contrib/bind9/lib/isc/unix/socket.c

   1 /*
   2  * Copyright (C) 2004-2012  Internet Systems Consortium, Inc. ("ISC")
   3  * Copyright (C) 1998-2003  Internet Software Consortium.
   4  *
   5  * Permission to use, copy, modify, and/or distribute this software for any
   6  * purpose with or without fee is hereby granted, provided that the above
   7  * copyright notice and this permission notice appear in all copies.
   8  *
   9  * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
  10  * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
  11  * AND FITNESS.  IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
  12  * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
  13  * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
  14  * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
  15  * PERFORMANCE OF THIS SOFTWARE.
  16  */
  17
  18 /* $Id$ */
  19
  20 /*! \file */
  21
  22 #include <config.h>
  23
  24 #include <sys/param.h>
  25 #include <sys/types.h>
  26 #include <sys/socket.h>
  27 #include <sys/stat.h>
  28 #include <sys/time.h>
  29 #include <sys/uio.h>
  30
  31 #include <errno.h>
  32 #include <fcntl.h>
  33 #include <stddef.h>
  34 #include <stdlib.h>
  35 #include <string.h>
  36 #include <unistd.h>
  37
  38 #include <isc/buffer.h>
  39 #include <isc/bufferlist.h>
  40 #include <isc/condition.h>
  41 #include <isc/formatcheck.h>
  42 #include <isc/list.h>
  43 #include <isc/log.h>
  44 #include <isc/mem.h>
  45 #include <isc/msgs.h>
  46 #include <isc/mutex.h>
  47 #include <isc/net.h>
  48 #include <isc/once.h>
  49 #include <isc/platform.h>
  50 #include <isc/print.h>
  51 #include <isc/region.h>
  52 #include <isc/socket.h>
  53 #include <isc/stats.h>
  54 #include <isc/strerror.h>
  55 #include <isc/task.h>
  56 #include <isc/thread.h>
  57 #include <isc/util.h>
  58 #include <isc/xml.h>
  59
  60 #ifdef ISC_PLATFORM_HAVESYSUNH
  61 #include <sys/un.h>
  62 #endif
  63 #ifdef ISC_PLATFORM_HAVEKQUEUE
  64 #include <sys/event.h>
  65 #endif
  66 #ifdef ISC_PLATFORM_HAVEEPOLL
  67 #include <sys/epoll.h>
  68 #endif
  69 #ifdef ISC_PLATFORM_HAVEDEVPOLL
  70 #if defined(HAVE_SYS_DEVPOLL_H)
  71 #include <sys/devpoll.h>
  72 #elif defined(HAVE_DEVPOLL_H)
  73 #include <devpoll.h>
  74 #endif
  75 #endif
  76
  77 #include "errno2result.h"
  78
  79 /* See task.c about the following definition: */
  80 #ifdef BIND9
  81 #ifdef ISC_PLATFORM_USETHREADS
  82 #define USE_WATCHER_THREAD
  83 #else
  84 #define USE_SHARED_MANAGER
  85 #endif  /* ISC_PLATFORM_USETHREADS */
  86 #endif  /* BIND9 */
  87
  88 #ifndef USE_WATCHER_THREAD
  89 #include "socket_p.h"
  90 #include "../task_p.h"
  91 #endif /* USE_WATCHER_THREAD */
  92
  93 #if defined(SO_BSDCOMPAT) && defined(__linux__)
  94 #include <sys/utsname.h>
  95 #endif
  96
  97 /*%
  98  * Choose the most preferable multiplex method.
  99  */
 100 #ifdef ISC_PLATFORM_HAVEKQUEUE
 101 #define USE_KQUEUE
 102 #elif defined (ISC_PLATFORM_HAVEEPOLL)
 103 #define USE_EPOLL
 104 #elif defined (ISC_PLATFORM_HAVEDEVPOLL)
 105 #define USE_DEVPOLL
 106 typedef struct {
 107         unsigned int want_read : 1,
 108                 want_write : 1;
 109 } pollinfo_t;
 110 #else
 111 #define USE_SELECT
 112 #endif  /* ISC_PLATFORM_HAVEKQUEUE */
 113
 114 #ifndef USE_WATCHER_THREAD
 115 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
 116 struct isc_socketwait {
 117         int nevents;
 118 };
 119 #elif defined (USE_SELECT)
 120 struct isc_socketwait {
 121         fd_set *readset;
 122         fd_set *writeset;
 123         int nfds;
 124         int maxfd;
 125 };
 126 #endif  /* USE_KQUEUE */
 127 #endif /* !USE_WATCHER_THREAD */
 128
 129 /*%
 130  * Maximum number of allowable open sockets.  This is also the maximum
 131  * allowable socket file descriptor.
 132  *
 133  * Care should be taken before modifying this value for select():
 134  * The API standard doesn't ensure select() accept more than (the system default
 135  * of) FD_SETSIZE descriptors, and the default size should in fact be fine in
 136  * the vast majority of cases.  This constant should therefore be increased only
 137  * when absolutely necessary and possible, i.e., the server is exhausting all
 138  * available file descriptors (up to FD_SETSIZE) and the select() function
 139  * and FD_xxx macros support larger values than FD_SETSIZE (which may not
 140  * always by true, but we keep using some of them to ensure as much
 141  * portability as possible).  Note also that overall server performance
 142  * may be rather worsened with a larger value of this constant due to
 143  * inherent scalability problems of select().
 144  *
 145  * As a special note, this value shouldn't have to be touched if
 146  * this is a build for an authoritative only DNS server.
 147  */
 148 #ifndef ISC_SOCKET_MAXSOCKETS
 149 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
 150 #define ISC_SOCKET_MAXSOCKETS 4096
 151 #elif defined(USE_SELECT)
 152 #define ISC_SOCKET_MAXSOCKETS FD_SETSIZE
 153 #endif  /* USE_KQUEUE... */
 154 #endif  /* ISC_SOCKET_MAXSOCKETS */
 155
 156 #ifdef USE_SELECT
 157 /*%
 158  * Mac OS X needs a special definition to support larger values in select().
 159  * We always define this because a larger value can be specified run-time.
 160  */
 161 #ifdef __APPLE__
 162 #define _DARWIN_UNLIMITED_SELECT
 163 #endif  /* __APPLE__ */
 164 #endif  /* USE_SELECT */
 165
 166 #ifdef ISC_SOCKET_USE_POLLWATCH
 167 /*%
 168  * If this macro is defined, enable workaround for a Solaris /dev/poll kernel
 169  * bug: DP_POLL ioctl could keep sleeping even if socket I/O is possible for
 170  * some of the specified FD.  The idea is based on the observation that it's
 171  * likely for a busy server to keep receiving packets.  It specifically works
 172  * as follows: the socket watcher is first initialized with the state of
 173  * "poll_idle".  While it's in the idle state it keeps sleeping until a socket
 174  * event occurs.  When it wakes up for a socket I/O event, it moves to the
 175  * poll_active state, and sets the poll timeout to a short period
 176  * (ISC_SOCKET_POLLWATCH_TIMEOUT msec).  If timeout occurs in this state, the
 177  * watcher goes to the poll_checking state with the same timeout period.
 178  * In this state, the watcher tries to detect whether this is a break
 179  * during intermittent events or the kernel bug is triggered.  If the next
 180  * polling reports an event within the short period, the previous timeout is
 181  * likely to be a kernel bug, and so the watcher goes back to the active state.
 182  * Otherwise, it moves to the idle state again.
 183  *
 184  * It's not clear whether this is a thread-related bug, but since we've only
 185  * seen this with threads, this workaround is used only when enabling threads.
 186  */
 187
 188 typedef enum { poll_idle, poll_active, poll_checking } pollstate_t;
 189
 190 #ifndef ISC_SOCKET_POLLWATCH_TIMEOUT
 191 #define ISC_SOCKET_POLLWATCH_TIMEOUT 10
 192 #endif  /* ISC_SOCKET_POLLWATCH_TIMEOUT */
 193 #endif  /* ISC_SOCKET_USE_POLLWATCH */
 194
 195 /*%
 196  * Size of per-FD lock buckets.
 197  */
 198 #ifdef ISC_PLATFORM_USETHREADS
 199 #define FDLOCK_COUNT            1024
 200 #define FDLOCK_ID(fd)           ((fd) % FDLOCK_COUNT)
 201 #else
 202 #define FDLOCK_COUNT            1
 203 #define FDLOCK_ID(fd)           0
 204 #endif  /* ISC_PLATFORM_USETHREADS */
 205
 206 /*%
 207  * Maximum number of events communicated with the kernel.  There should normally
 208  * be no need for having a large number.
 209  */
 210 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
 211 #ifndef ISC_SOCKET_MAXEVENTS
 212 #define ISC_SOCKET_MAXEVENTS    64
 213 #endif
 214 #endif
 215
 216 /*%
 217  * Some systems define the socket length argument as an int, some as size_t,
 218  * some as socklen_t.  This is here so it can be easily changed if needed.
 219  */
 220 #ifndef ISC_SOCKADDR_LEN_T
 221 #define ISC_SOCKADDR_LEN_T unsigned int
 222 #endif
 223
 224 /*%
 225  * Define what the possible "soft" errors can be.  These are non-fatal returns
 226  * of various network related functions, like recv() and so on.
 227  *
 228  * For some reason, BSDI (and perhaps others) will sometimes return <0
 229  * from recv() but will have errno==0.  This is broken, but we have to
 230  * work around it here.
 231  */
 232 #define SOFT_ERROR(e)   ((e) == EAGAIN || \
 233                          (e) == EWOULDBLOCK || \
 234                          (e) == EINTR || \
 235                          (e) == 0)
 236
 237 #define DLVL(x) ISC_LOGCATEGORY_GENERAL, ISC_LOGMODULE_SOCKET, ISC_LOG_DEBUG(x)
 238
 239 /*!<
 240  * DLVL(90)  --  Function entry/exit and other tracing.
 241  * DLVL(70)  --  Socket "correctness" -- including returning of events, etc.
 242  * DLVL(60)  --  Socket data send/receive
 243  * DLVL(50)  --  Event tracing, including receiving/sending completion events.
 244  * DLVL(20)  --  Socket creation/destruction.
 245  */
 246 #define TRACE_LEVEL             90
 247 #define CORRECTNESS_LEVEL       70
 248 #define IOEVENT_LEVEL           60
 249 #define EVENT_LEVEL             50
 250 #define CREATION_LEVEL          20
 251
 252 #define TRACE           DLVL(TRACE_LEVEL)
 253 #define CORRECTNESS     DLVL(CORRECTNESS_LEVEL)
 254 #define IOEVENT         DLVL(IOEVENT_LEVEL)
 255 #define EVENT           DLVL(EVENT_LEVEL)
 256 #define CREATION        DLVL(CREATION_LEVEL)
 257
 258 typedef isc_event_t intev_t;
 259
 260 #define SOCKET_MAGIC            ISC_MAGIC('I', 'O', 'i', 'o')
 261 #define VALID_SOCKET(s)         ISC_MAGIC_VALID(s, SOCKET_MAGIC)
 262
 263 /*!
 264  * IPv6 control information.  If the socket is an IPv6 socket we want
 265  * to collect the destination address and interface so the client can
 266  * set them on outgoing packets.
 267  */
 268 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
 269 #ifndef USE_CMSG
 270 #define USE_CMSG        1
 271 #endif
 272 #endif
 273
 274 /*%
 275  * NetBSD and FreeBSD can timestamp packets.  XXXMLG Should we have
 276  * a setsockopt() like interface to request timestamps, and if the OS
 277  * doesn't do it for us, call gettimeofday() on every UDP receive?
 278  */
 279 #ifdef SO_TIMESTAMP
 280 #ifndef USE_CMSG
 281 #define USE_CMSG        1
 282 #endif
 283 #endif
 284
 285 /*%
 286  * The size to raise the receive buffer to (from BIND 8).
 287  */
 288 #define RCVBUFSIZE (32*1024)
 289
 290 /*%
 291  * The number of times a send operation is repeated if the result is EINTR.
 292  */
 293 #define NRETRIES 10
 294
 295 typedef struct isc__socket isc__socket_t;
 296 typedef struct isc__socketmgr isc__socketmgr_t;
 297
 298 #define NEWCONNSOCK(ev) ((isc__socket_t *)(ev)->newsocket)
 299
 300 struct isc__socket {
 301         /* Not locked. */
 302         isc_socket_t            common;
 303         isc__socketmgr_t        *manager;
 304         isc_mutex_t             lock;
 305         isc_sockettype_t        type;
 306         const isc_statscounter_t        *statsindex;
 307
 308         /* Locked by socket lock. */
 309         ISC_LINK(isc__socket_t) link;
 310         unsigned int            references;
 311         int                     fd;
 312         int                     pf;
 313         char                            name[16];
 314         void *                          tag;
 315
 316         ISC_LIST(isc_socketevent_t)             send_list;
 317         ISC_LIST(isc_socketevent_t)             recv_list;
 318         ISC_LIST(isc_socket_newconnev_t)        accept_list;
 319         isc_socket_connev_t                    *connect_ev;
 320
 321         /*
 322          * Internal events.  Posted when a descriptor is readable or
 323          * writable.  These are statically allocated and never freed.
 324          * They will be set to non-purgable before use.
 325          */
 326         intev_t                 readable_ev;
 327         intev_t                 writable_ev;
 328
 329         isc_sockaddr_t          peer_address;  /* remote address */
 330
 331         unsigned int            pending_recv : 1,
 332                                 pending_send : 1,
 333                                 pending_accept : 1,
 334                                 listener : 1, /* listener socket */
 335                                 connected : 1,
 336                                 connecting : 1, /* connect pending */
 337                                 bound : 1; /* bound to local addr */
 338
 339 #ifdef ISC_NET_RECVOVERFLOW
 340         unsigned char           overflow; /* used for MSG_TRUNC fake */
 341 #endif
 342
 343         char                    *recvcmsgbuf;
 344         ISC_SOCKADDR_LEN_T      recvcmsgbuflen;
 345         char                    *sendcmsgbuf;
 346         ISC_SOCKADDR_LEN_T      sendcmsgbuflen;
 347
 348         void                    *fdwatcharg;
 349         isc_sockfdwatch_t       fdwatchcb;
 350         int                     fdwatchflags;
 351         isc_task_t              *fdwatchtask;
 352 };
 353
 354 #define SOCKET_MANAGER_MAGIC    ISC_MAGIC('I', 'O', 'm', 'g')
 355 #define VALID_MANAGER(m)        ISC_MAGIC_VALID(m, SOCKET_MANAGER_MAGIC)
 356
 357 struct isc__socketmgr {
 358         /* Not locked. */
 359         isc_socketmgr_t         common;
 360         isc_mem_t              *mctx;
 361         isc_mutex_t             lock;
 362         isc_mutex_t             *fdlock;
 363         isc_stats_t             *stats;
 364 #ifdef USE_KQUEUE
 365         int                     kqueue_fd;
 366         int                     nevents;
 367         struct kevent           *events;
 368 #endif  /* USE_KQUEUE */
 369 #ifdef USE_EPOLL
 370         int                     epoll_fd;
 371         int                     nevents;
 372         struct epoll_event      *events;
 373 #endif  /* USE_EPOLL */
 374 #ifdef USE_DEVPOLL
 375         int                     devpoll_fd;
 376         int                     nevents;
 377         struct pollfd           *events;
 378 #endif  /* USE_DEVPOLL */
 379 #ifdef USE_SELECT
 380         int                     fd_bufsize;
 381 #endif  /* USE_SELECT */
 382         unsigned int            maxsocks;
 383 #ifdef ISC_PLATFORM_USETHREADS
 384         int                     pipe_fds[2];
 385 #endif
 386
 387         /* Locked by fdlock. */
 388         isc__socket_t          **fds;
 389         int                     *fdstate;
 390 #ifdef USE_DEVPOLL
 391         pollinfo_t              *fdpollinfo;
 392 #endif
 393
 394         /* Locked by manager lock. */
 395         ISC_LIST(isc__socket_t) socklist;
 396 #ifdef USE_SELECT
 397         fd_set                  *read_fds;
 398         fd_set                  *read_fds_copy;
 399         fd_set                  *write_fds;
 400         fd_set                  *write_fds_copy;
 401         int                     maxfd;
 402 #endif  /* USE_SELECT */
 403         int                     reserved;       /* unlocked */
 404 #ifdef USE_WATCHER_THREAD
 405         isc_thread_t            watcher;
 406         isc_condition_t         shutdown_ok;
 407 #else /* USE_WATCHER_THREAD */
 408         unsigned int            refs;
 409 #endif /* USE_WATCHER_THREAD */
 410         int                     maxudp;
 411 };
 412
 413 #ifdef USE_SHARED_MANAGER
 414 static isc__socketmgr_t *socketmgr = NULL;
 415 #endif /* USE_SHARED_MANAGER */
 416
 417 #define CLOSED                  0       /* this one must be zero */
 418 #define MANAGED                 1
 419 #define CLOSE_PENDING           2
 420
 421 /*
 422  * send() and recv() iovec counts
 423  */
 424 #define MAXSCATTERGATHER_SEND   (ISC_SOCKET_MAXSCATTERGATHER)
 425 #ifdef ISC_NET_RECVOVERFLOW
 426 # define MAXSCATTERGATHER_RECV  (ISC_SOCKET_MAXSCATTERGATHER + 1)
 427 #else
 428 # define MAXSCATTERGATHER_RECV  (ISC_SOCKET_MAXSCATTERGATHER)
 429 #endif
 430
 431 static void send_recvdone_event(isc__socket_t *, isc_socketevent_t **);
 432 static void send_senddone_event(isc__socket_t *, isc_socketevent_t **);
 433 static void free_socket(isc__socket_t **);
 434 static isc_result_t allocate_socket(isc__socketmgr_t *, isc_sockettype_t,
 435                                     isc__socket_t **);
 436 static void destroy(isc__socket_t **);
 437 static void internal_accept(isc_task_t *, isc_event_t *);
 438 static void internal_connect(isc_task_t *, isc_event_t *);
 439 static void internal_recv(isc_task_t *, isc_event_t *);
 440 static void internal_send(isc_task_t *, isc_event_t *);
 441 static void internal_fdwatch_write(isc_task_t *, isc_event_t *);
 442 static void internal_fdwatch_read(isc_task_t *, isc_event_t *);
 443 static void process_cmsg(isc__socket_t *, struct msghdr *, isc_socketevent_t *);
 444 static void build_msghdr_send(isc__socket_t *, isc_socketevent_t *,
 445                               struct msghdr *, struct iovec *, size_t *);
 446 static void build_msghdr_recv(isc__socket_t *, isc_socketevent_t *,
 447                               struct msghdr *, struct iovec *, size_t *);
 448 #ifdef USE_WATCHER_THREAD
 449 static isc_boolean_t process_ctlfd(isc__socketmgr_t *manager);
 450 #endif
 451
 452 /*%
 453  * The following can be either static or public, depending on build environment.
 454  */
 455
 456 #ifdef BIND9
 457 #define ISC_SOCKETFUNC_SCOPE
 458 #else
 459 #define ISC_SOCKETFUNC_SCOPE static
 460 #endif
 461
 462 ISC_SOCKETFUNC_SCOPE isc_result_t
 463 isc__socket_create(isc_socketmgr_t *manager, int pf, isc_sockettype_t type,
 464                    isc_socket_t **socketp);
 465 ISC_SOCKETFUNC_SCOPE void
 466 isc__socket_attach(isc_socket_t *sock, isc_socket_t **socketp);
 467 ISC_SOCKETFUNC_SCOPE void
 468 isc__socket_detach(isc_socket_t **socketp);
 469 ISC_SOCKETFUNC_SCOPE isc_result_t
 470 isc__socketmgr_create(isc_mem_t *mctx, isc_socketmgr_t **managerp);
 471 ISC_SOCKETFUNC_SCOPE isc_result_t
 472 isc__socketmgr_create2(isc_mem_t *mctx, isc_socketmgr_t **managerp,
 473                        unsigned int maxsocks);
 474 ISC_SOCKETFUNC_SCOPE void
 475 isc__socketmgr_destroy(isc_socketmgr_t **managerp);
 476 ISC_SOCKETFUNC_SCOPE isc_result_t
 477 isc__socket_recvv(isc_socket_t *sock, isc_bufferlist_t *buflist,
 478                  unsigned int minimum, isc_task_t *task,
 479                   isc_taskaction_t action, const void *arg);
 480 ISC_SOCKETFUNC_SCOPE isc_result_t
 481 isc__socket_recv(isc_socket_t *sock, isc_region_t *region,
 482                  unsigned int minimum, isc_task_t *task,
 483                  isc_taskaction_t action, const void *arg);
 484 ISC_SOCKETFUNC_SCOPE isc_result_t
 485 isc__socket_recv2(isc_socket_t *sock, isc_region_t *region,
 486                   unsigned int minimum, isc_task_t *task,
 487                   isc_socketevent_t *event, unsigned int flags);
 488 ISC_SOCKETFUNC_SCOPE isc_result_t
 489 isc__socket_send(isc_socket_t *sock, isc_region_t *region,
 490                  isc_task_t *task, isc_taskaction_t action, const void *arg);
 491 ISC_SOCKETFUNC_SCOPE isc_result_t
 492 isc__socket_sendto(isc_socket_t *sock, isc_region_t *region,
 493                    isc_task_t *task, isc_taskaction_t action, const void *arg,
 494                    isc_sockaddr_t *address, struct in6_pktinfo *pktinfo);
 495 ISC_SOCKETFUNC_SCOPE isc_result_t
 496 isc__socket_sendv(isc_socket_t *sock, isc_bufferlist_t *buflist,
 497                   isc_task_t *task, isc_taskaction_t action, const void *arg);
 498 ISC_SOCKETFUNC_SCOPE isc_result_t
 499 isc__socket_sendtov(isc_socket_t *sock, isc_bufferlist_t *buflist,
 500                     isc_task_t *task, isc_taskaction_t action, const void *arg,
 501                     isc_sockaddr_t *address, struct in6_pktinfo *pktinfo);
 502 ISC_SOCKETFUNC_SCOPE isc_result_t
 503 isc__socket_sendto2(isc_socket_t *sock, isc_region_t *region,
 504                     isc_task_t *task,
 505                     isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
 506                     isc_socketevent_t *event, unsigned int flags);
 507 ISC_SOCKETFUNC_SCOPE void
 508 isc__socket_cleanunix(isc_sockaddr_t *sockaddr, isc_boolean_t active);
 509 ISC_SOCKETFUNC_SCOPE isc_result_t
 510 isc__socket_permunix(isc_sockaddr_t *sockaddr, isc_uint32_t perm,
 511                      isc_uint32_t owner, isc_uint32_t group);
 512 ISC_SOCKETFUNC_SCOPE isc_result_t
 513 isc__socket_bind(isc_socket_t *sock, isc_sockaddr_t *sockaddr,
 514                  unsigned int options);
 515 ISC_SOCKETFUNC_SCOPE isc_result_t
 516 isc__socket_filter(isc_socket_t *sock, const char *filter);
 517 ISC_SOCKETFUNC_SCOPE isc_result_t
 518 isc__socket_listen(isc_socket_t *sock, unsigned int backlog);
 519 ISC_SOCKETFUNC_SCOPE isc_result_t
 520 isc__socket_accept(isc_socket_t *sock,
 521                    isc_task_t *task, isc_taskaction_t action, const void *arg);
 522 ISC_SOCKETFUNC_SCOPE isc_result_t
 523 isc__socket_connect(isc_socket_t *sock, isc_sockaddr_t *addr,
 524                     isc_task_t *task, isc_taskaction_t action,
 525                     const void *arg);
 526 ISC_SOCKETFUNC_SCOPE isc_result_t
 527 isc__socket_getpeername(isc_socket_t *sock, isc_sockaddr_t *addressp);
 528 ISC_SOCKETFUNC_SCOPE isc_result_t
 529 isc__socket_getsockname(isc_socket_t *sock, isc_sockaddr_t *addressp);
 530 ISC_SOCKETFUNC_SCOPE void
 531 isc__socket_cancel(isc_socket_t *sock, isc_task_t *task, unsigned int how);
 532 ISC_SOCKETFUNC_SCOPE isc_sockettype_t
 533 isc__socket_gettype(isc_socket_t *sock);
 534 ISC_SOCKETFUNC_SCOPE isc_boolean_t
 535 isc__socket_isbound(isc_socket_t *sock);
 536 ISC_SOCKETFUNC_SCOPE void
 537 isc__socket_ipv6only(isc_socket_t *sock, isc_boolean_t yes);
 538 #if defined(HAVE_LIBXML2) && defined(BIND9)
 539 ISC_SOCKETFUNC_SCOPE void
 540 isc__socketmgr_renderxml(isc_socketmgr_t *mgr0, xmlTextWriterPtr writer);
 541 #endif
 542
 543 ISC_SOCKETFUNC_SCOPE isc_result_t
 544 isc__socket_fdwatchcreate(isc_socketmgr_t *manager, int fd, int flags,
 545                           isc_sockfdwatch_t callback, void *cbarg,
 546                           isc_task_t *task, isc_socket_t **socketp);
 547 ISC_SOCKETFUNC_SCOPE isc_result_t
 548 isc__socket_fdwatchpoke(isc_socket_t *sock, int flags);
 549
 550 static struct {
 551         isc_socketmethods_t methods;
 552
 553         /*%
 554          * The following are defined just for avoiding unused static functions.
 555          */
 556 #ifndef BIND9
 557         void *recvv, *send, *sendv, *sendto2, *cleanunix, *permunix, *filter,
 558                 *listen, *accept, *getpeername, *isbound;
 559 #endif
 560 } socketmethods = {
 561         {
 562                 isc__socket_attach,
 563                 isc__socket_detach,
 564                 isc__socket_bind,
 565                 isc__socket_sendto,
 566                 isc__socket_connect,
 567                 isc__socket_recv,
 568                 isc__socket_cancel,
 569                 isc__socket_getsockname,
 570                 isc__socket_gettype,
 571                 isc__socket_ipv6only,
 572                 isc__socket_fdwatchpoke
 573         }
 574 #ifndef BIND9
 575         ,
 576         (void *)isc__socket_recvv, (void *)isc__socket_send,
 577         (void *)isc__socket_sendv, (void *)isc__socket_sendto2,
 578         (void *)isc__socket_cleanunix, (void *)isc__socket_permunix,
 579         (void *)isc__socket_filter, (void *)isc__socket_listen,
 580         (void *)isc__socket_accept, (void *)isc__socket_getpeername,
 581         (void *)isc__socket_isbound
 582 #endif
 583 };
 584
 585 static isc_socketmgrmethods_t socketmgrmethods = {
 586         isc__socketmgr_destroy,
 587         isc__socket_create,
 588         isc__socket_fdwatchcreate
 589 };
 590
 591 #define SELECT_POKE_SHUTDOWN            (-1)
 592 #define SELECT_POKE_NOTHING             (-2)
 593 #define SELECT_POKE_READ                (-3)
 594 #define SELECT_POKE_ACCEPT              (-3) /*%< Same as _READ */
 595 #define SELECT_POKE_WRITE               (-4)
 596 #define SELECT_POKE_CONNECT             (-4) /*%< Same as _WRITE */
 597 #define SELECT_POKE_CLOSE               (-5)
 598
 599 #define SOCK_DEAD(s)                    ((s)->references == 0)
 600
 601 /*%
 602  * Shortcut index arrays to get access to statistics counters.
