src/backend/port/win32/socket.c

   1 /*-------------------------------------------------------------------------
   2  *
   3  * socket.c
   4  *        Microsoft Windows Win32 Socket Functions
   5  *
   6  * Portions Copyright (c) 1996-2009, PostgreSQL Global Development Group
   7  *
   8  * IDENTIFICATION
   9  *        $PostgreSQL$
  10  *
  11  *-------------------------------------------------------------------------
  12  */
  13
  14 #include "postgres.h"
  15
  16 #undef socket
  17 #undef accept
  18 #undef connect
  19 #undef select
  20 #undef recv
  21 #undef send
  22
  23 /*
  24  * Blocking socket functions implemented so they listen on both
  25  * the socket and the signal event, required for signal handling.
  26  */
  27
  28 /*
  29  * Convert the last socket error code into errno
  30  */
  31 static void
  32 TranslateSocketError(void)
  33 {
  34         switch (WSAGetLastError())
  35         {
  36                 case WSANOTINITIALISED:
  37                 case WSAENETDOWN:
  38                 case WSAEINPROGRESS:
  39                 case WSAEINVAL:
  40                 case WSAESOCKTNOSUPPORT:
  41                 case WSAEFAULT:
  42                 case WSAEINVALIDPROVIDER:
  43                 case WSAEINVALIDPROCTABLE:
  44                 case WSAEMSGSIZE:
  45                         errno = EINVAL;
  46                         break;
  47                 case WSAEAFNOSUPPORT:
  48                         errno = EAFNOSUPPORT;
  49                         break;
  50                 case WSAEMFILE:
  51                         errno = EMFILE;
  52                         break;
  53                 case WSAENOBUFS:
  54                         errno = ENOBUFS;
  55                         break;
  56                 case WSAEPROTONOSUPPORT:
  57                 case WSAEPROTOTYPE:
  58                         errno = EPROTONOSUPPORT;
  59                         break;
  60                 case WSAECONNREFUSED:
  61                         errno = ECONNREFUSED;
  62                         break;
  63                 case WSAEINTR:
  64                         errno = EINTR;
  65                         break;
  66                 case WSAENOTSOCK:
  67                         errno = EBADFD;
  68                         break;
  69                 case WSAEOPNOTSUPP:
  70                         errno = EOPNOTSUPP;
  71                         break;
  72                 case WSAEWOULDBLOCK:
  73                         errno = EWOULDBLOCK;
  74                         break;
  75                 case WSAEACCES:
  76                         errno = EACCES;
  77                         break;
  78                 case WSAENOTCONN:
  79                 case WSAENETRESET:
  80                 case WSAECONNRESET:
  81                 case WSAESHUTDOWN:
  82                 case WSAECONNABORTED:
  83                 case WSAEDISCON:
  84                         errno = ECONNREFUSED;           /* ENOTCONN? */
  85                         break;
  86                 default:
  87                         ereport(NOTICE,
  88                                         (errmsg_internal("Unknown win32 socket error code: %i", WSAGetLastError())));
  89                         errno = EINVAL;
  90         }
  91 }
  92
  93 static int
  94 pgwin32_poll_signals(void)
  95 {
  96         if (UNBLOCKED_SIGNAL_QUEUE())
  97         {
  98                 pgwin32_dispatch_queued_signals();
  99                 errno = EINTR;
 100                 return 1;
 101         }
 102         return 0;
 103 }
 104
 105 static int
 106 isDataGram(SOCKET s)
 107 {
 108         int                     type;
 109         int                     typelen = sizeof(type);
 110
 111         if (getsockopt(s, SOL_SOCKET, SO_TYPE, (char *) &type, &typelen))
 112                 return 1;
 113
 114         return (type == SOCK_DGRAM) ? 1 : 0;
 115 }
 116
 117 int
 118 pgwin32_waitforsinglesocket(SOCKET s, int what, int timeout)
 119 {
 120         static HANDLE waitevent = INVALID_HANDLE_VALUE;
 121         static SOCKET current_socket = -1;
 122         static int      isUDP = 0;
 123         HANDLE          events[2];
 124         int                     r;
 125
 126         if (waitevent == INVALID_HANDLE_VALUE)
 127         {
 128                 waitevent = CreateEvent(NULL, TRUE, FALSE, NULL);
 129
 130                 if (waitevent == INVALID_HANDLE_VALUE)
 131                         ereport(ERROR,
 132                                         (errmsg_internal("Failed to create socket waiting event: %i", (int) GetLastError())));
 133         }
 134         else if (!ResetEvent(waitevent))
 135                 ereport(ERROR,
 136                                 (errmsg_internal("Failed to reset socket waiting event: %i", (int) GetLastError())));
 137
 138         /*
 139          * make sure we don't multiplex this kernel event object with a different
 140          * socket from a previous call
 141          */
 142
 143         if (current_socket != s)
 144         {
 145                 if (current_socket != -1)
 146                         WSAEventSelect(current_socket, waitevent, 0);
 147                 isUDP = isDataGram(s);
 148         }
 149
 150         current_socket = s;
 151
 152         if (WSAEventSelect(s, waitevent, what) == SOCKET_ERROR)
 153         {
 154                 TranslateSocketError();
 155                 return 0;
 156         }
 157
 158         events[0] = pgwin32_signal_event;
 159         events[1] = waitevent;
 160
 161         /*
 162          * Just a workaround of unknown locking problem with writing in UDP socket
 163          * under high load: Client's pgsql backend sleeps infinitely in
 164          * WaitForMultipleObjectsEx, pgstat process sleeps in pgwin32_select().
