contrib/ntp/ntpd/ntp_io.c

   1 /*
   2  * ntp_io.c - input/output routines for ntpd.   The socket-opening code
   3  *                 was shamelessly stolen from ntpd.
   4  */
   5
   6 #ifdef HAVE_CONFIG_H
   7 # include <config.h>
   8 #endif
   9
  10 #include <stdio.h>
  11 #include <signal.h>
  12 #ifdef HAVE_FNMATCH_H
  13 # include <fnmatch.h>
  14 # if !defined(FNM_CASEFOLD) && defined(FNM_IGNORECASE)
  15 #  define FNM_CASEFOLD FNM_IGNORECASE
  16 # endif
  17 #endif
  18 #ifdef HAVE_SYS_PARAM_H
  19 # include <sys/param.h>
  20 #endif
  21 #ifdef HAVE_SYS_IOCTL_H
  22 # include <sys/ioctl.h>
  23 #endif
  24 #ifdef HAVE_SYS_SOCKIO_H        /* UXPV: SIOC* #defines (Frank Vance <fvance@waii.com>) */
  25 # include <sys/sockio.h>
  26 #endif
  27 #ifdef HAVE_SYS_UIO_H
  28 # include <sys/uio.h>
  29 #endif
  30
  31 #include "ntp_machine.h"
  32 #include "ntpd.h"
  33 #include "ntp_io.h"
  34 #include "iosignal.h"
  35 #include "ntp_lists.h"
  36 #include "ntp_refclock.h"
  37 #include "ntp_stdlib.h"
  38 #include "ntp_worker.h"
  39 #include "ntp_request.h"
  40 #include "ntp_assert.h"
  41 #include "timevalops.h"
  42 #include "timespecops.h"
  43 #include "ntpd-opts.h"
  44 #include "safecast.h"
  45
  46 /* Don't include ISC's version of IPv6 variables and structures */
  47 #define ISC_IPV6_H 1
  48 #include <isc/mem.h>
  49 #include <isc/interfaceiter.h>
  50 #include <isc/netaddr.h>
  51 #include <isc/result.h>
  52 #include <isc/sockaddr.h>
  53
  54 #ifdef SIM
  55 #include "ntpsim.h"
  56 #endif
  57
  58 #ifdef HAS_ROUTING_SOCKET
  59 # include <net/route.h>
  60 # ifdef HAVE_RTNETLINK
  61 #  include <linux/rtnetlink.h>
  62 # endif
  63 #endif
  64
  65 /*
  66  * setsockopt does not always have the same arg declaration
  67  * across all platforms. If it's not defined we make it empty
  68  */
  69
  70 #ifndef SETSOCKOPT_ARG_CAST
  71 #define SETSOCKOPT_ARG_CAST
  72 #endif
  73
  74 extern int listen_to_virtual_ips;
  75
  76 #ifndef IPTOS_DSCP_EF
  77 #define IPTOS_DSCP_EF 0xb8
  78 #endif
  79 int qos = IPTOS_DSCP_EF;        /* QoS RFC3246 */
  80
  81 #ifdef LEAP_SMEAR
  82 /* TODO burnicki: This should be moved to ntp_timer.c, but if we do so
  83  * we get a linker error. Since we're running out of time before the leap
  84  * second occurs, we let it here where it just works.
  85  */
  86 int leap_smear_intv;
  87 #endif
  88
  89 /*
  90  * NIC rule entry
  91  */
  92 typedef struct nic_rule_tag nic_rule;
  93
  94 struct nic_rule_tag {
  95         nic_rule *      next;
  96         nic_rule_action action;
  97         nic_rule_match  match_type;
  98         char *          if_name;
  99         sockaddr_u      addr;
 100         int             prefixlen;
 101 };
 102
 103 /*
 104  * NIC rule listhead.  Entries are added at the head so that the first
 105  * match in the list is the last matching rule specified.
 106  */
 107 nic_rule *nic_rule_list;
 108
 109
 110 #if defined(SO_BINTIME) && defined(SCM_BINTIME) && defined(CMSG_FIRSTHDR)
 111 #  define HAVE_PACKET_TIMESTAMP
 112 #  define HAVE_BINTIME
 113 #  ifdef BINTIME_CTLMSGBUF_SIZE
 114 #   define CMSG_BUFSIZE BINTIME_CTLMSGBUF_SIZE
 115 #  else
 116 #   define CMSG_BUFSIZE  1536 /* moderate default */
 117 #  endif
 118 #elif defined(SO_TIMESTAMPNS) && defined(SCM_TIMESTAMPNS) && defined(CMSG_FIRSTHDR)
 119 #  define HAVE_PACKET_TIMESTAMP
 120 #  define HAVE_TIMESTAMPNS
 121 #  ifdef TIMESTAMPNS_CTLMSGBUF_SIZE
 122 #   define CMSG_BUFSIZE TIMESTAMPNS_CTLMSGBUF_SIZE
 123 #  else
 124 #   define CMSG_BUFSIZE  1536 /* moderate default */
 125 #  endif
 126 #elif defined(SO_TIMESTAMP) && defined(SCM_TIMESTAMP) && defined(CMSG_FIRSTHDR)
 127 #  define HAVE_PACKET_TIMESTAMP
 128 #  define HAVE_TIMESTAMP
 129 #  ifdef TIMESTAMP_CTLMSGBUF_SIZE
 130 #   define CMSG_BUFSIZE TIMESTAMP_CTLMSGBUF_SIZE
 131 #  else
 132 #   define CMSG_BUFSIZE  1536 /* moderate default */
 133 #  endif
 134 #else
 135 /* fill in for old/other timestamp interfaces */
 136 #endif
 137
 138 #if defined(SYS_WINNT)
 139 #include "win32_io.h"
 140 #include <isc/win32os.h>
 141 #endif
 142
 143 /*
 144  * We do asynchronous input using the SIGIO facility.  A number of
 145  * recvbuf buffers are preallocated for input.  In the signal
 146  * handler we poll to see which sockets are ready and read the
 147  * packets from them into the recvbuf's along with a time stamp and
 148  * an indication of the source host and the interface it was received
 149  * through.  This allows us to get as accurate receive time stamps
 150  * as possible independent of other processing going on.
 151  *
 152  * We watch the number of recvbufs available to the signal handler
 153  * and allocate more when this number drops below the low water
 154  * mark.  If the signal handler should run out of buffers in the
 155  * interim it will drop incoming frames, the idea being that it is
 156  * better to drop a packet than to be inaccurate.
 157  */
 158
 159
 160 /*
 161  * Other statistics of possible interest
 162  */
 163 volatile u_long packets_dropped;        /* total number of packets dropped on reception */
 164 volatile u_long packets_ignored;        /* packets received on wild card interface */
 165 volatile u_long packets_received;       /* total number of packets received */
 166          u_long packets_sent;           /* total number of packets sent */
 167          u_long packets_notsent;        /* total number of packets which couldn't be sent */
 168
 169 volatile u_long handler_calls;  /* number of calls to interrupt handler */
 170 volatile u_long handler_pkts;   /* number of pkts received by handler */
 171 u_long io_timereset;            /* time counters were reset */
 172
 173 /*
 174  * Interface stuff
 175  */
 176 endpt * any_interface;          /* wildcard ipv4 interface */
 177 endpt * any6_interface;         /* wildcard ipv6 interface */
 178 endpt * loopback_interface;     /* loopback ipv4 interface */
 179
 180 isc_boolean_t broadcast_client_enabled; /* is broadcast client enabled */
 181 u_int sys_ifnum;                        /* next .ifnum to assign */
 182 int ninterfaces;                        /* Total number of interfaces */
 183
 184 int disable_dynamic_updates;            /* scan interfaces once only */
 185
 186 #ifdef REFCLOCK
 187 /*
 188  * Refclock stuff.      We keep a chain of structures with data concerning
 189  * the guys we are doing I/O for.
 190  */
 191 static  struct refclockio *refio;
 192 #endif /* REFCLOCK */
 193
 194 /*
 195  * File descriptor masks etc. for call to select
 196  * Not needed for I/O Completion Ports or anything outside this file
 197  */
 198 static fd_set activefds;
 199 static int maxactivefd;
 200
 201 /*
 202  * bit alternating value to detect verified interfaces during an update cycle
 203  */
 204 static  u_short         sys_interphase = 0;
 205
 206 static endpt *  new_interface(endpt *);
 207 static void     add_interface(endpt *);
 208 static int      update_interfaces(u_short, interface_receiver_t,
 209                                   void *);
 210 static void     remove_interface(endpt *);
 211 static endpt *  create_interface(u_short, endpt *);
 212
 213 static int      is_wildcard_addr        (const sockaddr_u *);
 214
 215 /*
 216  * Multicast functions
 217  */
 218 static  isc_boolean_t   addr_ismulticast        (sockaddr_u *);
 219 static  isc_boolean_t   is_anycast              (sockaddr_u *,
 220                                                  const char *);
 221
 222 /*
 223  * Not all platforms support multicast
 224  */
 225 #ifdef MCAST
 226 static  isc_boolean_t   socket_multicast_enable (endpt *, sockaddr_u *);
 227 static  isc_boolean_t   socket_multicast_disable(endpt *, sockaddr_u *);
 228 #endif
 229
 230 #ifdef DEBUG
 231 static void interface_dump      (const endpt *);
 232 static void sockaddr_dump       (const sockaddr_u *);
 233 static void print_interface     (const endpt *, const char *, const char *);
 234 #define DPRINT_INTERFACE(level, args) do { if (debug >= (level)) { print_interface args; } } while (0)
 235 #else
 236 #define DPRINT_INTERFACE(level, args) do {} while (0)
 237 #endif
 238
 239 typedef struct vsock vsock_t;
 240 enum desc_type { FD_TYPE_SOCKET, FD_TYPE_FILE };
 241
 242 struct vsock {
 243         vsock_t *       link;
 244         SOCKET          fd;
 245         enum desc_type  type;
 246 };
 247
 248 vsock_t *fd_list;
 249
 250 #if !defined(HAVE_IO_COMPLETION_PORT) && defined(HAS_ROUTING_SOCKET)
 251 /*
 252  * async notification processing (e. g. routing sockets)
 253  */
 254 /*
 255  * support for receiving data on fd that is not a refclock or a socket
 256  * like e. g. routing sockets
 257  */
 258 struct asyncio_reader {
 259         struct asyncio_reader *link;                /* the list this is being kept in */
 260         SOCKET fd;                                  /* fd to be read */
 261         void  *data;                                /* possibly local data */
 262         void (*receiver)(struct asyncio_reader *);  /* input handler */
 263 };
 264
 265 struct asyncio_reader *asyncio_reader_list;
 266
 267 static void delete_asyncio_reader (struct asyncio_reader *);
 268 static struct asyncio_reader *new_asyncio_reader (void);
 269 static void add_asyncio_reader (struct asyncio_reader *, enum desc_type);
 270 static void remove_asyncio_reader (struct asyncio_reader *);
 271
 272 #endif /* !defined(HAVE_IO_COMPLETION_PORT) && defined(HAS_ROUTING_SOCKET) */
 273
 274 static void init_async_notifications (void);
 275
 276 static  int     addr_eqprefix   (const sockaddr_u *, const sockaddr_u *,
 277                                  int);
 278 static int      addr_samesubnet (const sockaddr_u *, const sockaddr_u *,
 279                                  const sockaddr_u *, const sockaddr_u *);
 280 static  int     create_sockets  (u_short);
 281 static  SOCKET  open_socket     (sockaddr_u *, int, int, endpt *);
 282 static  void    set_reuseaddr   (int);
 283 static  isc_boolean_t   socket_broadcast_enable  (struct interface *, SOCKET, sockaddr_u *);
 284
 285 #if !defined(HAVE_IO_COMPLETION_PORT) && !defined(HAVE_SIGNALED_IO)
 286 static  char *  fdbits          (int, const fd_set *);
 287 #endif
 288 #ifdef  OS_MISSES_SPECIFIC_ROUTE_UPDATES
 289 static  isc_boolean_t   socket_broadcast_disable (struct interface *, sockaddr_u *);
 290 #endif
 291
 292 typedef struct remaddr remaddr_t;
 293
 294 struct remaddr {
 295         remaddr_t *             link;
 296         sockaddr_u              addr;
 297         endpt *                 ep;
 298 };
 299
 300 remaddr_t *     remoteaddr_list;
 301 endpt *         ep_list;        /* complete endpt list */
 302 endpt *         mc4_list;       /* IPv4 mcast-capable unicast endpts */
 303 endpt *         mc6_list;       /* IPv6 mcast-capable unicast endpts */
 304
 305 static endpt *  wildipv4;
 306 static endpt *  wildipv6;
 307
 308 #ifdef SYS_WINNT
 309 int accept_wildcard_if_for_winnt;
 310 #else
 311 const int accept_wildcard_if_for_winnt = FALSE;
 312 #endif
 313
 314 static void     add_fd_to_list          (SOCKET, enum desc_type);
 315 static endpt *  find_addr_in_list       (sockaddr_u *);
 316 static endpt *  find_flagged_addr_in_list(sockaddr_u *, u_int32);
 317 static void     delete_addr_from_list   (sockaddr_u *);
 318 static void     delete_interface_from_list(endpt *);
 319 static void     close_and_delete_fd_from_list(SOCKET);
 320 static void     add_addr_to_list        (sockaddr_u *, endpt *);
 321 static void     create_wildcards        (u_short);
 322 static endpt *  findlocalinterface      (sockaddr_u *, int, int);
 323 static endpt *  findclosestinterface    (sockaddr_u *, int);
 324 #ifdef DEBUG
 325 static const char *     action_text     (nic_rule_action);
 326 #endif
 327 static nic_rule_action  interface_action(char *, sockaddr_u *, u_int32);
 328 static void             convert_isc_if  (isc_interface_t *,
 329                                          endpt *, u_short);
 330 static void             calc_addr_distance(sockaddr_u *,
 331                                            const sockaddr_u *,
 332                                            const sockaddr_u *);
 333 static int              cmp_addr_distance(const sockaddr_u *,
 334                                           const sockaddr_u *);
 335
 336 /*
 337  * Routines to read the ntp packets
 338  */
 339 #if !defined(HAVE_IO_COMPLETION_PORT)
 340 static inline int       read_network_packet     (SOCKET, struct interface *, l_fp);
 341 static void             ntpd_addremove_io_fd    (int, int, int);
 342 static void             input_handler_scan      (const l_fp*, const fd_set*);
 343 static int/*BOOL*/      sanitize_fdset          (int errc);
 344 #ifdef REFCLOCK
 345 static inline int       read_refclock_packet    (SOCKET, struct refclockio *, l_fp);
 346 #endif
 347 #ifdef HAVE_SIGNALED_IO
 348 static void             input_handler           (l_fp*);
 349 #endif
 350 #endif
 351
 352
 353 #ifndef HAVE_IO_COMPLETION_PORT
 354 void
 355 maintain_activefds(
 356         int fd,
 357         int closing
 358         )
 359 {
 360         int i;
 361
 362         if (fd < 0 || fd >= FD_SETSIZE) {
 363                 msyslog(LOG_ERR,
 364                         "Too many sockets in use, FD_SETSIZE %d exceeded by fd %d",
 365                         FD_SETSIZE, fd);
 366                 exit(1);
 367         }
 368
 369         if (!closing) {
 370                 FD_SET(fd, &activefds);
 371                 maxactivefd = max(fd, maxactivefd);
 372         } else {
 373                 FD_CLR(fd, &activefds);
 374                 if (maxactivefd && fd == maxactivefd) {
 375                         for (i = maxactivefd - 1; i >= 0; i--)
 376                                 if (FD_ISSET(i, &activefds)) {
 377                                         maxactivefd = i;
 378                                         break;
 379                                 }
 380                         INSIST(fd != maxactivefd);
 381                 }
 382         }
 383 }
 384 #endif  /* !HAVE_IO_COMPLETION_PORT */
 385
 386
 387 #ifdef DEBUG_TIMING
 388 /*
 389  * collect timing information for various processing
 390  * paths. currently we only pass them on to the file
 391  * for later processing. this could also do histogram
 392  * based analysis in other to reduce the load (and skew)
 393  * dur to the file output
 394  */
 395 void
 396 collect_timing(struct recvbuf *rb, const char *tag, int count, l_fp *dts)
 397 {
 398         char buf[256];
 399
 400         snprintf(buf, sizeof(buf), "%s %d %s %s",
 401                  (rb != NULL)
 402                      ? ((rb->dstadr != NULL)
 403                             ? stoa(&rb->recv_srcadr)
 404                             : "-REFCLOCK-")
 405                      : "-",
 406                  count, lfptoa(dts, 9), tag);
 407         record_timing_stats(buf);
 408 }
 409 #endif
 410
 411 /*
 412  * About dynamic interfaces, sockets, reception and more...
 413  *
 414  * the code solves following tasks:
 415  *
 416  *   - keep a current list of active interfaces in order
 417  *     to bind to to the interface address on NTP_PORT so that
 418  *     all wild and specific bindings for NTP_PORT are taken by ntpd
 419  *     to avoid other daemons messing with the time or sockets.
 420  *   - all interfaces keep a list of peers that are referencing
 421  *     the interface in order to quickly re-assign the peers to
 422  *     new interface in case an interface is deleted (=> gone from system or
 423  *     down)
 424  *   - have a preconfigured socket ready with the right local address
 425  *     for transmission and reception
 426  *   - have an address list for all destination addresses used within ntpd
 427  *     to find the "right" preconfigured socket.
 428  *   - facilitate updating the internal interface list with respect to
 429  *     the current kernel state
 430  *
 431  * special issues:
 432  *
 433  *   - mapping of multicast addresses to the interface affected is not always
 434  *     one to one - especially on hosts with multiple interfaces
 435  *     the code here currently allocates a separate interface entry for those
 436  *     multicast addresses
 437  *     iff it is able to bind to a *new* socket with the multicast address (flags |= MCASTIF)
 438  *     in case of failure the multicast address is bound to an existing interface.
 439  *   - on some systems it is perfectly legal to assign the same address to
 440  *     multiple interfaces. Therefore this code does not keep a list of interfaces
 441  *     but a list of interfaces that represent a unique address as determined by the kernel
 442  *     by the procedure in findlocalinterface. Thus it is perfectly legal to see only
 443  *     one representative of a group of real interfaces if they share the same address.
 444  *
 445  * Frank Kardel 20050910
 446  */
 447
 448 /*
 449  * init_io - initialize I/O module.
 450  */
 451 void
 452 init_io(void)
 453 {
 454         /* Init buffer free list and stat counters */
 455         init_recvbuff(RECV_INIT);
 456         /* update interface every 5 minutes as default */
 457         interface_interval = 300;
 458
 459 #ifdef WORK_PIPE
 460         addremove_io_fd = &ntpd_addremove_io_fd;
 461 #endif
 462
 463 #if defined(SYS_WINNT)
 464         init_io_completion_port();
 465 #elif defined(HAVE_SIGNALED_IO)
 466         (void) set_signal(input_handler);
 467 #endif
 468 }
 469
 470
 471 static void
 472 ntpd_addremove_io_fd(
 473         int     fd,
 474         int     is_pipe,
 475         int     remove_it
 476         )
 477 {
 478         UNUSED_ARG(is_pipe);
 479
 480 #ifdef HAVE_SIGNALED_IO
 481         if (!remove_it)
 482                 init_socket_sig(fd);
 483 #endif /* not HAVE_SIGNALED_IO */
 484
 485         maintain_activefds(fd, remove_it);
 486 }
 487
 488
 489 /*
 490  * io_open_sockets - call socket creation routine
 491  */
 492 void
 493 io_open_sockets(void)
 494 {
 495         static int already_opened;
 496
 497         if (already_opened || HAVE_OPT( SAVECONFIGQUIT ))
 498                 return;
 499
 500         already_opened = 1;
 501
 502         /*
 503          * Create the sockets
 504          */
 505         BLOCKIO();
 506         create_sockets(NTP_PORT);
 507         UNBLOCKIO();
 508
 509         init_async_notifications();
 510
 511         DPRINTF(3, ("io_open_sockets: maxactivefd %d\n", maxactivefd));
 512 }
 513
 514
 515 #ifdef DEBUG
 516 /*
 517  * function to dump the contents of the interface structure
 518  * for debugging use only.
 519  * We face a dilemma here -- sockets are FDs under POSIX and
 520  * actually HANDLES under Windows. So we use '%lld' as format
 521  * and cast the value to 'long long'; this should not hurt
 522  * with UNIX-like systems and does not truncate values on Win64.