 603  */
 604 enum {
 605         STATID_OPEN = 0,
 606         STATID_OPENFAIL = 1,
 607         STATID_CLOSE = 2,
 608         STATID_BINDFAIL = 3,
 609         STATID_CONNECTFAIL = 4,
 610         STATID_CONNECT = 5,
 611         STATID_ACCEPTFAIL = 6,
 612         STATID_ACCEPT = 7,
 613         STATID_SENDFAIL = 8,
 614         STATID_RECVFAIL = 9
 615 };
 616 static const isc_statscounter_t upd4statsindex[] = {
 617         isc_sockstatscounter_udp4open,
 618         isc_sockstatscounter_udp4openfail,
 619         isc_sockstatscounter_udp4close,
 620         isc_sockstatscounter_udp4bindfail,
 621         isc_sockstatscounter_udp4connectfail,
 622         isc_sockstatscounter_udp4connect,
 623         -1,
 624         -1,
 625         isc_sockstatscounter_udp4sendfail,
 626         isc_sockstatscounter_udp4recvfail
 627 };
 628 static const isc_statscounter_t upd6statsindex[] = {
 629         isc_sockstatscounter_udp6open,
 630         isc_sockstatscounter_udp6openfail,
 631         isc_sockstatscounter_udp6close,
 632         isc_sockstatscounter_udp6bindfail,
 633         isc_sockstatscounter_udp6connectfail,
 634         isc_sockstatscounter_udp6connect,
 635         -1,
 636         -1,
 637         isc_sockstatscounter_udp6sendfail,
 638         isc_sockstatscounter_udp6recvfail
 639 };
 640 static const isc_statscounter_t tcp4statsindex[] = {
 641         isc_sockstatscounter_tcp4open,
 642         isc_sockstatscounter_tcp4openfail,
 643         isc_sockstatscounter_tcp4close,
 644         isc_sockstatscounter_tcp4bindfail,
 645         isc_sockstatscounter_tcp4connectfail,
 646         isc_sockstatscounter_tcp4connect,
 647         isc_sockstatscounter_tcp4acceptfail,
 648         isc_sockstatscounter_tcp4accept,
 649         isc_sockstatscounter_tcp4sendfail,
 650         isc_sockstatscounter_tcp4recvfail
 651 };
 652 static const isc_statscounter_t tcp6statsindex[] = {
 653         isc_sockstatscounter_tcp6open,
 654         isc_sockstatscounter_tcp6openfail,
 655         isc_sockstatscounter_tcp6close,
 656         isc_sockstatscounter_tcp6bindfail,
 657         isc_sockstatscounter_tcp6connectfail,
 658         isc_sockstatscounter_tcp6connect,
 659         isc_sockstatscounter_tcp6acceptfail,
 660         isc_sockstatscounter_tcp6accept,
 661         isc_sockstatscounter_tcp6sendfail,
 662         isc_sockstatscounter_tcp6recvfail
 663 };
 664 static const isc_statscounter_t unixstatsindex[] = {
 665         isc_sockstatscounter_unixopen,
 666         isc_sockstatscounter_unixopenfail,
 667         isc_sockstatscounter_unixclose,
 668         isc_sockstatscounter_unixbindfail,
 669         isc_sockstatscounter_unixconnectfail,
 670         isc_sockstatscounter_unixconnect,
 671         isc_sockstatscounter_unixacceptfail,
 672         isc_sockstatscounter_unixaccept,
 673         isc_sockstatscounter_unixsendfail,
 674         isc_sockstatscounter_unixrecvfail
 675 };
 676 static const isc_statscounter_t fdwatchstatsindex[] = {
 677         -1,
 678         -1,
 679         isc_sockstatscounter_fdwatchclose,
 680         isc_sockstatscounter_fdwatchbindfail,
 681         isc_sockstatscounter_fdwatchconnectfail,
 682         isc_sockstatscounter_fdwatchconnect,
 683         -1,
 684         -1,
 685         isc_sockstatscounter_fdwatchsendfail,
 686         isc_sockstatscounter_fdwatchrecvfail
 687 };
 688
 689 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL) || \
 690     defined(USE_WATCHER_THREAD)
 691 static void
 692 manager_log(isc__socketmgr_t *sockmgr,
 693             isc_logcategory_t *category, isc_logmodule_t *module, int level,
 694             const char *fmt, ...) ISC_FORMAT_PRINTF(5, 6);
 695 static void
 696 manager_log(isc__socketmgr_t *sockmgr,
 697             isc_logcategory_t *category, isc_logmodule_t *module, int level,
 698             const char *fmt, ...)
 699 {
 700         char msgbuf[2048];
 701         va_list ap;
 702
 703         if (! isc_log_wouldlog(isc_lctx, level))
 704                 return;
 705
 706         va_start(ap, fmt);
 707         vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
 708         va_end(ap);
 709
 710         isc_log_write(isc_lctx, category, module, level,
 711                       "sockmgr %p: %s", sockmgr, msgbuf);
 712 }
 713 #endif
 714
 715 static void
 716 socket_log(isc__socket_t *sock, isc_sockaddr_t *address,
 717            isc_logcategory_t *category, isc_logmodule_t *module, int level,
 718            isc_msgcat_t *msgcat, int msgset, int message,
 719            const char *fmt, ...) ISC_FORMAT_PRINTF(9, 10);
 720 static void
 721 socket_log(isc__socket_t *sock, isc_sockaddr_t *address,
 722            isc_logcategory_t *category, isc_logmodule_t *module, int level,
 723            isc_msgcat_t *msgcat, int msgset, int message,
 724            const char *fmt, ...)
 725 {
 726         char msgbuf[2048];
 727         char peerbuf[ISC_SOCKADDR_FORMATSIZE];
 728         va_list ap;
 729
 730         if (! isc_log_wouldlog(isc_lctx, level))
 731                 return;
 732
 733         va_start(ap, fmt);
 734         vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
 735         va_end(ap);
 736
 737         if (address == NULL) {
 738                 isc_log_iwrite(isc_lctx, category, module, level,
 739                                msgcat, msgset, message,
 740                                "socket %p: %s", sock, msgbuf);
 741         } else {
 742                 isc_sockaddr_format(address, peerbuf, sizeof(peerbuf));
 743                 isc_log_iwrite(isc_lctx, category, module, level,
 744                                msgcat, msgset, message,
 745                                "socket %p %s: %s", sock, peerbuf, msgbuf);
 746         }
 747 }
 748
 749 #if defined(_AIX) && defined(ISC_NET_BSD44MSGHDR) && \
 750     defined(USE_CMSG) && defined(IPV6_RECVPKTINFO)
 751 /*
 752  * AIX has a kernel bug where IPV6_RECVPKTINFO gets cleared by
 753  * setting IPV6_V6ONLY.
 754  */
 755 static void
 756 FIX_IPV6_RECVPKTINFO(isc__socket_t *sock)
 757 {
 758         char strbuf[ISC_STRERRORSIZE];
 759         int on = 1;
 760
 761         if (sock->pf != AF_INET6 || sock->type != isc_sockettype_udp)
 762                 return;
 763
 764         if (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_RECVPKTINFO,
 765                        (void *)&on, sizeof(on)) < 0) {
 766
 767                 isc__strerror(errno, strbuf, sizeof(strbuf));
 768                 UNEXPECTED_ERROR(__FILE__, __LINE__,
 769                                  "setsockopt(%d, IPV6_RECVPKTINFO) "
 770                                  "%s: %s", sock->fd,
 771                                  isc_msgcat_get(isc_msgcat,
 772                                                 ISC_MSGSET_GENERAL,
 773                                                 ISC_MSG_FAILED,
 774                                                 "failed"),
 775                                  strbuf);
 776         }
 777 }
 778 #else
 779 #define FIX_IPV6_RECVPKTINFO(sock) (void)0
 780 #endif
 781
 782 /*%
 783  * Increment socket-related statistics counters.
 784  */
 785 static inline void
 786 inc_stats(isc_stats_t *stats, isc_statscounter_t counterid) {
 787         REQUIRE(counterid != -1);
 788
 789         if (stats != NULL)
 790                 isc_stats_increment(stats, counterid);
 791 }
 792
 793 static inline isc_result_t
 794 watch_fd(isc__socketmgr_t *manager, int fd, int msg) {
 795         isc_result_t result = ISC_R_SUCCESS;
 796
 797 #ifdef USE_KQUEUE
 798         struct kevent evchange;
 799
 800         memset(&evchange, 0, sizeof(evchange));
 801         if (msg == SELECT_POKE_READ)
 802                 evchange.filter = EVFILT_READ;
 803         else
 804                 evchange.filter = EVFILT_WRITE;
 805         evchange.flags = EV_ADD;
 806         evchange.ident = fd;
 807         if (kevent(manager->kqueue_fd, &evchange, 1, NULL, 0, NULL) != 0)
 808                 result = isc__errno2result(errno);
 809
 810         return (result);
 811 #elif defined(USE_EPOLL)
 812         struct epoll_event event;
 813
 814         if (msg == SELECT_POKE_READ)
 815                 event.events = EPOLLIN;
 816         else
 817                 event.events = EPOLLOUT;
 818         memset(&event.data, 0, sizeof(event.data));
 819         event.data.fd = fd;
 820         if (epoll_ctl(manager->epoll_fd, EPOLL_CTL_ADD, fd, &event) == -1 &&
 821             errno != EEXIST) {
 822                 result = isc__errno2result(errno);
 823         }
 824
 825         return (result);
 826 #elif defined(USE_DEVPOLL)
 827         struct pollfd pfd;
 828         int lockid = FDLOCK_ID(fd);
 829
 830         memset(&pfd, 0, sizeof(pfd));
 831         if (msg == SELECT_POKE_READ)
 832                 pfd.events = POLLIN;
 833         else
 834                 pfd.events = POLLOUT;
 835         pfd.fd = fd;
 836         pfd.revents = 0;
 837         LOCK(&manager->fdlock[lockid]);
 838         if (write(manager->devpoll_fd, &pfd, sizeof(pfd)) == -1)
 839                 result = isc__errno2result(errno);
 840         else {
 841                 if (msg == SELECT_POKE_READ)
 842                         manager->fdpollinfo[fd].want_read = 1;
 843                 else
 844                         manager->fdpollinfo[fd].want_write = 1;
 845         }
 846         UNLOCK(&manager->fdlock[lockid]);
 847
 848         return (result);
 849 #elif defined(USE_SELECT)
 850         LOCK(&manager->lock);
 851         if (msg == SELECT_POKE_READ)
 852                 FD_SET(fd, manager->read_fds);
 853         if (msg == SELECT_POKE_WRITE)
 854                 FD_SET(fd, manager->write_fds);
 855         UNLOCK(&manager->lock);
 856
 857         return (result);
 858 #endif
 859 }
 860
 861 static inline isc_result_t
 862 unwatch_fd(isc__socketmgr_t *manager, int fd, int msg) {
 863         isc_result_t result = ISC_R_SUCCESS;
 864
 865 #ifdef USE_KQUEUE
 866         struct kevent evchange;
 867
 868         memset(&evchange, 0, sizeof(evchange));
 869         if (msg == SELECT_POKE_READ)
 870                 evchange.filter = EVFILT_READ;
 871         else
 872                 evchange.filter = EVFILT_WRITE;
 873         evchange.flags = EV_DELETE;
 874         evchange.ident = fd;
 875         if (kevent(manager->kqueue_fd, &evchange, 1, NULL, 0, NULL) != 0)
 876                 result = isc__errno2result(errno);
 877
 878         return (result);
 879 #elif defined(USE_EPOLL)
 880         struct epoll_event event;
 881
 882         if (msg == SELECT_POKE_READ)
 883                 event.events = EPOLLIN;
 884         else
 885                 event.events = EPOLLOUT;
 886         memset(&event.data, 0, sizeof(event.data));
 887         event.data.fd = fd;
 888         if (epoll_ctl(manager->epoll_fd, EPOLL_CTL_DEL, fd, &event) == -1 &&
 889             errno != ENOENT) {
 890                 char strbuf[ISC_STRERRORSIZE];
 891                 isc__strerror(errno, strbuf, sizeof(strbuf));
 892                 UNEXPECTED_ERROR(__FILE__, __LINE__,
 893                                  "epoll_ctl(DEL), %d: %s", fd, strbuf);
 894                 result = ISC_R_UNEXPECTED;
 895         }
 896         return (result);
 897 #elif defined(USE_DEVPOLL)
 898         struct pollfd pfds[2];
 899         size_t writelen = sizeof(pfds[0]);
 900         int lockid = FDLOCK_ID(fd);
 901
 902         memset(pfds, 0, sizeof(pfds));
 903         pfds[0].events = POLLREMOVE;
 904         pfds[0].fd = fd;
 905
 906         /*
 907          * Canceling read or write polling via /dev/poll is tricky.  Since it
 908          * only provides a way of canceling per FD, we may need to re-poll the
 909          * socket for the other operation.
 910          */
 911         LOCK(&manager->fdlock[lockid]);
 912         if (msg == SELECT_POKE_READ &&
 913             manager->fdpollinfo[fd].want_write == 1) {
 914                 pfds[1].events = POLLOUT;
 915                 pfds[1].fd = fd;
 916                 writelen += sizeof(pfds[1]);
 917         }
 918         if (msg == SELECT_POKE_WRITE &&
 919             manager->fdpollinfo[fd].want_read == 1) {
 920                 pfds[1].events = POLLIN;
 921                 pfds[1].fd = fd;
 922                 writelen += sizeof(pfds[1]);
 923         }
 924
 925         if (write(manager->devpoll_fd, pfds, writelen) == -1)
 926                 result = isc__errno2result(errno);
 927         else {
 928                 if (msg == SELECT_POKE_READ)
 929                         manager->fdpollinfo[fd].want_read = 0;
 930                 else
 931                         manager->fdpollinfo[fd].want_write = 0;
 932         }
 933         UNLOCK(&manager->fdlock[lockid]);
 934
 935         return (result);
 936 #elif defined(USE_SELECT)
 937         LOCK(&manager->lock);
 938         if (msg == SELECT_POKE_READ)
 939                 FD_CLR(fd, manager->read_fds);
 940         else if (msg == SELECT_POKE_WRITE)
 941                 FD_CLR(fd, manager->write_fds);
 942         UNLOCK(&manager->lock);
 943
 944         return (result);
 945 #endif
 946 }
 947
 948 static void
 949 wakeup_socket(isc__socketmgr_t *manager, int fd, int msg) {
 950         isc_result_t result;
 951         int lockid = FDLOCK_ID(fd);
 952
 953         /*
 954          * This is a wakeup on a socket.  If the socket is not in the
 955          * process of being closed, start watching it for either reads
 956          * or writes.
 957          */
 958
 959         INSIST(fd >= 0 && fd < (int)manager->maxsocks);
 960
 961         if (msg == SELECT_POKE_CLOSE) {
 962                 /* No one should be updating fdstate, so no need to lock it */
 963                 INSIST(manager->fdstate[fd] == CLOSE_PENDING);
 964                 manager->fdstate[fd] = CLOSED;
 965                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
 966                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
 967                 (void)close(fd);
 968                 return;
 969         }
 970
 971         LOCK(&manager->fdlock[lockid]);
 972         if (manager->fdstate[fd] == CLOSE_PENDING) {
 973                 UNLOCK(&manager->fdlock[lockid]);
 974
 975                 /*
 976                  * We accept (and ignore) any error from unwatch_fd() as we are
 977                  * closing the socket, hoping it doesn't leave dangling state in
 978                  * the kernel.
 979                  * Note that unwatch_fd() must be called after releasing the
 980                  * fdlock; otherwise it could cause deadlock due to a lock order
 981                  * reversal.
 982                  */
 983                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
 984                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
 985                 return;
 986         }
 987         if (manager->fdstate[fd] != MANAGED) {
 988                 UNLOCK(&manager->fdlock[lockid]);
 989                 return;
 990         }
 991         UNLOCK(&manager->fdlock[lockid]);
 992
 993         /*
 994          * Set requested bit.
 995          */
 996         result = watch_fd(manager, fd, msg);
 997         if (result != ISC_R_SUCCESS) {
 998                 /*
 999                  * XXXJT: what should we do?  Ignoring the failure of watching
1000                  * a socket will make the application dysfunctional, but there
1001                  * seems to be no reasonable recovery process.
1002                  */
1003                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
1004                               ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
1005                               "failed to start watching FD (%d): %s",
1006                               fd, isc_result_totext(result));
1007         }
1008 }
1009
1010 #ifdef USE_WATCHER_THREAD
1011 /*
1012  * Poke the select loop when there is something for us to do.
1013  * The write is required (by POSIX) to complete.  That is, we
1014  * will not get partial writes.
1015  */
1016 static void
1017 select_poke(isc__socketmgr_t *mgr, int fd, int msg) {
1018         int cc;
1019         int buf[2];
1020         char strbuf[ISC_STRERRORSIZE];
1021
1022         buf[0] = fd;
1023         buf[1] = msg;
1024
1025         do {
1026                 cc = write(mgr->pipe_fds[1], buf, sizeof(buf));
1027 #ifdef ENOSR
1028                 /*
1029                  * Treat ENOSR as EAGAIN but loop slowly as it is
1030                  * unlikely to clear fast.
1031                  */
1032                 if (cc < 0 && errno == ENOSR) {
1033                         sleep(1);
1034                         errno = EAGAIN;
1035                 }
1036 #endif
1037         } while (cc < 0 && SOFT_ERROR(errno));
1038
1039         if (cc < 0) {
1040                 isc__strerror(errno, strbuf, sizeof(strbuf));
1041                 FATAL_ERROR(__FILE__, __LINE__,
1042                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
1043                                            ISC_MSG_WRITEFAILED,
1044                                            "write() failed "
1045                                            "during watcher poke: %s"),
1046                             strbuf);
1047         }
1048
1049         INSIST(cc == sizeof(buf));
1050 }
1051
1052 /*
1053  * Read a message on the internal fd.
1054  */
1055 static void
1056 select_readmsg(isc__socketmgr_t *mgr, int *fd, int *msg) {
1057         int buf[2];
1058         int cc;
1059         char strbuf[ISC_STRERRORSIZE];
1060
1061         cc = read(mgr->pipe_fds[0], buf, sizeof(buf));
1062         if (cc < 0) {
1063                 *msg = SELECT_POKE_NOTHING;
1064                 *fd = -1;       /* Silence compiler. */
1065                 if (SOFT_ERROR(errno))
1066                         return;
1067
1068                 isc__strerror(errno, strbuf, sizeof(strbuf));
1069                 FATAL_ERROR(__FILE__, __LINE__,
1070                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
1071                                            ISC_MSG_READFAILED,
1072                                            "read() failed "
1073                                            "during watcher poke: %s"),
1074                             strbuf);
1075
1076                 return;
1077         }
1078         INSIST(cc == sizeof(buf));
1079
1080         *fd = buf[0];
1081         *msg = buf[1];
1082 }
1083 #else /* USE_WATCHER_THREAD */
1084 /*
1085  * Update the state of the socketmgr when something changes.
1086  */
1087 static void
1088 select_poke(isc__socketmgr_t *manager, int fd, int msg) {
1089         if (msg == SELECT_POKE_SHUTDOWN)
1090                 return;
1091         else if (fd >= 0)
1092                 wakeup_socket(manager, fd, msg);
1093         return;
1094 }
1095 #endif /* USE_WATCHER_THREAD */
1096
1097 /*
1098  * Make a fd non-blocking.
1099  */
1100 static isc_result_t
1101 make_nonblock(int fd) {
1102         int ret;
1103         int flags;
1104         char strbuf[ISC_STRERRORSIZE];
1105 #ifdef USE_FIONBIO_IOCTL
1106         int on = 1;
1107
1108         ret = ioctl(fd, FIONBIO, (char *)&on);
1109 #else
1110         flags = fcntl(fd, F_GETFL, 0);
1111         flags |= PORT_NONBLOCK;
1112         ret = fcntl(fd, F_SETFL, flags);
1113 #endif
1114
1115         if (ret == -1) {
1116                 isc__strerror(errno, strbuf, sizeof(strbuf));
1117                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1118 #ifdef USE_FIONBIO_IOCTL
1119                                  "ioctl(%d, FIONBIO, &on): %s", fd,
1120 #else
1121                                  "fcntl(%d, F_SETFL, %d): %s", fd, flags,
1122 #endif
1123                                  strbuf);
1124
1125                 return (ISC_R_UNEXPECTED);
1126         }
1127
1128         return (ISC_R_SUCCESS);
1129 }
1130
1131 #ifdef USE_CMSG
1132 /*
1133  * Not all OSes support advanced CMSG macros: CMSG_LEN and CMSG_SPACE.
1134  * In order to ensure as much portability as possible, we provide wrapper
1135  * functions of these macros.
1136  * Note that cmsg_space() could run slow on OSes that do not have
1137  * CMSG_SPACE.
1138  */
1139 static inline ISC_SOCKADDR_LEN_T
1140 cmsg_len(ISC_SOCKADDR_LEN_T len) {
1141 #ifdef CMSG_LEN
1142         return (CMSG_LEN(len));
1143 #else
1144         ISC_SOCKADDR_LEN_T hdrlen;
1145
1146         /*
1147          * Cast NULL so that any pointer arithmetic performed by CMSG_DATA
1148          * is correct.
1149          */
1150         hdrlen = (ISC_SOCKADDR_LEN_T)CMSG_DATA(((struct cmsghdr *)NULL));
1151         return (hdrlen + len);
1152 #endif
1153 }
1154
1155 static inline ISC_SOCKADDR_LEN_T
1156 cmsg_space(ISC_SOCKADDR_LEN_T len) {
1157 #ifdef CMSG_SPACE
1158         return (CMSG_SPACE(len));
1159 #else
1160         struct msghdr msg;
1161         struct cmsghdr *cmsgp;
1162         /*
1163          * XXX: The buffer length is an ad-hoc value, but should be enough
1164          * in a practical sense.
1165          */
1166         char dummybuf[sizeof(struct cmsghdr) + 1024];
1167
1168         memset(&msg, 0, sizeof(msg));
1169         msg.msg_control = dummybuf;
1170         msg.msg_controllen = sizeof(dummybuf);
1171
1172         cmsgp = (struct cmsghdr *)dummybuf;
1173         cmsgp->cmsg_len = cmsg_len(len);
1174
1175         cmsgp = CMSG_NXTHDR(&msg, cmsgp);
1176         if (cmsgp != NULL)
1177                 return ((char *)cmsgp - (char *)msg.msg_control);
1178         else
1179                 return (0);
1180 #endif
1181 }
1182 #endif /* USE_CMSG */
1183
1184 /*
1185  * Process control messages received on a socket.
1186  */
1187 static void
1188 process_cmsg(isc__socket_t *sock, struct msghdr *msg, isc_socketevent_t *dev) {
1189 #ifdef USE_CMSG
1190         struct cmsghdr *cmsgp;
1191 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1192         struct in6_pktinfo *pktinfop;
1193 #endif
1194 #ifdef SO_TIMESTAMP
1195         struct timeval *timevalp;
1196 #endif
1197 #endif
1198
1199         /*
1200          * sock is used only when ISC_NET_BSD44MSGHDR and USE_CMSG are defined.
1201          * msg and dev are used only when ISC_NET_BSD44MSGHDR is defined.
1202          * They are all here, outside of the CPP tests, because it is
1203          * more consistent with the usual ISC coding style.
1204          */
1205         UNUSED(sock);
1206         UNUSED(msg);
1207         UNUSED(dev);
1208
1209 #ifdef ISC_NET_BSD44MSGHDR
1210
1211 #ifdef MSG_TRUNC
1212         if ((msg->msg_flags & MSG_TRUNC) == MSG_TRUNC)
1213                 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
1214 #endif
1215
1216 #ifdef MSG_CTRUNC
1217         if ((msg->msg_flags & MSG_CTRUNC) == MSG_CTRUNC)
1218                 dev->attributes |= ISC_SOCKEVENTATTR_CTRUNC;
1219 #endif
1220
1221 #ifndef USE_CMSG
1222         return;
1223 #else
1224         if (msg->msg_controllen == 0U || msg->msg_control == NULL)
1225                 return;
1226
1227 #ifdef SO_TIMESTAMP
1228         timevalp = NULL;
1229 #endif
1230 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1231         pktinfop = NULL;
1232 #endif
1233
1234         cmsgp = CMSG_FIRSTHDR(msg);
1235         while (cmsgp != NULL) {
1236                 socket_log(sock, NULL, TRACE,
1237                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PROCESSCMSG,
1238                            "processing cmsg %p", cmsgp);
1239
1240 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1241                 if (cmsgp->cmsg_level == IPPROTO_IPV6
1242                     && cmsgp->cmsg_type == IPV6_PKTINFO) {
1243
1244                         pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
1245                         memcpy(&dev->pktinfo, pktinfop,
1246                                sizeof(struct in6_pktinfo));
1247                         dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
1248                         socket_log(sock, NULL, TRACE,
1249                                    isc_msgcat, ISC_MSGSET_SOCKET,
1250                                    ISC_MSG_IFRECEIVED,
1251                                    "interface received on ifindex %u",
1252                                    dev->pktinfo.ipi6_ifindex);
1253                         if (IN6_IS_ADDR_MULTICAST(&pktinfop->ipi6_addr))
1254                                 dev->attributes |= ISC_SOCKEVENTATTR_MULTICAST;
1255                         goto next;
1256                 }
1257 #endif
1258
1259 #ifdef SO_TIMESTAMP
1260                 if (cmsgp->cmsg_level == SOL_SOCKET
1261                     && cmsgp->cmsg_type == SCM_TIMESTAMP) {
1262                         timevalp = (struct timeval *)CMSG_DATA(cmsgp);
1263                         dev->timestamp.seconds = timevalp->tv_sec;
1264                         dev->timestamp.nanoseconds = timevalp->tv_usec * 1000;
1265                         dev->attributes |= ISC_SOCKEVENTATTR_TIMESTAMP;
1266                         goto next;
1267                 }
1268 #endif
1269
1270         next:
1271                 cmsgp = CMSG_NXTHDR(msg, cmsgp);
1272         }
1273 #endif /* USE_CMSG */
1274
1275 #endif /* ISC_NET_BSD44MSGHDR */
1276 }
1277
1278 /*
1279  * Construct an iov array and attach it to the msghdr passed in.  This is
1280  * the SEND constructor, which will use the used region of the buffer
1281  * (if using a buffer list) or will use the internal region (if a single
1282  * buffer I/O is requested).
1283  *
1284  * Nothing can be NULL, and the done event must list at least one buffer
1285  * on the buffer linked list for this function to be meaningful.
1286  *
1287  * If write_countp != NULL, *write_countp will hold the number of bytes
1288  * this transaction can send.
1289  */
1290 static void
1291 build_msghdr_send(isc__socket_t *sock, isc_socketevent_t *dev,
1292                   struct msghdr *msg, struct iovec *iov, size_t *write_countp)
1293 {
1294         unsigned int iovcount;
1295         isc_buffer_t *buffer;
1296         isc_region_t used;
1297         size_t write_count;
1298         size_t skip_count;
1299
1300         memset(msg, 0, sizeof(*msg));
1301
1302         if (!sock->connected) {
1303                 msg->msg_name = (void *)&dev->address.type.sa;
1304                 msg->msg_namelen = dev->address.length;
1305         } else {
1306                 msg->msg_name = NULL;
1307                 msg->msg_namelen = 0;
1308         }
1309
1310         buffer = ISC_LIST_HEAD(dev->bufferlist);
1311         write_count = 0;
1312         iovcount = 0;
1313
1314         /*
1315          * Single buffer I/O?  Skip what we've done so far in this region.
1316          */
1317         if (buffer == NULL) {
1318                 write_count = dev->region.length - dev->n;
1319                 iov[0].iov_base = (void *)(dev->region.base + dev->n);
1320                 iov[0].iov_len = write_count;
1321                 iovcount = 1;
1322
1323                 goto config;
1324         }
1325
1326         /*
1327          * Multibuffer I/O.
1328          * Skip the data in the buffer list that we have already written.
1329          */
1330         skip_count = dev->n;
1331         while (buffer != NULL) {
1332                 REQUIRE(ISC_BUFFER_VALID(buffer));
1333                 if (skip_count < isc_buffer_usedlength(buffer))
1334                         break;
1335                 skip_count -= isc_buffer_usedlength(buffer);
1336                 buffer = ISC_LIST_NEXT(buffer, link);
1337         }
1338
1339         while (buffer != NULL) {
1340                 INSIST(iovcount < MAXSCATTERGATHER_SEND);
1341
1342                 isc_buffer_usedregion(buffer, &used);
1343
1344                 if (used.length > 0) {
1345                         iov[iovcount].iov_base = (void *)(used.base
1346                                                           + skip_count);
1347                         iov[iovcount].iov_len = used.length - skip_count;
1348                         write_count += (used.length - skip_count);
1349                         skip_count = 0;
1350                         iovcount++;
1351                 }
1352                 buffer = ISC_LIST_NEXT(buffer, link);
1353         }
1354
1355         INSIST(skip_count == 0U);
1356
1357  config:
1358         msg->msg_iov = iov;
1359         msg->msg_iovlen = iovcount;
1360
1361 #ifdef ISC_NET_BSD44MSGHDR
1362         msg->msg_control = NULL;
1363         msg->msg_controllen = 0;
1364         msg->msg_flags = 0;
1365 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
1366         if ((sock->type == isc_sockettype_udp)
1367             && ((dev->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0)) {
1368 #if defined(IPV6_USE_MIN_MTU)
1369                 int use_min_mtu = 1;    /* -1, 0, 1 */
1370 #endif
1371                 struct cmsghdr *cmsgp;
1372                 struct in6_pktinfo *pktinfop;
1373
1374                 socket_log(sock, NULL, TRACE,
1375                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_SENDTODATA,
1376                            "sendto pktinfo data, ifindex %u",
1377                            dev->pktinfo.ipi6_ifindex);
1378
1379                 msg->msg_controllen = cmsg_space(sizeof(struct in6_pktinfo));
1380                 INSIST(msg->msg_controllen <= sock->sendcmsgbuflen);
1381                 msg->msg_control = (void *)sock->sendcmsgbuf;
1382
1383                 cmsgp = (struct cmsghdr *)sock->sendcmsgbuf;
1384                 cmsgp->cmsg_level = IPPROTO_IPV6;
1385                 cmsgp->cmsg_type = IPV6_PKTINFO;
1386                 cmsgp->cmsg_len = cmsg_len(sizeof(struct in6_pktinfo));
1387                 pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
1388                 memcpy(pktinfop, &dev->pktinfo, sizeof(struct in6_pktinfo));
1389 #if defined(IPV6_USE_MIN_MTU)
1390                 /*
1391                  * Set IPV6_USE_MIN_MTU as a per packet option as FreeBSD
1392                  * ignores setsockopt(IPV6_USE_MIN_MTU) when IPV6_PKTINFO
1393                  * is used.
