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 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 };
  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 &eth_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 (&eth_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 = wait4pid(spawn(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);
 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, 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         argc -= optind;
 245         argv += optind - 1;
 246
 247         /* Now: argv[0]:junk argv[1]:intf argv[2]:script argv[3]:NULL */
 248         /* We need to make space for script argument: */
 249         argv[0] = argv[1];
 250         argv[1] = argv[2];
 251         /* Now: argv[0]:intf argv[1]:script argv[2]:junk argv[3]:NULL */
 252 #define argv_intf (argv[0])
 253
 254         xsetenv("interface", argv_intf);
 255
 256         // initialize the interface (modprobe, ifup, etc)
 257         if (run(argv, "init", NULL))
 258                 return EXIT_FAILURE;
 259
 260         // initialize saddr
 261         // saddr is: { u16 sa_family; u8 sa_data[14]; }
 262         //memset(&saddr, 0, sizeof(saddr));
 263         //TODO: are we leaving sa_family == 0 (AF_UNSPEC)?!
 264         safe_strncpy(saddr.sa_data, argv_intf, sizeof(saddr.sa_data));
 265
 266         // bind to the interface's ARP socket
 267         xbind(sock_fd, &saddr, sizeof(saddr));
 268
 269         // get the interface's ethernet address
 270         //memset(&ifr, 0, sizeof(ifr));
 271         strncpy_IFNAMSIZ(ifr.ifr_name, argv_intf);
 272         xioctl(sock_fd, SIOCGIFHWADDR, &ifr);
 273         memcpy(&eth_addr, &ifr.ifr_hwaddr.sa_data, ETH_ALEN);
 274
 275         // start with some stable ip address, either a function of
 276         // the hardware address or else the last address we used.
 277         // we are taking low-order four bytes, as top-order ones
 278         // aren't random enough.
 279         // NOTE: the sequence of addresses we try changes only
 280         // depending on when we detect conflicts.
 281         {
 282                 uint32_t t;
 283                 move_from_unaligned32(t, ((char *)&eth_addr + 2));
 284                 srand(t);
 285         }
 286         if (ip.s_addr == 0)
 287                 ip.s_addr = pick();
 288
 289         // FIXME cases to handle:
 290         //  - zcip already running!
 291         //  - link already has local address... just defend/update
 292
 293         // daemonize now; don't delay system startup
 294         if (!FOREGROUND) {
 295 #if BB_MMU
 296                 bb_daemonize(0 /*was: DAEMON_CHDIR_ROOT*/);
 297 #endif
 298                 bb_info_msg("start, interface %s", argv_intf);
 299         }
 300
 301         // run the dynamic address negotiation protocol,
 302         // restarting after address conflicts:
 303         //  - start with some address we want to try
 304         //  - short random delay
 305         //  - arp probes to see if another host uses it
 306         //  - arp announcements that we're claiming it
 307         //  - use it
 308         //  - defend it, within limits
 309         // exit if:
 310         // - address is successfully obtained and -q was given:
 311         //   run "<script> config", then exit with exitcode 0
 312         // - poll error (when does this happen?)
 313         // - read error (when does this happen?)
 314         // - sendto error (in arp()) (when does this happen?)
 315         // - revents & POLLERR (link down). run "<script> deconfig" first
 316         state = PROBE;
 317         while (1) {
 318                 struct pollfd fds[1];
 319                 unsigned deadline_us;
 320                 struct arp_packet p;
 321                 int source_ip_conflict;
 322                 int target_ip_conflict;
 323
 324                 fds[0].fd = sock_fd;
 325                 fds[0].events = POLLIN;
 326                 fds[0].revents = 0;
 327
 328                 // poll, being ready to adjust current timeout
 329                 if (!timeout_ms) {
 330                         timeout_ms = random_delay_ms(PROBE_WAIT);
 331                         // FIXME setsockopt(sock_fd, SO_ATTACH_FILTER, ...) to
 332                         // make the kernel filter out all packets except
 333                         // ones we'd care about.
 334                 }
 335                 // set deadline_us to the point in time when we timeout
 336                 deadline_us = MONOTONIC_US() + timeout_ms * 1000;
 337
 338                 VDBG("...wait %d %s nprobes=%u, nclaims=%u\n",
 339                                 timeout_ms, argv_intf, nprobes, nclaims);
 340
 341                 switch (safe_poll(fds, 1, timeout_ms)) {
 342
 343                 default:
 344                         //bb_perror_msg("poll"); - done in safe_poll
 345                         return EXIT_FAILURE;
 346
 347                 // timeout
 348                 case 0:
 349                         VDBG("state = %d\n", state);
 350                         switch (state) {
 351                         case PROBE:
 352                                 // timeouts in the PROBE state mean no conflicting ARP packets
 353                                 // have been received, so we can progress through the states
 354                                 if (nprobes < PROBE_NUM) {
 355                                         nprobes++;
 356                                         VDBG("probe/%u %s@%s\n",
 357                                                         nprobes, argv_intf, inet_ntoa(ip));
 358                                         arp(/* ARPOP_REQUEST, */
 359                                                         /* &eth_addr, */ null_ip,
 360                                                         &null_addr, ip);
 361                                         timeout_ms = PROBE_MIN * 1000;
 362                                         timeout_ms += random_delay_ms(PROBE_MAX - PROBE_MIN);
 363                                 }
 364                                 else {
 365                                         // Switch to announce state.
