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

   1 /*
   2  * Copyright (C) 2004-2014  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                                 dupped : 1;
 339
 340 #ifdef ISC_NET_RECVOVERFLOW
 341         unsigned char           overflow; /* used for MSG_TRUNC fake */
 342 #endif
 343
 344         char                    *recvcmsgbuf;
 345         ISC_SOCKADDR_LEN_T      recvcmsgbuflen;
 346         char                    *sendcmsgbuf;
 347         ISC_SOCKADDR_LEN_T      sendcmsgbuflen;
 348
 349         void                    *fdwatcharg;
 350         isc_sockfdwatch_t       fdwatchcb;
 351         int                     fdwatchflags;
 352         isc_task_t              *fdwatchtask;
 353 };
 354
 355 #define SOCKET_MANAGER_MAGIC    ISC_MAGIC('I', 'O', 'm', 'g')
 356 #define VALID_MANAGER(m)        ISC_MAGIC_VALID(m, SOCKET_MANAGER_MAGIC)
 357
 358 struct isc__socketmgr {
 359         /* Not locked. */
 360         isc_socketmgr_t         common;
 361         isc_mem_t              *mctx;
 362         isc_mutex_t             lock;
 363         isc_mutex_t             *fdlock;
 364         isc_stats_t             *stats;
 365 #ifdef USE_KQUEUE
 366         int                     kqueue_fd;
 367         int                     nevents;
 368         struct kevent           *events;
 369 #endif  /* USE_KQUEUE */
 370 #ifdef USE_EPOLL
 371         int                     epoll_fd;
 372         int                     nevents;
 373         struct epoll_event      *events;
 374 #endif  /* USE_EPOLL */
 375 #ifdef USE_DEVPOLL
 376         int                     devpoll_fd;
 377         int                     nevents;
 378         struct pollfd           *events;
 379 #endif  /* USE_DEVPOLL */
 380 #ifdef USE_SELECT
 381         int                     fd_bufsize;
 382 #endif  /* USE_SELECT */
 383         unsigned int            maxsocks;
 384 #ifdef ISC_PLATFORM_USETHREADS
 385         int                     pipe_fds[2];
 386 #endif
 387
 388         /* Locked by fdlock. */
 389         isc__socket_t          **fds;
 390         int                     *fdstate;
 391 #ifdef USE_DEVPOLL
 392         pollinfo_t              *fdpollinfo;
 393 #endif
 394
 395         /* Locked by manager lock. */
 396         ISC_LIST(isc__socket_t) socklist;
 397 #ifdef USE_SELECT
 398         fd_set                  *read_fds;
 399         fd_set                  *read_fds_copy;
 400         fd_set                  *write_fds;
 401         fd_set                  *write_fds_copy;
 402         int                     maxfd;
 403 #endif  /* USE_SELECT */
 404         int                     reserved;       /* unlocked */
 405 #ifdef USE_WATCHER_THREAD
 406         isc_thread_t            watcher;
 407         isc_condition_t         shutdown_ok;
 408 #else /* USE_WATCHER_THREAD */
 409         unsigned int            refs;
 410 #endif /* USE_WATCHER_THREAD */
 411         int                     maxudp;
 412 };
 413
 414 #ifdef USE_SHARED_MANAGER
 415 static isc__socketmgr_t *socketmgr = NULL;
 416 #endif /* USE_SHARED_MANAGER */
 417
 418 #define CLOSED                  0       /* this one must be zero */
 419 #define MANAGED                 1
 420 #define CLOSE_PENDING           2
 421
 422 /*
 423  * send() and recv() iovec counts
 424  */
 425 #define MAXSCATTERGATHER_SEND   (ISC_SOCKET_MAXSCATTERGATHER)
 426 #ifdef ISC_NET_RECVOVERFLOW
 427 # define MAXSCATTERGATHER_RECV  (ISC_SOCKET_MAXSCATTERGATHER + 1)
 428 #else
 429 # define MAXSCATTERGATHER_RECV  (ISC_SOCKET_MAXSCATTERGATHER)
 430 #endif
 431
 432 static isc_result_t socket_create(isc_socketmgr_t *manager0, int pf,
 433                                   isc_sockettype_t type,
 434                                   isc_socket_t **socketp,
 435                                   isc_socket_t *dup_socket);
 436 static void send_recvdone_event(isc__socket_t *, isc_socketevent_t **);
 437 static void send_senddone_event(isc__socket_t *, isc_socketevent_t **);
 438 static void free_socket(isc__socket_t **);
 439 static isc_result_t allocate_socket(isc__socketmgr_t *, isc_sockettype_t,
 440                                     isc__socket_t **);
 441 static void destroy(isc__socket_t **);
 442 static void internal_accept(isc_task_t *, isc_event_t *);
 443 static void internal_connect(isc_task_t *, isc_event_t *);
 444 static void internal_recv(isc_task_t *, isc_event_t *);
 445 static void internal_send(isc_task_t *, isc_event_t *);
 446 static void internal_fdwatch_write(isc_task_t *, isc_event_t *);
 447 static void internal_fdwatch_read(isc_task_t *, isc_event_t *);
 448 static void process_cmsg(isc__socket_t *, struct msghdr *, isc_socketevent_t *);
 449 static void build_msghdr_send(isc__socket_t *, isc_socketevent_t *,
 450                               struct msghdr *, struct iovec *, size_t *);
 451 static void build_msghdr_recv(isc__socket_t *, isc_socketevent_t *,
 452                               struct msghdr *, struct iovec *, size_t *);
 453 #ifdef USE_WATCHER_THREAD
 454 static isc_boolean_t process_ctlfd(isc__socketmgr_t *manager);
 455 #endif
 456
 457 /*%
 458  * The following can be either static or public, depending on build environment.
 459  */
 460
 461 #ifdef BIND9
 462 #define ISC_SOCKETFUNC_SCOPE
 463 #else
 464 #define ISC_SOCKETFUNC_SCOPE static
 465 #endif
 466
 467 ISC_SOCKETFUNC_SCOPE isc_result_t
 468 isc__socket_create(isc_socketmgr_t *manager, int pf, isc_sockettype_t type,
 469                    isc_socket_t **socketp);
 470 ISC_SOCKETFUNC_SCOPE void
 471 isc__socket_attach(isc_socket_t *sock, isc_socket_t **socketp);
 472 ISC_SOCKETFUNC_SCOPE void
 473 isc__socket_detach(isc_socket_t **socketp);
 474 ISC_SOCKETFUNC_SCOPE isc_result_t
 475 isc__socketmgr_create(isc_mem_t *mctx, isc_socketmgr_t **managerp);
 476 ISC_SOCKETFUNC_SCOPE isc_result_t
 477 isc__socketmgr_create2(isc_mem_t *mctx, isc_socketmgr_t **managerp,
 478                        unsigned int maxsocks);
 479 ISC_SOCKETFUNC_SCOPE void
 480 isc__socketmgr_destroy(isc_socketmgr_t **managerp);
 481 ISC_SOCKETFUNC_SCOPE isc_result_t
 482 isc__socket_recvv(isc_socket_t *sock, isc_bufferlist_t *buflist,
 483                  unsigned int minimum, isc_task_t *task,
 484                   isc_taskaction_t action, const void *arg);
 485 ISC_SOCKETFUNC_SCOPE isc_result_t
 486 isc__socket_recv(isc_socket_t *sock, isc_region_t *region,
 487                  unsigned int minimum, isc_task_t *task,
 488                  isc_taskaction_t action, const void *arg);
 489 ISC_SOCKETFUNC_SCOPE isc_result_t
 490 isc__socket_recv2(isc_socket_t *sock, isc_region_t *region,
 491                   unsigned int minimum, isc_task_t *task,
 492                   isc_socketevent_t *event, unsigned int flags);
 493 ISC_SOCKETFUNC_SCOPE isc_result_t
 494 isc__socket_send(isc_socket_t *sock, isc_region_t *region,
 495                  isc_task_t *task, isc_taskaction_t action, const void *arg);
 496 ISC_SOCKETFUNC_SCOPE isc_result_t
 497 isc__socket_sendto(isc_socket_t *sock, isc_region_t *region,
 498                    isc_task_t *task, isc_taskaction_t action, const void *arg,
 499                    isc_sockaddr_t *address, struct in6_pktinfo *pktinfo);
 500 ISC_SOCKETFUNC_SCOPE isc_result_t
 501 isc__socket_sendv(isc_socket_t *sock, isc_bufferlist_t *buflist,
 502                   isc_task_t *task, isc_taskaction_t action, const void *arg);
 503 ISC_SOCKETFUNC_SCOPE isc_result_t
 504 isc__socket_sendtov(isc_socket_t *sock, isc_bufferlist_t *buflist,
 505                     isc_task_t *task, isc_taskaction_t action, const void *arg,
 506                     isc_sockaddr_t *address, struct in6_pktinfo *pktinfo);
 507 ISC_SOCKETFUNC_SCOPE isc_result_t
 508 isc__socket_sendtov2(isc_socket_t *sock, isc_bufferlist_t *buflist,
 509                      isc_task_t *task, isc_taskaction_t action, const void *arg,
 510                      isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
 511                      unsigned int flags);
 512 ISC_SOCKETFUNC_SCOPE isc_result_t
 513 isc__socket_sendto2(isc_socket_t *sock, isc_region_t *region,
 514                     isc_task_t *task,
 515                     isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
 516                     isc_socketevent_t *event, unsigned int flags);
 517 ISC_SOCKETFUNC_SCOPE void
 518 isc__socket_cleanunix(isc_sockaddr_t *sockaddr, isc_boolean_t active);
 519 ISC_SOCKETFUNC_SCOPE isc_result_t
 520 isc__socket_permunix(isc_sockaddr_t *sockaddr, isc_uint32_t perm,
 521                      isc_uint32_t owner, isc_uint32_t group);
 522 ISC_SOCKETFUNC_SCOPE isc_result_t
 523 isc__socket_bind(isc_socket_t *sock, isc_sockaddr_t *sockaddr,
 524                  unsigned int options);
 525 ISC_SOCKETFUNC_SCOPE isc_result_t
 526 isc__socket_filter(isc_socket_t *sock, const char *filter);
 527 ISC_SOCKETFUNC_SCOPE isc_result_t
 528 isc__socket_listen(isc_socket_t *sock, unsigned int backlog);
 529 ISC_SOCKETFUNC_SCOPE isc_result_t
 530 isc__socket_accept(isc_socket_t *sock,
 531                    isc_task_t *task, isc_taskaction_t action, const void *arg);
 532 ISC_SOCKETFUNC_SCOPE isc_result_t
 533 isc__socket_connect(isc_socket_t *sock, isc_sockaddr_t *addr,
 534                     isc_task_t *task, isc_taskaction_t action,
 535                     const void *arg);
 536 ISC_SOCKETFUNC_SCOPE isc_result_t
 537 isc__socket_getpeername(isc_socket_t *sock, isc_sockaddr_t *addressp);
 538 ISC_SOCKETFUNC_SCOPE isc_result_t
 539 isc__socket_getsockname(isc_socket_t *sock, isc_sockaddr_t *addressp);
 540 ISC_SOCKETFUNC_SCOPE void
 541 isc__socket_cancel(isc_socket_t *sock, isc_task_t *task, unsigned int how);
 542 ISC_SOCKETFUNC_SCOPE isc_sockettype_t
 543 isc__socket_gettype(isc_socket_t *sock);
 544 ISC_SOCKETFUNC_SCOPE isc_boolean_t
 545 isc__socket_isbound(isc_socket_t *sock);
 546 ISC_SOCKETFUNC_SCOPE void
 547 isc__socket_ipv6only(isc_socket_t *sock, isc_boolean_t yes);
 548 #if defined(HAVE_LIBXML2) && defined(BIND9)
 549 ISC_SOCKETFUNC_SCOPE void
 550 isc__socketmgr_renderxml(isc_socketmgr_t *mgr0, xmlTextWriterPtr writer);
 551 #endif
 552
 553 ISC_SOCKETFUNC_SCOPE isc_result_t
 554 isc__socket_fdwatchcreate(isc_socketmgr_t *manager, int fd, int flags,
 555                           isc_sockfdwatch_t callback, void *cbarg,
 556                           isc_task_t *task, isc_socket_t **socketp);
 557 ISC_SOCKETFUNC_SCOPE isc_result_t
 558 isc__socket_fdwatchpoke(isc_socket_t *sock, int flags);
 559 ISC_SOCKETFUNC_SCOPE isc_result_t
 560 isc__socket_dup(isc_socket_t *sock, isc_socket_t **socketp);
 561 ISC_SOCKETFUNC_SCOPE int
 562 isc__socket_getfd(isc_socket_t *sock);
 563
 564 static struct {
 565         isc_socketmethods_t methods;
 566
 567         /*%
 568          * The following are defined just for avoiding unused static functions.
 569          */
 570 #ifndef BIND9
 571         void *recvv, *send, *sendv, *sendto2, *sendtov, *cleanunix, *permunix,
 572              *filter, *listen, *accept, *getpeername, *isbound;
 573 #endif
 574 } socketmethods = {
 575         {
 576                 isc__socket_attach,
 577                 isc__socket_detach,
 578                 isc__socket_bind,
 579                 isc__socket_sendto,
 580                 isc__socket_sendto2,
 581                 isc__socket_connect,
 582                 isc__socket_recv,
 583                 isc__socket_recv2,
 584                 isc__socket_cancel,
 585                 isc__socket_getsockname,
 586                 isc__socket_gettype,
 587                 isc__socket_ipv6only,
 588                 isc__socket_fdwatchpoke,
 589                 isc__socket_dup,
 590                 isc__socket_getfd
 591         }
 592 #ifndef BIND9
 593         ,
 594         (void *)isc__socket_recvv, (void *)isc__socket_send,
 595         (void *)isc__socket_sendv, (void *)isc__socket_sendto2,
 596         (void *)isc__socket_sendtov,
 597         (void *)isc__socket_cleanunix, (void *)isc__socket_permunix,
 598         (void *)isc__socket_filter, (void *)isc__socket_listen,
 599         (void *)isc__socket_accept, (void *)isc__socket_getpeername,
 600         (void *)isc__socket_isbound
 601 #endif
 602 };
 603
 604 static isc_socketmgrmethods_t socketmgrmethods = {
 605         isc__socketmgr_destroy,
 606         isc__socket_create,
 607         isc__socket_fdwatchcreate
 608 };
 609
 610 #define SELECT_POKE_SHUTDOWN            (-1)
 611 #define SELECT_POKE_NOTHING             (-2)
 612 #define SELECT_POKE_READ                (-3)
 613 #define SELECT_POKE_ACCEPT              (-3) /*%< Same as _READ */
 614 #define SELECT_POKE_WRITE               (-4)
 615 #define SELECT_POKE_CONNECT             (-4) /*%< Same as _WRITE */
 616 #define SELECT_POKE_CLOSE               (-5)
 617
 618 #define SOCK_DEAD(s)                    ((s)->references == 0)
 619
 620 /*%
 621  * Shortcut index arrays to get access to statistics counters.
 622  */
 623 enum {
 624         STATID_OPEN = 0,
 625         STATID_OPENFAIL = 1,
 626         STATID_CLOSE = 2,
 627         STATID_BINDFAIL = 3,
 628         STATID_CONNECTFAIL = 4,
 629         STATID_CONNECT = 5,
 630         STATID_ACCEPTFAIL = 6,
 631         STATID_ACCEPT = 7,
 632         STATID_SENDFAIL = 8,
 633         STATID_RECVFAIL = 9
 634 };
 635 static const isc_statscounter_t udp4statsindex[] = {
 636         isc_sockstatscounter_udp4open,
 637         isc_sockstatscounter_udp4openfail,
 638         isc_sockstatscounter_udp4close,
 639         isc_sockstatscounter_udp4bindfail,
 640         isc_sockstatscounter_udp4connectfail,
 641         isc_sockstatscounter_udp4connect,
 642         -1,
 643         -1,
 644         isc_sockstatscounter_udp4sendfail,
 645         isc_sockstatscounter_udp4recvfail
 646 };
 647 static const isc_statscounter_t udp6statsindex[] = {
 648         isc_sockstatscounter_udp6open,
 649         isc_sockstatscounter_udp6openfail,
 650         isc_sockstatscounter_udp6close,
 651         isc_sockstatscounter_udp6bindfail,
 652         isc_sockstatscounter_udp6connectfail,
 653         isc_sockstatscounter_udp6connect,
 654         -1,
 655         -1,
 656         isc_sockstatscounter_udp6sendfail,
 657         isc_sockstatscounter_udp6recvfail
 658 };
 659 static const isc_statscounter_t tcp4statsindex[] = {
 660         isc_sockstatscounter_tcp4open,
 661         isc_sockstatscounter_tcp4openfail,
 662         isc_sockstatscounter_tcp4close,
 663         isc_sockstatscounter_tcp4bindfail,
 664         isc_sockstatscounter_tcp4connectfail,
 665         isc_sockstatscounter_tcp4connect,
 666         isc_sockstatscounter_tcp4acceptfail,
 667         isc_sockstatscounter_tcp4accept,
 668         isc_sockstatscounter_tcp4sendfail,
 669         isc_sockstatscounter_tcp4recvfail
 670 };
 671 static const isc_statscounter_t tcp6statsindex[] = {
 672         isc_sockstatscounter_tcp6open,
 673         isc_sockstatscounter_tcp6openfail,
 674         isc_sockstatscounter_tcp6close,
 675         isc_sockstatscounter_tcp6bindfail,
 676         isc_sockstatscounter_tcp6connectfail,
 677         isc_sockstatscounter_tcp6connect,
 678         isc_sockstatscounter_tcp6acceptfail,
 679         isc_sockstatscounter_tcp6accept,
 680         isc_sockstatscounter_tcp6sendfail,
 681         isc_sockstatscounter_tcp6recvfail
 682 };
 683 static const isc_statscounter_t unixstatsindex[] = {
 684         isc_sockstatscounter_unixopen,
 685         isc_sockstatscounter_unixopenfail,
 686         isc_sockstatscounter_unixclose,
 687         isc_sockstatscounter_unixbindfail,
 688         isc_sockstatscounter_unixconnectfail,
 689         isc_sockstatscounter_unixconnect,
 690         isc_sockstatscounter_unixacceptfail,
 691         isc_sockstatscounter_unixaccept,
 692         isc_sockstatscounter_unixsendfail,
 693         isc_sockstatscounter_unixrecvfail
 694 };
 695 static const isc_statscounter_t fdwatchstatsindex[] = {
 696         -1,
 697         -1,
 698         isc_sockstatscounter_fdwatchclose,
 699         isc_sockstatscounter_fdwatchbindfail,
 700         isc_sockstatscounter_fdwatchconnectfail,
 701         isc_sockstatscounter_fdwatchconnect,
 702         -1,
 703         -1,
 704         isc_sockstatscounter_fdwatchsendfail,
 705         isc_sockstatscounter_fdwatchrecvfail
 706 };
 707
 708 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL) || \
 709     defined(USE_WATCHER_THREAD)
 710 static void
 711 manager_log(isc__socketmgr_t *sockmgr,
 712             isc_logcategory_t *category, isc_logmodule_t *module, int level,
 713             const char *fmt, ...) ISC_FORMAT_PRINTF(5, 6);
 714 static void
 715 manager_log(isc__socketmgr_t *sockmgr,
 716             isc_logcategory_t *category, isc_logmodule_t *module, int level,
 717             const char *fmt, ...)
 718 {
 719         char msgbuf[2048];
 720         va_list ap;
 721
 722         if (! isc_log_wouldlog(isc_lctx, level))
 723                 return;
 724
 725         va_start(ap, fmt);
 726         vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
 727         va_end(ap);
 728
 729         isc_log_write(isc_lctx, category, module, level,
 730                       "sockmgr %p: %s", sockmgr, msgbuf);
 731 }
 732 #endif
 733
 734 static void
 735 socket_log(isc__socket_t *sock, isc_sockaddr_t *address,
 736            isc_logcategory_t *category, isc_logmodule_t *module, int level,
 737            isc_msgcat_t *msgcat, int msgset, int message,
 738            const char *fmt, ...) ISC_FORMAT_PRINTF(9, 10);
 739 static void
 740 socket_log(isc__socket_t *sock, isc_sockaddr_t *address,
 741            isc_logcategory_t *category, isc_logmodule_t *module, int level,
 742            isc_msgcat_t *msgcat, int msgset, int message,
 743            const char *fmt, ...)
 744 {
 745         char msgbuf[2048];
 746         char peerbuf[ISC_SOCKADDR_FORMATSIZE];
 747         va_list ap;
 748
 749         if (! isc_log_wouldlog(isc_lctx, level))
 750                 return;
 751
 752         va_start(ap, fmt);
 753         vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
 754         va_end(ap);
 755
 756         if (address == NULL) {
 757                 isc_log_iwrite(isc_lctx, category, module, level,
 758                                msgcat, msgset, message,
 759                                "socket %p: %s", sock, msgbuf);
 760         } else {
 761                 isc_sockaddr_format(address, peerbuf, sizeof(peerbuf));
 762                 isc_log_iwrite(isc_lctx, category, module, level,
 763                                msgcat, msgset, message,
 764                                "socket %p %s: %s", sock, peerbuf, msgbuf);
 765         }
 766 }
 767
 768 #if defined(_AIX) && defined(ISC_NET_BSD44MSGHDR) && \
 769     defined(USE_CMSG) && defined(IPV6_RECVPKTINFO)
 770 /*
 771  * AIX has a kernel bug where IPV6_RECVPKTINFO gets cleared by
 772  * setting IPV6_V6ONLY.
 773  */
 774 static void
 775 FIX_IPV6_RECVPKTINFO(isc__socket_t *sock)
 776 {
 777         char strbuf[ISC_STRERRORSIZE];
 778         int on = 1;
 779
 780         if (sock->pf != AF_INET6 || sock->type != isc_sockettype_udp)
 781                 return;
 782
 783         if (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_RECVPKTINFO,
 784                        (void *)&on, sizeof(on)) < 0) {
 785
 786                 isc__strerror(errno, strbuf, sizeof(strbuf));
 787                 UNEXPECTED_ERROR(__FILE__, __LINE__,
 788                                  "setsockopt(%d, IPV6_RECVPKTINFO) "
 789                                  "%s: %s", sock->fd,
 790                                  isc_msgcat_get(isc_msgcat,
 791                                                 ISC_MSGSET_GENERAL,
 792                                                 ISC_MSG_FAILED,
 793                                                 "failed"),
 794                                  strbuf);
 795         }
 796 }
 797 #else
 798 #define FIX_IPV6_RECVPKTINFO(sock) (void)0
 799 #endif
 800
 801 /*%
 802  * Increment socket-related statistics counters.
 803  */
 804 static inline void
 805 inc_stats(isc_stats_t *stats, isc_statscounter_t counterid) {
 806         REQUIRE(counterid != -1);
 807
 808         if (stats != NULL)
 809                 isc_stats_increment(stats, counterid);
 810 }
 811
 812 static inline isc_result_t
 813 watch_fd(isc__socketmgr_t *manager, int fd, int msg) {
 814         isc_result_t result = ISC_R_SUCCESS;
 815
 816 #ifdef USE_KQUEUE
 817         struct kevent evchange;
 818
 819         memset(&evchange, 0, sizeof(evchange));
 820         if (msg == SELECT_POKE_READ)
 821                 evchange.filter = EVFILT_READ;
 822         else
 823                 evchange.filter = EVFILT_WRITE;
 824         evchange.flags = EV_ADD;
 825         evchange.ident = fd;
 826         if (kevent(manager->kqueue_fd, &evchange, 1, NULL, 0, NULL) != 0)
 827                 result = isc__errno2result(errno);
 828
 829         return (result);
 830 #elif defined(USE_EPOLL)
 831         struct epoll_event event;
 832
 833         if (msg == SELECT_POKE_READ)
 834                 event.events = EPOLLIN;
 835         else
 836                 event.events = EPOLLOUT;
 837         memset(&event.data, 0, sizeof(event.data));
 838         event.data.fd = fd;
 839         if (epoll_ctl(manager->epoll_fd, EPOLL_CTL_ADD, fd, &event) == -1 &&
 840             errno != EEXIST) {
 841                 result = isc__errno2result(errno);
 842         }
 843
 844         return (result);
 845 #elif defined(USE_DEVPOLL)
 846         struct pollfd pfd;
 847         int lockid = FDLOCK_ID(fd);
 848
 849         memset(&pfd, 0, sizeof(pfd));
 850         if (msg == SELECT_POKE_READ)
 851                 pfd.events = POLLIN;
 852         else
 853                 pfd.events = POLLOUT;
 854         pfd.fd = fd;
 855         pfd.revents = 0;
 856         LOCK(&manager->fdlock[lockid]);
 857         if (write(manager->devpoll_fd, &pfd, sizeof(pfd)) == -1)
 858                 result = isc__errno2result(errno);
 859         else {
 860                 if (msg == SELECT_POKE_READ)
 861                         manager->fdpollinfo[fd].want_read = 1;
 862                 else
 863                         manager->fdpollinfo[fd].want_write = 1;
 864         }
 865         UNLOCK(&manager->fdlock[lockid]);
 866
 867         return (result);
 868 #elif defined(USE_SELECT)
 869         LOCK(&manager->lock);
 870         if (msg == SELECT_POKE_READ)
 871                 FD_SET(fd, manager->read_fds);
 872         if (msg == SELECT_POKE_WRITE)
 873                 FD_SET(fd, manager->write_fds);
 874         UNLOCK(&manager->lock);
 875
 876         return (result);
 877 #endif
 878 }
 879
 880 static inline isc_result_t
 881 unwatch_fd(isc__socketmgr_t *manager, int fd, int msg) {
 882         isc_result_t result = ISC_R_SUCCESS;
 883
 884 #ifdef USE_KQUEUE
 885         struct kevent evchange;
 886
 887         memset(&evchange, 0, sizeof(evchange));
 888         if (msg == SELECT_POKE_READ)
 889                 evchange.filter = EVFILT_READ;
 890         else
 891                 evchange.filter = EVFILT_WRITE;
 892         evchange.flags = EV_DELETE;
 893         evchange.ident = fd;
 894         if (kevent(manager->kqueue_fd, &evchange, 1, NULL, 0, NULL) != 0)
 895                 result = isc__errno2result(errno);
 896
 897         return (result);
 898 #elif defined(USE_EPOLL)
 899         struct epoll_event event;
 900
 901         if (msg == SELECT_POKE_READ)
 902                 event.events = EPOLLIN;
 903         else
 904                 event.events = EPOLLOUT;
 905         memset(&event.data, 0, sizeof(event.data));
 906         event.data.fd = fd;
 907         if (epoll_ctl(manager->epoll_fd, EPOLL_CTL_DEL, fd, &event) == -1 &&
 908             errno != ENOENT) {
 909                 char strbuf[ISC_STRERRORSIZE];
 910                 isc__strerror(errno, strbuf, sizeof(strbuf));
 911                 UNEXPECTED_ERROR(__FILE__, __LINE__,
 912                                  "epoll_ctl(DEL), %d: %s", fd, strbuf);
 913                 result = ISC_R_UNEXPECTED;
 914         }
 915         return (result);
 916 #elif defined(USE_DEVPOLL)
 917         struct pollfd pfds[2];
 918         size_t writelen = sizeof(pfds[0]);
 919         int lockid = FDLOCK_ID(fd);
 920
 921         memset(pfds, 0, sizeof(pfds));
 922         pfds[0].events = POLLREMOVE;
 923         pfds[0].fd = fd;
 924
 925         /*
 926          * Canceling read or write polling via /dev/poll is tricky.  Since it
 927          * only provides a way of canceling per FD, we may need to re-poll the
 928          * socket for the other operation.