1394                  */
1395                 cmsgp = (struct cmsghdr *)(sock->sendcmsgbuf +
1396                                            msg->msg_controllen);
1397                 msg->msg_controllen += cmsg_space(sizeof(use_min_mtu));
1398                 INSIST(msg->msg_controllen <= sock->sendcmsgbuflen);
1399
1400                 cmsgp->cmsg_level = IPPROTO_IPV6;
1401                 cmsgp->cmsg_type = IPV6_USE_MIN_MTU;
1402                 cmsgp->cmsg_len = cmsg_len(sizeof(use_min_mtu));
1403                 memcpy(CMSG_DATA(cmsgp), &use_min_mtu, sizeof(use_min_mtu));
1404 #endif
1405         }
1406 #endif /* USE_CMSG && ISC_PLATFORM_HAVEIPV6 */
1407 #else /* ISC_NET_BSD44MSGHDR */
1408         msg->msg_accrights = NULL;
1409         msg->msg_accrightslen = 0;
1410 #endif /* ISC_NET_BSD44MSGHDR */
1411
1412         if (write_countp != NULL)
1413                 *write_countp = write_count;
1414 }
1415
1416 /*
1417  * Construct an iov array and attach it to the msghdr passed in.  This is
1418  * the RECV constructor, which will use the available region of the buffer
1419  * (if using a buffer list) or will use the internal region (if a single
1420  * buffer I/O is requested).
1421  *
1422  * Nothing can be NULL, and the done event must list at least one buffer
1423  * on the buffer linked list for this function to be meaningful.
1424  *
1425  * If read_countp != NULL, *read_countp will hold the number of bytes
1426  * this transaction can receive.
1427  */
1428 static void
1429 build_msghdr_recv(isc__socket_t *sock, isc_socketevent_t *dev,
1430                   struct msghdr *msg, struct iovec *iov, size_t *read_countp)
1431 {
1432         unsigned int iovcount;
1433         isc_buffer_t *buffer;
1434         isc_region_t available;
1435         size_t read_count;
1436
1437         memset(msg, 0, sizeof(struct msghdr));
1438
1439         if (sock->type == isc_sockettype_udp) {
1440                 memset(&dev->address, 0, sizeof(dev->address));
1441 #ifdef BROKEN_RECVMSG
1442                 if (sock->pf == AF_INET) {
1443                         msg->msg_name = (void *)&dev->address.type.sin;
1444                         msg->msg_namelen = sizeof(dev->address.type.sin6);
1445                 } else if (sock->pf == AF_INET6) {
1446                         msg->msg_name = (void *)&dev->address.type.sin6;
1447                         msg->msg_namelen = sizeof(dev->address.type.sin6);
1448 #ifdef ISC_PLATFORM_HAVESYSUNH
1449                 } else if (sock->pf == AF_UNIX) {
1450                         msg->msg_name = (void *)&dev->address.type.sunix;
1451                         msg->msg_namelen = sizeof(dev->address.type.sunix);
1452 #endif
1453                 } else {
1454                         msg->msg_name = (void *)&dev->address.type.sa;
1455                         msg->msg_namelen = sizeof(dev->address.type);
1456                 }
1457 #else
1458                 msg->msg_name = (void *)&dev->address.type.sa;
1459                 msg->msg_namelen = sizeof(dev->address.type);
1460 #endif
1461 #ifdef ISC_NET_RECVOVERFLOW
1462                 /* If needed, steal one iovec for overflow detection. */
1463                 maxiov--;
1464 #endif
1465         } else { /* TCP */
1466                 msg->msg_name = NULL;
1467                 msg->msg_namelen = 0;
1468                 dev->address = sock->peer_address;
1469         }
1470
1471         buffer = ISC_LIST_HEAD(dev->bufferlist);
1472         read_count = 0;
1473
1474         /*
1475          * Single buffer I/O?  Skip what we've done so far in this region.
1476          */
1477         if (buffer == NULL) {
1478                 read_count = dev->region.length - dev->n;
1479                 iov[0].iov_base = (void *)(dev->region.base + dev->n);
1480                 iov[0].iov_len = read_count;
1481                 iovcount = 1;
1482
1483                 goto config;
1484         }
1485
1486         /*
1487          * Multibuffer I/O.
1488          * Skip empty buffers.
1489          */
1490         while (buffer != NULL) {
1491                 REQUIRE(ISC_BUFFER_VALID(buffer));
1492                 if (isc_buffer_availablelength(buffer) != 0)
1493                         break;
1494                 buffer = ISC_LIST_NEXT(buffer, link);
1495         }
1496
1497         iovcount = 0;
1498         while (buffer != NULL) {
1499                 INSIST(iovcount < MAXSCATTERGATHER_RECV);
1500
1501                 isc_buffer_availableregion(buffer, &available);
1502
1503                 if (available.length > 0) {
1504                         iov[iovcount].iov_base = (void *)(available.base);
1505                         iov[iovcount].iov_len = available.length;
1506                         read_count += available.length;
1507                         iovcount++;
1508                 }
1509                 buffer = ISC_LIST_NEXT(buffer, link);
1510         }
1511
1512  config:
1513
1514         /*
1515          * If needed, set up to receive that one extra byte.  Note that
1516          * we know there is at least one iov left, since we stole it
1517          * at the top of this function.
1518          */
1519 #ifdef ISC_NET_RECVOVERFLOW
1520         if (sock->type == isc_sockettype_udp) {
1521                 iov[iovcount].iov_base = (void *)(&sock->overflow);
1522                 iov[iovcount].iov_len = 1;
1523                 iovcount++;
1524         }
1525 #endif
1526
1527         msg->msg_iov = iov;
1528         msg->msg_iovlen = iovcount;
1529
1530 #ifdef ISC_NET_BSD44MSGHDR
1531         msg->msg_control = NULL;
1532         msg->msg_controllen = 0;
1533         msg->msg_flags = 0;
1534 #if defined(USE_CMSG)
1535         if (sock->type == isc_sockettype_udp) {
1536                 msg->msg_control = sock->recvcmsgbuf;
1537                 msg->msg_controllen = sock->recvcmsgbuflen;
1538         }
1539 #endif /* USE_CMSG */
1540 #else /* ISC_NET_BSD44MSGHDR */
1541         msg->msg_accrights = NULL;
1542         msg->msg_accrightslen = 0;
1543 #endif /* ISC_NET_BSD44MSGHDR */
1544
1545         if (read_countp != NULL)
1546                 *read_countp = read_count;
1547 }
1548
1549 static void
1550 set_dev_address(isc_sockaddr_t *address, isc__socket_t *sock,
1551                 isc_socketevent_t *dev)
1552 {
1553         if (sock->type == isc_sockettype_udp) {
1554                 if (address != NULL)
1555                         dev->address = *address;
1556                 else
1557                         dev->address = sock->peer_address;
1558         } else if (sock->type == isc_sockettype_tcp) {
1559                 INSIST(address == NULL);
1560                 dev->address = sock->peer_address;
1561         }
1562 }
1563
1564 static void
1565 destroy_socketevent(isc_event_t *event) {
1566         isc_socketevent_t *ev = (isc_socketevent_t *)event;
1567
1568         INSIST(ISC_LIST_EMPTY(ev->bufferlist));
1569
1570         (ev->destroy)(event);
1571 }
1572
1573 static isc_socketevent_t *
1574 allocate_socketevent(isc__socket_t *sock, isc_eventtype_t eventtype,
1575                      isc_taskaction_t action, const void *arg)
1576 {
1577         isc_socketevent_t *ev;
1578
1579         ev = (isc_socketevent_t *)isc_event_allocate(sock->manager->mctx,
1580                                                      sock, eventtype,
1581                                                      action, arg,
1582                                                      sizeof(*ev));
1583
1584         if (ev == NULL)
1585                 return (NULL);
1586
1587         ev->result = ISC_R_UNSET;
1588         ISC_LINK_INIT(ev, ev_link);
1589         ISC_LIST_INIT(ev->bufferlist);
1590         ev->region.base = NULL;
1591         ev->n = 0;
1592         ev->offset = 0;
1593         ev->attributes = 0;
1594         ev->destroy = ev->ev_destroy;
1595         ev->ev_destroy = destroy_socketevent;
1596
1597         return (ev);
1598 }
1599
1600 #if defined(ISC_SOCKET_DEBUG)
1601 static void
1602 dump_msg(struct msghdr *msg) {
1603         unsigned int i;
1604
1605         printf("MSGHDR %p\n", msg);
1606         printf("\tname %p, namelen %ld\n", msg->msg_name,
1607                (long) msg->msg_namelen);
1608         printf("\tiov %p, iovlen %ld\n", msg->msg_iov,
1609                (long) msg->msg_iovlen);
1610         for (i = 0; i < (unsigned int)msg->msg_iovlen; i++)
1611                 printf("\t\t%d\tbase %p, len %ld\n", i,
1612                        msg->msg_iov[i].iov_base,
1613                        (long) msg->msg_iov[i].iov_len);
1614 #ifdef ISC_NET_BSD44MSGHDR
1615         printf("\tcontrol %p, controllen %ld\n", msg->msg_control,
1616                (long) msg->msg_controllen);
1617 #endif
1618 }
1619 #endif
1620
1621 #define DOIO_SUCCESS            0       /* i/o ok, event sent */
1622 #define DOIO_SOFT               1       /* i/o ok, soft error, no event sent */
1623 #define DOIO_HARD               2       /* i/o error, event sent */
1624 #define DOIO_EOF                3       /* EOF, no event sent */
1625
1626 static int
1627 doio_recv(isc__socket_t *sock, isc_socketevent_t *dev) {
1628         int cc;
1629         struct iovec iov[MAXSCATTERGATHER_RECV];
1630         size_t read_count;
1631         size_t actual_count;
1632         struct msghdr msghdr;
1633         isc_buffer_t *buffer;
1634         int recv_errno;
1635         char strbuf[ISC_STRERRORSIZE];
1636
1637         build_msghdr_recv(sock, dev, &msghdr, iov, &read_count);
1638
1639 #if defined(ISC_SOCKET_DEBUG)
1640         dump_msg(&msghdr);
1641 #endif
1642
1643         cc = recvmsg(sock->fd, &msghdr, 0);
1644         recv_errno = errno;
1645
1646 #if defined(ISC_SOCKET_DEBUG)
1647         dump_msg(&msghdr);
1648 #endif
1649
1650         if (cc < 0) {
1651                 if (SOFT_ERROR(recv_errno))
1652                         return (DOIO_SOFT);
1653
1654                 if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
1655                         isc__strerror(recv_errno, strbuf, sizeof(strbuf));
1656                         socket_log(sock, NULL, IOEVENT,
1657                                    isc_msgcat, ISC_MSGSET_SOCKET,
1658                                    ISC_MSG_DOIORECV,
1659                                   "doio_recv: recvmsg(%d) %d bytes, err %d/%s",
1660                                    sock->fd, cc, recv_errno, strbuf);
1661                 }
1662
1663 #define SOFT_OR_HARD(_system, _isc) \
1664         if (recv_errno == _system) { \
1665                 if (sock->connected) { \
1666                         dev->result = _isc; \
1667                         inc_stats(sock->manager->stats, \
1668                                   sock->statsindex[STATID_RECVFAIL]); \
1669                         return (DOIO_HARD); \
1670                 } \
1671                 return (DOIO_SOFT); \
1672         }
1673 #define ALWAYS_HARD(_system, _isc) \
1674         if (recv_errno == _system) { \
1675                 dev->result = _isc; \
1676                 inc_stats(sock->manager->stats, \
1677                           sock->statsindex[STATID_RECVFAIL]); \
1678                 return (DOIO_HARD); \
1679         }
1680
1681                 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1682                 SOFT_OR_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1683                 SOFT_OR_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1684                 SOFT_OR_HARD(EHOSTDOWN, ISC_R_HOSTDOWN);
1685                 /* HPUX 11.11 can return EADDRNOTAVAIL. */
1686                 SOFT_OR_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1687                 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1688                 /*
1689                  * HPUX returns EPROTO and EINVAL on receiving some ICMP/ICMPv6
1690                  * errors.
1691                  */
1692 #ifdef EPROTO
1693                 SOFT_OR_HARD(EPROTO, ISC_R_HOSTUNREACH);
1694 #endif
1695                 SOFT_OR_HARD(EINVAL, ISC_R_HOSTUNREACH);
1696
1697 #undef SOFT_OR_HARD
1698 #undef ALWAYS_HARD
1699
1700                 dev->result = isc__errno2result(recv_errno);
1701                 inc_stats(sock->manager->stats,
1702                           sock->statsindex[STATID_RECVFAIL]);
1703                 return (DOIO_HARD);
1704         }
1705
1706         /*
1707          * On TCP and UNIX sockets, zero length reads indicate EOF,
1708          * while on UDP sockets, zero length reads are perfectly valid,
1709          * although strange.
1710          */
1711         switch (sock->type) {
1712         case isc_sockettype_tcp:
1713         case isc_sockettype_unix:
1714                 if (cc == 0)
1715                         return (DOIO_EOF);
1716                 break;
1717         case isc_sockettype_udp:
1718                 break;
1719         case isc_sockettype_fdwatch:
1720         default:
1721                 INSIST(0);
1722         }
1723
1724         if (sock->type == isc_sockettype_udp) {
1725                 dev->address.length = msghdr.msg_namelen;
1726                 if (isc_sockaddr_getport(&dev->address) == 0) {
1727                         if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
1728                                 socket_log(sock, &dev->address, IOEVENT,
1729                                            isc_msgcat, ISC_MSGSET_SOCKET,
1730                                            ISC_MSG_ZEROPORT,
1731                                            "dropping source port zero packet");
1732                         }
1733                         return (DOIO_SOFT);
1734                 }
1735                 /*
1736                  * Simulate a firewall blocking UDP responses bigger than
1737                  * 512 bytes.
1738                  */
1739                 if (sock->manager->maxudp != 0 && cc > sock->manager->maxudp)
1740                         return (DOIO_SOFT);
1741         }
1742
1743         socket_log(sock, &dev->address, IOEVENT,
1744                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PKTRECV,
1745                    "packet received correctly");
1746
1747         /*
1748          * Overflow bit detection.  If we received MORE bytes than we should,
1749          * this indicates an overflow situation.  Set the flag in the
1750          * dev entry and adjust how much we read by one.
1751          */
1752 #ifdef ISC_NET_RECVOVERFLOW
1753         if ((sock->type == isc_sockettype_udp) && ((size_t)cc > read_count)) {
1754                 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
1755                 cc--;
1756         }
1757 #endif
1758
1759         /*
1760          * If there are control messages attached, run through them and pull
1761          * out the interesting bits.
1762          */
1763         if (sock->type == isc_sockettype_udp)
1764                 process_cmsg(sock, &msghdr, dev);
1765
1766         /*
1767          * update the buffers (if any) and the i/o count
1768          */
1769         dev->n += cc;
1770         actual_count = cc;
1771         buffer = ISC_LIST_HEAD(dev->bufferlist);
1772         while (buffer != NULL && actual_count > 0U) {
1773                 REQUIRE(ISC_BUFFER_VALID(buffer));
1774                 if (isc_buffer_availablelength(buffer) <= actual_count) {
1775                         actual_count -= isc_buffer_availablelength(buffer);
1776                         isc_buffer_add(buffer,
1777                                        isc_buffer_availablelength(buffer));
1778                 } else {
1779                         isc_buffer_add(buffer, actual_count);
1780                         actual_count = 0;
1781                         POST(actual_count);
1782                         break;
1783                 }
1784                 buffer = ISC_LIST_NEXT(buffer, link);
1785                 if (buffer == NULL) {
1786                         INSIST(actual_count == 0U);
1787                 }
1788         }
1789
1790         /*
1791          * If we read less than we expected, update counters,
1792          * and let the upper layer poke the descriptor.
1793          */
1794         if (((size_t)cc != read_count) && (dev->n < dev->minimum))
1795                 return (DOIO_SOFT);
1796
1797         /*
1798          * Full reads are posted, or partials if partials are ok.
1799          */
1800         dev->result = ISC_R_SUCCESS;
1801         return (DOIO_SUCCESS);
1802 }
1803
1804 /*
1805  * Returns:
1806  *      DOIO_SUCCESS    The operation succeeded.  dev->result contains
1807  *                      ISC_R_SUCCESS.
1808  *
1809  *      DOIO_HARD       A hard or unexpected I/O error was encountered.
1810  *                      dev->result contains the appropriate error.
1811  *
1812  *      DOIO_SOFT       A soft I/O error was encountered.  No senddone
1813  *                      event was sent.  The operation should be retried.
1814  *
1815  *      No other return values are possible.
1816  */
1817 static int
1818 doio_send(isc__socket_t *sock, isc_socketevent_t *dev) {
1819         int cc;
1820         struct iovec iov[MAXSCATTERGATHER_SEND];
1821         size_t write_count;
1822         struct msghdr msghdr;
1823         char addrbuf[ISC_SOCKADDR_FORMATSIZE];
1824         int attempts = 0;
1825         int send_errno;
1826         char strbuf[ISC_STRERRORSIZE];
1827
1828         build_msghdr_send(sock, dev, &msghdr, iov, &write_count);
1829
1830  resend:
1831         cc = sendmsg(sock->fd, &msghdr, 0);
1832         send_errno = errno;
1833
1834         /*
1835          * Check for error or block condition.
1836          */
1837         if (cc < 0) {
1838                 if (send_errno == EINTR && ++attempts < NRETRIES)
1839                         goto resend;
1840
1841                 if (SOFT_ERROR(send_errno))
1842                         return (DOIO_SOFT);
1843
1844 #define SOFT_OR_HARD(_system, _isc) \
1845         if (send_errno == _system) { \
1846                 if (sock->connected) { \
1847                         dev->result = _isc; \
1848                         inc_stats(sock->manager->stats, \
1849                                   sock->statsindex[STATID_SENDFAIL]); \
1850                         return (DOIO_HARD); \
1851                 } \
1852                 return (DOIO_SOFT); \
1853         }
1854 #define ALWAYS_HARD(_system, _isc) \
1855         if (send_errno == _system) { \
1856                 dev->result = _isc; \
1857                 inc_stats(sock->manager->stats, \
1858                           sock->statsindex[STATID_SENDFAIL]); \
1859                 return (DOIO_HARD); \
1860         }
1861
1862                 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1863                 ALWAYS_HARD(EACCES, ISC_R_NOPERM);
1864                 ALWAYS_HARD(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
1865                 ALWAYS_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1866                 ALWAYS_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1867 #ifdef EHOSTDOWN
1868                 ALWAYS_HARD(EHOSTDOWN, ISC_R_HOSTUNREACH);
1869 #endif
1870                 ALWAYS_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1871                 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1872                 ALWAYS_HARD(EPERM, ISC_R_HOSTUNREACH);
1873                 ALWAYS_HARD(EPIPE, ISC_R_NOTCONNECTED);
1874                 ALWAYS_HARD(ECONNRESET, ISC_R_CONNECTIONRESET);
1875
1876 #undef SOFT_OR_HARD
1877 #undef ALWAYS_HARD
1878
1879                 /*
1880                  * The other error types depend on whether or not the
1881                  * socket is UDP or TCP.  If it is UDP, some errors
1882                  * that we expect to be fatal under TCP are merely
1883                  * annoying, and are really soft errors.
1884                  *
1885                  * However, these soft errors are still returned as
1886                  * a status.
1887                  */
1888                 isc_sockaddr_format(&dev->address, addrbuf, sizeof(addrbuf));
1889                 isc__strerror(send_errno, strbuf, sizeof(strbuf));
1890                 UNEXPECTED_ERROR(__FILE__, __LINE__, "internal_send: %s: %s",
1891                                  addrbuf, strbuf);
1892                 dev->result = isc__errno2result(send_errno);
1893                 inc_stats(sock->manager->stats,
1894                           sock->statsindex[STATID_SENDFAIL]);
1895                 return (DOIO_HARD);
1896         }
1897
1898         if (cc == 0) {
1899                 inc_stats(sock->manager->stats,
1900                           sock->statsindex[STATID_SENDFAIL]);
1901                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1902                                  "doio_send: send() %s 0",
1903                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1904                                                 ISC_MSG_RETURNED, "returned"));
1905         }
1906
1907         /*
1908          * If we write less than we expected, update counters, poke.
1909          */
1910         dev->n += cc;
1911         if ((size_t)cc != write_count)
1912                 return (DOIO_SOFT);
1913
1914         /*
1915          * Exactly what we wanted to write.  We're done with this
1916          * entry.  Post its completion event.
1917          */
1918         dev->result = ISC_R_SUCCESS;
1919         return (DOIO_SUCCESS);
1920 }
1921
1922 /*
1923  * Kill.
1924  *
1925  * Caller must ensure that the socket is not locked and no external
1926  * references exist.
1927  */
1928 static void
1929 closesocket(isc__socketmgr_t *manager, isc__socket_t *sock, int fd) {
1930         isc_sockettype_t type = sock->type;
1931         int lockid = FDLOCK_ID(fd);
1932
1933         /*
1934          * No one has this socket open, so the watcher doesn't have to be
1935          * poked, and the socket doesn't have to be locked.
1936          */
1937         LOCK(&manager->fdlock[lockid]);
1938         manager->fds[fd] = NULL;
1939         if (type == isc_sockettype_fdwatch)
1940                 manager->fdstate[fd] = CLOSED;
1941         else
1942                 manager->fdstate[fd] = CLOSE_PENDING;
1943         UNLOCK(&manager->fdlock[lockid]);
1944         if (type == isc_sockettype_fdwatch) {
1945                 /*
1946                  * The caller may close the socket once this function returns,
1947                  * and `fd' may be reassigned for a new socket.  So we do
1948                  * unwatch_fd() here, rather than defer it via select_poke().
1949                  * Note: this may complicate data protection among threads and
1950                  * may reduce performance due to additional locks.  One way to
1951                  * solve this would be to dup() the watched descriptor, but we
1952                  * take a simpler approach at this moment.
1953                  */
1954                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
1955                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
1956         } else
1957                 select_poke(manager, fd, SELECT_POKE_CLOSE);
1958
1959         inc_stats(manager->stats, sock->statsindex[STATID_CLOSE]);
1960
1961         /*
1962          * update manager->maxfd here (XXX: this should be implemented more
1963          * efficiently)
1964          */
1965 #ifdef USE_SELECT
1966         LOCK(&manager->lock);
1967         if (manager->maxfd == fd) {
1968                 int i;
1969
1970                 manager->maxfd = 0;
1971                 for (i = fd - 1; i >= 0; i--) {
1972                         lockid = FDLOCK_ID(i);
1973
1974                         LOCK(&manager->fdlock[lockid]);
1975                         if (manager->fdstate[i] == MANAGED) {
1976                                 manager->maxfd = i;
1977                                 UNLOCK(&manager->fdlock[lockid]);
1978                                 break;
1979                         }
1980                         UNLOCK(&manager->fdlock[lockid]);
1981                 }
1982 #ifdef ISC_PLATFORM_USETHREADS
1983                 if (manager->maxfd < manager->pipe_fds[0])
1984                         manager->maxfd = manager->pipe_fds[0];
1985 #endif
1986         }
1987         UNLOCK(&manager->lock);
1988 #endif  /* USE_SELECT */
1989 }
1990
1991 static void
1992 destroy(isc__socket_t **sockp) {
1993         int fd;
1994         isc__socket_t *sock = *sockp;
1995         isc__socketmgr_t *manager = sock->manager;
1996
1997         socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
1998                    ISC_MSG_DESTROYING, "destroying");
1999
2000         INSIST(ISC_LIST_EMPTY(sock->accept_list));
2001         INSIST(ISC_LIST_EMPTY(sock->recv_list));
2002         INSIST(ISC_LIST_EMPTY(sock->send_list));
2003         INSIST(sock->connect_ev == NULL);
2004         REQUIRE(sock->fd == -1 || sock->fd < (int)manager->maxsocks);
2005
2006         if (sock->fd >= 0) {
2007                 fd = sock->fd;
2008                 sock->fd = -1;
2009                 closesocket(manager, sock, fd);
2010         }
2011
2012         LOCK(&manager->lock);
2013
2014         ISC_LIST_UNLINK(manager->socklist, sock, link);
2015
2016 #ifdef USE_WATCHER_THREAD
2017         if (ISC_LIST_EMPTY(manager->socklist))
2018                 SIGNAL(&manager->shutdown_ok);
2019 #endif /* USE_WATCHER_THREAD */
2020
2021         /* can't unlock manager as its memory context is still used */
2022         free_socket(sockp);
2023
2024         UNLOCK(&manager->lock);
2025 }
2026
2027 static isc_result_t
2028 allocate_socket(isc__socketmgr_t *manager, isc_sockettype_t type,
2029                 isc__socket_t **socketp)
2030 {
2031         isc__socket_t *sock;
2032         isc_result_t result;
2033         ISC_SOCKADDR_LEN_T cmsgbuflen;
2034
2035         sock = isc_mem_get(manager->mctx, sizeof(*sock));
2036
2037         if (sock == NULL)
2038                 return (ISC_R_NOMEMORY);
2039
2040         sock->common.magic = 0;
2041         sock->common.impmagic = 0;
2042         sock->references = 0;
2043
2044         sock->manager = manager;
2045         sock->type = type;
2046         sock->fd = -1;
2047         sock->statsindex = NULL;
2048
2049         ISC_LINK_INIT(sock, link);
2050
2051         sock->recvcmsgbuf = NULL;
2052         sock->sendcmsgbuf = NULL;
2053
2054         /*
2055          * set up cmsg buffers
2056          */
2057         cmsgbuflen = 0;
2058 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
2059         cmsgbuflen += cmsg_space(sizeof(struct in6_pktinfo));
2060 #endif
2061 #if defined(USE_CMSG) && defined(SO_TIMESTAMP)
2062         cmsgbuflen += cmsg_space(sizeof(struct timeval));
2063 #endif
2064         sock->recvcmsgbuflen = cmsgbuflen;
2065         if (sock->recvcmsgbuflen != 0U) {
2066                 sock->recvcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
2067                 if (sock->recvcmsgbuf == NULL) {
2068                         result = ISC_R_NOMEMORY;
2069                         goto error;
2070                 }
2071         }
2072
2073         cmsgbuflen = 0;
2074 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
2075         cmsgbuflen += cmsg_space(sizeof(struct in6_pktinfo));
2076 #if defined(IPV6_USE_MIN_MTU)
2077         /*
2078          * Provide space for working around FreeBSD's broken IPV6_USE_MIN_MTU
2079          * support.
2080          */
2081         cmsgbuflen += cmsg_space(sizeof(int));
2082 #endif
2083 #endif
2084         sock->sendcmsgbuflen = cmsgbuflen;
2085         if (sock->sendcmsgbuflen != 0U) {
2086                 sock->sendcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
2087                 if (sock->sendcmsgbuf == NULL) {
2088                         result = ISC_R_NOMEMORY;
2089                         goto error;
2090                 }
2091         }
2092
2093         memset(sock->name, 0, sizeof(sock->name));
2094         sock->tag = NULL;
2095
2096         /*
2097          * set up list of readers and writers to be initially empty
2098          */
2099         ISC_LIST_INIT(sock->recv_list);
2100         ISC_LIST_INIT(sock->send_list);
2101         ISC_LIST_INIT(sock->accept_list);
2102         sock->connect_ev = NULL;
2103         sock->pending_recv = 0;
2104         sock->pending_send = 0;
2105         sock->pending_accept = 0;
2106         sock->listener = 0;
2107         sock->connected = 0;
2108         sock->connecting = 0;
2109         sock->bound = 0;
2110
2111         /*
2112          * initialize the lock
2113          */
2114         result = isc_mutex_init(&sock->lock);
2115         if (result != ISC_R_SUCCESS) {
2116                 sock->common.magic = 0;
2117                 sock->common.impmagic = 0;
2118                 goto error;
2119         }
2120
2121         /*
2122          * Initialize readable and writable events
2123          */
2124         ISC_EVENT_INIT(&sock->readable_ev, sizeof(intev_t),
2125                        ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTR,
2126                        NULL, sock, sock, NULL, NULL);
2127         ISC_EVENT_INIT(&sock->writable_ev, sizeof(intev_t),
2128                        ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTW,
2129                        NULL, sock, sock, NULL, NULL);
2130
2131         sock->common.magic = ISCAPI_SOCKET_MAGIC;
2132         sock->common.impmagic = SOCKET_MAGIC;
2133         *socketp = sock;
2134
2135         return (ISC_R_SUCCESS);
2136
2137  error:
2138         if (sock->recvcmsgbuf != NULL)
2139                 isc_mem_put(manager->mctx, sock->recvcmsgbuf,
2140                             sock->recvcmsgbuflen);
2141         if (sock->sendcmsgbuf != NULL)
2142                 isc_mem_put(manager->mctx, sock->sendcmsgbuf,
2143                             sock->sendcmsgbuflen);
2144         isc_mem_put(manager->mctx, sock, sizeof(*sock));
2145
2146         return (result);
2147 }
2148
2149 /*
2150  * This event requires that the various lists be empty, that the reference
2151  * count be 1, and that the magic number is valid.  The other socket bits,
2152  * like the lock, must be initialized as well.  The fd associated must be
2153  * marked as closed, by setting it to -1 on close, or this routine will
2154  * also close the socket.
