implement a native version of poll for Win32

2008-09-22  Paolo Bonzini  <bonzini@gnu.org>

	* lib/poll.c: Rewrite.
	* modules/poll: Depend on alloca.

1 files changed, 393 insertions, 52 deletions

diff --git a/lib/poll.c b/lib/poll.c
index e0714f09a9..b4001283d3 100644
--- a/lib/poll.c
+++ b/lib/poll.c
@@ -20,14 +20,27 @@
    Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.  */
 
 #include <config.h>
+#include <alloca.h>
 
 #include <sys/types.h>
 #include "poll.h"
 #include <errno.h>
 #include <limits.h>
+#include <assert.h>
+
+#if (defined _WIN32 || defined __WIN32__) && ! defined __CYGWIN__
+#define WIN32_NATIVE
+#include <winsock2.h>
+#include <windows.h>
+#include <io.h>
+#include <stdio.h>
+#include <conio.h>
+#else
+#include <sys/time.h>
 #include <sys/socket.h>
 #include <sys/select.h>
 #include <unistd.h>
+#endif
 
 #ifdef HAVE_SYS_IOCTL_H
 #include <sys/ioctl.h>
@@ -36,7 +49,6 @@
 #include <sys/filio.h>
 #endif
 
-#include <sys/time.h>
 #include <time.h>
 
 #ifndef INFTIM
@@ -48,12 +60,228 @@
 #define MSG_PEEK 0
 #endif
 
+#ifdef WIN32_NATIVE
+
+/* Declare data structures for ntdll functions.  */
+typedef struct _FILE_PIPE_LOCAL_INFORMATION {
+  ULONG NamedPipeType;
+  ULONG NamedPipeConfiguration;
+  ULONG MaximumInstances;
+  ULONG CurrentInstances;
+  ULONG InboundQuota;
+  ULONG ReadDataAvailable;
+  ULONG OutboundQuota;
+  ULONG WriteQuotaAvailable;
+  ULONG NamedPipeState;
+  ULONG NamedPipeEnd;
+} FILE_PIPE_LOCAL_INFORMATION, *PFILE_PIPE_LOCAL_INFORMATION;
+
+typedef struct _IO_STATUS_BLOCK
+{
+  union {
+    DWORD Status;
+    PVOID Pointer;
+  } u;
+  ULONG_PTR Information;
+} IO_STATUS_BLOCK, *PIO_STATUS_BLOCK;
+
+typedef enum _FILE_INFORMATION_CLASS {
+  FilePipeLocalInformation = 24
+} FILE_INFORMATION_CLASS, *PFILE_INFORMATION_CLASS;
+
+typedef DWORD (WINAPI *PNtQueryInformationFile)
+	 (HANDLE, IO_STATUS_BLOCK *, VOID *, ULONG, FILE_INFORMATION_CLASS);
+
+#ifndef PIPE_BUF
+#define PIPE_BUF	512
+#endif
+
+/* Compute revents values for file handle H.  */
+
+static int
+win32_compute_revents (HANDLE h, int sought)
+{
+  int i, ret, happened;
+  INPUT_RECORD *irbuffer;
+  DWORD avail, nbuffer;
+  BOOL bRet;
+  IO_STATUS_BLOCK iosb;
+  FILE_PIPE_LOCAL_INFORMATION fpli;
+  static PNtQueryInformationFile NtQueryInformationFile;
+  static BOOL once_only;
+
+  switch (GetFileType (h))
+    {
+    case FILE_TYPE_PIPE:
+      if (!once_only)
+	{
+	  NtQueryInformationFile = (PNtQueryInformationFile)
+	    GetProcAddress (GetModuleHandle ("ntdll.dll"),
+			    "NtQueryInformationFile");
+	  once_only = TRUE;
+	}
+
+      happened = 0;
+      if (PeekNamedPipe (h, NULL, 0, NULL, &avail, NULL) != 0)
+	{
+	  if (avail)
+	    happened |= sought & (POLLIN | POLLRDNORM);
+	}
+
+      else
+	{
+	  /* It was the write-end of the pipe.  Check if it is writable.
+	     If NtQueryInformationFile fails, optimistically assume the pipe is
+	     writable.  This could happen on Win9x, where NtQueryInformationFile
+	     is not available, or if we inherit a pipe that doesn't permit
+	     FILE_READ_ATTRIBUTES access on the write end (I think this should
+	     not happen since WinXP SP2; WINE seems fine too).  Otherwise,
+	     ensure that enough space is available for atomic writes.  */
+          memset (&iosb, 0, sizeof (iosb));
+          memset (&fpli, 0, sizeof (fpli));
+
+          if (!NtQueryInformationFile