 523  */
 524 void
 525 interface_dump(const endpt *itf)
 526 {
 527         printf("Dumping interface: %p\n", itf);
 528         printf("fd = %lld\n", (long long)itf->fd);
 529         printf("bfd = %lld\n", (long long)itf->bfd);
 530         printf("sin = %s,\n", stoa(&itf->sin));
 531         sockaddr_dump(&itf->sin);
 532         printf("bcast = %s,\n", stoa(&itf->bcast));
 533         sockaddr_dump(&itf->bcast);
 534         printf("mask = %s,\n", stoa(&itf->mask));
 535         sockaddr_dump(&itf->mask);
 536         printf("name = %s\n", itf->name);
 537         printf("flags = 0x%08x\n", itf->flags);
 538         printf("last_ttl = %d\n", itf->last_ttl);
 539         printf("addr_refid = %08x\n", itf->addr_refid);
 540         printf("num_mcast = %d\n", itf->num_mcast);
 541         printf("received = %ld\n", itf->received);
 542         printf("sent = %ld\n", itf->sent);
 543         printf("notsent = %ld\n", itf->notsent);
 544         printf("ifindex = %u\n", itf->ifindex);
 545         printf("peercnt = %u\n", itf->peercnt);
 546         printf("phase = %u\n", itf->phase);
 547 }
 548
 549 /*
 550  * sockaddr_dump - hex dump the start of a sockaddr_u
 551  */
 552 static void
 553 sockaddr_dump(const sockaddr_u *psau)
 554 {
 555         /* Limit the size of the sockaddr_in6 hex dump */
 556         const int maxsize = min(32, sizeof(psau->sa6));
 557         const u_char *  cp;
 558         int             i;
 559
 560         /* XXX: Should we limit maxsize based on psau->saX.sin_family? */
 561         cp = (const void *)&psau->sa6;
 562
 563         for(i = 0; i < maxsize; i++) {
 564                 printf("%02x", *cp++);
 565                 if (!((i + 1) % 4))
 566                         printf(" ");
 567         }
 568         printf("\n");
 569 }
 570
 571 /*
 572  * print_interface - helper to output debug information
 573  */
 574 static void
 575 print_interface(const endpt *iface, const char *pfx, const char *sfx)
 576 {
 577         printf("%sinterface #%d: fd=%lld, bfd=%lld, name=%s, flags=0x%x, ifindex=%u, sin=%s",
 578                pfx,
 579                iface->ifnum,
 580                (long long)iface->fd,
 581                (long long)iface->bfd,
 582                iface->name,
 583                iface->flags,
 584                iface->ifindex,
 585                stoa(&iface->sin));
 586         if (AF_INET == iface->family) {
 587                 if (iface->flags & INT_BROADCAST)
 588                         printf(", bcast=%s", stoa(&iface->bcast));
 589                 printf(", mask=%s", stoa(&iface->mask));
 590         }
 591         printf(", %s:%s",
 592                (iface->ignore_packets)
 593                    ? "Disabled"
 594                    : "Enabled",
 595                sfx);
 596         if (debug > 4)  /* in-depth debugging only */
 597                 interface_dump(iface);
 598 }
 599 #endif
 600
 601 #if !defined(HAVE_IO_COMPLETION_PORT) && defined(HAS_ROUTING_SOCKET)
 602 /*
 603  * create an asyncio_reader structure
 604  */
 605 static struct asyncio_reader *
 606 new_asyncio_reader(void)
 607 {
 608         struct asyncio_reader *reader;
 609
 610         reader = emalloc_zero(sizeof(*reader));
 611         reader->fd = INVALID_SOCKET;
 612
 613         return reader;
 614 }
 615
 616 /*
 617  * delete a reader
 618  */
 619 static void
 620 delete_asyncio_reader(
 621         struct asyncio_reader *reader
 622         )
 623 {
 624         free(reader);
 625 }
 626
 627 /*
 628  * add asynchio_reader
 629  */
 630 static void
 631 add_asyncio_reader(
 632         struct asyncio_reader * reader,
 633         enum desc_type          type)
 634 {
 635         LINK_SLIST(asyncio_reader_list, reader, link);
 636         add_fd_to_list(reader->fd, type);
 637 }
 638
 639 /*
 640  * remove asynchio_reader
 641  */
 642 static void
 643 remove_asyncio_reader(
 644         struct asyncio_reader *reader
 645         )
 646 {
 647         struct asyncio_reader *unlinked;
 648
 649         UNLINK_SLIST(unlinked, asyncio_reader_list, reader, link,
 650             struct asyncio_reader);
 651
 652         if (reader->fd != INVALID_SOCKET)
 653                 close_and_delete_fd_from_list(reader->fd);
 654
 655         reader->fd = INVALID_SOCKET;
 656 }
 657 #endif /* !defined(HAVE_IO_COMPLETION_PORT) && defined(HAS_ROUTING_SOCKET) */
 658
 659
 660 /* compare two sockaddr prefixes */
 661 static int
 662 addr_eqprefix(
 663         const sockaddr_u *      a,
 664         const sockaddr_u *      b,
 665         int                     prefixlen
 666         )
 667 {
 668         isc_netaddr_t           isc_a;
 669         isc_netaddr_t           isc_b;
 670         isc_sockaddr_t          isc_sa;
 671
 672         ZERO(isc_sa);
 673         memcpy(&isc_sa.type, a, min(sizeof(isc_sa.type), sizeof(*a)));
 674         isc_netaddr_fromsockaddr(&isc_a, &isc_sa);
 675
 676         ZERO(isc_sa);
 677         memcpy(&isc_sa.type, b, min(sizeof(isc_sa.type), sizeof(*b)));
 678         isc_netaddr_fromsockaddr(&isc_b, &isc_sa);
 679
 680         return (int)isc_netaddr_eqprefix(&isc_a, &isc_b,
 681                                          (u_int)prefixlen);
 682 }
 683
 684
 685 static int
 686 addr_samesubnet(
 687         const sockaddr_u *      a,
 688         const sockaddr_u *      a_mask,
 689         const sockaddr_u *      b,
 690         const sockaddr_u *      b_mask
 691         )
 692 {
 693         const u_int32 * pa;
 694         const u_int32 * pa_limit;
 695         const u_int32 * pb;
 696         const u_int32 * pm;
 697         size_t          loops;
 698
 699         REQUIRE(AF(a) == AF(a_mask));
 700         REQUIRE(AF(b) == AF(b_mask));
 701         /*
 702          * With address and mask families verified to match, comparing
 703          * the masks also validates the address's families match.
 704          */
 705         if (!SOCK_EQ(a_mask, b_mask))
 706                 return FALSE;
 707
 708         if (IS_IPV6(a)) {
 709                 loops = sizeof(NSRCADR6(a)) / sizeof(*pa);
 710                 pa = (const void *)&NSRCADR6(a);
 711                 pb = (const void *)&NSRCADR6(b);
 712                 pm = (const void *)&NSRCADR6(a_mask);
 713         } else {
 714                 loops = sizeof(NSRCADR(a)) / sizeof(*pa);
 715                 pa = (const void *)&NSRCADR(a);
 716                 pb = (const void *)&NSRCADR(b);
 717                 pm = (const void *)&NSRCADR(a_mask);
 718         }
 719         for (pa_limit = pa + loops; pa < pa_limit; pa++, pb++, pm++)
 720                 if ((*pa & *pm) != (*pb & *pm))
 721                         return FALSE;
 722
 723         return TRUE;
 724 }
 725
 726
 727 /*
 728  * interface list enumerator - visitor pattern
 729  */
 730 void
 731 interface_enumerate(
 732         interface_receiver_t    receiver,
 733         void *                  data
 734         )
 735 {
 736         interface_info_t ifi;
 737
 738         ifi.action = IFS_EXISTS;
 739         for (ifi.ep = ep_list; ifi.ep != NULL; ifi.ep = ifi.ep->elink)
 740                 (*receiver)(data, &ifi);
 741 }
 742
 743 /*
 744  * do standard initialization of interface structure
 745  */
 746 static void
 747 init_interface(
 748         endpt *ep
 749         )
 750 {
 751         ZERO(*ep);
 752         ep->fd = INVALID_SOCKET;
 753         ep->bfd = INVALID_SOCKET;
 754         ep->phase = sys_interphase;
 755 }
 756
 757
 758 /*
 759  * create new interface structure initialize from
 760  * template structure or via standard initialization
 761  * function
 762  */
 763 static struct interface *
 764 new_interface(
 765         struct interface *interface
 766         )
 767 {
 768         struct interface *      iface;
 769
 770         iface = emalloc(sizeof(*iface));
 771
 772         if (NULL == interface)
 773                 init_interface(iface);
 774         else                            /* use the template */
 775                 memcpy(iface, interface, sizeof(*iface));
 776
 777         /* count every new instance of an interface in the system */
 778         iface->ifnum = sys_ifnum++;
 779         iface->starttime = current_time;
 780
 781 #   ifdef HAVE_IO_COMPLETION_PORT
 782         if (!io_completion_port_add_interface(iface)) {
 783                 msyslog(LOG_EMERG, "cannot register interface with IO engine -- will exit now");
 784                 exit(1);
 785         }
 786 #   endif
 787         return iface;
 788 }
 789
 790
 791 /*
 792  * return interface storage into free memory pool
 793  */
 794 static void
 795 delete_interface(
 796         endpt *ep
 797         )
 798 {
 799 #    ifdef HAVE_IO_COMPLETION_PORT
 800         io_completion_port_remove_interface(ep);
 801 #    endif
 802         free(ep);
 803 }
 804
 805
 806 /*
 807  * link interface into list of known interfaces
 808  */
 809 static void
 810 add_interface(
 811         endpt * ep
 812         )
 813 {
 814         endpt **        pmclisthead;
 815         endpt *         scan;
 816         endpt *         scan_next;
 817         endpt *         unlinked;
 818         sockaddr_u *    addr;
 819         int             ep_local;
 820         int             scan_local;
 821         int             same_subnet;
 822         int             ep_univ_iid;    /* iface ID from MAC address */
 823         int             scan_univ_iid;  /* see RFC 4291 */
 824         int             ep_privacy;     /* random local iface ID */
 825         int             scan_privacy;   /* see RFC 4941 */
 826         int             rc;
 827
 828         /* Calculate the refid */
 829         ep->addr_refid = addr2refid(&ep->sin);
 830         /* link at tail so ntpdc -c ifstats index increases each row */
 831         LINK_TAIL_SLIST(ep_list, ep, elink, endpt);
 832         ninterfaces++;
 833 #ifdef MCAST
 834         /* the rest is for enabled multicast-capable addresses only */
 835         if (ep->ignore_packets || !(INT_MULTICAST & ep->flags) ||
 836             INT_LOOPBACK & ep->flags)
 837                 return;
 838 # ifndef INCLUDE_IPV6_MULTICAST_SUPPORT
 839         if (AF_INET6 == ep->family)
 840                 return;
 841 # endif
 842         pmclisthead = (AF_INET == ep->family)
 843                          ? &mc4_list
 844                          : &mc6_list;
 845
 846         if (AF_INET6 == ep->family) {
 847                 ep_local =
 848                     IN6_IS_ADDR_LINKLOCAL(PSOCK_ADDR6(&ep->sin)) ||
 849                     IN6_IS_ADDR_SITELOCAL(PSOCK_ADDR6(&ep->sin));
 850                 ep_univ_iid = IS_IID_UNIV(&ep->sin);
 851                 ep_privacy = !!(INT_PRIVACY & ep->flags);
 852         } else {
 853                 ep_local = FALSE;
 854                 ep_univ_iid = FALSE;
 855                 ep_privacy = FALSE;
 856         }
 857         DPRINTF(4, ("add_interface mcast-capable %s%s%s%s\n",
 858                     stoa(&ep->sin),
 859                     (ep_local) ? " link/scope-local" : "",
 860                     (ep_univ_iid) ? " univ-IID" : "",
 861                     (ep_privacy) ? " privacy" : ""));
 862         /*
 863          * If we have multiple local addresses on the same network
 864          * interface, and some are link- or site-local, do not multicast
 865          * out from the link-/site-local addresses by default, to avoid
 866          * duplicate manycastclient associations between v6 peers using
 867          * link-local and global addresses.  link-local can still be
 868          * chosen using "nic ignore myv6globalprefix::/64".
 869          * Similarly, if we have multiple global addresses from the same
 870          * prefix on the same network interface, multicast from one,
 871          * preferring EUI-64, then static, then least RFC 4941 privacy
 872          * addresses.
 873          */
 874         for (scan = *pmclisthead; scan != NULL; scan = scan_next) {
 875                 scan_next = scan->mclink;
 876                 if (ep->family != scan->family)
 877                         continue;
 878                 if (strcmp(ep->name, scan->name))
 879                         continue;
 880                 same_subnet = addr_samesubnet(&ep->sin, &ep->mask,
 881                                               &scan->sin, &scan->mask);
 882                 if (AF_INET6 == ep->family) {
 883                         addr = &scan->sin;
 884                         scan_local =
 885                             IN6_IS_ADDR_LINKLOCAL(PSOCK_ADDR6(addr)) ||
 886                             IN6_IS_ADDR_SITELOCAL(PSOCK_ADDR6(addr));
 887                         scan_univ_iid = IS_IID_UNIV(addr);
 888                         scan_privacy = !!(INT_PRIVACY & scan->flags);
 889                 } else {
 890                         scan_local = FALSE;
 891                         scan_univ_iid = FALSE;
 892                         scan_privacy = FALSE;
 893                 }
 894                 DPRINTF(4, ("add_interface mcast-capable scan %s%s%s%s\n",
 895                             stoa(&scan->sin),
 896                             (scan_local) ? " link/scope-local" : "",
 897                             (scan_univ_iid) ? " univ-IID" : "",
 898                             (scan_privacy) ? " privacy" : ""));
 899                 if ((ep_local && !scan_local) || (same_subnet &&
 900                     ((ep_privacy && !scan_privacy) ||
 901                      (!ep_univ_iid && scan_univ_iid)))) {
 902                         DPRINTF(4, ("did not add %s to %s of IPv6 multicast-capable list which already has %s\n",
 903                                 stoa(&ep->sin),
 904                                 (ep_local)
 905                                     ? "tail"
 906                                     : "head",
 907                                 stoa(&scan->sin)));
 908                         return;
 909                 }
 910                 if ((scan_local && !ep_local) || (same_subnet &&
 911                     ((scan_privacy && !ep_privacy) ||
 912                      (!scan_univ_iid && ep_univ_iid)))) {
 913                         UNLINK_SLIST(unlinked, *pmclisthead,
 914                                      scan, mclink, endpt);
 915                         DPRINTF(4, ("%s %s from IPv6 multicast-capable list to add %s\n",
 916                                 (unlinked != scan)
 917                                     ? "Failed to remove"
 918                                     : "removed",
 919                                 stoa(&scan->sin), stoa(&ep->sin)));
 920                 }
 921         }
 922         /*
 923          * Add link/site local at the tail of the multicast-
 924          * capable unicast interfaces list, so that ntpd will
 925          * send from global addresses before link-/site-local
 926          * ones.
 927          */
 928         if (ep_local)
 929                 LINK_TAIL_SLIST(*pmclisthead, ep, mclink, endpt);
 930         else
 931                 LINK_SLIST(*pmclisthead, ep, mclink);
 932         DPRINTF(4, ("added %s to %s of IPv%s multicast-capable unicast local address list\n",
 933                 stoa(&ep->sin),
 934                 (ep_local)
 935                     ? "tail"
 936                     : "head",
 937                 (AF_INET == ep->family)
 938                     ? "4"
 939                     : "6"));
 940
 941         if (INVALID_SOCKET == ep->fd)
 942                 return;
 943
 944         /*
 945          * select the local address from which to send to multicast.
 946          */
 947         switch (AF(&ep->sin)) {
 948
 949         case AF_INET :
 950                 rc = setsockopt(ep->fd, IPPROTO_IP,
 951                                 IP_MULTICAST_IF,
 952                                 (void *)&NSRCADR(&ep->sin),
 953                                 sizeof(NSRCADR(&ep->sin)));
 954                 if (rc)
 955                         msyslog(LOG_ERR,
 956                                 "setsockopt IP_MULTICAST_IF %s fails: %m",
 957                                 stoa(&ep->sin));
 958                 break;
 959
 960 # ifdef INCLUDE_IPV6_MULTICAST_SUPPORT
 961         case AF_INET6 :
 962                 rc = setsockopt(ep->fd, IPPROTO_IPV6,
 963                                  IPV6_MULTICAST_IF,
 964                                  (void *)&ep->ifindex,
 965                                  sizeof(ep->ifindex));
 966                 /* do not complain if bound addr scope is ifindex */
 967                 if (rc && ep->ifindex != SCOPE(&ep->sin))
 968                         msyslog(LOG_ERR,
 969                                 "setsockopt IPV6_MULTICAST_IF %u for %s fails: %m",
 970                                 ep->ifindex, stoa(&ep->sin));
 971                 break;
 972 # endif
 973         }
 974 #endif  /* MCAST */
 975 }
 976
 977
 978 /*
 979  * remove interface from known interface list and clean up
 980  * associated resources
 981  */
 982 static void
 983 remove_interface(
 984         endpt * ep
 985         )
 986 {
 987         endpt *         unlinked;
 988         endpt **        pmclisthead;
 989         sockaddr_u      resmask;
 990
 991         UNLINK_SLIST(unlinked, ep_list, ep, elink, endpt);
 992         if (!ep->ignore_packets && INT_MULTICAST & ep->flags) {
 993                 pmclisthead = (AF_INET == ep->family)
 994                                  ? &mc4_list
 995                                  : &mc6_list;
 996                 UNLINK_SLIST(unlinked, *pmclisthead, ep, mclink, endpt);
 997                 DPRINTF(4, ("%s %s IPv%s multicast-capable unicast local address list\n",
 998                         stoa(&ep->sin),
 999                         (unlinked != NULL)
1000                             ? "removed from"
1001                             : "not found on",
1002                         (AF_INET == ep->family)
1003                             ? "4"
1004                             : "6"));
1005         }
1006         delete_interface_from_list(ep);
1007
1008         if (ep->fd != INVALID_SOCKET) {
1009                 msyslog(LOG_INFO,
1010                         "Deleting interface #%d %s, %s#%d, interface stats: received=%ld, sent=%ld, dropped=%ld, active_time=%ld secs",
1011                         ep->ifnum,
1012                         ep->name,
1013                         stoa(&ep->sin),
1014                         SRCPORT(&ep->sin),
1015                         ep->received,
1016                         ep->sent,
1017                         ep->notsent,
1018                         current_time - ep->starttime);
1019 #           ifdef HAVE_IO_COMPLETION_PORT
1020                 io_completion_port_remove_socket(ep->fd, ep);
1021 #           endif
1022                 close_and_delete_fd_from_list(ep->fd);
1023                 ep->fd = INVALID_SOCKET;
1024         }
1025
1026         if (ep->bfd != INVALID_SOCKET) {
1027                 msyslog(LOG_INFO,
1028                         "stop listening for broadcasts to %s on interface #%d %s",
1029                         stoa(&ep->bcast), ep->ifnum, ep->name);
1030 #           ifdef HAVE_IO_COMPLETION_PORT
1031                 io_completion_port_remove_socket(ep->bfd, ep);
1032 #           endif
1033                 close_and_delete_fd_from_list(ep->bfd);
1034                 ep->bfd = INVALID_SOCKET;
1035         }
1036 #   ifdef HAVE_IO_COMPLETION_PORT
1037         io_completion_port_remove_interface(ep);
1038 #   endif
1039
1040         ninterfaces--;
1041         mon_clearinterface(ep);
1042
1043         /* remove restrict interface entry */
1044         SET_HOSTMASK(&resmask, AF(&ep->sin));
1045         hack_restrict(RESTRICT_REMOVEIF, &ep->sin, &resmask,
1046                       -3, RESM_NTPONLY | RESM_INTERFACE, RES_IGNORE, 0);
1047 }
1048
1049
1050 static void
1051 log_listen_address(
1052         endpt * ep
1053         )
1054 {
1055         msyslog(LOG_INFO, "%s on %d %s %s",
1056                 (ep->ignore_packets)
1057                     ? "Listen and drop"
1058                     : "Listen normally",
1059                 ep->ifnum,
1060                 ep->name,
1061                 sptoa(&ep->sin));
1062 }
1063
1064
1065 static void
1066 create_wildcards(
1067         u_short port
1068         )
1069 {
1070         int                     v4wild;
1071 #ifdef INCLUDE_IPV6_SUPPORT
1072         int                     v6wild;
1073 #endif
1074         sockaddr_u              wildaddr;
1075         nic_rule_action         action;
1076         struct interface *      wildif;
1077
1078         /*
1079          * silence "potentially uninitialized" warnings from VC9
1080          * failing to follow the logic.  Ideally action could remain
1081          * uninitialized, and the memset be the first statement under
1082          * the first if (v4wild).
1083          */
1084         action = ACTION_LISTEN;
1085         ZERO(wildaddr);
1086
1087 #ifdef INCLUDE_IPV6_SUPPORT
1088         /*
1089          * create pseudo-interface with wildcard IPv6 address
1090          */
1091         v6wild = ipv6_works;
1092         if (v6wild) {
1093                 /* set wildaddr to the v6 wildcard address :: */
1094                 ZERO(wildaddr);
1095                 AF(&wildaddr) = AF_INET6;
1096                 SET_ADDR6N(&wildaddr, in6addr_any);
1097                 SET_PORT(&wildaddr, port);
1098                 SET_SCOPE(&wildaddr, 0);
1099
1100                 /* check for interface/nic rules affecting the wildcard */
1101                 action = interface_action(NULL, &wildaddr, 0);
1102                 v6wild = (ACTION_IGNORE != action);
1103         }
1104         if (v6wild) {
1105                 wildif = new_interface(NULL);
1106
1107                 strlcpy(wildif->name, "v6wildcard", sizeof(wildif->name));
1108                 memcpy(&wildif->sin, &wildaddr, sizeof(wildif->sin));
1109                 wildif->family = AF_INET6;
1110                 AF(&wildif->mask) = AF_INET6;
1111                 SET_ONESMASK(&wildif->mask);
1112
1113                 wildif->flags = INT_UP | INT_WILDCARD;
1114                 wildif->ignore_packets = (ACTION_DROP == action);
1115
1116                 wildif->fd = open_socket(&wildif->sin, 0, 1, wildif);
1117
1118                 if (wildif->fd != INVALID_SOCKET) {
1119                         wildipv6 = wildif;
1120                         any6_interface = wildif;
1121                         add_addr_to_list(&wildif->sin, wildif);
1122                         add_interface(wildif);
1123                         log_listen_address(wildif);
1124                 } else {
1125                         msyslog(LOG_ERR,
1126                                 "unable to bind to wildcard address %s - another process may be running - EXITING",
1127                                 stoa(&wildif->sin));
1128                         exit(1);
1129                 }
1130                 DPRINT_INTERFACE(2, (wildif, "created ", "\n"));
1131         }
1132 #endif
1133
1134         /*
1135          * create pseudo-interface with wildcard IPv4 address
1136          */
1137         v4wild = ipv4_works;
1138         if (v4wild) {
1139                 /* set wildaddr to the v4 wildcard address 0.0.0.0 */
1140                 AF(&wildaddr) = AF_INET;
1141                 SET_ADDR4N(&wildaddr, INADDR_ANY);
1142                 SET_PORT(&wildaddr, port);
1143
1144                 /* check for interface/nic rules affecting the wildcard */
1145                 action = interface_action(NULL, &wildaddr, 0);
1146                 v4wild = (ACTION_IGNORE != action);
1147         }
1148         if (v4wild) {
1149                 wildif = new_interface(NULL);
1150
1151                 strlcpy(wildif->name, "v4wildcard", sizeof(wildif->name));
1152                 memcpy(&wildif->sin, &wildaddr, sizeof(wildif->sin));
1153                 wildif->family = AF_INET;
1154                 AF(&wildif->mask) = AF_INET;
1155                 SET_ONESMASK(&wildif->mask);
1156
1157                 wildif->flags = INT_BROADCAST | INT_UP | INT_WILDCARD;
1158                 wildif->ignore_packets = (ACTION_DROP == action);
1159 #if defined(MCAST)
1160                 /*
1161                  * enable multicast reception on the broadcast socket
1162                  */
1163                 AF(&wildif->bcast) = AF_INET;
1164                 SET_ADDR4N(&wildif->bcast, INADDR_ANY);
1165                 SET_PORT(&wildif->bcast, port);
1166 #endif /* MCAST */
1167                 wildif->fd = open_socket(&wildif->sin, 0, 1, wildif);
1168
1169                 if (wildif->fd != INVALID_SOCKET) {
1170                         wildipv4 = wildif;
1171                         any_interface = wildif;
1172
1173                         add_addr_to_list(&wildif->sin, wildif);
1174                         add_interface(wildif);
1175                         log_listen_address(wildif);
1176                 } else {
1177                         msyslog(LOG_ERR,
1178                                 "unable to bind to wildcard address %s - another process may be running - EXITING",
1179                                 stoa(&wildif->sin));
1180                         exit(1);
1181                 }
1182                 DPRINT_INTERFACE(2, (wildif, "created ", "\n"));
1183         }
1184 }
1185
1186
1187 /*
1188  * add_nic_rule() -- insert a rule entry at the head of nic_rule_list.