 165          * So, we will wait with small timeout(0.1 sec) and if sockect is still
 166          * blocked, try WSASend (see comments in pgwin32_select) and wait again.
 167          */
 168         if ((what & FD_WRITE) && isUDP)
 169         {
 170                 for (;;)
 171                 {
 172                         r = WaitForMultipleObjectsEx(2, events, FALSE, 100, TRUE);
 173
 174                         if (r == WAIT_TIMEOUT)
 175                         {
 176                                 char            c;
 177                                 WSABUF          buf;
 178                                 DWORD           sent;
 179
 180                                 buf.buf = &c;
 181                                 buf.len = 0;
 182
 183                                 r = WSASend(s, &buf, 1, &sent, 0, NULL, NULL);
 184                                 if (r == 0)             /* Completed - means things are fine! */
 185                                         return 1;
 186                                 else if (WSAGetLastError() != WSAEWOULDBLOCK)
 187                                 {
 188                                         TranslateSocketError();
 189                                         return 0;
 190                                 }
 191                         }
 192                         else
 193                                 break;
 194                 }
 195         }
 196         else
 197                 r = WaitForMultipleObjectsEx(2, events, FALSE, timeout, TRUE);
 198
 199         if (r == WAIT_OBJECT_0 || r == WAIT_IO_COMPLETION)
 200         {
 201                 pgwin32_dispatch_queued_signals();
 202                 errno = EINTR;
 203                 return 0;
 204         }
 205         if (r == WAIT_OBJECT_0 + 1)
 206                 return 1;
 207         if (r == WAIT_TIMEOUT)
 208                 return 0;
 209         ereport(ERROR,
 210                         (errmsg_internal("Bad return from WaitForMultipleObjects: %i (%i)", r, (int) GetLastError())));
 211         return 0;
 212 }
 213
 214 /*
 215  * Create a socket, setting it to overlapped and non-blocking
 216  */
 217 SOCKET
 218 pgwin32_socket(int af, int type, int protocol)
 219 {
 220         SOCKET          s;
 221         unsigned long on = 1;
 222
 223         s = WSASocket(af, type, protocol, NULL, 0, WSA_FLAG_OVERLAPPED);
 224         if (s == INVALID_SOCKET)
 225         {
 226                 TranslateSocketError();
 227                 return INVALID_SOCKET;
 228         }
 229
 230         if (ioctlsocket(s, FIONBIO, &on))
 231         {
 232                 TranslateSocketError();
 233                 return INVALID_SOCKET;
 234         }
 235         errno = 0;
 236
 237         return s;
 238 }
 239
 240
 241 SOCKET
 242 pgwin32_accept(SOCKET s, struct sockaddr * addr, int *addrlen)
 243 {
 244         SOCKET          rs;
 245
 246         /*
 247          * Poll for signals, but don't return with EINTR, since we don't handle
 248          * that in pqcomm.c
 249          */
 250         pgwin32_poll_signals();
 251
 252         rs = WSAAccept(s, addr, addrlen, NULL, 0);
 253         if (rs == INVALID_SOCKET)
 254         {
 255                 TranslateSocketError();
 256                 return INVALID_SOCKET;
 257         }
 258         return rs;
 259 }
 260
 261
 262 /* No signal delivery during connect. */
 263 int
 264 pgwin32_connect(SOCKET s, const struct sockaddr * addr, int addrlen)
 265 {
 266         int                     r;
 267
 268         r = WSAConnect(s, addr, addrlen, NULL, NULL, NULL, NULL);
 269         if (r == 0)
 270                 return 0;
 271
 272         if (WSAGetLastError() != WSAEWOULDBLOCK)
 273         {
 274                 TranslateSocketError();
 275                 return -1;
 276         }
 277
 278         while (pgwin32_waitforsinglesocket(s, FD_CONNECT, INFINITE) == 0)
 279         {
 280                 /* Loop endlessly as long as we are just delivering signals */