 929          */
 930         LOCK(&manager->fdlock[lockid]);
 931         if (msg == SELECT_POKE_READ &&
 932             manager->fdpollinfo[fd].want_write == 1) {
 933                 pfds[1].events = POLLOUT;
 934                 pfds[1].fd = fd;
 935                 writelen += sizeof(pfds[1]);
 936         }
 937         if (msg == SELECT_POKE_WRITE &&
 938             manager->fdpollinfo[fd].want_read == 1) {
 939                 pfds[1].events = POLLIN;
 940                 pfds[1].fd = fd;
 941                 writelen += sizeof(pfds[1]);
 942         }
 943
 944         if (write(manager->devpoll_fd, pfds, writelen) == -1)
 945                 result = isc__errno2result(errno);
 946         else {
 947                 if (msg == SELECT_POKE_READ)
 948                         manager->fdpollinfo[fd].want_read = 0;
 949                 else
 950                         manager->fdpollinfo[fd].want_write = 0;
 951         }
 952         UNLOCK(&manager->fdlock[lockid]);
 953
 954         return (result);
 955 #elif defined(USE_SELECT)
 956         LOCK(&manager->lock);
 957         if (msg == SELECT_POKE_READ)
 958                 FD_CLR(fd, manager->read_fds);
 959         else if (msg == SELECT_POKE_WRITE)
 960                 FD_CLR(fd, manager->write_fds);
 961         UNLOCK(&manager->lock);
 962
 963         return (result);
 964 #endif
 965 }
 966
 967 static void
 968 wakeup_socket(isc__socketmgr_t *manager, int fd, int msg) {
 969         isc_result_t result;
 970         int lockid = FDLOCK_ID(fd);
 971
 972         /*
 973          * This is a wakeup on a socket.  If the socket is not in the
 974          * process of being closed, start watching it for either reads
 975          * or writes.
 976          */
 977
 978         INSIST(fd >= 0 && fd < (int)manager->maxsocks);
 979
 980         if (msg == SELECT_POKE_CLOSE) {
 981                 /* No one should be updating fdstate, so no need to lock it */
 982                 INSIST(manager->fdstate[fd] == CLOSE_PENDING);
 983                 manager->fdstate[fd] = CLOSED;
 984                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
 985                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
 986                 (void)close(fd);
 987                 return;
 988         }
 989
 990         LOCK(&manager->fdlock[lockid]);
 991         if (manager->fdstate[fd] == CLOSE_PENDING) {
 992                 UNLOCK(&manager->fdlock[lockid]);
 993
 994                 /*
 995                  * We accept (and ignore) any error from unwatch_fd() as we are
 996                  * closing the socket, hoping it doesn't leave dangling state in
 997                  * the kernel.
 998                  * Note that unwatch_fd() must be called after releasing the
 999                  * fdlock; otherwise it could cause deadlock due to a lock order
1000                  * reversal.
1001                  */
1002                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
1003                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
1004                 return;
1005         }
1006         if (manager->fdstate[fd] != MANAGED) {
1007                 UNLOCK(&manager->fdlock[lockid]);
1008                 return;
1009         }
1010         UNLOCK(&manager->fdlock[lockid]);
1011
1012         /*
1013          * Set requested bit.
1014          */
1015         result = watch_fd(manager, fd, msg);
1016         if (result != ISC_R_SUCCESS) {
1017                 /*
1018                  * XXXJT: what should we do?  Ignoring the failure of watching
1019                  * a socket will make the application dysfunctional, but there
1020                  * seems to be no reasonable recovery process.
1021                  */
1022                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
1023                               ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
1024                               "failed to start watching FD (%d): %s",
1025                               fd, isc_result_totext(result));
1026         }
1027 }
1028
1029 #ifdef USE_WATCHER_THREAD
1030 /*
1031  * Poke the select loop when there is something for us to do.
1032  * The write is required (by POSIX) to complete.  That is, we
1033  * will not get partial writes.
1034  */
1035 static void
1036 select_poke(isc__socketmgr_t *mgr, int fd, int msg) {
1037         int cc;
1038         int buf[2];
1039         char strbuf[ISC_STRERRORSIZE];
1040
1041         buf[0] = fd;
1042         buf[1] = msg;
1043
1044         do {
1045                 cc = write(mgr->pipe_fds[1], buf, sizeof(buf));
1046 #ifdef ENOSR
1047                 /*
1048                  * Treat ENOSR as EAGAIN but loop slowly as it is
1049                  * unlikely to clear fast.
1050                  */
1051                 if (cc < 0 && errno == ENOSR) {
1052                         sleep(1);
1053                         errno = EAGAIN;
1054                 }
1055 #endif
1056         } while (cc < 0 && SOFT_ERROR(errno));
1057
1058         if (cc < 0) {
1059                 isc__strerror(errno, strbuf, sizeof(strbuf));
1060                 FATAL_ERROR(__FILE__, __LINE__,
1061                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
1062                                            ISC_MSG_WRITEFAILED,
1063                                            "write() failed "
1064                                            "during watcher poke: %s"),
1065                             strbuf);
1066         }
1067
1068         INSIST(cc == sizeof(buf));
1069 }
1070
1071 /*
1072  * Read a message on the internal fd.
1073  */
1074 static void
1075 select_readmsg(isc__socketmgr_t *mgr, int *fd, int *msg) {
1076         int buf[2];
1077         int cc;
1078         char strbuf[ISC_STRERRORSIZE];
1079
1080         cc = read(mgr->pipe_fds[0], buf, sizeof(buf));
1081         if (cc < 0) {
1082                 *msg = SELECT_POKE_NOTHING;
1083                 *fd = -1;       /* Silence compiler. */
1084                 if (SOFT_ERROR(errno))
1085                         return;
1086
1087                 isc__strerror(errno, strbuf, sizeof(strbuf));
1088                 FATAL_ERROR(__FILE__, __LINE__,
1089                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
1090                                            ISC_MSG_READFAILED,
1091                                            "read() failed "
1092                                            "during watcher poke: %s"),
1093                             strbuf);
1094
1095                 return;
1096         }
1097         INSIST(cc == sizeof(buf));
1098
1099         *fd = buf[0];
1100         *msg = buf[1];
1101 }
1102 #else /* USE_WATCHER_THREAD */
1103 /*
1104  * Update the state of the socketmgr when something changes.
1105  */
1106 static void
1107 select_poke(isc__socketmgr_t *manager, int fd, int msg) {
1108         if (msg == SELECT_POKE_SHUTDOWN)
1109                 return;
1110         else if (fd >= 0)
1111                 wakeup_socket(manager, fd, msg);
1112         return;
1113 }
1114 #endif /* USE_WATCHER_THREAD */
1115
1116 /*
1117  * Make a fd non-blocking.
1118  */
1119 static isc_result_t
1120 make_nonblock(int fd) {
1121         int ret;
1122         int flags;
1123         char strbuf[ISC_STRERRORSIZE];
1124 #ifdef USE_FIONBIO_IOCTL
1125         int on = 1;
1126
1127         ret = ioctl(fd, FIONBIO, (char *)&on);
1128 #else
1129         flags = fcntl(fd, F_GETFL, 0);
1130         flags |= PORT_NONBLOCK;
1131         ret = fcntl(fd, F_SETFL, flags);
1132 #endif
1133
1134         if (ret == -1) {
1135                 isc__strerror(errno, strbuf, sizeof(strbuf));
1136                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1137 #ifdef USE_FIONBIO_IOCTL
1138                                  "ioctl(%d, FIONBIO, &on): %s", fd,
1139 #else
1140                                  "fcntl(%d, F_SETFL, %d): %s", fd, flags,
1141 #endif
1142                                  strbuf);
1143
1144                 return (ISC_R_UNEXPECTED);
1145         }
1146
1147         return (ISC_R_SUCCESS);
1148 }
1149
1150 #ifdef USE_CMSG
1151 /*
1152  * Not all OSes support advanced CMSG macros: CMSG_LEN and CMSG_SPACE.
1153  * In order to ensure as much portability as possible, we provide wrapper
1154  * functions of these macros.
1155  * Note that cmsg_space() could run slow on OSes that do not have
1156  * CMSG_SPACE.
1157  */
1158 static inline ISC_SOCKADDR_LEN_T
1159 cmsg_len(ISC_SOCKADDR_LEN_T len) {
1160 #ifdef CMSG_LEN
1161         return (CMSG_LEN(len));
1162 #else
1163         ISC_SOCKADDR_LEN_T hdrlen;
1164
1165         /*
1166          * Cast NULL so that any pointer arithmetic performed by CMSG_DATA
1167          * is correct.
1168          */
1169         hdrlen = (ISC_SOCKADDR_LEN_T)CMSG_DATA(((struct cmsghdr *)NULL));
1170         return (hdrlen + len);
1171 #endif
1172 }
1173
1174 static inline ISC_SOCKADDR_LEN_T
1175 cmsg_space(ISC_SOCKADDR_LEN_T len) {
1176 #ifdef CMSG_SPACE
1177         return (CMSG_SPACE(len));
1178 #else
1179         struct msghdr msg;
1180         struct cmsghdr *cmsgp;
1181         /*
1182          * XXX: The buffer length is an ad-hoc value, but should be enough
1183          * in a practical sense.
1184          */
1185         char dummybuf[sizeof(struct cmsghdr) + 1024];
1186
1187         memset(&msg, 0, sizeof(msg));
1188         msg.msg_control = dummybuf;
1189         msg.msg_controllen = sizeof(dummybuf);
1190
1191         cmsgp = (struct cmsghdr *)dummybuf;
1192         cmsgp->cmsg_len = cmsg_len(len);
1193
1194         cmsgp = CMSG_NXTHDR(&msg, cmsgp);
1195         if (cmsgp != NULL)
1196                 return ((char *)cmsgp - (char *)msg.msg_control);
1197         else
1198                 return (0);
1199 #endif
1200 }
1201 #endif /* USE_CMSG */
1202
1203 /*
1204  * Process control messages received on a socket.
1205  */
1206 static void
1207 process_cmsg(isc__socket_t *sock, struct msghdr *msg, isc_socketevent_t *dev) {
1208 #ifdef USE_CMSG
1209         struct cmsghdr *cmsgp;
1210 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1211         struct in6_pktinfo *pktinfop;
1212 #endif
1213 #ifdef SO_TIMESTAMP
1214         void *timevalp;
1215 #endif
1216 #endif
1217
1218         /*
1219          * sock is used only when ISC_NET_BSD44MSGHDR and USE_CMSG are defined.
1220          * msg and dev are used only when ISC_NET_BSD44MSGHDR is defined.
1221          * They are all here, outside of the CPP tests, because it is
1222          * more consistent with the usual ISC coding style.
1223          */
1224         UNUSED(sock);
1225         UNUSED(msg);
1226         UNUSED(dev);
1227
1228 #ifdef ISC_NET_BSD44MSGHDR
1229
1230 #ifdef MSG_TRUNC
1231         if ((msg->msg_flags & MSG_TRUNC) == MSG_TRUNC)
1232                 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
1233 #endif
1234
1235 #ifdef MSG_CTRUNC
1236         if ((msg->msg_flags & MSG_CTRUNC) == MSG_CTRUNC)
1237                 dev->attributes |= ISC_SOCKEVENTATTR_CTRUNC;
1238 #endif
1239
1240 #ifndef USE_CMSG
1241         return;
1242 #else
1243         if (msg->msg_controllen == 0U || msg->msg_control == NULL)
1244                 return;
1245
1246 #ifdef SO_TIMESTAMP
1247         timevalp = NULL;
1248 #endif
1249 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1250         pktinfop = NULL;
1251 #endif
1252
1253         cmsgp = CMSG_FIRSTHDR(msg);
1254         while (cmsgp != NULL) {
1255                 socket_log(sock, NULL, TRACE,
1256                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PROCESSCMSG,
1257                            "processing cmsg %p", cmsgp);
1258
1259 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1260                 if (cmsgp->cmsg_level == IPPROTO_IPV6
1261                     && cmsgp->cmsg_type == IPV6_PKTINFO) {
1262
1263                         pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
1264                         memmove(&dev->pktinfo, pktinfop,
1265                                 sizeof(struct in6_pktinfo));
1266                         dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
1267                         socket_log(sock, NULL, TRACE,
1268                                    isc_msgcat, ISC_MSGSET_SOCKET,
1269                                    ISC_MSG_IFRECEIVED,
1270                                    "interface received on ifindex %u",
1271                                    dev->pktinfo.ipi6_ifindex);
1272                         if (IN6_IS_ADDR_MULTICAST(&pktinfop->ipi6_addr))
1273                                 dev->attributes |= ISC_SOCKEVENTATTR_MULTICAST;
1274                         goto next;
1275                 }
1276 #endif
1277
1278 #ifdef SO_TIMESTAMP
1279                 if (cmsgp->cmsg_level == SOL_SOCKET
1280                     && cmsgp->cmsg_type == SCM_TIMESTAMP) {
1281                         struct timeval tv;
1282                         timevalp = CMSG_DATA(cmsgp);
1283                         memmove(&tv, timevalp, sizeof(tv));
1284                         dev->timestamp.seconds = tv.tv_sec;
1285                         dev->timestamp.nanoseconds = tv.tv_usec * 1000;
1286                         dev->attributes |= ISC_SOCKEVENTATTR_TIMESTAMP;
1287                         goto next;
1288                 }
1289 #endif
1290
1291         next:
1292                 cmsgp = CMSG_NXTHDR(msg, cmsgp);
1293         }
1294 #endif /* USE_CMSG */
1295
1296 #endif /* ISC_NET_BSD44MSGHDR */
1297 }
1298
1299 /*
1300  * Construct an iov array and attach it to the msghdr passed in.  This is
1301  * the SEND constructor, which will use the used region of the buffer
1302  * (if using a buffer list) or will use the internal region (if a single
1303  * buffer I/O is requested).
1304  *
1305  * Nothing can be NULL, and the done event must list at least one buffer
1306  * on the buffer linked list for this function to be meaningful.
1307  *
1308  * If write_countp != NULL, *write_countp will hold the number of bytes
1309  * this transaction can send.
1310  */
1311 static void
1312 build_msghdr_send(isc__socket_t *sock, isc_socketevent_t *dev,
1313                   struct msghdr *msg, struct iovec *iov, size_t *write_countp)
1314 {
1315         unsigned int iovcount;
1316         isc_buffer_t *buffer;
1317         isc_region_t used;
1318         size_t write_count;
1319         size_t skip_count;
1320
1321         memset(msg, 0, sizeof(*msg));
1322
1323         if (!sock->connected) {
1324                 msg->msg_name = (void *)&dev->address.type.sa;
1325                 msg->msg_namelen = dev->address.length;
1326         } else {
1327                 msg->msg_name = NULL;
1328                 msg->msg_namelen = 0;
1329         }
1330
1331         buffer = ISC_LIST_HEAD(dev->bufferlist);
1332         write_count = 0;
1333         iovcount = 0;
1334
1335         /*
1336          * Single buffer I/O?  Skip what we've done so far in this region.
1337          */
1338         if (buffer == NULL) {
1339                 write_count = dev->region.length - dev->n;
1340                 iov[0].iov_base = (void *)(dev->region.base + dev->n);
1341                 iov[0].iov_len = write_count;
1342                 iovcount = 1;
1343
1344                 goto config;
1345         }
1346
1347         /*
1348          * Multibuffer I/O.
1349          * Skip the data in the buffer list that we have already written.
1350          */
1351         skip_count = dev->n;
1352         while (buffer != NULL) {
1353                 REQUIRE(ISC_BUFFER_VALID(buffer));
1354                 if (skip_count < isc_buffer_usedlength(buffer))
1355                         break;
1356                 skip_count -= isc_buffer_usedlength(buffer);
1357                 buffer = ISC_LIST_NEXT(buffer, link);
1358         }
1359
1360         while (buffer != NULL) {
1361                 INSIST(iovcount < MAXSCATTERGATHER_SEND);
1362
1363                 isc_buffer_usedregion(buffer, &used);
1364
1365                 if (used.length > 0) {
1366                         iov[iovcount].iov_base = (void *)(used.base
1367                                                           + skip_count);
1368                         iov[iovcount].iov_len = used.length - skip_count;
1369                         write_count += (used.length - skip_count);
1370                         skip_count = 0;
1371                         iovcount++;
1372                 }
1373                 buffer = ISC_LIST_NEXT(buffer, link);
1374         }
1375
1376         INSIST(skip_count == 0U);
1377
1378  config:
1379         msg->msg_iov = iov;
1380         msg->msg_iovlen = iovcount;
1381
1382 #ifdef ISC_NET_BSD44MSGHDR
1383         msg->msg_control = NULL;
1384         msg->msg_controllen = 0;
1385         msg->msg_flags = 0;
1386 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
1387         if ((sock->type == isc_sockettype_udp)
1388             && ((dev->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0)) {
1389 #if defined(IPV6_USE_MIN_MTU)
1390                 int use_min_mtu = 1;    /* -1, 0, 1 */
1391 #endif
1392                 struct cmsghdr *cmsgp;
1393                 struct in6_pktinfo *pktinfop;
1394
1395                 socket_log(sock, NULL, TRACE,
1396                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_SENDTODATA,
1397                            "sendto pktinfo data, ifindex %u",
1398                            dev->pktinfo.ipi6_ifindex);
1399
1400                 msg->msg_controllen = cmsg_space(sizeof(struct in6_pktinfo));
1401                 INSIST(msg->msg_controllen <= sock->sendcmsgbuflen);
1402                 msg->msg_control = (void *)sock->sendcmsgbuf;
1403
1404                 cmsgp = (struct cmsghdr *)sock->sendcmsgbuf;
1405                 cmsgp->cmsg_level = IPPROTO_IPV6;
1406                 cmsgp->cmsg_type = IPV6_PKTINFO;
1407                 cmsgp->cmsg_len = cmsg_len(sizeof(struct in6_pktinfo));
1408                 pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
1409                 memmove(pktinfop, &dev->pktinfo, sizeof(struct in6_pktinfo));
1410 #if defined(IPV6_USE_MIN_MTU)
1411                 /*
1412                  * Set IPV6_USE_MIN_MTU as a per packet option as FreeBSD
1413                  * ignores setsockopt(IPV6_USE_MIN_MTU) when IPV6_PKTINFO
1414                  * is used.
1415                  */
1416                 cmsgp = (struct cmsghdr *)(sock->sendcmsgbuf +
1417                                            msg->msg_controllen);
1418                 msg->msg_controllen += cmsg_space(sizeof(use_min_mtu));
1419                 INSIST(msg->msg_controllen <= sock->sendcmsgbuflen);
1420
1421                 cmsgp->cmsg_level = IPPROTO_IPV6;
1422                 cmsgp->cmsg_type = IPV6_USE_MIN_MTU;
1423                 cmsgp->cmsg_len = cmsg_len(sizeof(use_min_mtu));
1424                 memmove(CMSG_DATA(cmsgp), &use_min_mtu, sizeof(use_min_mtu));
1425 #endif
1426         }
1427 #endif /* USE_CMSG && ISC_PLATFORM_HAVEIPV6 */
1428 #else /* ISC_NET_BSD44MSGHDR */
1429         msg->msg_accrights = NULL;
1430         msg->msg_accrightslen = 0;
1431 #endif /* ISC_NET_BSD44MSGHDR */
1432
1433         if (write_countp != NULL)
1434                 *write_countp = write_count;
1435 }
1436
1437 /*
1438  * Construct an iov array and attach it to the msghdr passed in.  This is
1439  * the RECV constructor, which will use the available region of the buffer
1440  * (if using a buffer list) or will use the internal region (if a single
1441  * buffer I/O is requested).
1442  *
1443  * Nothing can be NULL, and the done event must list at least one buffer
1444  * on the buffer linked list for this function to be meaningful.
1445  *
1446  * If read_countp != NULL, *read_countp will hold the number of bytes
1447  * this transaction can receive.
1448  */
1449 static void
1450 build_msghdr_recv(isc__socket_t *sock, isc_socketevent_t *dev,
1451                   struct msghdr *msg, struct iovec *iov, size_t *read_countp)
1452 {
1453         unsigned int iovcount;
1454         isc_buffer_t *buffer;
1455         isc_region_t available;
1456         size_t read_count;
1457
1458         memset(msg, 0, sizeof(struct msghdr));
1459
1460         if (sock->type == isc_sockettype_udp) {
1461                 memset(&dev->address, 0, sizeof(dev->address));
1462 #ifdef BROKEN_RECVMSG
1463                 if (sock->pf == AF_INET) {
1464                         msg->msg_name = (void *)&dev->address.type.sin;
1465                         msg->msg_namelen = sizeof(dev->address.type.sin6);
1466                 } else if (sock->pf == AF_INET6) {
1467                         msg->msg_name = (void *)&dev->address.type.sin6;
1468                         msg->msg_namelen = sizeof(dev->address.type.sin6);
1469 #ifdef ISC_PLATFORM_HAVESYSUNH
1470                 } else if (sock->pf == AF_UNIX) {
1471                         msg->msg_name = (void *)&dev->address.type.sunix;
1472                         msg->msg_namelen = sizeof(dev->address.type.sunix);
1473 #endif
1474                 } else {
1475                         msg->msg_name = (void *)&dev->address.type.sa;
1476                         msg->msg_namelen = sizeof(dev->address.type);
1477                 }
1478 #else
1479                 msg->msg_name = (void *)&dev->address.type.sa;
1480                 msg->msg_namelen = sizeof(dev->address.type);
1481 #endif
1482 #ifdef ISC_NET_RECVOVERFLOW
1483                 /* If needed, steal one iovec for overflow detection. */
1484                 maxiov--;
1485 #endif
1486         } else { /* TCP */
1487                 msg->msg_name = NULL;
1488                 msg->msg_namelen = 0;
1489                 dev->address = sock->peer_address;
1490         }
1491
1492         buffer = ISC_LIST_HEAD(dev->bufferlist);
1493         read_count = 0;
1494
1495         /*
1496          * Single buffer I/O?  Skip what we've done so far in this region.
1497          */
1498         if (buffer == NULL) {
1499                 read_count = dev->region.length - dev->n;
1500                 iov[0].iov_base = (void *)(dev->region.base + dev->n);
1501                 iov[0].iov_len = read_count;
1502                 iovcount = 1;
1503
1504                 goto config;
1505         }
1506
1507         /*
1508          * Multibuffer I/O.
1509          * Skip empty buffers.
1510          */
1511         while (buffer != NULL) {
1512                 REQUIRE(ISC_BUFFER_VALID(buffer));
1513                 if (isc_buffer_availablelength(buffer) != 0)
1514                         break;
1515                 buffer = ISC_LIST_NEXT(buffer, link);
1516         }
1517
1518         iovcount = 0;
1519         while (buffer != NULL) {
1520                 INSIST(iovcount < MAXSCATTERGATHER_RECV);
1521
1522                 isc_buffer_availableregion(buffer, &available);
1523
1524                 if (available.length > 0) {
1525                         iov[iovcount].iov_base = (void *)(available.base);
1526                         iov[iovcount].iov_len = available.length;
1527                         read_count += available.length;
1528                         iovcount++;
1529                 }
1530                 buffer = ISC_LIST_NEXT(buffer, link);
1531         }
1532
1533  config:
1534
1535         /*
1536          * If needed, set up to receive that one extra byte.  Note that
1537          * we know there is at least one iov left, since we stole it
1538          * at the top of this function.
1539          */
1540 #ifdef ISC_NET_RECVOVERFLOW
1541         if (sock->type == isc_sockettype_udp) {
1542                 iov[iovcount].iov_base = (void *)(&sock->overflow);
1543                 iov[iovcount].iov_len = 1;
1544                 iovcount++;
1545         }
1546 #endif
1547
1548         msg->msg_iov = iov;
1549         msg->msg_iovlen = iovcount;
1550
1551 #ifdef ISC_NET_BSD44MSGHDR
1552         msg->msg_control = NULL;
1553         msg->msg_controllen = 0;
1554         msg->msg_flags = 0;
1555 #if defined(USE_CMSG)
1556         if (sock->type == isc_sockettype_udp) {
1557                 msg->msg_control = sock->recvcmsgbuf;
1558                 msg->msg_controllen = sock->recvcmsgbuflen;
1559         }
1560 #endif /* USE_CMSG */
1561 #else /* ISC_NET_BSD44MSGHDR */
1562         msg->msg_accrights = NULL;
1563         msg->msg_accrightslen = 0;
1564 #endif /* ISC_NET_BSD44MSGHDR */
1565
1566         if (read_countp != NULL)
1567                 *read_countp = read_count;
1568 }
1569
1570 static void
1571 set_dev_address(isc_sockaddr_t *address, isc__socket_t *sock,
1572                 isc_socketevent_t *dev)
1573 {
1574         if (sock->type == isc_sockettype_udp) {
1575                 if (address != NULL)
1576                         dev->address = *address;
1577                 else
1578                         dev->address = sock->peer_address;
1579         } else if (sock->type == isc_sockettype_tcp) {
1580                 INSIST(address == NULL);
1581                 dev->address = sock->peer_address;
1582         }
1583 }
1584
1585 static void
1586 destroy_socketevent(isc_event_t *event) {
1587         isc_socketevent_t *ev = (isc_socketevent_t *)event;
1588
1589         INSIST(ISC_LIST_EMPTY(ev->bufferlist));
1590
1591         (ev->destroy)(event);
1592 }
1593
1594 static isc_socketevent_t *
1595 allocate_socketevent(isc__socket_t *sock, isc_eventtype_t eventtype,
1596                      isc_taskaction_t action, const void *arg)
1597 {
1598         isc_socketevent_t *ev;
1599
1600         ev = (isc_socketevent_t *)isc_event_allocate(sock->manager->mctx,
1601                                                      sock, eventtype,
1602                                                      action, arg,
1603                                                      sizeof(*ev));
1604
1605         if (ev == NULL)
1606                 return (NULL);
1607
1608         ev->result = ISC_R_UNSET;
1609         ISC_LINK_INIT(ev, ev_link);
1610         ISC_LIST_INIT(ev->bufferlist);
1611         ev->region.base = NULL;
1612         ev->n = 0;
1613         ev->offset = 0;
1614         ev->attributes = 0;
1615         ev->destroy = ev->ev_destroy;
1616         ev->ev_destroy = destroy_socketevent;
1617
1618         return (ev);
1619 }
1620
1621 #if defined(ISC_SOCKET_DEBUG)
1622 static void
1623 dump_msg(struct msghdr *msg) {
1624         unsigned int i;
1625
1626         printf("MSGHDR %p\n", msg);
1627         printf("\tname %p, namelen %ld\n", msg->msg_name,
1628                (long) msg->msg_namelen);
1629         printf("\tiov %p, iovlen %ld\n", msg->msg_iov,
1630                (long) msg->msg_iovlen);
1631         for (i = 0; i < (unsigned int)msg->msg_iovlen; i++)
1632                 printf("\t\t%d\tbase %p, len %ld\n", i,
1633                        msg->msg_iov[i].iov_base,
1634                        (long) msg->msg_iov[i].iov_len);
1635 #ifdef ISC_NET_BSD44MSGHDR
1636         printf("\tcontrol %p, controllen %ld\n", msg->msg_control,
1637                (long) msg->msg_controllen);
1638 #endif
1639 }
1640 #endif
1641
1642 #define DOIO_SUCCESS            0       /* i/o ok, event sent */
1643 #define DOIO_SOFT               1       /* i/o ok, soft error, no event sent */
1644 #define DOIO_HARD               2       /* i/o error, event sent */
1645 #define DOIO_EOF                3       /* EOF, no event sent */
1646
1647 static int
1648 doio_recv(isc__socket_t *sock, isc_socketevent_t *dev) {
1649         int cc;
1650         struct iovec iov[MAXSCATTERGATHER_RECV];
1651         size_t read_count;
1652         size_t actual_count;
1653         struct msghdr msghdr;
1654         isc_buffer_t *buffer;
1655         int recv_errno;
1656         char strbuf[ISC_STRERRORSIZE];
1657
1658         build_msghdr_recv(sock, dev, &msghdr, iov, &read_count);
1659
1660 #if defined(ISC_SOCKET_DEBUG)
1661         dump_msg(&msghdr);
1662 #endif
1663
1664         cc = recvmsg(sock->fd, &msghdr, 0);
1665         recv_errno = errno;
1666
1667 #if defined(ISC_SOCKET_DEBUG)
1668         dump_msg(&msghdr);
1669 #endif
1670
1671         if (cc < 0) {
1672                 if (SOFT_ERROR(recv_errno))
1673                         return (DOIO_SOFT);
1674
1675                 if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
1676                         isc__strerror(recv_errno, strbuf, sizeof(strbuf));
1677                         socket_log(sock, NULL, IOEVENT,
1678                                    isc_msgcat, ISC_MSGSET_SOCKET,
1679                                    ISC_MSG_DOIORECV,
1680                                   "doio_recv: recvmsg(%d) %d bytes, err %d/%s",
1681                                    sock->fd, cc, recv_errno, strbuf);
1682                 }
1683
1684 #define SOFT_OR_HARD(_system, _isc) \
1685         if (recv_errno == _system) { \
1686                 if (sock->connected) { \
1687                         dev->result = _isc; \
1688                         inc_stats(sock->manager->stats, \
1689                                   sock->statsindex[STATID_RECVFAIL]); \
1690                         return (DOIO_HARD); \
1691                 } \
1692                 return (DOIO_SOFT); \
1693         }
1694 #define ALWAYS_HARD(_system, _isc) \
1695         if (recv_errno == _system) { \
1696                 dev->result = _isc; \
1697                 inc_stats(sock->manager->stats, \
1698                           sock->statsindex[STATID_RECVFAIL]); \
1699                 return (DOIO_HARD); \
1700         }
1701
1702                 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1703                 SOFT_OR_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1704                 SOFT_OR_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1705                 SOFT_OR_HARD(EHOSTDOWN, ISC_R_HOSTDOWN);
1706                 /* HPUX 11.11 can return EADDRNOTAVAIL. */
1707                 SOFT_OR_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1708                 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1709                 /* Should never get this one but it was seen. */
1710 #ifdef ENOPROTOOPT
1711                 SOFT_OR_HARD(ENOPROTOOPT, ISC_R_HOSTUNREACH);
1712 #endif
1713                 /*
1714                  * HPUX returns EPROTO and EINVAL on receiving some ICMP/ICMPv6
1715                  * errors.