1189  */
1190 void
1191 add_nic_rule(
1192         nic_rule_match  match_type,
1193         const char *    if_name,        /* interface name or numeric address */
1194         int             prefixlen,
1195         nic_rule_action action
1196         )
1197 {
1198         nic_rule *      rule;
1199         isc_boolean_t   is_ip;
1200
1201         rule = emalloc_zero(sizeof(*rule));
1202         rule->match_type = match_type;
1203         rule->prefixlen = prefixlen;
1204         rule->action = action;
1205
1206         if (MATCH_IFNAME == match_type) {
1207                 REQUIRE(NULL != if_name);
1208                 rule->if_name = estrdup(if_name);
1209         } else if (MATCH_IFADDR == match_type) {
1210                 REQUIRE(NULL != if_name);
1211                 /* set rule->addr */
1212                 is_ip = is_ip_address(if_name, AF_UNSPEC, &rule->addr);
1213                 REQUIRE(is_ip);
1214         } else
1215                 REQUIRE(NULL == if_name);
1216
1217         LINK_SLIST(nic_rule_list, rule, next);
1218 }
1219
1220
1221 #ifdef DEBUG
1222 static const char *
1223 action_text(
1224         nic_rule_action action
1225         )
1226 {
1227         const char *t;
1228
1229         switch (action) {
1230
1231         default:
1232                 t = "ERROR";    /* quiet uninit warning */
1233                 DPRINTF(1, ("fatal: unknown nic_rule_action %d\n",
1234                             action));
1235                 ENSURE(0);
1236                 break;
1237
1238         case ACTION_LISTEN:
1239                 t = "listen";
1240                 break;
1241
1242         case ACTION_IGNORE:
1243                 t = "ignore";
1244                 break;
1245
1246         case ACTION_DROP:
1247                 t = "drop";
1248                 break;
1249         }
1250
1251         return t;
1252 }
1253 #endif  /* DEBUG */
1254
1255
1256 static nic_rule_action
1257 interface_action(
1258         char *          if_name,
1259         sockaddr_u *    if_addr,
1260         u_int32         if_flags
1261         )
1262 {
1263         nic_rule *      rule;
1264         int             isloopback;
1265         int             iswildcard;
1266
1267         DPRINTF(4, ("interface_action: interface %s ",
1268                     (if_name != NULL) ? if_name : "wildcard"));
1269
1270         iswildcard = is_wildcard_addr(if_addr);
1271         isloopback = !!(INT_LOOPBACK & if_flags);
1272
1273         /*
1274          * Find any matching NIC rule from --interface / -I or ntp.conf
1275          * interface/nic rules.
1276          */
1277         for (rule = nic_rule_list; rule != NULL; rule = rule->next) {
1278
1279                 switch (rule->match_type) {
1280
1281                 case MATCH_ALL:
1282                         /* loopback and wildcard excluded from "all" */
1283                         if (isloopback || iswildcard)
1284                                 break;
1285                         DPRINTF(4, ("nic all %s\n",
1286                             action_text(rule->action)));
1287                         return rule->action;
1288
1289                 case MATCH_IPV4:
1290                         if (IS_IPV4(if_addr)) {
1291                                 DPRINTF(4, ("nic ipv4 %s\n",
1292                                     action_text(rule->action)));
1293                                 return rule->action;
1294                         }
1295                         break;
1296
1297                 case MATCH_IPV6:
1298                         if (IS_IPV6(if_addr)) {
1299                                 DPRINTF(4, ("nic ipv6 %s\n",
1300                                     action_text(rule->action)));
1301                                 return rule->action;
1302                         }
1303                         break;
1304
1305                 case MATCH_WILDCARD:
1306                         if (iswildcard) {
1307                                 DPRINTF(4, ("nic wildcard %s\n",
1308                                     action_text(rule->action)));
1309                                 return rule->action;
1310                         }
1311                         break;
1312
1313                 case MATCH_IFADDR:
1314                         if (rule->prefixlen != -1) {
1315                                 if (addr_eqprefix(if_addr, &rule->addr,
1316                                                   rule->prefixlen)) {
1317
1318                                         DPRINTF(4, ("subnet address match - %s\n",
1319                                             action_text(rule->action)));
1320                                         return rule->action;
1321                                 }
1322                         } else
1323                                 if (SOCK_EQ(if_addr, &rule->addr)) {
1324
1325                                         DPRINTF(4, ("address match - %s\n",
1326                                             action_text(rule->action)));
1327                                         return rule->action;
1328                                 }
1329                         break;
1330
1331                 case MATCH_IFNAME:
1332                         if (if_name != NULL
1333 #if defined(HAVE_FNMATCH) && defined(FNM_CASEFOLD)
1334                             && !fnmatch(rule->if_name, if_name, FNM_CASEFOLD)
1335 #else
1336                             && !strcasecmp(if_name, rule->if_name)
1337 #endif
1338                             ) {
1339
1340                                 DPRINTF(4, ("interface name match - %s\n",
1341                                     action_text(rule->action)));
1342                                 return rule->action;
1343                         }
1344                         break;
1345                 }
1346         }
1347
1348         /*
1349          * Unless explicitly disabled such as with "nic ignore ::1"
1350          * listen on loopback addresses.  Since ntpq and ntpdc query
1351          * "localhost" by default, which typically resolves to ::1 and
1352          * 127.0.0.1, it's useful to default to listening on both.
1353          */
1354         if (isloopback) {
1355                 DPRINTF(4, ("default loopback listen\n"));
1356                 return ACTION_LISTEN;
1357         }
1358
1359         /*
1360          * Treat wildcard addresses specially.  If there is no explicit
1361          * "nic ... wildcard" or "nic ... 0.0.0.0" or "nic ... ::" rule
1362          * default to drop.
1363          */
1364         if (iswildcard) {
1365                 DPRINTF(4, ("default wildcard drop\n"));
1366                 return ACTION_DROP;
1367         }
1368
1369         /*
1370          * Check for "virtual IP" (colon in the interface name) after
1371          * the rules so that "ntpd --interface eth0:1 -novirtualips"
1372          * does indeed listen on eth0:1's addresses.
1373          */
1374         if (!listen_to_virtual_ips && if_name != NULL
1375             && (strchr(if_name, ':') != NULL)) {
1376
1377                 DPRINTF(4, ("virtual ip - ignore\n"));
1378                 return ACTION_IGNORE;
1379         }
1380
1381         /*
1382          * If there are no --interface/-I command-line options and no
1383          * interface/nic rules in ntp.conf, the default action is to
1384          * listen.  In the presence of rules from either, the default
1385          * is to ignore.  This implements ntpd's traditional listen-
1386          * every default with no interface listen configuration, and
1387          * ensures a single -I eth0 or "nic listen eth0" means do not
1388          * listen on any other addresses.
1389          */
1390         if (NULL == nic_rule_list) {
1391                 DPRINTF(4, ("default listen\n"));
1392                 return ACTION_LISTEN;
1393         }
1394
1395         DPRINTF(4, ("implicit ignore\n"));
1396         return ACTION_IGNORE;
1397 }
1398
1399
1400 static void
1401 convert_isc_if(
1402         isc_interface_t *isc_if,
1403         endpt *itf,
1404         u_short port
1405         )
1406 {
1407         const u_char v6loop[16] = {0, 0, 0, 0, 0, 0, 0, 0,
1408                                    0, 0, 0, 0, 0, 0, 0, 1};
1409
1410         strlcpy(itf->name, isc_if->name, sizeof(itf->name));
1411         itf->ifindex = isc_if->ifindex;
1412         itf->family = (u_short)isc_if->af;
1413         AF(&itf->sin) = itf->family;
1414         AF(&itf->mask) = itf->family;
1415         AF(&itf->bcast) = itf->family;
1416         SET_PORT(&itf->sin, port);
1417         SET_PORT(&itf->mask, port);
1418         SET_PORT(&itf->bcast, port);
1419
1420         if (IS_IPV4(&itf->sin)) {
1421                 NSRCADR(&itf->sin) = isc_if->address.type.in.s_addr;
1422                 NSRCADR(&itf->mask) = isc_if->netmask.type.in.s_addr;
1423
1424                 if (isc_if->flags & INTERFACE_F_BROADCAST) {
1425                         itf->flags |= INT_BROADCAST;
1426                         NSRCADR(&itf->bcast) =
1427                             isc_if->broadcast.type.in.s_addr;
1428                 }
1429         }
1430 #ifdef INCLUDE_IPV6_SUPPORT
1431         else if (IS_IPV6(&itf->sin)) {
1432                 SET_ADDR6N(&itf->sin, isc_if->address.type.in6);
1433                 SET_ADDR6N(&itf->mask, isc_if->netmask.type.in6);
1434
1435                 SET_SCOPE(&itf->sin, isc_if->address.zone);
1436         }
1437 #endif /* INCLUDE_IPV6_SUPPORT */
1438
1439
1440         /* Process the rest of the flags */
1441
1442         itf->flags |=
1443                   ((INTERFACE_F_UP & isc_if->flags)
1444                         ? INT_UP : 0)
1445                 | ((INTERFACE_F_LOOPBACK & isc_if->flags)
1446                         ? INT_LOOPBACK : 0)
1447                 | ((INTERFACE_F_POINTTOPOINT & isc_if->flags)
1448                         ? INT_PPP : 0)
1449                 | ((INTERFACE_F_MULTICAST & isc_if->flags)
1450                         ? INT_MULTICAST : 0)
1451                 | ((INTERFACE_F_PRIVACY & isc_if->flags)
1452                         ? INT_PRIVACY : 0)
1453                 ;
1454
1455         /*
1456          * Clear the loopback flag if the address is not localhost.
1457          * http://bugs.ntp.org/1683
1458          */
1459         if (INT_LOOPBACK & itf->flags) {
1460                 if (AF_INET == itf->family) {
1461                         if (127 != (SRCADR(&itf->sin) >> 24))
1462                                 itf->flags &= ~INT_LOOPBACK;
1463                 } else {
1464                         if (memcmp(v6loop, NSRCADR6(&itf->sin),
1465                                    sizeof(NSRCADR6(&itf->sin))))
1466                                 itf->flags &= ~INT_LOOPBACK;
1467                 }
1468         }
1469 }
1470
1471
1472 /*
1473  * refresh_interface
1474  *
1475  * some OSes have been observed to keep
1476  * cached routes even when more specific routes
1477  * become available.
1478  * this can be mitigated by re-binding
1479  * the socket.
1480  */
1481 static int
1482 refresh_interface(
1483         struct interface * interface
1484         )
1485 {
1486 #ifdef  OS_MISSES_SPECIFIC_ROUTE_UPDATES
1487         if (interface->fd != INVALID_SOCKET) {
1488                 int bcast = (interface->flags & INT_BCASTXMIT) != 0;
1489                 /* as we forcibly close() the socket remove the
1490                    broadcast permission indication */
1491                 if (bcast)
1492                         socket_broadcast_disable(interface, &interface->sin);
1493
1494                 close_and_delete_fd_from_list(interface->fd);
1495
1496                 /* create new socket picking up a new first hop binding
1497                    at connect() time */
1498                 interface->fd = open_socket(&interface->sin,
1499                                             bcast, 0, interface);
1500                  /*
1501                   * reset TTL indication so TTL is is set again
1502                   * next time around
1503                   */
1504                 interface->last_ttl = 0;
1505                 return (interface->fd != INVALID_SOCKET);
1506         } else
1507                 return 0;       /* invalid sockets are not refreshable */
1508 #else /* !OS_MISSES_SPECIFIC_ROUTE_UPDATES */
1509         return (interface->fd != INVALID_SOCKET);
1510 #endif /* !OS_MISSES_SPECIFIC_ROUTE_UPDATES */
1511 }
1512
1513 /*
1514  * interface_update - externally callable update function
1515  */
1516 void
1517 interface_update(
1518         interface_receiver_t    receiver,
1519         void *                  data)
1520 {
1521         int new_interface_found;
1522
1523         if (disable_dynamic_updates)
1524                 return;
1525
1526         BLOCKIO();
1527         new_interface_found = update_interfaces(NTP_PORT, receiver, data);
1528         UNBLOCKIO();
1529
1530         if (!new_interface_found)
1531                 return;
1532
1533 #ifdef DEBUG
1534         msyslog(LOG_DEBUG, "new interface(s) found: waking up resolver");
1535 #endif
1536         interrupt_worker_sleep();
1537 }
1538
1539
1540 /*
1541  * sau_from_netaddr() - convert network address on-wire formats.
1542  * Convert from libisc's isc_netaddr_t to NTP's sockaddr_u
1543  */
1544 void
1545 sau_from_netaddr(
1546         sockaddr_u *psau,
1547         const isc_netaddr_t *pna
1548         )
1549 {
1550         ZERO_SOCK(psau);
1551         AF(psau) = (u_short)pna->family;
1552         switch (pna->family) {
1553
1554         case AF_INET:
1555                 memcpy(&psau->sa4.sin_addr, &pna->type.in,
1556                        sizeof(psau->sa4.sin_addr));
1557                 break;
1558
1559         case AF_INET6:
1560                 memcpy(&psau->sa6.sin6_addr, &pna->type.in6,
1561                        sizeof(psau->sa6.sin6_addr));
1562                 break;
1563         }
1564 }
1565
1566
1567 static int
1568 is_wildcard_addr(
1569         const sockaddr_u *psau
1570         )
1571 {
1572         if (IS_IPV4(psau) && !NSRCADR(psau))
1573                 return 1;
1574
1575 #ifdef INCLUDE_IPV6_SUPPORT
1576         if (IS_IPV6(psau) && S_ADDR6_EQ(psau, &in6addr_any))
1577                 return 1;
1578 #endif
1579
1580         return 0;
1581 }
1582
1583
1584 #ifdef OS_NEEDS_REUSEADDR_FOR_IFADDRBIND
1585 /*
1586  * enable/disable re-use of wildcard address socket
1587  */
1588 static void
1589 set_wildcard_reuse(
1590         u_short family,
1591         int     on
1592         )
1593 {
1594         struct interface *any;
1595         SOCKET fd = INVALID_SOCKET;
1596
1597         any = ANY_INTERFACE_BYFAM(family);
1598         if (any != NULL)
1599                 fd = any->fd;
1600
1601         if (fd != INVALID_SOCKET) {
1602                 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
1603                                (void *)&on, sizeof(on)))
1604                         msyslog(LOG_ERR,
1605                                 "set_wildcard_reuse: setsockopt(SO_REUSEADDR, %s) failed: %m",
1606                                 on ? "on" : "off");
1607
1608                 DPRINTF(4, ("set SO_REUSEADDR to %s on %s\n",
1609                             on ? "on" : "off",
1610                             stoa(&any->sin)));
1611         }
1612 }
1613 #endif /* OS_NEEDS_REUSEADDR_FOR_IFADDRBIND */
1614
1615
1616 static isc_boolean_t
1617 check_flags6(
1618         sockaddr_u *psau,
1619         const char *name,
1620         u_int32 flags6
1621         )
1622 {
1623 #if defined(INCLUDE_IPV6_SUPPORT) && defined(SIOCGIFAFLAG_IN6)
1624         struct in6_ifreq ifr6;
1625         int fd;
1626
1627         if (psau->sa.sa_family != AF_INET6)
1628                 return ISC_FALSE;
1629         if ((fd = socket(AF_INET6, SOCK_DGRAM, 0)) < 0)
1630                 return ISC_FALSE;
1631         ZERO(ifr6);
1632         memcpy(&ifr6.ifr_addr, &psau->sa6, sizeof(ifr6.ifr_addr));
1633         strlcpy(ifr6.ifr_name, name, sizeof(ifr6.ifr_name));
1634         if (ioctl(fd, SIOCGIFAFLAG_IN6, &ifr6) < 0) {
1635                 close(fd);
1636                 return ISC_FALSE;
1637         }
1638         close(fd);
1639         if ((ifr6.ifr_ifru.ifru_flags6 & flags6) != 0)
1640                 return ISC_TRUE;
1641 #endif  /* INCLUDE_IPV6_SUPPORT && SIOCGIFAFLAG_IN6 */
1642         return ISC_FALSE;
1643 }
1644
1645 static isc_boolean_t
1646 is_anycast(
1647         sockaddr_u *psau,
1648         const char *name
1649         )
1650 {
1651 #ifdef IN6_IFF_ANYCAST
1652         return check_flags6(psau, name, IN6_IFF_ANYCAST);
1653 #else
1654         return ISC_FALSE;
1655 #endif
1656 }
1657
1658 static isc_boolean_t
1659 is_valid(
1660         sockaddr_u *psau,
1661         const char *name
1662         )
1663 {
1664         u_int32 flags6;
1665
1666         flags6 = 0;
1667 #ifdef IN6_IFF_DEPARTED
1668         flags6 |= IN6_IFF_DEPARTED;
1669 #endif
1670 #ifdef IN6_IFF_DETACHED
1671         flags6 |= IN6_IFF_DETACHED;
1672 #endif
1673 #ifdef IN6_IFF_TENTATIVE
1674         flags6 |= IN6_IFF_TENTATIVE;
1675 #endif
1676         return check_flags6(psau, name, flags6) ? ISC_FALSE : ISC_TRUE;
1677 }
1678
1679 /*
1680  * update_interface strategy
1681  *
1682  * toggle configuration phase
1683  *
1684  * Phase 1:
1685  * forall currently existing interfaces
1686  *   if address is known:
1687  *      drop socket - rebind again
1688  *
1689  *   if address is NOT known:
1690  *      attempt to create a new interface entry
1691  *
1692  * Phase 2:
1693  * forall currently known non MCAST and WILDCARD interfaces
1694  *   if interface does not match configuration phase (not seen in phase 1):
1695  *      remove interface from known interface list
1696  *      forall peers associated with this interface
1697  *         disconnect peer from this interface
1698  *
1699  * Phase 3:
1700  *   attempt to re-assign interfaces to peers
1701  *
1702  */
1703
1704 static int
1705 update_interfaces(
1706         u_short                 port,
1707         interface_receiver_t    receiver,
1708         void *                  data
1709         )
1710 {
1711         isc_mem_t *             mctx = (void *)-1;
1712         interface_info_t        ifi;
1713         isc_interfaceiter_t *   iter;
1714         isc_result_t            result;
1715         isc_interface_t         isc_if;
1716         int                     new_interface_found;
1717         unsigned int            family;
1718         endpt                   enumep;
1719         endpt *                 ep;
1720         endpt *                 next_ep;
1721
1722         DPRINTF(3, ("update_interfaces(%d)\n", port));
1723
1724         /*
1725          * phase one - scan interfaces
1726          * - create those that are not found
1727          * - update those that are found
1728          */
1729
1730         new_interface_found = FALSE;
1731         iter = NULL;
1732         result = isc_interfaceiter_create(mctx, &iter);
1733
1734         if (result != ISC_R_SUCCESS)
1735                 return 0;
1736
1737         /*
1738          * Toggle system interface scan phase to find untouched
1739          * interfaces to be deleted.
1740          */
1741         sys_interphase ^= 0x1;
1742
1743         for (result = isc_interfaceiter_first(iter);
1744              ISC_R_SUCCESS == result;
1745              result = isc_interfaceiter_next(iter)) {
1746
1747                 result = isc_interfaceiter_current(iter, &isc_if);
1748
1749                 if (result != ISC_R_SUCCESS)
1750                         break;
1751
1752                 /* See if we have a valid family to use */
1753                 family = isc_if.address.family;
1754                 if (AF_INET != family && AF_INET6 != family)
1755                         continue;
1756                 if (AF_INET == family && !ipv4_works)
1757                         continue;
1758                 if (AF_INET6 == family && !ipv6_works)
1759                         continue;
1760
1761                 /* create prototype */
1762                 init_interface(&enumep);
1763
1764                 convert_isc_if(&isc_if, &enumep, port);
1765
1766                 DPRINT_INTERFACE(4, (&enumep, "examining ", "\n"));
1767
1768                 /*
1769                  * Check if and how we are going to use the interface.
