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:     "\n$LOGGING=none           Suppress logging"
  37 //usage:     "\n$LOGGING=syslog         Log to syslog"
  38 //usage:     "\n"
  39 //usage:     "\nWith no -q, runs continuously monitoring for ARP conflicts,"
  40 //usage:     "\nexits only on I/O errors (link down etc)"
  41
  42 #include "libbb.h"
  43 #include <netinet/ether.h>
  44 #include <net/if.h>
  45 #include <net/if_arp.h>
  46 #include <linux/sockios.h>
  47
  48 #include <syslog.h>
  49
  50 /* We don't need more than 32 bits of the counter */
  51 #define MONOTONIC_US() ((unsigned)monotonic_us())
  52
  53 struct arp_packet {
  54         struct ether_header eth;
  55         struct ether_arp arp;
  56 } PACKED;
  57
  58 enum {
  59         /* 169.254.0.0 */
  60         LINKLOCAL_ADDR = 0xa9fe0000,
  61
  62         /* 0-1 seconds before sending 1st probe */
  63         PROBE_WAIT = 1,
  64         /* 1-2 seconds between probes */
  65         PROBE_MIN = 1,
  66         PROBE_MAX = 2,
  67         PROBE_NUM = 3,          /* total probes to send */
  68         ANNOUNCE_INTERVAL = 2,  /* 2 seconds between announces */
  69         ANNOUNCE_NUM = 3,       /* announces to send */
  70         /* if probe/announce sees a conflict, multiply RANDOM(NUM_CONFLICT) by... */
  71         CONFLICT_MULTIPLIER = 2,
  72         /* if we monitor and see a conflict, how long is defend state? */
  73         DEFEND_INTERVAL = 10
  74 };
  75
  76 /* States during the configuration process. */
  77 enum {
  78         PROBE = 0,
  79         ANNOUNCE,
  80         MONITOR,
  81         DEFEND
  82 };
  83
  84 #define VDBG(...) do { } while (0)
  85
  86
  87 enum {
  88         sock_fd = 3
  89 };
  90
  91 struct globals {
  92         struct sockaddr iface_sockaddr;
  93         struct ether_addr eth_addr;
  94         uint32_t localnet_ip;
  95 } FIX_ALIASING;
  96 #define G (*(struct globals*)&bb_common_bufsiz1)
  97 #define INIT_G() do { } while (0)
  98
  99
 100 /**
 101  * Pick a random link local IP address on 169.254/16, except that
 102  * the first and last 256 addresses are reserved.
 103  */
 104 static uint32_t pick_nip(void)
 105 {
 106         unsigned tmp;
 107
 108         do {
 109                 tmp = rand() & IN_CLASSB_HOST;
 110         } while (tmp > (IN_CLASSB_HOST - 0x0200));
 111         return htonl((G.localnet_ip + 0x0100) + tmp);
 112 }
 113
 114 static const char *nip_to_a(uint32_t nip)
 115 {
 116         struct in_addr in;
 117         in.s_addr = nip;
 118         return inet_ntoa(in);
 119 }
 120
 121 /**
 122  * Broadcast an ARP packet.
 123  */
 124 static void arp(
 125         /* int op, - always ARPOP_REQUEST */
 126         /* const struct ether_addr *source_eth, - always &G.eth_addr */
 127                                         uint32_t source_nip,
 128         const struct ether_addr *target_eth, uint32_t target_nip)
 129 {
 130         enum { op = ARPOP_REQUEST };
 131 #define source_eth (&G.eth_addr)
 132
 133         struct arp_packet p;
 134         memset(&p, 0, sizeof(p));
 135
 136         // ether header
 137         p.eth.ether_type = htons(ETHERTYPE_ARP);
 138         memcpy(p.eth.ether_shost, source_eth, ETH_ALEN);
 139         memset(p.eth.ether_dhost, 0xff, ETH_ALEN);
 140
 141         // arp request
 142         p.arp.arp_hrd = htons(ARPHRD_ETHER);
 143         p.arp.arp_pro = htons(ETHERTYPE_IP);
 144         p.arp.arp_hln = ETH_ALEN;
 145         p.arp.arp_pln = 4;
 146         p.arp.arp_op = htons(op);
 147         memcpy(&p.arp.arp_sha, source_eth, ETH_ALEN);
 148         memcpy(&p.arp.arp_spa, &source_nip, 4);
 149         memcpy(&p.arp.arp_tha, target_eth, ETH_ALEN);
 150         memcpy(&p.arp.arp_tpa, &target_nip, 4);
 151
 152         // send it
 153         // Even though sock_fd is already bound to G.iface_sockaddr, just send()
 154         // won't work, because "socket is not connected"
 155         // (and connect() won't fix that, "operation not supported").