1716                  */
1717 #ifdef EPROTO
1718                 SOFT_OR_HARD(EPROTO, ISC_R_HOSTUNREACH);
1719 #endif
1720                 SOFT_OR_HARD(EINVAL, ISC_R_HOSTUNREACH);
1721
1722 #undef SOFT_OR_HARD
1723 #undef ALWAYS_HARD
1724
1725                 dev->result = isc__errno2result(recv_errno);
1726                 inc_stats(sock->manager->stats,
1727                           sock->statsindex[STATID_RECVFAIL]);
1728                 return (DOIO_HARD);
1729         }
1730
1731         /*
1732          * On TCP and UNIX sockets, zero length reads indicate EOF,
1733          * while on UDP sockets, zero length reads are perfectly valid,
1734          * although strange.
1735          */
1736         switch (sock->type) {
1737         case isc_sockettype_tcp:
1738         case isc_sockettype_unix:
1739                 if (cc == 0)
1740                         return (DOIO_EOF);
1741                 break;
1742         case isc_sockettype_udp:
1743                 break;
1744         case isc_sockettype_fdwatch:
1745         default:
1746                 INSIST(0);
1747         }
1748
1749         if (sock->type == isc_sockettype_udp) {
1750                 dev->address.length = msghdr.msg_namelen;
1751                 if (isc_sockaddr_getport(&dev->address) == 0) {
1752                         if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
1753                                 socket_log(sock, &dev->address, IOEVENT,
1754                                            isc_msgcat, ISC_MSGSET_SOCKET,
1755                                            ISC_MSG_ZEROPORT,
1756                                            "dropping source port zero packet");
1757                         }
1758                         return (DOIO_SOFT);
1759                 }
1760                 /*
1761                  * Simulate a firewall blocking UDP responses bigger than
1762                  * 512 bytes.
1763                  */
1764                 if (sock->manager->maxudp != 0 && cc > sock->manager->maxudp)
1765                         return (DOIO_SOFT);
1766         }
1767
1768         socket_log(sock, &dev->address, IOEVENT,
1769                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PKTRECV,
1770                    "packet received correctly");
1771
1772         /*
1773          * Overflow bit detection.  If we received MORE bytes than we should,
1774          * this indicates an overflow situation.  Set the flag in the
1775          * dev entry and adjust how much we read by one.
1776          */
1777 #ifdef ISC_NET_RECVOVERFLOW
1778         if ((sock->type == isc_sockettype_udp) && ((size_t)cc > read_count)) {
1779                 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
1780                 cc--;
1781         }
1782 #endif
1783
1784         /*
1785          * If there are control messages attached, run through them and pull
1786          * out the interesting bits.
1787          */
1788         if (sock->type == isc_sockettype_udp)
1789                 process_cmsg(sock, &msghdr, dev);
1790
1791         /*
1792          * update the buffers (if any) and the i/o count
1793          */
1794         dev->n += cc;
1795         actual_count = cc;
1796         buffer = ISC_LIST_HEAD(dev->bufferlist);
1797         while (buffer != NULL && actual_count > 0U) {
1798                 REQUIRE(ISC_BUFFER_VALID(buffer));
1799                 if (isc_buffer_availablelength(buffer) <= actual_count) {
1800                         actual_count -= isc_buffer_availablelength(buffer);
1801                         isc_buffer_add(buffer,
1802                                        isc_buffer_availablelength(buffer));
1803                 } else {
1804                         isc_buffer_add(buffer, actual_count);
1805                         actual_count = 0;
1806                         POST(actual_count);
1807                         break;
1808                 }
1809                 buffer = ISC_LIST_NEXT(buffer, link);
1810                 if (buffer == NULL) {
1811                         INSIST(actual_count == 0U);
1812                 }
1813         }
1814
1815         /*
1816          * If we read less than we expected, update counters,
1817          * and let the upper layer poke the descriptor.
1818          */
1819         if (((size_t)cc != read_count) && (dev->n < dev->minimum))
1820                 return (DOIO_SOFT);
1821
1822         /*
1823          * Full reads are posted, or partials if partials are ok.
1824          */
1825         dev->result = ISC_R_SUCCESS;
1826         return (DOIO_SUCCESS);
1827 }
1828
1829 /*
1830  * Returns:
1831  *      DOIO_SUCCESS    The operation succeeded.  dev->result contains
1832  *                      ISC_R_SUCCESS.
1833  *
1834  *      DOIO_HARD       A hard or unexpected I/O error was encountered.
1835  *                      dev->result contains the appropriate error.
1836  *
1837  *      DOIO_SOFT       A soft I/O error was encountered.  No senddone
1838  *                      event was sent.  The operation should be retried.
1839  *
1840  *      No other return values are possible.
1841  */
1842 static int
1843 doio_send(isc__socket_t *sock, isc_socketevent_t *dev) {
1844         int cc;
1845         struct iovec iov[MAXSCATTERGATHER_SEND];
1846         size_t write_count;
1847         struct msghdr msghdr;
1848         char addrbuf[ISC_SOCKADDR_FORMATSIZE];
1849         int attempts = 0;
1850         int send_errno;
1851         char strbuf[ISC_STRERRORSIZE];
1852
1853         build_msghdr_send(sock, dev, &msghdr, iov, &write_count);
1854
1855  resend:
1856         cc = sendmsg(sock->fd, &msghdr, 0);
1857         send_errno = errno;
1858
1859         /*
1860          * Check for error or block condition.
1861          */
1862         if (cc < 0) {
1863                 if (send_errno == EINTR && ++attempts < NRETRIES)
1864                         goto resend;
1865
1866                 if (SOFT_ERROR(send_errno))
1867                         return (DOIO_SOFT);
1868
1869 #define SOFT_OR_HARD(_system, _isc) \
1870         if (send_errno == _system) { \
1871                 if (sock->connected) { \
1872                         dev->result = _isc; \
1873                         inc_stats(sock->manager->stats, \
1874                                   sock->statsindex[STATID_SENDFAIL]); \
1875                         return (DOIO_HARD); \
1876                 } \
1877                 return (DOIO_SOFT); \
1878         }
1879 #define ALWAYS_HARD(_system, _isc) \
1880         if (send_errno == _system) { \
1881                 dev->result = _isc; \
1882                 inc_stats(sock->manager->stats, \
1883                           sock->statsindex[STATID_SENDFAIL]); \
1884                 return (DOIO_HARD); \
1885         }
1886
1887                 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1888                 ALWAYS_HARD(EACCES, ISC_R_NOPERM);
1889                 ALWAYS_HARD(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
1890                 ALWAYS_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1891                 ALWAYS_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1892 #ifdef EHOSTDOWN
1893                 ALWAYS_HARD(EHOSTDOWN, ISC_R_HOSTUNREACH);
1894 #endif
1895                 ALWAYS_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1896                 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1897                 ALWAYS_HARD(EPERM, ISC_R_HOSTUNREACH);
1898                 ALWAYS_HARD(EPIPE, ISC_R_NOTCONNECTED);
1899                 ALWAYS_HARD(ECONNRESET, ISC_R_CONNECTIONRESET);
1900
1901 #undef SOFT_OR_HARD
1902 #undef ALWAYS_HARD
1903
1904                 /*
1905                  * The other error types depend on whether or not the
1906                  * socket is UDP or TCP.  If it is UDP, some errors
1907                  * that we expect to be fatal under TCP are merely
1908                  * annoying, and are really soft errors.
1909                  *
1910                  * However, these soft errors are still returned as
1911                  * a status.
1912                  */
1913                 isc_sockaddr_format(&dev->address, addrbuf, sizeof(addrbuf));
1914                 isc__strerror(send_errno, strbuf, sizeof(strbuf));
1915                 UNEXPECTED_ERROR(__FILE__, __LINE__, "internal_send: %s: %s",
1916                                  addrbuf, strbuf);
1917                 dev->result = isc__errno2result(send_errno);
1918                 inc_stats(sock->manager->stats,
1919                           sock->statsindex[STATID_SENDFAIL]);
1920                 return (DOIO_HARD);
1921         }
1922
1923         if (cc == 0) {
1924                 inc_stats(sock->manager->stats,
1925                           sock->statsindex[STATID_SENDFAIL]);
1926                 UNEXPECTED_ERROR(__FILE__, __LINE__,
1927                                  "doio_send: send() %s 0",
1928                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1929                                                 ISC_MSG_RETURNED, "returned"));
1930         }
1931
1932         /*
1933          * If we write less than we expected, update counters, poke.
1934          */
1935         dev->n += cc;
1936         if ((size_t)cc != write_count)
1937                 return (DOIO_SOFT);
1938
1939         /*
1940          * Exactly what we wanted to write.  We're done with this
1941          * entry.  Post its completion event.
1942          */
1943         dev->result = ISC_R_SUCCESS;
1944         return (DOIO_SUCCESS);
1945 }
1946
1947 /*
1948  * Kill.
1949  *
1950  * Caller must ensure that the socket is not locked and no external
1951  * references exist.
1952  */
1953 static void
1954 closesocket(isc__socketmgr_t *manager, isc__socket_t *sock, int fd) {
1955         isc_sockettype_t type = sock->type;
1956         int lockid = FDLOCK_ID(fd);
1957
1958         /*
1959          * No one has this socket open, so the watcher doesn't have to be
1960          * poked, and the socket doesn't have to be locked.
1961          */
1962         LOCK(&manager->fdlock[lockid]);
1963         manager->fds[fd] = NULL;
1964         if (type == isc_sockettype_fdwatch)
1965                 manager->fdstate[fd] = CLOSED;
1966         else
1967                 manager->fdstate[fd] = CLOSE_PENDING;
1968         UNLOCK(&manager->fdlock[lockid]);
1969         if (type == isc_sockettype_fdwatch) {
1970                 /*
1971                  * The caller may close the socket once this function returns,
1972                  * and `fd' may be reassigned for a new socket.  So we do
1973                  * unwatch_fd() here, rather than defer it via select_poke().
1974                  * Note: this may complicate data protection among threads and
1975                  * may reduce performance due to additional locks.  One way to
1976                  * solve this would be to dup() the watched descriptor, but we
1977                  * take a simpler approach at this moment.
1978                  */
1979                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
1980                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
1981         } else
1982                 select_poke(manager, fd, SELECT_POKE_CLOSE);
1983
1984         inc_stats(manager->stats, sock->statsindex[STATID_CLOSE]);
1985
1986         /*
1987          * update manager->maxfd here (XXX: this should be implemented more
1988          * efficiently)
1989          */
1990 #ifdef USE_SELECT
1991         LOCK(&manager->lock);
1992         if (manager->maxfd == fd) {
1993                 int i;
1994
1995                 manager->maxfd = 0;
1996                 for (i = fd - 1; i >= 0; i--) {
1997                         lockid = FDLOCK_ID(i);
1998
1999                         LOCK(&manager->fdlock[lockid]);
2000                         if (manager->fdstate[i] == MANAGED) {
2001                                 manager->maxfd = i;
2002                                 UNLOCK(&manager->fdlock[lockid]);
2003                                 break;
2004                         }
2005                         UNLOCK(&manager->fdlock[lockid]);
2006                 }
2007 #ifdef ISC_PLATFORM_USETHREADS
2008                 if (manager->maxfd < manager->pipe_fds[0])
2009                         manager->maxfd = manager->pipe_fds[0];
2010 #endif
2011         }
2012         UNLOCK(&manager->lock);
2013 #endif  /* USE_SELECT */
2014 }
2015
2016 static void
2017 destroy(isc__socket_t **sockp) {
2018         int fd;
2019         isc__socket_t *sock = *sockp;
2020         isc__socketmgr_t *manager = sock->manager;
2021
2022         socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
2023                    ISC_MSG_DESTROYING, "destroying");
2024
2025         INSIST(ISC_LIST_EMPTY(sock->accept_list));
2026         INSIST(ISC_LIST_EMPTY(sock->recv_list));
2027         INSIST(ISC_LIST_EMPTY(sock->send_list));
2028         INSIST(sock->connect_ev == NULL);
2029         REQUIRE(sock->fd == -1 || sock->fd < (int)manager->maxsocks);
2030
2031         if (sock->fd >= 0) {
2032                 fd = sock->fd;
2033                 sock->fd = -1;
2034                 closesocket(manager, sock, fd);
2035         }
2036
2037         LOCK(&manager->lock);
2038
2039         ISC_LIST_UNLINK(manager->socklist, sock, link);
2040
2041 #ifdef USE_WATCHER_THREAD
2042         if (ISC_LIST_EMPTY(manager->socklist))
2043                 SIGNAL(&manager->shutdown_ok);
2044 #endif /* USE_WATCHER_THREAD */
2045
2046         /* can't unlock manager as its memory context is still used */
2047         free_socket(sockp);
2048
2049         UNLOCK(&manager->lock);
2050 }
2051
2052 static isc_result_t
2053 allocate_socket(isc__socketmgr_t *manager, isc_sockettype_t type,
2054                 isc__socket_t **socketp)
2055 {
2056         isc__socket_t *sock;
2057         isc_result_t result;
2058         ISC_SOCKADDR_LEN_T cmsgbuflen;
2059
2060         sock = isc_mem_get(manager->mctx, sizeof(*sock));
2061
2062         if (sock == NULL)
2063                 return (ISC_R_NOMEMORY);
2064
2065         sock->common.magic = 0;
2066         sock->common.impmagic = 0;
2067         sock->references = 0;
2068
2069         sock->manager = manager;
2070         sock->type = type;
2071         sock->fd = -1;
2072         sock->dupped = 0;
2073         sock->statsindex = NULL;
2074
2075         ISC_LINK_INIT(sock, link);
2076
2077         sock->recvcmsgbuf = NULL;
2078         sock->sendcmsgbuf = NULL;
2079
2080         /*
2081          * Set up cmsg buffers.
2082          */
2083         cmsgbuflen = 0;
2084 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
2085         cmsgbuflen += cmsg_space(sizeof(struct in6_pktinfo));
2086 #endif
2087 #if defined(USE_CMSG) && defined(SO_TIMESTAMP)
2088         cmsgbuflen += cmsg_space(sizeof(struct timeval));
2089 #endif
2090         sock->recvcmsgbuflen = cmsgbuflen;
2091         if (sock->recvcmsgbuflen != 0U) {
2092                 sock->recvcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
2093                 if (sock->recvcmsgbuf == NULL) {
2094                         result = ISC_R_NOMEMORY;
2095                         goto error;
2096                 }
2097         }
2098
2099         cmsgbuflen = 0;
2100 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
2101         cmsgbuflen += cmsg_space(sizeof(struct in6_pktinfo));
2102 #if defined(IPV6_USE_MIN_MTU)
2103         /*
2104          * Provide space for working around FreeBSD's broken IPV6_USE_MIN_MTU
2105          * support.
2106          */
2107         cmsgbuflen += cmsg_space(sizeof(int));
2108 #endif
2109 #endif
2110         sock->sendcmsgbuflen = cmsgbuflen;
2111         if (sock->sendcmsgbuflen != 0U) {
2112                 sock->sendcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
2113                 if (sock->sendcmsgbuf == NULL) {
2114                         result = ISC_R_NOMEMORY;
2115                         goto error;
2116                 }
2117         }
2118
2119         memset(sock->name, 0, sizeof(sock->name));
2120         sock->tag = NULL;
2121
2122         /*
2123          * Set up list of readers and writers to be initially empty.
2124          */
2125         ISC_LIST_INIT(sock->recv_list);
2126         ISC_LIST_INIT(sock->send_list);
2127         ISC_LIST_INIT(sock->accept_list);
2128         sock->connect_ev = NULL;
2129         sock->pending_recv = 0;
2130         sock->pending_send = 0;
2131         sock->pending_accept = 0;
2132         sock->listener = 0;
2133         sock->connected = 0;
2134         sock->connecting = 0;
2135         sock->bound = 0;
2136
2137         /*
2138          * Initialize the lock.
2139          */
2140         result = isc_mutex_init(&sock->lock);
2141         if (result != ISC_R_SUCCESS) {
2142                 sock->common.magic = 0;
2143                 sock->common.impmagic = 0;
2144                 goto error;
2145         }
2146
2147         /*
2148          * Initialize readable and writable events.
2149          */
2150         ISC_EVENT_INIT(&sock->readable_ev, sizeof(intev_t),
2151                        ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTR,
2152                        NULL, sock, sock, NULL, NULL);
2153         ISC_EVENT_INIT(&sock->writable_ev, sizeof(intev_t),
2154                        ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTW,
2155                        NULL, sock, sock, NULL, NULL);
2156
2157         sock->common.magic = ISCAPI_SOCKET_MAGIC;
2158         sock->common.impmagic = SOCKET_MAGIC;
2159         *socketp = sock;
2160
2161         return (ISC_R_SUCCESS);
2162
2163  error:
2164         if (sock->recvcmsgbuf != NULL)
2165                 isc_mem_put(manager->mctx, sock->recvcmsgbuf,
2166                             sock->recvcmsgbuflen);
2167         if (sock->sendcmsgbuf != NULL)
2168                 isc_mem_put(manager->mctx, sock->sendcmsgbuf,
2169                             sock->sendcmsgbuflen);
2170         isc_mem_put(manager->mctx, sock, sizeof(*sock));
2171
2172         return (result);
2173 }
2174
2175 /*
2176  * This event requires that the various lists be empty, that the reference
2177  * count be 1, and that the magic number is valid.  The other socket bits,
2178  * like the lock, must be initialized as well.  The fd associated must be
2179  * marked as closed, by setting it to -1 on close, or this routine will
2180  * also close the socket.
2181  */
2182 static void
2183 free_socket(isc__socket_t **socketp) {
2184         isc__socket_t *sock = *socketp;
2185
2186         INSIST(sock->references == 0);
2187         INSIST(VALID_SOCKET(sock));
2188         INSIST(!sock->connecting);
2189         INSIST(!sock->pending_recv);
2190         INSIST(!sock->pending_send);
2191         INSIST(!sock->pending_accept);
2192         INSIST(ISC_LIST_EMPTY(sock->recv_list));
2193         INSIST(ISC_LIST_EMPTY(sock->send_list));
2194         INSIST(ISC_LIST_EMPTY(sock->accept_list));
2195         INSIST(!ISC_LINK_LINKED(sock, link));
2196
2197         if (sock->recvcmsgbuf != NULL)
2198                 isc_mem_put(sock->manager->mctx, sock->recvcmsgbuf,
2199                             sock->recvcmsgbuflen);
2200         if (sock->sendcmsgbuf != NULL)
2201                 isc_mem_put(sock->manager->mctx, sock->sendcmsgbuf,
2202                             sock->sendcmsgbuflen);
2203
2204         sock->common.magic = 0;
2205         sock->common.impmagic = 0;
2206
2207         DESTROYLOCK(&sock->lock);
2208
2209         isc_mem_put(sock->manager->mctx, sock, sizeof(*sock));
2210
2211         *socketp = NULL;
2212 }
2213
2214 #ifdef SO_BSDCOMPAT
2215 /*
2216  * This really should not be necessary to do.  Having to workout
2217  * which kernel version we are on at run time so that we don't cause
2218  * the kernel to issue a warning about us using a deprecated socket option.
2219  * Such warnings should *never* be on by default in production kernels.
2220  *
2221  * We can't do this a build time because executables are moved between
2222  * machines and hence kernels.
2223  *
2224  * We can't just not set SO_BSDCOMAT because some kernels require it.
2225  */
2226
2227 static isc_once_t         bsdcompat_once = ISC_ONCE_INIT;
2228 isc_boolean_t bsdcompat = ISC_TRUE;
2229
2230 static void
2231 clear_bsdcompat(void) {
2232 #ifdef __linux__
2233          struct utsname buf;
2234          char *endp;
2235          long int major;
2236          long int minor;
2237
2238          uname(&buf);    /* Can only fail if buf is bad in Linux. */
2239
2240          /* Paranoia in parsing can be increased, but we trust uname(). */
2241          major = strtol(buf.release, &endp, 10);
2242          if (*endp == '.') {
2243                 minor = strtol(endp+1, &endp, 10);
2244                 if ((major > 2) || ((major == 2) && (minor >= 4))) {
2245                         bsdcompat = ISC_FALSE;
2246                 }
2247          }
2248 #endif /* __linux __ */
2249 }
2250 #endif
2251
2252 static void
2253 use_min_mtu(isc__socket_t *sock) {
2254 #if !defined(IPV6_USE_MIN_MTU) && !defined(IPV6_MTU)
2255         UNUSED(sock);
2256 #endif
2257 #ifdef IPV6_USE_MIN_MTU
2258         /* use minimum MTU */
2259         if (sock->pf == AF_INET6) {
2260                 int on = 1;
2261                 (void)setsockopt(sock->fd, IPPROTO_IPV6, IPV6_USE_MIN_MTU,
2262                                 (void *)&on, sizeof(on));
2263         }
2264 #endif
2265 #if defined(IPV6_MTU)
2266         /*
2267          * Use minimum MTU on IPv6 sockets.
2268          */
2269         if (sock->pf == AF_INET6) {
2270                 int mtu = 1280;
2271                 (void)setsockopt(sock->fd, IPPROTO_IPV6, IPV6_MTU,
2272                                  &mtu, sizeof(mtu));
2273         }
2274 #endif
2275 }
2276
2277 static isc_result_t
2278 opensocket(isc__socketmgr_t *manager, isc__socket_t *sock,
2279            isc__socket_t *dup_socket)
2280 {
2281         isc_result_t result;
2282         char strbuf[ISC_STRERRORSIZE];
2283         const char *err = "socket";
2284         int tries = 0;
2285 #if defined(USE_CMSG) || defined(SO_BSDCOMPAT) || defined(SO_NOSIGPIPE)
2286         int on = 1;
2287 #endif
2288 #if defined(SO_RCVBUF)
2289         ISC_SOCKADDR_LEN_T optlen;
2290         int size;
2291 #endif
2292
2293  again:
2294         if (dup_socket == NULL) {
2295                 switch (sock->type) {
2296                 case isc_sockettype_udp:
2297                         sock->fd = socket(sock->pf, SOCK_DGRAM, IPPROTO_UDP);
2298                         break;
2299                 case isc_sockettype_tcp:
2300                         sock->fd = socket(sock->pf, SOCK_STREAM, IPPROTO_TCP);
2301                         break;
2302                 case isc_sockettype_unix:
2303                         sock->fd = socket(sock->pf, SOCK_STREAM, 0);
2304                         break;
2305                 case isc_sockettype_fdwatch:
2306                         /*
2307                          * We should not be called for isc_sockettype_fdwatch
2308                          * sockets.
2309                          */
2310                         INSIST(0);
2311                         break;
2312                 }
2313         } else {
2314                 sock->fd = dup(dup_socket->fd);
2315                 sock->dupped = 1;
2316                 sock->bound = dup_socket->bound;
2317         }
2318         if (sock->fd == -1 && errno == EINTR && tries++ < 42)
2319                 goto again;
2320
2321 #ifdef F_DUPFD
2322         /*
2323          * Leave a space for stdio and TCP to work in.
2324          */
2325         if (manager->reserved != 0 && sock->type == isc_sockettype_udp &&
2326             sock->fd >= 0 && sock->fd < manager->reserved) {
2327                 int new, tmp;
2328                 new = fcntl(sock->fd, F_DUPFD, manager->reserved);
2329                 tmp = errno;
2330                 (void)close(sock->fd);
2331                 errno = tmp;
2332                 sock->fd = new;
2333                 err = "isc_socket_create: fcntl/reserved";
2334         } else if (sock->fd >= 0 && sock->fd < 20) {
2335                 int new, tmp;
2336                 new = fcntl(sock->fd, F_DUPFD, 20);
2337                 tmp = errno;
2338                 (void)close(sock->fd);
2339                 errno = tmp;
2340                 sock->fd = new;
2341                 err = "isc_socket_create: fcntl";
2342         }
2343 #endif
2344
2345         if (sock->fd >= (int)manager->maxsocks) {
2346                 (void)close(sock->fd);
2347                 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2348                                ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2349                                isc_msgcat, ISC_MSGSET_SOCKET,
2350                                ISC_MSG_TOOMANYFDS,
2351                                "socket: file descriptor exceeds limit (%d/%u)",
2352                                sock->fd, manager->maxsocks);
2353                 return (ISC_R_NORESOURCES);
2354         }
2355
2356         if (sock->fd < 0) {
2357                 switch (errno) {
2358                 case EMFILE:
2359                 case ENFILE:
2360                         isc__strerror(errno, strbuf, sizeof(strbuf));
2361                         isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2362                                        ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2363                                        isc_msgcat, ISC_MSGSET_SOCKET,
2364                                        ISC_MSG_TOOMANYFDS,
2365                                        "%s: %s", err, strbuf);
2366                         /* fallthrough */
2367                 case ENOBUFS:
2368                         return (ISC_R_NORESOURCES);
2369
2370                 case EPROTONOSUPPORT:
2371                 case EPFNOSUPPORT:
2372                 case EAFNOSUPPORT:
2373                 /*
2374                  * Linux 2.2 (and maybe others) return EINVAL instead of
2375                  * EAFNOSUPPORT.