1770                  */
1771                 switch (interface_action(enumep.name, &enumep.sin,
1772                                          enumep.flags)) {
1773
1774                 case ACTION_IGNORE:
1775                         DPRINTF(4, ("ignoring interface %s (%s) - by nic rules\n",
1776                                     enumep.name, stoa(&enumep.sin)));
1777                         continue;
1778
1779                 case ACTION_LISTEN:
1780                         DPRINTF(4, ("listen interface %s (%s) - by nic rules\n",
1781                                     enumep.name, stoa(&enumep.sin)));
1782                         enumep.ignore_packets = ISC_FALSE;
1783                         break;
1784
1785                 case ACTION_DROP:
1786                         DPRINTF(4, ("drop on interface %s (%s) - by nic rules\n",
1787                                     enumep.name, stoa(&enumep.sin)));
1788                         enumep.ignore_packets = ISC_TRUE;
1789                         break;
1790                 }
1791
1792                  /* interfaces must be UP to be usable */
1793                 if (!(enumep.flags & INT_UP)) {
1794                         DPRINTF(4, ("skipping interface %s (%s) - DOWN\n",
1795                                     enumep.name, stoa(&enumep.sin)));
1796                         continue;
1797                 }
1798
1799                 /*
1800                  * skip any interfaces UP and bound to a wildcard
1801                  * address - some dhcp clients produce that in the
1802                  * wild
1803                  */
1804                 if (is_wildcard_addr(&enumep.sin))
1805                         continue;
1806
1807                 if (is_anycast(&enumep.sin, isc_if.name))
1808                         continue;
1809
1810                 /*
1811                  * skip any address that is an invalid state to be used
1812                  */
1813                 if (!is_valid(&enumep.sin, isc_if.name))
1814                         continue;
1815
1816                 /*
1817                  * map to local *address* in order to map all duplicate
1818                  * interfaces to an endpt structure with the appropriate
1819                  * socket.  Our name space is (ip-address), NOT
1820                  * (interface name, ip-address).
1821                  */
1822                 ep = getinterface(&enumep.sin, INT_WILDCARD);
1823
1824                 if (ep != NULL && refresh_interface(ep)) {
1825                         /*
1826                          * found existing and up to date interface -
1827                          * mark present.
1828                          */
1829                         if (ep->phase != sys_interphase) {
1830                                 /*
1831                                  * On a new round we reset the name so
1832                                  * the interface name shows up again if
1833                                  * this address is no longer shared.
1834                                  * We reset ignore_packets from the
1835                                  * new prototype to respect any runtime
1836                                  * changes to the nic rules.
1837                                  */
1838                                 strlcpy(ep->name, enumep.name,
1839                                         sizeof(ep->name));
1840                                 ep->ignore_packets =
1841                                             enumep.ignore_packets;
1842                         } else {
1843                                 /* name collision - rename interface */
1844                                 strlcpy(ep->name, "*multiple*",
1845                                         sizeof(ep->name));
1846                         }
1847
1848                         DPRINT_INTERFACE(4, (ep, "updating ",
1849                                              " present\n"));
1850
1851                         if (ep->ignore_packets !=
1852                             enumep.ignore_packets) {
1853                                 /*
1854                                  * We have conflicting configurations
1855                                  * for the interface address. This is
1856                                  * caused by using -I <interfacename>
1857                                  * for an interface that shares its
1858                                  * address with other interfaces. We
1859                                  * can not disambiguate incoming
1860                                  * packets delivered to this socket
1861                                  * without extra syscalls/features.
1862                                  * These are not (commonly) available.
1863                                  * Note this is a more unusual
1864                                  * configuration where several
1865                                  * interfaces share an address but
1866                                  * filtering via interface name is
1867                                  * attempted.  We resolve the
1868                                  * configuration conflict by disabling
1869                                  * the processing of received packets.
1870                                  * This leads to no service on the
1871                                  * interface address where the conflict
1872                                  * occurs.
1873                                  */
1874                                 msyslog(LOG_ERR,
1875                                         "WARNING: conflicting enable configuration for interfaces %s and %s for address %s - unsupported configuration - address DISABLED",
1876                                         enumep.name, ep->name,
1877                                         stoa(&enumep.sin));
1878
1879                                 ep->ignore_packets = ISC_TRUE;
1880                         }
1881
1882                         ep->phase = sys_interphase;
1883
1884                         ifi.action = IFS_EXISTS;
1885                         ifi.ep = ep;
1886                         if (receiver != NULL)
1887                                 (*receiver)(data, &ifi);
1888                 } else {
1889                         /*
1890                          * This is new or refreshing failed - add to
1891                          * our interface list.  If refreshing failed we
1892                          * will delete the interface structure in phase
1893                          * 2 as the interface was not marked current.
1894                          * We can bind to the address as the refresh
1895                          * code already closed the offending socket
1896                          */
1897                         ep = create_interface(port, &enumep);
1898
1899                         if (ep != NULL) {
1900                                 ifi.action = IFS_CREATED;
1901                                 ifi.ep = ep;
1902                                 if (receiver != NULL)
1903                                         (*receiver)(data, &ifi);
1904
1905                                 new_interface_found = TRUE;
1906                                 DPRINT_INTERFACE(3,
1907                                         (ep, "updating ",
1908                                          " new - created\n"));
1909                         } else {
1910                                 DPRINT_INTERFACE(3,
1911                                         (&enumep, "updating ",
1912                                          " new - creation FAILED"));
1913
1914                                 msyslog(LOG_INFO,
1915                                         "failed to init interface for address %s",
1916                                         stoa(&enumep.sin));
1917                                 continue;
1918                         }
1919                 }
1920         }
1921
1922         isc_interfaceiter_destroy(&iter);
1923
1924         /*
1925          * phase 2 - delete gone interfaces - reassigning peers to
1926          * other interfaces
1927          */
1928         for (ep = ep_list; ep != NULL; ep = next_ep) {
1929                 next_ep = ep->elink;
1930
1931                 /*
1932                  * if phase does not match sys_phase this interface was
1933                  * not enumerated during the last interface scan - so it
1934                  * is gone and will be deleted here unless it did not
1935                  * originate from interface enumeration (INT_WILDCARD,
1936                  * INT_MCASTIF).
1937                  */
1938                 if (((INT_WILDCARD | INT_MCASTIF) & ep->flags) ||
1939                     ep->phase == sys_interphase)
1940                         continue;
1941
1942                 DPRINT_INTERFACE(3, (ep, "updating ",
1943                                      "GONE - deleting\n"));
1944                 remove_interface(ep);
1945
1946                 ifi.action = IFS_DELETED;
1947                 ifi.ep = ep;
1948                 if (receiver != NULL)
1949                         (*receiver)(data, &ifi);
1950
1951                 /* disconnect peers from deleted endpt. */
1952                 while (ep->peers != NULL)
1953                         set_peerdstadr(ep->peers, NULL);
1954
1955                 /*
1956                  * update globals in case we lose
1957                  * a loopback interface
1958                  */
1959                 if (ep == loopback_interface)
1960                         loopback_interface = NULL;
1961
1962                 delete_interface(ep);
1963         }
1964
1965         /*
1966          * phase 3 - re-configure as the world has possibly changed
1967          *
1968          * never ever make this conditional again - it is needed to track
1969          * routing updates. see bug #2506
1970          */
1971         refresh_all_peerinterfaces();
1972
1973         if (broadcast_client_enabled)
1974                 io_setbclient();
1975
1976         if (sys_bclient)
1977                 io_setbclient();
1978
1979 #ifdef MCAST
1980         /*
1981          * Check multicast interfaces and try to join multicast groups if
1982          * not joined yet.
1983          */
1984         for (ep = ep_list; ep != NULL; ep = ep->elink) {
1985                 remaddr_t *entry;
1986
1987                 if (!(INT_MCASTIF & ep->flags) || (INT_MCASTOPEN & ep->flags))
1988                         continue;
1989
1990                 /* Find remote address that was linked to this interface */
1991                 for (entry = remoteaddr_list;
1992                      entry != NULL;
1993                      entry = entry->link) {
1994                         if (entry->ep == ep) {
1995                                 if (socket_multicast_enable(ep, &entry->addr)) {
1996                                         msyslog(LOG_INFO,
1997                                                 "Joined %s socket to multicast group %s",
1998                                                 stoa(&ep->sin),
1999                                                 stoa(&entry->addr));
2000                                 }
2001                                 break;
2002                         }
2003                 }
2004         }
2005 #endif /* MCAST */
2006
2007         return new_interface_found;
2008 }
2009
2010
2011 /*
2012  * create_sockets - create a socket for each interface plus a default
2013  *                      socket for when we don't know where to send
2014  */
2015 static int
2016 create_sockets(
2017         u_short port
2018         )
2019 {
2020 #ifndef HAVE_IO_COMPLETION_PORT
2021         /*
2022          * I/O Completion Ports don't care about the select and FD_SET
2023          */
2024         maxactivefd = 0;
2025         FD_ZERO(&activefds);
2026 #endif
2027
2028         DPRINTF(2, ("create_sockets(%d)\n", port));
2029
2030         create_wildcards(port);
2031
2032         update_interfaces(port, NULL, NULL);
2033
2034         /*
2035          * Now that we have opened all the sockets, turn off the reuse
2036          * flag for security.
2037          */
2038         set_reuseaddr(0);
2039
2040         DPRINTF(2, ("create_sockets: Total interfaces = %d\n", ninterfaces));
2041
2042         return ninterfaces;
2043 }
2044
2045 /*
2046  * create_interface - create a new interface for a given prototype
2047  *                    binding the socket.
2048  */
2049 static struct interface *
2050 create_interface(
2051         u_short                 port,
2052         struct interface *      protot
2053         )
2054 {
2055         sockaddr_u      resmask;
2056         endpt *         iface;
2057 #if defined(MCAST) && defined(MULTICAST_NONEWSOCKET)
2058         remaddr_t *     entry;
2059         remaddr_t *     next_entry;
2060 #endif
2061         DPRINTF(2, ("create_interface(%s#%d)\n", stoa(&protot->sin),
2062                     port));
2063
2064         /* build an interface */
2065         iface = new_interface(protot);
2066
2067         /*
2068          * create socket
2069          */
2070         iface->fd = open_socket(&iface->sin, 0, 0, iface);
2071
2072         if (iface->fd != INVALID_SOCKET)
2073                 log_listen_address(iface);
2074
2075         if ((INT_BROADCAST & iface->flags)
2076             && iface->bfd != INVALID_SOCKET)
2077                 msyslog(LOG_INFO, "Listening on broadcast address %s#%d",
2078                         stoa((&iface->bcast)), port);
2079
2080         if (INVALID_SOCKET == iface->fd
2081             && INVALID_SOCKET == iface->bfd) {
2082                 msyslog(LOG_ERR, "unable to create socket on %s (%d) for %s#%d",
2083                         iface->name,
2084                         iface->ifnum,
2085                         stoa((&iface->sin)),
2086                         port);
2087                 delete_interface(iface);
2088                 return NULL;
2089         }
2090
2091         /*
2092          * Blacklist our own addresses, no use talking to ourself
2093          */
2094         SET_HOSTMASK(&resmask, AF(&iface->sin));
2095         hack_restrict(RESTRICT_FLAGS, &iface->sin, &resmask,
2096                       -4, RESM_NTPONLY | RESM_INTERFACE, RES_IGNORE, 0);
2097
2098         /*
2099          * set globals with the first found
2100          * loopback interface of the appropriate class
2101          */
2102         if (NULL == loopback_interface && AF_INET == iface->family
2103             && (INT_LOOPBACK & iface->flags))
2104                 loopback_interface = iface;
2105
2106         /*
2107          * put into our interface list
2108          */
2109         add_addr_to_list(&iface->sin, iface);
2110         add_interface(iface);
2111
2112 #if defined(MCAST) && defined(MULTICAST_NONEWSOCKET)
2113         /*
2114          * Join any previously-configured compatible multicast groups.
2115          */
2116         if (INT_MULTICAST & iface->flags &&
2117             !((INT_LOOPBACK | INT_WILDCARD) & iface->flags) &&
2118             !iface->ignore_packets) {
2119                 for (entry = remoteaddr_list;
2120                      entry != NULL;
2121                      entry = next_entry) {
2122                         next_entry = entry->link;
2123                         if (AF(&iface->sin) != AF(&entry->addr) ||
2124                             !IS_MCAST(&entry->addr))
2125                                 continue;
2126                         if (socket_multicast_enable(iface,
2127                                                     &entry->addr))
2128                                 msyslog(LOG_INFO,
2129                                         "Joined %s socket to multicast group %s",
2130                                         stoa(&iface->sin),
2131                                         stoa(&entry->addr));
2132                         else
2133                                 msyslog(LOG_ERR,
2134                                         "Failed to join %s socket to multicast group %s",
2135                                         stoa(&iface->sin),
2136                                         stoa(&entry->addr));
2137                 }
2138         }
2139 #endif  /* MCAST && MCAST_NONEWSOCKET */
2140
2141         DPRINT_INTERFACE(2, (iface, "created ", "\n"));
2142         return iface;
2143 }
2144
2145
2146 #ifdef SO_EXCLUSIVEADDRUSE
2147 static void
2148 set_excladdruse(
2149         SOCKET fd
2150         )
2151 {
2152         int one = 1;
2153         int failed;
2154 #ifdef SYS_WINNT
2155         DWORD err;
2156 #endif
2157
2158         failed = setsockopt(fd, SOL_SOCKET, SO_EXCLUSIVEADDRUSE,
2159                             (void *)&one, sizeof(one));
2160
2161         if (!failed)
2162                 return;
2163
2164 #ifdef SYS_WINNT
2165         /*
2166          * Prior to Windows XP setting SO_EXCLUSIVEADDRUSE can fail with
2167          * error WSAINVAL depending on service pack level and whether
2168          * the user account is in the Administrators group.  Do not
2169          * complain if it fails that way on versions prior to XP (5.1).
2170          */
2171         err = GetLastError();
2172
2173         if (isc_win32os_versioncheck(5, 1, 0, 0) < 0    /* < 5.1/XP */
2174             && WSAEINVAL == err)
2175                 return;
2176
2177         SetLastError(err);
2178 #endif
2179         msyslog(LOG_ERR,
2180                 "setsockopt(%d, SO_EXCLUSIVEADDRUSE, on): %m",
2181                 (int)fd);
2182 }
2183 #endif  /* SO_EXCLUSIVEADDRUSE */
2184
2185
2186 /*
2187  * set_reuseaddr() - set/clear REUSEADDR on all sockets
2188  *                      NB possible hole - should we be doing this on broadcast
2189  *                      fd's also?
2190  */
2191 static void
2192 set_reuseaddr(
2193         int flag
2194         )
2195 {
2196 #ifndef SO_EXCLUSIVEADDRUSE
2197         endpt *ep;
2198
2199         for (ep = ep_list; ep != NULL; ep = ep->elink) {
2200                 if (ep->flags & INT_WILDCARD)
2201                         continue;
2202
2203                 /*
2204                  * if ep->fd  is INVALID_SOCKET, we might have a adapter
2205                  * configured but not present
2206                  */
2207                 DPRINTF(4, ("setting SO_REUSEADDR on %.16s@%s to %s\n",
2208                             ep->name, stoa(&ep->sin),
2209                             flag ? "on" : "off"));
2210
2211                 if (ep->fd != INVALID_SOCKET) {
2212                         if (setsockopt(ep->fd, SOL_SOCKET, SO_REUSEADDR,
2213                                        (void *)&flag, sizeof(flag))) {
2214                                 msyslog(LOG_ERR, "set_reuseaddr: setsockopt(%s, SO_REUSEADDR, %s) failed: %m",
2215                                         stoa(&ep->sin), flag ? "on" : "off");
2216                         }
2217                 }
2218         }
2219 #endif /* ! SO_EXCLUSIVEADDRUSE */
2220 }
2221
2222 /*
2223  * This is just a wrapper around an internal function so we can
2224  * make other changes as necessary later on
2225  */
2226 void
2227 enable_broadcast(
2228         struct interface *      iface,
2229         sockaddr_u *            baddr
2230         )
2231 {
2232 #ifdef OPEN_BCAST_SOCKET
2233         socket_broadcast_enable(iface, iface->fd, baddr);
2234 #endif
2235 }
2236
2237 #ifdef OPEN_BCAST_SOCKET
2238 /*
2239  * Enable a broadcast address to a given socket
2240  * The socket is in the ep_list all we need to do is enable
2241  * broadcasting. It is not this function's job to select the socket
2242  */
2243 static isc_boolean_t
2244 socket_broadcast_enable(
2245         struct interface *      iface,
2246         SOCKET                  fd,
2247         sockaddr_u *            baddr
2248         )
2249 {
2250 #ifdef SO_BROADCAST
2251         int on = 1;
2252
2253         if (IS_IPV4(baddr)) {
2254                 /* if this interface can support broadcast, set SO_BROADCAST */
2255                 if (setsockopt(fd, SOL_SOCKET, SO_BROADCAST,
2256                                (void *)&on, sizeof(on)))
2257                         msyslog(LOG_ERR,
2258                                 "setsockopt(SO_BROADCAST) enable failure on address %s: %m",
2259                                 stoa(baddr));
2260                 else
2261                         DPRINTF(2, ("Broadcast enabled on socket %d for address %s\n",
2262                                     fd, stoa(baddr)));
2263         }
2264         iface->flags |= INT_BCASTXMIT;
2265         return ISC_TRUE;
2266 #else
2267         return ISC_FALSE;
2268 #endif /* SO_BROADCAST */
2269 }
2270
2271 #ifdef  OS_MISSES_SPECIFIC_ROUTE_UPDATES
2272 /*
2273  * Remove a broadcast address from a given socket
2274  * The socket is in the ep_list all we need to do is disable
2275  * broadcasting. It is not this function's job to select the socket
2276  */
2277 static isc_boolean_t
2278 socket_broadcast_disable(
2279         struct interface *      iface,
2280         sockaddr_u *            baddr
2281         )
2282 {
2283 #ifdef SO_BROADCAST
2284         int off = 0;    /* This seems to be OK as an int */
2285
2286         if (IS_IPV4(baddr) && setsockopt(iface->fd, SOL_SOCKET,
2287             SO_BROADCAST, (void *)&off, sizeof(off)))
2288                 msyslog(LOG_ERR,
2289                         "setsockopt(SO_BROADCAST) disable failure on address %s: %m",
2290                         stoa(baddr));
2291
2292         iface->flags &= ~INT_BCASTXMIT;
2293         return ISC_TRUE;
2294 #else
2295         return ISC_FALSE;
2296 #endif /* SO_BROADCAST */
2297 }
2298 #endif /* OS_MISSES_SPECIFIC_ROUTE_UPDATES */
2299
2300 #endif /* OPEN_BCAST_SOCKET */
2301
2302 /*
2303  * return the broadcast client flag value
2304  */
2305 isc_boolean_t
2306 get_broadcastclient_flag(void)
2307 {
2308         return (broadcast_client_enabled);
2309 }
2310
2311 /*
2312  * Check to see if the address is a multicast address
2313  */
2314 static isc_boolean_t
2315 addr_ismulticast(
2316         sockaddr_u *maddr
2317         )
2318 {
2319         isc_boolean_t result;
2320
2321 #ifndef INCLUDE_IPV6_MULTICAST_SUPPORT
2322         /*
2323          * If we don't have IPV6 support any IPV6 addr is not multicast
2324          */
2325         if (IS_IPV6(maddr))
2326                 result = ISC_FALSE;
2327         else
2328 #endif
2329                 result = IS_MCAST(maddr);
2330
2331         if (!result)
2332                 DPRINTF(4, ("address %s is not multicast\n",
2333                             stoa(maddr)));
2334
2335         return result;
2336 }
2337
2338 /*
2339  * Multicast servers need to set the appropriate Multicast interface
2340  * socket option in order for it to know which interface to use for
2341  * send the multicast packet.
