_die(bb_msg_read_error);
489                         }
490                         if (p.eth.ether_type != htons(ETHERTYPE_ARP))
491                                 continue;
492 #ifdef DEBUG
493                         {
494                                 struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
495                                 struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
496                                 struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
497                                 struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
498                                 VDBG("%s recv arp type=%d, op=%d,\n",
499                                         argv_intf, ntohs(p.eth.ether_type),
500                                         ntohs(p.arp.arp_op));
501                                 VDBG("\tsource=%s %s\n",
502                                         ether_ntoa(sha),
503                                         inet_ntoa(*spa));
504                                 VDBG("\ttarget=%s %s\n",
505                                         ether_ntoa(tha),
506                                         inet_ntoa(*tpa));
507                         }
508 #endif
509                         if (p.arp.arp_op != htons(ARPOP_REQUEST)
510                          && p.arp.arp_op != htons(ARPOP_REPLY)
511                         ) {
512                                 continue;
513                         }
514
515                         source_ip_conflict = 0;
516                         target_ip_conflict = 0;
517
518                         if (memcmp(&p.arp.arp_sha, &eth_addr, ETH_ALEN) != 0) {
519                                 if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr))) {
520                                         /* A probe or reply with source_ip == chosen ip */
521                                         source_ip_conflict = 1;
522                                 }
523                                 if (p.arp.arp_op == htons(ARPOP_REQUEST)
524                                  && memcmp(p.arp.arp_spa, &null_ip, sizeof(struct in_addr)) == 0
525                                  && memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0
526                                 ) {
527                                         /* A probe with source_ip == 0.0.0.0, target_ip == chosen ip:
528                                          * another host trying to claim this ip!
529                                          */
530                                         target_ip_conflict = 1;
531                                 }
532                         }
533
534                         VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n",
535                                 state, source_ip_conflict, target_ip_conflict);
536                         switch (state) {
537                         case PROBE:
538                         case ANNOUNCE:
539                                 // When probing or announcing, check for source IP conflicts
540                                 // and other hosts doing ARP probes (target IP conflicts).
541                                 if (source_ip_conflict || target_ip_conflict) {
542                                         conflicts++;
543                                         if (conflicts >= MAX_CONFLICTS) {
544                                                 VDBG("%s ratelimit\n", argv_intf);
545                                                 timeout_ms = RATE_LIMIT_INTERVAL * 1000;
546                                                 state = RATE_LIMIT_PROBE;
547                                         }
548
549                                         // restart the whole protocol
550                                         ip.s_addr = pick_nip();
551                                         timeout_ms = 0;
552                                         nprobes = 0;
553                                         nclaims = 0;
554                                 }
555                                 break;
556                         case MONITOR:
557                                 // If a conflict, we try to defend with a single ARP probe.
558                                 if (source_ip_conflict) {
559                                         VDBG("monitor conflict -- defending\n");
560                                         state = DEFEND;
561                                         timeout_ms = DEFEND_INTERVAL * 1000;
562                                         arp(/* ARPOP_REQUEST, */
563                                                 /* &eth_addr, */ ip,
564                                                 &eth_addr, ip);
565                                 }
566                                 break;
567                         case DEFEND:
568                                 // Well, we tried.  Start over (on conflict).
569                                 if (source_ip_conflict) {
570                                         state = PROBE;
571                                         VDBG("defend conflict -- starting over\n");
572                                         ready = 0;
573                                         run(argv, "deconfig", &ip);
574
575                                         // restart the whole protocol
576                                         ip.s_addr = pick_nip();
577                                         timeout_ms = 0;
578                                         nprobes = 0;
579                                         nclaims = 0;
580                                 }
581                                 break;
582                         default:
583                                 // Invalid, should never happen.  Restart the whole protocol.
584                                 VDBG("invalid state -- starting over\n");
585                                 state = PROBE;
586                                 ip.s_addr = pick_nip();
587                                 timeout_ms = 0;
588                                 nprobes = 0;
589                                 nclaims = 0;
590                                 break;
591                         } // switch state
592                         break; // case 1 (packets arriving)
593                 } // switch poll
594         } // while (1)
595 #undef argv_intf
596 }