2376                  */
2377                 case EINVAL:
2378                         return (ISC_R_FAMILYNOSUPPORT);
2379
2380                 default:
2381                         isc__strerror(errno, strbuf, sizeof(strbuf));
2382                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2383                                          "%s() %s: %s", err,
2384                                          isc_msgcat_get(isc_msgcat,
2385                                                         ISC_MSGSET_GENERAL,
2386                                                         ISC_MSG_FAILED,
2387                                                         "failed"),
2388                                          strbuf);
2389                         return (ISC_R_UNEXPECTED);
2390                 }
2391         }
2392
2393         if (dup_socket != NULL)
2394                 goto setup_done;
2395
2396         result = make_nonblock(sock->fd);
2397         if (result != ISC_R_SUCCESS) {
2398                 (void)close(sock->fd);
2399                 return (result);
2400         }
2401
2402 #ifdef SO_BSDCOMPAT
2403         RUNTIME_CHECK(isc_once_do(&bsdcompat_once,
2404                                   clear_bsdcompat) == ISC_R_SUCCESS);
2405         if (sock->type != isc_sockettype_unix && bsdcompat &&
2406             setsockopt(sock->fd, SOL_SOCKET, SO_BSDCOMPAT,
2407                        (void *)&on, sizeof(on)) < 0) {
2408                 isc__strerror(errno, strbuf, sizeof(strbuf));
2409                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2410                                  "setsockopt(%d, SO_BSDCOMPAT) %s: %s",
2411                                  sock->fd,
2412                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2413                                                 ISC_MSG_FAILED, "failed"),
2414                                  strbuf);
2415                 /* Press on... */
2416         }
2417 #endif
2418
2419 #ifdef SO_NOSIGPIPE
2420         if (setsockopt(sock->fd, SOL_SOCKET, SO_NOSIGPIPE,
2421                        (void *)&on, sizeof(on)) < 0) {
2422                 isc__strerror(errno, strbuf, sizeof(strbuf));
2423                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2424                                  "setsockopt(%d, SO_NOSIGPIPE) %s: %s",
2425                                  sock->fd,
2426                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2427                                                 ISC_MSG_FAILED, "failed"),
2428                                  strbuf);
2429                 /* Press on... */
2430         }
2431 #endif
2432
2433         /*
2434          * Use minimum mtu if possible.
2435          */
2436         use_min_mtu(sock);
2437
2438 #if defined(USE_CMSG) || defined(SO_RCVBUF)
2439         if (sock->type == isc_sockettype_udp) {
2440
2441 #if defined(USE_CMSG)
2442 #if defined(SO_TIMESTAMP)
2443                 if (setsockopt(sock->fd, SOL_SOCKET, SO_TIMESTAMP,
2444                                (void *)&on, sizeof(on)) < 0
2445                     && errno != ENOPROTOOPT) {
2446                         isc__strerror(errno, strbuf, sizeof(strbuf));
2447                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2448                                          "setsockopt(%d, SO_TIMESTAMP) %s: %s",
2449                                          sock->fd,
2450                                          isc_msgcat_get(isc_msgcat,
2451                                                         ISC_MSGSET_GENERAL,
2452                                                         ISC_MSG_FAILED,
2453                                                         "failed"),
2454                                          strbuf);
2455                         /* Press on... */
2456                 }
2457 #endif /* SO_TIMESTAMP */
2458
2459 #if defined(ISC_PLATFORM_HAVEIPV6)
2460                 if (sock->pf == AF_INET6 && sock->recvcmsgbuflen == 0U) {
2461                         /*
2462                          * Warn explicitly because this anomaly can be hidden
2463                          * in usual operation (and unexpectedly appear later).
2464                          */
2465                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2466                                          "No buffer available to receive "
2467                                          "IPv6 destination");
2468                 }
2469 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
2470 #ifdef IPV6_RECVPKTINFO
2471                 /* RFC 3542 */
2472                 if ((sock->pf == AF_INET6)
2473                     && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_RECVPKTINFO,
2474                                    (void *)&on, sizeof(on)) < 0)) {
2475                         isc__strerror(errno, strbuf, sizeof(strbuf));
2476                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2477                                          "setsockopt(%d, IPV6_RECVPKTINFO) "
2478                                          "%s: %s", sock->fd,
2479                                          isc_msgcat_get(isc_msgcat,
2480                                                         ISC_MSGSET_GENERAL,
2481                                                         ISC_MSG_FAILED,
2482                                                         "failed"),
2483                                          strbuf);
2484                 }
2485 #else
2486                 /* RFC 2292 */
2487                 if ((sock->pf == AF_INET6)
2488                     && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_PKTINFO,
2489                                    (void *)&on, sizeof(on)) < 0)) {
2490                         isc__strerror(errno, strbuf, sizeof(strbuf));
2491                         UNEXPECTED_ERROR(__FILE__, __LINE__,
2492                                          "setsockopt(%d, IPV6_PKTINFO) %s: %s",
2493                                          sock->fd,
2494                                          isc_msgcat_get(isc_msgcat,
2495                                                         ISC_MSGSET_GENERAL,
2496                                                         ISC_MSG_FAILED,
2497                                                         "failed"),
2498                                          strbuf);
2499                 }
2500 #endif /* IPV6_RECVPKTINFO */
2501 #endif /* ISC_PLATFORM_HAVEIN6PKTINFO */
2502 #if defined(IPV6_MTU_DISCOVER) && defined(IPV6_PMTUDISC_DONT)
2503                 /*
2504                  * Turn off Path MTU discovery on IPv6/UDP sockets.
2505                  */
2506                 if (sock->pf == AF_INET6) {
2507                         int action = IPV6_PMTUDISC_DONT;
2508                         (void)setsockopt(sock->fd, IPPROTO_IPV6,
2509                                          IPV6_MTU_DISCOVER, &action,
2510                                          sizeof(action));
2511                 }
2512 #endif
2513 #endif /* ISC_PLATFORM_HAVEIPV6 */
2514 #endif /* defined(USE_CMSG) */
2515
2516 #if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
2517                 /*
2518                  * Turn off Path MTU discovery on IPv4/UDP sockets.
2519                  */
2520                 if (sock->pf == AF_INET) {
2521                         int action = IP_PMTUDISC_DONT;
2522                         (void)setsockopt(sock->fd, IPPROTO_IP, IP_MTU_DISCOVER,
2523                                          &action, sizeof(action));
2524                 }
2525 #endif
2526 #if defined(IP_DONTFRAG)
2527                 /*
2528                  * Turn off Path MTU discovery on IPv4/UDP sockets.
2529                  */
2530                 if (sock->pf == AF_INET) {
2531                         int off = 0;
2532                         (void)setsockopt(sock->fd, IPPROTO_IP, IP_DONTFRAG,
2533                                          &off, sizeof(off));
2534                 }
2535 #endif
2536
2537 #if defined(SO_RCVBUF)
2538                 optlen = sizeof(size);
2539                 if (getsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
2540                                (void *)&size, &optlen) >= 0 &&
2541                      size < RCVBUFSIZE) {
2542                         size = RCVBUFSIZE;
2543                         if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
2544                                        (void *)&size, sizeof(size)) == -1) {
2545                                 isc__strerror(errno, strbuf, sizeof(strbuf));
2546                                 UNEXPECTED_ERROR(__FILE__, __LINE__,
2547                                         "setsockopt(%d, SO_RCVBUF, %d) %s: %s",
2548                                         sock->fd, size,
2549                                         isc_msgcat_get(isc_msgcat,
2550                                                        ISC_MSGSET_GENERAL,
2551                                                        ISC_MSG_FAILED,
2552                                                        "failed"),
2553                                         strbuf);
2554                         }
2555                 }
2556 #endif
2557         }
2558 #endif /* defined(USE_CMSG) || defined(SO_RCVBUF) */
2559
2560 setup_done:
2561         inc_stats(manager->stats, sock->statsindex[STATID_OPEN]);
2562
2563         return (ISC_R_SUCCESS);
2564 }
2565
2566 /*
2567  * Create a 'type' socket or duplicate an existing socket, managed
2568  * by 'manager'.  Events will be posted to 'task' and when dispatched
2569  * 'action' will be called with 'arg' as the arg value.  The new
2570  * socket is returned in 'socketp'.
2571  */
2572 static isc_result_t
2573 socket_create(isc_socketmgr_t *manager0, int pf, isc_sockettype_t type,
2574               isc_socket_t **socketp, isc_socket_t *dup_socket)
2575 {
2576         isc__socket_t *sock = NULL;
2577         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
2578         isc_result_t result;
2579         int lockid;
2580
2581         REQUIRE(VALID_MANAGER(manager));
2582         REQUIRE(socketp != NULL && *socketp == NULL);
2583         REQUIRE(type != isc_sockettype_fdwatch);
2584
2585         result = allocate_socket(manager, type, &sock);
2586         if (result != ISC_R_SUCCESS)
2587                 return (result);
2588
2589         switch (sock->type) {
2590         case isc_sockettype_udp:
2591                 sock->statsindex =
2592                         (pf == AF_INET) ? udp4statsindex : udp6statsindex;
2593                 break;
2594         case isc_sockettype_tcp:
2595                 sock->statsindex =
2596                         (pf == AF_INET) ? tcp4statsindex : tcp6statsindex;
2597                 break;
2598         case isc_sockettype_unix:
2599                 sock->statsindex = unixstatsindex;
2600                 break;
2601         default:
2602                 INSIST(0);
2603         }
2604
2605         sock->pf = pf;
2606
2607         result = opensocket(manager, sock, (isc__socket_t *)dup_socket);
2608         if (result != ISC_R_SUCCESS) {
2609                 inc_stats(manager->stats, sock->statsindex[STATID_OPENFAIL]);
2610                 free_socket(&sock);
2611                 return (result);
2612         }
2613
2614         sock->common.methods = (isc_socketmethods_t *)&socketmethods;
2615         sock->references = 1;
2616         *socketp = (isc_socket_t *)sock;
2617
2618         /*
2619          * Note we don't have to lock the socket like we normally would because
2620          * there are no external references to it yet.
2621          */
2622
2623         lockid = FDLOCK_ID(sock->fd);
2624         LOCK(&manager->fdlock[lockid]);
2625         manager->fds[sock->fd] = sock;
2626         manager->fdstate[sock->fd] = MANAGED;
2627 #ifdef USE_DEVPOLL
2628         INSIST(sock->manager->fdpollinfo[sock->fd].want_read == 0 &&
2629                sock->manager->fdpollinfo[sock->fd].want_write == 0);
2630 #endif
2631         UNLOCK(&manager->fdlock[lockid]);
2632
2633         LOCK(&manager->lock);
2634         ISC_LIST_APPEND(manager->socklist, sock, link);
2635 #ifdef USE_SELECT
2636         if (manager->maxfd < sock->fd)
2637                 manager->maxfd = sock->fd;
2638 #endif
2639         UNLOCK(&manager->lock);
2640
2641         socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
2642                    ISC_MSG_CREATED, dup_socket != NULL ? "dupped" : "created");
2643
2644         return (ISC_R_SUCCESS);
2645 }
2646
2647 /*%
2648  * Create a new 'type' socket managed by 'manager'.  Events
2649  * will be posted to 'task' and when dispatched 'action' will be
2650  * called with 'arg' as the arg value.  The new socket is returned
2651  * in 'socketp'.
2652  */
2653 ISC_SOCKETFUNC_SCOPE isc_result_t
2654 isc__socket_create(isc_socketmgr_t *manager0, int pf, isc_sockettype_t type,
2655                    isc_socket_t **socketp)
2656 {
2657         return (socket_create(manager0, pf, type, socketp, NULL));
2658 }
2659
2660 /*%
2661  * Duplicate an existing socket.  The new socket is returned
2662  * in 'socketp'.
2663  */
2664 ISC_SOCKETFUNC_SCOPE isc_result_t
2665 isc__socket_dup(isc_socket_t *sock0, isc_socket_t **socketp) {
2666         isc__socket_t *sock = (isc__socket_t *)sock0;
2667
2668         REQUIRE(VALID_SOCKET(sock));
2669         REQUIRE(socketp != NULL && *socketp == NULL);
2670
2671         return (socket_create((isc_socketmgr_t *) sock->manager,
2672                               sock->pf, sock->type, socketp,
2673                               sock0));
2674 }
2675
2676 #ifdef BIND9
2677 ISC_SOCKETFUNC_SCOPE isc_result_t
2678 isc__socket_open(isc_socket_t *sock0) {
2679         isc_result_t result;
2680         isc__socket_t *sock = (isc__socket_t *)sock0;
2681
2682         REQUIRE(VALID_SOCKET(sock));
2683
2684         LOCK(&sock->lock);
2685         REQUIRE(sock->references == 1);
2686         REQUIRE(sock->type != isc_sockettype_fdwatch);
2687         UNLOCK(&sock->lock);
2688         /*
2689          * We don't need to retain the lock hereafter, since no one else has
2690          * this socket.
2691          */
2692         REQUIRE(sock->fd == -1);
2693
2694         result = opensocket(sock->manager, sock, NULL);
2695         if (result != ISC_R_SUCCESS)
2696                 sock->fd = -1;
2697
2698         if (result == ISC_R_SUCCESS) {
2699                 int lockid = FDLOCK_ID(sock->fd);
2700
2701                 LOCK(&sock->manager->fdlock[lockid]);
2702                 sock->manager->fds[sock->fd] = sock;
2703                 sock->manager->fdstate[sock->fd] = MANAGED;
2704 #ifdef USE_DEVPOLL
2705                 INSIST(sock->manager->fdpollinfo[sock->fd].want_read == 0 &&
2706                        sock->manager->fdpollinfo[sock->fd].want_write == 0);
2707 #endif
2708                 UNLOCK(&sock->manager->fdlock[lockid]);
2709
2710 #ifdef USE_SELECT
2711                 LOCK(&sock->manager->lock);
2712                 if (sock->manager->maxfd < sock->fd)
2713                         sock->manager->maxfd = sock->fd;
2714                 UNLOCK(&sock->manager->lock);
2715 #endif
2716         }
2717
2718         return (result);
2719 }
2720 #endif  /* BIND9 */
2721
2722 /*
2723  * Create a new 'type' socket managed by 'manager'.  Events
2724  * will be posted to 'task' and when dispatched 'action' will be
2725  * called with 'arg' as the arg value.  The new socket is returned
2726  * in 'socketp'.
2727  */
2728 ISC_SOCKETFUNC_SCOPE isc_result_t
2729 isc__socket_fdwatchcreate(isc_socketmgr_t *manager0, int fd, int flags,
2730                           isc_sockfdwatch_t callback, void *cbarg,
2731                           isc_task_t *task, isc_socket_t **socketp)
2732 {
2733         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
2734         isc__socket_t *sock = NULL;
2735         isc_result_t result;
2736         int lockid;
2737
2738         REQUIRE(VALID_MANAGER(manager));
2739         REQUIRE(socketp != NULL && *socketp == NULL);
2740
2741         result = allocate_socket(manager, isc_sockettype_fdwatch, &sock);
2742         if (result != ISC_R_SUCCESS)
2743                 return (result);
2744
2745         sock->fd = fd;
2746         sock->fdwatcharg = cbarg;
2747         sock->fdwatchcb = callback;
2748         sock->fdwatchflags = flags;
2749         sock->fdwatchtask = task;
2750         sock->statsindex = fdwatchstatsindex;
2751
2752         sock->common.methods = (isc_socketmethods_t *)&socketmethods;
2753         sock->references = 1;
2754         *socketp = (isc_socket_t *)sock;
2755
2756         /*
2757          * Note we don't have to lock the socket like we normally would because
2758          * there are no external references to it yet.
2759          */
2760
2761         lockid = FDLOCK_ID(sock->fd);
2762         LOCK(&manager->fdlock[lockid]);
2763         manager->fds[sock->fd] = sock;
2764         manager->fdstate[sock->fd] = MANAGED;
2765         UNLOCK(&manager->fdlock[lockid]);
2766
2767         LOCK(&manager->lock);
2768         ISC_LIST_APPEND(manager->socklist, sock, link);
2769 #ifdef USE_SELECT
2770         if (manager->maxfd < sock->fd)
2771                 manager->maxfd = sock->fd;
2772 #endif
2773         UNLOCK(&manager->lock);
2774
2775         if (flags & ISC_SOCKFDWATCH_READ)
2776                 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
2777         if (flags & ISC_SOCKFDWATCH_WRITE)
2778                 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
2779
2780         socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
2781                    ISC_MSG_CREATED, "fdwatch-created");
2782
2783         return (ISC_R_SUCCESS);
2784 }
2785
2786 /*
2787  * Indicate to the manager that it should watch the socket again.
2788  * This can be used to restart watching if the previous event handler
2789  * didn't indicate there was more data to be processed.  Primarily
2790  * it is for writing but could be used for reading if desired
2791  */
2792
2793 ISC_SOCKETFUNC_SCOPE isc_result_t
2794 isc__socket_fdwatchpoke(isc_socket_t *sock0, int flags)
2795 {
2796         isc__socket_t *sock = (isc__socket_t *)sock0;
2797
2798         REQUIRE(VALID_SOCKET(sock));
2799
2800         /*
2801          * We check both flags first to allow us to get the lock
2802          * once but only if we need it.
2803          */
2804
2805         if ((flags & (ISC_SOCKFDWATCH_READ | ISC_SOCKFDWATCH_WRITE)) != 0) {
2806                 LOCK(&sock->lock);
2807                 if (((flags & ISC_SOCKFDWATCH_READ) != 0) &&
2808                     !sock->pending_recv)
2809                         select_poke(sock->manager, sock->fd,
2810                                     SELECT_POKE_READ);
2811                 if (((flags & ISC_SOCKFDWATCH_WRITE) != 0) &&
2812                     !sock->pending_send)
2813                         select_poke(sock->manager, sock->fd,
2814                                     SELECT_POKE_WRITE);
2815                 UNLOCK(&sock->lock);
2816         }
2817
2818         socket_log(sock, NULL, TRACE, isc_msgcat, ISC_MSGSET_SOCKET,
2819                    ISC_MSG_POKED, "fdwatch-poked flags: %d", flags);
2820
2821         return (ISC_R_SUCCESS);
2822 }
2823
2824 /*
2825  * Attach to a socket.  Caller must explicitly detach when it is done.
2826  */
2827 ISC_SOCKETFUNC_SCOPE void
2828 isc__socket_attach(isc_socket_t *sock0, isc_socket_t **socketp) {
2829         isc__socket_t *sock = (isc__socket_t *)sock0;
2830
2831         REQUIRE(VALID_SOCKET(sock));
2832         REQUIRE(socketp != NULL && *socketp == NULL);
2833
2834         LOCK(&sock->lock);
2835         sock->references++;
2836         UNLOCK(&sock->lock);
2837
2838         *socketp = (isc_socket_t *)sock;
2839 }
2840
2841 /*
2842  * Dereference a socket.  If this is the last reference to it, clean things
2843  * up by destroying the socket.
2844  */
2845 ISC_SOCKETFUNC_SCOPE void
2846 isc__socket_detach(isc_socket_t **socketp) {
2847         isc__socket_t *sock;
2848         isc_boolean_t kill_socket = ISC_FALSE;
2849
2850         REQUIRE(socketp != NULL);
2851         sock = (isc__socket_t *)*socketp;
2852         REQUIRE(VALID_SOCKET(sock));
2853
2854         LOCK(&sock->lock);
2855         REQUIRE(sock->references > 0);
2856         sock->references--;
2857         if (sock->references == 0)
2858                 kill_socket = ISC_TRUE;
2859         UNLOCK(&sock->lock);
2860
2861         if (kill_socket)
2862                 destroy(&sock);
2863
2864         *socketp = NULL;
2865 }
2866
2867 #ifdef BIND9
2868 ISC_SOCKETFUNC_SCOPE isc_result_t
2869 isc__socket_close(isc_socket_t *sock0) {
2870         isc__socket_t *sock = (isc__socket_t *)sock0;
2871         int fd;
2872         isc__socketmgr_t *manager;
2873
2874         fflush(stdout);
2875         REQUIRE(VALID_SOCKET(sock));
2876
2877         LOCK(&sock->lock);
2878
2879         REQUIRE(sock->references == 1);
2880         REQUIRE(sock->type != isc_sockettype_fdwatch);
2881         REQUIRE(sock->fd >= 0 && sock->fd < (int)sock->manager->maxsocks);
2882
2883         INSIST(!sock->connecting);
2884         INSIST(!sock->pending_recv);
2885         INSIST(!sock->pending_send);
2886         INSIST(!sock->pending_accept);
2887         INSIST(ISC_LIST_EMPTY(sock->recv_list));
2888         INSIST(ISC_LIST_EMPTY(sock->send_list));
2889         INSIST(ISC_LIST_EMPTY(sock->accept_list));
2890         INSIST(sock->connect_ev == NULL);
2891
2892         manager = sock->manager;
2893         fd = sock->fd;
2894         sock->fd = -1;
2895         sock->dupped = 0;
2896         memset(sock->name, 0, sizeof(sock->name));
2897         sock->tag = NULL;
2898         sock->listener = 0;
2899         sock->connected = 0;
2900         sock->connecting = 0;
2901         sock->bound = 0;
2902         isc_sockaddr_any(&sock->peer_address);
2903
2904         UNLOCK(&sock->lock);
2905
2906         closesocket(manager, sock, fd);
2907
2908         return (ISC_R_SUCCESS);
2909 }
2910 #endif  /* BIND9 */
2911
2912 /*
2913  * I/O is possible on a given socket.  Schedule an event to this task that
2914  * will call an internal function to do the I/O.  This will charge the
2915  * task with the I/O operation and let our select loop handler get back
2916  * to doing something real as fast as possible.
2917  *
2918  * The socket and manager must be locked before calling this function.
2919  */
2920 static void
2921 dispatch_recv(isc__socket_t *sock) {
2922         intev_t *iev;
2923         isc_socketevent_t *ev;
2924         isc_task_t *sender;
2925
2926         INSIST(!sock->pending_recv);
2927
2928         if (sock->type != isc_sockettype_fdwatch) {
2929                 ev = ISC_LIST_HEAD(sock->recv_list);
2930                 if (ev == NULL)
2931                         return;
2932                 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2933                            "dispatch_recv:  event %p -> task %p",
2934                            ev, ev->ev_sender);
2935                 sender = ev->ev_sender;
2936         } else {
2937                 sender = sock->fdwatchtask;
2938         }
2939
2940         sock->pending_recv = 1;
2941         iev = &sock->readable_ev;
2942
2943         sock->references++;
2944         iev->ev_sender = sock;
2945         if (sock->type == isc_sockettype_fdwatch)
2946                 iev->ev_action = internal_fdwatch_read;
2947         else
2948                 iev->ev_action = internal_recv;
2949         iev->ev_arg = sock;
2950
2951         isc_task_send(sender, (isc_event_t **)&iev);
2952 }
2953
2954 static void
2955 dispatch_send(isc__socket_t *sock) {
2956         intev_t *iev;
2957         isc_socketevent_t *ev;
2958         isc_task_t *sender;
2959
2960         INSIST(!sock->pending_send);
2961
2962         if (sock->type != isc_sockettype_fdwatch) {
2963                 ev = ISC_LIST_HEAD(sock->send_list);
2964                 if (ev == NULL)
2965                         return;
2966                 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2967                            "dispatch_send:  event %p -> task %p",
2968                            ev, ev->ev_sender);
2969                 sender = ev->ev_sender;
2970         } else {
2971                 sender = sock->fdwatchtask;
2972         }
2973
2974         sock->pending_send = 1;
2975         iev = &sock->writable_ev;
2976
2977         sock->references++;
2978         iev->ev_sender = sock;
2979         if (sock->type == isc_sockettype_fdwatch)
2980                 iev->ev_action = internal_fdwatch_write;
2981         else
2982                 iev->ev_action = internal_send;
2983         iev->ev_arg = sock;
2984
2985         isc_task_send(sender, (isc_event_t **)&iev);
2986 }
2987
2988 /*
2989  * Dispatch an internal accept event.
2990  */
2991 static void
2992 dispatch_accept(isc__socket_t *sock) {
2993         intev_t *iev;
2994         isc_socket_newconnev_t *ev;
2995
2996         INSIST(!sock->pending_accept);
2997
2998         /*
2999          * Are there any done events left, or were they all canceled
3000          * before the manager got the socket lock?
3001          */
3002         ev = ISC_LIST_HEAD(sock->accept_list);
3003         if (ev == NULL)
3004                 return;
3005
3006         sock->pending_accept = 1;
3007         iev = &sock->readable_ev;
3008
3009         sock->references++;  /* keep socket around for this internal event */
3010         iev->ev_sender = sock;
3011         iev->ev_action = internal_accept;
3012         iev->ev_arg = sock;
3013
3014         isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
3015 }
3016
3017 static void
3018 dispatch_connect(isc__socket_t *sock) {
3019         intev_t *iev;
3020         isc_socket_connev_t *ev;
3021
3022         iev = &sock->writable_ev;
3023
3024         ev = sock->connect_ev;
3025         INSIST(ev != NULL); /* XXX */
3026
3027         INSIST(sock->connecting);
3028
3029         sock->references++;  /* keep socket around for this internal event */
3030         iev->ev_sender = sock;
3031         iev->ev_action = internal_connect;
3032         iev->ev_arg = sock;
3033
3034         isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
3035 }
3036
3037 /*
3038  * Dequeue an item off the given socket's read queue, set the result code
3039  * in the done event to the one provided, and send it to the task it was
3040  * destined for.
3041  *
3042  * If the event to be sent is on a list, remove it before sending.  If
3043  * asked to, send and detach from the socket as well.
3044  *
3045  * Caller must have the socket locked if the event is attached to the socket.
3046  */
3047 static void
3048 send_recvdone_event(isc__socket_t *sock, isc_socketevent_t **dev) {
3049         isc_task_t *task;
3050
3051         task = (*dev)->ev_sender;
3052
3053         (*dev)->ev_sender = sock;
3054
3055         if (ISC_LINK_LINKED(*dev, ev_link))
3056                 ISC_LIST_DEQUEUE(sock->recv_list, *dev, ev_link);
3057
3058         if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
3059             == ISC_SOCKEVENTATTR_ATTACHED)
3060                 isc_task_sendanddetach(&task, (isc_event_t **)dev);
3061         else
3062                 isc_task_send(task, (isc_event_t **)dev);
3063 }
3064
3065 /*
3066  * See comments for send_recvdone_event() above.
3067  *
3068  * Caller must have the socket locked if the event is attached to the socket.
3069  */
3070 static void
3071 send_senddone_event(isc__socket_t *sock, isc_socketevent_t **dev) {
3072         isc_task_t *task;
3073
3074         INSIST(dev != NULL && *dev != NULL);
3075
3076         task = (*dev)->ev_sender;
3077         (*dev)->ev_sender = sock;
3078
3079         if (ISC_LINK_LINKED(*dev, ev_link))
3080                 ISC_LIST_DEQUEUE(sock->send_list, *dev, ev_link);
3081
3082         if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
3083             == ISC_SOCKEVENTATTR_ATTACHED)
3084                 isc_task_sendanddetach(&task, (isc_event_t **)dev);
3085         else
3086                 isc_task_send(task, (isc_event_t **)dev);
3087 }
3088
3089 /*
3090  * Call accept() on a socket, to get the new file descriptor.  The listen
3091  * socket is used as a prototype to create a new isc_socket_t.  The new
3092  * socket has one outstanding reference.  The task receiving the event
3093  * will be detached from just after the event is delivered.
3094  *
3095  * On entry to this function, the event delivered is the internal
3096  * readable event, and the first item on the accept_list should be
3097  * the done event we want to send.  If the list is empty, this is a no-op,
3098  * so just unlock and return.
3099  */
3100 static void
3101 internal_accept(isc_task_t *me, isc_event_t *ev) {
3102         isc__socket_t *sock;
3103         isc__socketmgr_t *manager;
3104         isc_socket_newconnev_t *dev;
3105         isc_task_t *task;
3106         ISC_SOCKADDR_LEN_T addrlen;
3107         int fd;
3108         isc_result_t result = ISC_R_SUCCESS;
3109         char strbuf[ISC_STRERRORSIZE];
3110         const char *err = "accept";
3111
3112         UNUSED(me);
3113
3114         sock = ev->ev_sender;
3115         INSIST(VALID_SOCKET(sock));
3116
3117         LOCK(&sock->lock);
3118         socket_log(sock, NULL, TRACE,
3119                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTLOCK,
3120                    "internal_accept called, locked socket");
3121
3122         manager = sock->manager;
3123         INSIST(VALID_MANAGER(manager));
3124
3125         INSIST(sock->listener);
3126         INSIST(sock->pending_accept == 1);
3127         sock->pending_accept = 0;
3128
3129         INSIST(sock->references > 0);
3130         sock->references--;  /* the internal event is done with this socket */
3131         if (sock->references == 0) {
3132                 UNLOCK(&sock->lock);
3133                 destroy(&sock);
3134                 return;
3135         }
3136
3137         /*
3138          * Get the first item off the accept list.