2155  */
2156 static void
2157 free_socket(isc__socket_t **socketp) {
2158         isc__socket_t *sock = *socketp;
2159
2160         INSIST(sock->references == 0);
2161         INSIST(VALID_SOCKET(sock));
2162         INSIST(!sock->connecting);
2163         INSIST(!sock->pending_recv);
2164         INSIST(!sock->pending_send);
2165         INSIST(!sock->pending_accept);
2166         INSIST(ISC_LIST_EMPTY(sock->recv_list));
2167         INSIST(ISC_LIST_EMPTY(sock->send_list));
2168         INSIST(ISC_LIST_EMPTY(sock->accept_list));
2169         INSIST(!ISC_LINK_LINKED(sock, link));
2170
2171         if (sock->recvcmsgbuf != NULL)
2172                 isc_mem_put(sock->manager->mctx, sock->recvcmsgbuf,
2173                             sock->recvcmsgbuflen);
2174         if (sock->sendcmsgbuf != NULL)
2175                 isc_mem_put(sock->manager->mctx, sock->sendcmsgbuf,
2176                             sock->sendcmsgbuflen);
2177
2178         sock->common.magic = 0;
2179         sock->common.impmagic = 0;
2180
2181         DESTROYLOCK(&sock->lock);
2182
2183         isc_mem_put(sock->manager->mctx, sock, sizeof(*sock));
2184
2185         *socketp = NULL;
2186 }
2187
2188 #ifdef SO_BSDCOMPAT
2189 /*
2190  * This really should not be necessary to do.  Having to workout
2191  * which kernel version we are on at run time so that we don't cause
2192  * the kernel to issue a warning about us using a deprecated socket option.
2193  * Such warnings should *never* be on by default in production kernels.
2194  *
2195  * We can't do this a build time because executables are moved between
2196  * machines and hence kernels.
2197  *
2198  * We can't just not set SO_BSDCOMAT because some kernels require it.
2199  */
2200
2201 static isc_once_t         bsdcompat_once = ISC_ONCE_INIT;
2202 isc_boolean_t bsdcompat = ISC_TRUE;
2203
2204 static void
2205 clear_bsdcompat(void) {
2206 #ifdef __linux__
2207          struct utsname buf;
2208          char *endp;
2209          long int major;
2210          long int minor;
2211
2212          uname(&buf);    /* Can only fail if buf is bad in Linux. */
2213
2214          /* Paranoia in parsing can be increased, but we trust uname(). */
2215          major = strtol(buf.release, &endp, 10);
2216          if (*endp == '.') {
2217                 minor = strtol(endp+1, &endp, 10);
2218                 if ((major > 2) || ((major == 2) && (minor >= 4))) {
2219                         bsdcompat = ISC_FALSE;
2220                 }
2221          }
2222 #endif /* __linux __ */
2223 }
2224 #endif
2225
2226 static isc_result_t
2227 opensocket(isc__socketmgr_t *manager, isc__socket_t *sock) {
2228         isc_result_t result;
2229         char strbuf[ISC_STRERRORSIZE];
2230         const char *err = "socket";
2231         int tries = 0;
2232 #if defined(USE_CMSG) || defined(SO_BSDCOMPAT)
2233         int on = 1;
2234 #endif
2235 #if defined(SO_RCVBUF)
2236         ISC_SOCKADDR_LEN_T optlen;
2237         int size;
2238 #endif
2239
2240  again:
2241         switch (sock->type) {
2242         case isc_sockettype_udp:
2243                 sock->fd = socket(sock->pf, SOCK_DGRAM, IPPROTO_UDP);
2244                 break;
2245         case isc_sockettype_tcp:
2246                 sock->fd = socket(sock->pf, SOCK_STREAM, IPPROTO_TCP);
2247                 break;
2248         case isc_sockettype_unix:
2249                 sock->fd = socket(sock->pf, SOCK_STREAM, 0);
2250                 break;
2251         case isc_sockettype_fdwatch:
2252                 /*
2253                  * We should not be called for isc_sockettype_fdwatch sockets.
2254                  */
2255                 INSIST(0);
2256                 break;
2257         }
2258         if (sock->fd == -1 && errno == EINTR && tries++ < 42)
2259                 goto again;
2260
2261 #ifdef F_DUPFD
2262         /*
2263          * Leave a space for stdio and TCP to work in.
2264          */
2265         if (manager->reserved != 0 && sock->type == isc_sockettype_udp &&
2266             sock->fd >= 0 && sock->fd < manager->reserved) {
2267                 int new, tmp;
2268                 new = fcntl(sock->fd, F_DUPFD, manager->reserved);
2269                 tmp = errno;
2270                 (void)close(sock->fd);
2271                 errno = tmp;
2272                 sock->fd = new;
2273                 err = "isc_socket_create: fcntl/reserved";
2274         } else if (sock->fd >= 0 && sock->fd < 20) {
2275                 int new, tmp;
2276                 new = fcntl(sock->fd, F_DUPFD, 20);
2277                 tmp = errno;
2278                 (void)close(sock->fd);
2279                 errno = tmp;
2280                 sock->fd = new;
2281                 err = "isc_socket_create: fcntl";
2282         }
2283 #endif
2284
2285         if (sock->fd >= (int)manager->maxsocks) {
2286                 (void)close(sock->fd);
2287                 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2288                                ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2289                                isc_msgcat, ISC_MSGSET_SOCKET,
2290                                ISC_MSG_TOOMANYFDS,
2291                                "socket: file descriptor exceeds limit (%d/%u)",
2292                                sock->fd, manager->maxsocks);
2293                 return (ISC_R_NORESOURCES);
2294         }
2295
2296         if (sock->fd < 0) {
2297                 switch (errno) {
2298                 case EMFILE:
2299                 case ENFILE:
2300                         isc__strerror(errno, strbuf, sizeof(strbuf));
2301                         isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2302                                        ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2303                                        isc_msgcat, ISC_MSGSET_SOCKET,
2304                                        ISC_MSG_TOOMANYFDS,
2305                                        "%s: %s", err, strbuf);
2306                         /* fallthrough */
2307                 case ENOBUFS:
2308                         return (ISC_R_NORESOURCES);
2309
2310                 case EPROTONOSUPPORT:
2311                 case EPFNOSUPPORT:
2312                 case EAFNOSUPPORT:
2313                 /*
2314                  * Linux 2.2 (and maybe others) return EINVAL instead of
2315                  * EAFNOSUPPORT.
2316                  */
2317                 case EINVAL:
2318                         return (ISC_R_FAMILYNOSUPPORT);
2319
2320                 default:
2321                         isc__strerror(errno, strbuf, sizeof(strbuf));
2322                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2323                                          "%s() %s: %s", err,
2324                                          isc_msgcat_get(isc_msgcat,
2325                                                         ISC_MSGSET_GENERAL,
2326                                                         ISC_MSG_FAILED,
2327                                                         "failed"),
2328                                          strbuf);
2329                         return (ISC_R_UNEXPECTED);
2330                 }
2331         }
2332
2333         result = make_nonblock(sock->fd);
2334         if (result != ISC_R_SUCCESS) {
2335                 (void)close(sock->fd);
2336                 return (result);
2337         }
2338
2339 #ifdef SO_BSDCOMPAT
2340         RUNTIME_CHECK(isc_once_do(&bsdcompat_once,
2341                                   clear_bsdcompat) == ISC_R_SUCCESS);
2342         if (sock->type != isc_sockettype_unix && bsdcompat &&
2343             setsockopt(sock->fd, SOL_SOCKET, SO_BSDCOMPAT,
2344                        (void *)&on, sizeof(on)) < 0) {
2345                 isc__strerror(errno, strbuf, sizeof(strbuf));
2346                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2347                                  "setsockopt(%d, SO_BSDCOMPAT) %s: %s",
2348                                  sock->fd,
2349                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2350                                                 ISC_MSG_FAILED, "failed"),
2351                                  strbuf);
2352                 /* Press on... */
2353         }
2354 #endif
2355
2356 #ifdef SO_NOSIGPIPE
2357         if (setsockopt(sock->fd, SOL_SOCKET, SO_NOSIGPIPE,
2358                        (void *)&on, sizeof(on)) < 0) {
2359                 isc__strerror(errno, strbuf, sizeof(strbuf));
2360                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2361                                  "setsockopt(%d, SO_NOSIGPIPE) %s: %s",
2362                                  sock->fd,
2363                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2364                                                 ISC_MSG_FAILED, "failed"),
2365                                  strbuf);
2366                 /* Press on... */
2367         }
2368 #endif
2369
2370 #if defined(USE_CMSG) || defined(SO_RCVBUF)
2371         if (sock->type == isc_sockettype_udp) {
2372
2373 #if defined(USE_CMSG)
2374 #if defined(SO_TIMESTAMP)
2375                 if (setsockopt(sock->fd, SOL_SOCKET, SO_TIMESTAMP,
2376                                (void *)&on, sizeof(on)) < 0
2377                     && errno != ENOPROTOOPT) {
2378                         isc__strerror(errno, strbuf, sizeof(strbuf));
2379                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2380                                          "setsockopt(%d, SO_TIMESTAMP) %s: %s",
2381                                          sock->fd,
2382                                          isc_msgcat_get(isc_msgcat,
2383                                                         ISC_MSGSET_GENERAL,
2384                                                         ISC_MSG_FAILED,
2385                                                         "failed"),
2386                                          strbuf);
2387                         /* Press on... */
2388                 }
2389 #endif /* SO_TIMESTAMP */
2390
2391 #if defined(ISC_PLATFORM_HAVEIPV6)
2392                 if (sock->pf == AF_INET6 && sock->recvcmsgbuflen == 0U) {
2393                         /*
2394                          * Warn explicitly because this anomaly can be hidden
2395                          * in usual operation (and unexpectedly appear later).
2396                          */
2397                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2398                                          "No buffer available to receive "
2399                                          "IPv6 destination");
2400                 }
2401 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
2402 #ifdef IPV6_RECVPKTINFO
2403                 /* RFC 3542 */
2404                 if ((sock->pf == AF_INET6)
2405                     && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_RECVPKTINFO,
2406                                    (void *)&on, sizeof(on)) < 0)) {
2407                         isc__strerror(errno, strbuf, sizeof(strbuf));
2408                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2409                                          "setsockopt(%d, IPV6_RECVPKTINFO) "
2410                                          "%s: %s", sock->fd,
2411                                          isc_msgcat_get(isc_msgcat,
2412                                                         ISC_MSGSET_GENERAL,
2413                                                         ISC_MSG_FAILED,
2414                                                         "failed"),
2415                                          strbuf);
2416                 }
2417 #else
2418                 /* RFC 2292 */
2419                 if ((sock->pf == AF_INET6)
2420                     && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_PKTINFO,
2421                                    (void *)&on, sizeof(on)) < 0)) {
2422                         isc__strerror(errno, strbuf, sizeof(strbuf));
2423                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2424                                          "setsockopt(%d, IPV6_PKTINFO) %s: %s",
2425                                          sock->fd,
2426                                          isc_msgcat_get(isc_msgcat,
2427                                                         ISC_MSGSET_GENERAL,
2428                                                         ISC_MSG_FAILED,
2429                                                         "failed"),
2430                                          strbuf);
2431                 }
2432 #endif /* IPV6_RECVPKTINFO */
2433 #endif /* ISC_PLATFORM_HAVEIN6PKTINFO */
2434 #ifdef IPV6_USE_MIN_MTU        /* RFC 3542, not too common yet*/
2435                 /* use minimum MTU */
2436                 if (sock->pf == AF_INET6 &&
2437                     setsockopt(sock->fd, IPPROTO_IPV6, IPV6_USE_MIN_MTU,
2438                                (void *)&on, sizeof(on)) < 0) {
2439                         isc__strerror(errno, strbuf, sizeof(strbuf));
2440                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2441                                          "setsockopt(%d, IPV6_USE_MIN_MTU) "
2442                                          "%s: %s", sock->fd,
2443                                          isc_msgcat_get(isc_msgcat,
2444                                                         ISC_MSGSET_GENERAL,
2445                                                         ISC_MSG_FAILED,
2446                                                         "failed"),
2447                                          strbuf);
2448                 }
2449 #endif
2450 #if defined(IPV6_MTU)
2451                 /*
2452                  * Use minimum MTU on IPv6 sockets.
2453                  */
2454                 if (sock->pf == AF_INET6) {
2455                         int mtu = 1280;
2456                         (void)setsockopt(sock->fd, IPPROTO_IPV6, IPV6_MTU,
2457                                          &mtu, sizeof(mtu));
2458                 }
2459 #endif
2460 #if defined(IPV6_MTU_DISCOVER) && defined(IPV6_PMTUDISC_DONT)
2461                 /*
2462                  * Turn off Path MTU discovery on IPv6/UDP sockets.
2463                  */
2464                 if (sock->pf == AF_INET6) {
2465                         int action = IPV6_PMTUDISC_DONT;
2466                         (void)setsockopt(sock->fd, IPPROTO_IPV6,
2467                                          IPV6_MTU_DISCOVER, &action,
2468                                          sizeof(action));
2469                 }
2470 #endif
2471 #endif /* ISC_PLATFORM_HAVEIPV6 */
2472 #endif /* defined(USE_CMSG) */
2473
2474 #if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
2475                 /*
2476                  * Turn off Path MTU discovery on IPv4/UDP sockets.
2477                  */
2478                 if (sock->pf == AF_INET) {
2479                         int action = IP_PMTUDISC_DONT;
2480                         (void)setsockopt(sock->fd, IPPROTO_IP, IP_MTU_DISCOVER,
2481                                          &action, sizeof(action));
2482                 }
2483 #endif
2484 #if defined(IP_DONTFRAG)
2485                 /*
2486                  * Turn off Path MTU discovery on IPv4/UDP sockets.
2487                  */
2488                 if (sock->pf == AF_INET) {
2489                         int off = 0;
2490                         (void)setsockopt(sock->fd, IPPROTO_IP, IP_DONTFRAG,
2491                                          &off, sizeof(off));
2492                 }
2493 #endif
2494
2495 #if defined(SO_RCVBUF)
2496                 optlen = sizeof(size);
2497                 if (getsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
2498                                (void *)&size, &optlen) >= 0 &&
2499                      size < RCVBUFSIZE) {
2500                         size = RCVBUFSIZE;
2501                         if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
2502                                        (void *)&size, sizeof(size)) == -1) {
2503                                 isc__strerror(errno, strbuf, sizeof(strbuf));
2504                                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2505                                         "setsockopt(%d, SO_RCVBUF, %d) %s: %s",
2506                                         sock->fd, size,
2507                                         isc_msgcat_get(isc_msgcat,
2508                                                        ISC_MSGSET_GENERAL,
2509                                                        ISC_MSG_FAILED,
2510                                                        "failed"),
2511                                         strbuf);
2512                         }
2513                 }
2514 #endif
2515         }
2516 #endif /* defined(USE_CMSG) || defined(SO_RCVBUF) */
2517
2518         inc_stats(manager->stats, sock->statsindex[STATID_OPEN]);
2519
2520         return (ISC_R_SUCCESS);
2521 }
2522
2523 /*%
2524  * Create a new 'type' socket managed by 'manager'.  Events
2525  * will be posted to 'task' and when dispatched 'action' will be
2526  * called with 'arg' as the arg value.  The new socket is returned
2527  * in 'socketp'.
2528  */
2529 ISC_SOCKETFUNC_SCOPE isc_result_t
2530 isc__socket_create(isc_socketmgr_t *manager0, int pf, isc_sockettype_t type,
2531                    isc_socket_t **socketp)
2532 {
2533         isc__socket_t *sock = NULL;
2534         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
2535         isc_result_t result;
2536         int lockid;
2537
2538         REQUIRE(VALID_MANAGER(manager));
2539         REQUIRE(socketp != NULL && *socketp == NULL);
2540         REQUIRE(type != isc_sockettype_fdwatch);
2541
2542         result = allocate_socket(manager, type, &sock);
2543         if (result != ISC_R_SUCCESS)
2544                 return (result);
2545
2546         switch (sock->type) {
2547         case isc_sockettype_udp:
2548                 sock->statsindex =
2549                         (pf == AF_INET) ? upd4statsindex : upd6statsindex;
2550                 break;
2551         case isc_sockettype_tcp:
2552                 sock->statsindex =
2553                         (pf == AF_INET) ? tcp4statsindex : tcp6statsindex;
2554                 break;
2555         case isc_sockettype_unix:
2556                 sock->statsindex = unixstatsindex;
2557                 break;
2558         default:
2559                 INSIST(0);
2560         }
2561
2562         sock->pf = pf;
2563         result = opensocket(manager, sock);
2564         if (result != ISC_R_SUCCESS) {
2565                 inc_stats(manager->stats, sock->statsindex[STATID_OPENFAIL]);
2566                 free_socket(&sock);
2567                 return (result);
2568         }
2569
2570         sock->common.methods = (isc_socketmethods_t *)&socketmethods;
2571         sock->references = 1;
2572         *socketp = (isc_socket_t *)sock;
2573
2574         /*
2575          * Note we don't have to lock the socket like we normally would because
2576          * there are no external references to it yet.
2577          */
2578
2579         lockid = FDLOCK_ID(sock->fd);
2580         LOCK(&manager->fdlock[lockid]);
2581         manager->fds[sock->fd] = sock;
2582         manager->fdstate[sock->fd] = MANAGED;
2583 #ifdef USE_DEVPOLL
2584         INSIST(sock->manager->fdpollinfo[sock->fd].want_read == 0 &&
2585                sock->manager->fdpollinfo[sock->fd].want_write == 0);
2586 #endif
2587         UNLOCK(&manager->fdlock[lockid]);
2588
2589         LOCK(&manager->lock);
2590         ISC_LIST_APPEND(manager->socklist, sock, link);
2591 #ifdef USE_SELECT
2592         if (manager->maxfd < sock->fd)
2593                 manager->maxfd = sock->fd;
2594 #endif
2595         UNLOCK(&manager->lock);
2596
2597         socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
2598                    ISC_MSG_CREATED, "created");
2599
2600         return (ISC_R_SUCCESS);
2601 }
2602
2603 #ifdef BIND9
2604 ISC_SOCKETFUNC_SCOPE isc_result_t
2605 isc__socket_open(isc_socket_t *sock0) {
2606         isc_result_t result;
2607         isc__socket_t *sock = (isc__socket_t *)sock0;
2608
2609         REQUIRE(VALID_SOCKET(sock));
2610
2611         LOCK(&sock->lock);
2612         REQUIRE(sock->references == 1);
2613         REQUIRE(sock->type != isc_sockettype_fdwatch);
2614         UNLOCK(&sock->lock);
2615         /*
2616          * We don't need to retain the lock hereafter, since no one else has
2617          * this socket.
2618          */
2619         REQUIRE(sock->fd == -1);
2620
2621         result = opensocket(sock->manager, sock);
2622         if (result != ISC_R_SUCCESS)
2623                 sock->fd = -1;
2624
2625         if (result == ISC_R_SUCCESS) {
2626                 int lockid = FDLOCK_ID(sock->fd);
2627
2628                 LOCK(&sock->manager->fdlock[lockid]);
2629                 sock->manager->fds[sock->fd] = sock;
2630                 sock->manager->fdstate[sock->fd] = MANAGED;
2631 #ifdef USE_DEVPOLL
2632                 INSIST(sock->manager->fdpollinfo[sock->fd].want_read == 0 &&
2633                        sock->manager->fdpollinfo[sock->fd].want_write == 0);
2634 #endif
2635                 UNLOCK(&sock->manager->fdlock[lockid]);
2636
2637 #ifdef USE_SELECT
2638                 LOCK(&sock->manager->lock);
2639                 if (sock->manager->maxfd < sock->fd)
2640                         sock->manager->maxfd = sock->fd;
2641                 UNLOCK(&sock->manager->lock);
2642 #endif
2643         }
2644
2645         return (result);
2646 }
2647 #endif  /* BIND9 */
2648
2649 /*
2650  * Create a new 'type' socket managed by 'manager'.  Events
2651  * will be posted to 'task' and when dispatched 'action' will be
2652  * called with 'arg' as the arg value.  The new socket is returned
2653  * in 'socketp'.
2654  */
2655 ISC_SOCKETFUNC_SCOPE isc_result_t
2656 isc__socket_fdwatchcreate(isc_socketmgr_t *manager0, int fd, int flags,
2657                           isc_sockfdwatch_t callback, void *cbarg,
2658                           isc_task_t *task, isc_socket_t **socketp)
2659 {
2660         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
2661         isc__socket_t *sock = NULL;
2662         isc_result_t result;
2663         int lockid;
2664
2665         REQUIRE(VALID_MANAGER(manager));
2666         REQUIRE(socketp != NULL && *socketp == NULL);
2667
2668         result = allocate_socket(manager, isc_sockettype_fdwatch, &sock);
2669         if (result != ISC_R_SUCCESS)
2670                 return (result);
2671
2672         sock->fd = fd;
2673         sock->fdwatcharg = cbarg;
2674         sock->fdwatchcb = callback;
2675         sock->fdwatchflags = flags;
2676         sock->fdwatchtask = task;
2677         sock->statsindex = fdwatchstatsindex;
2678
2679         sock->common.methods = (isc_socketmethods_t *)&socketmethods;
2680         sock->references = 1;
2681         *socketp = (isc_socket_t *)sock;
2682
2683         /*
2684          * Note we don't have to lock the socket like we normally would because
2685          * there are no external references to it yet.
2686          */
2687
2688         lockid = FDLOCK_ID(sock->fd);
2689         LOCK(&manager->fdlock[lockid]);
2690         manager->fds[sock->fd] = sock;
2691         manager->fdstate[sock->fd] = MANAGED;
2692         UNLOCK(&manager->fdlock[lockid]);
2693
2694         LOCK(&manager->lock);
2695         ISC_LIST_APPEND(manager->socklist, sock, link);
2696 #ifdef USE_SELECT
2697         if (manager->maxfd < sock->fd)
2698                 manager->maxfd = sock->fd;
2699 #endif
2700         UNLOCK(&manager->lock);
2701
2702         if (flags & ISC_SOCKFDWATCH_READ)
2703                 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
2704         if (flags & ISC_SOCKFDWATCH_WRITE)
2705                 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
2706
2707         socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
2708                    ISC_MSG_CREATED, "fdwatch-created");
2709
2710         return (ISC_R_SUCCESS);
2711 }
2712
2713 /*
2714  * Indicate to the manager that it should watch the socket again.
2715  * This can be used to restart watching if the previous event handler
2716  * didn't indicate there was more data to be processed.  Primarily
2717  * it is for writing but could be used for reading if desired
2718  */
2719
2720 ISC_SOCKETFUNC_SCOPE isc_result_t
2721 isc__socket_fdwatchpoke(isc_socket_t *sock0, int flags)
2722 {
2723         isc__socket_t *sock = (isc__socket_t *)sock0;
2724
2725         REQUIRE(VALID_SOCKET(sock));
2726
2727         /*
2728          * We check both flags first to allow us to get the lock
2729          * once but only if we need it.
2730          */
2731
2732         if ((flags & (ISC_SOCKFDWATCH_READ | ISC_SOCKFDWATCH_WRITE)) != 0) {
2733                 LOCK(&sock->lock);
2734                 if (((flags & ISC_SOCKFDWATCH_READ) != 0) &&
2735                     !sock->pending_recv)
2736                         select_poke(sock->manager, sock->fd,
2737                                     SELECT_POKE_READ);
2738                 if (((flags & ISC_SOCKFDWATCH_WRITE) != 0) &&
2739                     !sock->pending_send)
2740                         select_poke(sock->manager, sock->fd,
2741                                     SELECT_POKE_WRITE);
2742                 UNLOCK(&sock->lock);
2743         }
2744
2745         socket_log(sock, NULL, TRACE, isc_msgcat, ISC_MSGSET_SOCKET,
2746                    ISC_MSG_POKED, "fdwatch-poked flags: %d", flags);
2747
2748         return (ISC_R_SUCCESS);
2749 }
2750
2751 /*
2752  * Attach to a socket.  Caller must explicitly detach when it is done.
2753  */
2754 ISC_SOCKETFUNC_SCOPE void
2755 isc__socket_attach(isc_socket_t *sock0, isc_socket_t **socketp) {
2756         isc__socket_t *sock = (isc__socket_t *)sock0;
2757
2758         REQUIRE(VALID_SOCKET(sock));
2759         REQUIRE(socketp != NULL && *socketp == NULL);
2760
2761         LOCK(&sock->lock);
2762         sock->references++;
2763         UNLOCK(&sock->lock);
2764
2765         *socketp = (isc_socket_t *)sock;
2766 }
2767
2768 /*
2769  * Dereference a socket.  If this is the last reference to it, clean things
2770  * up by destroying the socket.
2771  */
2772 ISC_SOCKETFUNC_SCOPE void
2773 isc__socket_detach(isc_socket_t **socketp) {
2774         isc__socket_t *sock;
2775         isc_boolean_t kill_socket = ISC_FALSE;
2776
2777         REQUIRE(socketp != NULL);
2778         sock = (isc__socket_t *)*socketp;
2779         REQUIRE(VALID_SOCKET(sock));
2780
2781         LOCK(&sock->lock);
2782         REQUIRE(sock->references > 0);
2783         sock->references--;
2784         if (sock->references == 0)
2785                 kill_socket = ISC_TRUE;
2786         UNLOCK(&sock->lock);
2787
2788         if (kill_socket)
2789                 destroy(&sock);
2790
2791         *socketp = NULL;
2792 }
2793
2794 #ifdef BIND9
2795 ISC_SOCKETFUNC_SCOPE isc_result_t
2796 isc__socket_close(isc_socket_t *sock0) {
2797         isc__socket_t *sock = (isc__socket_t *)sock0;
2798         int fd;
2799         isc__socketmgr_t *manager;
2800
2801         REQUIRE(VALID_SOCKET(sock));
2802
2803         LOCK(&sock->lock);
2804
2805         REQUIRE(sock->references == 1);
2806         REQUIRE(sock->type != isc_sockettype_fdwatch);
2807         REQUIRE(sock->fd >= 0 && sock->fd < (int)sock->manager->maxsocks);
2808
2809         INSIST(!sock->connecting);
2810         INSIST(!sock->pending_recv);
2811         INSIST(!sock->pending_send);
2812         INSIST(!sock->pending_accept);
2813         INSIST(ISC_LIST_EMPTY(sock->recv_list));
2814         INSIST(ISC_LIST_EMPTY(sock->send_list));
2815         INSIST(ISC_LIST_EMPTY(sock->accept_list));
2816         INSIST(sock->connect_ev == NULL);
2817
2818         manager = sock->manager;
2819         fd = sock->fd;
2820         sock->fd = -1;
2821         memset(sock->name, 0, sizeof(sock->name));
2822         sock->tag = NULL;
2823         sock->listener = 0;
2824         sock->connected = 0;
2825         sock->connecting = 0;
2826         sock->bound = 0;
2827         isc_sockaddr_any(&sock->peer_address);
2828
2829         UNLOCK(&sock->lock);
2830
2831         closesocket(manager, sock, fd);
2832
2833         return (ISC_R_SUCCESS);
2834 }
2835 #endif  /* BIND9 */
2836
2837 /*
2838  * I/O is possible on a given socket.  Schedule an event to this task that
2839  * will call an internal function to do the I/O.  This will charge the
2840  * task with the I/O operation and let our select loop handler get back
2841  * to doing something real as fast as possible.
2842  *
2843  * The socket and manager must be locked before calling this function.