 156         // Thus we sendto() to G.iface_sockaddr. I wonder which sockaddr
 157         // (from bind() or from sendto()?) kernel actually uses
 158         // to determine iface to emit the packet from...
 159         xsendto(sock_fd, &p, sizeof(p), &G.iface_sockaddr, sizeof(G.iface_sockaddr));
 160 #undef source_eth
 161 }
 162
 163 /**
 164  * Run a script.
 165  * argv[0]:intf argv[1]:script_name argv[2]:junk argv[3]:NULL
 166  */
 167 static int run(char *argv[3], const char *param, uint32_t nip)
 168 {
 169         int status;
 170         const char *addr = addr; /* for gcc */
 171         const char *fmt = "%s %s %s" + 3;
 172
 173         argv[2] = (char*)param;
 174
 175         VDBG("%s run %s %s\n", argv[0], argv[1], argv[2]);
 176
 177         if (nip != 0) {
 178                 addr = nip_to_a(nip);
 179                 xsetenv("ip", addr);
 180                 fmt -= 3;
 181         }
 182         bb_info_msg(fmt, argv[2], argv[0], addr);
 183
 184         status = spawn_and_wait(argv + 1);
 185         if (status < 0) {
 186                 bb_perror_msg("%s %s %s" + 3, argv[2], argv[0]);
 187                 return -errno;
 188         }
 189         if (status != 0)
 190                 bb_error_msg("script %s %s failed, exitcode=%d", argv[1], argv[2], status & 0xff);
 191         return status;
 192 }
 193
 194 /**
 195  * Return milliseconds of random delay, up to "secs" seconds.
 196  */
 197 static ALWAYS_INLINE unsigned random_delay_ms(unsigned secs)
 198 {
 199         return rand() % (secs * 1000);
 200 }
 201
 202 /**
 203  * main program
 204  */
 205 int zcip_main(int argc, char **argv) MAIN_EXTERNALLY_VISIBLE;
 206 int zcip_main(int argc UNUSED_PARAM, char **argv)
 207 {
 208         int state;
 209         char *r_opt;
 210         const char *l_opt = "169.254.0.0";
 211         unsigned opts;
 212
 213         // ugly trick, but I want these zeroed in one go
 214         struct {
 215                 const struct ether_addr null_addr;
 216                 struct ifreq ifr;
 217                 uint32_t chosen_nip;
 218                 int timeout_ms; // must be signed
 219                 unsigned conflicts;
 220                 unsigned nsent;
 221                 int verbose;
 222         } L;
 223 #define null_addr  (L.null_addr )
 224 #define ifr        (L.ifr       )
 225 #define chosen_nip (L.chosen_nip)
 226 #define timeout_ms (L.timeout_ms)
 227 #define conflicts  (L.conflicts )
 228 #define nsent      (L.nsent     )
 229 #define verbose    (L.verbose   )
 230
 231         memset(&L, 0, sizeof(L));
 232         INIT_G();
 233
 234 #define FOREGROUND (opts & 1)
 235 #define QUIT       (opts & 2)
 236         // parse commandline: prog [options] ifname script
 237         // exactly 2 args; -v accumulates and implies -f
 238         opt_complementary = "=2:vv:vf";
 239         opts = getopt32(argv, "fqr:l:v", &r_opt, &l_opt, &verbose);
 240 #if !BB_MMU
 241         // on NOMMU reexec early (or else we will rerun things twice)
 242         if (!FOREGROUND)
 243                 bb_daemonize_or_rexec(0 /*was: DAEMON_CHDIR_ROOT*/, argv);
 244 #endif
 245         // open an ARP socket
 246         // (need to do it before openlog to prevent openlog from taking
 247         // fd 3 (sock_fd==3))
 248         xmove_fd(xsocket(AF_PACKET, SOCK_PACKET, htons(ETH_P_ARP)), sock_fd);
 249         if (!FOREGROUND) {
 250                 // do it before all bb_xx_msg calls
 251                 openlog(applet_name, 0, LOG_DAEMON);
 252                 logmode |= LOGMODE_SYSLOG;
 253         }
 254         bb_logenv_override();
 255
 256         { // -l n.n.n.n
 257                 struct in_addr net;
 258                 if (inet_aton(l_opt, &net) == 0
 259                  || (net.s_addr & htonl(IN_CLASSB_NET)) != net.s_addr
 260                 ) {