 281         }
 282
 283         return 0;
 284 }
 285
 286 int
 287 pgwin32_recv(SOCKET s, char *buf, int len, int f)
 288 {
 289         WSABUF          wbuf;
 290         int                     r;
 291         DWORD           b;
 292         DWORD           flags = f;
 293         int                     n;
 294
 295         if (pgwin32_poll_signals())
 296                 return -1;
 297
 298         wbuf.len = len;
 299         wbuf.buf = buf;
 300
 301         r = WSARecv(s, &wbuf, 1, &b, &flags, NULL, NULL);
 302         if (r != SOCKET_ERROR && b > 0)
 303                 /* Read succeeded right away */
 304                 return b;
 305
 306         if (r == SOCKET_ERROR &&
 307                 WSAGetLastError() != WSAEWOULDBLOCK)
 308         {
 309                 TranslateSocketError();
 310                 return -1;
 311         }
 312
 313         /* No error, zero bytes (win2000+) or error+WSAEWOULDBLOCK (<=nt4) */
 314
 315         for (n = 0; n < 5; n++)
 316         {
 317                 if (pgwin32_waitforsinglesocket(s, FD_READ | FD_CLOSE | FD_ACCEPT,
 318                                                                                 INFINITE) == 0)
 319                         return -1;                      /* errno already set */
 320
 321                 r = WSARecv(s, &wbuf, 1, &b, &flags, NULL, NULL);
 322                 if (r == SOCKET_ERROR)
 323                 {
 324                         if (WSAGetLastError() == WSAEWOULDBLOCK)
 325                         {
 326                                 /*
 327                                  * There seem to be cases on win2k (at least) where WSARecv
 328                                  * can return WSAEWOULDBLOCK even when
 329                                  * pgwin32_waitforsinglesocket claims the socket is readable.
 330                                  * In this case, just sleep for a moment and try again. We try
 331                                  * up to 5 times - if it fails more than that it's not likely
 332                                  * to ever come back.
 333                                  */
 334                                 pg_usleep(10000);
 335                                 continue;
 336                         }
 337                         TranslateSocketError();
 338                         return -1;
 339                 }
 340                 return b;
 341         }
 342         ereport(NOTICE,
 343           (errmsg_internal("Failed to read from ready socket (after retries)")));
 344         errno = EWOULDBLOCK;
 345         return -1;
 346 }
 347
 348 int
 349 pgwin32_send(SOCKET s, char *buf, int len, int flags)
 350 {
 351         WSABUF          wbuf;
 352         int                     r;
 353         DWORD           b;
 354
 355         if (pgwin32_poll_signals())
 356                 return -1;
 357
 358         wbuf.len = len;
 359         wbuf.buf = buf;
 360
 361         /*
 362          * Readiness of socket to send data to UDP socket may be not true: socket
 363          * can become busy again! So loop until send or error occurs.
 364          */
 365         for (;;)
 366         {
 367                 r = WSASend(s, &wbuf, 1, &b, flags, NULL, NULL);
 368                 if (r != SOCKET_ERROR && b > 0)
 369                         /* Write succeeded right away */
 370                         return b;
 371
 372                 if (r == SOCKET_ERROR &&
 373                         WSAGetLastError() != WSAEWOULDBLOCK)
 374                 {
 375                         TranslateSocketError();
 376                         return -1;
 377                 }
 378
 379                 /* No error, zero bytes (win2000+) or error+WSAEWOULDBLOCK (<=nt4) */
 380
 381                 if (pgwin32_waitforsinglesocket(s, FD_WRITE | FD_CLOSE, INFINITE) == 0)
 382                         return -1;
 383         }
 384
 385         return -1;
 386 }
 387
 388
 389 /*
 390  * Wait for activity on one or more sockets.