2844  */
2845 static void
2846 dispatch_recv(isc__socket_t *sock) {
2847         intev_t *iev;
2848         isc_socketevent_t *ev;
2849         isc_task_t *sender;
2850
2851         INSIST(!sock->pending_recv);
2852
2853         if (sock->type != isc_sockettype_fdwatch) {
2854                 ev = ISC_LIST_HEAD(sock->recv_list);
2855                 if (ev == NULL)
2856                         return;
2857                 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2858                            "dispatch_recv:  event %p -> task %p",
2859                            ev, ev->ev_sender);
2860                 sender = ev->ev_sender;
2861         } else {
2862                 sender = sock->fdwatchtask;
2863         }
2864
2865         sock->pending_recv = 1;
2866         iev = &sock->readable_ev;
2867
2868         sock->references++;
2869         iev->ev_sender = sock;
2870         if (sock->type == isc_sockettype_fdwatch)
2871                 iev->ev_action = internal_fdwatch_read;
2872         else
2873                 iev->ev_action = internal_recv;
2874         iev->ev_arg = sock;
2875
2876         isc_task_send(sender, (isc_event_t **)&iev);
2877 }
2878
2879 static void
2880 dispatch_send(isc__socket_t *sock) {
2881         intev_t *iev;
2882         isc_socketevent_t *ev;
2883         isc_task_t *sender;
2884
2885         INSIST(!sock->pending_send);
2886
2887         if (sock->type != isc_sockettype_fdwatch) {
2888                 ev = ISC_LIST_HEAD(sock->send_list);
2889                 if (ev == NULL)
2890                         return;
2891                 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2892                            "dispatch_send:  event %p -> task %p",
2893                            ev, ev->ev_sender);
2894                 sender = ev->ev_sender;
2895         } else {
2896                 sender = sock->fdwatchtask;
2897         }
2898
2899         sock->pending_send = 1;
2900         iev = &sock->writable_ev;
2901
2902         sock->references++;
2903         iev->ev_sender = sock;
2904         if (sock->type == isc_sockettype_fdwatch)
2905                 iev->ev_action = internal_fdwatch_write;
2906         else
2907                 iev->ev_action = internal_send;
2908         iev->ev_arg = sock;
2909
2910         isc_task_send(sender, (isc_event_t **)&iev);
2911 }
2912
2913 /*
2914  * Dispatch an internal accept event.
2915  */
2916 static void
2917 dispatch_accept(isc__socket_t *sock) {
2918         intev_t *iev;
2919         isc_socket_newconnev_t *ev;
2920
2921         INSIST(!sock->pending_accept);
2922
2923         /*
2924          * Are there any done events left, or were they all canceled
2925          * before the manager got the socket lock?
2926          */
2927         ev = ISC_LIST_HEAD(sock->accept_list);
2928         if (ev == NULL)
2929                 return;
2930
2931         sock->pending_accept = 1;
2932         iev = &sock->readable_ev;
2933
2934         sock->references++;  /* keep socket around for this internal event */
2935         iev->ev_sender = sock;
2936         iev->ev_action = internal_accept;
2937         iev->ev_arg = sock;
2938
2939         isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
2940 }
2941
2942 static void
2943 dispatch_connect(isc__socket_t *sock) {
2944         intev_t *iev;
2945         isc_socket_connev_t *ev;
2946
2947         iev = &sock->writable_ev;
2948
2949         ev = sock->connect_ev;
2950         INSIST(ev != NULL); /* XXX */
2951
2952         INSIST(sock->connecting);
2953
2954         sock->references++;  /* keep socket around for this internal event */
2955         iev->ev_sender = sock;
2956         iev->ev_action = internal_connect;
2957         iev->ev_arg = sock;
2958
2959         isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
2960 }
2961
2962 /*
2963  * Dequeue an item off the given socket's read queue, set the result code
2964  * in the done event to the one provided, and send it to the task it was
2965  * destined for.
2966  *
2967  * If the event to be sent is on a list, remove it before sending.  If
2968  * asked to, send and detach from the socket as well.
2969  *
2970  * Caller must have the socket locked if the event is attached to the socket.
2971  */
2972 static void
2973 send_recvdone_event(isc__socket_t *sock, isc_socketevent_t **dev) {
2974         isc_task_t *task;
2975
2976         task = (*dev)->ev_sender;
2977
2978         (*dev)->ev_sender = sock;
2979
2980         if (ISC_LINK_LINKED(*dev, ev_link))
2981                 ISC_LIST_DEQUEUE(sock->recv_list, *dev, ev_link);
2982
2983         if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
2984             == ISC_SOCKEVENTATTR_ATTACHED)
2985                 isc_task_sendanddetach(&task, (isc_event_t **)dev);
2986         else
2987                 isc_task_send(task, (isc_event_t **)dev);
2988 }
2989
2990 /*
2991  * See comments for send_recvdone_event() above.
2992  *
2993  * Caller must have the socket locked if the event is attached to the socket.
2994  */
2995 static void
2996 send_senddone_event(isc__socket_t *sock, isc_socketevent_t **dev) {
2997         isc_task_t *task;
2998
2999         INSIST(dev != NULL && *dev != NULL);
3000
3001         task = (*dev)->ev_sender;
3002         (*dev)->ev_sender = sock;
3003
3004         if (ISC_LINK_LINKED(*dev, ev_link))
3005                 ISC_LIST_DEQUEUE(sock->send_list, *dev, ev_link);
3006
3007         if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
3008             == ISC_SOCKEVENTATTR_ATTACHED)
3009                 isc_task_sendanddetach(&task, (isc_event_t **)dev);
3010         else
3011                 isc_task_send(task, (isc_event_t **)dev);
3012 }
3013
3014 /*
3015  * Call accept() on a socket, to get the new file descriptor.  The listen
3016  * socket is used as a prototype to create a new isc_socket_t.  The new
3017  * socket has one outstanding reference.  The task receiving the event
3018  * will be detached from just after the event is delivered.
3019  *
3020  * On entry to this function, the event delivered is the internal
3021  * readable event, and the first item on the accept_list should be
3022  * the done event we want to send.  If the list is empty, this is a no-op,
3023  * so just unlock and return.
3024  */
3025 static void
3026 internal_accept(isc_task_t *me, isc_event_t *ev) {
3027         isc__socket_t *sock;
3028         isc__socketmgr_t *manager;
3029         isc_socket_newconnev_t *dev;
3030         isc_task_t *task;
3031         ISC_SOCKADDR_LEN_T addrlen;
3032         int fd;
3033         isc_result_t result = ISC_R_SUCCESS;
3034         char strbuf[ISC_STRERRORSIZE];
3035         const char *err = "accept";
3036
3037         UNUSED(me);
3038
3039         sock = ev->ev_sender;
3040         INSIST(VALID_SOCKET(sock));
3041
3042         LOCK(&sock->lock);
3043         socket_log(sock, NULL, TRACE,
3044                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTLOCK,
3045                    "internal_accept called, locked socket");
3046
3047         manager = sock->manager;
3048         INSIST(VALID_MANAGER(manager));
3049
3050         INSIST(sock->listener);
3051         INSIST(sock->pending_accept == 1);
3052         sock->pending_accept = 0;
3053
3054         INSIST(sock->references > 0);
3055         sock->references--;  /* the internal event is done with this socket */
3056         if (sock->references == 0) {
3057                 UNLOCK(&sock->lock);
3058                 destroy(&sock);
3059                 return;
3060         }
3061
3062         /*
3063          * Get the first item off the accept list.
3064          * If it is empty, unlock the socket and return.
3065          */
3066         dev = ISC_LIST_HEAD(sock->accept_list);
3067         if (dev == NULL) {
3068                 UNLOCK(&sock->lock);
3069                 return;
3070         }
3071
3072         /*
3073          * Try to accept the new connection.  If the accept fails with
3074          * EAGAIN or EINTR, simply poke the watcher to watch this socket
3075          * again.  Also ignore ECONNRESET, which has been reported to
3076          * be spuriously returned on Linux 2.2.19 although it is not
3077          * a documented error for accept().  ECONNABORTED has been
3078          * reported for Solaris 8.  The rest are thrown in not because
3079          * we have seen them but because they are ignored by other
3080          * daemons such as BIND 8 and Apache.
3081          */
3082
3083         addrlen = sizeof(NEWCONNSOCK(dev)->peer_address.type);
3084         memset(&NEWCONNSOCK(dev)->peer_address.type, 0, addrlen);
3085         fd = accept(sock->fd, &NEWCONNSOCK(dev)->peer_address.type.sa,
3086                     (void *)&addrlen);
3087
3088 #ifdef F_DUPFD
3089         /*
3090          * Leave a space for stdio to work in.
3091          */
3092         if (fd >= 0 && fd < 20) {
3093                 int new, tmp;
3094                 new = fcntl(fd, F_DUPFD, 20);
3095                 tmp = errno;
3096                 (void)close(fd);
3097                 errno = tmp;
3098                 fd = new;
3099                 err = "accept/fcntl";
3100         }
3101 #endif
3102
3103         if (fd < 0) {
3104                 if (SOFT_ERROR(errno))
3105                         goto soft_error;
3106                 switch (errno) {
3107                 case ENFILE:
3108                 case EMFILE:
3109                         isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
3110                                        ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
3111                                        isc_msgcat, ISC_MSGSET_SOCKET,
3112                                        ISC_MSG_TOOMANYFDS,
3113                                        "%s: too many open file descriptors",
3114                                        err);
3115                         goto soft_error;
3116
3117                 case ENOBUFS:
3118                 case ENOMEM:
3119                 case ECONNRESET:
3120                 case ECONNABORTED:
3121                 case EHOSTUNREACH:
3122                 case EHOSTDOWN:
3123                 case ENETUNREACH:
3124                 case ENETDOWN:
3125                 case ECONNREFUSED:
3126 #ifdef EPROTO
3127                 case EPROTO:
3128 #endif
3129 #ifdef ENONET
3130                 case ENONET:
3131 #endif
3132                         goto soft_error;
3133                 default:
3134                         break;
3135                 }
3136                 isc__strerror(errno, strbuf, sizeof(strbuf));
3137                 UNEXPECTED_ERROR(__FILE__, __LINE__,
3138                                  "internal_accept: %s() %s: %s", err,
3139                                  isc_msgcat_get(isc_msgcat,
3140                                                 ISC_MSGSET_GENERAL,
3141                                                 ISC_MSG_FAILED,
3142                                                 "failed"),
3143                                  strbuf);
3144                 fd = -1;
3145                 result = ISC_R_UNEXPECTED;
3146         } else {
3147                 if (addrlen == 0U) {
3148                         UNEXPECTED_ERROR(__FILE__, __LINE__,
3149                                          "internal_accept(): "
3150                                          "accept() failed to return "
3151                                          "remote address");
3152
3153                         (void)close(fd);
3154                         goto soft_error;
3155                 } else if (NEWCONNSOCK(dev)->peer_address.type.sa.sa_family !=
3156                            sock->pf)
3157                 {
3158                         UNEXPECTED_ERROR(__FILE__, __LINE__,
3159                                          "internal_accept(): "
3160                                          "accept() returned peer address "
3161                                          "family %u (expected %u)",
3162                                          NEWCONNSOCK(dev)->peer_address.
3163                                          type.sa.sa_family,
3164                                          sock->pf);
3165                         (void)close(fd);
3166                         goto soft_error;
3167                 } else if (fd >= (int)manager->maxsocks) {
3168                         isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
3169                                        ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
3170                                        isc_msgcat, ISC_MSGSET_SOCKET,
3171                                        ISC_MSG_TOOMANYFDS,
3172                                        "accept: "
3173                                        "file descriptor exceeds limit (%d/%u)",
3174                                        fd, manager->maxsocks);
3175                         (void)close(fd);
3176                         goto soft_error;
3177                 }
3178         }
3179
3180         if (fd != -1) {
3181                 NEWCONNSOCK(dev)->peer_address.length = addrlen;
3182                 NEWCONNSOCK(dev)->pf = sock->pf;
3183         }
3184
3185         /*
3186          * Pull off the done event.
3187          */
3188         ISC_LIST_UNLINK(sock->accept_list, dev, ev_link);
3189
3190         /*
3191          * Poke watcher if there are more pending accepts.
3192          */
3193         if (!ISC_LIST_EMPTY(sock->accept_list))
3194                 select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
3195
3196         UNLOCK(&sock->lock);
3197
3198         if (fd != -1) {
3199                 result = make_nonblock(fd);
3200                 if (result != ISC_R_SUCCESS) {
3201                         (void)close(fd);
3202                         fd = -1;
3203                 }
3204         }
3205
3206         /*
3207          * -1 means the new socket didn't happen.
3208          */
3209         if (fd != -1) {
3210                 int lockid = FDLOCK_ID(fd);
3211
3212                 LOCK(&manager->fdlock[lockid]);
3213                 manager->fds[fd] = NEWCONNSOCK(dev);
3214                 manager->fdstate[fd] = MANAGED;
3215                 UNLOCK(&manager->fdlock[lockid]);
3216
3217                 LOCK(&manager->lock);
3218                 ISC_LIST_APPEND(manager->socklist, NEWCONNSOCK(dev), link);
3219
3220                 NEWCONNSOCK(dev)->fd = fd;
3221                 NEWCONNSOCK(dev)->bound = 1;
3222                 NEWCONNSOCK(dev)->connected = 1;
3223
3224                 /*
3225                  * Save away the remote address
3226                  */
3227                 dev->address = NEWCONNSOCK(dev)->peer_address;
3228
3229 #ifdef USE_SELECT
3230                 if (manager->maxfd < fd)
3231                         manager->maxfd = fd;
3232 #endif
3233
3234                 socket_log(sock, &NEWCONNSOCK(dev)->peer_address, CREATION,
3235                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTEDCXN,
3236                            "accepted connection, new socket %p",
3237                            dev->newsocket);
3238
3239                 UNLOCK(&manager->lock);
3240
3241                 inc_stats(manager->stats, sock->statsindex[STATID_ACCEPT]);
3242         } else {
3243                 inc_stats(manager->stats, sock->statsindex[STATID_ACCEPTFAIL]);
3244                 NEWCONNSOCK(dev)->references--;
3245                 free_socket((isc__socket_t **)&dev->newsocket);
3246         }
3247
3248         /*
3249          * Fill in the done event details and send it off.
3250          */
3251         dev->result = result;
3252         task = dev->ev_sender;
3253         dev->ev_sender = sock;
3254
3255         isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
3256         return;
3257
3258  soft_error:
3259         select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
3260         UNLOCK(&sock->lock);
3261
3262         inc_stats(manager->stats, sock->statsindex[STATID_ACCEPTFAIL]);
3263         return;
3264 }
3265
3266 static void
3267 internal_recv(isc_task_t *me, isc_event_t *ev) {
3268         isc_socketevent_t *dev;
3269         isc__socket_t *sock;
3270
3271         INSIST(ev->ev_type == ISC_SOCKEVENT_INTR);
3272
3273         sock = ev->ev_sender;
3274         INSIST(VALID_SOCKET(sock));
3275
3276         LOCK(&sock->lock);
3277         socket_log(sock, NULL, IOEVENT,
3278                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALRECV,
3279                    "internal_recv: task %p got event %p", me, ev);
3280
3281         INSIST(sock->pending_recv == 1);
3282         sock->pending_recv = 0;
3283
3284         INSIST(sock->references > 0);
3285         sock->references--;  /* the internal event is done with this socket */
3286         if (sock->references == 0) {
3287                 UNLOCK(&sock->lock);
3288                 destroy(&sock);
3289                 return;
3290         }
3291
3292         /*
3293          * Try to do as much I/O as possible on this socket.  There are no
3294          * limits here, currently.
3295          */
3296         dev = ISC_LIST_HEAD(sock->recv_list);
3297         while (dev != NULL) {
3298                 switch (doio_recv(sock, dev)) {
3299                 case DOIO_SOFT:
3300                         goto poke;
3301
3302                 case DOIO_EOF:
3303                         /*
3304                          * read of 0 means the remote end was closed.
3305                          * Run through the event queue and dispatch all
3306                          * the events with an EOF result code.
3307                          */
3308                         do {
3309                                 dev->result = ISC_R_EOF;
3310                                 send_recvdone_event(sock, &dev);
3311                                 dev = ISC_LIST_HEAD(sock->recv_list);
3312                         } while (dev != NULL);
3313                         goto poke;
3314
3315                 case DOIO_SUCCESS:
3316                 case DOIO_HARD:
3317                         send_recvdone_event(sock, &dev);
3318                         break;
3319                 }
3320
3321                 dev = ISC_LIST_HEAD(sock->recv_list);
3322         }
3323
3324  poke:
3325         if (!ISC_LIST_EMPTY(sock->recv_list))
3326                 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
3327
3328         UNLOCK(&sock->lock);
3329 }
3330
3331 static void
3332 internal_send(isc_task_t *me, isc_event_t *ev) {
3333         isc_socketevent_t *dev;
3334         isc__socket_t *sock;
3335
3336         INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
3337
3338         /*
3339          * Find out what socket this is and lock it.
3340          */
3341         sock = (isc__socket_t *)ev->ev_sender;
3342         INSIST(VALID_SOCKET(sock));
3343
3344         LOCK(&sock->lock);
3345         socket_log(sock, NULL, IOEVENT,
3346                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALSEND,
3347                    "internal_send: task %p got event %p", me, ev);
3348
3349         INSIST(sock->pending_send == 1);
3350         sock->pending_send = 0;
3351
3352         INSIST(sock->references > 0);
3353         sock->references--;  /* the internal event is done with this socket */
3354         if (sock->references == 0) {
3355                 UNLOCK(&sock->lock);
3356                 destroy(&sock);
3357                 return;
3358         }
3359
3360         /*
3361          * Try to do as much I/O as possible on this socket.  There are no
3362          * limits here, currently.
3363          */
3364         dev = ISC_LIST_HEAD(sock->send_list);
3365         while (dev != NULL) {
3366                 switch (doio_send(sock, dev)) {
3367                 case DOIO_SOFT:
3368                         goto poke;
3369
3370                 case DOIO_HARD:
3371                 case DOIO_SUCCESS:
3372                         send_senddone_event(sock, &dev);
3373                         break;
3374                 }
3375
3376                 dev = ISC_LIST_HEAD(sock->send_list);
3377         }
3378
3379  poke:
3380         if (!ISC_LIST_EMPTY(sock->send_list))
3381                 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
3382
3383         UNLOCK(&sock->lock);
3384 }
3385
3386 static void
3387 internal_fdwatch_write(isc_task_t *me, isc_event_t *ev) {
3388         isc__socket_t *sock;
3389         int more_data;
3390
3391         INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
3392
3393         /*
3394          * Find out what socket this is and lock it.
3395          */
3396         sock = (isc__socket_t *)ev->ev_sender;
3397         INSIST(VALID_SOCKET(sock));
3398
3399         LOCK(&sock->lock);
3400         socket_log(sock, NULL, IOEVENT,
3401                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALSEND,
3402                    "internal_fdwatch_write: task %p got event %p", me, ev);
3403
3404         INSIST(sock->pending_send == 1);
3405
3406         UNLOCK(&sock->lock);
3407         more_data = (sock->fdwatchcb)(me, (isc_socket_t *)sock,
3408                                       sock->fdwatcharg, ISC_SOCKFDWATCH_WRITE);
3409         LOCK(&sock->lock);
3410
3411         sock->pending_send = 0;
3412
3413         INSIST(sock->references > 0);
3414         sock->references--;  /* the internal event is done with this socket */
3415         if (sock->references == 0) {
3416                 UNLOCK(&sock->lock);
3417                 destroy(&sock);
3418                 return;
3419         }
3420
3421         if (more_data)
3422                 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
3423
3424         UNLOCK(&sock->lock);
3425 }
3426
3427 static void
3428 internal_fdwatch_read(isc_task_t *me, isc_event_t *ev) {
3429         isc__socket_t *sock;
3430         int more_data;
3431
3432         INSIST(ev->ev_type == ISC_SOCKEVENT_INTR);
3433
3434         /*
3435          * Find out what socket this is and lock it.
3436          */
3437         sock = (isc__socket_t *)ev->ev_sender;
3438         INSIST(VALID_SOCKET(sock));
3439
3440         LOCK(&sock->lock);
3441         socket_log(sock, NULL, IOEVENT,
3442                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALRECV,
3443                    "internal_fdwatch_read: task %p got event %p", me, ev);
3444
3445         INSIST(sock->pending_recv == 1);
3446
3447         UNLOCK(&sock->lock);
3448         more_data = (sock->fdwatchcb)(me, (isc_socket_t *)sock,
3449                                       sock->fdwatcharg, ISC_SOCKFDWATCH_READ);
3450         LOCK(&sock->lock);
3451
3452         sock->pending_recv = 0;
3453
3454         INSIST(sock->references > 0);
3455         sock->references--;  /* the internal event is done with this socket */
3456         if (sock->references == 0) {
3457                 UNLOCK(&sock->lock);
3458                 destroy(&sock);
3459                 return;
3460         }
3461
3462         if (more_data)
3463                 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
3464
3465         UNLOCK(&sock->lock);
3466 }
3467
3468 /*
3469  * Process read/writes on each fd here.  Avoid locking
3470  * and unlocking twice if both reads and writes are possible.
3471  */
3472 static void
3473 process_fd(isc__socketmgr_t *manager, int fd, isc_boolean_t readable,
3474            isc_boolean_t writeable)
3475 {
3476         isc__socket_t *sock;
3477         isc_boolean_t unlock_sock;
3478         isc_boolean_t unwatch_read = ISC_FALSE, unwatch_write = ISC_FALSE;
3479         int lockid = FDLOCK_ID(fd);
3480
3481         /*
3482          * If the socket is going to be closed, don't do more I/O.
3483          */
3484         LOCK(&manager->fdlock[lockid]);
3485         if (manager->fdstate[fd] == CLOSE_PENDING) {
3486                 UNLOCK(&manager->fdlock[lockid]);
3487
3488                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
3489                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
3490                 return;
3491         }
3492
3493         sock = manager->fds[fd];
3494         unlock_sock = ISC_FALSE;
3495         if (readable) {
3496                 if (sock == NULL) {
3497                         unwatch_read = ISC_TRUE;
3498                         goto check_write;
3499                 }
3500                 unlock_sock = ISC_TRUE;
3501                 LOCK(&sock->lock);
3502                 if (!SOCK_DEAD(sock)) {
3503                         if (sock->listener)
3504                                 dispatch_accept(sock);
3505                         else
3506                                 dispatch_recv(sock);
3507                 }
3508                 unwatch_read = ISC_TRUE;
3509         }
3510 check_write:
3511         if (writeable) {
3512                 if (sock == NULL) {
3513                         unwatch_write = ISC_TRUE;
3514                         goto unlock_fd;
3515                 }
3516                 if (!unlock_sock) {
3517                         unlock_sock = ISC_TRUE;
3518                         LOCK(&sock->lock);
3519                 }
3520                 if (!SOCK_DEAD(sock)) {
3521                         if (sock->connecting)
3522                                 dispatch_connect(sock);
3523                         else
3524                                 dispatch_send(sock);
3525                 }
3526                 unwatch_write = ISC_TRUE;
3527         }
3528         if (unlock_sock)
3529                 UNLOCK(&sock->lock);
3530
3531  unlock_fd:
3532         UNLOCK(&manager->fdlock[lockid]);
3533         if (unwatch_read)
3534                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
3535         if (unwatch_write)
3536                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
3537
3538 }
3539
3540 #ifdef USE_KQUEUE
3541 static isc_boolean_t
3542 process_fds(isc__socketmgr_t *manager, struct kevent *events, int nevents) {
3543         int i;
3544         isc_boolean_t readable, writable;
3545         isc_boolean_t done = ISC_FALSE;
3546 #ifdef USE_WATCHER_THREAD
3547         isc_boolean_t have_ctlevent = ISC_FALSE;
3548 #endif
3549
3550         if (nevents == manager->nevents) {
3551                 /*
3552                  * This is not an error, but something unexpected.  If this
3553                  * happens, it may indicate the need for increasing
3554                  * ISC_SOCKET_MAXEVENTS.
3555                  */
3556                 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3557                             ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3558                             "maximum number of FD events (%d) received",
3559                             nevents);
3560         }
3561
3562         for (i = 0; i < nevents; i++) {
3563                 REQUIRE(events[i].ident < manager->maxsocks);
3564 #ifdef USE_WATCHER_THREAD
3565                 if (events[i].ident == (uintptr_t)manager->pipe_fds[0]) {
3566                         have_ctlevent = ISC_TRUE;
3567                         continue;
3568                 }
3569 #endif
3570                 readable = ISC_TF(events[i].filter == EVFILT_READ);
3571                 writable = ISC_TF(events[i].filter == EVFILT_WRITE);
3572                 process_fd(manager, events[i].ident, readable, writable);
3573         }
3574
3575 #ifdef USE_WATCHER_THREAD
3576         if (have_ctlevent)
3577                 done = process_ctlfd(manager);
3578 #endif
3579
3580         return (done);
3581 }
3582 #elif defined(USE_EPOLL)
3583 static isc_boolean_t
3584 process_fds(isc__socketmgr_t *manager, struct epoll_event *events, int nevents)
3585 {
3586         int i;
3587         isc_boolean_t done = ISC_FALSE;
3588 #ifdef USE_WATCHER_THREAD
3589         isc_boolean_t have_ctlevent = ISC_FALSE;
3590 #endif
3591
3592         if (nevents == manager->nevents) {
3593                 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3594                             ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3595                             "maximum number of FD events (%d) received",
3596                             nevents);
3597         }
3598
3599         for (i = 0; i < nevents; i++) {
3600                 REQUIRE(events[i].data.fd < (int)manager->maxsocks);
3601 #ifdef USE_WATCHER_THREAD
3602                 if (events[i].data.fd == manager->pipe_fds[0]) {
3603                         have_ctlevent = ISC_TRUE;
3604                         continue;
3605                 }
3606 #endif
3607                 if ((events[i].events & EPOLLERR) != 0 ||
3608                     (events[i].events & EPOLLHUP) != 0) {
3609                         /*
3610                          * epoll does not set IN/OUT bits on an erroneous
3611                          * condition, so we need to try both anyway.  This is a
3612                          * bit inefficient, but should be okay for such rare
3613                          * events.  Note also that the read or write attempt
3614                          * won't block because we use non-blocking sockets.
3615                          */
3616                         events[i].events |= (EPOLLIN | EPOLLOUT);
3617                 }
3618                 process_fd(manager, events[i].data.fd,
3619                            (events[i].events & EPOLLIN) != 0,
3620                            (events[i].events & EPOLLOUT) != 0);
3621         }
3622
3623 #ifdef USE_WATCHER_THREAD
3624         if (have_ctlevent)
3625                 done = process_ctlfd(manager);
3626 #endif
3627
3628         return (done);
3629 }
3630 #elif defined(USE_DEVPOLL)
3631 static isc_boolean_t
3632 process_fds(isc__socketmgr_t *manager, struct pollfd *events, int nevents) {
3633         int i;
3634         isc_boolean_t done = ISC_FALSE;
3635 #ifdef USE_WATCHER_THREAD
3636         isc_boolean_t have_ctlevent = ISC_FALSE;
3637 #endif
3638
3639         if (nevents == manager->nevents) {
3640                 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3641                             ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3642                             "maximum number of FD events (%d) received",
3643                             nevents);
3644         }
3645
3646         for (i = 0; i < nevents; i++) {
3647                 REQUIRE(events[i].fd < (int)manager->maxsocks);
3648 #ifdef USE_WATCHER_THREAD
3649                 if (events[i].fd == manager->pipe_fds[0]) {
3650                         have_ctlevent = ISC_TRUE;
3651                         continue;
3652                 }
3653 #endif
3654                 process_fd(manager, events[i].fd,
3655                            (events[i].events & POLLIN) != 0,
3656                            (events[i].events & POLLOUT) != 0);
3657         }
3658
3659 #ifdef USE_WATCHER_THREAD
3660         if (have_ctlevent)
3661                 done = process_ctlfd(manager);
3662 #endif
3663
3664         return (done);
3665 }
3666 #elif defined(USE_SELECT)
3667 static void
3668 process_fds(isc__socketmgr_t *manager, int maxfd, fd_set *readfds,
3669             fd_set *writefds)
3670 {
3671         int i;
3672
3673         REQUIRE(maxfd <= (int)manager->maxsocks);
3674
3675         for (i = 0; i < maxfd; i++) {
3676 #ifdef USE_WATCHER_THREAD
3677                 if (i == manager->pipe_fds[0] || i == manager->pipe_fds[1])
3678                         continue;
3679 #endif /* USE_WATCHER_THREAD */
3680                 process_fd(manager, i, FD_ISSET(i, readfds),
3681                            FD_ISSET(i, writefds));
3682         }
3683 }
3684 #endif
3685
3686 #ifdef USE_WATCHER_THREAD
3687 static isc_boolean_t
3688 process_ctlfd(isc__socketmgr_t *manager) {
3689         int msg, fd;
3690
3691         for (;;) {
3692                 select_readmsg(manager, &fd, &msg);
3693
3694                 manager_log(manager, IOEVENT,
3695                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
3696                                            ISC_MSG_WATCHERMSG,
3697                                            "watcher got message %d "
3698                                            "for socket %d"), msg, fd);
3699
3700                 /*
3701                  * Nothing to read?
3702                  */
3703                 if (msg == SELECT_POKE_NOTHING)
3704                         break;
3705
3706                 /*
3707                  * Handle shutdown message.  We really should
3708                  * jump out of this loop right away, but
3709                  * it doesn't matter if we have to do a little
3710                  * more work first.