 261                         bb_error_msg_and_die("invalid network address");
 262                 }
 263                 G.localnet_ip = ntohl(net.s_addr);
 264         }
 265         if (opts & 4) { // -r n.n.n.n
 266                 struct in_addr ip;
 267                 if (inet_aton(r_opt, &ip) == 0
 268                  || (ntohl(ip.s_addr) & IN_CLASSB_NET) != G.localnet_ip
 269                 ) {
 270                         bb_error_msg_and_die("invalid link address");
 271                 }
 272                 chosen_nip = ip.s_addr;
 273         }
 274         argv += optind - 1;
 275
 276         /* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
 277         /* We need to make space for script argument: */
 278         argv[0] = argv[1];
 279         argv[1] = argv[2];
 280         /* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
 281 #define argv_intf (argv[0])
 282
 283         xsetenv("interface", argv_intf);
 284
 285         // initialize the interface (modprobe, ifup, etc)
 286         if (run(argv, "init", 0))
 287                 return EXIT_FAILURE;
 288
 289         // initialize G.iface_sockaddr
 290         // G.iface_sockaddr is: { u16 sa_family; u8 sa_data[14]; }
 291         //memset(&G.iface_sockaddr, 0, sizeof(G.iface_sockaddr));
 292         //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
 293         safe_strncpy(G.iface_sockaddr.sa_data, argv_intf, sizeof(G.iface_sockaddr.sa_data));
 294
 295         // bind to the interface's ARP socket
 296         xbind(sock_fd, &G.iface_sockaddr, sizeof(G.iface_sockaddr));
 297
 298         // get the interface's ethernet address
 299         //memset(&ifr, 0, sizeof(ifr));
 300         strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
 301         xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
 302         memcpy(&G.eth_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 303
 304         // start with some stable ip address, either a function of
 305         // the hardware address or else the last address we used.
 306         // we are taking low-order four bytes, as top-order ones
 307         // aren't random enough.
 308         // NOTE: the sequence of addresses we try changes only
 309         // depending on when we detect conflicts.
 310         {
 311                 uint32_t t;
 312                 move_from_unaligned32(t, ((char *)&G.eth_addr + 2));
 313                 srand(t);
 314         }
 315         if (chosen_nip == 0)
 316                 chosen_nip = pick_nip();
 317
 318         // FIXME cases to handle:
 319         //  - zcip already running!
 320         //  - link already has local address... just defend/update
 321
 322         // daemonize now; don't delay system startup
 323         if (!FOREGROUND) {
 324 #if BB_MMU
 325                 bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
 326 #endif
 327                 bb_info_msg("start, interface %s", argv_intf);
 328         }
 329
 330         // run the dynamic address negotiation protocol,
 331         // restarting after address conflicts:
 332         //  - start with some address we want to try
 333         //  - short random delay
 334         //  - arp probes to see if another host uses it
 335         //    00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff, ARP (0x0806): arp who-has 169.254.194.171 tell 0.0.0.0
 336         //  - arp announcements that we're claiming it
 337         //    00:04:e2:64:23:c2 > ff:ff:ff:ff:ff:ff, ARP (0x0806): arp who-has 169.254.194.171 (00:04:e2:64:23:c2) tell 169.254.194.171
 338         //  - use it
 339         //  - defend it, within limits
 340         // exit if:
 341         // - address is successfully obtained and -q was given:
 342         //   run "<script> config", then exit with exitcode 0
 343         // - poll error (when does this happen?)