 391  * While waiting, allow signals to run
 392  *
 393  * NOTE! Currently does not implement exceptfds check,
 394  * since it is not used in postgresql!
 395  */
 396 int
 397 pgwin32_select(int nfds, fd_set *readfds, fd_set *writefds, fd_set *exceptfds, const struct timeval * timeout)
 398 {
 399         WSAEVENT        events[FD_SETSIZE * 2]; /* worst case is readfds totally
 400                                                                                  * different from writefds, so
 401                                                                                  * 2*FD_SETSIZE sockets */
 402         SOCKET          sockets[FD_SETSIZE * 2];
 403         int                     numevents = 0;
 404         int                     i;
 405         int                     r;
 406         DWORD           timeoutval = WSA_INFINITE;
 407         FD_SET          outreadfds;
 408         FD_SET          outwritefds;
 409         int                     nummatches = 0;
 410
 411         Assert(exceptfds == NULL);
 412
 413         if (pgwin32_poll_signals())
 414                 return -1;
 415
 416         FD_ZERO(&outreadfds);
 417         FD_ZERO(&outwritefds);
 418
 419         /*
 420          * Write FDs are different in the way that it is only flagged by
 421          * WSASelectEvent() if we have tried to write to them first. So try an
 422          * empty write
 423          */
 424         if (writefds)
 425         {
 426                 for (i = 0; i < writefds->fd_count; i++)
 427                 {
 428                         char            c;
 429                         WSABUF          buf;
 430                         DWORD           sent;
 431
 432                         buf.buf = &c;
 433                         buf.len = 0;
 434
 435                         r = WSASend(writefds->fd_array[i], &buf, 1, &sent, 0, NULL, NULL);
 436                         if (r == 0)                     /* Completed - means things are fine! */
 437                                 FD_SET          (writefds->fd_array[i], &outwritefds);
 438
 439                         else
 440                         {                                       /* Not completed */
 441                                 if (WSAGetLastError() != WSAEWOULDBLOCK)
 442
 443                                         /*
 444                                          * Not completed, and not just "would block", so an error
 445                                          * occured
 446                                          */
 447                                         FD_SET          (writefds->fd_array[i], &outwritefds);
 448                         }
 449                 }
 450                 if (outwritefds.fd_count > 0)
 451                 {
 452                         memcpy(writefds, &outwritefds, sizeof(fd_set));
 453                         if (readfds)
 454                                 FD_ZERO(readfds);
 455                         return outwritefds.fd_count;
 456                 }
 457         }
 458
 459
 460         /* Now set up for an actual select */
 461
 462         if (timeout != NULL)
 463         {
 464                 /* timeoutval is in milliseconds */
 465                 timeoutval = timeout->tv_sec * 1000 + timeout->tv_usec / 1000;
 466         }
 467
 468         if (readfds != NULL)
 469         {
 470                 for (i = 0; i < readfds->fd_count; i++)
 471                 {
 472                         events[numevents] = WSACreateEvent();
 473                         sockets[numevents] = readfds->fd_array[i];
 474                         numevents++;
 475                 }
 476         }
 477         if (writefds != NULL)
 478         {
 479                 for (i = 0; i < writefds->fd_count; i++)
 480                 {
 481                         if (!readfds ||
 482                                 !FD_ISSET(writefds->fd_array[i], readfds))
 483                         {
 484                                 /* If the socket is not in the read list */
 485                                 events[numevents] = WSACreateEvent();
 486                                 sockets[numevents] = writefds->fd_array[i];
 487                                 numevents++;
 488                         }
 489                 }
 490         }
 491
 492         for (i = 0; i < numevents; i++)
 493         {
 494                 int                     flags = 0;
 495
 496                 if (readfds && FD_ISSET(sockets[i], readfds))
 497                         flags |= FD_READ | FD_ACCEPT | FD_CLOSE;
 498
 499                 if (writefds && FD_ISSET(sockets[i], writefds))
 500                         flags |= FD_WRITE | FD_CLOSE;
 501
 502                 if (WSAEventSelect(sockets[i], events[i], flags) == SOCKET_ERROR)
 503                 {
 504                         TranslateSocketError();
 505                         for (i = 0; i < numevents; i++)
 506                                 WSACloseEvent(events[i]);
 507                         return -1;
 508                 }
 509         }
 510
 511         events[numevents] = pgwin32_signal_event;
 512         r = WaitForMultipleObjectsEx(numevents + 1, events, FALSE, timeoutval, TRUE);
 513         if (r != WAIT_TIMEOUT && r != WAIT_IO_COMPLETION && r != (WAIT_OBJECT_0 + numevents))
 514         {
 515                 /*
 516                  * We scan all events, even those not signalled, in case more than one
 517                  * event has been tagged but Wait.. can only return one.