 366                                         state = ANNOUNCE;
 367                                         nclaims = 0;
 368                                         VDBG("announce/%u %s@%s\n",
 369                                                         nclaims, argv_intf, inet_ntoa(ip));
 370                                         arp(/* ARPOP_REQUEST, */
 371                                                         /* &eth_addr, */ ip,
 372                                                         &eth_addr, ip);
 373                                         timeout_ms = ANNOUNCE_INTERVAL * 1000;
 374                                 }
 375                                 break;
 376                         case RATE_LIMIT_PROBE:
 377                                 // timeouts in the RATE_LIMIT_PROBE state mean no conflicting ARP packets
 378                                 // have been received, so we can move immediately to the announce state
 379                                 state = ANNOUNCE;
 380                                 nclaims = 0;
 381                                 VDBG("announce/%u %s@%s\n",
 382                                                 nclaims, argv_intf, inet_ntoa(ip));
 383                                 arp(/* ARPOP_REQUEST, */
 384                                                 /* &eth_addr, */ ip,
 385                                                 &eth_addr, ip);
 386                                 timeout_ms = ANNOUNCE_INTERVAL * 1000;
 387                                 break;
 388                         case ANNOUNCE:
 389                                 // timeouts in the ANNOUNCE state mean no conflicting ARP packets
 390                                 // have been received, so we can progress through the states
 391                                 if (nclaims < ANNOUNCE_NUM) {
 392                                         nclaims++;
 393                                         VDBG("announce/%u %s@%s\n",
 394                                                         nclaims, argv_intf, inet_ntoa(ip));
 395                                         arp(/* ARPOP_REQUEST, */
 396                                                         /* &eth_addr, */ ip,
 397                                                         &eth_addr, ip);
 398                                         timeout_ms = ANNOUNCE_INTERVAL * 1000;
 399                                 }
 400                                 else {
 401                                         // Switch to monitor state.
 402                                         state = MONITOR;
 403                                         // link is ok to use earlier
 404                                         // FIXME update filters
 405                                         run(argv, "config", &ip);
 406                                         ready = 1;
 407                                         conflicts = 0;
 408                                         timeout_ms = -1; // Never timeout in the monitor state.
 409
 410                                         // NOTE: all other exit paths
 411                                         // should deconfig ...
 412                                         if (QUIT)
 413                                                 return EXIT_SUCCESS;
 414                                 }
 415                                 break;
 416                         case DEFEND:
 417                                 // We won!  No ARP replies, so just go back to monitor.
 418                                 state = MONITOR;
 419                                 timeout_ms = -1;
 420                                 conflicts = 0;
 421                                 break;
 422                         default:
 423                                 // Invalid, should never happen.  Restart the whole protocol.
 424                                 state = PROBE;
 425                                 ip.s_addr = pick();
 426                                 timeout_ms = 0;
 427                                 nprobes = 0;
 428                                 nclaims = 0;
 429                                 break;
 430                         } // switch (state)
 431                         break; // case 0 (timeout)
 432
 433                 // packets arriving, or link went down
 434                 case 1:
 435                         // We need to adjust the timeout in case we didn't receive
 436                         // a conflicting packet.
 437                         if (timeout_ms > 0) {
 438                                 unsigned diff = deadline_us - MONOTONIC_US();
 439                                 if ((int)(diff) < 0) {
 440                                         // Current time is greater than the expected timeout time.
 441                                         // Should never happen.
 442                                         VDBG("missed an expected timeout\n");
 443                                         timeout_ms = 0;
 444                                 } else {
 445                                         VDBG("adjusting timeout\n");
 446                                         timeout_ms = (diff / 1000) | 1; /* never 0 */
 447                                 }
 448                         }
 449
 450                         if ((fds[0].revents & POLLIN) == 0) {
 451                                 if (fds[0].revents & POLLERR) {
 452                                         // FIXME: links routinely go down;
 453                                         // this shouldn't necessarily exit.