3711                  */
3712                 if (msg == SELECT_POKE_SHUTDOWN)
3713                         return (ISC_TRUE);
3714
3715                 /*
3716                  * This is a wakeup on a socket.  Look
3717                  * at the event queue for both read and write,
3718                  * and decide if we need to watch on it now
3719                  * or not.
3720                  */
3721                 wakeup_socket(manager, fd, msg);
3722         }
3723
3724         return (ISC_FALSE);
3725 }
3726
3727 /*
3728  * This is the thread that will loop forever, always in a select or poll
3729  * call.
3730  *
3731  * When select returns something to do, track down what thread gets to do
3732  * this I/O and post the event to it.
3733  */
3734 static isc_threadresult_t
3735 watcher(void *uap) {
3736         isc__socketmgr_t *manager = uap;
3737         isc_boolean_t done;
3738         int cc;
3739 #ifdef USE_KQUEUE
3740         const char *fnname = "kevent()";
3741 #elif defined (USE_EPOLL)
3742         const char *fnname = "epoll_wait()";
3743 #elif defined(USE_DEVPOLL)
3744         const char *fnname = "ioctl(DP_POLL)";
3745         struct dvpoll dvp;
3746 #elif defined (USE_SELECT)
3747         const char *fnname = "select()";
3748         int maxfd;
3749         int ctlfd;
3750 #endif
3751         char strbuf[ISC_STRERRORSIZE];
3752 #ifdef ISC_SOCKET_USE_POLLWATCH
3753         pollstate_t pollstate = poll_idle;
3754 #endif
3755
3756 #if defined (USE_SELECT)
3757         /*
3758          * Get the control fd here.  This will never change.
3759          */
3760         ctlfd = manager->pipe_fds[0];
3761 #endif
3762         done = ISC_FALSE;
3763         while (!done) {
3764                 do {
3765 #ifdef USE_KQUEUE
3766                         cc = kevent(manager->kqueue_fd, NULL, 0,
3767                                     manager->events, manager->nevents, NULL);
3768 #elif defined(USE_EPOLL)
3769                         cc = epoll_wait(manager->epoll_fd, manager->events,
3770                                         manager->nevents, -1);
3771 #elif defined(USE_DEVPOLL)
3772                         dvp.dp_fds = manager->events;
3773                         dvp.dp_nfds = manager->nevents;
3774 #ifndef ISC_SOCKET_USE_POLLWATCH
3775                         dvp.dp_timeout = -1;
3776 #else
3777                         if (pollstate == poll_idle)
3778                                 dvp.dp_timeout = -1;
3779                         else
3780                                 dvp.dp_timeout = ISC_SOCKET_POLLWATCH_TIMEOUT;
3781 #endif  /* ISC_SOCKET_USE_POLLWATCH */
3782                         cc = ioctl(manager->devpoll_fd, DP_POLL, &dvp);
3783 #elif defined(USE_SELECT)
3784                         LOCK(&manager->lock);
3785                         memcpy(manager->read_fds_copy, manager->read_fds,
3786                                manager->fd_bufsize);
3787                         memcpy(manager->write_fds_copy, manager->write_fds,
3788                                manager->fd_bufsize);
3789                         maxfd = manager->maxfd + 1;
3790                         UNLOCK(&manager->lock);
3791
3792                         cc = select(maxfd, manager->read_fds_copy,
3793                                     manager->write_fds_copy, NULL, NULL);
3794 #endif  /* USE_KQUEUE */
3795
3796                         if (cc < 0 && !SOFT_ERROR(errno)) {
3797                                 isc__strerror(errno, strbuf, sizeof(strbuf));
3798                                 FATAL_ERROR(__FILE__, __LINE__,
3799                                             "%s %s: %s", fnname,
3800                                             isc_msgcat_get(isc_msgcat,
3801                                                            ISC_MSGSET_GENERAL,
3802                                                            ISC_MSG_FAILED,
3803                                                            "failed"), strbuf);
3804                         }
3805
3806 #if defined(USE_DEVPOLL) && defined(ISC_SOCKET_USE_POLLWATCH)
3807                         if (cc == 0) {
3808                                 if (pollstate == poll_active)
3809                                         pollstate = poll_checking;
3810                                 else if (pollstate == poll_checking)
3811                                         pollstate = poll_idle;
3812                         } else if (cc > 0) {
3813                                 if (pollstate == poll_checking) {
3814                                         /*
3815                                          * XXX: We'd like to use a more
3816                                          * verbose log level as it's actually an
3817                                          * unexpected event, but the kernel bug
3818                                          * reportedly happens pretty frequently
3819                                          * (and it can also be a false positive)
3820                                          * so it would be just too noisy.
3821                                          */
3822                                         manager_log(manager,
3823                                                     ISC_LOGCATEGORY_GENERAL,
3824                                                     ISC_LOGMODULE_SOCKET,
3825                                                     ISC_LOG_DEBUG(1),
3826                                                     "unexpected POLL timeout");
3827                                 }
3828                                 pollstate = poll_active;
3829                         }
3830 #endif
3831                 } while (cc < 0);
3832
3833 #if defined(USE_KQUEUE) || defined (USE_EPOLL) || defined (USE_DEVPOLL)
3834                 done = process_fds(manager, manager->events, cc);
3835 #elif defined(USE_SELECT)
3836                 process_fds(manager, maxfd, manager->read_fds_copy,
3837                             manager->write_fds_copy);
3838
3839                 /*
3840                  * Process reads on internal, control fd.
3841                  */
3842                 if (FD_ISSET(ctlfd, manager->read_fds_copy))
3843                         done = process_ctlfd(manager);
3844 #endif
3845         }
3846
3847         manager_log(manager, TRACE, "%s",
3848                     isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3849                                    ISC_MSG_EXITING, "watcher exiting"));
3850
3851         return ((isc_threadresult_t)0);
3852 }
3853 #endif /* USE_WATCHER_THREAD */
3854
3855 #ifdef BIND9
3856 ISC_SOCKETFUNC_SCOPE void
3857 isc__socketmgr_setreserved(isc_socketmgr_t *manager0, isc_uint32_t reserved) {
3858         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
3859
3860         REQUIRE(VALID_MANAGER(manager));
3861
3862         manager->reserved = reserved;
3863 }
3864
3865 ISC_SOCKETFUNC_SCOPE void
3866 isc___socketmgr_maxudp(isc_socketmgr_t *manager0, int maxudp) {
3867         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
3868
3869         REQUIRE(VALID_MANAGER(manager));
3870
3871         manager->maxudp = maxudp;
3872 }
3873 #endif  /* BIND9 */
3874
3875 /*
3876  * Create a new socket manager.
3877  */
3878
3879 static isc_result_t
3880 setup_watcher(isc_mem_t *mctx, isc__socketmgr_t *manager) {
3881         isc_result_t result;
3882 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
3883         char strbuf[ISC_STRERRORSIZE];
3884 #endif
3885
3886 #ifdef USE_KQUEUE
3887         manager->nevents = ISC_SOCKET_MAXEVENTS;
3888         manager->events = isc_mem_get(mctx, sizeof(struct kevent) *
3889                                       manager->nevents);
3890         if (manager->events == NULL)
3891                 return (ISC_R_NOMEMORY);
3892         manager->kqueue_fd = kqueue();
3893         if (manager->kqueue_fd == -1) {
3894                 result = isc__errno2result(errno);
3895                 isc__strerror(errno, strbuf, sizeof(strbuf));
3896                 UNEXPECTED_ERROR(__FILE__, __LINE__,
3897                                  "kqueue %s: %s",
3898                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3899                                                 ISC_MSG_FAILED, "failed"),
3900                                  strbuf);
3901                 isc_mem_put(mctx, manager->events,
3902                             sizeof(struct kevent) * manager->nevents);
3903                 return (result);
3904         }
3905
3906 #ifdef USE_WATCHER_THREAD
3907         result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3908         if (result != ISC_R_SUCCESS) {
3909                 close(manager->kqueue_fd);
3910                 isc_mem_put(mctx, manager->events,
3911                             sizeof(struct kevent) * manager->nevents);
3912                 return (result);
3913         }
3914 #endif  /* USE_WATCHER_THREAD */
3915 #elif defined(USE_EPOLL)
3916         manager->nevents = ISC_SOCKET_MAXEVENTS;
3917         manager->events = isc_mem_get(mctx, sizeof(struct epoll_event) *
3918                                       manager->nevents);
3919         if (manager->events == NULL)
3920                 return (ISC_R_NOMEMORY);
3921         manager->epoll_fd = epoll_create(manager->nevents);
3922         if (manager->epoll_fd == -1) {
3923                 result = isc__errno2result(errno);
3924                 isc__strerror(errno, strbuf, sizeof(strbuf));
3925                 UNEXPECTED_ERROR(__FILE__, __LINE__,
3926                                  "epoll_create %s: %s",
3927                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3928                                                 ISC_MSG_FAILED, "failed"),
3929                                  strbuf);
3930                 isc_mem_put(mctx, manager->events,
3931                             sizeof(struct epoll_event) * manager->nevents);
3932                 return (result);
3933         }
3934 #ifdef USE_WATCHER_THREAD
3935         result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3936         if (result != ISC_R_SUCCESS) {
3937                 close(manager->epoll_fd);
3938                 isc_mem_put(mctx, manager->events,
3939                             sizeof(struct epoll_event) * manager->nevents);
3940                 return (result);
3941         }
3942 #endif  /* USE_WATCHER_THREAD */
3943 #elif defined(USE_DEVPOLL)
3944         /*
3945          * XXXJT: /dev/poll seems to reject large numbers of events,
3946          * so we should be careful about redefining ISC_SOCKET_MAXEVENTS.
3947          */
3948         manager->nevents = ISC_SOCKET_MAXEVENTS;
3949         manager->events = isc_mem_get(mctx, sizeof(struct pollfd) *
3950                                       manager->nevents);
3951         if (manager->events == NULL)
3952                 return (ISC_R_NOMEMORY);
3953         /*
3954          * Note: fdpollinfo should be able to support all possible FDs, so
3955          * it must have maxsocks entries (not nevents).
3956          */
3957         manager->fdpollinfo = isc_mem_get(mctx, sizeof(pollinfo_t) *
3958                                           manager->maxsocks);
3959         if (manager->fdpollinfo == NULL) {
3960                 isc_mem_put(mctx, manager->events,
3961                             sizeof(struct pollfd) * manager->nevents);
3962                 return (ISC_R_NOMEMORY);
3963         }
3964         memset(manager->fdpollinfo, 0, sizeof(pollinfo_t) * manager->maxsocks);
3965         manager->devpoll_fd = open("/dev/poll", O_RDWR);
3966         if (manager->devpoll_fd == -1) {
3967                 result = isc__errno2result(errno);
3968                 isc__strerror(errno, strbuf, sizeof(strbuf));
3969                 UNEXPECTED_ERROR(__FILE__, __LINE__,
3970                                  "open(/dev/poll) %s: %s",
3971                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3972                                                 ISC_MSG_FAILED, "failed"),
3973                                  strbuf);
3974                 isc_mem_put(mctx, manager->events,
3975                             sizeof(struct pollfd) * manager->nevents);
3976                 isc_mem_put(mctx, manager->fdpollinfo,
3977                             sizeof(pollinfo_t) * manager->maxsocks);
3978                 return (result);
3979         }
3980 #ifdef USE_WATCHER_THREAD
3981         result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3982         if (result != ISC_R_SUCCESS) {
3983                 close(manager->devpoll_fd);
3984                 isc_mem_put(mctx, manager->events,
3985                             sizeof(struct pollfd) * manager->nevents);
3986                 isc_mem_put(mctx, manager->fdpollinfo,
3987                             sizeof(pollinfo_t) * manager->maxsocks);
3988                 return (result);
3989         }
3990 #endif  /* USE_WATCHER_THREAD */
3991 #elif defined(USE_SELECT)
3992         UNUSED(result);
3993
3994 #if ISC_SOCKET_MAXSOCKETS > FD_SETSIZE
3995         /*
3996          * Note: this code should also cover the case of MAXSOCKETS <=
3997          * FD_SETSIZE, but we separate the cases to avoid possible portability
3998          * issues regarding howmany() and the actual representation of fd_set.
3999          */
4000         manager->fd_bufsize = howmany(manager->maxsocks, NFDBITS) *
4001                 sizeof(fd_mask);
4002 #else
4003         manager->fd_bufsize = sizeof(fd_set);
4004 #endif
4005
4006         manager->read_fds = NULL;
4007         manager->read_fds_copy = NULL;
4008         manager->write_fds = NULL;
4009         manager->write_fds_copy = NULL;
4010
4011         manager->read_fds = isc_mem_get(mctx, manager->fd_bufsize);
4012         if (manager->read_fds != NULL)
4013                 manager->read_fds_copy = isc_mem_get(mctx, manager->fd_bufsize);
4014         if (manager->read_fds_copy != NULL)
4015                 manager->write_fds = isc_mem_get(mctx, manager->fd_bufsize);
4016         if (manager->write_fds != NULL) {
4017                 manager->write_fds_copy = isc_mem_get(mctx,
4018                                                       manager->fd_bufsize);
4019         }
4020         if (manager->write_fds_copy == NULL) {
4021                 if (manager->write_fds != NULL) {
4022                         isc_mem_put(mctx, manager->write_fds,
4023                                     manager->fd_bufsize);
4024                 }
4025                 if (manager->read_fds_copy != NULL) {
4026                         isc_mem_put(mctx, manager->read_fds_copy,
4027                                     manager->fd_bufsize);
4028                 }
4029                 if (manager->read_fds != NULL) {
4030                         isc_mem_put(mctx, manager->read_fds,
4031                                     manager->fd_bufsize);
4032                 }
4033                 return (ISC_R_NOMEMORY);
4034         }
4035         memset(manager->read_fds, 0, manager->fd_bufsize);
4036         memset(manager->write_fds, 0, manager->fd_bufsize);
4037
4038 #ifdef USE_WATCHER_THREAD
4039         (void)watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
4040         manager->maxfd = manager->pipe_fds[0];
4041 #else /* USE_WATCHER_THREAD */
4042         manager->maxfd = 0;
4043 #endif /* USE_WATCHER_THREAD */
4044 #endif  /* USE_KQUEUE */
4045
4046         return (ISC_R_SUCCESS);
4047 }
4048
4049 static void
4050 cleanup_watcher(isc_mem_t *mctx, isc__socketmgr_t *manager) {
4051 #ifdef USE_WATCHER_THREAD
4052         isc_result_t result;
4053
4054         result = unwatch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
4055         if (result != ISC_R_SUCCESS) {
4056                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4057                                  "epoll_ctl(DEL) %s",
4058                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4059                                                 ISC_MSG_FAILED, "failed"));
4060         }
4061 #endif  /* USE_WATCHER_THREAD */
4062
4063 #ifdef USE_KQUEUE
4064         close(manager->kqueue_fd);
4065         isc_mem_put(mctx, manager->events,
4066                     sizeof(struct kevent) * manager->nevents);
4067 #elif defined(USE_EPOLL)
4068         close(manager->epoll_fd);
4069         isc_mem_put(mctx, manager->events,
4070                     sizeof(struct epoll_event) * manager->nevents);
4071 #elif defined(USE_DEVPOLL)
4072         close(manager->devpoll_fd);
4073         isc_mem_put(mctx, manager->events,
4074                     sizeof(struct pollfd) * manager->nevents);
4075         isc_mem_put(mctx, manager->fdpollinfo,
4076                     sizeof(pollinfo_t) * manager->maxsocks);
4077 #elif defined(USE_SELECT)
4078         if (manager->read_fds != NULL)
4079                 isc_mem_put(mctx, manager->read_fds, manager->fd_bufsize);
4080         if (manager->read_fds_copy != NULL)
4081                 isc_mem_put(mctx, manager->read_fds_copy, manager->fd_bufsize);
4082         if (manager->write_fds != NULL)
4083                 isc_mem_put(mctx, manager->write_fds, manager->fd_bufsize);
4084         if (manager->write_fds_copy != NULL)
4085                 isc_mem_put(mctx, manager->write_fds_copy, manager->fd_bufsize);
4086 #endif  /* USE_KQUEUE */
4087 }
4088
4089 ISC_SOCKETFUNC_SCOPE isc_result_t
4090 isc__socketmgr_create(isc_mem_t *mctx, isc_socketmgr_t **managerp) {
4091         return (isc__socketmgr_create2(mctx, managerp, 0));
4092 }
4093
4094 ISC_SOCKETFUNC_SCOPE isc_result_t
4095 isc__socketmgr_create2(isc_mem_t *mctx, isc_socketmgr_t **managerp,
4096                        unsigned int maxsocks)
4097 {
4098         int i;
4099         isc__socketmgr_t *manager;
4100 #ifdef USE_WATCHER_THREAD
4101         char strbuf[ISC_STRERRORSIZE];
4102 #endif
4103         isc_result_t result;
4104
4105         REQUIRE(managerp != NULL && *managerp == NULL);
4106
4107 #ifdef USE_SHARED_MANAGER
4108         if (socketmgr != NULL) {
4109                 /* Don't allow maxsocks to be updated */
4110                 if (maxsocks > 0 && socketmgr->maxsocks != maxsocks)
4111                         return (ISC_R_EXISTS);
4112
4113                 socketmgr->refs++;
4114                 *managerp = (isc_socketmgr_t *)socketmgr;
4115                 return (ISC_R_SUCCESS);
4116         }
4117 #endif /* USE_SHARED_MANAGER */
4118
4119         if (maxsocks == 0)
4120                 maxsocks = ISC_SOCKET_MAXSOCKETS;
4121
4122         manager = isc_mem_get(mctx, sizeof(*manager));
4123         if (manager == NULL)
4124                 return (ISC_R_NOMEMORY);
4125
4126         /* zero-clear so that necessary cleanup on failure will be easy */
4127         memset(manager, 0, sizeof(*manager));
4128         manager->maxsocks = maxsocks;
4129         manager->reserved = 0;
4130         manager->maxudp = 0;
4131         manager->fds = isc_mem_get(mctx,
4132                                    manager->maxsocks * sizeof(isc__socket_t *));
4133         if (manager->fds == NULL) {
4134                 result = ISC_R_NOMEMORY;
4135                 goto free_manager;
4136         }
4137         manager->fdstate = isc_mem_get(mctx, manager->maxsocks * sizeof(int));
4138         if (manager->fdstate == NULL) {
4139                 result = ISC_R_NOMEMORY;
4140                 goto free_manager;
4141         }
4142         manager->stats = NULL;
4143
4144         manager->common.methods = &socketmgrmethods;
4145         manager->common.magic = ISCAPI_SOCKETMGR_MAGIC;
4146         manager->common.impmagic = SOCKET_MANAGER_MAGIC;
4147         manager->mctx = NULL;
4148         memset(manager->fds, 0, manager->maxsocks * sizeof(isc_socket_t *));
4149         ISC_LIST_INIT(manager->socklist);
4150         result = isc_mutex_init(&manager->lock);
4151         if (result != ISC_R_SUCCESS)
4152                 goto free_manager;
4153         manager->fdlock = isc_mem_get(mctx, FDLOCK_COUNT * sizeof(isc_mutex_t));
4154         if (manager->fdlock == NULL) {
4155                 result = ISC_R_NOMEMORY;
4156                 goto cleanup_lock;
4157         }
4158         for (i = 0; i < FDLOCK_COUNT; i++) {
4159                 result = isc_mutex_init(&manager->fdlock[i]);
4160                 if (result != ISC_R_SUCCESS) {
4161                         while (--i >= 0)
4162                                 DESTROYLOCK(&manager->fdlock[i]);
4163                         isc_mem_put(mctx, manager->fdlock,
4164                                     FDLOCK_COUNT * sizeof(isc_mutex_t));
4165                         manager->fdlock = NULL;
4166                         goto cleanup_lock;
4167                 }
4168         }
4169
4170 #ifdef USE_WATCHER_THREAD
4171         if (isc_condition_init(&manager->shutdown_ok) != ISC_R_SUCCESS) {
4172                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4173                                  "isc_condition_init() %s",
4174                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4175                                                 ISC_MSG_FAILED, "failed"));
4176                 result = ISC_R_UNEXPECTED;
4177                 goto cleanup_lock;
4178         }
4179
4180         /*
4181          * Create the special fds that will be used to wake up the
4182          * select/poll loop when something internal needs to be done.
4183          */
4184         if (pipe(manager->pipe_fds) != 0) {
4185                 isc__strerror(errno, strbuf, sizeof(strbuf));
4186                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4187                                  "pipe() %s: %s",
4188                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4189                                                 ISC_MSG_FAILED, "failed"),
4190                                  strbuf);
4191                 result = ISC_R_UNEXPECTED;
4192                 goto cleanup_condition;
4193         }
4194
4195         RUNTIME_CHECK(make_nonblock(manager->pipe_fds[0]) == ISC_R_SUCCESS);
4196 #if 0
4197         RUNTIME_CHECK(make_nonblock(manager->pipe_fds[1]) == ISC_R_SUCCESS);
4198 #endif
4199 #endif  /* USE_WATCHER_THREAD */
4200
4201 #ifdef USE_SHARED_MANAGER
4202         manager->refs = 1;
4203 #endif /* USE_SHARED_MANAGER */
4204
4205         /*
4206          * Set up initial state for the select loop
4207          */
4208         result = setup_watcher(mctx, manager);
4209         if (result != ISC_R_SUCCESS)
4210                 goto cleanup;
4211         memset(manager->fdstate, 0, manager->maxsocks * sizeof(int));
4212 #ifdef USE_WATCHER_THREAD
4213         /*
4214          * Start up the select/poll thread.
4215          */
4216         if (isc_thread_create(watcher, manager, &manager->watcher) !=
4217             ISC_R_SUCCESS) {
4218                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4219                                  "isc_thread_create() %s",
4220                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4221                                                 ISC_MSG_FAILED, "failed"));
4222                 cleanup_watcher(mctx, manager);
4223                 result = ISC_R_UNEXPECTED;
4224                 goto cleanup;
4225         }
4226 #endif /* USE_WATCHER_THREAD */
4227         isc_mem_attach(mctx, &manager->mctx);
4228
4229 #ifdef USE_SHARED_MANAGER
4230         socketmgr = manager;
4231 #endif /* USE_SHARED_MANAGER */
4232         *managerp = (isc_socketmgr_t *)manager;
4233
4234         return (ISC_R_SUCCESS);
4235
4236 cleanup:
4237 #ifdef USE_WATCHER_THREAD
4238         (void)close(manager->pipe_fds[0]);
4239         (void)close(manager->pipe_fds[1]);
4240 #endif  /* USE_WATCHER_THREAD */
4241
4242 #ifdef USE_WATCHER_THREAD
4243 cleanup_condition:
4244         (void)isc_condition_destroy(&manager->shutdown_ok);
4245 #endif  /* USE_WATCHER_THREAD */
4246
4247
4248 cleanup_lock:
4249         if (manager->fdlock != NULL) {
4250                 for (i = 0; i < FDLOCK_COUNT; i++)
4251                         DESTROYLOCK(&manager->fdlock[i]);
4252         }
4253         DESTROYLOCK(&manager->lock);
4254
4255 free_manager:
4256         if (manager->fdlock != NULL) {
4257                 isc_mem_put(mctx, manager->fdlock,
4258                             FDLOCK_COUNT * sizeof(isc_mutex_t));
4259         }
4260         if (manager->fdstate != NULL) {
4261                 isc_mem_put(mctx, manager->fdstate,
4262                             manager->maxsocks * sizeof(int));
4263         }
4264         if (manager->fds != NULL) {
4265                 isc_mem_put(mctx, manager->fds,
4266                             manager->maxsocks * sizeof(isc_socket_t *));
4267         }
4268         isc_mem_put(mctx, manager, sizeof(*manager));
4269
4270         return (result);
4271 }
4272
4273 #ifdef BIND9
4274 isc_result_t
4275 isc__socketmgr_getmaxsockets(isc_socketmgr_t *manager0, unsigned int *nsockp) {
4276         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
4277         REQUIRE(VALID_MANAGER(manager));
4278         REQUIRE(nsockp != NULL);
4279
4280         *nsockp = manager->maxsocks;
4281
4282         return (ISC_R_SUCCESS);
4283 }
4284
4285 void
4286 isc__socketmgr_setstats(isc_socketmgr_t *manager0, isc_stats_t *stats) {
4287         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
4288
4289         REQUIRE(VALID_MANAGER(manager));
4290         REQUIRE(ISC_LIST_EMPTY(manager->socklist));
4291         REQUIRE(manager->stats == NULL);
4292         REQUIRE(isc_stats_ncounters(stats) == isc_sockstatscounter_max);
4293
4294         isc_stats_attach(stats, &manager->stats);
4295 }
4296 #endif
4297
4298 ISC_SOCKETFUNC_SCOPE void
4299 isc__socketmgr_destroy(isc_socketmgr_t **managerp) {
4300         isc__socketmgr_t *manager;
4301         int i;
4302         isc_mem_t *mctx;
4303
4304         /*
4305          * Destroy a socket manager.
4306          */
4307
4308         REQUIRE(managerp != NULL);
4309         manager = (isc__socketmgr_t *)*managerp;
4310         REQUIRE(VALID_MANAGER(manager));
4311
4312 #ifdef USE_SHARED_MANAGER
4313         manager->refs--;
4314         if (manager->refs > 0) {
4315                 *managerp = NULL;
4316                 return;
4317         }
4318         socketmgr = NULL;
4319 #endif /* USE_SHARED_MANAGER */
4320
4321         LOCK(&manager->lock);
4322
4323         /*
4324          * Wait for all sockets to be destroyed.
4325          */
4326         while (!ISC_LIST_EMPTY(manager->socklist)) {
4327 #ifdef USE_WATCHER_THREAD
4328                 manager_log(manager, CREATION, "%s",
4329                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
4330                                            ISC_MSG_SOCKETSREMAIN,
4331                                            "sockets exist"));
4332                 WAIT(&manager->shutdown_ok, &manager->lock);
4333 #else /* USE_WATCHER_THREAD */
4334                 UNLOCK(&manager->lock);
4335                 isc__taskmgr_dispatch(NULL);
4336                 LOCK(&manager->lock);
4337 #endif /* USE_WATCHER_THREAD */
4338         }
4339
4340         UNLOCK(&manager->lock);
4341
4342         /*
4343          * Here, poke our select/poll thread.  Do this by closing the write
4344          * half of the pipe, which will send EOF to the read half.
4345          * This is currently a no-op in the non-threaded case.
4346          */
4347         select_poke(manager, 0, SELECT_POKE_SHUTDOWN);
4348
4349 #ifdef USE_WATCHER_THREAD
4350         /*
4351          * Wait for thread to exit.
4352          */
4353         if (isc_thread_join(manager->watcher, NULL) != ISC_R_SUCCESS)
4354                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4355                                  "isc_thread_join() %s",
4356                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4357                                                 ISC_MSG_FAILED, "failed"));
4358 #endif /* USE_WATCHER_THREAD */
4359
4360         /*
4361          * Clean up.
4362          */
4363         cleanup_watcher(manager->mctx, manager);
4364
4365 #ifdef USE_WATCHER_THREAD
4366         (void)close(manager->pipe_fds[0]);
4367         (void)close(manager->pipe_fds[1]);
4368         (void)isc_condition_destroy(&manager->shutdown_ok);
4369 #endif /* USE_WATCHER_THREAD */
4370
4371         for (i = 0; i < (int)manager->maxsocks; i++)
4372                 if (manager->fdstate[i] == CLOSE_PENDING) /* no need to lock */
4373                         (void)close(i);
4374
4375         isc_mem_put(manager->mctx, manager->fds,
4376                     manager->maxsocks * sizeof(isc__socket_t *));
4377         isc_mem_put(manager->mctx, manager->fdstate,
4378                     manager->maxsocks * sizeof(int));
4379
4380         if (manager->stats != NULL)
4381                 isc_stats_detach(&manager->stats);
4382
4383         if (manager->fdlock != NULL) {
4384                 for (i = 0; i < FDLOCK_COUNT; i++)
4385                         DESTROYLOCK(&manager->fdlock[i]);
4386                 isc_mem_put(manager->mctx, manager->fdlock,
4387                             FDLOCK_COUNT * sizeof(isc_mutex_t));
4388         }
4389         DESTROYLOCK(&manager->lock);
4390         manager->common.magic = 0;
4391         manager->common.impmagic = 0;
4392         mctx= manager->mctx;
4393         isc_mem_put(mctx, manager, sizeof(*manager));
4394
4395         isc_mem_detach(&mctx);
4396
4397         *managerp = NULL;
4398
4399 #ifdef USE_SHARED_MANAGER
4400         socketmgr = NULL;
4401 #endif
4402 }
4403
4404 static isc_result_t
4405 socket_recv(isc__socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
4406             unsigned int flags)
4407 {
4408         int io_state;
4409         isc_boolean_t have_lock = ISC_FALSE;
4410         isc_task_t *ntask = NULL;
4411         isc_result_t result = ISC_R_SUCCESS;
4412
4413         dev->ev_sender = task;
4414
4415         if (sock->type == isc_sockettype_udp) {
4416                 io_state = doio_recv(sock, dev);
4417         } else {
4418                 LOCK(&sock->lock);
4419                 have_lock = ISC_TRUE;
4420
4421                 if (ISC_LIST_EMPTY(sock->recv_list))
4422                         io_state = doio_recv(sock, dev);
4423                 else
4424                         io_state = DOIO_SOFT;
4425         }
4426
4427         switch (io_state) {
4428         case DOIO_SOFT:
4429                 /*
4430                  * We couldn't read all or part of the request right now, so
4431                  * queue it.