3139          * If it is empty, unlock the socket and return.
3140          */
3141         dev = ISC_LIST_HEAD(sock->accept_list);
3142         if (dev == NULL) {
3143                 UNLOCK(&sock->lock);
3144                 return;
3145         }
3146
3147         /*
3148          * Try to accept the new connection.  If the accept fails with
3149          * EAGAIN or EINTR, simply poke the watcher to watch this socket
3150          * again.  Also ignore ECONNRESET, which has been reported to
3151          * be spuriously returned on Linux 2.2.19 although it is not
3152          * a documented error for accept().  ECONNABORTED has been
3153          * reported for Solaris 8.  The rest are thrown in not because
3154          * we have seen them but because they are ignored by other
3155          * daemons such as BIND 8 and Apache.
3156          */
3157
3158         addrlen = sizeof(NEWCONNSOCK(dev)->peer_address.type);
3159         memset(&NEWCONNSOCK(dev)->peer_address.type, 0, addrlen);
3160         fd = accept(sock->fd, &NEWCONNSOCK(dev)->peer_address.type.sa,
3161                     (void *)&addrlen);
3162
3163 #ifdef F_DUPFD
3164         /*
3165          * Leave a space for stdio to work in.
3166          */
3167         if (fd >= 0 && fd < 20) {
3168                 int new, tmp;
3169                 new = fcntl(fd, F_DUPFD, 20);
3170                 tmp = errno;
3171                 (void)close(fd);
3172                 errno = tmp;
3173                 fd = new;
3174                 err = "accept/fcntl";
3175         }
3176 #endif
3177
3178         if (fd < 0) {
3179                 if (SOFT_ERROR(errno))
3180                         goto soft_error;
3181                 switch (errno) {
3182                 case ENFILE:
3183                 case EMFILE:
3184                         isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
3185                                        ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
3186                                        isc_msgcat, ISC_MSGSET_SOCKET,
3187                                        ISC_MSG_TOOMANYFDS,
3188                                        "%s: too many open file descriptors",
3189                                        err);
3190                         goto soft_error;
3191
3192                 case ENOBUFS:
3193                 case ENOMEM:
3194                 case ECONNRESET:
3195                 case ECONNABORTED:
3196                 case EHOSTUNREACH:
3197                 case EHOSTDOWN:
3198                 case ENETUNREACH:
3199                 case ENETDOWN:
3200                 case ECONNREFUSED:
3201 #ifdef EPROTO
3202                 case EPROTO:
3203 #endif
3204 #ifdef ENONET
3205                 case ENONET:
3206 #endif
3207                         goto soft_error;
3208                 default:
3209                         break;
3210                 }
3211                 isc__strerror(errno, strbuf, sizeof(strbuf));
3212                 UNEXPECTED_ERROR(__FILE__, __LINE__,
3213                                  "internal_accept: %s() %s: %s", err,
3214                                  isc_msgcat_get(isc_msgcat,
3215                                                 ISC_MSGSET_GENERAL,
3216                                                 ISC_MSG_FAILED,
3217                                                 "failed"),
3218                                  strbuf);
3219                 fd = -1;
3220                 result = ISC_R_UNEXPECTED;
3221         } else {
3222                 if (addrlen == 0U) {
3223                         UNEXPECTED_ERROR(__FILE__, __LINE__,
3224                                          "internal_accept(): "
3225                                          "accept() failed to return "
3226                                          "remote address");
3227
3228                         (void)close(fd);
3229                         goto soft_error;
3230                 } else if (NEWCONNSOCK(dev)->peer_address.type.sa.sa_family !=
3231                            sock->pf)
3232                 {
3233                         UNEXPECTED_ERROR(__FILE__, __LINE__,
3234                                          "internal_accept(): "
3235                                          "accept() returned peer address "
3236                                          "family %u (expected %u)",
3237                                          NEWCONNSOCK(dev)->peer_address.
3238                                          type.sa.sa_family,
3239                                          sock->pf);
3240                         (void)close(fd);
3241                         goto soft_error;
3242                 } else if (fd >= (int)manager->maxsocks) {
3243                         isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
3244                                        ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
3245                                        isc_msgcat, ISC_MSGSET_SOCKET,
3246                                        ISC_MSG_TOOMANYFDS,
3247                                        "accept: "
3248                                        "file descriptor exceeds limit (%d/%u)",
3249                                        fd, manager->maxsocks);
3250                         (void)close(fd);
3251                         goto soft_error;
3252                 }
3253         }
3254
3255         if (fd != -1) {
3256                 NEWCONNSOCK(dev)->peer_address.length = addrlen;
3257                 NEWCONNSOCK(dev)->pf = sock->pf;
3258         }
3259
3260         /*
3261          * Pull off the done event.
3262          */
3263         ISC_LIST_UNLINK(sock->accept_list, dev, ev_link);
3264
3265         /*
3266          * Poke watcher if there are more pending accepts.
3267          */
3268         if (!ISC_LIST_EMPTY(sock->accept_list))
3269                 select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
3270
3271         UNLOCK(&sock->lock);
3272
3273         if (fd != -1) {
3274                 result = make_nonblock(fd);
3275                 if (result != ISC_R_SUCCESS) {
3276                         (void)close(fd);
3277                         fd = -1;
3278                 }
3279         }
3280
3281         /*
3282          * -1 means the new socket didn't happen.
3283          */
3284         if (fd != -1) {
3285                 int lockid = FDLOCK_ID(fd);
3286
3287                 NEWCONNSOCK(dev)->fd = fd;
3288                 NEWCONNSOCK(dev)->bound = 1;
3289                 NEWCONNSOCK(dev)->connected = 1;
3290
3291                 /*
3292                  * Use minimum mtu if possible.
3293                  */
3294                 use_min_mtu(NEWCONNSOCK(dev));
3295
3296                 /*
3297                  * Save away the remote address
3298                  */
3299                 dev->address = NEWCONNSOCK(dev)->peer_address;
3300
3301                 LOCK(&manager->fdlock[lockid]);
3302                 manager->fds[fd] = NEWCONNSOCK(dev);
3303                 manager->fdstate[fd] = MANAGED;
3304                 UNLOCK(&manager->fdlock[lockid]);
3305
3306                 LOCK(&manager->lock);
3307
3308 #ifdef USE_SELECT
3309                 if (manager->maxfd < fd)
3310                         manager->maxfd = fd;
3311 #endif
3312
3313                 socket_log(sock, &NEWCONNSOCK(dev)->peer_address, CREATION,
3314                            isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTEDCXN,
3315                            "accepted connection, new socket %p",
3316                            dev->newsocket);
3317
3318                 ISC_LIST_APPEND(manager->socklist, NEWCONNSOCK(dev), link);
3319
3320                 UNLOCK(&manager->lock);
3321
3322                 inc_stats(manager->stats, sock->statsindex[STATID_ACCEPT]);
3323         } else {
3324                 inc_stats(manager->stats, sock->statsindex[STATID_ACCEPTFAIL]);
3325                 NEWCONNSOCK(dev)->references--;
3326                 free_socket((isc__socket_t **)&dev->newsocket);
3327         }
3328
3329         /*
3330          * Fill in the done event details and send it off.
3331          */
3332         dev->result = result;
3333         task = dev->ev_sender;
3334         dev->ev_sender = sock;
3335
3336         isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
3337         return;
3338
3339  soft_error:
3340         select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
3341         UNLOCK(&sock->lock);
3342
3343         inc_stats(manager->stats, sock->statsindex[STATID_ACCEPTFAIL]);
3344         return;
3345 }
3346
3347 static void
3348 internal_recv(isc_task_t *me, isc_event_t *ev) {
3349         isc_socketevent_t *dev;
3350         isc__socket_t *sock;
3351
3352         INSIST(ev->ev_type == ISC_SOCKEVENT_INTR);
3353
3354         sock = ev->ev_sender;
3355         INSIST(VALID_SOCKET(sock));
3356
3357         LOCK(&sock->lock);
3358         socket_log(sock, NULL, IOEVENT,
3359                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALRECV,
3360                    "internal_recv: task %p got event %p", me, ev);
3361
3362         INSIST(sock->pending_recv == 1);
3363         sock->pending_recv = 0;
3364
3365         INSIST(sock->references > 0);
3366         sock->references--;  /* the internal event is done with this socket */
3367         if (sock->references == 0) {
3368                 UNLOCK(&sock->lock);
3369                 destroy(&sock);
3370                 return;
3371         }
3372
3373         /*
3374          * Try to do as much I/O as possible on this socket.  There are no
3375          * limits here, currently.
3376          */
3377         dev = ISC_LIST_HEAD(sock->recv_list);
3378         while (dev != NULL) {
3379                 switch (doio_recv(sock, dev)) {
3380                 case DOIO_SOFT:
3381                         goto poke;
3382
3383                 case DOIO_EOF:
3384                         /*
3385                          * read of 0 means the remote end was closed.
3386                          * Run through the event queue and dispatch all
3387                          * the events with an EOF result code.
3388                          */
3389                         do {
3390                                 dev->result = ISC_R_EOF;
3391                                 send_recvdone_event(sock, &dev);
3392                                 dev = ISC_LIST_HEAD(sock->recv_list);
3393                         } while (dev != NULL);
3394                         goto poke;
3395
3396                 case DOIO_SUCCESS:
3397                 case DOIO_HARD:
3398                         send_recvdone_event(sock, &dev);
3399                         break;
3400                 }
3401
3402                 dev = ISC_LIST_HEAD(sock->recv_list);
3403         }
3404
3405  poke:
3406         if (!ISC_LIST_EMPTY(sock->recv_list))
3407                 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
3408
3409         UNLOCK(&sock->lock);
3410 }
3411
3412 static void
3413 internal_send(isc_task_t *me, isc_event_t *ev) {
3414         isc_socketevent_t *dev;
3415         isc__socket_t *sock;
3416
3417         INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
3418
3419         /*
3420          * Find out what socket this is and lock it.
3421          */
3422         sock = (isc__socket_t *)ev->ev_sender;
3423         INSIST(VALID_SOCKET(sock));
3424
3425         LOCK(&sock->lock);
3426         socket_log(sock, NULL, IOEVENT,
3427                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALSEND,
3428                    "internal_send: task %p got event %p", me, ev);
3429
3430         INSIST(sock->pending_send == 1);
3431         sock->pending_send = 0;
3432
3433         INSIST(sock->references > 0);
3434         sock->references--;  /* the internal event is done with this socket */
3435         if (sock->references == 0) {
3436                 UNLOCK(&sock->lock);
3437                 destroy(&sock);
3438                 return;
3439         }
3440
3441         /*
3442          * Try to do as much I/O as possible on this socket.  There are no
3443          * limits here, currently.
3444          */
3445         dev = ISC_LIST_HEAD(sock->send_list);
3446         while (dev != NULL) {
3447                 switch (doio_send(sock, dev)) {
3448                 case DOIO_SOFT:
3449                         goto poke;
3450
3451                 case DOIO_HARD:
3452                 case DOIO_SUCCESS:
3453                         send_senddone_event(sock, &dev);
3454                         break;
3455                 }
3456
3457                 dev = ISC_LIST_HEAD(sock->send_list);
3458         }
3459
3460  poke:
3461         if (!ISC_LIST_EMPTY(sock->send_list))
3462                 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
3463
3464         UNLOCK(&sock->lock);
3465 }
3466
3467 static void
3468 internal_fdwatch_write(isc_task_t *me, isc_event_t *ev) {
3469         isc__socket_t *sock;
3470         int more_data;
3471
3472         INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
3473
3474         /*
3475          * Find out what socket this is and lock it.
3476          */
3477         sock = (isc__socket_t *)ev->ev_sender;
3478         INSIST(VALID_SOCKET(sock));
3479
3480         LOCK(&sock->lock);
3481         socket_log(sock, NULL, IOEVENT,
3482                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALSEND,
3483                    "internal_fdwatch_write: task %p got event %p", me, ev);
3484
3485         INSIST(sock->pending_send == 1);
3486
3487         UNLOCK(&sock->lock);
3488         more_data = (sock->fdwatchcb)(me, (isc_socket_t *)sock,
3489                                       sock->fdwatcharg, ISC_SOCKFDWATCH_WRITE);
3490         LOCK(&sock->lock);
3491
3492         sock->pending_send = 0;
3493
3494         INSIST(sock->references > 0);
3495         sock->references--;  /* the internal event is done with this socket */
3496         if (sock->references == 0) {
3497                 UNLOCK(&sock->lock);
3498                 destroy(&sock);
3499                 return;
3500         }
3501
3502         if (more_data)
3503                 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
3504
3505         UNLOCK(&sock->lock);
3506 }
3507
3508 static void
3509 internal_fdwatch_read(isc_task_t *me, isc_event_t *ev) {
3510         isc__socket_t *sock;
3511         int more_data;
3512
3513         INSIST(ev->ev_type == ISC_SOCKEVENT_INTR);
3514
3515         /*
3516          * Find out what socket this is and lock it.
3517          */
3518         sock = (isc__socket_t *)ev->ev_sender;
3519         INSIST(VALID_SOCKET(sock));
3520
3521         LOCK(&sock->lock);
3522         socket_log(sock, NULL, IOEVENT,
3523                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALRECV,
3524                    "internal_fdwatch_read: task %p got event %p", me, ev);
3525
3526         INSIST(sock->pending_recv == 1);
3527
3528         UNLOCK(&sock->lock);
3529         more_data = (sock->fdwatchcb)(me, (isc_socket_t *)sock,
3530                                       sock->fdwatcharg, ISC_SOCKFDWATCH_READ);
3531         LOCK(&sock->lock);
3532
3533         sock->pending_recv = 0;
3534
3535         INSIST(sock->references > 0);
3536         sock->references--;  /* the internal event is done with this socket */
3537         if (sock->references == 0) {
3538                 UNLOCK(&sock->lock);
3539                 destroy(&sock);
3540                 return;
3541         }
3542
3543         if (more_data)
3544                 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
3545
3546         UNLOCK(&sock->lock);
3547 }
3548
3549 /*
3550  * Process read/writes on each fd here.  Avoid locking
3551  * and unlocking twice if both reads and writes are possible.
3552  */
3553 static void
3554 process_fd(isc__socketmgr_t *manager, int fd, isc_boolean_t readable,
3555            isc_boolean_t writeable)
3556 {
3557         isc__socket_t *sock;
3558         isc_boolean_t unlock_sock;
3559         isc_boolean_t unwatch_read = ISC_FALSE, unwatch_write = ISC_FALSE;
3560         int lockid = FDLOCK_ID(fd);
3561
3562         /*
3563          * If the socket is going to be closed, don't do more I/O.
3564          */
3565         LOCK(&manager->fdlock[lockid]);
3566         if (manager->fdstate[fd] == CLOSE_PENDING) {
3567                 UNLOCK(&manager->fdlock[lockid]);
3568
3569                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
3570                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
3571                 return;
3572         }
3573
3574         sock = manager->fds[fd];
3575         unlock_sock = ISC_FALSE;
3576         if (readable) {
3577                 if (sock == NULL) {
3578                         unwatch_read = ISC_TRUE;
3579                         goto check_write;
3580                 }
3581                 unlock_sock = ISC_TRUE;
3582                 LOCK(&sock->lock);
3583                 if (!SOCK_DEAD(sock)) {
3584                         if (sock->listener)
3585                                 dispatch_accept(sock);
3586                         else
3587                                 dispatch_recv(sock);
3588                 }
3589                 unwatch_read = ISC_TRUE;
3590         }
3591 check_write:
3592         if (writeable) {
3593                 if (sock == NULL) {
3594                         unwatch_write = ISC_TRUE;
3595                         goto unlock_fd;
3596                 }
3597                 if (!unlock_sock) {
3598                         unlock_sock = ISC_TRUE;
3599                         LOCK(&sock->lock);
3600                 }
3601                 if (!SOCK_DEAD(sock)) {
3602                         if (sock->connecting)
3603                                 dispatch_connect(sock);
3604                         else
3605                                 dispatch_send(sock);
3606                 }
3607                 unwatch_write = ISC_TRUE;
3608         }
3609         if (unlock_sock)
3610                 UNLOCK(&sock->lock);
3611
3612  unlock_fd:
3613         UNLOCK(&manager->fdlock[lockid]);
3614         if (unwatch_read)
3615                 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
3616         if (unwatch_write)
3617                 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
3618
3619 }
3620
3621 #ifdef USE_KQUEUE
3622 static isc_boolean_t
3623 process_fds(isc__socketmgr_t *manager, struct kevent *events, int nevents) {
3624         int i;
3625         isc_boolean_t readable, writable;
3626         isc_boolean_t done = ISC_FALSE;
3627 #ifdef USE_WATCHER_THREAD
3628         isc_boolean_t have_ctlevent = ISC_FALSE;
3629 #endif
3630
3631         if (nevents == manager->nevents) {
3632                 /*
3633                  * This is not an error, but something unexpected.  If this
3634                  * happens, it may indicate the need for increasing
3635                  * ISC_SOCKET_MAXEVENTS.
3636                  */
3637                 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3638                             ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3639                             "maximum number of FD events (%d) received",
3640                             nevents);
3641         }
3642
3643         for (i = 0; i < nevents; i++) {
3644                 REQUIRE(events[i].ident < manager->maxsocks);
3645 #ifdef USE_WATCHER_THREAD
3646                 if (events[i].ident == (uintptr_t)manager->pipe_fds[0]) {
3647                         have_ctlevent = ISC_TRUE;
3648                         continue;
3649                 }
3650 #endif
3651                 readable = ISC_TF(events[i].filter == EVFILT_READ);
3652                 writable = ISC_TF(events[i].filter == EVFILT_WRITE);
3653                 process_fd(manager, events[i].ident, readable, writable);
3654         }
3655
3656 #ifdef USE_WATCHER_THREAD
3657         if (have_ctlevent)
3658                 done = process_ctlfd(manager);
3659 #endif
3660
3661         return (done);
3662 }
3663 #elif defined(USE_EPOLL)
3664 static isc_boolean_t
3665 process_fds(isc__socketmgr_t *manager, struct epoll_event *events, int nevents)
3666 {
3667         int i;
3668         isc_boolean_t done = ISC_FALSE;
3669 #ifdef USE_WATCHER_THREAD
3670         isc_boolean_t have_ctlevent = ISC_FALSE;
3671 #endif
3672
3673         if (nevents == manager->nevents) {
3674                 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3675                             ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3676                             "maximum number of FD events (%d) received",
3677                             nevents);
3678         }
3679
3680         for (i = 0; i < nevents; i++) {
3681                 REQUIRE(events[i].data.fd < (int)manager->maxsocks);
3682 #ifdef USE_WATCHER_THREAD
3683                 if (events[i].data.fd == manager->pipe_fds[0]) {
3684                         have_ctlevent = ISC_TRUE;
3685                         continue;
3686                 }
3687 #endif
3688                 if ((events[i].events & EPOLLERR) != 0 ||
3689                     (events[i].events & EPOLLHUP) != 0) {
3690                         /*
3691                          * epoll does not set IN/OUT bits on an erroneous
3692                          * condition, so we need to try both anyway.  This is a
3693                          * bit inefficient, but should be okay for such rare
3694                          * events.  Note also that the read or write attempt
3695                          * won't block because we use non-blocking sockets.
3696                          */
3697                         events[i].events |= (EPOLLIN | EPOLLOUT);
3698                 }
3699                 process_fd(manager, events[i].data.fd,
3700                            (events[i].events & EPOLLIN) != 0,
3701                            (events[i].events & EPOLLOUT) != 0);
3702         }
3703
3704 #ifdef USE_WATCHER_THREAD
3705         if (have_ctlevent)
3706                 done = process_ctlfd(manager);
3707 #endif
3708
3709         return (done);
3710 }
3711 #elif defined(USE_DEVPOLL)
3712 static isc_boolean_t
3713 process_fds(isc__socketmgr_t *manager, struct pollfd *events, int nevents) {
3714         int i;
3715         isc_boolean_t done = ISC_FALSE;
3716 #ifdef USE_WATCHER_THREAD
3717         isc_boolean_t have_ctlevent = ISC_FALSE;
3718 #endif
3719
3720         if (nevents == manager->nevents) {
3721                 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3722                             ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3723                             "maximum number of FD events (%d) received",
3724                             nevents);
3725         }
3726
3727         for (i = 0; i < nevents; i++) {
3728                 REQUIRE(events[i].fd < (int)manager->maxsocks);
3729 #ifdef USE_WATCHER_THREAD
3730                 if (events[i].fd == manager->pipe_fds[0]) {
3731                         have_ctlevent = ISC_TRUE;
3732                         continue;
3733                 }
3734 #endif
3735                 process_fd(manager, events[i].fd,
3736                            (events[i].events & POLLIN) != 0,
3737                            (events[i].events & POLLOUT) != 0);
3738         }
3739
3740 #ifdef USE_WATCHER_THREAD
3741         if (have_ctlevent)
3742                 done = process_ctlfd(manager);
3743 #endif
3744
3745         return (done);
3746 }
3747 #elif defined(USE_SELECT)
3748 static void
3749 process_fds(isc__socketmgr_t *manager, int maxfd, fd_set *readfds,
3750             fd_set *writefds)
3751 {
3752         int i;
3753
3754         REQUIRE(maxfd <= (int)manager->maxsocks);
3755
3756         for (i = 0; i < maxfd; i++) {
3757 #ifdef USE_WATCHER_THREAD
3758                 if (i == manager->pipe_fds[0] || i == manager->pipe_fds[1])
3759                         continue;
3760 #endif /* USE_WATCHER_THREAD */
3761                 process_fd(manager, i, FD_ISSET(i, readfds),
3762                            FD_ISSET(i, writefds));
3763         }
3764 }
3765 #endif
3766
3767 #ifdef USE_WATCHER_THREAD
3768 static isc_boolean_t
3769 process_ctlfd(isc__socketmgr_t *manager) {
3770         int msg, fd;
3771
3772         for (;;) {
3773                 select_readmsg(manager, &fd, &msg);
3774
3775                 manager_log(manager, IOEVENT,
3776                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
3777                                            ISC_MSG_WATCHERMSG,
3778                                            "watcher got message %d "
3779                                            "for socket %d"), msg, fd);
3780
3781                 /*
3782                  * Nothing to read?
3783                  */
3784                 if (msg == SELECT_POKE_NOTHING)
3785                         break;
3786
3787                 /*
3788                  * Handle shutdown message.  We really should
3789                  * jump out of this loop right away, but
3790                  * it doesn't matter if we have to do a little
3791                  * more work first.
3792                  */
3793                 if (msg == SELECT_POKE_SHUTDOWN)
3794                         return (ISC_TRUE);
3795
3796                 /*
3797                  * This is a wakeup on a socket.  Look
3798                  * at the event queue for both read and write,
3799                  * and decide if we need to watch on it now
3800                  * or not.
3801                  */
3802                 wakeup_socket(manager, fd, msg);
3803         }
3804
3805         return (ISC_FALSE);
3806 }
3807
3808 /*
3809  * This is the thread that will loop forever, always in a select or poll
3810  * call.
3811  *
3812  * When select returns something to do, track down what thread gets to do
3813  * this I/O and post the event to it.
3814  */
3815 static isc_threadresult_t
3816 watcher(void *uap) {
3817         isc__socketmgr_t *manager = uap;
3818         isc_boolean_t done;
3819         int cc;
3820 #ifdef USE_KQUEUE
3821         const char *fnname = "kevent()";
3822 #elif defined (USE_EPOLL)
3823         const char *fnname = "epoll_wait()";
3824 #elif defined(USE_DEVPOLL)
3825         const char *fnname = "ioctl(DP_POLL)";
3826         struct dvpoll dvp;
3827 #elif defined (USE_SELECT)
3828         const char *fnname = "select()";
3829         int maxfd;
3830         int ctlfd;
3831 #endif
3832         char strbuf[ISC_STRERRORSIZE];
3833 #ifdef ISC_SOCKET_USE_POLLWATCH
3834         pollstate_t pollstate = poll_idle;
3835 #endif
3836
3837 #if defined (USE_SELECT)
3838         /*
3839          * Get the control fd here.  This will never change.
3840          */
3841         ctlfd = manager->pipe_fds[0];
3842 #endif
3843         done = ISC_FALSE;
3844         while (!done) {
3845                 do {
3846 #ifdef USE_KQUEUE
3847                         cc = kevent(manager->kqueue_fd, NULL, 0,
3848                                     manager->events, manager->nevents, NULL);
3849 #elif defined(USE_EPOLL)
3850                         cc = epoll_wait(manager->epoll_fd, manager->events,
3851                                         manager->nevents, -1);
3852 #elif defined(USE_DEVPOLL)
3853                         dvp.dp_fds = manager->events;
3854                         dvp.dp_nfds = manager->nevents;
3855 #ifndef ISC_SOCKET_USE_POLLWATCH
3856                         dvp.dp_timeout = -1;
3857 #else
3858                         if (pollstate == poll_idle)
3859                                 dvp.dp_timeout = -1;
3860                         else
3861                                 dvp.dp_timeout = ISC_SOCKET_POLLWATCH_TIMEOUT;
3862 #endif  /* ISC_SOCKET_USE_POLLWATCH */
3863                         cc = ioctl(manager->devpoll_fd, DP_POLL, &dvp);
3864 #elif defined(USE_SELECT)
3865                         LOCK(&manager->lock);
3866                         memmove(manager->read_fds_copy, manager->read_fds,
3867                                 manager->fd_bufsize);
3868                         memmove(manager->write_fds_copy, manager->write_fds,
3869                                 manager->fd_bufsize);
3870                         maxfd = manager->maxfd + 1;
3871                         UNLOCK(&manager->lock);
3872
3873                         cc = select(maxfd, manager->read_fds_copy,
3874                                     manager->write_fds_copy, NULL, NULL);
3875 #endif  /* USE_KQUEUE */
3876
3877                         if (cc < 0 && !SOFT_ERROR(errno)) {
3878                                 isc__strerror(errno, strbuf, sizeof(strbuf));
3879                                 FATAL_ERROR(__FILE__, __LINE__,
3880                                             "%s %s: %s", fnname,
3881                                             isc_msgcat_get(isc_msgcat,
3882                                                            ISC_MSGSET_GENERAL,
3883                                                            ISC_MSG_FAILED,
3884                                                            "failed"), strbuf);
3885                         }
3886
3887 #if defined(USE_DEVPOLL) && defined(ISC_SOCKET_USE_POLLWATCH)
3888                         if (cc == 0) {
3889                                 if (pollstate == poll_active)
3890                                         pollstate = poll_checking;
3891                                 else if (pollstate == poll_checking)
3892                                         pollstate = poll_idle;
3893                         } else if (cc > 0) {
3894                                 if (pollstate == poll_checking) {
3895                                         /*
3896                                          * XXX: We'd like to use a more
3897                                          * verbose log level as it's actually an
3898                                          * unexpected event, but the kernel bug
3899                                          * reportedly happens pretty frequently
3900                                          * (and it can also be a false positive)
3901                                          * so it would be just too noisy.
3902                                          */
3903                                         manager_log(manager,
3904                                                     ISC_LOGCATEGORY_GENERAL,
3905                                                     ISC_LOGMODULE_SOCKET,
3906                                                     ISC_LOG_DEBUG(1),
3907                                                     "unexpected POLL timeout");
3908                                 }
3909                                 pollstate = poll_active;
3910                         }
3911 #endif
3912                 } while (cc < 0);
3913
3914 #if defined(USE_KQUEUE) || defined (USE_EPOLL) || defined (USE_DEVPOLL)
3915                 done = process_fds(manager, manager->events, cc);
3916 #elif defined(USE_SELECT)
3917                 process_fds(manager, maxfd, manager->read_fds_copy,
3918                             manager->write_fds_copy);
3919
3920                 /*
3921                  * Process reads on internal, control fd.