 518                  */
 519                 WSANETWORKEVENTS resEvents;
 520
 521                 for (i = 0; i < numevents; i++)
 522                 {
 523                         ZeroMemory(&resEvents, sizeof(resEvents));
 524                         if (WSAEnumNetworkEvents(sockets[i], events[i], &resEvents) == SOCKET_ERROR)
 525                                 ereport(FATAL,
 526                                                 (errmsg_internal("failed to enumerate network events: %i", (int) GetLastError())));
 527                         /* Read activity? */
 528                         if (readfds && FD_ISSET(sockets[i], readfds))
 529                         {
 530                                 if ((resEvents.lNetworkEvents & FD_READ) ||
 531                                         (resEvents.lNetworkEvents & FD_ACCEPT) ||
 532                                         (resEvents.lNetworkEvents & FD_CLOSE))
 533                                 {
 534                                         FD_SET          (sockets[i], &outreadfds);
 535
 536                                         nummatches++;
 537                                 }
 538                         }
 539                         /* Write activity? */
 540                         if (writefds && FD_ISSET(sockets[i], writefds))
 541                         {
 542                                 if ((resEvents.lNetworkEvents & FD_WRITE) ||
 543                                         (resEvents.lNetworkEvents & FD_CLOSE))
 544                                 {
 545                                         FD_SET          (sockets[i], &outwritefds);
 546
 547                                         nummatches++;
 548                                 }
 549                         }
 550                 }
 551         }
 552
 553         /* Clean up all handles */
 554         for (i = 0; i < numevents; i++)
 555         {
 556                 WSAEventSelect(sockets[i], events[i], 0);
 557                 WSACloseEvent(events[i]);
 558         }
 559
 560         if (r == WSA_WAIT_TIMEOUT)
 561         {
 562                 if (readfds)
 563                         FD_ZERO(readfds);
 564                 if (writefds)
 565                         FD_ZERO(writefds);
 566                 return 0;
 567         }
 568
 569         if (r == WAIT_OBJECT_0 + numevents)
 570         {
 571                 pgwin32_dispatch_queued_signals();
 572                 errno = EINTR;
 573                 if (readfds)
 574                         FD_ZERO(readfds);
 575                 if (writefds)
 576                         FD_ZERO(writefds);
 577                 return -1;
 578         }
 579
 580         /* Overwrite socket sets with our resulting values */
 581         if (readfds)
 582                 memcpy(readfds, &outreadfds, sizeof(fd_set));
 583         if (writefds)
 584                 memcpy(writefds, &outwritefds, sizeof(fd_set));
 585         return nummatches;
 586 }
 587
 588
 589 /*
 590  * Return win32 error string, since strerror can't
 591  * handle winsock codes
 592  */
 593 static char wserrbuf[256];
 594 const char *
 595 pgwin32_socket_strerror(int err)
 596 {
 597         static HANDLE handleDLL = INVALID_HANDLE_VALUE;
 598
 599         if (handleDLL == INVALID_HANDLE_VALUE)
 600         {
 601                 handleDLL = LoadLibraryEx("netmsg.dll", NULL, DONT_RESOLVE_DLL_REFERENCES | LOAD_LIBRARY_AS_DATAFILE);
 602                 if (handleDLL == NULL)
 603                         ereport(FATAL,
 604                                         (errmsg_internal("Failed to load netmsg.dll: %i", (int) GetLastError())));
 605         }
 606
 607         ZeroMemory(&wserrbuf, sizeof(wserrbuf));
 608         if (FormatMessage(FORMAT_MESSAGE_IGNORE_INSERTS | FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_FROM_HMODULE,
 609                                           handleDLL,
 610                                           err,
 611                                           MAKELANGID(LANG_ENGLISH, SUBLANG_DEFAULT),
 612                                           wserrbuf,
 613                                           sizeof(wserrbuf) - 1,
 614                                           NULL) == 0)
 615         {
 616                 /* Failed to get id */
 617                 sprintf(wserrbuf, "Unknown winsock error %i", err);
 618         }
 619         return wserrbuf;
 620 }