2342  */
2343 void
2344 enable_multicast_if(
2345         struct interface *      iface,
2346         sockaddr_u *            maddr
2347         )
2348 {
2349 #ifdef MCAST
2350 #ifdef IP_MULTICAST_LOOP
2351         TYPEOF_IP_MULTICAST_LOOP off = 0;
2352 #endif
2353 #if defined(INCLUDE_IPV6_MULTICAST_SUPPORT) && defined(IPV6_MULTICAST_LOOP)
2354         u_int off6 = 0;
2355 #endif
2356
2357         REQUIRE(AF(maddr) == AF(&iface->sin));
2358
2359         switch (AF(&iface->sin)) {
2360
2361         case AF_INET:
2362 #ifdef IP_MULTICAST_LOOP
2363                 /*
2364                  * Don't send back to itself, but allow failure to set
2365                  */
2366                 if (setsockopt(iface->fd, IPPROTO_IP,
2367                                IP_MULTICAST_LOOP,
2368                                (void *)&off,
2369                                sizeof(off))) {
2370
2371                         msyslog(LOG_ERR,
2372                                 "setsockopt IP_MULTICAST_LOOP failed: %m on socket %d, addr %s for multicast address %s",
2373                                 iface->fd, stoa(&iface->sin),
2374                                 stoa(maddr));
2375                 }
2376 #endif
2377                 break;
2378
2379         case AF_INET6:
2380 #ifdef INCLUDE_IPV6_MULTICAST_SUPPORT
2381 #ifdef IPV6_MULTICAST_LOOP
2382                 /*
2383                  * Don't send back to itself, but allow failure to set
2384                  */
2385                 if (setsockopt(iface->fd, IPPROTO_IPV6,
2386                                IPV6_MULTICAST_LOOP,
2387                                (void *) &off6, sizeof(off6))) {
2388
2389                         msyslog(LOG_ERR,
2390                                 "setsockopt IPV6_MULTICAST_LOOP failed: %m on socket %d, addr %s for multicast address %s",
2391                                 iface->fd, stoa(&iface->sin),
2392                                 stoa(maddr));
2393                 }
2394 #endif
2395                 break;
2396 #else
2397                 return;
2398 #endif  /* INCLUDE_IPV6_MULTICAST_SUPPORT */
2399         }
2400         return;
2401 #endif
2402 }
2403
2404 /*
2405  * Add a multicast address to a given socket
2406  * The socket is in the ep_list all we need to do is enable
2407  * multicasting. It is not this function's job to select the socket
2408  */
2409 #if defined(MCAST)
2410 static isc_boolean_t
2411 socket_multicast_enable(
2412         endpt *         iface,
2413         sockaddr_u *    maddr
2414         )
2415 {
2416         struct ip_mreq          mreq;
2417 # ifdef INCLUDE_IPV6_MULTICAST_SUPPORT
2418         struct ipv6_mreq        mreq6;
2419 # endif
2420         switch (AF(maddr)) {
2421
2422         case AF_INET:
2423                 ZERO(mreq);
2424                 mreq.imr_multiaddr = SOCK_ADDR4(maddr);
2425                 mreq.imr_interface.s_addr = htonl(INADDR_ANY);
2426                 if (setsockopt(iface->fd,
2427                                IPPROTO_IP,
2428                                IP_ADD_MEMBERSHIP,
2429                                (void *)&mreq,
2430                                sizeof(mreq))) {
2431                         DPRINTF(2, (
2432                                 "setsockopt IP_ADD_MEMBERSHIP failed: %m on socket %d, addr %s for %x / %x (%s)",
2433                                 iface->fd, stoa(&iface->sin),
2434                                 mreq.imr_multiaddr.s_addr,
2435                                 mreq.imr_interface.s_addr,
2436                                 stoa(maddr)));
2437                         return ISC_FALSE;
2438                 }
2439                 DPRINTF(4, ("Added IPv4 multicast membership on socket %d, addr %s for %x / %x (%s)\n",
2440                             iface->fd, stoa(&iface->sin),
2441                             mreq.imr_multiaddr.s_addr,
2442                             mreq.imr_interface.s_addr, stoa(maddr)));
2443                 break;
2444
2445         case AF_INET6:
2446 # ifdef INCLUDE_IPV6_MULTICAST_SUPPORT
2447                 /*
2448                  * Enable reception of multicast packets.
2449                  * If the address is link-local we can get the
2450                  * interface index from the scope id. Don't do this
2451                  * for other types of multicast addresses. For now let
2452                  * the kernel figure it out.
2453                  */
2454                 ZERO(mreq6);
2455                 mreq6.ipv6mr_multiaddr = SOCK_ADDR6(maddr);
2456                 mreq6.ipv6mr_interface = iface->ifindex;
2457
2458                 if (setsockopt(iface->fd, IPPROTO_IPV6,
2459                                IPV6_JOIN_GROUP, (void *)&mreq6,
2460                                sizeof(mreq6))) {
2461                         DPRINTF(2, (
2462                                 "setsockopt IPV6_JOIN_GROUP failed: %m on socket %d, addr %s for interface %u (%s)",
2463                                 iface->fd, stoa(&iface->sin),
2464                                 mreq6.ipv6mr_interface, stoa(maddr)));
2465                         return ISC_FALSE;
2466                 }
2467                 DPRINTF(4, ("Added IPv6 multicast group on socket %d, addr %s for interface %u (%s)\n",
2468                             iface->fd, stoa(&iface->sin),
2469                             mreq6.ipv6mr_interface, stoa(maddr)));
2470 # else
2471                 return ISC_FALSE;
2472 # endif /* INCLUDE_IPV6_MULTICAST_SUPPORT */
2473         }
2474         iface->flags |= INT_MCASTOPEN;
2475         iface->num_mcast++;
2476
2477         return ISC_TRUE;
2478 }
2479 #endif  /* MCAST */
2480
2481
2482 /*
2483  * Remove a multicast address from a given socket
2484  * The socket is in the ep_list all we need to do is disable
2485  * multicasting. It is not this function's job to select the socket
2486  */
2487 #ifdef MCAST
2488 static isc_boolean_t
2489 socket_multicast_disable(
2490         struct interface *      iface,
2491         sockaddr_u *            maddr
2492         )
2493 {
2494 # ifdef INCLUDE_IPV6_MULTICAST_SUPPORT
2495         struct ipv6_mreq mreq6;
2496 # endif
2497         struct ip_mreq mreq;
2498
2499         ZERO(mreq);
2500
2501         if (find_addr_in_list(maddr) == NULL) {
2502                 DPRINTF(4, ("socket_multicast_disable(%s): not found\n",
2503                             stoa(maddr)));
2504                 return ISC_TRUE;
2505         }
2506
2507         switch (AF(maddr)) {
2508
2509         case AF_INET:
2510                 mreq.imr_multiaddr = SOCK_ADDR4(maddr);
2511                 mreq.imr_interface = SOCK_ADDR4(&iface->sin);
2512                 if (setsockopt(iface->fd, IPPROTO_IP,
2513                                IP_DROP_MEMBERSHIP, (void *)&mreq,
2514                                sizeof(mreq))) {
2515
2516                         msyslog(LOG_ERR,
2517                                 "setsockopt IP_DROP_MEMBERSHIP failed: %m on socket %d, addr %s for %x / %x (%s)",
2518                                 iface->fd, stoa(&iface->sin),
2519                                 SRCADR(maddr), SRCADR(&iface->sin),
2520                                 stoa(maddr));
2521                         return ISC_FALSE;
2522                 }
2523                 break;
2524         case AF_INET6:
2525 # ifdef INCLUDE_IPV6_MULTICAST_SUPPORT
2526                 /*
2527                  * Disable reception of multicast packets
2528                  * If the address is link-local we can get the
2529                  * interface index from the scope id.  Don't do this
2530                  * for other types of multicast addresses. For now let
2531                  * the kernel figure it out.
2532                  */
2533                 mreq6.ipv6mr_multiaddr = SOCK_ADDR6(maddr);
2534                 mreq6.ipv6mr_interface = iface->ifindex;
2535
2536                 if (setsockopt(iface->fd, IPPROTO_IPV6,
2537                                IPV6_LEAVE_GROUP, (void *)&mreq6,
2538                                sizeof(mreq6))) {
2539
2540                         msyslog(LOG_ERR,
2541                                 "setsockopt IPV6_LEAVE_GROUP failure: %m on socket %d, addr %s for %d (%s)",
2542                                 iface->fd, stoa(&iface->sin),
2543                                 iface->ifindex, stoa(maddr));
2544                         return ISC_FALSE;
2545                 }
2546                 break;
2547 # else
2548                 return ISC_FALSE;
2549 # endif /* INCLUDE_IPV6_MULTICAST_SUPPORT */
2550         }
2551
2552         iface->num_mcast--;
2553         if (!iface->num_mcast)
2554                 iface->flags &= ~INT_MCASTOPEN;
2555
2556         return ISC_TRUE;
2557 }
2558 #endif  /* MCAST */
2559
2560 /*
2561  * io_setbclient - open the broadcast client sockets
2562  */
2563 void
2564 io_setbclient(void)
2565 {
2566 #ifdef OPEN_BCAST_SOCKET
2567         struct interface *      interf;
2568         unsigned int            nif;
2569
2570         nif = 0;
2571         set_reuseaddr(1);
2572
2573         for (interf = ep_list;
2574              interf != NULL;
2575              interf = interf->elink) {
2576
2577                 if (interf->flags & (INT_WILDCARD | INT_LOOPBACK))
2578                         continue;
2579
2580                 /* use only allowed addresses */
2581                 if (interf->ignore_packets)
2582                         continue;
2583
2584                 /* Need a broadcast-capable interface */
2585                 if (!(interf->flags & INT_BROADCAST))
2586                         continue;
2587
2588                 /* Only IPv4 addresses are valid for broadcast */
2589                 REQUIRE(IS_IPV4(&interf->bcast));
2590
2591                 /* Do we already have the broadcast address open? */
2592                 if (interf->flags & INT_BCASTOPEN) {
2593                         /*
2594                          * account for already open interfaces to avoid
2595                          * misleading warning below
2596                          */
2597                         nif++;
2598                         continue;
2599                 }
2600
2601                 /*
2602                  * Try to open the broadcast address
2603                  */
2604                 interf->family = AF_INET;
2605                 interf->bfd = open_socket(&interf->bcast, 1, 0, interf);
2606
2607                 /*
2608                  * If we succeeded then we use it otherwise enable
2609                  * broadcast on the interface address
2610                  */
2611                 if (interf->bfd != INVALID_SOCKET) {
2612                         nif++;
2613                         interf->flags |= INT_BCASTOPEN;
2614                         msyslog(LOG_INFO,
2615                                 "Listen for broadcasts to %s on interface #%d %s",
2616                                 stoa(&interf->bcast), interf->ifnum, interf->name);
2617                 } else switch (errno) {
2618                         /* Silently ignore EADDRINUSE as we probably
2619                          * opened the socket already for an address in
2620                          * the same network */
2621                 case EADDRINUSE:
2622                         /* Some systems cannot bind a socket to a broadcast
2623                          * address, as that is not a valid host address. */
2624                 case EADDRNOTAVAIL:
2625 #                   ifdef SYS_WINNT     /*TODO: use for other systems, too? */
2626                         /* avoid recurrence here -- if we already have a
2627                          * regular socket, it's quite useless to try this
2628                          * again.
2629                          */
2630                         if (interf->fd != INVALID_SOCKET) {
2631                                 interf->flags |= INT_BCASTOPEN;
2632                                 nif++;
2633                         }
2634 #                   endif
2635                         break;
2636
2637                 default:
2638                         msyslog(LOG_INFO,
2639                                 "failed to listen for broadcasts to %s on interface #%d %s",
2640                                 stoa(&interf->bcast), interf->ifnum, interf->name);
2641                         break;
2642                 }
2643         }
2644         set_reuseaddr(0);
2645         if (nif != 0) {
2646                 broadcast_client_enabled = ISC_TRUE;
2647                 DPRINTF(1, ("io_setbclient: listening to %d broadcast addresses\n", nif));
2648         } else {
2649                 broadcast_client_enabled = ISC_FALSE;
2650                 msyslog(LOG_ERR,
2651                         "Unable to listen for broadcasts, no broadcast interfaces available");
2652         }
2653 #else
2654         msyslog(LOG_ERR,
2655                 "io_setbclient: Broadcast Client disabled by build");
2656 #endif  /* OPEN_BCAST_SOCKET */
2657 }
2658
2659 /*
2660  * io_unsetbclient - close the broadcast client sockets
2661  */
2662 void
2663 io_unsetbclient(void)
2664 {
2665         endpt *ep;
2666
2667         for (ep = ep_list; ep != NULL; ep = ep->elink) {
2668                 if (INT_WILDCARD & ep->flags)
2669                         continue;
2670                 if (!(INT_BCASTOPEN & ep->flags))
2671                         continue;
2672
2673                 if (ep->bfd != INVALID_SOCKET) {
2674                         /* destroy broadcast listening socket */
2675                         msyslog(LOG_INFO,
2676                                 "stop listening for broadcasts to %s on interface #%d %s",
2677                                 stoa(&ep->bcast), ep->ifnum, ep->name);
2678 #                   ifdef HAVE_IO_COMPLETION_PORT
2679                         io_completion_port_remove_socket(ep->bfd, ep);
2680 #                   endif
2681                         close_and_delete_fd_from_list(ep->bfd);
2682                         ep->bfd = INVALID_SOCKET;
2683                 }
2684                 ep->flags &= ~INT_BCASTOPEN;
2685         }
2686         broadcast_client_enabled = ISC_FALSE;
2687 }
2688
2689 /*
2690  * io_multicast_add() - add multicast group address
2691  */
2692 void
2693 io_multicast_add(
2694         sockaddr_u *addr
2695         )
2696 {
2697 #ifdef MCAST
2698         endpt * ep;
2699         endpt * one_ep;
2700
2701         /*
2702          * Check to see if this is a multicast address
2703          */
2704         if (!addr_ismulticast(addr))
2705                 return;
2706
2707         /* If we already have it we can just return */
2708         if (NULL != find_flagged_addr_in_list(addr, INT_MCASTOPEN)) {
2709                 msyslog(LOG_INFO,
2710                         "Duplicate request found for multicast address %s",
2711                         stoa(addr));
2712                 return;
2713         }
2714
2715 # ifndef MULTICAST_NONEWSOCKET
2716         ep = new_interface(NULL);
2717
2718         /*
2719          * Open a new socket for the multicast address
2720          */
2721         ep->sin = *addr;
2722         SET_PORT(&ep->sin, NTP_PORT);
2723         ep->family = AF(&ep->sin);
2724         AF(&ep->mask) = ep->family;
2725         SET_ONESMASK(&ep->mask);
2726
2727         set_reuseaddr(1);
2728         ep->bfd = INVALID_SOCKET;
2729         ep->fd = open_socket(&ep->sin, 0, 0, ep);
2730         if (ep->fd != INVALID_SOCKET) {
2731                 ep->ignore_packets = ISC_FALSE;
2732                 ep->flags |= INT_MCASTIF;
2733                 ep->ifindex = SCOPE(addr);
2734
2735                 strlcpy(ep->name, "multicast", sizeof(ep->name));
2736                 DPRINT_INTERFACE(2, (ep, "multicast add ", "\n"));
2737                 add_interface(ep);
2738                 log_listen_address(ep);
2739         } else {
2740                 /* bind failed, re-use wildcard interface */
2741                 delete_interface(ep);
2742
2743                 if (IS_IPV4(addr))
2744                         ep = wildipv4;
2745                 else if (IS_IPV6(addr))
2746                         ep = wildipv6;
2747                 else
2748                         ep = NULL;
2749
2750                 if (ep != NULL) {
2751                         /* HACK ! -- stuff in an address */
2752                         /* because we don't bind addr? DH */
2753                         ep->bcast = *addr;
2754                         msyslog(LOG_ERR,
2755                                 "multicast address %s using wildcard interface #%d %s",
2756                                 stoa(addr), ep->ifnum, ep->name);
2757                 } else {
2758                         msyslog(LOG_ERR,
2759                                 "No multicast socket available to use for address %s",
2760                                 stoa(addr));
2761                         return;
2762                 }
2763         }
2764         {       /* in place of the { following for in #else clause */
2765                 one_ep = ep;
2766 # else  /* MULTICAST_NONEWSOCKET follows */
2767         /*
2768          * For the case where we can't use a separate socket (Windows)
2769          * join each applicable endpoint socket to the group address.
2770          */
2771         if (IS_IPV4(addr))
2772                 one_ep = wildipv4;
2773         else
2774                 one_ep = wildipv6;
2775         for (ep = ep_list; ep != NULL; ep = ep->elink) {
2776                 if (ep->ignore_packets || AF(&ep->sin) != AF(addr) ||
2777                     !(INT_MULTICAST & ep->flags) ||
2778                     (INT_LOOPBACK | INT_WILDCARD) & ep->flags)
2779                         continue;
2780                 one_ep = ep;
2781 # endif /* MULTICAST_NONEWSOCKET */
2782                 if (socket_multicast_enable(ep, addr))
2783                         msyslog(LOG_INFO,
2784                                 "Joined %s socket to multicast group %s",
2785                                 stoa(&ep->sin),
2786                                 stoa(addr));
2787         }
2788
2789         add_addr_to_list(addr, one_ep);
2790 #else   /* !MCAST  follows*/
2791         msyslog(LOG_ERR,
2792                 "Can not add multicast address %s: no multicast support",
2793                 stoa(addr));
2794 #endif
2795         return;
2796 }
2797
2798
2799 /*
2800  * io_multicast_del() - delete multicast group address
2801  */
2802 void
2803 io_multicast_del(
2804         sockaddr_u *    addr
2805         )
2806 {
2807 #ifdef MCAST
2808         endpt *iface;
2809
2810         /*
2811          * Check to see if this is a multicast address
2812          */
2813         if (!addr_ismulticast(addr)) {
2814                 msyslog(LOG_ERR, "invalid multicast address %s",
2815                         stoa(addr));
2816                 return;
2817         }
2818
2819         /*
2820          * Disable reception of multicast packets
2821          */
2822         while ((iface = find_flagged_addr_in_list(addr, INT_MCASTOPEN))
2823                != NULL)
2824                 socket_multicast_disable(iface, addr);
2825
2826         delete_addr_from_list(addr);
2827
2828 #else /* not MCAST */
2829         msyslog(LOG_ERR,
2830                 "Can not delete multicast address %s: no multicast support",
2831                 stoa(addr));
2832 #endif /* not MCAST */
2833 }
2834
2835
2836 /*
2837  * open_socket - open a socket, returning the file descriptor
2838  */
2839
2840 static SOCKET
2841 open_socket(
2842         sockaddr_u *    addr,
2843         int             bcast,
2844         int             turn_off_reuse,
2845         endpt *         interf
2846         )
2847 {
2848         SOCKET  fd;
2849         int     errval;
2850         /*
2851          * int is OK for REUSEADR per
2852          * http://www.kohala.com/start/mcast.api.txt
2853          */
2854         int     on = 1;
2855         int     off = 0;
2856
2857         if (IS_IPV6(addr) && !ipv6_works)
2858                 return INVALID_SOCKET;
2859
2860         /* create a datagram (UDP) socket */
2861         fd = socket(AF(addr), SOCK_DGRAM, 0);
2862         if (INVALID_SOCKET == fd) {
2863                 errval = socket_errno();
2864                 msyslog(LOG_ERR,
2865                         "socket(AF_INET%s, SOCK_DGRAM, 0) failed on address %s: %m",
2866                         IS_IPV6(addr) ? "6" : "", stoa(addr));
2867
2868                 if (errval == EPROTONOSUPPORT ||
2869                     errval == EAFNOSUPPORT ||
2870                     errval == EPFNOSUPPORT)
2871                         return (INVALID_SOCKET);
2872
2873                 errno = errval;
2874                 msyslog(LOG_ERR,
2875                         "unexpected socket() error %m code %d (not EPROTONOSUPPORT nor EAFNOSUPPORT nor EPFNOSUPPORT) - exiting",
2876                         errno);
2877                 exit(1);
2878         }
2879
2880 #ifdef SYS_WINNT
2881         connection_reset_fix(fd, addr);
2882 #endif
2883         /*
2884          * Fixup the file descriptor for some systems
2885          * See bug #530 for details of the issue.
2886          */
2887         fd = move_fd(fd);
2888
2889         /*
2890          * set SO_REUSEADDR since we will be binding the same port
2891          * number on each interface according to turn_off_reuse.
2892          * This is undesirable on Windows versions starting with
2893          * Windows XP (numeric version 5.1).
2894          */
2895 #ifdef SYS_WINNT
2896         if (isc_win32os_versioncheck(5, 1, 0, 0) < 0)  /* before 5.1 */
2897 #endif
2898                 if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
2899                                (void *)((turn_off_reuse)
2900                                             ? &off
2901                                             : &on),
2902                                sizeof(on))) {
2903
2904                         msyslog(LOG_ERR,
2905                                 "setsockopt SO_REUSEADDR %s fails for address %s: %m",
2906                                 (turn_off_reuse)
2907                                     ? "off"
2908                                     : "on",
2909                                 stoa(addr));
2910                         closesocket(fd);
2911                         return INVALID_SOCKET;
2912                 }
2913 #ifdef SO_EXCLUSIVEADDRUSE
2914         /*
2915          * setting SO_EXCLUSIVEADDRUSE on the wildcard we open
2916          * first will cause more specific binds to fail.
2917          */
2918         if (!(interf->flags & INT_WILDCARD))
2919                 set_excladdruse(fd);
2920 #endif
2921
2922         /*
2923          * IPv4 specific options go here
2924          */
2925         if (IS_IPV4(addr)) {
2926 #if defined(IPPROTO_IP) && defined(IP_TOS)
2927                 if (setsockopt(fd, IPPROTO_IP, IP_TOS, (void *)&qos,
2928                                sizeof(qos)))
2929                         msyslog(LOG_ERR,
2930                                 "setsockopt IP_TOS (%02x) fails on address %s: %m",
2931                                 qos, stoa(addr));
2932 #endif /* IPPROTO_IP && IP_TOS */
2933                 if (bcast)
2934                         socket_broadcast_enable(interf, fd, addr);
2935         }
2936
2937         /*
2938          * IPv6 specific options go here
2939          */
2940         if (IS_IPV6(addr)) {
2941 #if defined(IPPROTO_IPV6) && defined(IPV6_TCLASS)
2942                 if (setsockopt(fd, IPPROTO_IPV6, IPV6_TCLASS, (void *)&qos,
2943                                sizeof(qos)))
2944                         msyslog(LOG_ERR,
2945                                 "setsockopt IPV6_TCLASS (%02x) fails on address %s: %m",
2946                                 qos, stoa(addr));
2947 #endif /* IPPROTO_IPV6 && IPV6_TCLASS */
2948 #ifdef IPV6_V6ONLY
2949                 if (isc_net_probe_ipv6only() == ISC_R_SUCCESS
2950                     && setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY,
2951                     (void *)&on, sizeof(on)))
2952                         msyslog(LOG_ERR,
2953                                 "setsockopt IPV6_V6ONLY on fails on address %s: %m",
2954                                 stoa(addr));
2955 #endif
2956 #ifdef IPV6_BINDV6ONLY
2957                 if (setsockopt(fd, IPPROTO_IPV6, IPV6_BINDV6ONLY,
2958                     (void *)&on, sizeof(on)))
2959                         msyslog(LOG_ERR,
2960                                 "setsockopt IPV6_BINDV6ONLY on fails on address %s: %m",
2961                                 stoa(addr));
2962 #endif
2963         }
2964
2965 #ifdef OS_NEEDS_REUSEADDR_FOR_IFADDRBIND
2966         /*
2967          * some OSes don't allow binding to more specific
2968          * addresses if a wildcard address already bound
2969          * to the port and SO_REUSEADDR is not set
2970          */
2971         if (!is_wildcard_addr(addr))
2972                 set_wildcard_reuse(AF(addr), 1);
2973 #endif
2974
2975         /*
2976          * bind the local address.