3922                  */
3923                 if (FD_ISSET(ctlfd, manager->read_fds_copy))
3924                         done = process_ctlfd(manager);
3925 #endif
3926         }
3927
3928         manager_log(manager, TRACE, "%s",
3929                     isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3930                                    ISC_MSG_EXITING, "watcher exiting"));
3931
3932         return ((isc_threadresult_t)0);
3933 }
3934 #endif /* USE_WATCHER_THREAD */
3935
3936 #ifdef BIND9
3937 ISC_SOCKETFUNC_SCOPE void
3938 isc__socketmgr_setreserved(isc_socketmgr_t *manager0, isc_uint32_t reserved) {
3939         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
3940
3941         REQUIRE(VALID_MANAGER(manager));
3942
3943         manager->reserved = reserved;
3944 }
3945
3946 ISC_SOCKETFUNC_SCOPE void
3947 isc___socketmgr_maxudp(isc_socketmgr_t *manager0, int maxudp) {
3948         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
3949
3950         REQUIRE(VALID_MANAGER(manager));
3951
3952         manager->maxudp = maxudp;
3953 }
3954 #endif  /* BIND9 */
3955
3956 /*
3957  * Create a new socket manager.
3958  */
3959
3960 static isc_result_t
3961 setup_watcher(isc_mem_t *mctx, isc__socketmgr_t *manager) {
3962         isc_result_t result;
3963 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
3964         char strbuf[ISC_STRERRORSIZE];
3965 #endif
3966
3967 #ifdef USE_KQUEUE
3968         manager->nevents = ISC_SOCKET_MAXEVENTS;
3969         manager->events = isc_mem_get(mctx, sizeof(struct kevent) *
3970                                       manager->nevents);
3971         if (manager->events == NULL)
3972                 return (ISC_R_NOMEMORY);
3973         manager->kqueue_fd = kqueue();
3974         if (manager->kqueue_fd == -1) {
3975                 result = isc__errno2result(errno);
3976                 isc__strerror(errno, strbuf, sizeof(strbuf));
3977                 UNEXPECTED_ERROR(__FILE__, __LINE__,
3978                                  "kqueue %s: %s",
3979                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3980                                                 ISC_MSG_FAILED, "failed"),
3981                                  strbuf);
3982                 isc_mem_put(mctx, manager->events,
3983                             sizeof(struct kevent) * manager->nevents);
3984                 return (result);
3985         }
3986
3987 #ifdef USE_WATCHER_THREAD
3988         result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3989         if (result != ISC_R_SUCCESS) {
3990                 close(manager->kqueue_fd);
3991                 isc_mem_put(mctx, manager->events,
3992                             sizeof(struct kevent) * manager->nevents);
3993                 return (result);
3994         }
3995 #endif  /* USE_WATCHER_THREAD */
3996 #elif defined(USE_EPOLL)
3997         manager->nevents = ISC_SOCKET_MAXEVENTS;
3998         manager->events = isc_mem_get(mctx, sizeof(struct epoll_event) *
3999                                       manager->nevents);
4000         if (manager->events == NULL)
4001                 return (ISC_R_NOMEMORY);
4002         manager->epoll_fd = epoll_create(manager->nevents);
4003         if (manager->epoll_fd == -1) {
4004                 result = isc__errno2result(errno);
4005                 isc__strerror(errno, strbuf, sizeof(strbuf));
4006                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4007                                  "epoll_create %s: %s",
4008                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4009                                                 ISC_MSG_FAILED, "failed"),
4010                                  strbuf);
4011                 isc_mem_put(mctx, manager->events,
4012                             sizeof(struct epoll_event) * manager->nevents);
4013                 return (result);
4014         }
4015 #ifdef USE_WATCHER_THREAD
4016         result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
4017         if (result != ISC_R_SUCCESS) {
4018                 close(manager->epoll_fd);
4019                 isc_mem_put(mctx, manager->events,
4020                             sizeof(struct epoll_event) * manager->nevents);
4021                 return (result);
4022         }
4023 #endif  /* USE_WATCHER_THREAD */
4024 #elif defined(USE_DEVPOLL)
4025         /*
4026          * XXXJT: /dev/poll seems to reject large numbers of events,
4027          * so we should be careful about redefining ISC_SOCKET_MAXEVENTS.
4028          */
4029         manager->nevents = ISC_SOCKET_MAXEVENTS;
4030         manager->events = isc_mem_get(mctx, sizeof(struct pollfd) *
4031                                       manager->nevents);
4032         if (manager->events == NULL)
4033                 return (ISC_R_NOMEMORY);
4034         /*
4035          * Note: fdpollinfo should be able to support all possible FDs, so
4036          * it must have maxsocks entries (not nevents).
4037          */
4038         manager->fdpollinfo = isc_mem_get(mctx, sizeof(pollinfo_t) *
4039                                           manager->maxsocks);
4040         if (manager->fdpollinfo == NULL) {
4041                 isc_mem_put(mctx, manager->events,
4042                             sizeof(struct pollfd) * manager->nevents);
4043                 return (ISC_R_NOMEMORY);
4044         }
4045         memset(manager->fdpollinfo, 0, sizeof(pollinfo_t) * manager->maxsocks);
4046         manager->devpoll_fd = open("/dev/poll", O_RDWR);
4047         if (manager->devpoll_fd == -1) {
4048                 result = isc__errno2result(errno);
4049                 isc__strerror(errno, strbuf, sizeof(strbuf));
4050                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4051                                  "open(/dev/poll) %s: %s",
4052                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4053                                                 ISC_MSG_FAILED, "failed"),
4054                                  strbuf);
4055                 isc_mem_put(mctx, manager->events,
4056                             sizeof(struct pollfd) * manager->nevents);
4057                 isc_mem_put(mctx, manager->fdpollinfo,
4058                             sizeof(pollinfo_t) * manager->maxsocks);
4059                 return (result);
4060         }
4061 #ifdef USE_WATCHER_THREAD
4062         result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
4063         if (result != ISC_R_SUCCESS) {
4064                 close(manager->devpoll_fd);
4065                 isc_mem_put(mctx, manager->events,
4066                             sizeof(struct pollfd) * manager->nevents);
4067                 isc_mem_put(mctx, manager->fdpollinfo,
4068                             sizeof(pollinfo_t) * manager->maxsocks);
4069                 return (result);
4070         }
4071 #endif  /* USE_WATCHER_THREAD */
4072 #elif defined(USE_SELECT)
4073         UNUSED(result);
4074
4075 #if ISC_SOCKET_MAXSOCKETS > FD_SETSIZE
4076         /*
4077          * Note: this code should also cover the case of MAXSOCKETS <=
4078          * FD_SETSIZE, but we separate the cases to avoid possible portability
4079          * issues regarding howmany() and the actual representation of fd_set.
4080          */
4081         manager->fd_bufsize = howmany(manager->maxsocks, NFDBITS) *
4082                 sizeof(fd_mask);
4083 #else
4084         manager->fd_bufsize = sizeof(fd_set);
4085 #endif
4086
4087         manager->read_fds = NULL;
4088         manager->read_fds_copy = NULL;
4089         manager->write_fds = NULL;
4090         manager->write_fds_copy = NULL;
4091
4092         manager->read_fds = isc_mem_get(mctx, manager->fd_bufsize);
4093         if (manager->read_fds != NULL)
4094                 manager->read_fds_copy = isc_mem_get(mctx, manager->fd_bufsize);
4095         if (manager->read_fds_copy != NULL)
4096                 manager->write_fds = isc_mem_get(mctx, manager->fd_bufsize);
4097         if (manager->write_fds != NULL) {
4098                 manager->write_fds_copy = isc_mem_get(mctx,
4099                                                       manager->fd_bufsize);
4100         }
4101         if (manager->write_fds_copy == NULL) {
4102                 if (manager->write_fds != NULL) {
4103                         isc_mem_put(mctx, manager->write_fds,
4104                                     manager->fd_bufsize);
4105                 }
4106                 if (manager->read_fds_copy != NULL) {
4107                         isc_mem_put(mctx, manager->read_fds_copy,
4108                                     manager->fd_bufsize);
4109                 }
4110                 if (manager->read_fds != NULL) {
4111                         isc_mem_put(mctx, manager->read_fds,
4112                                     manager->fd_bufsize);
4113                 }
4114                 return (ISC_R_NOMEMORY);
4115         }
4116         memset(manager->read_fds, 0, manager->fd_bufsize);
4117         memset(manager->write_fds, 0, manager->fd_bufsize);
4118
4119 #ifdef USE_WATCHER_THREAD
4120         (void)watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
4121         manager->maxfd = manager->pipe_fds[0];
4122 #else /* USE_WATCHER_THREAD */
4123         manager->maxfd = 0;
4124 #endif /* USE_WATCHER_THREAD */
4125 #endif  /* USE_KQUEUE */
4126
4127         return (ISC_R_SUCCESS);
4128 }
4129
4130 static void
4131 cleanup_watcher(isc_mem_t *mctx, isc__socketmgr_t *manager) {
4132 #ifdef USE_WATCHER_THREAD
4133         isc_result_t result;
4134
4135         result = unwatch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
4136         if (result != ISC_R_SUCCESS) {
4137                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4138                                  "epoll_ctl(DEL) %s",
4139                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4140                                                 ISC_MSG_FAILED, "failed"));
4141         }
4142 #endif  /* USE_WATCHER_THREAD */
4143
4144 #ifdef USE_KQUEUE
4145         close(manager->kqueue_fd);
4146         isc_mem_put(mctx, manager->events,
4147                     sizeof(struct kevent) * manager->nevents);
4148 #elif defined(USE_EPOLL)
4149         close(manager->epoll_fd);
4150         isc_mem_put(mctx, manager->events,
4151                     sizeof(struct epoll_event) * manager->nevents);
4152 #elif defined(USE_DEVPOLL)
4153         close(manager->devpoll_fd);
4154         isc_mem_put(mctx, manager->events,
4155                     sizeof(struct pollfd) * manager->nevents);
4156         isc_mem_put(mctx, manager->fdpollinfo,
4157                     sizeof(pollinfo_t) * manager->maxsocks);
4158 #elif defined(USE_SELECT)
4159         if (manager->read_fds != NULL)
4160                 isc_mem_put(mctx, manager->read_fds, manager->fd_bufsize);
4161         if (manager->read_fds_copy != NULL)
4162                 isc_mem_put(mctx, manager->read_fds_copy, manager->fd_bufsize);
4163         if (manager->write_fds != NULL)
4164                 isc_mem_put(mctx, manager->write_fds, manager->fd_bufsize);
4165         if (manager->write_fds_copy != NULL)
4166                 isc_mem_put(mctx, manager->write_fds_copy, manager->fd_bufsize);
4167 #endif  /* USE_KQUEUE */
4168 }
4169
4170 ISC_SOCKETFUNC_SCOPE isc_result_t
4171 isc__socketmgr_create(isc_mem_t *mctx, isc_socketmgr_t **managerp) {
4172         return (isc__socketmgr_create2(mctx, managerp, 0));
4173 }
4174
4175 ISC_SOCKETFUNC_SCOPE isc_result_t
4176 isc__socketmgr_create2(isc_mem_t *mctx, isc_socketmgr_t **managerp,
4177                        unsigned int maxsocks)
4178 {
4179         int i;
4180         isc__socketmgr_t *manager;
4181 #ifdef USE_WATCHER_THREAD
4182         char strbuf[ISC_STRERRORSIZE];
4183 #endif
4184         isc_result_t result;
4185
4186         REQUIRE(managerp != NULL && *managerp == NULL);
4187
4188 #ifdef USE_SHARED_MANAGER
4189         if (socketmgr != NULL) {
4190                 /* Don't allow maxsocks to be updated */
4191                 if (maxsocks > 0 && socketmgr->maxsocks != maxsocks)
4192                         return (ISC_R_EXISTS);
4193
4194                 socketmgr->refs++;
4195                 *managerp = (isc_socketmgr_t *)socketmgr;
4196                 return (ISC_R_SUCCESS);
4197         }
4198 #endif /* USE_SHARED_MANAGER */
4199
4200         if (maxsocks == 0)
4201                 maxsocks = ISC_SOCKET_MAXSOCKETS;
4202
4203         manager = isc_mem_get(mctx, sizeof(*manager));
4204         if (manager == NULL)
4205                 return (ISC_R_NOMEMORY);
4206
4207         /* zero-clear so that necessary cleanup on failure will be easy */
4208         memset(manager, 0, sizeof(*manager));
4209         manager->maxsocks = maxsocks;
4210         manager->reserved = 0;
4211         manager->maxudp = 0;
4212         manager->fds = isc_mem_get(mctx,
4213                                    manager->maxsocks * sizeof(isc__socket_t *));
4214         if (manager->fds == NULL) {
4215                 result = ISC_R_NOMEMORY;
4216                 goto free_manager;
4217         }
4218         manager->fdstate = isc_mem_get(mctx, manager->maxsocks * sizeof(int));
4219         if (manager->fdstate == NULL) {
4220                 result = ISC_R_NOMEMORY;
4221                 goto free_manager;
4222         }
4223         manager->stats = NULL;
4224
4225         manager->common.methods = &socketmgrmethods;
4226         manager->common.magic = ISCAPI_SOCKETMGR_MAGIC;
4227         manager->common.impmagic = SOCKET_MANAGER_MAGIC;
4228         manager->mctx = NULL;
4229         memset(manager->fds, 0, manager->maxsocks * sizeof(isc_socket_t *));
4230         ISC_LIST_INIT(manager->socklist);
4231         result = isc_mutex_init(&manager->lock);
4232         if (result != ISC_R_SUCCESS)
4233                 goto free_manager;
4234         manager->fdlock = isc_mem_get(mctx, FDLOCK_COUNT * sizeof(isc_mutex_t));
4235         if (manager->fdlock == NULL) {
4236                 result = ISC_R_NOMEMORY;
4237                 goto cleanup_lock;
4238         }
4239         for (i = 0; i < FDLOCK_COUNT; i++) {
4240                 result = isc_mutex_init(&manager->fdlock[i]);
4241                 if (result != ISC_R_SUCCESS) {
4242                         while (--i >= 0)
4243                                 DESTROYLOCK(&manager->fdlock[i]);
4244                         isc_mem_put(mctx, manager->fdlock,
4245                                     FDLOCK_COUNT * sizeof(isc_mutex_t));
4246                         manager->fdlock = NULL;
4247                         goto cleanup_lock;
4248                 }
4249         }
4250
4251 #ifdef USE_WATCHER_THREAD
4252         if (isc_condition_init(&manager->shutdown_ok) != ISC_R_SUCCESS) {
4253                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4254                                  "isc_condition_init() %s",
4255                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4256                                                 ISC_MSG_FAILED, "failed"));
4257                 result = ISC_R_UNEXPECTED;
4258                 goto cleanup_lock;
4259         }
4260
4261         /*
4262          * Create the special fds that will be used to wake up the
4263          * select/poll loop when something internal needs to be done.
4264          */
4265         if (pipe(manager->pipe_fds) != 0) {
4266                 isc__strerror(errno, strbuf, sizeof(strbuf));
4267                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4268                                  "pipe() %s: %s",
4269                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4270                                                 ISC_MSG_FAILED, "failed"),
4271                                  strbuf);
4272                 result = ISC_R_UNEXPECTED;
4273                 goto cleanup_condition;
4274         }
4275
4276         RUNTIME_CHECK(make_nonblock(manager->pipe_fds[0]) == ISC_R_SUCCESS);
4277 #if 0
4278         RUNTIME_CHECK(make_nonblock(manager->pipe_fds[1]) == ISC_R_SUCCESS);
4279 #endif
4280 #endif  /* USE_WATCHER_THREAD */
4281
4282 #ifdef USE_SHARED_MANAGER
4283         manager->refs = 1;
4284 #endif /* USE_SHARED_MANAGER */
4285
4286         /*
4287          * Set up initial state for the select loop
4288          */
4289         result = setup_watcher(mctx, manager);
4290         if (result != ISC_R_SUCCESS)
4291                 goto cleanup;
4292         memset(manager->fdstate, 0, manager->maxsocks * sizeof(int));
4293 #ifdef USE_WATCHER_THREAD
4294         /*
4295          * Start up the select/poll thread.
4296          */
4297         if (isc_thread_create(watcher, manager, &manager->watcher) !=
4298             ISC_R_SUCCESS) {
4299                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4300                                  "isc_thread_create() %s",
4301                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4302                                                 ISC_MSG_FAILED, "failed"));
4303                 cleanup_watcher(mctx, manager);
4304                 result = ISC_R_UNEXPECTED;
4305                 goto cleanup;
4306         }
4307 #endif /* USE_WATCHER_THREAD */
4308         isc_mem_attach(mctx, &manager->mctx);
4309
4310 #ifdef USE_SHARED_MANAGER
4311         socketmgr = manager;
4312 #endif /* USE_SHARED_MANAGER */
4313         *managerp = (isc_socketmgr_t *)manager;
4314
4315         return (ISC_R_SUCCESS);
4316
4317 cleanup:
4318 #ifdef USE_WATCHER_THREAD
4319         (void)close(manager->pipe_fds[0]);
4320         (void)close(manager->pipe_fds[1]);
4321 #endif  /* USE_WATCHER_THREAD */
4322
4323 #ifdef USE_WATCHER_THREAD
4324 cleanup_condition:
4325         (void)isc_condition_destroy(&manager->shutdown_ok);
4326 #endif  /* USE_WATCHER_THREAD */
4327
4328
4329 cleanup_lock:
4330         if (manager->fdlock != NULL) {
4331                 for (i = 0; i < FDLOCK_COUNT; i++)
4332                         DESTROYLOCK(&manager->fdlock[i]);
4333         }
4334         DESTROYLOCK(&manager->lock);
4335
4336 free_manager:
4337         if (manager->fdlock != NULL) {
4338                 isc_mem_put(mctx, manager->fdlock,
4339                             FDLOCK_COUNT * sizeof(isc_mutex_t));
4340         }
4341         if (manager->fdstate != NULL) {
4342                 isc_mem_put(mctx, manager->fdstate,
4343                             manager->maxsocks * sizeof(int));
4344         }
4345         if (manager->fds != NULL) {
4346                 isc_mem_put(mctx, manager->fds,
4347                             manager->maxsocks * sizeof(isc_socket_t *));
4348         }
4349         isc_mem_put(mctx, manager, sizeof(*manager));
4350
4351         return (result);
4352 }
4353
4354 #ifdef BIND9
4355 isc_result_t
4356 isc__socketmgr_getmaxsockets(isc_socketmgr_t *manager0, unsigned int *nsockp) {
4357         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
4358         REQUIRE(VALID_MANAGER(manager));
4359         REQUIRE(nsockp != NULL);
4360
4361         *nsockp = manager->maxsocks;
4362
4363         return (ISC_R_SUCCESS);
4364 }
4365
4366 void
4367 isc__socketmgr_setstats(isc_socketmgr_t *manager0, isc_stats_t *stats) {
4368         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
4369
4370         REQUIRE(VALID_MANAGER(manager));
4371         REQUIRE(ISC_LIST_EMPTY(manager->socklist));
4372         REQUIRE(manager->stats == NULL);
4373         REQUIRE(isc_stats_ncounters(stats) == isc_sockstatscounter_max);
4374
4375         isc_stats_attach(stats, &manager->stats);
4376 }
4377 #endif
4378
4379 ISC_SOCKETFUNC_SCOPE void
4380 isc__socketmgr_destroy(isc_socketmgr_t **managerp) {
4381         isc__socketmgr_t *manager;
4382         int i;
4383         isc_mem_t *mctx;
4384
4385         /*
4386          * Destroy a socket manager.
4387          */
4388
4389         REQUIRE(managerp != NULL);
4390         manager = (isc__socketmgr_t *)*managerp;
4391         REQUIRE(VALID_MANAGER(manager));
4392
4393 #ifdef USE_SHARED_MANAGER
4394         manager->refs--;
4395         if (manager->refs > 0) {
4396                 *managerp = NULL;
4397                 return;
4398         }
4399         socketmgr = NULL;
4400 #endif /* USE_SHARED_MANAGER */
4401
4402         LOCK(&manager->lock);
4403
4404         /*
4405          * Wait for all sockets to be destroyed.
4406          */
4407         while (!ISC_LIST_EMPTY(manager->socklist)) {
4408 #ifdef USE_WATCHER_THREAD
4409                 manager_log(manager, CREATION, "%s",
4410                             isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
4411                                            ISC_MSG_SOCKETSREMAIN,
4412                                            "sockets exist"));
4413                 WAIT(&manager->shutdown_ok, &manager->lock);
4414 #else /* USE_WATCHER_THREAD */
4415                 UNLOCK(&manager->lock);
4416                 isc__taskmgr_dispatch(NULL);
4417                 LOCK(&manager->lock);
4418 #endif /* USE_WATCHER_THREAD */
4419         }
4420
4421         UNLOCK(&manager->lock);
4422
4423         /*
4424          * Here, poke our select/poll thread.  Do this by closing the write
4425          * half of the pipe, which will send EOF to the read half.
4426          * This is currently a no-op in the non-threaded case.
4427          */
4428         select_poke(manager, 0, SELECT_POKE_SHUTDOWN);
4429
4430 #ifdef USE_WATCHER_THREAD
4431         /*
4432          * Wait for thread to exit.
4433          */
4434         if (isc_thread_join(manager->watcher, NULL) != ISC_R_SUCCESS)
4435                 UNEXPECTED_ERROR(__FILE__, __LINE__,
4436                                  "isc_thread_join() %s",
4437                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4438                                                 ISC_MSG_FAILED, "failed"));
4439 #endif /* USE_WATCHER_THREAD */
4440
4441         /*
4442          * Clean up.
4443          */
4444         cleanup_watcher(manager->mctx, manager);
4445
4446 #ifdef USE_WATCHER_THREAD
4447         (void)close(manager->pipe_fds[0]);
4448         (void)close(manager->pipe_fds[1]);
4449         (void)isc_condition_destroy(&manager->shutdown_ok);
4450 #endif /* USE_WATCHER_THREAD */
4451
4452         for (i = 0; i < (int)manager->maxsocks; i++)
4453                 if (manager->fdstate[i] == CLOSE_PENDING) /* no need to lock */
4454                         (void)close(i);
4455
4456         isc_mem_put(manager->mctx, manager->fds,
4457                     manager->maxsocks * sizeof(isc__socket_t *));
4458         isc_mem_put(manager->mctx, manager->fdstate,
4459                     manager->maxsocks * sizeof(int));
4460
4461         if (manager->stats != NULL)
4462                 isc_stats_detach(&manager->stats);
4463
4464         if (manager->fdlock != NULL) {
4465                 for (i = 0; i < FDLOCK_COUNT; i++)
4466                         DESTROYLOCK(&manager->fdlock[i]);
4467                 isc_mem_put(manager->mctx, manager->fdlock,
4468                             FDLOCK_COUNT * sizeof(isc_mutex_t));
4469         }
4470         DESTROYLOCK(&manager->lock);
4471         manager->common.magic = 0;
4472         manager->common.impmagic = 0;
4473         mctx= manager->mctx;
4474         isc_mem_put(mctx, manager, sizeof(*manager));
4475
4476         isc_mem_detach(&mctx);
4477
4478         *managerp = NULL;
4479
4480 #ifdef USE_SHARED_MANAGER
4481         socketmgr = NULL;
4482 #endif
4483 }
4484
4485 static isc_result_t
4486 socket_recv(isc__socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
4487             unsigned int flags)
4488 {
4489         int io_state;
4490         isc_boolean_t have_lock = ISC_FALSE;
4491         isc_task_t *ntask = NULL;
4492         isc_result_t result = ISC_R_SUCCESS;
4493
4494         dev->ev_sender = task;
4495
4496         if (sock->type == isc_sockettype_udp) {
4497                 io_state = doio_recv(sock, dev);
4498         } else {
4499                 LOCK(&sock->lock);
4500                 have_lock = ISC_TRUE;
4501
4502                 if (ISC_LIST_EMPTY(sock->recv_list))
4503                         io_state = doio_recv(sock, dev);
4504                 else
4505                         io_state = DOIO_SOFT;
4506         }
4507
4508         switch (io_state) {
4509         case DOIO_SOFT:
4510                 /*
4511                  * We couldn't read all or part of the request right now, so
4512                  * queue it.
4513                  *
4514                  * Attach to socket and to task
4515                  */
4516                 isc_task_attach(task, &ntask);
4517                 dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
4518
4519                 if (!have_lock) {
4520                         LOCK(&sock->lock);
4521                         have_lock = ISC_TRUE;
4522                 }
4523
4524                 /*
4525                  * Enqueue the request.  If the socket was previously not being
4526                  * watched, poke the watcher to start paying attention to it.
4527                  */
4528                 if (ISC_LIST_EMPTY(sock->recv_list) && !sock->pending_recv)
4529                         select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
4530                 ISC_LIST_ENQUEUE(sock->recv_list, dev, ev_link);
4531
4532                 socket_log(sock, NULL, EVENT, NULL, 0, 0,
4533                            "socket_recv: event %p -> task %p",
4534                            dev, ntask);
4535
4536                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
4537                         result = ISC_R_INPROGRESS;
4538                 break;
4539
4540         case DOIO_EOF:
4541                 dev->result = ISC_R_EOF;
4542                 /* fallthrough */
4543
4544         case DOIO_HARD:
4545         case DOIO_SUCCESS:
4546                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
4547                         send_recvdone_event(sock, &dev);
4548                 break;
4549         }
4550
4551         if (have_lock)
4552                 UNLOCK(&sock->lock);
4553
4554         return (result);
4555 }
4556
4557 ISC_SOCKETFUNC_SCOPE isc_result_t
4558 isc__socket_recvv(isc_socket_t *sock0, isc_bufferlist_t *buflist,
4559                   unsigned int minimum, isc_task_t *task,
4560                   isc_taskaction_t action, const void *arg)
4561 {
4562         isc__socket_t *sock = (isc__socket_t *)sock0;
4563         isc_socketevent_t *dev;
4564         isc__socketmgr_t *manager;
4565         unsigned int iocount;
4566         isc_buffer_t *buffer;
4567
4568         REQUIRE(VALID_SOCKET(sock));
4569         REQUIRE(buflist != NULL);
4570         REQUIRE(!ISC_LIST_EMPTY(*buflist));
4571         REQUIRE(task != NULL);
4572         REQUIRE(action != NULL);
4573
4574         manager = sock->manager;
4575         REQUIRE(VALID_MANAGER(manager));
4576
4577         iocount = isc_bufferlist_availablecount(buflist);
4578         REQUIRE(iocount > 0);
4579
4580         INSIST(sock->bound);
4581
4582         dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
4583         if (dev == NULL)
4584                 return (ISC_R_NOMEMORY);
4585
4586         /*
4587          * UDP sockets are always partial read
4588          */
4589         if (sock->type == isc_sockettype_udp)
4590                 dev->minimum = 1;
4591         else {
4592                 if (minimum == 0)
4593                         dev->minimum = iocount;
4594                 else
4595                         dev->minimum = minimum;
4596         }
4597
4598         /*
4599          * Move each buffer from the passed in list to our internal one.