 344         // - read error (when does this happen?)
 345         // - sendto error (in arp()) (when does this happen?)
 346         // - revents & POLLERR (link down). run "<script> deconfig" first
 347         state = PROBE;
 348         while (1) {
 349                 struct pollfd fds[1];
 350                 unsigned deadline_us;
 351                 struct arp_packet p;
 352                 int ip_conflict;
 353
 354                 fds[0].fd = sock_fd;
 355                 fds[0].events = POLLIN;
 356                 fds[0].revents = 0;
 357
 358                 // poll, being ready to adjust current timeout
 359                 if (!timeout_ms) {
 360                         timeout_ms = random_delay_ms(PROBE_WAIT);
 361                         // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
 362                         // make the kernel filter out all packets except
 363                         // ones we'd care about.
 364                 }
 365                 // set deadline_us to the point in time when we timeout
 366                 deadline_us = MONOTONIC_US() + timeout_ms * 1000;
 367
 368                 VDBG("...wait %d %s nsent=%u\n",
 369                                 timeout_ms, argv_intf, nsent);
 370
 371                 switch (safe_poll(fds, 1, timeout_ms)) {
 372
 373                 default:
 374                         //bb_perror_msg("poll"); - done in safe_poll
 375                         return EXIT_FAILURE;
 376
 377                 // timeout
 378                 case 0:
 379                         VDBG("state = %d\n", state);
 380                         switch (state) {
 381                         case PROBE:
 382                                 // timeouts in the PROBE state mean no conflicting ARP packets
 383                                 // have been received, so we can progress through the states
 384                                 if (nsent < PROBE_NUM) {
 385                                         nsent++;
 386                                         VDBG("probe/%u %s@%s\n",
 387                                                         nsent, argv_intf, nip_to_a(chosen_nip));
 388                                         timeout_ms = PROBE_MIN * 1000;
 389                                         timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
 390                                         arp(/* ARPOP_REQUEST, */
 391                                                         /* &G.eth_addr, */ 0,
 392                                                         &null_addr, chosen_nip);
 393                                         break;
 394                                 }
 395                                 // Switch to announce state
 396                                 nsent = 0;
 397                                 state = ANNOUNCE;
 398                                 goto send_announce;
 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 (nsent < ANNOUNCE_NUM) {
 403  send_announce:
 404                                         nsent++;
 405                                         VDBG("announce/%u %s@%s\n",
 406                                                         nsent, argv_intf, nip_to_a(chosen_nip));
 407                                         timeout_ms = ANNOUNCE_INTERVAL * 1000;
 408                                         arp(/* ARPOP_REQUEST, */
 409                                                         /* &G.eth_addr, */ chosen_nip,
 410                                                         &G.eth_addr, chosen_nip);
 411                                         break;
 412                                 }
 413                                 // Switch to monitor state
 414                                 // FIXME update filters
 415                                 run(argv, "config", chosen_nip);
 416                                 // NOTE: all other exit paths should deconfig...
 417                                 if (QUIT)
 418                                         return EXIT_SUCCESS;
 419                                 conflicts = 0;
 420                                 timeout_ms = -1; // Never timeout in the monitor state.
 421                                 state = MONITOR;
 422                                 break;
 423                         case DEFEND:
 424                                 // Defend period ended with no ARP replies - we won
 425                                 conflicts = 0;
 426                                 timeout_ms = -1;
 427                                 state = MONITOR;
 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                                         diff = 0;
 441                                 }
 442                                 VDBG("adjusting timeout\n");
 443                                 timeout_ms = (diff / 1000) | 1; // never 0
 444                         }
 445
 446                         if ((fds[0].revents & POLLIN) == 0) {
 447                                 if (fds[0].revents & POLLERR) {
 448                                         // FIXME: links routinely go down;
 449                                         // this shouldn't necessarily exit.