2977          */
2978         errval = bind(fd, &addr->sa, SOCKLEN(addr));
2979
2980 #ifdef OS_NEEDS_REUSEADDR_FOR_IFADDRBIND
2981         if (!is_wildcard_addr(addr))
2982                 set_wildcard_reuse(AF(addr), 0);
2983 #endif
2984
2985         if (errval < 0) {
2986                 /*
2987                  * Don't log this under all conditions
2988                  */
2989                 if (turn_off_reuse == 0
2990 #ifdef DEBUG
2991                     || debug > 1
2992 #endif
2993                     ) {
2994                         msyslog(LOG_ERR,
2995                                 "bind(%d) AF_INET%s %s#%d%s flags 0x%x failed: %m",
2996                                 fd, IS_IPV6(addr) ? "6" : "",
2997                                 stoa(addr), SRCPORT(addr),
2998                                 IS_MCAST(addr) ? " (multicast)" : "",
2999                                 interf->flags);
3000                 }
3001
3002                 closesocket(fd);
3003
3004                 return INVALID_SOCKET;
3005         }
3006
3007 #ifdef HAVE_TIMESTAMP
3008         {
3009                 if (setsockopt(fd, SOL_SOCKET, SO_TIMESTAMP,
3010                                (void *)&on, sizeof(on)))
3011                         msyslog(LOG_DEBUG,
3012                                 "setsockopt SO_TIMESTAMP on fails on address %s: %m",
3013                                 stoa(addr));
3014                 else
3015                         DPRINTF(4, ("setsockopt SO_TIMESTAMP enabled on fd %d address %s\n",
3016                                     fd, stoa(addr)));
3017         }
3018 #endif
3019 #ifdef HAVE_TIMESTAMPNS
3020         {
3021                 if (setsockopt(fd, SOL_SOCKET, SO_TIMESTAMPNS,
3022                                (void *)&on, sizeof(on)))
3023                         msyslog(LOG_DEBUG,
3024                                 "setsockopt SO_TIMESTAMPNS on fails on address %s: %m",
3025                                 stoa(addr));
3026                 else
3027                         DPRINTF(4, ("setsockopt SO_TIMESTAMPNS enabled on fd %d address %s\n",
3028                                     fd, stoa(addr)));
3029         }
3030 #endif
3031 #ifdef HAVE_BINTIME
3032         {
3033                 if (setsockopt(fd, SOL_SOCKET, SO_BINTIME,
3034                                (void *)&on, sizeof(on)))
3035                         msyslog(LOG_DEBUG,
3036                                 "setsockopt SO_BINTIME on fails on address %s: %m",
3037                                 stoa(addr));
3038                 else
3039                         DPRINTF(4, ("setsockopt SO_BINTIME enabled on fd %d address %s\n",
3040                                     fd, stoa(addr)));
3041         }
3042 #endif
3043
3044         DPRINTF(4, ("bind(%d) AF_INET%s, addr %s%%%d#%d, flags 0x%x\n",
3045                    fd, IS_IPV6(addr) ? "6" : "", stoa(addr),
3046                    SCOPE(addr), SRCPORT(addr), interf->flags));
3047
3048         make_socket_nonblocking(fd);
3049
3050 #ifdef HAVE_SIGNALED_IO
3051         init_socket_sig(fd);
3052 #endif /* not HAVE_SIGNALED_IO */
3053
3054         add_fd_to_list(fd, FD_TYPE_SOCKET);
3055
3056 #if !defined(SYS_WINNT) && !defined(VMS)
3057         DPRINTF(4, ("flags for fd %d: 0x%x\n", fd,
3058                     fcntl(fd, F_GETFL, 0)));
3059 #endif /* SYS_WINNT || VMS */
3060
3061 #if defined(HAVE_IO_COMPLETION_PORT)
3062 /*
3063  * Add the socket to the completion port
3064  */
3065         if (!io_completion_port_add_socket(fd, interf, bcast)) {
3066                 msyslog(LOG_ERR, "unable to set up io completion port - EXITING");
3067                 exit(1);
3068         }
3069 #endif
3070         return fd;
3071 }
3072
3073
3074
3075 /* XXX ELIMINATE sendpkt similar in ntpq.c, ntpdc.c, ntp_io.c, ntptrace.c */
3076 /*
3077  * sendpkt - send a packet to the specified destination. Maintain a
3078  * send error cache so that only the first consecutive error for a
3079  * destination is logged.
3080  */
3081 void
3082 sendpkt(
3083         sockaddr_u *            dest,
3084         struct interface *      ep,
3085         int                     ttl,
3086         struct pkt *            pkt,
3087         int                     len
3088         )
3089 {
3090         endpt * src;
3091         int     ismcast;
3092         int     cc;
3093         int     rc;
3094         u_char  cttl;
3095         l_fp    fp_zero = { 0, 0 };
3096
3097         ismcast = IS_MCAST(dest);
3098         if (!ismcast)
3099                 src = ep;
3100         else
3101                 src = (IS_IPV4(dest))
3102                           ? mc4_list
3103                           : mc6_list;
3104
3105         if (NULL == src) {
3106                 /*
3107                  * unbound peer - drop request and wait for better
3108                  * network conditions
3109                  */
3110                 DPRINTF(2, ("%ssendpkt(dst=%s, ttl=%d, len=%d): no interface - IGNORED\n",
3111                             ismcast ? "\tMCAST\t***** " : "",
3112                             stoa(dest), ttl, len));
3113                 return;
3114         }
3115
3116         do {
3117                 DPRINTF(2, ("%ssendpkt(%d, dst=%s, src=%s, ttl=%d, len=%d)\n",
3118                             ismcast ? "\tMCAST\t***** " : "", src->fd,
3119                             stoa(dest), stoa(&src->sin), ttl, len));
3120 #ifdef MCAST
3121                 /*
3122                  * for the moment we use the bcast option to set multicast ttl
3123                  */
3124                 if (ismcast && ttl > 0 && ttl != src->last_ttl) {
3125                         /*
3126                          * set the multicast ttl for outgoing packets
3127                          */
3128                         switch (AF(&src->sin)) {
3129
3130                         case AF_INET :
3131                                 cttl = (u_char)ttl;
3132                                 rc = setsockopt(src->fd, IPPROTO_IP,
3133                                                 IP_MULTICAST_TTL,
3134                                                 (void *)&cttl,
3135                                                 sizeof(cttl));
3136                                 break;
3137
3138 # ifdef INCLUDE_IPV6_SUPPORT
3139                         case AF_INET6 :
3140                                 rc = setsockopt(src->fd, IPPROTO_IPV6,
3141                                                  IPV6_MULTICAST_HOPS,
3142                                                  (void *)&ttl,
3143                                                  sizeof(ttl));
3144                                 break;
3145 # endif /* INCLUDE_IPV6_SUPPORT */
3146
3147                         default:
3148                                 rc = 0;
3149                         }
3150
3151                         if (!rc)
3152                                 src->last_ttl = ttl;
3153                         else
3154                                 msyslog(LOG_ERR,
3155                                         "setsockopt IP_MULTICAST_TTL/IPV6_MULTICAST_HOPS fails on address %s: %m",
3156                                         stoa(&src->sin));
3157                 }
3158 #endif  /* MCAST */
3159
3160 #ifdef SIM
3161                 cc = simulate_server(dest, src, pkt);
3162 #elif defined(HAVE_IO_COMPLETION_PORT)
3163                 cc = io_completion_port_sendto(src, src->fd, pkt,
3164                         (size_t)len, (sockaddr_u *)&dest->sa);
3165 #else
3166                 cc = sendto(src->fd, (char *)pkt, (u_int)len, 0,
3167                             &dest->sa, SOCKLEN(dest));
3168 #endif
3169                 if (cc == -1) {
3170                         src->notsent++;
3171                         packets_notsent++;
3172                 } else  {
3173                         src->sent++;
3174                         packets_sent++;
3175                 }
3176                 if (ismcast)
3177                         src = src->mclink;
3178         } while (ismcast && src != NULL);
3179
3180         /* HMS: pkt->rootdisp is usually random here */
3181         record_raw_stats(src ? &src->sin : NULL, dest,
3182                         &pkt->org, &pkt->rec, &pkt->xmt, &fp_zero,
3183                         PKT_MODE(pkt->li_vn_mode),
3184                         PKT_VERSION(pkt->li_vn_mode),
3185                         PKT_LEAP(pkt->li_vn_mode),
3186                         pkt->stratum,
3187                         pkt->ppoll, pkt->precision,
3188                         pkt->rootdelay, pkt->rootdisp, pkt->refid,
3189                         len - MIN_V4_PKT_LEN, (u_char *)&pkt->exten);
3190
3191         return;
3192 }
3193
3194
3195 #if !defined(HAVE_IO_COMPLETION_PORT)
3196 #if !defined(HAVE_SIGNALED_IO)
3197 /*
3198  * fdbits - generate ascii representation of fd_set (FAU debug support)
3199  * HFDF format - highest fd first.
3200  */
3201 static char *
3202 fdbits(
3203         int             count,
3204         const fd_set*   set
3205         )
3206 {
3207         static char buffer[256];
3208         char * buf = buffer;
3209
3210         count = min(count,  255);
3211
3212         while (count >= 0) {
3213                 *buf++ = FD_ISSET(count, set) ? '#' : '-';
3214                 count--;
3215         }
3216         *buf = '\0';
3217
3218         return buffer;
3219 }
3220 #endif
3221
3222 #ifdef REFCLOCK
3223 /*
3224  * Routine to read the refclock packets for a specific interface
3225  * Return the number of bytes read. That way we know if we should
3226  * read it again or go on to the next one if no bytes returned
3227  */
3228 static inline int
3229 read_refclock_packet(
3230         SOCKET                  fd,
3231         struct refclockio *     rp,
3232         l_fp                    ts
3233         )
3234 {
3235         u_int                   read_count;
3236         int                     buflen;
3237         int                     saved_errno;
3238         int                     consumed;
3239         struct recvbuf *        rb;
3240
3241         rb = get_free_recv_buffer();
3242
3243         if (NULL == rb) {
3244                 /*
3245                  * No buffer space available - just drop the packet
3246                  */
3247                 char buf[RX_BUFF_SIZE];
3248
3249                 buflen = read(fd, buf, sizeof buf);
3250                 packets_dropped++;
3251                 return (buflen);
3252         }
3253
3254         /* TALOS-CAN-0064: avoid signed/unsigned clashes that can lead
3255          * to buffer overrun and memory corruption
3256          */
3257         if (rp->datalen <= 0 || (size_t)rp->datalen > sizeof(rb->recv_space))
3258                 read_count = sizeof(rb->recv_space);
3259         else
3260                 read_count = (u_int)rp->datalen;
3261         do {
3262                 buflen = read(fd, (char *)&rb->recv_space, read_count);
3263         } while (buflen < 0 && EINTR == errno);
3264
3265         if (buflen <= 0) {
3266                 saved_errno = errno;
3267                 freerecvbuf(rb);
3268                 errno = saved_errno;
3269                 return buflen;
3270         }
3271
3272         /*
3273          * Got one. Mark how and when it got here,
3274          * put it on the full list and do bookkeeping.
3275          */
3276         rb->recv_length = buflen;
3277         rb->recv_peer = rp->srcclock;
3278         rb->dstadr = 0;
3279         rb->fd = fd;
3280         rb->recv_time = ts;
3281         rb->receiver = rp->clock_recv;
3282
3283         consumed = indicate_refclock_packet(rp, rb);
3284         if (!consumed) {
3285                 rp->recvcount++;
3286                 packets_received++;
3287         }
3288
3289         return buflen;
3290 }
3291 #endif  /* REFCLOCK */
3292
3293
3294 #ifdef HAVE_PACKET_TIMESTAMP
3295 /*
3296  * extract timestamps from control message buffer
3297  */
3298 static l_fp
3299 fetch_timestamp(
3300         struct recvbuf *        rb,
3301         struct msghdr *         msghdr,
3302         l_fp                    ts
3303         )
3304 {
3305         struct cmsghdr *        cmsghdr;
3306         unsigned long           ticks;
3307         double                  fuzz;
3308         l_fp                    lfpfuzz;
3309         l_fp                    nts;
3310 #ifdef DEBUG_TIMING
3311         l_fp                    dts;
3312 #endif
3313
3314         cmsghdr = CMSG_FIRSTHDR(msghdr);
3315         while (cmsghdr != NULL) {
3316                 switch (cmsghdr->cmsg_type)
3317                 {
3318 #ifdef HAVE_BINTIME
3319                 case SCM_BINTIME:
3320 #endif  /* HAVE_BINTIME */
3321 #ifdef HAVE_TIMESTAMPNS
3322                 case SCM_TIMESTAMPNS:
3323 #endif  /* HAVE_TIMESTAMPNS */
3324 #ifdef HAVE_TIMESTAMP
3325                 case SCM_TIMESTAMP:
3326 #endif  /* HAVE_TIMESTAMP */
3327 #if defined(HAVE_BINTIME) || defined (HAVE_TIMESTAMPNS) || defined(HAVE_TIMESTAMP)
3328                         switch (cmsghdr->cmsg_type)
3329                         {
3330 #ifdef HAVE_BINTIME
3331                         case SCM_BINTIME:
3332                                 {
3333                                         struct bintime  pbt;
3334                                         memcpy(&pbt, CMSG_DATA(cmsghdr), sizeof(pbt));
3335                                         /*
3336                                          * bintime documentation is at http://phk.freebsd.dk/pubs/timecounter.pdf
3337                                          */
3338                                         nts.l_i = pbt.sec + JAN_1970;
3339                                         nts.l_uf = (u_int32)(pbt.frac >> 32);
3340                                         if (sys_tick > measured_tick &&
3341                                             sys_tick > 1e-9) {
3342                                                 ticks = (unsigned long)(nts.l_uf / (unsigned long)(sys_tick * FRAC));
3343                                                 nts.l_uf = (unsigned long)(ticks * (unsigned long)(sys_tick * FRAC));
3344                                         }
3345                                         DPRINTF(4, ("fetch_timestamp: system bintime network time stamp: %ld.%09lu\n",
3346                                                     pbt.sec, (unsigned long)((nts.l_uf / FRAC) * 1e9)));
3347                                 }
3348                                 break;
3349 #endif  /* HAVE_BINTIME */
3350 #ifdef HAVE_TIMESTAMPNS
3351                         case SCM_TIMESTAMPNS:
3352                                 {
3353                                         struct timespec pts;
3354                                         memcpy(&pts, CMSG_DATA(cmsghdr), sizeof(pts));
3355                                         if (sys_tick > measured_tick &&
3356                                             sys_tick > 1e-9) {
3357                                                 ticks = (unsigned long)((pts.tv_nsec * 1e-9) /
3358                                                                         sys_tick);
3359                                                 pts.tv_nsec = (long)(ticks * 1e9 *
3360                                                                      sys_tick);
3361                                         }
3362                                         DPRINTF(4, ("fetch_timestamp: system nsec network time stamp: %ld.%09ld\n",
3363                                                     pts.tv_sec, pts.tv_nsec));
3364                                         nts = tspec_stamp_to_lfp(pts);
3365                                 }
3366                                 break;
3367 #endif  /* HAVE_TIMESTAMPNS */
3368 #ifdef HAVE_TIMESTAMP
3369                         case SCM_TIMESTAMP:
3370                                 {
3371                                         struct timeval  ptv;
3372                                         memcpy(&ptv, CMSG_DATA(cmsghdr), sizeof(ptv));
3373                                         if (sys_tick > measured_tick &&
3374                                             sys_tick > 1e-6) {
3375                                                 ticks = (unsigned long)((ptv.tv_usec * 1e-6) /
3376                                                                         sys_tick);
3377                                                 ptv.tv_usec = (long)(ticks * 1e6 *
3378                                                                     sys_tick);
3379                                         }
3380                                         DPRINTF(4, ("fetch_timestamp: system usec network time stamp: %jd.%06ld\n",
3381                                                     (intmax_t)ptv.tv_sec, (long)ptv.tv_usec));
3382                                         nts = tval_stamp_to_lfp(ptv);
3383                                 }
3384                                 break;
3385 #endif  /* HAVE_TIMESTAMP */
3386                         }
3387                         fuzz = ntp_random() * 2. / FRAC * sys_fuzz;
3388                         DTOLFP(fuzz, &lfpfuzz);
3389                         L_ADD(&nts, &lfpfuzz);
3390 #ifdef DEBUG_TIMING
3391                         dts = ts;
3392                         L_SUB(&dts, &nts);
3393                         collect_timing(rb, "input processing delay", 1,
3394                                        &dts);
3395                         DPRINTF(4, ("fetch_timestamp: timestamp delta: %s (incl. fuzz)\n",
3396                                     lfptoa(&dts, 9)));
3397 #endif  /* DEBUG_TIMING */
3398                         ts = nts;  /* network time stamp */
3399                         break;
3400 #endif  /* HAVE_BINTIME || HAVE_TIMESTAMPNS || HAVE_TIMESTAMP */
3401
3402                 default:
3403                         DPRINTF(4, ("fetch_timestamp: skipping control message 0x%x\n",
3404                                     cmsghdr->cmsg_type));
3405                 }
3406                 cmsghdr = CMSG_NXTHDR(msghdr, cmsghdr);
3407         }
3408         return ts;
3409 }
3410 #endif  /* HAVE_PACKET_TIMESTAMP */
3411
3412
3413 /*
3414  * Routine to read the network NTP packets for a specific interface
3415  * Return the number of bytes read. That way we know if we should
3416  * read it again or go on to the next one if no bytes returned
3417  */
3418 static inline int
3419 read_network_packet(
3420         SOCKET                  fd,
3421         struct interface *      itf,
3422         l_fp                    ts
3423         )
3424 {
3425         GETSOCKNAME_SOCKLEN_TYPE fromlen;
3426         int buflen;
3427         register struct recvbuf *rb;
3428 #ifdef HAVE_PACKET_TIMESTAMP
3429         struct msghdr msghdr;
3430         struct iovec iovec;
3431         char control[CMSG_BUFSIZE];
3432 #endif
3433
3434         /*
3435          * Get a buffer and read the frame.  If we
3436          * haven't got a buffer, or this is received
3437          * on a disallowed socket, just dump the
3438          * packet.
3439          */
3440
3441         rb = get_free_recv_buffer();
3442         if (NULL == rb || itf->ignore_packets) {
3443                 char buf[RX_BUFF_SIZE];
3444                 sockaddr_u from;
3445
3446                 if (rb != NULL)
3447                         freerecvbuf(rb);
3448
3449                 fromlen = sizeof(from);
3450                 buflen = recvfrom(fd, buf, sizeof(buf), 0,
3451                                   &from.sa, &fromlen);
3452                 DPRINTF(4, ("%s on (%lu) fd=%d from %s\n",
3453                         (itf->ignore_packets)
3454                             ? "ignore"
3455                             : "drop",
3456                         free_recvbuffs(), fd, stoa(&from)));
3457                 if (itf->ignore_packets)
3458                         packets_ignored++;
3459                 else
3460                         packets_dropped++;
3461                 return (buflen);
3462         }
3463
3464         fromlen = sizeof(rb->recv_srcadr);
3465
3466 #ifndef HAVE_PACKET_TIMESTAMP
3467         rb->recv_length = recvfrom(fd, (char *)&rb->recv_space,
3468                                    sizeof(rb->recv_space), 0,
3469                                    &rb->recv_srcadr.sa, &fromlen);
3470 #else
3471         iovec.iov_base        = &rb->recv_space;
3472         iovec.iov_len         = sizeof(rb->recv_space);
3473         msghdr.msg_name       = &rb->recv_srcadr;
3474         msghdr.msg_namelen    = fromlen;
3475         msghdr.msg_iov        = &iovec;
3476         msghdr.msg_iovlen     = 1;
3477         msghdr.msg_control    = (void *)&control;
3478         msghdr.msg_controllen = sizeof(control);
3479         msghdr.msg_flags      = 0;
3480         rb->recv_length       = recvmsg(fd, &msghdr, 0);
3481 #endif
3482
3483         buflen = rb->recv_length;
3484
3485         if (buflen == 0 || (buflen == -1 &&
3486             (EWOULDBLOCK == errno
3487 #ifdef EAGAIN
3488              || EAGAIN == errno
3489 #endif
3490              ))) {
3491                 freerecvbuf(rb);
3492                 return (buflen);
3493         } else if (buflen < 0) {
3494                 msyslog(LOG_ERR, "recvfrom(%s) fd=%d: %m",
3495                         stoa(&rb->recv_srcadr), fd);
3496                 DPRINTF(5, ("read_network_packet: fd=%d dropped (bad recvfrom)\n",
3497                             fd));
3498                 freerecvbuf(rb);
3499                 return (buflen);
3500         }
3501
3502         DPRINTF(3, ("read_network_packet: fd=%d length %d from %s\n",
3503                     fd, buflen, stoa(&rb->recv_srcadr)));
3504
3505 #ifdef ENABLE_BUG3020_FIX
3506         if (ISREFCLOCKADR(&rb->recv_srcadr)) {
3507                 msyslog(LOG_ERR, "recvfrom(%s) fd=%d: refclock srcadr on a network interface!",
3508                         stoa(&rb->recv_srcadr), fd);
3509                 DPRINTF(1, ("read_network_packet: fd=%d dropped (refclock srcadr))\n",
3510                             fd));
3511                 packets_dropped++;
3512                 freerecvbuf(rb);
3513                 return (buflen);
3514         }
3515 #endif
3516
3517         /*
3518         ** Bug 2672: Some OSes (MacOSX and Linux) don't block spoofed ::1
3519         */
3520
3521         if (AF_INET6 == itf->family) {
3522                 DPRINTF(2, ("Got an IPv6 packet, from <%s> (%d) to <%s> (%d)\n",
3523                         stoa(&rb->recv_srcadr),
3524                         IN6_IS_ADDR_LOOPBACK(PSOCK_ADDR6(&rb->recv_srcadr)),
3525                         stoa(&itf->sin),
3526                         !IN6_IS_ADDR_LOOPBACK(PSOCK_ADDR6(&itf->sin))
3527                         ));
3528
3529                 if (   IN6_IS_ADDR_LOOPBACK(PSOCK_ADDR6(&rb->recv_srcadr))
3530                     && !IN6_IS_ADDR_LOOPBACK(PSOCK_ADDR6(&itf->sin))
3531                    ) {
3532                         packets_dropped++;
3533                         DPRINTF(2, ("DROPPING that packet\n"));
3534                         freerecvbuf(rb);
3535                         return buflen;
3536                 }
3537                 DPRINTF(2, ("processing that packet\n"));
3538         }
3539
3540         /*
3541          * Got one.  Mark how and when it got here,
3542          * put it on the full list and do bookkeeping.