4600          */
4601         buffer = ISC_LIST_HEAD(*buflist);
4602         while (buffer != NULL) {
4603                 ISC_LIST_DEQUEUE(*buflist, buffer, link);
4604                 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
4605                 buffer = ISC_LIST_HEAD(*buflist);
4606         }
4607
4608         return (socket_recv(sock, dev, task, 0));
4609 }
4610
4611 ISC_SOCKETFUNC_SCOPE isc_result_t
4612 isc__socket_recv(isc_socket_t *sock0, isc_region_t *region,
4613                  unsigned int minimum, isc_task_t *task,
4614                  isc_taskaction_t action, const void *arg)
4615 {
4616         isc__socket_t *sock = (isc__socket_t *)sock0;
4617         isc_socketevent_t *dev;
4618         isc__socketmgr_t *manager;
4619
4620         REQUIRE(VALID_SOCKET(sock));
4621         REQUIRE(action != NULL);
4622
4623         manager = sock->manager;
4624         REQUIRE(VALID_MANAGER(manager));
4625
4626         INSIST(sock->bound);
4627
4628         dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
4629         if (dev == NULL)
4630                 return (ISC_R_NOMEMORY);
4631
4632         return (isc__socket_recv2(sock0, region, minimum, task, dev, 0));
4633 }
4634
4635 ISC_SOCKETFUNC_SCOPE isc_result_t
4636 isc__socket_recv2(isc_socket_t *sock0, isc_region_t *region,
4637                   unsigned int minimum, isc_task_t *task,
4638                   isc_socketevent_t *event, unsigned int flags)
4639 {
4640         isc__socket_t *sock = (isc__socket_t *)sock0;
4641
4642         event->ev_sender = sock;
4643         event->result = ISC_R_UNSET;
4644         ISC_LIST_INIT(event->bufferlist);
4645         event->region = *region;
4646         event->n = 0;
4647         event->offset = 0;
4648         event->attributes = 0;
4649
4650         /*
4651          * UDP sockets are always partial read.
4652          */
4653         if (sock->type == isc_sockettype_udp)
4654                 event->minimum = 1;
4655         else {
4656                 if (minimum == 0)
4657                         event->minimum = region->length;
4658                 else
4659                         event->minimum = minimum;
4660         }
4661
4662         return (socket_recv(sock, event, task, flags));
4663 }
4664
4665 static isc_result_t
4666 socket_send(isc__socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
4667             isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
4668             unsigned int flags)
4669 {
4670         int io_state;
4671         isc_boolean_t have_lock = ISC_FALSE;
4672         isc_task_t *ntask = NULL;
4673         isc_result_t result = ISC_R_SUCCESS;
4674
4675         dev->ev_sender = task;
4676
4677         set_dev_address(address, sock, dev);
4678         if (pktinfo != NULL) {
4679                 dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
4680                 dev->pktinfo = *pktinfo;
4681
4682                 if (!isc_sockaddr_issitelocal(&dev->address) &&
4683                     !isc_sockaddr_islinklocal(&dev->address)) {
4684                         socket_log(sock, NULL, TRACE, isc_msgcat,
4685                                    ISC_MSGSET_SOCKET, ISC_MSG_PKTINFOPROVIDED,
4686                                    "pktinfo structure provided, ifindex %u "
4687                                    "(set to 0)", pktinfo->ipi6_ifindex);
4688
4689                         /*
4690                          * Set the pktinfo index to 0 here, to let the
4691                          * kernel decide what interface it should send on.
4692                          */
4693                         dev->pktinfo.ipi6_ifindex = 0;
4694                 }
4695         }
4696
4697         if (sock->type == isc_sockettype_udp)
4698                 io_state = doio_send(sock, dev);
4699         else {
4700                 LOCK(&sock->lock);
4701                 have_lock = ISC_TRUE;
4702
4703                 if (ISC_LIST_EMPTY(sock->send_list))
4704                         io_state = doio_send(sock, dev);
4705                 else
4706                         io_state = DOIO_SOFT;
4707         }
4708
4709         switch (io_state) {
4710         case DOIO_SOFT:
4711                 /*
4712                  * We couldn't send all or part of the request right now, so
4713                  * queue it unless ISC_SOCKFLAG_NORETRY is set.
4714                  */
4715                 if ((flags & ISC_SOCKFLAG_NORETRY) == 0) {
4716                         isc_task_attach(task, &ntask);
4717                         dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
4718
4719                         if (!have_lock) {
4720                                 LOCK(&sock->lock);
4721                                 have_lock = ISC_TRUE;
4722                         }
4723
4724                         /*
4725                          * Enqueue the request.  If the socket was previously
4726                          * not being watched, poke the watcher to start
4727                          * paying attention to it.
4728                          */
4729                         if (ISC_LIST_EMPTY(sock->send_list) &&
4730                             !sock->pending_send)
4731                                 select_poke(sock->manager, sock->fd,
4732                                             SELECT_POKE_WRITE);
4733                         ISC_LIST_ENQUEUE(sock->send_list, dev, ev_link);
4734
4735                         socket_log(sock, NULL, EVENT, NULL, 0, 0,
4736                                    "socket_send: event %p -> task %p",
4737                                    dev, ntask);
4738
4739                         if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
4740                                 result = ISC_R_INPROGRESS;
4741                         break;
4742                 }
4743
4744         case DOIO_HARD:
4745         case DOIO_SUCCESS:
4746                 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
4747                         send_senddone_event(sock, &dev);
4748                 break;
4749         }
4750
4751         if (have_lock)
4752                 UNLOCK(&sock->lock);
4753
4754         return (result);
4755 }
4756
4757 ISC_SOCKETFUNC_SCOPE isc_result_t
4758 isc__socket_send(isc_socket_t *sock, isc_region_t *region,
4759                  isc_task_t *task, isc_taskaction_t action, const void *arg)
4760 {
4761         /*
4762          * REQUIRE() checking is performed in isc_socket_sendto().
4763          */
4764         return (isc__socket_sendto(sock, region, task, action, arg, NULL,
4765                                    NULL));
4766 }
4767
4768 ISC_SOCKETFUNC_SCOPE isc_result_t
4769 isc__socket_sendto(isc_socket_t *sock0, isc_region_t *region,
4770                    isc_task_t *task, isc_taskaction_t action, const void *arg,
4771                    isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
4772 {
4773         isc__socket_t *sock = (isc__socket_t *)sock0;
4774         isc_socketevent_t *dev;
4775         isc__socketmgr_t *manager;
4776
4777         REQUIRE(VALID_SOCKET(sock));
4778         REQUIRE(region != NULL);
4779         REQUIRE(task != NULL);
4780         REQUIRE(action != NULL);
4781
4782         manager = sock->manager;
4783         REQUIRE(VALID_MANAGER(manager));
4784
4785         INSIST(sock->bound);
4786
4787         dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
4788         if (dev == NULL)
4789                 return (ISC_R_NOMEMORY);
4790
4791         dev->region = *region;
4792
4793         return (socket_send(sock, dev, task, address, pktinfo, 0));
4794 }
4795
4796 ISC_SOCKETFUNC_SCOPE isc_result_t
4797 isc__socket_sendv(isc_socket_t *sock, isc_bufferlist_t *buflist,
4798                   isc_task_t *task, isc_taskaction_t action, const void *arg)
4799 {
4800         return (isc__socket_sendtov2(sock, buflist, task, action, arg, NULL,
4801                                      NULL, 0));
4802 }
4803
4804 ISC_SOCKETFUNC_SCOPE isc_result_t
4805 isc__socket_sendtov(isc_socket_t *sock, isc_bufferlist_t *buflist,
4806                     isc_task_t *task, isc_taskaction_t action, const void *arg,
4807                     isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
4808 {
4809         return (isc__socket_sendtov2(sock, buflist, task, action, arg, address,
4810                                      pktinfo, 0));
4811 }
4812
4813 ISC_SOCKETFUNC_SCOPE isc_result_t
4814 isc__socket_sendtov2(isc_socket_t *sock0, isc_bufferlist_t *buflist,
4815                      isc_task_t *task, isc_taskaction_t action, const void *arg,
4816                      isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
4817                      unsigned int flags)
4818 {
4819         isc__socket_t *sock = (isc__socket_t *)sock0;
4820         isc_socketevent_t *dev;
4821         isc__socketmgr_t *manager;
4822         unsigned int iocount;
4823         isc_buffer_t *buffer;
4824
4825         REQUIRE(VALID_SOCKET(sock));
4826         REQUIRE(buflist != NULL);
4827         REQUIRE(!ISC_LIST_EMPTY(*buflist));
4828         REQUIRE(task != NULL);
4829         REQUIRE(action != NULL);
4830
4831         manager = sock->manager;
4832         REQUIRE(VALID_MANAGER(manager));
4833
4834         iocount = isc_bufferlist_usedcount(buflist);
4835         REQUIRE(iocount > 0);
4836
4837         dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
4838         if (dev == NULL)
4839                 return (ISC_R_NOMEMORY);
4840
4841         /*
4842          * Move each buffer from the passed in list to our internal one.
4843          */
4844         buffer = ISC_LIST_HEAD(*buflist);
4845         while (buffer != NULL) {
4846                 ISC_LIST_DEQUEUE(*buflist, buffer, link);
4847                 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
4848                 buffer = ISC_LIST_HEAD(*buflist);
4849         }
4850
4851         return (socket_send(sock, dev, task, address, pktinfo, flags));
4852 }
4853
4854 ISC_SOCKETFUNC_SCOPE isc_result_t
4855 isc__socket_sendto2(isc_socket_t *sock0, isc_region_t *region,
4856                     isc_task_t *task,
4857                     isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
4858                     isc_socketevent_t *event, unsigned int flags)
4859 {
4860         isc__socket_t *sock = (isc__socket_t *)sock0;
4861
4862         REQUIRE(VALID_SOCKET(sock));
4863         REQUIRE((flags & ~(ISC_SOCKFLAG_IMMEDIATE|ISC_SOCKFLAG_NORETRY)) == 0);
4864         if ((flags & ISC_SOCKFLAG_NORETRY) != 0)
4865                 REQUIRE(sock->type == isc_sockettype_udp);
4866         event->ev_sender = sock;
4867         event->result = ISC_R_UNSET;
4868         ISC_LIST_INIT(event->bufferlist);
4869         event->region = *region;
4870         event->n = 0;
4871         event->offset = 0;
4872         event->attributes = 0;
4873
4874         return (socket_send(sock, event, task, address, pktinfo, flags));
4875 }
4876
4877 ISC_SOCKETFUNC_SCOPE void
4878 isc__socket_cleanunix(isc_sockaddr_t *sockaddr, isc_boolean_t active) {
4879 #ifdef ISC_PLATFORM_HAVESYSUNH
4880         int s;
4881         struct stat sb;
4882         char strbuf[ISC_STRERRORSIZE];
4883
4884         if (sockaddr->type.sa.sa_family != AF_UNIX)
4885                 return;
4886
4887 #ifndef S_ISSOCK
4888 #if defined(S_IFMT) && defined(S_IFSOCK)
4889 #define S_ISSOCK(mode) ((mode & S_IFMT)==S_IFSOCK)
4890 #elif defined(_S_IFMT) && defined(S_IFSOCK)
4891 #define S_ISSOCK(mode) ((mode & _S_IFMT)==S_IFSOCK)
4892 #endif
4893 #endif
4894
4895 #ifndef S_ISFIFO
4896 #if defined(S_IFMT) && defined(S_IFIFO)
4897 #define S_ISFIFO(mode) ((mode & S_IFMT)==S_IFIFO)
4898 #elif defined(_S_IFMT) && defined(S_IFIFO)
4899 #define S_ISFIFO(mode) ((mode & _S_IFMT)==S_IFIFO)
4900 #endif
4901 #endif
4902
4903 #if !defined(S_ISFIFO) && !defined(S_ISSOCK)
4904 #error You need to define S_ISFIFO and S_ISSOCK as appropriate for your platform.  See <sys/stat.h>.
4905 #endif
4906
4907 #ifndef S_ISFIFO
4908 #define S_ISFIFO(mode) 0
4909 #endif
4910
4911 #ifndef S_ISSOCK
4912 #define S_ISSOCK(mode) 0
4913 #endif
4914
4915         if (active) {
4916                 if (stat(sockaddr->type.sunix.sun_path, &sb) < 0) {
4917                         isc__strerror(errno, strbuf, sizeof(strbuf));
4918                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4919                                       ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4920                                       "isc_socket_cleanunix: stat(%s): %s",
4921                                       sockaddr->type.sunix.sun_path, strbuf);
4922                         return;
4923                 }
4924                 if (!(S_ISSOCK(sb.st_mode) || S_ISFIFO(sb.st_mode))) {
4925                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4926                                       ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4927                                       "isc_socket_cleanunix: %s: not a socket",
4928                                       sockaddr->type.sunix.sun_path);
4929                         return;
4930                 }
4931                 if (unlink(sockaddr->type.sunix.sun_path) < 0) {
4932                         isc__strerror(errno, strbuf, sizeof(strbuf));
4933                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4934                                       ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4935                                       "isc_socket_cleanunix: unlink(%s): %s",
4936                                       sockaddr->type.sunix.sun_path, strbuf);
4937                 }
4938                 return;
4939         }
4940
4941         s = socket(AF_UNIX, SOCK_STREAM, 0);
4942         if (s < 0) {
4943                 isc__strerror(errno, strbuf, sizeof(strbuf));
4944                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4945                               ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4946                               "isc_socket_cleanunix: socket(%s): %s",
4947                               sockaddr->type.sunix.sun_path, strbuf);
4948                 return;
4949         }
4950
4951         if (stat(sockaddr->type.sunix.sun_path, &sb) < 0) {
4952                 switch (errno) {
4953                 case ENOENT:    /* We exited cleanly last time */
4954                         break;
4955                 default:
4956                         isc__strerror(errno, strbuf, sizeof(strbuf));
4957                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4958                                       ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4959                                       "isc_socket_cleanunix: stat(%s): %s",
4960                                       sockaddr->type.sunix.sun_path, strbuf);
4961                         break;
4962                 }
4963                 goto cleanup;
4964         }
4965
4966         if (!(S_ISSOCK(sb.st_mode) || S_ISFIFO(sb.st_mode))) {
4967                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4968                               ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4969                               "isc_socket_cleanunix: %s: not a socket",
4970                               sockaddr->type.sunix.sun_path);
4971                 goto cleanup;
4972         }
4973
4974         if (connect(s, (struct sockaddr *)&sockaddr->type.sunix,
4975                     sizeof(sockaddr->type.sunix)) < 0) {
4976                 switch (errno) {
4977                 case ECONNREFUSED:
4978                 case ECONNRESET:
4979                         if (unlink(sockaddr->type.sunix.sun_path) < 0) {
4980                                 isc__strerror(errno, strbuf, sizeof(strbuf));
4981                                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4982                                               ISC_LOGMODULE_SOCKET,
4983                                               ISC_LOG_WARNING,
4984                                               "isc_socket_cleanunix: "
4985                                               "unlink(%s): %s",
4986                                               sockaddr->type.sunix.sun_path,
4987                                               strbuf);
4988                         }
4989                         break;
4990                 default:
4991                         isc__strerror(errno, strbuf, sizeof(strbuf));
4992                         isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4993                                       ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4994                                       "isc_socket_cleanunix: connect(%s): %s",
4995                                       sockaddr->type.sunix.sun_path, strbuf);
4996                         break;
4997                 }
4998         }
4999  cleanup:
5000         close(s);
5001 #else
5002         UNUSED(sockaddr);
5003         UNUSED(active);
5004 #endif
5005 }
5006
5007 ISC_SOCKETFUNC_SCOPE isc_result_t
5008 isc__socket_permunix(isc_sockaddr_t *sockaddr, isc_uint32_t perm,
5009                     isc_uint32_t owner, isc_uint32_t group)
5010 {
5011 #ifdef ISC_PLATFORM_HAVESYSUNH
5012         isc_result_t result = ISC_R_SUCCESS;
5013         char strbuf[ISC_STRERRORSIZE];
5014         char path[sizeof(sockaddr->type.sunix.sun_path)];
5015 #ifdef NEED_SECURE_DIRECTORY
5016         char *slash;
5017 #endif
5018
5019         REQUIRE(sockaddr->type.sa.sa_family == AF_UNIX);
5020         INSIST(strlen(sockaddr->type.sunix.sun_path) < sizeof(path));
5021         strcpy(path, sockaddr->type.sunix.sun_path);
5022
5023 #ifdef NEED_SECURE_DIRECTORY
5024         slash = strrchr(path, '/');
5025         if (slash != NULL) {
5026                 if (slash != path)
5027                         *slash = '\0';
5028                 else
5029                         strcpy(path, "/");
5030         } else
5031                 strcpy(path, ".");
5032 #endif
5033
5034         if (chmod(path, perm) < 0) {
5035                 isc__strerror(errno, strbuf, sizeof(strbuf));
5036                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
5037                               ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
5038                               "isc_socket_permunix: chmod(%s, %d): %s",
5039                               path, perm, strbuf);
5040                 result = ISC_R_FAILURE;
5041         }
5042         if (chown(path, owner, group) < 0) {
5043                 isc__strerror(errno, strbuf, sizeof(strbuf));
5044                 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
5045                               ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
5046                               "isc_socket_permunix: chown(%s, %d, %d): %s",
5047                               path, owner, group,
5048                               strbuf);
5049                 result = ISC_R_FAILURE;
5050         }
5051         return (result);
5052 #else
5053         UNUSED(sockaddr);
5054         UNUSED(perm);
5055         UNUSED(owner);
5056         UNUSED(group);
5057         return (ISC_R_NOTIMPLEMENTED);
5058 #endif
5059 }
5060
5061 ISC_SOCKETFUNC_SCOPE isc_result_t
5062 isc__socket_bind(isc_socket_t *sock0, isc_sockaddr_t *sockaddr,
5063                  unsigned int options) {
5064         isc__socket_t *sock = (isc__socket_t *)sock0;
5065         char strbuf[ISC_STRERRORSIZE];
5066         int on = 1;
5067
5068         REQUIRE(VALID_SOCKET(sock));
5069
5070         LOCK(&sock->lock);
5071
5072         INSIST(!sock->bound);
5073         INSIST(!sock->dupped);
5074
5075         if (sock->pf != sockaddr->type.sa.sa_family) {
5076                 UNLOCK(&sock->lock);
5077                 return (ISC_R_FAMILYMISMATCH);
5078         }
5079
5080         /*
5081          * Only set SO_REUSEADDR when we want a specific port.
5082          */
5083 #ifdef AF_UNIX
5084         if (sock->pf == AF_UNIX)
5085                 goto bind_socket;
5086 #endif
5087         if ((options & ISC_SOCKET_REUSEADDRESS) != 0 &&
5088             isc_sockaddr_getport(sockaddr) != (in_port_t)0 &&
5089             setsockopt(sock->fd, SOL_SOCKET, SO_REUSEADDR, (void *)&on,
5090                        sizeof(on)) < 0) {
5091                 UNEXPECTED_ERROR(__FILE__, __LINE__,
5092                                  "setsockopt(%d) %s", sock->fd,
5093                                  isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
5094                                                 ISC_MSG_FAILED, "failed"));
5095                 /* Press on... */
5096         }
5097 #ifdef AF_UNIX
5098  bind_socket:
5099 #endif
5100         if (bind(sock->fd, &sockaddr->type.sa, sockaddr->length) < 0) {
5101                 inc_stats(sock->manager->stats,
5102                           sock->statsindex[STATID_BINDFAIL]);
5103
5104                 UNLOCK(&sock->lock);
5105                 switch (errno) {
5106                 case EACCES:
5107                         return (ISC_R_NOPERM);
5108                 case EADDRNOTAVAIL:
5109                         return (ISC_R_ADDRNOTAVAIL);
5110                 case EADDRINUSE:
5111                         return (ISC_R_ADDRINUSE);
5112                 case EINVAL:
5113                         return (ISC_R_BOUND);
5114                 default:
5115                         isc__strerror(errno, strbuf, sizeof(strbuf));
5116                         UNEXPECTED_ERROR(__FILE__, __LINE__, "bind: %s",
5117                                          strbuf);
5118                         return (ISC_R_UNEXPECTED);
5119                 }
5120         }
5121
5122         socket_log(sock, sockaddr, TRACE,
5123                    isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_BOUND, "bound");
5124         sock->bound = 1;
5125
5126         UNLOCK(&sock->lock);
5127         return (ISC_R_SUCCESS);
5128 }
5129
5130 /*
5131  * Enable this only for specific OS versions, and only when they have repaired
5132  * their problems with it.  Until then, this is is broken and needs to be
5133  * diabled by default.  See RT22589 for details.
5134  */
5135 #undef ENABLE_ACCEPTFILTER
5136
5137 ISC_SOCKETFUNC_SCOPE isc_result_t
5138 isc__socket_filter(isc_socket_t *sock0, const char *filter) {
5139         isc__socket_t *sock = (isc__socket_t *)sock0;
5140 #if defined(SO_ACCEPTFILTER) && defined(ENABLE_ACCEPTFILTER)
5141         char strbuf[ISC_STRERRORSIZE];
5142         struct accept_filter_arg afa;
5143 #else
5144         UNUSED(sock);
5145         UNUSED(filter);
5146 #endif
5147
5148         REQUIRE(VALID_SOCKET(sock));
5149
5150 #if defined(SO_ACCEPTFILTER) && defined(ENABLE_ACCEPTFILTER)
5151         bzero(&afa, sizeof(afa));
5152         strncpy(afa.af_name, filter, sizeof(afa.af_name));
5153         if (setsockopt(sock->fd, SOL_SOCKET, SO_ACCEPTFILTER,
5154                          &afa, sizeof(afa)) == -1) {
5155                 isc__strerror(errno, strbuf, sizeof(strbuf));
5156                 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
5157                            ISC_MSG_FILTER, "setsockopt(SO_ACCEPTFILTER): %s",
5158                            strbuf);
5159                 return (ISC_R_FAILURE);
5160         }
5161         return (ISC_R_SUCCESS);
5162 #else
5163         return (ISC_R_NOTIMPLEMENTED);
5164 #endif
5165 }
5166
5167 /*
5168  * Set up to listen on a given socket.  We do this by creating an internal
5169  * event that will be dispatched when the socket has read activity.  The
5170  * watcher will send the internal event to the task when there is a new
5171  * connection.
5172  *
5173  * Unlike in read, we don't preallocate a done event here.  Every time there
5174  * is a new connection we'll have to allocate a new one anyway, so we might
5175  * as well keep things simple rather than having to track them.
5176  */
5177 ISC_SOCKETFUNC_SCOPE isc_result_t
5178 isc__socket_listen(isc_socket_t *sock0, unsigned int backlog) {
5179         isc__socket_t *sock = (isc__socket_t *)sock0;
5180         char strbuf[ISC_STRERRORSIZE];
5181
5182         REQUIRE(VALID_SOCKET(sock));
5183
5184         LOCK(&sock->lock);
5185
5186         REQUIRE(!sock->listener);
5187         REQUIRE(sock->bound);
5188         REQUIRE(sock->type == isc_sockettype_tcp ||
5189                 sock->type == isc_sockettype_unix);
5190
5191         if (backlog == 0)
5192                 backlog = SOMAXCONN;
5193
5194         if (listen(sock->fd, (int)backlog) < 0) {
5195                 UNLOCK(&sock->lock);
5196                 isc__strerror(errno, strbuf, sizeof(strbuf));
5197
5198                 UNEXPECTED_ERROR(__FILE__, __LINE__, "listen: %s", strbuf);
5199
5200                 return (ISC_R_UNEXPECTED);
5201         }
5202
5203         sock->listener = 1;
5204
5205         UNLOCK(&sock->lock);
5206         return (ISC_R_SUCCESS);
5207 }
5208
5209 /*
5210  * This should try to do aggressive accept() XXXMLG
5211  */
5212 ISC_SOCKETFUNC_SCOPE isc_result_t
5213 isc__socket_accept(isc_socket_t *sock0,
5214                   isc_task_t *task, isc_taskaction_t action, const void *arg)
5215 {
5216         isc__socket_t *sock = (isc__socket_t *)sock0;
5217         isc_socket_newconnev_t *dev;
5218         isc__socketmgr_t *manager;
5219         isc_task_t *ntask = NULL;
5220         isc__socket_t *nsock;
5221         isc_result_t result;
5222         isc_boolean_t do_poke = ISC_FALSE;
5223
5224         REQUIRE(VALID_SOCKET(sock));
5225         manager = sock->manager;
5226         REQUIRE(VALID_MANAGER(manager));
5227
5228         LOCK(&sock->lock);
5229
5230         REQUIRE(sock->listener);
5231
5232         /*
5233          * Sender field is overloaded here with the task we will be sending
5234          * this event to.  Just before the actual event is delivered the
5235          * actual ev_sender will be touched up to be the socket.
5236          */
5237         dev = (isc_socket_newconnev_t *)
5238                 isc_event_allocate(manager->mctx, task, ISC_SOCKEVENT_NEWCONN,
5239                                    action, arg, sizeof(*dev));
5240         if (dev == NULL) {
5241                 UNLOCK(&sock->lock);
5242                 return (ISC_R_NOMEMORY);
5243         }
5244         ISC_LINK_INIT(dev, ev_link);
5245
5246         result = allocate_socket(manager, sock->type, &nsock);
5247         if (result != ISC_R_SUCCESS) {
5248                 isc_event_free(ISC_EVENT_PTR(&dev));
5249                 UNLOCK(&sock->lock);
5250                 return (result);
5251         }
5252
5253         /*
5254          * Attach to socket and to task.
5255          */
5256         isc_task_attach(task, &ntask);
5257         if (isc_task_exiting(ntask)) {
5258                 free_socket(&nsock);
5259                 isc_task_detach(&ntask);
5260                 isc_event_free(ISC_EVENT_PTR(&dev));
5261                 UNLOCK(&sock->lock);
5262                 return (ISC_R_SHUTTINGDOWN);
5263         }
5264         nsock->references++;
5265         nsock->statsindex = sock->statsindex;
5266
5267         dev->ev_sender = ntask;
5268         dev->newsocket = (isc_socket_t *)nsock;
5269
5270         /*
5271          * Poke watcher here.  We still have the socket locked, so there
5272          * is no race condition.  We will keep the lock for such a short
5273          * bit of time waking it up now or later won't matter all that much.
5274          */
5275         if (ISC_LIST_EMPTY(sock->accept_list))
5276                 do_poke = ISC_TRUE;
5277
5278         ISC_LIST_ENQUEUE(sock->accept_list, dev, ev_link);
5279
5280         if (do_poke)
5281                 select_poke(manager, sock->fd, SELECT_POKE_ACCEPT);
5282
5283         UNLOCK(&sock->lock);
5284         return (ISC_R_SUCCESS);
5285 }
5286
5287 ISC_SOCKETFUNC_SCOPE isc_result_t
5288 isc__socket_connect(isc_socket_t *sock0, isc_sockaddr_t *addr,
5289                    isc_task_t *task, isc_taskaction_t action, const void *arg)
5290 {
5291         isc__socket_t *sock = (isc__socket_t *)sock0;
5292         isc_socket_connev_t *dev;
5293         isc_task_t *ntask = NULL;
5294         isc__socketmgr_t *manager;
5295         int cc;
5296         char strbuf[ISC_STRERRORSIZE];
5297         char addrbuf[ISC_SOCKADDR_FORMATSIZE];
5298
5299         REQUIRE(VALID_SOCKET(sock));
5300         REQUIRE(addr != NULL);
5301         REQUIRE(task != NULL);
5302         REQUIRE(action != NULL);
5303
5304         manager = sock->manager;
5305         REQUIRE(VALID_MANAGER(manager));
5306         REQUIRE(addr != NULL);
5307
5308         if (isc_sockaddr_ismulticast(addr))
5309                 return (ISC_R_MULTICAST);
5310
5311         LOCK(&sock->lock);
5312
5313         REQUIRE(!sock->connecting);
5314
5315         dev = (isc_socket_connev_t *)isc_event_allocate(manager->mctx, sock,
5316                                                         ISC_SOCKEVENT_CONNECT,
5317                                                         action, arg,
5318                                                         sizeof(*dev));
5319         if (dev == NULL) {
5320                 UNLOCK(&sock->lock);
5321                 return (ISC_R_NOMEMORY);
5322         }
5323         ISC_LINK_INIT(dev, ev_link);
5324
5325         /*
5326          * Try to do the connect right away, as there can be only one
5327          * outstanding, and it might happen to complete.
5328          */
5329         sock->peer_address = *addr;
5330         cc = connect(sock->fd, &addr->type.sa, addr->length);
5331         if (cc < 0) {
5332                 /*
5333                  * HP-UX "fails" to connect a UDP socket and sets errno to
5334                  * EINPROGRESS if it's non-blocking.  We'd rather regard this as
5335                  * a success and let the user detect it if it's really an error
5336                  * at the time of sending a packet on the socket.