4432                  *
4433                  * Attach to socket and to task
4434                  */
4435                 isc_task_attach(task, &ntask);
4436                 dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
4437
4438                 if (!have_lock) {
4439                         LOCK(&sock->lock);
4440                         have_lock = ISC_TRUE;
4441                 }
4442
4443                 /*
4444                  * Enqueue the request.  If the socket was previously not being
4445                  * watched, poke the watcher to start paying attention to it.
4446                  */
4447                 if (ISC_LIST_EMPTY(sock->recv_list) && !sock->pending_recv)
4448                         select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
4449                 ISC_LIST_ENQUEUE(sock->recv_list, dev, ev_link);
4450
4451                 socket_log(sock, NULL, EVENT, NULL, 0, 0,
4452                            "socket_recv: event %p -> task %p",
4453                            dev, ntask);
4454
4455                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
4456                         result = ISC_R_INPROGRESS;
4457                 break;
4458
4459         case DOIO_EOF:
4460                 dev->result = ISC_R_EOF;
4461                 /* fallthrough */
4462
4463         case DOIO_HARD:
4464         case DOIO_SUCCESS:
4465                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
4466                         send_recvdone_event(sock, &dev);
4467                 break;
4468         }
4469
4470         if (have_lock)
4471                 UNLOCK(&sock->lock);
4472
4473         return (result);
4474 }
4475
4476 ISC_SOCKETFUNC_SCOPE isc_result_t
4477 isc__socket_recvv(isc_socket_t *sock0, isc_bufferlist_t *buflist,
4478                   unsigned int minimum, isc_task_t *task,
4479                   isc_taskaction_t action, const void *arg)
4480 {
4481         isc__socket_t *sock = (isc__socket_t *)sock0;
4482         isc_socketevent_t *dev;
4483         isc__socketmgr_t *manager;
4484         unsigned int iocount;
4485         isc_buffer_t *buffer;
4486
4487         REQUIRE(VALID_SOCKET(sock));
4488         REQUIRE(buflist != NULL);
4489         REQUIRE(!ISC_LIST_EMPTY(*buflist));
4490         REQUIRE(task != NULL);
4491         REQUIRE(action != NULL);
4492
4493         manager = sock->manager;
4494         REQUIRE(VALID_MANAGER(manager));
4495
4496         iocount = isc_bufferlist_availablecount(buflist);
4497         REQUIRE(iocount > 0);
4498
4499         INSIST(sock->bound);
4500
4501         dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
4502         if (dev == NULL)
4503                 return (ISC_R_NOMEMORY);
4504
4505         /*
4506          * UDP sockets are always partial read
4507          */
4508         if (sock->type == isc_sockettype_udp)
4509                 dev->minimum = 1;
4510         else {
4511                 if (minimum == 0)
4512                         dev->minimum = iocount;
4513                 else
4514                         dev->minimum = minimum;
4515         }
4516
4517         /*
4518          * Move each buffer from the passed in list to our internal one.
4519          */
4520         buffer = ISC_LIST_HEAD(*buflist);
4521         while (buffer != NULL) {
4522                 ISC_LIST_DEQUEUE(*buflist, buffer, link);
4523                 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
4524                 buffer = ISC_LIST_HEAD(*buflist);
4525         }
4526
4527         return (socket_recv(sock, dev, task, 0));
4528 }
4529
4530 ISC_SOCKETFUNC_SCOPE isc_result_t
4531 isc__socket_recv(isc_socket_t *sock0, isc_region_t *region,
4532                  unsigned int minimum, isc_task_t *task,
4533                  isc_taskaction_t action, const void *arg)
4534 {
4535         isc__socket_t *sock = (isc__socket_t *)sock0;
4536         isc_socketevent_t *dev;
4537         isc__socketmgr_t *manager;
4538
4539         REQUIRE(VALID_SOCKET(sock));
4540         REQUIRE(action != NULL);
4541
4542         manager = sock->manager;
4543         REQUIRE(VALID_MANAGER(manager));
4544
4545         INSIST(sock->bound);
4546
4547         dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
4548         if (dev == NULL)
4549                 return (ISC_R_NOMEMORY);
4550
4551         return (isc__socket_recv2(sock0, region, minimum, task, dev, 0));
4552 }
4553
4554 ISC_SOCKETFUNC_SCOPE isc_result_t
4555 isc__socket_recv2(isc_socket_t *sock0, isc_region_t *region,
4556                   unsigned int minimum, isc_task_t *task,
4557                   isc_socketevent_t *event, unsigned int flags)
4558 {
4559         isc__socket_t *sock = (isc__socket_t *)sock0;
4560
4561         event->ev_sender = sock;
4562         event->result = ISC_R_UNSET;
4563         ISC_LIST_INIT(event->bufferlist);
4564         event->region = *region;
4565         event->n = 0;
4566         event->offset = 0;
4567         event->attributes = 0;
4568
4569         /*
4570          * UDP sockets are always partial read.
4571          */
4572         if (sock->type == isc_sockettype_udp)
4573                 event->minimum = 1;
4574         else {
4575                 if (minimum == 0)
4576                         event->minimum = region->length;
4577                 else
4578                         event->minimum = minimum;
4579         }
4580
4581         return (socket_recv(sock, event, task, flags));
4582 }
4583
4584 static isc_result_t
4585 socket_send(isc__socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
4586             isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
4587             unsigned int flags)
4588 {
4589         int io_state;
4590         isc_boolean_t have_lock = ISC_FALSE;
4591         isc_task_t *ntask = NULL;
4592         isc_result_t result = ISC_R_SUCCESS;
4593
4594         dev->ev_sender = task;
4595
4596         set_dev_address(address, sock, dev);
4597         if (pktinfo != NULL) {
4598                 dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
4599                 dev->pktinfo = *pktinfo;
4600
4601                 if (!isc_sockaddr_issitelocal(&dev->address) &&
4602                     !isc_sockaddr_islinklocal(&dev->address)) {
4603                         socket_log(sock, NULL, TRACE, isc_msgcat,
4604                                    ISC_MSGSET_SOCKET, ISC_MSG_PKTINFOPROVIDED,
4605                                    "pktinfo structure provided, ifindex %u "
4606                                    "(set to 0)", pktinfo->ipi6_ifindex);
4607
4608                         /*
4609                          * Set the pktinfo index to 0 here, to let the
4610                          * kernel decide what interface it should send on.
4611                          */
4612                         dev->pktinfo.ipi6_ifindex = 0;
4613                 }
4614         }
4615
4616         if (sock->type == isc_sockettype_udp)
4617                 io_state = doio_send(sock, dev);
4618         else {
4619                 LOCK(&sock->lock);
4620                 have_lock = ISC_TRUE;
4621
4622                 if (ISC_LIST_EMPTY(sock->send_list))
4623                         io_state = doio_send(sock, dev);
4624                 else
4625                         io_state = DOIO_SOFT;
4626         }
4627
4628         switch (io_state) {
4629         case DOIO_SOFT:
4630                 /*
4631                  * We couldn't send all or part of the request right now, so
4632                  * queue it unless ISC_SOCKFLAG_NORETRY is set.
4633                  */
4634                 if ((flags & ISC_SOCKFLAG_NORETRY) == 0) {
4635                         isc_task_attach(task, &ntask);
4636                         dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
4637
4638                         if (!have_lock) {
4639                                 LOCK(&sock->lock);
4640                                 have_lock = ISC_TRUE;
4641                         }
4642
4643                         /*
4644                          * Enqueue the request.  If the socket was previously
4645                          * not being watched, poke the watcher to start
4646                          * paying attention to it.
4647                          */
4648                         if (ISC_LIST_EMPTY(sock->send_list) &&
4649                             !sock->pending_send)
4650                                 select_poke(sock->manager, sock->fd,
4651                                             SELECT_POKE_WRITE);
4652                         ISC_LIST_ENQUEUE(sock->send_list, dev, ev_link);
4653
4654                         socket_log(sock, NULL, EVENT, NULL, 0, 0,
4655                                    "socket_send: event %p -> task %p",
4656                                    dev, ntask);
4657
4658                         if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
4659                                 result = ISC_R_INPROGRESS;
4660                         break;
4661                 }
4662
4663         case DOIO_HARD:
4664         case DOIO_SUCCESS:
4665                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
4666                         send_senddone_event(sock, &dev);
4667                 break;
4668         }
4669
4670         if (have_lock)
4671                 UNLOCK(&sock->lock);
4672
4673         return (result);
4674 }
4675
4676 ISC_SOCKETFUNC_SCOPE isc_result_t
4677 isc__socket_send(isc_socket_t *sock, isc_region_t *region,
4678                  isc_task_t *task, isc_taskaction_t action, const void *arg)
4679 {
4680         /*
4681          * REQUIRE() checking is performed in isc_socket_sendto().
4682          */
4683         return (isc__socket_sendto(sock, region, task, action, arg, NULL,
4684                                    NULL));
4685 }
4686
4687 ISC_SOCKETFUNC_SCOPE isc_result_t
4688 isc__socket_sendto(isc_socket_t *sock0, isc_region_t *region,
4689                    isc_task_t *task, isc_taskaction_t action, const void *arg,
4690                    isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
4691 {
4692         isc__socket_t *sock = (isc__socket_t *)sock0;
4693         isc_socketevent_t *dev;
4694         isc__socketmgr_t *manager;
4695
4696         REQUIRE(VALID_SOCKET(sock));
4697         REQUIRE(region != NULL);
4698         REQUIRE(task != NULL);
4699         REQUIRE(action != NULL);
4700
4701         manager = sock->manager;
4702         REQUIRE(VALID_MANAGER(manager));
4703
4704         INSIST(sock->bound);
4705
4706         dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
4707         if (dev == NULL)
4708                 return (ISC_R_NOMEMORY);
4709
4710         dev->region = *region;
4711
4712         return (socket_send(sock, dev, task, address, pktinfo, 0));
4713 }
4714
4715 ISC_SOCKETFUNC_SCOPE isc_result_t
4716 isc__socket_sendv(isc_socket_t *sock, isc_bufferlist_t *buflist,
4717                   isc_task_t *task, isc_taskaction_t action, const void *arg)
4718 {
4719         return (isc__socket_sendtov(sock, buflist, task, action, arg, NULL,
4720                                     NULL));
4721 }
4722
4723 ISC_SOCKETFUNC_SCOPE isc_result_t
4724 isc__socket_sendtov(isc_socket_t *sock0, isc_bufferlist_t *buflist,
4725                     isc_task_t *task, isc_taskaction_t action, const void *arg,
4726                     isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
4727 {
4728         isc__socket_t *sock = (isc__socket_t *)sock0;
4729         isc_socketevent_t *dev;
4730         isc__socketmgr_t *manager;
4731         unsigned int iocount;
4732         isc_buffer_t *buffer;
4733
4734         REQUIRE(VALID_SOCKET(sock));
4735         REQUIRE(buflist != NULL);
4736         REQUIRE(!ISC_LIST_EMPTY(*buflist));
4737         REQUIRE(task != NULL);
4738         REQUIRE(action != NULL);
4739
4740         manager = sock->manager;
4741         REQUIRE(VALID_MANAGER(manager));
4742
4743         iocount = isc_bufferlist_usedcount(buflist);
4744         REQUIRE(iocount > 0);
4745
4746         dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
4747         if (dev == NULL)
4748                 return (ISC_R_NOMEMORY);
4749
4750         /*
4751          * Move each buffer from the passed in list to our internal one.
4752          */
4753         buffer = ISC_LIST_HEAD(*buflist);
4754         while (buffer != NULL) {
4755                 ISC_LIST_DEQUEUE(*buflist, buffer, link);
4756                 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
4757                 buffer = ISC_LIST_HEAD(*buflist);
4758         }
4759
4760         return (socket_send(sock, dev, task, address, pktinfo, 0));
4761 }
4762
4763 ISC_SOCKETFUNC_SCOPE isc_result_t
4764 isc__socket_sendto2(isc_socket_t *sock0, isc_region_t *region,
4765                     isc_task_t *task,
4766                     isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
4767                     isc_socketevent_t *event, unsigned int flags)
4768 {
4769         isc__socket_t *sock = (isc__socket_t *)sock0;
4770
4771         REQUIRE(VALID_SOCKET(sock));
4772         REQUIRE((flags & ~(ISC_SOCKFLAG_IMMEDIATE|ISC_SOCKFLAG_NORETRY)) == 0);
4773         if ((flags & ISC_SOCKFLAG_NORETRY) != 0)
4774                 REQUIRE(sock->type == isc_sockettype_udp);
4775         event->ev_sender = sock;
4776         event->result = ISC_R_UNSET;
4777         ISC_LIST_INIT(event->bufferlist);
4778         event->region = *region;
4779         event->n = 0;
4780         event->offset = 0;
4781         event->attributes = 0;
4782
4783         return (socket_send(sock, event, task, address, pktinfo, flags));
4784 }
4785
4786 ISC_SOCKETFUNC_SCOPE void
4787 isc__socket_cleanunix(isc_sockaddr_t *sockaddr, isc_boolean_t active) {
4788 #ifdef ISC_PLATFORM_HAVESYSUNH
4789         int s;
4790         struct stat sb;
4791         char strbuf[ISC_STRERRORSIZE];
4792
4793         if (sockaddr->type.sa.sa_family != AF_UNIX)
4794                 return;
4795
4796 #ifndef S_ISSOCK
4797 #if defined(S_IFMT) && defined(S_IFSOCK)
4798 #define S_ISSOCK(mode) ((mode & S_IFMT)==S_IFSOCK)
4799 #elif defined(_S_IFMT) && defined(S_IFSOCK)
4800 #define S_ISSOCK(mode) ((mode & _S_IFMT)==S_IFSOCK)
4801 #endif
4802 #endif
4803
4804 #ifndef S_ISFIFO
4805 #if defined(S_IFMT) && defined(S_IFIFO)
4806 #define S_ISFIFO(mode) ((mode & S_IFMT)==S_IFIFO)
4807 #elif defined(_S_IFMT) && defined(S_IFIFO)
4808 #define S_ISFIFO(mode) ((mode & _S_IFMT)==S_IFIFO)
4809 #endif
4810 #endif
4811
4812 #if !defined(S_ISFIFO) && !defined(S_ISSOCK)
4813 #error You need to define S_ISFIFO and S_ISSOCK as appropriate for your platform.  See <sys/stat.h>.
4814 #endif
4815
4816 #ifndef S_ISFIFO
4817 #define S_ISFIFO(mode) 0
4818 #endif
4819
4820 #ifndef S_ISSOCK
4821 #define S_ISSOCK(mode) 0
4822 #endif
4823
4824         if (active) {
4825                 if (stat(sockaddr->type.sunix.sun_path, &sb) < 0) {
4826                         isc__strerror(errno, strbuf, sizeof(strbuf));
4827                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4828                                       ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4829                                       "isc_socket_cleanunix: stat(%s): %s",
4830                                       sockaddr->type.sunix.sun_path, strbuf);
4831                         return;
4832                 }
4833                 if (!(S_ISSOCK(sb.st_mode) || S_ISFIFO(sb.st_mode))) {
4834                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4835                                       ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4836                                       "isc_socket_cleanunix: %s: not a socket",
4837                                       sockaddr->type.sunix.sun_path);
4838                         return;
4839                 }
4840                 if (unlink(sockaddr->type.sunix.sun_path) < 0) {
4841                         isc__strerror(errno, strbuf, sizeof(strbuf));
4842                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4843                                       ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4844                                       "isc_socket_cleanunix: unlink(%s): %s",
4845                                       sockaddr->type.sunix.sun_path, strbuf);
4846                 }
4847                 return;
4848         }
4849
4850         s = socket(AF_UNIX, SOCK_STREAM, 0);
4851         if (s < 0) {
4852                 isc__strerror(errno, strbuf, sizeof(strbuf));
4853                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4854                               ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4855                               "isc_socket_cleanunix: socket(%s): %s",
4856                               sockaddr->type.sunix.sun_path, strbuf);
4857                 return;
4858         }
4859
4860         if (stat(sockaddr->type.sunix.sun_path, &sb) < 0) {
4861                 switch (errno) {
4862                 case ENOENT:    /* We exited cleanly last time */
4863                         break;
4864                 default:
4865                         isc__strerror(errno, strbuf, sizeof(strbuf));
4866                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4867                                       ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4868                                       "isc_socket_cleanunix: stat(%s): %s",
4869                                       sockaddr->type.sunix.sun_path, strbuf);
4870                         break;
4871                 }
4872                 goto cleanup;
4873         }
4874
4875         if (!(S_ISSOCK(sb.st_mode) || S_ISFIFO(sb.st_mode))) {
4876                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4877                               ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4878                               "isc_socket_cleanunix: %s: not a socket",
4879                               sockaddr->type.sunix.sun_path);
4880                 goto cleanup;
4881         }
4882
4883         if (connect(s, (struct sockaddr *)&sockaddr->type.sunix,
4884                     sizeof(sockaddr->type.sunix)) < 0) {
4885                 switch (errno) {
4886                 case ECONNREFUSED:
4887                 case ECONNRESET:
4888                         if (unlink(sockaddr->type.sunix.sun_path) < 0) {
4889                                 isc__strerror(errno, strbuf, sizeof(strbuf));
4890                                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4891                                               ISC_LOGMODULE_SOCKET,
4892                                               ISC_LOG_WARNING,
4893                                               "isc_socket_cleanunix: "
4894                                               "unlink(%s): %s",
4895                                               sockaddr->type.sunix.sun_path,
4896                                               strbuf);
4897                         }
4898                         break;
4899                 default:
4900                         isc__strerror(errno, strbuf, sizeof(strbuf));
4901                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4902                                       ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4903                                       "isc_socket_cleanunix: connect(%s): %s",
4904                                       sockaddr->type.sunix.sun_path, strbuf);
4905                         break;
4906                 }
4907         }
4908  cleanup:
4909         close(s);
4910 #else
4911         UNUSED(sockaddr);
4912         UNUSED(active);
4913 #endif
4914 }
4915
4916 ISC_SOCKETFUNC_SCOPE isc_result_t
4917 isc__socket_permunix(isc_sockaddr_t *sockaddr, isc_uint32_t perm,
4918                     isc_uint32_t owner, isc_uint32_t group)
4919 {
4920 #ifdef ISC_PLATFORM_HAVESYSUNH
4921         isc_result_t result = ISC_R_SUCCESS;
4922         char strbuf[ISC_STRERRORSIZE];
4923         char path[sizeof(sockaddr->type.sunix.sun_path)];
4924 #ifdef NEED_SECURE_DIRECTORY
4925         char *slash;
4926 #endif
4927
4928         REQUIRE(sockaddr->type.sa.sa_family == AF_UNIX);
4929         INSIST(strlen(sockaddr->type.sunix.sun_path) < sizeof(path));
4930         strcpy(path, sockaddr->type.sunix.sun_path);
4931
4932 #ifdef NEED_SECURE_DIRECTORY
4933         slash = strrchr(path, '/');
4934         if (slash != NULL) {
4935                 if (slash != path)
4936                         *slash = '\0';
4937                 else
4938                         strcpy(path, "/");
4939         } else
4940                 strcpy(path, ".");
4941 #endif
4942
4943         if (chmod(path, perm) < 0) {
4944                 isc__strerror(errno, strbuf, sizeof(strbuf));
4945                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4946                               ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4947                               "isc_socket_permunix: chmod(%s, %d): %s",
4948                               path, perm, strbuf);
4949                 result = ISC_R_FAILURE;
4950         }
4951         if (chown(path, owner, group) < 0) {
4952                 isc__strerror(errno, strbuf, sizeof(strbuf));
4953                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4954                               ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4955                               "isc_socket_permunix: chown(%s, %d, %d): %s",
4956                               path, owner, group,
4957                               strbuf);
4958                 result = ISC_R_FAILURE;
4959         }
4960         return (result);
4961 #else
4962         UNUSED(sockaddr);
4963         UNUSED(perm);
4964         UNUSED(owner);
4965         UNUSED(group);
4966         return (ISC_R_NOTIMPLEMENTED);
4967 #endif
4968 }
4969
4970 ISC_SOCKETFUNC_SCOPE isc_result_t
4971 isc__socket_bind(isc_socket_t *sock0, isc_sockaddr_t *sockaddr,
4972                  unsigned int options) {
4973         isc__socket_t *sock = (isc__socket_t *)sock0;
4974         char strbuf[ISC_STRERRORSIZE];
4975         int on = 1;
4976
4977         REQUIRE(VALID_SOCKET(sock));
4978
4979         LOCK(&sock->lock);
4980
4981         INSIST(!sock->bound);
4982
4983         if (sock->pf != sockaddr->type.sa.sa_family) {
4984                 UNLOCK(&sock->lock);
4985                 return (ISC_R_FAMILYMISMATCH);
4986         }
4987         /*
4988          * Only set SO_REUSEADDR when we want a specific port.
4989          */
4990 #ifdef AF_UNIX
4991         if (sock->pf == AF_UNIX)
4992                 goto bind_socket;
4993 #endif
4994         if ((options & ISC_SOCKET_REUSEADDRESS) != 0 &&
4995             isc_sockaddr_getport(sockaddr) != (in_port_t)0 &&
4996             setsockopt(sock->fd, SOL_SOCKET, SO_REUSEADDR, (void *)&on,
4997                        sizeof(on)) < 0) {
4998                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4999                                  "setsockopt(%d) %s", sock->fd,
5000                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
5001                                                 ISC_MSG_FAILED, "failed"));
5002                 /* Press on... */
5003         }
5004 #ifdef AF_UNIX
5005  bind_socket:
5006 #endif
5007         if (bind(sock->fd, &sockaddr->type.sa, sockaddr->length) < 0) {
5008                 inc_stats(sock->manager->stats,
5009                           sock->statsindex[STATID_BINDFAIL]);
5010
5011                 UNLOCK(&sock->lock);
5012                 switch (errno) {
5013                 case EACCES:
5014                         return (ISC_R_NOPERM);
5015                 case EADDRNOTAVAIL:
5016                         return (ISC_R_ADDRNOTAVAIL);
5017                 case EADDRINUSE:
5018                         return (ISC_R_ADDRINUSE);
5019                 case EINVAL:
5020                         return (ISC_R_BOUND);
5021                 default:
5022                         isc__strerror(errno, strbuf, sizeof(strbuf));
5023                         UNEXPECTED_ERROR(__FILE__, __LINE__, "bind: %s",
5024                                          strbuf);
5025                         return (ISC_R_UNEXPECTED);
5026                 }
5027         }
5028
5029         socket_log(sock, sockaddr, TRACE,
5030                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_BOUND, "bound");
5031         sock->bound = 1;
5032
5033         UNLOCK(&sock->lock);
5034         return (ISC_R_SUCCESS);
5035 }
5036
5037 /*
5038  * Enable this only for specific OS versions, and only when they have repaired
5039  * their problems with it.  Until then, this is is broken and needs to be
5040  * diabled by default.  See RT22589 for details.
5041  */
5042 #undef ENABLE_ACCEPTFILTER
5043
5044 ISC_SOCKETFUNC_SCOPE isc_result_t
5045 isc__socket_filter(isc_socket_t *sock0, const char *filter) {
5046         isc__socket_t *sock = (isc__socket_t *)sock0;
5047 #if defined(SO_ACCEPTFILTER) && defined(ENABLE_ACCEPTFILTER)
5048         char strbuf[ISC_STRERRORSIZE];
5049         struct accept_filter_arg afa;
5050 #else
5051         UNUSED(sock);
5052         UNUSED(filter);
5053 #endif
5054
5055         REQUIRE(VALID_SOCKET(sock));
5056
5057 #if defined(SO_ACCEPTFILTER) && defined(ENABLE_ACCEPTFILTER)
5058         bzero(&afa, sizeof(afa));
5059         strncpy(afa.af_name, filter, sizeof(afa.af_name));
5060         if (setsockopt(sock->fd, SOL_SOCKET, SO_ACCEPTFILTER,
5061                          &afa, sizeof(afa)) == -1) {
5062                 isc__strerror(errno, strbuf, sizeof(strbuf));
5063                 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
5064                            ISC_MSG_FILTER, "setsockopt(SO_ACCEPTFILTER): %s",
5065                            strbuf);
5066                 return (ISC_R_FAILURE);
5067         }
5068         return (ISC_R_SUCCESS);
5069 #else
5070         return (ISC_R_NOTIMPLEMENTED);
5071 #endif
5072 }
5073
5074 /*
5075  * Set up to listen on a given socket.  We do this by creating an internal
5076  * event that will be dispatched when the socket has read activity.  The
5077  * watcher will send the internal event to the task when there is a new
5078  * connection.
5079  *
5080  * Unlike in read, we don't preallocate a done event here.  Every time there
5081  * is a new connection we'll have to allocate a new one anyway, so we might
5082  * as well keep things simple rather than having to track them.
5083  */
5084 ISC_SOCKETFUNC_SCOPE isc_result_t
5085 isc__socket_listen(isc_socket_t *sock0, unsigned int backlog) {
5086         isc__socket_t *sock = (isc__socket_t *)sock0;
5087         char strbuf[ISC_STRERRORSIZE];
5088
5089         REQUIRE(VALID_SOCKET(sock));
5090
5091         LOCK(&sock->lock);
5092
5093         REQUIRE(!sock->listener);
5094         REQUIRE(sock->bound);
5095         REQUIRE(sock->type == isc_sockettype_tcp ||
5096                 sock->type == isc_sockettype_unix);
5097
5098         if (backlog == 0)
5099                 backlog = SOMAXCONN;
5100
5101         if (listen(sock->fd, (int)backlog) < 0) {
5102                 UNLOCK(&sock->lock);
5103                 isc__strerror(errno, strbuf, sizeof(strbuf));
5104
5105                 UNEXPECTED_ERROR(__FILE__, __LINE__, "listen: %s", strbuf);
5106
5107                 return (ISC_R_UNEXPECTED);
5108         }
5109
5110         sock->listener = 1;
5111
5112         UNLOCK(&sock->lock);
5113         return (ISC_R_SUCCESS);
5114 }
5115
5116 /*
5117  * This should try to do aggressive accept() XXXMLG
5118  */
5119 ISC_SOCKETFUNC_SCOPE isc_result_t
5120 isc__socket_accept(isc_socket_t *sock0,
5121                   isc_task_t *task, isc_taskaction_t action, const void *arg)
5122 {
5123         isc__socket_t *sock = (isc__socket_t *)sock0;
5124         isc_socket_newconnev_t *dev;
5125         isc__socketmgr_t *manager;
5126         isc_task_t *ntask = NULL;
5127         isc__socket_t *nsock;
5128         isc_result_t result;
5129         isc_boolean_t do_poke = ISC_FALSE;
5130
5131         REQUIRE(VALID_SOCKET(sock));
5132         manager = sock->manager;
5133         REQUIRE(VALID_MANAGER(manager));
5134
5135         LOCK(&sock->lock);
5136
5137         REQUIRE(sock->listener);
5138
5139         /*
5140          * Sender field is overloaded here with the task we will be sending
5141          * this event to.  Just before the actual event is delivered the
5142          * actual ev_sender will be touched up to be the socket.
5143          */
5144         dev = (isc_socket_newconnev_t *)
5145                 isc_event_allocate(manager->mctx, task, ISC_SOCKEVENT_NEWCONN,
5146                                    action, arg, sizeof(*dev));
5147         if (dev == NULL) {
5148                 UNLOCK(&sock->lock);
5149                 return (ISC_R_NOMEMORY);
5150         }
5151         ISC_LINK_INIT(dev, ev_link);
5152
5153         result = allocate_socket(manager, sock->type, &nsock);
5154         if (result != ISC_R_SUCCESS) {
5155                 isc_event_free(ISC_EVENT_PTR(&dev));
5156                 UNLOCK(&sock->lock);
5157                 return (result);
5158         }
5159
5160         /*
5161          * Attach to socket and to task.
5162          */
5163         isc_task_attach(task, &ntask);
5164         if (isc_task_exiting(ntask)) {
5165                 free_socket(&nsock);
5166                 isc_task_detach(&ntask);
5167                 isc_event_free(ISC_EVENT_PTR(&dev));
5168                 UNLOCK(&sock->lock);
5169                 return (ISC_R_SHUTTINGDOWN);
5170         }
5171         nsock->references++;
5172         nsock->statsindex = sock->statsindex;
5173
5174         dev->ev_sender = ntask;
5175         dev->newsocket = (isc_socket_t *)nsock;
5176
5177         /*
5178          * Poke watcher here.  We still have the socket locked, so there
5179          * is no race condition.  We will keep the lock for such a short
5180          * bit of time waking it up now or later won't matter all that much.