3543          */
3544         rb->dstadr = itf;
3545         rb->fd = fd;
3546 #ifdef HAVE_PACKET_TIMESTAMP
3547         /* pick up a network time stamp if possible */
3548         ts = fetch_timestamp(rb, &msghdr, ts);
3549 #endif
3550         rb->recv_time = ts;
3551         rb->receiver = receive;
3552
3553         add_full_recv_buffer(rb);
3554
3555         itf->received++;
3556         packets_received++;
3557         return (buflen);
3558 }
3559
3560 /*
3561  * attempt to handle io (select()/signaled IO)
3562  */
3563 void
3564 io_handler(void)
3565 {
3566 #  ifndef HAVE_SIGNALED_IO
3567         fd_set rdfdes;
3568         int nfound;
3569
3570         /*
3571          * Use select() on all on all input fd's for unlimited
3572          * time.  select() will terminate on SIGALARM or on the
3573          * reception of input.  Using select() means we can't do
3574          * robust signal handling and we get a potential race
3575          * between checking for alarms and doing the select().
3576          * Mostly harmless, I think.
3577          */
3578         /*
3579          * On VMS, I suspect that select() can't be interrupted
3580          * by a "signal" either, so I take the easy way out and
3581          * have select() time out after one second.
3582          * System clock updates really aren't time-critical,
3583          * and - lacking a hardware reference clock - I have
3584          * yet to learn about anything else that is.
3585          */
3586         ++handler_calls;
3587         rdfdes = activefds;
3588 #   if !defined(VMS) && !defined(SYS_VXWORKS)
3589         nfound = select(maxactivefd + 1, &rdfdes, NULL,
3590                         NULL, NULL);
3591 #   else        /* VMS, VxWorks */
3592         /* make select() wake up after one second */
3593         {
3594                 struct timeval t1;
3595                 t1.tv_sec  = 1;
3596                 t1.tv_usec = 0;
3597                 nfound = select(maxactivefd + 1,
3598                                 &rdfdes, NULL, NULL,
3599                                 &t1);
3600         }
3601 #   endif       /* VMS, VxWorks */
3602         if (nfound < 0 && sanitize_fdset(errno)) {
3603                 struct timeval t1;
3604                 t1.tv_sec  = 0;
3605                 t1.tv_usec = 0;
3606                 rdfdes = activefds;
3607                 nfound = select(maxactivefd + 1,
3608                                 &rdfdes, NULL, NULL,
3609                                 &t1);
3610         }
3611
3612         if (nfound > 0) {
3613                 l_fp ts;
3614
3615                 get_systime(&ts);
3616
3617                 input_handler_scan(&ts, &rdfdes);
3618         } else if (nfound == -1 && errno != EINTR) {
3619                 msyslog(LOG_ERR, "select() error: %m");
3620         }
3621 #   ifdef DEBUG
3622         else if (debug > 4) {
3623                 msyslog(LOG_DEBUG, "select(): nfound=%d, error: %m", nfound);
3624         } else {
3625                 DPRINTF(3, ("select() returned %d: %m\n", nfound));
3626         }
3627 #   endif /* DEBUG */
3628 #  else /* HAVE_SIGNALED_IO */
3629         wait_for_signal();
3630 #  endif /* HAVE_SIGNALED_IO */
3631 }
3632
3633 #ifdef HAVE_SIGNALED_IO
3634 /*
3635  * input_handler - receive packets asynchronously
3636  *
3637  * ALWAYS IN SIGNAL HANDLER CONTEXT -- only async-safe functions allowed!
3638  */
3639 static RETSIGTYPE
3640 input_handler(
3641         l_fp *  cts
3642         )
3643 {
3644         int             n;
3645         struct timeval  tvzero;
3646         fd_set          fds;
3647
3648         ++handler_calls;
3649
3650         /*
3651          * Do a poll to see who has data
3652          */
3653
3654         fds = activefds;
3655         tvzero.tv_sec = tvzero.tv_usec = 0;
3656
3657         n = select(maxactivefd + 1, &fds, NULL, NULL, &tvzero);
3658         if (n < 0 && sanitize_fdset(errno)) {
3659                 fds = activefds;
3660                 tvzero.tv_sec = tvzero.tv_usec = 0;
3661                 n = select(maxactivefd + 1, &fds, NULL, NULL, &tvzero);
3662         }
3663         if (n > 0)
3664                 input_handler_scan(cts, &fds);
3665 }
3666 #endif /* HAVE_SIGNALED_IO */
3667
3668
3669 /*
3670  * Try to sanitize the global FD set
3671  *
3672  * SIGNAL HANDLER CONTEXT if HAVE_SIGNALED_IO, ordinary userspace otherwise
3673  */
3674 static int/*BOOL*/
3675 sanitize_fdset(
3676         int     errc
3677         )
3678 {
3679         int j, b, maxscan;
3680
3681 #  ifndef HAVE_SIGNALED_IO
3682         /*
3683          * extended FAU debugging output
3684          */
3685         if (errc != EINTR) {
3686                 msyslog(LOG_ERR,
3687                         "select(%d, %s, 0L, 0L, &0.0) error: %m",
3688                         maxactivefd + 1,
3689                         fdbits(maxactivefd, &activefds));
3690         }
3691 #   endif
3692
3693         if (errc != EBADF)
3694                 return FALSE;
3695
3696         /* if we have oviously bad FDs, try to sanitize the FD set. */
3697         for (j = 0, maxscan = 0; j <= maxactivefd; j++) {
3698                 if (FD_ISSET(j, &activefds)) {
3699                         if (-1 != read(j, &b, 0)) {
3700                                 maxscan = j;
3701                                 continue;
3702                         }
3703 #                   ifndef HAVE_SIGNALED_IO
3704                         msyslog(LOG_ERR,
3705                                 "Removing bad file descriptor %d from select set",
3706                                 j);
3707 #                   endif
3708                         FD_CLR(j, &activefds);
3709                 }
3710         }
3711         if (maxactivefd != maxscan)
3712                 maxactivefd = maxscan;
3713         return TRUE;
3714 }
3715
3716 /*
3717  * scan the known FDs (clocks, servers, ...) for presence in a 'fd_set'.
3718  *
3719  * SIGNAL HANDLER CONTEXT if HAVE_SIGNALED_IO, ordinary userspace otherwise
3720  */
3721 static void
3722 input_handler_scan(
3723         const l_fp *    cts,
3724         const fd_set *  pfds
3725         )
3726 {
3727         int             buflen;
3728         u_int           idx;
3729         int             doing;
3730         SOCKET          fd;
3731         blocking_child *c;
3732         l_fp            ts;     /* Timestamp at BOselect() gob */
3733
3734 #if defined(DEBUG_TIMING)
3735         l_fp            ts_e;   /* Timestamp at EOselect() gob */
3736 #endif
3737         endpt *         ep;
3738 #ifdef REFCLOCK
3739         struct refclockio *rp;
3740         int             saved_errno;
3741         const char *    clk;
3742 #endif
3743 #ifdef HAS_ROUTING_SOCKET
3744         struct asyncio_reader * asyncio_reader;
3745         struct asyncio_reader * next_asyncio_reader;
3746 #endif
3747
3748         ++handler_pkts;
3749         ts = *cts;
3750
3751 #ifdef REFCLOCK
3752         /*
3753          * Check out the reference clocks first, if any
3754          */
3755
3756         for (rp = refio; rp != NULL; rp = rp->next) {
3757                 fd = rp->fd;
3758
3759                 if (!FD_ISSET(fd, pfds))
3760                         continue;
3761                 buflen = read_refclock_packet(fd, rp, ts);
3762                 /*
3763                  * The first read must succeed after select() indicates
3764                  * readability, or we've reached a permanent EOF.
3765                  * http://bugs.ntp.org/1732 reported ntpd munching CPU
3766                  * after a USB GPS was unplugged because select was
3767                  * indicating EOF but ntpd didn't remove the descriptor
3768                  * from the activefds set.
3769                  */
3770                 if (buflen < 0 && EAGAIN != errno) {
3771                         saved_errno = errno;
3772                         clk = refnumtoa(&rp->srcclock->srcadr);
3773                         errno = saved_errno;
3774                         msyslog(LOG_ERR, "%s read: %m", clk);
3775                         maintain_activefds(fd, TRUE);
3776                 } else if (0 == buflen) {
3777                         clk = refnumtoa(&rp->srcclock->srcadr);
3778                         msyslog(LOG_ERR, "%s read EOF", clk);
3779                         maintain_activefds(fd, TRUE);
3780                 } else {
3781                         /* drain any remaining refclock input */
3782                         do {
3783                                 buflen = read_refclock_packet(fd, rp, ts);
3784                         } while (buflen > 0);
3785                 }
3786         }
3787 #endif /* REFCLOCK */
3788
3789         /*
3790          * Loop through the interfaces looking for data to read.
3791          */
3792         for (ep = ep_list; ep != NULL; ep = ep->elink) {
3793                 for (doing = 0; doing < 2; doing++) {
3794                         if (!doing) {
3795                                 fd = ep->fd;
3796                         } else {
3797                                 if (!(ep->flags & INT_BCASTOPEN))
3798                                         break;
3799                                 fd = ep->bfd;
3800                         }
3801                         if (fd < 0)
3802                                 continue;
3803                         if (FD_ISSET(fd, pfds))
3804                                 do {
3805                                         buflen = read_network_packet(
3806                                                         fd, ep, ts);
3807                                 } while (buflen > 0);
3808                         /* Check more interfaces */
3809                 }
3810         }
3811
3812 #ifdef HAS_ROUTING_SOCKET
3813         /*
3814          * scan list of asyncio readers - currently only used for routing sockets
3815          */
3816         asyncio_reader = asyncio_reader_list;
3817
3818         while (asyncio_reader != NULL) {
3819                 /* callback may unlink and free asyncio_reader */
3820                 next_asyncio_reader = asyncio_reader->link;
3821                 if (FD_ISSET(asyncio_reader->fd, pfds))
3822                         (*asyncio_reader->receiver)(asyncio_reader);
3823                 asyncio_reader = next_asyncio_reader;
3824         }
3825 #endif /* HAS_ROUTING_SOCKET */
3826
3827         /*
3828          * Check for a response from a blocking child
3829          */
3830         for (idx = 0; idx < blocking_children_alloc; idx++) {
3831                 c = blocking_children[idx];
3832                 if (NULL == c || -1 == c->resp_read_pipe)
3833                         continue;
3834                 if (FD_ISSET(c->resp_read_pipe, pfds)) {
3835                         ++c->resp_ready_seen;
3836                         ++blocking_child_ready_seen;
3837                 }
3838         }
3839
3840         /* We've done our work */
3841 #if defined(DEBUG_TIMING)
3842         get_systime(&ts_e);
3843         /*
3844          * (ts_e - ts) is the amount of time we spent
3845          * processing this gob of file descriptors.  Log
3846          * it.
3847          */
3848         L_SUB(&ts_e, &ts);
3849         collect_timing(NULL, "input handler", 1, &ts_e);
3850         if (debug > 3)
3851                 msyslog(LOG_DEBUG,
3852                         "input_handler: Processed a gob of fd's in %s msec",
3853                         lfptoms(&ts_e, 6));
3854 #endif /* DEBUG_TIMING */
3855 }
3856 #endif /* !HAVE_IO_COMPLETION_PORT */
3857
3858 /*
3859  * find an interface suitable for the src address
3860  */
3861 endpt *
3862 select_peerinterface(
3863         struct peer *   peer,
3864         sockaddr_u *    srcadr,
3865         endpt *         dstadr
3866         )
3867 {
3868         endpt *ep;
3869 #ifndef SIM
3870         endpt *wild;
3871
3872         wild = ANY_INTERFACE_CHOOSE(srcadr);
3873
3874         /*
3875          * Initialize the peer structure and dance the interface jig.
3876          * Reference clocks step the loopback waltz, the others
3877          * squaredance around the interface list looking for a buddy. If
3878          * the dance peters out, there is always the wildcard interface.
3879          * This might happen in some systems and would preclude proper
3880          * operation with public key cryptography.
3881          */
3882         if (ISREFCLOCKADR(srcadr)) {
3883                 ep = loopback_interface;
3884         } else if (peer->cast_flags &
3885                    (MDF_BCLNT | MDF_ACAST | MDF_MCAST | MDF_BCAST)) {
3886                 ep = findbcastinter(srcadr);
3887                 if (ep != NULL)
3888                         DPRINTF(4, ("Found *-cast interface %s for address %s\n",
3889                                 stoa(&ep->sin), stoa(srcadr)));
3890                 else
3891                         DPRINTF(4, ("No *-cast local address found for address %s\n",
3892                                 stoa(srcadr)));
3893         } else {
3894                 ep = dstadr;
3895                 if (NULL == ep)
3896                         ep = wild;
3897         }
3898         /*
3899          * If it is a multicast address, findbcastinter() may not find
3900          * it.  For unicast, we get to find the interface when dstadr is
3901          * given to us as the wildcard (ANY_INTERFACE_CHOOSE).  Either
3902          * way, try a little harder.
3903          */
3904         if (wild == ep)
3905                 ep = findinterface(srcadr);
3906         /*
3907          * we do not bind to the wildcard interfaces for output
3908          * as our (network) source address would be undefined and
3909          * crypto will not work without knowing the own transmit address
3910          */
3911         if (ep != NULL && INT_WILDCARD & ep->flags)
3912                 if (!accept_wildcard_if_for_winnt)
3913                         ep = NULL;
3914 #else   /* SIM follows */
3915         ep = loopback_interface;
3916 #endif
3917
3918         return ep;
3919 }
3920
3921
3922 /*
3923  * findinterface - find local interface corresponding to address
3924  */
3925 endpt *
3926 findinterface(
3927         sockaddr_u *addr
3928         )
3929 {
3930         endpt *iface;
3931
3932         iface = findlocalinterface(addr, INT_WILDCARD, 0);
3933
3934         if (NULL == iface) {
3935                 DPRINTF(4, ("Found no interface for address %s - returning wildcard\n",
3936                             stoa(addr)));
3937
3938                 iface = ANY_INTERFACE_CHOOSE(addr);
3939         } else
3940                 DPRINTF(4, ("Found interface #%d %s for address %s\n",
3941                             iface->ifnum, iface->name, stoa(addr)));
3942
3943         return iface;
3944 }
3945
3946 /*
3947  * findlocalinterface - find local interface corresponding to addr,
3948  * which does not have any of flags set.  If bast is nonzero, addr is
3949  * a broadcast address.
3950  *
3951  * This code attempts to find the local sending address for an outgoing
3952  * address by connecting a new socket to destinationaddress:NTP_PORT
3953  * and reading the sockname of the resulting connect.
3954  * the complicated sequence simulates the routing table lookup
3955  * for to first hop without duplicating any of the routing logic into
3956  * ntpd. preferably we would have used an API call - but its not there -
3957  * so this is the best we can do here short of duplicating to entire routing
3958  * logic in ntpd which would be a silly and really unportable thing to do.
3959  *
3960  */
3961 static endpt *
3962 findlocalinterface(
3963         sockaddr_u *    addr,
3964         int             flags,
3965         int             bcast
3966         )
3967 {
3968         GETSOCKNAME_SOCKLEN_TYPE        sockaddrlen;
3969         endpt *                         iface;
3970         sockaddr_u                      saddr;
3971         SOCKET                          s;
3972         int                             rtn;
3973         int                             on;
3974
3975         DPRINTF(4, ("Finding interface for addr %s in list of addresses\n",
3976                     stoa(addr)));
3977
3978         /* [Bug 3437] The dummy POOL peer comes in with an AF of
3979          * zero. This is bound to fail, but on the way to nowhere it
3980          * triggers a security incident on SELinux.
3981          *
3982          * Checking the condition and failing early is probably a good
3983          * advice, and even saves us some syscalls in that case.
3984          * Thanks to Miroslav Lichvar for finding this.
3985          */
3986         if (AF_UNSPEC == AF(addr))
3987                 return NULL;
3988
3989         s = socket(AF(addr), SOCK_DGRAM, 0);
3990         if (INVALID_SOCKET == s)
3991                 return NULL;
3992
3993         /*
3994          * If we are looking for broadcast interface we need to set this
3995          * socket to allow broadcast
3996          */
3997         if (bcast) {
3998                 on = 1;
3999                 if (SOCKET_ERROR == setsockopt(s, SOL_SOCKET,
4000                                                 SO_BROADCAST,
4001                                                 (void *)&on,
4002                                                 sizeof(on))) {
4003                         closesocket(s);
4004                         return NULL;
4005                 }
4006         }
4007
4008         rtn = connect(s, &addr->sa, SOCKLEN(addr));
4009         if (SOCKET_ERROR == rtn) {
4010                 closesocket(s);
4011                 return NULL;
4012         }
4013
4014         sockaddrlen = sizeof(saddr);
4015         rtn = getsockname(s, &saddr.sa, &sockaddrlen);
4016         closesocket(s);
4017         if (SOCKET_ERROR == rtn)
4018                 return NULL;
4019
4020         DPRINTF(4, ("findlocalinterface: kernel maps %s to %s\n",
4021                     stoa(addr), stoa(&saddr)));
4022
4023         iface = getinterface(&saddr, flags);
4024
4025         /*
4026          * if we didn't find an exact match on saddr, find the closest
4027          * available local address.  This handles the case of the
4028          * address suggested by the kernel being excluded by nic rules
4029          * or the user's -I and -L options to ntpd.
4030          * See http://bugs.ntp.org/1184 and http://bugs.ntp.org/1683
4031          * for more background.
4032          */
4033         if (NULL == iface || iface->ignore_packets)
4034                 iface = findclosestinterface(&saddr,
4035                                              flags | INT_LOOPBACK);
4036
4037         /* Don't use an interface which will ignore replies */
4038         if (iface != NULL && iface->ignore_packets)
4039                 iface = NULL;
4040
4041         return iface;
4042 }
4043
4044
4045 /*
4046  * findclosestinterface
4047  *
4048  * If there are -I/--interface or -L/novirtualips command-line options,
4049  * or "nic" or "interface" rules in ntp.conf, findlocalinterface() may
4050  * find the kernel's preferred local address for a given peer address is
4051  * administratively unavailable to ntpd, and punt to this routine's more
4052  * expensive search.
4053  *
4054  * Find the numerically closest local address to the one connect()
4055  * suggested.  This matches an address on the same subnet first, as
4056  * needed by Bug 1184, and provides a consistent choice if there are
4057  * multiple feasible local addresses, regardless of the order ntpd
4058  * enumerated them.
4059  */
4060 endpt *
4061 findclosestinterface(
4062         sockaddr_u *    addr,
4063         int             flags
4064         )
4065 {
4066         endpt *         ep;
4067         endpt *         winner;
4068         sockaddr_u      addr_dist;
4069         sockaddr_u      min_dist;
4070
4071         ZERO_SOCK(&min_dist);
4072         winner = NULL;
4073
4074         for (ep = ep_list; ep != NULL; ep = ep->elink) {
4075                 if (ep->ignore_packets ||
4076                     AF(addr) != ep->family ||
4077                     flags & ep->flags)
4078                         continue;
4079
4080                 calc_addr_distance(&addr_dist, addr, &ep->sin);
4081                 if (NULL == winner ||
4082                     -1 == cmp_addr_distance(&addr_dist, &min_dist)) {
4083                         min_dist = addr_dist;
4084                         winner = ep;
4085                 }
4086         }
4087         if (NULL == winner)
4088                 DPRINTF(4, ("findclosestinterface(%s) failed\n",
4089                             stoa(addr)));
4090         else
4091                 DPRINTF(4, ("findclosestinterface(%s) -> %s\n",
4092                             stoa(addr), stoa(&winner->sin)));
4093
4094         return winner;
4095 }
4096
4097
4098 /*
4099  * calc_addr_distance - calculate the distance between two addresses,
4100  *                      the absolute value of the difference between
4101  *                      the addresses numerically, stored as an address.
4102  */
4103 static void
4104 calc_addr_distance(
4105         sockaddr_u *            dist,
4106         const sockaddr_u *      a1,
4107         const sockaddr_u *      a2
4108         )
4109 {
4110         u_int32 a1val;
4111         u_int32 a2val;
4112         u_int32 v4dist;
4113         int     found_greater;
4114         int     a1_greater;
4115         int     i;
4116
4117         REQUIRE(AF(a1) == AF(a2));
4118
4119         ZERO_SOCK(dist);
4120         AF(dist) = AF(a1);
4121
4122         /* v4 can be done a bit simpler */
4123         if (IS_IPV4(a1)) {
4124                 a1val = SRCADR(a1);
4125                 a2val = SRCADR(a2);
4126                 v4dist = (a1val > a2val)
4127                              ? a1val - a2val
4128                              : a2val - a1val;
4129                 SET_ADDR4(dist, v4dist);
4130
4131                 return;
4132         }
4133
4134         found_greater = FALSE;
4135         a1_greater = FALSE;     /* suppress pot. uninit. warning */
4136         for (i = 0; i < (int)sizeof(NSRCADR6(a1)); i++) {
4137                 if (!found_greater &&
4138                     NSRCADR6(a1)[i] != NSRCADR6(a2)[i]) {
4139                         found_greater = TRUE;
4140                         a1_greater = (NSRCADR6(a1)[i] > NSRCADR6(a2)[i]);
4141                 }
4142                 if (!found_greater) {
4143                         NSRCADR6(dist)[i] = 0;
4144                 } else {
4145                         if (a1_greater)
4146                                 NSRCADR6(dist)[i] = NSRCADR6(a1)[i] -
4147                                                     NSRCADR6(a2)[i];
4148                         else
4149                                 NSRCADR6(dist)[i] = NSRCADR6(a2)[i] -
4150                                                     NSRCADR6(a1)[i];
4151                 }
4152         }
4153 }
4154
4155
4156 /*
4157  * cmp_addr_distance - compare two address distances, returning -1, 0,
4158  *                     1 to indicate their relationship.