 450                                         bb_error_msg("iface %s is down", argv_intf);
 451                                         if (state >= MONITOR) {
 452                                                 // only if we are in MONITOR or DEFEND
 453                                                 run(argv, "deconfig", chosen_nip);
 454                                         }
 455                                         return EXIT_FAILURE;
 456                                 }
 457                                 continue;
 458                         }
 459
 460                         // read ARP packet
 461                         if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
 462                                 bb_perror_msg_and_die(bb_msg_read_error);
 463                         }
 464
 465                         if (p.eth.ether_type != htons(ETHERTYPE_ARP))
 466                                 continue;
 467                         if (p.arp.arp_op != htons(ARPOP_REQUEST)
 468                          && p.arp.arp_op != htons(ARPOP_REPLY)
 469                         ) {
 470                                 continue;
 471                         }
 472 #ifdef DEBUG
 473                         {
 474                                 struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
 475                                 struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
 476                                 struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
 477                                 struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
 478                                 VDBG("source=%s %s\n", ether_ntoa(sha), inet_ntoa(*spa));
 479                                 VDBG("target=%s %s\n", ether_ntoa(tha), inet_ntoa(*tpa));
 480                         }
 481 #endif
 482                         ip_conflict = 0;
 483                         if (memcmp(&p.arp.arp_sha, &G.eth_addr, ETH_ALEN) != 0) {
 484                                 if (memcmp(p.arp.arp_spa, &chosen_nip, 4) == 0) {
 485                                         // A probe or reply with source_ip == chosen ip
 486                                         ip_conflict = 1;
 487                                 }
 488                                 if (p.arp.arp_op == htons(ARPOP_REQUEST)
 489                                  && memcmp(p.arp.arp_spa, &const_int_0, 4) == 0
 490                                  && memcmp(p.arp.arp_tpa, &chosen_nip, 4) == 0
 491                                 ) {
 492                                         // A probe with source_ip == 0.0.0.0, target_ip == chosen ip:
 493                                         // another host trying to claim this ip!
 494                                         ip_conflict |= 2;
 495                                 }
 496                         }
 497                         VDBG("state:%d ip_conflict:%d\n", state, ip_conflict);
 498                         if (!ip_conflict)
 499                                 break;
 500
 501                         // Either src or target IP conflict exists
 502                         if (state <= ANNOUNCE) {
 503                                 // PROBE or ANNOUNCE
 504                                 conflicts++;
 505                                 timeout_ms = PROBE_MIN * 1000
 506                                         + CONFLICT_MULTIPLIER * random_delay_ms(conflicts);
 507                                 chosen_nip = pick_nip();
 508                                 nsent = 0;
 509                                 state = PROBE;
 510                                 break;
 511                         }
 512                         // MONITOR or DEFEND: only src IP conflict is a problem
 513                         if (ip_conflict & 1) {
 514                                 if (state == MONITOR) {
 515                                         // Src IP conflict, defend with a single ARP probe
 516                                         VDBG("monitor conflict - defending\n");
 517                                         timeout_ms = DEFEND_INTERVAL * 1000;
 518                                         state = DEFEND;
 519                                         arp(/* ARPOP_REQUEST, */
 520                                                 /* &G.eth_addr, */ chosen_nip,
 521                                                 &G.eth_addr, chosen_nip);
 522                                         break;
 523                                 }
 524                                 // state == DEFEND
 525                                 // Another src IP conflict, start over
 526                                 VDBG("defend conflict - starting over\n");
 527                                 run(argv, "deconfig", chosen_nip);
 528
 529                                 // restart the whole protocol
 530                                 timeout_ms = 0;
 531                                 chosen_nip = pick_nip();
 532                                 nsent = 0;
 533                                 state = PROBE;
 534                         }
 535                         break; // case 1 (packet arrived)
 536                 } // switch (poll)
 537         } // while (1)
 538 #undef argv_intf
 539 }