5337                  */
5338                 if (sock->type == isc_sockettype_udp && errno == EINPROGRESS) {
5339                         cc = 0;
5340                         goto success;
5341                 }
5342                 if (SOFT_ERROR(errno) || errno == EINPROGRESS)
5343                         goto queue;
5344
5345                 switch (errno) {
5346 #define ERROR_MATCH(a, b) case a: dev->result = b; goto err_exit;
5347                         ERROR_MATCH(EACCES, ISC_R_NOPERM);
5348                         ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
5349                         ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
5350                         ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
5351                         ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
5352 #ifdef EHOSTDOWN
5353                         ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
5354 #endif
5355                         ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
5356                         ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
5357                         ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
5358                         ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
5359                         ERROR_MATCH(ECONNRESET, ISC_R_CONNECTIONRESET);
5360 #undef ERROR_MATCH
5361                 }
5362
5363                 sock->connected = 0;
5364
5365                 isc__strerror(errno, strbuf, sizeof(strbuf));
5366                 isc_sockaddr_format(addr, addrbuf, sizeof(addrbuf));
5367                 UNEXPECTED_ERROR(__FILE__, __LINE__, "connect(%s) %d/%s",
5368                                  addrbuf, errno, strbuf);
5369
5370                 UNLOCK(&sock->lock);
5371                 inc_stats(sock->manager->stats,
5372                           sock->statsindex[STATID_CONNECTFAIL]);
5373                 isc_event_free(ISC_EVENT_PTR(&dev));
5374                 return (ISC_R_UNEXPECTED);
5375
5376         err_exit:
5377                 sock->connected = 0;
5378                 isc_task_send(task, ISC_EVENT_PTR(&dev));
5379
5380                 UNLOCK(&sock->lock);
5381                 inc_stats(sock->manager->stats,
5382                           sock->statsindex[STATID_CONNECTFAIL]);
5383                 return (ISC_R_SUCCESS);
5384         }
5385
5386         /*
5387          * If connect completed, fire off the done event.
5388          */
5389  success:
5390         if (cc == 0) {
5391                 sock->connected = 1;
5392                 sock->bound = 1;
5393                 dev->result = ISC_R_SUCCESS;
5394                 isc_task_send(task, ISC_EVENT_PTR(&dev));
5395
5396                 UNLOCK(&sock->lock);
5397
5398                 inc_stats(sock->manager->stats,
5399                           sock->statsindex[STATID_CONNECT]);
5400
5401                 return (ISC_R_SUCCESS);
5402         }
5403
5404  queue:
5405
5406         /*
5407          * Attach to task.
5408          */
5409         isc_task_attach(task, &ntask);
5410
5411         sock->connecting = 1;
5412
5413         dev->ev_sender = ntask;
5414
5415         /*
5416          * Poke watcher here.  We still have the socket locked, so there
5417          * is no race condition.  We will keep the lock for such a short
5418          * bit of time waking it up now or later won't matter all that much.
5419          */
5420         if (sock->connect_ev == NULL)
5421                 select_poke(manager, sock->fd, SELECT_POKE_CONNECT);
5422
5423         sock->connect_ev = dev;
5424
5425         UNLOCK(&sock->lock);
5426         return (ISC_R_SUCCESS);
5427 }
5428
5429 /*
5430  * Called when a socket with a pending connect() finishes.
5431  */
5432 static void
5433 internal_connect(isc_task_t *me, isc_event_t *ev) {
5434         isc__socket_t *sock;
5435         isc_socket_connev_t *dev;
5436         isc_task_t *task;
5437         int cc;
5438         ISC_SOCKADDR_LEN_T optlen;
5439         char strbuf[ISC_STRERRORSIZE];
5440         char peerbuf[ISC_SOCKADDR_FORMATSIZE];
5441
5442         UNUSED(me);
5443         INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
5444
5445         sock = ev->ev_sender;
5446         INSIST(VALID_SOCKET(sock));
5447
5448         LOCK(&sock->lock);
5449
5450         /*
5451          * When the internal event was sent the reference count was bumped
5452          * to keep the socket around for us.  Decrement the count here.
5453          */
5454         INSIST(sock->references > 0);
5455         sock->references--;
5456         if (sock->references == 0) {
5457                 UNLOCK(&sock->lock);
5458                 destroy(&sock);
5459                 return;
5460         }
5461
5462         /*
5463          * Has this event been canceled?
5464          */
5465         dev = sock->connect_ev;
5466         if (dev == NULL) {
5467                 INSIST(!sock->connecting);
5468                 UNLOCK(&sock->lock);
5469                 return;
5470         }
5471
5472         INSIST(sock->connecting);
5473         sock->connecting = 0;
5474
5475         /*
5476          * Get any possible error status here.
5477          */
5478         optlen = sizeof(cc);
5479         if (getsockopt(sock->fd, SOL_SOCKET, SO_ERROR,
5480                        (void *)&cc, (void *)&optlen) < 0)
5481                 cc = errno;
5482         else
5483                 errno = cc;
5484
5485         if (errno != 0) {
5486                 /*
5487                  * If the error is EAGAIN, just re-select on this
5488                  * fd and pretend nothing strange happened.
5489                  */
5490                 if (SOFT_ERROR(errno) || errno == EINPROGRESS) {
5491                         sock->connecting = 1;
5492                         select_poke(sock->manager, sock->fd,
5493                                     SELECT_POKE_CONNECT);
5494                         UNLOCK(&sock->lock);
5495
5496                         return;
5497                 }
5498
5499                 inc_stats(sock->manager->stats,
5500                           sock->statsindex[STATID_CONNECTFAIL]);
5501
5502                 /*
5503                  * Translate other errors into ISC_R_* flavors.
5504                  */
5505                 switch (errno) {
5506 #define ERROR_MATCH(a, b) case a: dev->result = b; break;
5507                         ERROR_MATCH(EACCES, ISC_R_NOPERM);
5508                         ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
5509                         ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
5510                         ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
5511                         ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
5512 #ifdef EHOSTDOWN
5513                         ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
5514 #endif
5515                         ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
5516                         ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
5517                         ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
5518                         ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
5519                         ERROR_MATCH(ETIMEDOUT, ISC_R_TIMEDOUT);
5520                         ERROR_MATCH(ECONNRESET, ISC_R_CONNECTIONRESET);
5521 #undef ERROR_MATCH
5522                 default:
5523                         dev->result = ISC_R_UNEXPECTED;
5524                         isc_sockaddr_format(&sock->peer_address, peerbuf,
5525                                             sizeof(peerbuf));
5526                         isc__strerror(errno, strbuf, sizeof(strbuf));
5527                         UNEXPECTED_ERROR(__FILE__, __LINE__,
5528                                          "internal_connect: connect(%s) %s",
5529                                          peerbuf, strbuf);
5530                 }
5531         } else {
5532                 inc_stats(sock->manager->stats,
5533                           sock->statsindex[STATID_CONNECT]);
5534                 dev->result = ISC_R_SUCCESS;
5535                 sock->connected = 1;
5536                 sock->bound = 1;
5537         }
5538
5539         sock->connect_ev = NULL;
5540
5541         UNLOCK(&sock->lock);
5542
5543         task = dev->ev_sender;
5544         dev->ev_sender = sock;
5545         isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
5546 }
5547
5548 ISC_SOCKETFUNC_SCOPE isc_result_t
5549 isc__socket_getpeername(isc_socket_t *sock0, isc_sockaddr_t *addressp) {
5550         isc__socket_t *sock = (isc__socket_t *)sock0;
5551         isc_result_t result;
5552
5553         REQUIRE(VALID_SOCKET(sock));
5554         REQUIRE(addressp != NULL);
5555
5556         LOCK(&sock->lock);
5557
5558         if (sock->connected) {
5559                 *addressp = sock->peer_address;
5560                 result = ISC_R_SUCCESS;
5561         } else {
5562                 result = ISC_R_NOTCONNECTED;
5563         }
5564
5565         UNLOCK(&sock->lock);
5566
5567         return (result);
5568 }
5569
5570 ISC_SOCKETFUNC_SCOPE isc_result_t
5571 isc__socket_getsockname(isc_socket_t *sock0, isc_sockaddr_t *addressp) {
5572         isc__socket_t *sock = (isc__socket_t *)sock0;
5573         ISC_SOCKADDR_LEN_T len;
5574         isc_result_t result;
5575         char strbuf[ISC_STRERRORSIZE];
5576
5577         REQUIRE(VALID_SOCKET(sock));
5578         REQUIRE(addressp != NULL);
5579
5580         LOCK(&sock->lock);
5581
5582         if (!sock->bound) {
5583                 result = ISC_R_NOTBOUND;
5584                 goto out;
5585         }
5586
5587         result = ISC_R_SUCCESS;
5588
5589         len = sizeof(addressp->type);
5590         if (getsockname(sock->fd, &addressp->type.sa, (void *)&len) < 0) {
5591                 isc__strerror(errno, strbuf, sizeof(strbuf));
5592                 UNEXPECTED_ERROR(__FILE__, __LINE__, "getsockname: %s",
5593                                  strbuf);
5594                 result = ISC_R_UNEXPECTED;
5595                 goto out;
5596         }
5597         addressp->length = (unsigned int)len;
5598
5599  out:
5600         UNLOCK(&sock->lock);
5601
5602         return (result);
5603 }
5604
5605 /*
5606  * Run through the list of events on this socket, and cancel the ones
5607  * queued for task "task" of type "how".  "how" is a bitmask.
5608  */
5609 ISC_SOCKETFUNC_SCOPE void
5610 isc__socket_cancel(isc_socket_t *sock0, isc_task_t *task, unsigned int how) {
5611         isc__socket_t *sock = (isc__socket_t *)sock0;
5612
5613         REQUIRE(VALID_SOCKET(sock));
5614
5615         /*
5616          * Quick exit if there is nothing to do.  Don't even bother locking
5617          * in this case.
5618          */
5619         if (how == 0)
5620                 return;
5621
5622         LOCK(&sock->lock);
5623
5624         /*
5625          * All of these do the same thing, more or less.
5626          * Each will:
5627          *      o If the internal event is marked as "posted" try to
5628          *        remove it from the task's queue.  If this fails, mark it
5629          *        as canceled instead, and let the task clean it up later.
5630          *      o For each I/O request for that task of that type, post
5631          *        its done event with status of "ISC_R_CANCELED".
5632          *      o Reset any state needed.
5633          */
5634         if (((how & ISC_SOCKCANCEL_RECV) == ISC_SOCKCANCEL_RECV)
5635             && !ISC_LIST_EMPTY(sock->recv_list)) {
5636                 isc_socketevent_t      *dev;
5637                 isc_socketevent_t      *next;
5638                 isc_task_t             *current_task;
5639
5640                 dev = ISC_LIST_HEAD(sock->recv_list);
5641
5642                 while (dev != NULL) {
5643                         current_task = dev->ev_sender;
5644                         next = ISC_LIST_NEXT(dev, ev_link);
5645
5646                         if ((task == NULL) || (task == current_task)) {
5647                                 dev->result = ISC_R_CANCELED;
5648                                 send_recvdone_event(sock, &dev);
5649                         }
5650                         dev = next;
5651                 }
5652         }
5653
5654         if (((how & ISC_SOCKCANCEL_SEND) == ISC_SOCKCANCEL_SEND)
5655             && !ISC_LIST_EMPTY(sock->send_list)) {
5656                 isc_socketevent_t      *dev;
5657                 isc_socketevent_t      *next;
5658                 isc_task_t             *current_task;
5659
5660                 dev = ISC_LIST_HEAD(sock->send_list);
5661
5662                 while (dev != NULL) {
5663                         current_task = dev->ev_sender;
5664                         next = ISC_LIST_NEXT(dev, ev_link);
5665
5666                         if ((task == NULL) || (task == current_task)) {
5667                                 dev->result = ISC_R_CANCELED;
5668                                 send_senddone_event(sock, &dev);
5669                         }
5670                         dev = next;
5671                 }
5672         }
5673
5674         if (((how & ISC_SOCKCANCEL_ACCEPT) == ISC_SOCKCANCEL_ACCEPT)
5675             && !ISC_LIST_EMPTY(sock->accept_list)) {
5676                 isc_socket_newconnev_t *dev;
5677                 isc_socket_newconnev_t *next;
5678                 isc_task_t             *current_task;
5679
5680                 dev = ISC_LIST_HEAD(sock->accept_list);
5681                 while (dev != NULL) {
5682                         current_task = dev->ev_sender;
5683                         next = ISC_LIST_NEXT(dev, ev_link);
5684
5685                         if ((task == NULL) || (task == current_task)) {
5686
5687                                 ISC_LIST_UNLINK(sock->accept_list, dev,
5688                                                 ev_link);
5689
5690                                 NEWCONNSOCK(dev)->references--;
5691                                 free_socket((isc__socket_t **)&dev->newsocket);
5692
5693                                 dev->result = ISC_R_CANCELED;
5694                                 dev->ev_sender = sock;
5695                                 isc_task_sendanddetach(&current_task,
5696                                                        ISC_EVENT_PTR(&dev));
5697                         }
5698
5699                         dev = next;
5700                 }
5701         }
5702
5703         /*
5704          * Connecting is not a list.
5705          */
5706         if (((how & ISC_SOCKCANCEL_CONNECT) == ISC_SOCKCANCEL_CONNECT)
5707             && sock->connect_ev != NULL) {
5708                 isc_socket_connev_t    *dev;
5709                 isc_task_t             *current_task;
5710
5711                 INSIST(sock->connecting);
5712                 sock->connecting = 0;
5713
5714                 dev = sock->connect_ev;
5715                 current_task = dev->ev_sender;
5716
5717                 if ((task == NULL) || (task == current_task)) {
5718                         sock->connect_ev = NULL;
5719
5720                         dev->result = ISC_R_CANCELED;
5721                         dev->ev_sender = sock;
5722                         isc_task_sendanddetach(&current_task,
5723                                                ISC_EVENT_PTR(&dev));
5724                 }
5725         }
5726
5727         UNLOCK(&sock->lock);
5728 }
5729
5730 ISC_SOCKETFUNC_SCOPE isc_sockettype_t
5731 isc__socket_gettype(isc_socket_t *sock0) {
5732         isc__socket_t *sock = (isc__socket_t *)sock0;
5733
5734         REQUIRE(VALID_SOCKET(sock));
5735
5736         return (sock->type);
5737 }
5738
5739 ISC_SOCKETFUNC_SCOPE isc_boolean_t
5740 isc__socket_isbound(isc_socket_t *sock0) {
5741         isc__socket_t *sock = (isc__socket_t *)sock0;
5742         isc_boolean_t val;
5743
5744         REQUIRE(VALID_SOCKET(sock));
5745
5746         LOCK(&sock->lock);
5747         val = ((sock->bound) ? ISC_TRUE : ISC_FALSE);
5748         UNLOCK(&sock->lock);
5749
5750         return (val);
5751 }
5752
5753 ISC_SOCKETFUNC_SCOPE void
5754 isc__socket_ipv6only(isc_socket_t *sock0, isc_boolean_t yes) {
5755         isc__socket_t *sock = (isc__socket_t *)sock0;
5756 #if defined(IPV6_V6ONLY)
5757         int onoff = yes ? 1 : 0;
5758 #else
5759         UNUSED(yes);
5760         UNUSED(sock);
5761 #endif
5762
5763         REQUIRE(VALID_SOCKET(sock));
5764         INSIST(!sock->dupped);
5765
5766 #ifdef IPV6_V6ONLY
5767         if (sock->pf == AF_INET6) {
5768                 if (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_V6ONLY,
5769                                (void *)&onoff, sizeof(int)) < 0) {
5770                         char strbuf[ISC_STRERRORSIZE];
5771                         isc__strerror(errno, strbuf, sizeof(strbuf));
5772                         UNEXPECTED_ERROR(__FILE__, __LINE__,
5773                                          "setsockopt(%d, IPV6_V6ONLY) "
5774                                          "%s: %s", sock->fd,
5775                                          isc_msgcat_get(isc_msgcat,
5776                                                         ISC_MSGSET_GENERAL,
5777                                                         ISC_MSG_FAILED,
5778                                                         "failed"),
5779                                          strbuf);
5780                 }
5781         }
5782         FIX_IPV6_RECVPKTINFO(sock);     /* AIX */
5783 #endif
5784 }
5785
5786 #ifndef USE_WATCHER_THREAD
5787 /*
5788  * In our assumed scenario, we can simply use a single static object.
5789  * XXX: this is not true if the application uses multiple threads with
5790  *      'multi-context' mode.  Fixing this is a future TODO item.
5791  */
5792 static isc_socketwait_t swait_private;
5793
5794 int
5795 isc__socketmgr_waitevents(isc_socketmgr_t *manager0, struct timeval *tvp,
5796                           isc_socketwait_t **swaitp)
5797 {
5798         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
5799
5800
5801         int n;
5802 #ifdef USE_KQUEUE
5803         struct timespec ts, *tsp;
5804 #endif
5805 #ifdef USE_EPOLL
5806         int timeout;
5807 #endif
5808 #ifdef USE_DEVPOLL
5809         struct dvpoll dvp;
5810 #endif
5811
5812         REQUIRE(swaitp != NULL && *swaitp == NULL);
5813
5814 #ifdef USE_SHARED_MANAGER
5815         if (manager == NULL)
5816                 manager = socketmgr;
5817 #endif
5818         if (manager == NULL)
5819                 return (0);
5820
5821 #ifdef USE_KQUEUE
5822         if (tvp != NULL) {
5823                 ts.tv_sec = tvp->tv_sec;
5824                 ts.tv_nsec = tvp->tv_usec * 1000;
5825                 tsp = &ts;
5826         } else
5827                 tsp = NULL;
5828         swait_private.nevents = kevent(manager->kqueue_fd, NULL, 0,
5829                                        manager->events, manager->nevents,
5830                                        tsp);
5831         n = swait_private.nevents;
5832 #elif defined(USE_EPOLL)
5833         if (tvp != NULL)
5834                 timeout = tvp->tv_sec * 1000 + (tvp->tv_usec + 999) / 1000;
5835         else
5836                 timeout = -1;
5837         swait_private.nevents = epoll_wait(manager->epoll_fd,
5838                                            manager->events,
5839                                            manager->nevents, timeout);
5840         n = swait_private.nevents;
5841 #elif defined(USE_DEVPOLL)
5842         dvp.dp_fds = manager->events;
5843         dvp.dp_nfds = manager->nevents;
5844         if (tvp != NULL) {
5845                 dvp.dp_timeout = tvp->tv_sec * 1000 +
5846                         (tvp->tv_usec + 999) / 1000;
5847         } else
5848                 dvp.dp_timeout = -1;
5849         swait_private.nevents = ioctl(manager->devpoll_fd, DP_POLL, &dvp);
5850         n = swait_private.nevents;
5851 #elif defined(USE_SELECT)
5852         memmove(manager->read_fds_copy, manager->read_fds, manager->fd_bufsize);
5853         memmove(manager->write_fds_copy, manager->write_fds,
5854                 manager->fd_bufsize);
5855
5856         swait_private.readset = manager->read_fds_copy;
5857         swait_private.writeset = manager->write_fds_copy;
5858         swait_private.maxfd = manager->maxfd + 1;
5859
5860         n = select(swait_private.maxfd, swait_private.readset,
5861                    swait_private.writeset, NULL, tvp);
5862 #endif
5863
5864         *swaitp = &swait_private;
5865         return (n);
5866 }
5867
5868 isc_result_t
5869 isc__socketmgr_dispatch(isc_socketmgr_t *manager0, isc_socketwait_t *swait) {
5870         isc__socketmgr_t *manager = (isc__socketmgr_t *)manager0;
5871
5872         REQUIRE(swait == &swait_private);
5873
5874 #ifdef USE_SHARED_MANAGER
5875         if (manager == NULL)
5876                 manager = socketmgr;
5877 #endif
5878         if (manager == NULL)
5879                 return (ISC_R_NOTFOUND);
5880
5881 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
5882         (void)process_fds(manager, manager->events, swait->nevents);
5883         return (ISC_R_SUCCESS);
5884 #elif defined(USE_SELECT)
5885         process_fds(manager, swait->maxfd, swait->readset, swait->writeset);
5886         return (ISC_R_SUCCESS);
5887 #endif
5888 }
5889 #endif /* USE_WATCHER_THREAD */
5890
5891 #ifdef BIND9
5892 void
5893 isc__socket_setname(isc_socket_t *socket0, const char *name, void *tag) {
5894         isc__socket_t *socket = (isc__socket_t *)socket0;
5895
5896         /*
5897          * Name 'socket'.
5898          */
5899
5900         REQUIRE(VALID_SOCKET(socket));
5901
5902         LOCK(&socket->lock);
5903         memset(socket->name, 0, sizeof(socket->name));
5904         strncpy(socket->name, name, sizeof(socket->name) - 1);
5905         socket->tag = tag;
5906         UNLOCK(&socket->lock);
5907 }
5908
5909 ISC_SOCKETFUNC_SCOPE const char *
5910 isc__socket_getname(isc_socket_t *socket0) {
5911         isc__socket_t *socket = (isc__socket_t *)socket0;
5912
5913         return (socket->name);
5914 }
5915
5916 void *
5917 isc__socket_gettag(isc_socket_t *socket0) {
5918         isc__socket_t *socket = (isc__socket_t *)socket0;
5919
5920         return (socket->tag);
5921 }
5922 #endif  /* BIND9 */
5923
5924 #ifdef USE_SOCKETIMPREGISTER
5925 isc_result_t
5926 isc__socket_register() {
5927         return (isc_socket_register(isc__socketmgr_create));
5928 }
5929 #endif
5930
5931 ISC_SOCKETFUNC_SCOPE int
5932 isc__socket_getfd(isc_socket_t *socket0) {
5933         isc__socket_t *socket = (isc__socket_t *)socket0;
5934
5935         return ((short) socket->fd);
5936 }
5937
5938 #if defined(HAVE_LIBXML2) && defined(BIND9)
5939
5940 static const char *
5941 _socktype(isc_sockettype_t type)
5942 {
5943         if (type == isc_sockettype_udp)
5944                 return ("udp");
5945         else if (type == isc_sockettype_tcp)
5946                 return ("tcp");
5947         else if (type == isc_sockettype_unix)
5948                 return ("unix");
5949         else if (type == isc_sockettype_fdwatch)
5950                 return ("fdwatch");
5951         else
5952                 return ("not-initialized");
5953 }
5954
5955 #define TRY0(a) do { xmlrc = (a); if (xmlrc < 0) goto error; } while(0)
5956 ISC_SOCKETFUNC_SCOPE int
5957 isc_socketmgr_renderxml(isc_socketmgr_t *mgr0, xmlTextWriterPtr writer) {
5958         isc__socketmgr_t *mgr = (isc__socketmgr_t *)mgr0;
5959         isc__socket_t *sock = NULL;
5960         char peerbuf[ISC_SOCKADDR_FORMATSIZE];
5961         isc_sockaddr_t addr;
5962         ISC_SOCKADDR_LEN_T len;
5963         int xmlrc;
5964
5965         LOCK(&mgr->lock);
5966
5967 #ifdef USE_SHARED_MANAGER
5968         TRY0(xmlTextWriterStartElement(writer, ISC_XMLCHAR "references"));
5969         TRY0(xmlTextWriterWriteFormatString(writer, "%d", mgr->refs));
5970         TRY0(xmlTextWriterEndElement(writer));
5971 #endif  /* USE_SHARED_MANAGER */
5972
5973         TRY0(xmlTextWriterStartElement(writer, ISC_XMLCHAR "sockets"));
5974         sock = ISC_LIST_HEAD(mgr->socklist);
5975         while (sock != NULL) {
5976                 LOCK(&sock->lock);
5977                 TRY0(xmlTextWriterStartElement(writer, ISC_XMLCHAR "socket"));
5978
5979                 TRY0(xmlTextWriterStartElement(writer, ISC_XMLCHAR "id"));
5980                 TRY0(xmlTextWriterWriteFormatString(writer, "%p", sock));
5981                 TRY0(xmlTextWriterEndElement(writer));
5982
5983                 if (sock->name[0] != 0) {
5984                         TRY0(xmlTextWriterStartElement(writer,
5985                                                        ISC_XMLCHAR "name"));
5986                         TRY0(xmlTextWriterWriteFormatString(writer, "%s",
5987                                                             sock->name));
5988                         TRY0(xmlTextWriterEndElement(writer)); /* name */
5989                 }
5990
5991                 TRY0(xmlTextWriterStartElement(writer,
5992                                                ISC_XMLCHAR "references"));
5993                 TRY0(xmlTextWriterWriteFormatString(writer, "%d",
5994                                                     sock->references));
5995                 TRY0(xmlTextWriterEndElement(writer));
5996
5997                 TRY0(xmlTextWriterWriteElement(writer, ISC_XMLCHAR "type",
5998                                           ISC_XMLCHAR _socktype(sock->type)));
5999
6000                 if (sock->connected) {
6001                         isc_sockaddr_format(&sock->peer_address, peerbuf,
6002                                             sizeof(peerbuf));
6003                         TRY0(xmlTextWriterWriteElement(writer,
6004                                                   ISC_XMLCHAR "peer-address",
6005                                                   ISC_XMLCHAR peerbuf));
6006                 }
6007
6008                 len = sizeof(addr);
6009                 if (getsockname(sock->fd, &addr.type.sa, (void *)&len) == 0) {
6010                         isc_sockaddr_format(&addr, peerbuf, sizeof(peerbuf));
6011                         TRY0(xmlTextWriterWriteElement(writer,
6012                                                   ISC_XMLCHAR "local-address",
6013                                                   ISC_XMLCHAR peerbuf));
6014                 }
6015
6016                 TRY0(xmlTextWriterStartElement(writer, ISC_XMLCHAR "states"));
6017                 if (sock->pending_recv)
6018                         TRY0(xmlTextWriterWriteElement(writer,
6019                                                 ISC_XMLCHAR "state",
6020                                                 ISC_XMLCHAR "pending-receive"));
6021                 if (sock->pending_send)
6022                         TRY0(xmlTextWriterWriteElement(writer,
6023                                                   ISC_XMLCHAR "state",
6024                                                   ISC_XMLCHAR "pending-send"));
6025                 if (sock->pending_accept)
6026                         TRY0(xmlTextWriterWriteElement(writer,
6027                                                  ISC_XMLCHAR "state",
6028                                                  ISC_XMLCHAR "pending_accept"));
6029                 if (sock->listener)
6030                         TRY0(xmlTextWriterWriteElement(writer,
6031                                                        ISC_XMLCHAR "state",
6032                                                        ISC_XMLCHAR "listener"));
6033                 if (sock->connected)
6034                         TRY0(xmlTextWriterWriteElement(writer,
6035                                                      ISC_XMLCHAR "state",
6036                                                      ISC_XMLCHAR "connected"));
6037                 if (sock->connecting)
6038                         TRY0(xmlTextWriterWriteElement(writer,
6039                                                     ISC_XMLCHAR "state",
6040                                                     ISC_XMLCHAR "connecting"));
6041                 if (sock->bound)
6042                         TRY0(xmlTextWriterWriteElement(writer,
6043                                                        ISC_XMLCHAR "state",
6044                                                        ISC_XMLCHAR "bound"));
6045
6046                 TRY0(xmlTextWriterEndElement(writer)); /* states */
6047
6048                 TRY0(xmlTextWriterEndElement(writer)); /* socket */
6049
6050                 UNLOCK(&sock->lock);
6051                 sock = ISC_LIST_NEXT(sock, link);
6052         }
6053         TRY0(xmlTextWriterEndElement(writer)); /* sockets */
6054
6055  error:
6056         if (sock != NULL)
6057                 UNLOCK(&sock->lock);
6058
6059         UNLOCK(&mgr->lock);
6060
6061         return (xmlrc);
6062 }
6063 #endif /* HAVE_LIBXML2 */