4159  */
4160 static int
4161 cmp_addr_distance(
4162         const sockaddr_u *      d1,
4163         const sockaddr_u *      d2
4164         )
4165 {
4166         int     i;
4167
4168         REQUIRE(AF(d1) == AF(d2));
4169
4170         if (IS_IPV4(d1)) {
4171                 if (SRCADR(d1) < SRCADR(d2))
4172                         return -1;
4173                 else if (SRCADR(d1) == SRCADR(d2))
4174                         return 0;
4175                 else
4176                         return 1;
4177         }
4178
4179         for (i = 0; i < (int)sizeof(NSRCADR6(d1)); i++) {
4180                 if (NSRCADR6(d1)[i] < NSRCADR6(d2)[i])
4181                         return -1;
4182                 else if (NSRCADR6(d1)[i] > NSRCADR6(d2)[i])
4183                         return 1;
4184         }
4185
4186         return 0;
4187 }
4188
4189
4190
4191 /*
4192  * fetch an interface structure the matches the
4193  * address and has the given flags NOT set
4194  */
4195 endpt *
4196 getinterface(
4197         sockaddr_u *    addr,
4198         u_int32         flags
4199         )
4200 {
4201         endpt *iface;
4202
4203         iface = find_addr_in_list(addr);
4204
4205         if (iface != NULL && (iface->flags & flags))
4206                 iface = NULL;
4207
4208         return iface;
4209 }
4210
4211
4212 /*
4213  * findbcastinter - find broadcast interface corresponding to address
4214  */
4215 endpt *
4216 findbcastinter(
4217         sockaddr_u *addr
4218         )
4219 {
4220         endpt * iface;
4221
4222         iface = NULL;
4223 #if !defined(MPE) && (defined(SIOCGIFCONF) || defined(SYS_WINNT))
4224         DPRINTF(4, ("Finding broadcast/multicast interface for addr %s in list of addresses\n",
4225                     stoa(addr)));
4226
4227         iface = findlocalinterface(addr, INT_LOOPBACK | INT_WILDCARD,
4228                                    1);
4229         if (iface != NULL) {
4230                 DPRINTF(4, ("Easily found bcast-/mcast- interface index #%d %s\n",
4231                             iface->ifnum, iface->name));
4232                 return iface;
4233         }
4234
4235         /*
4236          * plan B - try to find something reasonable in our lists in
4237          * case kernel lookup doesn't help
4238          */
4239         for (iface = ep_list; iface != NULL; iface = iface->elink) {
4240                 if (iface->flags & INT_WILDCARD)
4241                         continue;
4242
4243                 /* Don't bother with ignored interfaces */
4244                 if (iface->ignore_packets)
4245                         continue;
4246
4247                 /*
4248                  * First look if this is the correct family
4249                  */
4250                 if(AF(&iface->sin) != AF(addr))
4251                         continue;
4252
4253                 /* Skip the loopback addresses */
4254                 if (iface->flags & INT_LOOPBACK)
4255                         continue;
4256
4257                 /*
4258                  * If we are looking to match a multicast address and
4259                  * this interface is one...
4260                  */
4261                 if (addr_ismulticast(addr)
4262                     && (iface->flags & INT_MULTICAST)) {
4263 #ifdef INCLUDE_IPV6_SUPPORT
4264                         /*
4265                          * ...it is the winner unless we're looking for
4266                          * an interface to use for link-local multicast
4267                          * and its address is not link-local.
4268                          */
4269                         if (IS_IPV6(addr)
4270                             && IN6_IS_ADDR_MC_LINKLOCAL(PSOCK_ADDR6(addr))
4271                             && !IN6_IS_ADDR_LINKLOCAL(PSOCK_ADDR6(&iface->sin)))
4272                                 continue;
4273 #endif
4274                         break;
4275                 }
4276
4277                 /*
4278                  * We match only those interfaces marked as
4279                  * broadcastable and either the explicit broadcast
4280                  * address or the network portion of the IP address.
4281                  * Sloppy.
4282                  */
4283                 if (IS_IPV4(addr)) {
4284                         if (SOCK_EQ(&iface->bcast, addr))
4285                                 break;
4286
4287                         if ((NSRCADR(&iface->sin) & NSRCADR(&iface->mask))
4288                             == (NSRCADR(addr)     & NSRCADR(&iface->mask)))
4289                                 break;
4290                 }
4291 #ifdef INCLUDE_IPV6_SUPPORT
4292                 else if (IS_IPV6(addr)) {
4293                         if (SOCK_EQ(&iface->bcast, addr))
4294                                 break;
4295
4296                         if (SOCK_EQ(netof(&iface->sin), netof(addr)))
4297                                 break;
4298                 }
4299 #endif
4300         }
4301 #endif /* SIOCGIFCONF */
4302         if (NULL == iface) {
4303                 DPRINTF(4, ("No bcast interface found for %s\n",
4304                             stoa(addr)));
4305                 iface = ANY_INTERFACE_CHOOSE(addr);
4306         } else {
4307                 DPRINTF(4, ("Found bcast-/mcast- interface index #%d %s\n",
4308                             iface->ifnum, iface->name));
4309         }
4310
4311         return iface;
4312 }
4313
4314
4315 /*
4316  * io_clr_stats - clear I/O module statistics
4317  */
4318 void
4319 io_clr_stats(void)
4320 {
4321         packets_dropped = 0;
4322         packets_ignored = 0;
4323         packets_received = 0;
4324         packets_sent = 0;
4325         packets_notsent = 0;
4326
4327         handler_calls = 0;
4328         handler_pkts = 0;
4329         io_timereset = current_time;
4330 }
4331
4332
4333 #ifdef REFCLOCK
4334 /*
4335  * io_addclock - add a reference clock to the list and arrange that we
4336  *                               get SIGIO interrupts from it.
4337  */
4338 int
4339 io_addclock(
4340         struct refclockio *rio
4341         )
4342 {
4343         BLOCKIO();
4344
4345         /*
4346          * Stuff the I/O structure in the list and mark the descriptor
4347          * in use.  There is a harmless (I hope) race condition here.
4348          */
4349         rio->active = TRUE;
4350
4351 # ifdef HAVE_SIGNALED_IO
4352         if (init_clock_sig(rio)) {
4353                 UNBLOCKIO();
4354                 return 0;
4355         }
4356 # elif defined(HAVE_IO_COMPLETION_PORT)
4357         if (!io_completion_port_add_clock_io(rio)) {
4358                 UNBLOCKIO();
4359                 return 0;
4360         }
4361 # endif
4362
4363         /*
4364          * enqueue
4365          */
4366         LINK_SLIST(refio, rio, next);
4367
4368         /*
4369          * register fd
4370          */
4371         add_fd_to_list(rio->fd, FD_TYPE_FILE);
4372
4373         UNBLOCKIO();
4374         return 1;
4375 }
4376
4377
4378 /*
4379  * io_closeclock - close the clock in the I/O structure given
4380  */
4381 void
4382 io_closeclock(
4383         struct refclockio *rio
4384         )
4385 {
4386         struct refclockio *unlinked;
4387
4388         BLOCKIO();
4389
4390         /*
4391          * Remove structure from the list
4392          */
4393         rio->active = FALSE;
4394         UNLINK_SLIST(unlinked, refio, rio, next, struct refclockio);
4395         if (NULL != unlinked) {
4396                 /* Close the descriptor. The order of operations is
4397                  * important here in case of async / overlapped IO:
4398                  * only after we have removed the clock from the
4399                  * IO completion port we can be sure no further
4400                  * input is queued. So...
4401                  *  - we first disable feeding to the queu by removing
4402                  *    the clock from the IO engine
4403                  *  - close the file (which brings down any IO on it)
4404                  *  - clear the buffer from results for this fd
4405                  */
4406 #           ifdef HAVE_IO_COMPLETION_PORT
4407                 io_completion_port_remove_clock_io(rio);
4408 #           endif
4409                 close_and_delete_fd_from_list(rio->fd);
4410                 purge_recv_buffers_for_fd(rio->fd);
4411                 rio->fd = -1;
4412         }
4413
4414         UNBLOCKIO();
4415 }
4416 #endif  /* REFCLOCK */
4417
4418
4419 /*
4420  * On NT a SOCKET is an unsigned int so we cannot possibly keep it in
4421  * an array. So we use one of the ISC_LIST functions to hold the
4422  * socket value and use that when we want to enumerate it.
4423  *
4424  * This routine is called by the forked intres child process to close
4425  * all open sockets.  On Windows there's no need as intres runs in
4426  * the same process as a thread.
4427  */
4428 #ifndef SYS_WINNT
4429 void
4430 kill_asyncio(
4431         int     startfd
4432         )
4433 {
4434         BLOCKIO();
4435
4436         /*
4437          * In the child process we do not maintain activefds and
4438          * maxactivefd.  Zeroing maxactivefd disables code which
4439          * maintains it in close_and_delete_fd_from_list().
4440          */
4441         maxactivefd = 0;
4442
4443         while (fd_list != NULL)
4444                 close_and_delete_fd_from_list(fd_list->fd);
4445
4446         UNBLOCKIO();
4447 }
4448 #endif  /* !SYS_WINNT */
4449
4450
4451 /*
4452  * Add and delete functions for the list of open sockets
4453  */
4454 static void
4455 add_fd_to_list(
4456         SOCKET fd,
4457         enum desc_type type
4458         )
4459 {
4460         vsock_t *lsock = emalloc(sizeof(*lsock));
4461
4462         lsock->fd = fd;
4463         lsock->type = type;
4464
4465         LINK_SLIST(fd_list, lsock, link);
4466         maintain_activefds(fd, 0);
4467 }
4468
4469
4470 static void
4471 close_and_delete_fd_from_list(
4472         SOCKET fd
4473         )
4474 {
4475         vsock_t *lsock;
4476
4477         UNLINK_EXPR_SLIST(lsock, fd_list, fd ==
4478             UNLINK_EXPR_SLIST_CURRENT()->fd, link, vsock_t);
4479
4480         if (NULL == lsock)
4481                 return;
4482
4483         switch (lsock->type) {
4484
4485         case FD_TYPE_SOCKET:
4486                 closesocket(lsock->fd);
4487                 break;
4488
4489         case FD_TYPE_FILE:
4490                 closeserial((int)lsock->fd);
4491                 break;
4492
4493         default:
4494                 msyslog(LOG_ERR,
4495                         "internal error - illegal descriptor type %d - EXITING",
4496                         (int)lsock->type);
4497                 exit(1);
4498         }
4499
4500         free(lsock);
4501         /*
4502          * remove from activefds
4503          */
4504         maintain_activefds(fd, 1);
4505 }
4506
4507
4508 static void
4509 add_addr_to_list(
4510         sockaddr_u *    addr,
4511         endpt *         ep
4512         )
4513 {
4514         remaddr_t *laddr;
4515
4516 #ifdef DEBUG
4517         if (find_addr_in_list(addr) == NULL) {
4518 #endif
4519                 /* not there yet - add to list */
4520                 laddr = emalloc(sizeof(*laddr));
4521                 laddr->addr = *addr;
4522                 laddr->ep = ep;
4523
4524                 LINK_SLIST(remoteaddr_list, laddr, link);
4525
4526                 DPRINTF(4, ("Added addr %s to list of addresses\n",
4527                             stoa(addr)));
4528 #ifdef DEBUG
4529         } else
4530                 DPRINTF(4, ("WARNING: Attempt to add duplicate addr %s to address list\n",
4531                             stoa(addr)));
4532 #endif
4533 }
4534
4535
4536 static void
4537 delete_addr_from_list(
4538         sockaddr_u *addr
4539         )
4540 {
4541         remaddr_t *unlinked;
4542
4543         UNLINK_EXPR_SLIST(unlinked, remoteaddr_list, SOCK_EQ(addr,
4544                 &(UNLINK_EXPR_SLIST_CURRENT()->addr)), link, remaddr_t);
4545
4546         if (unlinked != NULL) {
4547                 DPRINTF(4, ("Deleted addr %s from list of addresses\n",
4548                         stoa(addr)));
4549                 free(unlinked);
4550         }
4551 }
4552
4553
4554 static void
4555 delete_interface_from_list(
4556         endpt *iface
4557         )
4558 {
4559         remaddr_t *unlinked;
4560
4561         for (;;) {
4562                 UNLINK_EXPR_SLIST(unlinked, remoteaddr_list, iface ==
4563                     UNLINK_EXPR_SLIST_CURRENT()->ep, link,
4564                     remaddr_t);
4565
4566                 if (unlinked == NULL)
4567                         break;
4568                 DPRINTF(4, ("Deleted addr %s for interface #%d %s from list of addresses\n",
4569                             stoa(&unlinked->addr), iface->ifnum,
4570                             iface->name));
4571                 free(unlinked);
4572         }
4573 }
4574
4575
4576 static struct interface *
4577 find_addr_in_list(
4578         sockaddr_u *addr
4579         )
4580 {
4581         remaddr_t *entry;
4582
4583         DPRINTF(4, ("Searching for addr %s in list of addresses - ",
4584                     stoa(addr)));
4585
4586         for (entry = remoteaddr_list;
4587              entry != NULL;
4588              entry = entry->link)
4589                 if (SOCK_EQ(&entry->addr, addr)) {
4590                         DPRINTF(4, ("FOUND\n"));
4591                         return entry->ep;
4592                 }
4593
4594         DPRINTF(4, ("NOT FOUND\n"));
4595         return NULL;
4596 }
4597
4598
4599 /*
4600  * Find the given address with the all given flags set in the list
4601  */
4602 static endpt *
4603 find_flagged_addr_in_list(
4604         sockaddr_u *    addr,
4605         u_int32         flags
4606         )
4607 {
4608         remaddr_t *entry;
4609
4610         DPRINTF(4, ("Finding addr %s with flags %d in list: ",
4611                     stoa(addr), flags));
4612
4613         for (entry = remoteaddr_list;
4614              entry != NULL;
4615              entry = entry->link)
4616
4617                 if (SOCK_EQ(&entry->addr, addr)
4618                     && (entry->ep->flags & flags) == flags) {
4619
4620                         DPRINTF(4, ("FOUND\n"));
4621                         return entry->ep;
4622                 }
4623
4624         DPRINTF(4, ("NOT FOUND\n"));
4625         return NULL;
4626 }
4627
4628
4629 const char *
4630 localaddrtoa(
4631         endpt *la
4632         )
4633 {
4634         return (NULL == la)
4635                    ? "<null>"
4636                    : stoa(&la->sin);
4637 }
4638
4639
4640 #ifdef HAS_ROUTING_SOCKET
4641 # ifndef UPDATE_GRACE
4642 #  define UPDATE_GRACE  2       /* wait UPDATE_GRACE seconds before scanning */
4643 # endif
4644
4645 static void
4646 process_routing_msgs(struct asyncio_reader *reader)
4647 {
4648         char buffer[5120];
4649         int cnt, msg_type;
4650 #ifdef HAVE_RTNETLINK
4651         struct nlmsghdr *nh;
4652 #else
4653         struct rt_msghdr rtm;
4654         char *p;
4655 #endif
4656
4657         if (disable_dynamic_updates) {
4658                 /*
4659                  * discard ourselves if we are not needed any more
4660                  * usually happens when running unprivileged
4661                  */
4662                 remove_asyncio_reader(reader);
4663                 delete_asyncio_reader(reader);
4664                 return;
4665         }
4666
4667         cnt = read(reader->fd, buffer, sizeof(buffer));
4668
4669         if (cnt < 0) {
4670                 if (errno == ENOBUFS) {
4671                         msyslog(LOG_ERR,
4672                                 "routing socket reports: %m");
4673                 } else {
4674                         msyslog(LOG_ERR,
4675                                 "routing socket reports: %m - disabling");
4676                         remove_asyncio_reader(reader);
4677                         delete_asyncio_reader(reader);
4678                 }
4679                 return;
4680         }
4681
4682         /*
4683          * process routing message
4684          */
4685 #ifdef HAVE_RTNETLINK
4686         for (nh = UA_PTR(struct nlmsghdr, buffer);
4687              NLMSG_OK(nh, cnt);
4688              nh = NLMSG_NEXT(nh, cnt)) {
4689                 msg_type = nh->nlmsg_type;
4690 #else
4691         for (p = buffer;
4692              (p + sizeof(struct rt_msghdr)) <= (buffer + cnt);
4693              p += rtm.rtm_msglen) {
4694                 memcpy(&rtm, p, sizeof(rtm));
4695                 if (rtm.rtm_version != RTM_VERSION) {
4696                         msyslog(LOG_ERR,
4697                                 "version mismatch (got %d - expected %d) on routing socket - disabling",
4698                                 rtm.rtm_version, RTM_VERSION);
4699
4700                         remove_asyncio_reader(reader);
4701                         delete_asyncio_reader(reader);
4702                         return;
4703                 }
4704                 msg_type = rtm.rtm_type;
4705 #endif
4706                 switch (msg_type) {
4707 #ifdef RTM_NEWADDR
4708                 case RTM_NEWADDR:
4709 #endif
4710 #ifdef RTM_DELADDR
4711                 case RTM_DELADDR:
4712 #endif
4713 #ifdef RTM_ADD
4714                 case RTM_ADD:
4715 #endif
4716 #ifdef RTM_DELETE
4717                 case RTM_DELETE:
4718 #endif
4719 #ifdef RTM_REDIRECT
4720                 case RTM_REDIRECT:
4721 #endif
4722 #ifdef RTM_CHANGE
4723                 case RTM_CHANGE:
4724 #endif
4725 #ifdef RTM_LOSING
4726                 case RTM_LOSING:
4727 #endif
4728 #ifdef RTM_IFINFO
4729                 case RTM_IFINFO:
4730 #endif
4731 #ifdef RTM_IFANNOUNCE
4732                 case RTM_IFANNOUNCE:
4733 #endif
4734 #ifdef RTM_NEWLINK
4735                 case RTM_NEWLINK:
4736 #endif
4737 #ifdef RTM_DELLINK
4738                 case RTM_DELLINK:
4739 #endif
4740 #ifdef RTM_NEWROUTE
4741                 case RTM_NEWROUTE:
4742 #endif
4743 #ifdef RTM_DELROUTE
4744                 case RTM_DELROUTE:
4745 #endif
4746                         /*
4747                          * we are keen on new and deleted addresses and
4748                          * if an interface goes up and down or routing
4749                          * changes
4750                          */
4751                         DPRINTF(3, ("routing message op = %d: scheduling interface update\n",
4752                                     msg_type));
4753                         timer_interfacetimeout(current_time + UPDATE_GRACE);
4754                         break;
4755 #ifdef HAVE_RTNETLINK
4756                 case NLMSG_DONE:
4757                         /* end of multipart message */
4758                         return;
4759 #endif
4760                 default:
4761                         /*
4762                          * the rest doesn't bother us.
4763                          */
4764                         DPRINTF(4, ("routing message op = %d: ignored\n",
4765                                     msg_type));
4766                         break;
4767                 }
4768         }
4769 }
4770
4771 /*
4772  * set up routing notifications
4773  */
4774 static void
4775 init_async_notifications()
4776 {
4777         struct asyncio_reader *reader;
4778 #ifdef HAVE_RTNETLINK
4779         int fd = socket(PF_NETLINK, SOCK_RAW, NETLINK_ROUTE);
4780         struct sockaddr_nl sa;
4781 #else
4782         int fd = socket(PF_ROUTE, SOCK_RAW, 0);
4783 #endif
4784         if (fd < 0) {
4785                 msyslog(LOG_ERR,
4786                         "unable to open routing socket (%m) - using polled interface update");
4787                 return;
4788         }
4789
4790         fd = move_fd(fd);
4791 #ifdef HAVE_RTNETLINK
4792         ZERO(sa);
4793         sa.nl_family = PF_NETLINK;
4794         sa.nl_groups = RTMGRP_LINK | RTMGRP_IPV4_IFADDR
4795                        | RTMGRP_IPV6_IFADDR | RTMGRP_IPV4_ROUTE
4796                        | RTMGRP_IPV4_MROUTE | RTMGRP_IPV6_ROUTE
4797                        | RTMGRP_IPV6_MROUTE;
4798         if (bind(fd, (struct sockaddr *)&sa, sizeof(sa)) < 0) {
4799                 msyslog(LOG_ERR,
4800                         "bind failed on routing socket (%m) - using polled interface update");
4801                 return;
4802         }
4803 #endif
4804         make_socket_nonblocking(fd);
4805 #if defined(HAVE_SIGNALED_IO)
4806         init_socket_sig(fd);
4807 #endif /* HAVE_SIGNALED_IO */
4808
4809         reader = new_asyncio_reader();
4810
4811         reader->fd = fd;
4812         reader->receiver = process_routing_msgs;
4813
4814         add_asyncio_reader(reader, FD_TYPE_SOCKET);
4815         msyslog(LOG_INFO,
4816                 "Listening on routing socket on fd #%d for interface updates",
4817                 fd);
4818 }
4819 #else
4820 /* HAS_ROUTING_SOCKET not defined */
4821 static void
4822 init_async_notifications(void)
4823 {
4824 }
4825 #endif
4826