5181          */
5182         if (ISC_LIST_EMPTY(sock->accept_list))
5183                 do_poke = ISC_TRUE;
5184
5185         ISC_LIST_ENQUEUE(sock->accept_list, dev, ev_link);
5186
5187         if (do_poke)
5188                 select_poke(manager, sock->fd, SELECT_POKE_ACCEPT);
5189
5190         UNLOCK(&sock->lock);
5191         return (ISC_R_SUCCESS);
5192 }
5193
5194 ISC_SOCKETFUNC_SCOPE isc_result_t
5195 isc__socket_connect(isc_socket_t *sock0, isc_sockaddr_t *addr,
5196                    isc_task_t *task, isc_taskaction_t action, const void *arg)
5197 {
5198         isc__socket_t *sock = (isc__socket_t *)sock0;
5199         isc_socket_connev_t *dev;
5200         isc_task_t *ntask = NULL;
5201         isc__socketmgr_t *manager;
5202         int cc;
5203         char strbuf[ISC_STRERRORSIZE];
5204         char addrbuf[ISC_SOCKADDR_FORMATSIZE];
5205
5206         REQUIRE(VALID_SOCKET(sock));
5207         REQUIRE(addr != NULL);
5208         REQUIRE(task != NULL);
5209         REQUIRE(action != NULL);
5210
5211         manager = sock->manager;
5212         REQUIRE(VALID_MANAGER(manager));
5213         REQUIRE(addr != NULL);
5214
5215         if (isc_sockaddr_ismulticast(addr))
5216                 return (ISC_R_MULTICAST);
5217
5218         LOCK(&sock->lock);
5219
5220         REQUIRE(!sock->connecting);
5221
5222         dev = (isc_socket_connev_t *)isc_event_allocate(manager->mctx, sock,
5223                                                         ISC_SOCKEVENT_CONNECT,
5224                                                         action, arg,
5225                                                         sizeof(*dev));
5226         if (dev == NULL) {
5227                 UNLOCK(&sock->lock);
5228                 return (ISC_R_NOMEMORY);
5229         }
5230         ISC_LINK_INIT(dev, ev_link);
5231
5232         /*
5233          * Try to do the connect right away, as there can be only one
5234          * outstanding, and it might happen to complete.
5235          */
5236         sock->peer_address = *addr;
5237         cc = connect(sock->fd, &addr->type.sa, addr->length);
5238         if (cc < 0) {
5239                 /*
5240                  * HP-UX "fails" to connect a UDP socket and sets errno to
5241                  * EINPROGRESS if it's non-blocking.  We'd rather regard this as
5242                  * a success and let the user detect it if it's really an error
5243                  * at the time of sending a packet on the socket.
5244                  */
5245                 if (sock->type == isc_sockettype_udp && errno == EINPROGRESS) {
5246                         cc = 0;
5247                         goto success;
5248                 }
5249                 if (SOFT_ERROR(errno) || errno == EINPROGRESS)
5250                         goto queue;
5251
5252                 switch (errno) {
5253 #define ERROR_MATCH(a, b) case a: dev->result = b; goto err_exit;
5254                         ERROR_MATCH(EACCES, ISC_R_NOPERM);
5255                         ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
5256                         ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
5257                         ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
5258                         ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
5259 #ifdef EHOSTDOWN
5260                         ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
5261 #endif
5262                         ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
5263                         ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
5264                         ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
5265                         ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
5266                         ERROR_MATCH(ECONNRESET, ISC_R_CONNECTIONRESET);
5267 #undef ERROR_MATCH
5268                 }
5269
5270                 sock->connected = 0;
5271
5272                 isc__strerror(errno, strbuf, sizeof(strbuf));
5273                 isc_sockaddr_format(addr, addrbuf, sizeof(addrbuf));
5274                 UNEXPECTED_ERROR(__FILE__, __LINE__, "connect(%s) %d/%s",
5275                                  addrbuf, errno, strbuf);
5276
5277                 UNLOCK(&sock->lock);
5278                 inc_stats(sock->manager->stats,
5279                           sock->statsindex[STATID_CONNECTFAIL]);
5280                 isc_event_free(ISC_EVENT_PTR(&dev));
5281                 return (ISC_R_UNEXPECTED);
5282
5283         err_exit:
5284                 sock->connected = 0;
5285                 isc_task_send(task, ISC_EVENT_PTR(&dev));
5286
5287                 UNLOCK(&sock->lock);
5288                 inc_stats(sock->manager->stats,
5289                           sock->statsindex[STATID_CONNECTFAIL]);
5290                 return (ISC_R_SUCCESS);
5291         }
5292
5293         /*
5294          * If connect completed, fire off the done event.
5295          */
5296  success:
5297         if (cc == 0) {
5298                 sock->connected = 1;
5299                 sock->bound = 1;
5300                 dev->result = ISC_R_SUCCESS;
5301                 isc_task_send(task, ISC_EVENT_PTR(&dev));
5302
5303                 UNLOCK(&sock->lock);
5304
5305                 inc_stats(sock->manager->stats,
5306                           sock->statsindex[STATID_CONNECT]);
5307
5308                 return (ISC_R_SUCCESS);
5309         }
5310
5311  queue:
5312
5313         /*
5314          * Attach to task.
5315          */
5316         isc_task_attach(task, &ntask);
5317
5318         sock->connecting = 1;
5319
5320         dev->ev_sender = ntask;
5321
5322         /*
5323          * Poke watcher here.  We still have the socket locked, so there
5324          * is no race condition.  We will keep the lock for such a short
5325          * bit of time waking it up now or later won't matter all that much.
5326          */
5327         if (sock->connect_ev == NULL)
5328                 select_poke(manager, sock->fd, SELECT_POKE_CONNECT);
5329
5330         sock->connect_ev = dev;
5331
5332         UNLOCK(&sock->lock);
5333         return (ISC_R_SUCCESS);
5334 }
5335
5336 /*
5337  * Called when a socket with a pending connect() finishes.
5338  */
5339 static void
5340 internal_connect(isc_task_t *me, isc_event_t *ev) {
5341         isc__socket_t *sock;
5342         isc_socket_connev_t *dev;
5343         isc_task_t *task;
5344         int cc;
5345         ISC_SOCKADDR_LEN_T optlen;
5346         char strbuf[ISC_STRERRORSIZE];
5347         char peerbuf[ISC_SOCKADDR_FORMATSIZE];
5348
5349         UNUSED(me);
5350         INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
5351
5352         sock = ev->ev_sender;
5353         INSIST(VALID_SOCKET(sock));
5354
5355         LOCK(&sock->lock);
5356
5357         /*
5358          * When the internal event was sent the reference count was bumped
5359          * to keep the socket around for us.  Decrement the count here.
5360          */
5361         INSIST(sock->references > 0);
5362         sock->references--;
5363         if (sock->references == 0) {
5364                 UNLOCK(&sock->lock);
5365                 destroy(&sock);
5366                 return;
5367         }
5368
5369         /*
5370          * Has this event been canceled?
5371          */
5372         dev = sock->connect_ev;
5373         if (dev == NULL) {
5374                 INSIST(!sock->connecting);
5375                 UNLOCK(&sock->lock);
5376                 return;
5377         }
5378
5379         INSIST(sock->connecting);
5380         sock->connecting = 0;
5381
5382         /*
5383          * Get any possible error status here.
5384          */
5385         optlen = sizeof(cc);
5386         if (getsockopt(sock->fd, SOL_SOCKET, SO_ERROR,
5387                        (void *)&cc, (void *)&optlen) < 0)
5388                 cc = errno;
5389         else
5390                 errno = cc;
5391
5392         if (errno != 0) {
5393                 /*
5394                  * If the error is EAGAIN, just re-select on this
5395                  * fd and pretend nothing strange happened.
5396                  */
5397                 if (SOFT_ERROR(errno) || errno == EINPROGRESS) {
5398                         sock->connecting = 1;
5399                         select_poke(sock->manager, sock->fd,
5400                                     SELECT_POKE_CONNECT);
5401                         UNLOCK(&sock->lock);
5402
5403                         return;
5404                 }
5405
5406                 inc_stats(sock->manager->stats,
5407                           sock->statsindex[STATID_CONNECTFAIL]);
5408
5409                 /*
5410                  * Translate other errors into ISC_R_* flavors.
5411                  */
5412                 switch (errno) {
5413 #define ERROR_MATCH(a, b) case a: dev->result = b; break;
5414                         ERROR_MATCH(EACCES, ISC_R_NOPERM);
5415                         ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
5416                         ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
5417                         ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
5418                         ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
5419 #ifdef EHOSTDOWN
5420                         ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
5421 #endif
5422                         ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
5423                         ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
5424                         ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
5425                         ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
5426                         ERROR_MATCH(ETIMEDOUT, ISC_R_TIMEDOUT);
5427                         ERROR_MATCH(ECONNRESET, ISC_R_CONNECTIONRESET);
5428 #undef ERROR_MATCH
5429                 default:
5430                         dev->result = ISC_R_UNEXPECTED;
5431                         isc_sockaddr_format(&sock->peer_address, peerbuf,
5432                                             sizeof(peerbuf));
5433                         isc__strerror(errno, strbuf, sizeof(strbuf));
5434                         UNEXPECTED_ERROR(__FILE__, __LINE__,
5435                                          "internal_connect: connect(%s) %s",
5436                                          peerbuf, strbuf);
5437                 }
5438         } else {
5439                 inc_stats(sock->manager->stats,
5440                           sock->statsindex[STATID_CONNECT]);
5441                 dev->result = ISC_R_SUCCESS;
5442                 sock->connected = 1;
5443                 sock->bound = 1;
5444         }
5445
5446         sock->connect_ev = NULL;
5447
5448         UNLOCK(&sock->lock);
5449
5450         task = dev->ev_sender;
5451         dev->ev_sender = sock;
5452         isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
5453 }
5454
5455 ISC_SOCKETFUNC_SCOPE isc_result_t
5456 isc__socket_getpeername(isc_socket_t *sock0, isc_sockaddr_t *addressp) {
5457         isc__socket_t *sock = (isc__socket_t *)sock0;
5458         isc_result_t result;
5459
5460         REQUIRE(VALID_SOCKET(sock));
5461         REQUIRE(addressp != NULL);
5462
5463         LOCK(&sock->lock);
5464
5465         if (sock->connected) {
5466                 *addressp = sock->peer_address;
5467                 result = ISC_R_SUCCESS;
5468         } else {
5469                 result = ISC_R_NOTCONNECTED;
5470         }
5471
5472         UNLOCK(&sock->lock);
5473
5474         return (result);
5475 }
5476
5477 ISC_SOCKETFUNC_SCOPE isc_result_t
5478 isc__socket_getsockname(isc_socket_t *sock0, isc_sockaddr_t *addressp) {
5479         isc__socket_t *sock = (isc__socket_t *)sock0;
5480         ISC_SOCKADDR_LEN_T len;
5481         isc_result_t result;
5482         char strbuf[ISC_STRERRORSIZE];
5483
5484         REQUIRE(VALID_SOCKET(sock));
5485         REQUIRE(addressp != NULL);
5486
5487         LOCK(&sock->lock);
5488
5489         if (!sock->bound) {
5490                 result = ISC_R_NOTBOUND;
5491                 goto out;
5492         }
5493
5494         result = ISC_R_SUCCESS;
5495
5496         len = sizeof(addressp->type);
5497         if (getsockname(sock->fd, &addressp->type.sa, (void *)&len) < 0) {
5498                 isc__strerror(errno, strbuf, sizeof(strbuf));
5499                 UNEXPECTED_ERROR(__FILE__, __LINE__, "getsockname: %s",
5500                                  strbuf);
5501                 result = ISC_R_UNEXPECTED;
5502                 goto out;
5503         }
5504         addressp->length = (unsigned int)len;
5505
5506  out:
5507         UNLOCK(&sock->lock);
5508
5509         return (result);
5510 }
5511
5512 /*
5513  * Run through the list of events on this socket, and cancel the ones
5514  * queued for task "task" of type "how".  "how" is a bitmask.
5515  */
5516 ISC_SOCKETFUNC_SCOPE void
5517 isc__socket_cancel(isc_socket_t *sock0, isc_task_t *task, unsigned int how) {
5518         isc__socket_t *sock = (isc__socket_t *)sock0;
5519
5520         REQUIRE(VALID_SOCKET(sock));
5521
5522         /*
5523          * Quick exit if there is nothing to do.  Don't even bother locking
5524          * in this case.
5525          */
5526         if (how == 0)
5527                 return;
5528
5529         LOCK(&sock->lock);
5530
5531         /*
5532          * All of these do the same thing, more or less.
5533          * Each will:
5534          *      o If the internal event is marked as "posted" try to
5535          *        remove it from the task's queue.  If this fails, mark it
5536          *        as canceled instead, and let the task clean it up later.
5537          *      o For each I/O request for that task of that type, post
5538          *        its done event with status of "ISC_R_CANCELED".
5539          *      o Reset any state needed.
5540          */
5541         if (((how & ISC_SOCKCANCEL_RECV) == ISC_SOCKCANCEL_RECV)
5542             && !ISC_LIST_EMPTY(sock->recv_list)) {
5543                 isc_socketevent_t      *dev;
5544                 isc_socketevent_t      *next;
5545                 isc_task_t             *current_task;
5546
5547                 dev = ISC_LIST_HEAD(sock->recv_list);
5548
5549                 while (dev != NULL) {
5550                         current_task = dev->ev_sender;
5551                         next = ISC_LIST_NEXT(dev, ev_link);
5552
5553                         if ((task == NULL) || (task == current_task)) {
5554                                 dev->result = ISC_R_CANCELED;
5555                                 send_recvdone_event(sock, &dev);
5556                         }
5557                         dev = next;
5558                 }
5559         }
5560
5561         if (((how & ISC_SOCKCANCEL_SEND) == ISC_SOCKCANCEL_SEND)
5562             && !ISC_LIST_EMPTY(sock->send_list)) {
5563                 isc_socketevent_t      *dev;
5564                 isc_socketevent_t      *next;
5565                 isc_task_t             *current_task;
5566
5567                 dev = ISC_LIST_HEAD(sock->send_list);
5568
5569                 while (dev != NULL) {
5570                         current_task = dev->ev_sender;
5571                         next = ISC_LIST_NEXT(dev, ev_link);
5572
5573                         if ((task == NULL) || (task == current_task)) {
5574                                 dev->result = ISC_R_CANCELED;
5575                                 send_senddone_event(sock, &dev);
5576                         }
5577                         dev = next;
5578                 }
5579         }
5580
5581         if (((how & ISC_SOCKCANCEL_ACCEPT) == ISC_SOCKCANCEL_ACCEPT)
5582             && !ISC_LIST_EMPTY(sock->accept_list)) {
5583                 isc_socket_newconnev_t *dev;
5584                 isc_socket_newconnev_t *next;
5585                 isc_task_t             *current_task;
5586
5587                 dev = ISC_LIST_HEAD(sock->accept_list);
5588                 while (dev != NULL) {
5589                         current_task = dev->ev_sender;
5590                         next = ISC_LIST_NEXT(dev, ev_link);
5591
5592                         if ((task == NULL) || (task == current_task)) {
5593
5594                                 ISC_LIST_UNLINK(sock->accept_list, dev,
5595                                                 ev_link);
5596
5597                                 NEWCONNSOCK(dev)->references--;
5598                                 free_socket((isc__socket_t **)&dev->newsocket);
5599
5600                                 dev->result = ISC_R_CANCELED;
5601                                 dev->ev_sender = sock;
5602                                 isc_task_sendanddetach(&current_task,
5603                                                        ISC_EVENT_PTR(&dev));
5604                         }
5605
5606                         dev = next;
5607                 }
5608         }
5609
5610         /*
5611          * Connecting is not a list.
5612          */
5613         if (((how & ISC_SOCKCANCEL_CONNECT) == ISC_SOCKCANCEL_CONNECT)
5614             && sock->connect_ev != NULL) {
5615                 isc_socket_connev_t    *dev;
5616                 isc_task_t             *current_task;
5617
5618                 INSIST(sock->connecting);
5619                 sock->connecting = 0;
5620
5621                 dev = sock->connect_ev;
5622                 current_task = dev->ev_sender;
5623
5624                 if ((task == NULL) || (task == current_task)) {
5625                         sock->connect_ev = NULL;
5626
5627                         dev->result = ISC_R_CANCELED;
5628                         dev->ev_sender = sock;
5629                         isc_task_sendanddetach(&current_task,
5630                                                ISC_EVENT_PTR(&dev));
5631                 }
5632         }
5633
5634         UNLOCK(&sock->lock);
5635 }
5636
5637 ISC_SOCKETFUNC_SCOPE isc_sockettype_t
5638 isc__socket_gettype(isc_socket_t *sock0) {
5639         isc__socket_t *sock = (isc__socket_t *)sock0;
5640
5641         REQUIRE(VALID_SOCKET(sock));
5642
5643         return (sock->type);
5644 }
5645
5646 ISC_SOCKETFUNC_SCOPE isc_boolean_t
5647 isc__socket_isbound(isc_socket_t *sock0) {
5648         isc__socket_t *sock = (isc__socket_t *)sock0;
5649         isc_boolean_t val;
5650
5651         REQUIRE(VALID_SOCKET(sock));
5652
5653         LOCK(&sock->lock);
5654         val = ((sock->bound) ? ISC_TRUE : ISC_FALSE);
5655         UNLOCK(&sock->lock);
5656
5657         return (val);
5658 }
5659
5660 ISC_SOCKETFUNC_SCOPE void
5661 isc__socket_ipv6only(isc_socket_t *sock0, isc_boolean_t yes) {
5662         isc__socket_t *sock = (isc__socket_t *)sock0;
5663 #if defined(IPV6_V6ONLY)
5664         int onoff = yes ? 1 : 0;
5665 #else
5666         UNUSED(yes);
5667         UNUSED(sock);
5668 #endif
5669
5670         REQUIRE(VALID_SOCKET(sock));
5671
5672 #ifdef IPV6_V6ONLY
5673         if (sock->pf == AF_INET6) {
5674                 if (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_V6ONLY,
5675                                (void *)&onoff, sizeof(int)) < 0) {
5676                         char strbuf[ISC_STRERRORSIZE];
5677                         isc__strerror(errno, strbuf, sizeof(strbuf));
5678                         UNEXPECTED_ERROR(__FILE__, __LINE__,
5679                                          "setsockopt(%d, IPV6_V6ONLY) "
5680                                          "%s: %s", sock->fd,
5681                                          isc_msgcat_get(isc_msgcat,
5682                                                         ISC_MSGSET_GENERAL,
5683                                                         ISC_MSG_FAILED,
5684                                                         "failed"),
5685                                          strbuf);
5686                 }
5687         }
5688         FIX_IPV6_RECVPKTINFO(sock);     /* AIX */
5689 #endif
5690 }
5691
5692 #ifndef USE_WATCHER_THREAD
5693 /*
5694  * In our assumed scenario, we can simply use a single static object.
5695  * XXX: this is not true if the application uses multiple threads with
5696  *      'multi-context' mode.  Fixing this is a future TODO item.
5697  */
5698 static isc_socketwait_t swait_private;
5699
5700 int
5701 isc__socketmgr_waitevents(isc_socketmgr_t *manager0, struct timeval *tvp,
5702                           isc_socketwait_t **swaitp)
5703 {
5704         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
5705
5706
5707         int n;
5708 #ifdef USE_KQUEUE
5709         struct timespec ts, *tsp;
5710 #endif
5711 #ifdef USE_EPOLL
5712         int timeout;
5713 #endif
5714 #ifdef USE_DEVPOLL
5715         struct dvpoll dvp;
5716 #endif
5717
5718         REQUIRE(swaitp != NULL && *swaitp == NULL);
5719
5720 #ifdef USE_SHARED_MANAGER
5721         if (manager == NULL)
5722                 manager = socketmgr;
5723 #endif
5724         if (manager == NULL)
5725                 return (0);
5726
5727 #ifdef USE_KQUEUE
5728         if (tvp != NULL) {
5729                 ts.tv_sec = tvp->tv_sec;
5730                 ts.tv_nsec = tvp->tv_usec * 1000;
5731                 tsp = &ts;
5732         } else
5733                 tsp = NULL;
5734         swait_private.nevents = kevent(manager->kqueue_fd, NULL, 0,
5735                                        manager->events, manager->nevents,
5736                                        tsp);
5737         n = swait_private.nevents;
5738 #elif defined(USE_EPOLL)
5739         if (tvp != NULL)
5740                 timeout = tvp->tv_sec * 1000 + (tvp->tv_usec + 999) / 1000;
5741         else
5742                 timeout = -1;
5743         swait_private.nevents = epoll_wait(manager->epoll_fd,
5744                                            manager->events,
5745                                            manager->nevents, timeout);
5746         n = swait_private.nevents;
5747 #elif defined(USE_DEVPOLL)
5748         dvp.dp_fds = manager->events;
5749         dvp.dp_nfds = manager->nevents;
5750         if (tvp != NULL) {
5751                 dvp.dp_timeout = tvp->tv_sec * 1000 +
5752                         (tvp->tv_usec + 999) / 1000;
5753         } else
5754                 dvp.dp_timeout = -1;
5755         swait_private.nevents = ioctl(manager->devpoll_fd, DP_POLL, &dvp);
5756         n = swait_private.nevents;
5757 #elif defined(USE_SELECT)
5758         memcpy(manager->read_fds_copy, manager->read_fds,  manager->fd_bufsize);
5759         memcpy(manager->write_fds_copy, manager->write_fds,
5760                manager->fd_bufsize);
5761
5762         swait_private.readset = manager->read_fds_copy;
5763         swait_private.writeset = manager->write_fds_copy;
5764         swait_private.maxfd = manager->maxfd + 1;
5765
5766         n = select(swait_private.maxfd, swait_private.readset,
5767                    swait_private.writeset, NULL, tvp);
5768 #endif
5769
5770         *swaitp = &swait_private;
5771         return (n);
5772 }
5773
5774 isc_result_t
5775 isc__socketmgr_dispatch(isc_socketmgr_t *manager0, isc_socketwait_t *swait) {
5776         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
5777
5778         REQUIRE(swait == &swait_private);
5779
5780 #ifdef USE_SHARED_MANAGER
5781         if (manager == NULL)
5782                 manager = socketmgr;
5783 #endif
5784         if (manager == NULL)
5785                 return (ISC_R_NOTFOUND);
5786
5787 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
5788         (void)process_fds(manager, manager->events, swait->nevents);
5789         return (ISC_R_SUCCESS);
5790 #elif defined(USE_SELECT)
5791         process_fds(manager, swait->maxfd, swait->readset, swait->writeset);
5792         return (ISC_R_SUCCESS);
5793 #endif
5794 }
5795 #endif /* USE_WATCHER_THREAD */
5796
5797 #ifdef BIND9
5798 void
5799 isc__socket_setname(isc_socket_t *socket0, const char *name, void *tag) {
5800         isc__socket_t *socket = (isc__socket_t *)socket0;
5801
5802         /*
5803          * Name 'socket'.
5804          */
5805
5806         REQUIRE(VALID_SOCKET(socket));
5807
5808         LOCK(&socket->lock);
5809         memset(socket->name, 0, sizeof(socket->name));
5810         strncpy(socket->name, name, sizeof(socket->name) - 1);
5811         socket->tag = tag;
5812         UNLOCK(&socket->lock);
5813 }
5814
5815 ISC_SOCKETFUNC_SCOPE const char *
5816 isc__socket_getname(isc_socket_t *socket0) {
5817         isc__socket_t *socket = (isc__socket_t *)socket0;
5818
5819         return (socket->name);
5820 }
5821
5822 void *
5823 isc__socket_gettag(isc_socket_t *socket0) {
5824         isc__socket_t *socket = (isc__socket_t *)socket0;
5825
5826         return (socket->tag);
5827 }
5828 #endif  /* BIND9 */
5829
5830 #ifdef USE_SOCKETIMPREGISTER
5831 isc_result_t
5832 isc__socket_register() {
5833         return (isc_socket_register(isc__socketmgr_create));
5834 }
5835 #endif
5836
5837 #if defined(HAVE_LIBXML2) && defined(BIND9)
5838
5839 static const char *
5840 _socktype(isc_sockettype_t type)
5841 {
5842         if (type == isc_sockettype_udp)
5843                 return ("udp");
5844         else if (type == isc_sockettype_tcp)
5845                 return ("tcp");
5846         else if (type == isc_sockettype_unix)
5847                 return ("unix");
5848         else if (type == isc_sockettype_fdwatch)
5849                 return ("fdwatch");
5850         else
5851                 return ("not-initialized");
5852 }
5853
5854 ISC_SOCKETFUNC_SCOPE void
5855 isc_socketmgr_renderxml(isc_socketmgr_t *mgr0, xmlTextWriterPtr writer) {
5856         isc__socketmgr_t *mgr = (isc__socketmgr_t *)mgr0;
5857         isc__socket_t *sock;
5858         char peerbuf[ISC_SOCKADDR_FORMATSIZE];
5859         isc_sockaddr_t addr;
5860         ISC_SOCKADDR_LEN_T len;
5861
5862         LOCK(&mgr->lock);
5863
5864 #ifdef USE_SHARED_MANAGER
5865         xmlTextWriterStartElement(writer, ISC_XMLCHAR "references");
5866         xmlTextWriterWriteFormatString(writer, "%d", mgr->refs);
5867         xmlTextWriterEndElement(writer);
5868 #endif  /* USE_SHARED_MANAGER */
5869
5870         xmlTextWriterStartElement(writer, ISC_XMLCHAR "sockets");
5871         sock = ISC_LIST_HEAD(mgr->socklist);
5872         while (sock != NULL) {
5873                 LOCK(&sock->lock);
5874                 xmlTextWriterStartElement(writer, ISC_XMLCHAR "socket");
5875
5876                 xmlTextWriterStartElement(writer, ISC_XMLCHAR "id");
5877                 xmlTextWriterWriteFormatString(writer, "%p", sock);
5878                 xmlTextWriterEndElement(writer);
5879
5880                 if (sock->name[0] != 0) {
5881                         xmlTextWriterStartElement(writer, ISC_XMLCHAR "name");
5882                         xmlTextWriterWriteFormatString(writer, "%s",
5883                                                        sock->name);
5884                         xmlTextWriterEndElement(writer); /* name */
5885                 }
5886
5887                 xmlTextWriterStartElement(writer, ISC_XMLCHAR "references");
5888                 xmlTextWriterWriteFormatString(writer, "%d", sock->references);
5889                 xmlTextWriterEndElement(writer);
5890
5891                 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "type",
5892                                           ISC_XMLCHAR _socktype(sock->type));
5893
5894                 if (sock->connected) {
5895                         isc_sockaddr_format(&sock->peer_address, peerbuf,
5896                                             sizeof(peerbuf));
5897                         xmlTextWriterWriteElement(writer,
5898                                                   ISC_XMLCHAR "peer-address",
5899                                                   ISC_XMLCHAR peerbuf);
5900                 }
5901
5902                 len = sizeof(addr);
5903                 if (getsockname(sock->fd, &addr.type.sa, (void *)&len) == 0) {
5904                         isc_sockaddr_format(&addr, peerbuf, sizeof(peerbuf));
5905                         xmlTextWriterWriteElement(writer,
5906                                                   ISC_XMLCHAR "local-address",
5907                                                   ISC_XMLCHAR peerbuf);
5908                 }
5909
5910                 xmlTextWriterStartElement(writer, ISC_XMLCHAR "states");
5911                 if (sock->pending_recv)
5912                         xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5913                                                 ISC_XMLCHAR "pending-receive");
5914                 if (sock->pending_send)
5915                         xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5916                                                   ISC_XMLCHAR "pending-send");
5917                 if (sock->pending_accept)
5918                         xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5919                                                  ISC_XMLCHAR "pending_accept");
5920                 if (sock->listener)
5921                         xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5922                                                   ISC_XMLCHAR "listener");
5923                 if (sock->connected)
5924                         xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5925                                                   ISC_XMLCHAR "connected");
5926                 if (sock->connecting)
5927                         xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5928                                                   ISC_XMLCHAR "connecting");
5929                 if (sock->bound)
5930                         xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5931                                                   ISC_XMLCHAR "bound");
5932
5933                 xmlTextWriterEndElement(writer); /* states */
5934
5935                 xmlTextWriterEndElement(writer); /* socket */
5936
5937                 UNLOCK(&sock->lock);
5938                 sock = ISC_LIST_NEXT(sock, link);
5939         }
5940         xmlTextWriterEndElement(writer); /* sockets */
5941
5942         UNLOCK(&mgr->lock);
5943 }
5944 #endif /* HAVE_LIBXML2 */