 454                                         bb_error_msg("iface %s is down", argv_intf);
 455                                         if (ready) {
 456                                                 run(argv, "deconfig", &ip);
 457                                         }
 458                                         return EXIT_FAILURE;
 459                                 }
 460                                 continue;
 461                         }
 462
 463                         // read ARP packet
 464                         if (safe_read(sock_fd, &p, sizeof(p)) < 0) {
 465                                 bb_perror_msg_and_die(bb_msg_read_error);
 466                         }
 467                         if (p.eth.ether_type != htons(ETHERTYPE_ARP))
 468                                 continue;
 469 #ifdef DEBUG
 470                         {
 471                                 struct ether_addr *sha = (struct ether_addr *) p.arp.arp_sha;
 472                                 struct ether_addr *tha = (struct ether_addr *) p.arp.arp_tha;
 473                                 struct in_addr *spa = (struct in_addr *) p.arp.arp_spa;
 474                                 struct in_addr *tpa = (struct in_addr *) p.arp.arp_tpa;
 475                                 VDBG("%s recv arp type=%d, op=%d,\n",
 476                                         argv_intf, ntohs(p.eth.ether_type),
 477                                         ntohs(p.arp.arp_op));
 478                                 VDBG("\tsource=%s %s\n",
 479                                         ether_ntoa(sha),
 480                                         inet_ntoa(*spa));
 481                                 VDBG("\ttarget=%s %s\n",
 482                                         ether_ntoa(tha),
 483                                         inet_ntoa(*tpa));
 484                         }
 485 #endif
 486                         if (p.arp.arp_op != htons(ARPOP_REQUEST)
 487                          && p.arp.arp_op != htons(ARPOP_REPLY))
 488                                 continue;
 489
 490                         source_ip_conflict = 0;
 491                         target_ip_conflict = 0;
 492
 493                         if (memcmp(p.arp.arp_spa, &ip.s_addr, sizeof(struct in_addr)) == 0
 494                          && memcmp(&p.arp.arp_sha, &eth_addr, ETH_ALEN) != 0
 495                         ) {
 496                                 source_ip_conflict = 1;
 497                         }
 498                         if (p.arp.arp_op == htons(ARPOP_REQUEST)
 499                          && memcmp(p.arp.arp_tpa, &ip.s_addr, sizeof(struct in_addr)) == 0
 500                          && memcmp(&p.arp.arp_tha, &eth_addr, ETH_ALEN) != 0
 501                         ) {
 502                                 target_ip_conflict = 1;
 503                         }
 504
 505                         VDBG("state = %d, source ip conflict = %d, target ip conflict = %d\n",
 506                                 state, source_ip_conflict, target_ip_conflict);
 507                         switch (state) {
 508                         case PROBE:
 509                         case ANNOUNCE:
 510                                 // When probing or announcing, check for source IP conflicts
 511                                 // and other hosts doing ARP probes (target IP conflicts).
 512                                 if (source_ip_conflict || target_ip_conflict) {
 513                                         conflicts++;
 514                                         if (conflicts >= MAX_CONFLICTS) {
 515                                                 VDBG("%s ratelimit\n", argv_intf);
 516                                                 timeout_ms = RATE_LIMIT_INTERVAL * 1000;
 517                                                 state = RATE_LIMIT_PROBE;
 518                                         }
 519
 520                                         // restart the whole protocol
 521                                         ip.s_addr = pick();
 522                                         timeout_ms = 0;
 523                                         nprobes = 0;
 524                                         nclaims = 0;
 525                                 }
 526                                 break;
 527                         case MONITOR:
 528                                 // If a conflict, we try to defend with a single ARP probe.
 529                                 if (source_ip_conflict) {
 530                                         VDBG("monitor conflict -- defending\n");
 531                                         state = DEFEND;
 532                                         timeout_ms = DEFEND_INTERVAL * 1000;
 533                                         arp(/* ARPOP_REQUEST, */
 534                                                 /* &eth_addr, */ ip,
 535                                                 &eth_addr, ip);
 536                                 }
 537                                 break;
 538                         case DEFEND:
 539                                 // Well, we tried.  Start over (on conflict).
 540                                 if (source_ip_conflict) {
 541                                         state = PROBE;
 542                                         VDBG("defend conflict -- starting over\n");
 543                                         ready = 0;
 544                                         run(argv, "deconfig", &ip);
 545
 546                                         // restart the whole protocol
 547                                         ip.s_addr = pick();
 548                                         timeout_ms = 0;
 549                                         nprobes = 0;
 550                                         nclaims = 0;
 551                                 }
 552                                 break;
 553                         default:
 554                                 // Invalid, should never happen.  Restart the whole protocol.
 555                                 VDBG("invalid state -- starting over\n");
 556                                 state = PROBE;
 557                                 ip.s_addr = pick();
 558                                 timeout_ms = 0;
 559                                 nprobes = 0;
 560                                 nclaims = 0;
 561                                 break;
 562                         } // switch state
 563                         break; // case 1 (packets arriving)
 564                 } // switch poll
 565         } // while (1)
 566 #undef argv_intf
 567 }