+              || NtQueryInformationFile (h, &iosb, &fpli, sizeof (fpli),
+				         FilePipeLocalInformation)
+	      || fpli.WriteQuotaAvailable >= PIPE_BUF
+	      || (fpli.OutboundQuota < PIPE_BUF &&
+	          fpli.WriteQuotaAvailable == fpli.OutboundQuota))
+	    happened |= sought & (POLLOUT | POLLWRNORM | POLLWRBAND);
+	}
+      return happened;
+
+    case FILE_TYPE_CHAR:
+      ret = WaitForSingleObject (h, 0);
+      if (ret == WAIT_OBJECT_0)
+        {
+	  nbuffer = avail = 0;
+	  bRet = GetNumberOfConsoleInputEvents (h, &nbuffer);
+	  if (!bRet || nbuffer == 0)
+	    return POLLHUP;
+
+	  irbuffer = (INPUT_RECORD *) alloca (nbuffer * sizeof (INPUT_RECORD));
+	  bRet = PeekConsoleInput (h, irbuffer, nbuffer, &avail);
+	  if (!bRet || avail == 0)
+	    return POLLHUP;
+
+	  for (i = 0; i < avail; i++)
+	    if (irbuffer[i].EventType == KEY_EVENT)
+	      return sought & ~(POLLPRI | POLLRDBAND);
+	}
+      break;
+
+    default:
+      ret = WaitForSingleObject (h, 0);
+      if (ret == WAIT_OBJECT_0)
+        return sought & ~(POLLPRI | POLLRDBAND);
+
+      break;
+    }
+
+  return sought & (POLLOUT | POLLWRNORM | POLLWRBAND);
+}
+
+/* Convert fd_sets returned by select into revents values.  */
+
+static int
+win32_compute_revents_socket (SOCKET h, int sought, long lNetworkEvents)
+{
+  int happened = 0;
+
+  if ((lNetworkEvents & (FD_READ | FD_ACCEPT | FD_CLOSE)) == FD_ACCEPT)
+    happened |= (POLLIN | POLLRDNORM) & sought;
+
+  else if (lNetworkEvents & (FD_READ | FD_ACCEPT | FD_CLOSE))
+    {
+      int r, error;
+
+      char data[64];
+      WSASetLastError (0);
+      r = recv (h, data, sizeof (data), MSG_PEEK);
+      error = WSAGetLastError ();
+      WSASetLastError (0);
+
+      if (r > 0 || error == WSAENOTCONN)
+        happened |= (POLLIN | POLLRDNORM) & sought;
+
+      /* Distinguish hung-up sockets from other errors.  */
+      else if (r == 0 || error == WSAESHUTDOWN || error == WSAECONNRESET
+	       || error == WSAECONNABORTED || error == WSAENETRESET)
+        happened |= POLLHUP;
+
+      else
+        happened |= POLLERR;
+    }
+
+  if (lNetworkEvents & (FD_WRITE | FD_CONNECT))
+    happened |= (POLLOUT | POLLWRNORM | POLLWRBAND) & sought;
+
+  if (lNetworkEvents & FD_OOB)
+    happened |= (POLLPRI | POLLRDBAND) & sought;
+
+  return happened;
+}
+
+#else /* !MinGW */
+
+/* Convert select(2) returned fd_sets into poll(2) revents values.  */
+static int
+compute_revents (int fd, int sought, fd_set *rfds, fd_set *wfds, fd_set *efds)
+{
+  int happened = 0;
+  if (FD_ISSET (fd, rfds))
+    {
+      int r;
+      int socket_errno;
+
+#if defined __MACH__ && defined __APPLE__
+      /* There is a bug in Mac OS X that causes it to ignore MSG_PEEK
+         for some kinds of descriptors.  Detect if this descriptor is a
+         connected socket, a server socket, or something else using a
+         0-byte recv, and use ioctl(2) to detect POLLHUP.  */
+      r = recv (fd, NULL, 0, MSG_PEEK);
+      socket_errno = (r < 0) ? errno : 0;
+      if (r == 0 || socket_errno == ENOTSOCK)
+	ioctl (fd, FIONREAD, &r);
+#else
+      char data[64];
+      r = recv (fd, data, sizeof (data), MSG_PEEK);
+      socket_errno = (r < 0) ? errno : 0;
+#endif
+      if (r == 0)
+	happened |= POLLHUP;
+
+      /* If the event happened on an unconnected server socket,
+         that's fine. */
+      else if (r > 0 || ( /* (r == -1) && */ socket_errno == ENOTCONN))
+	happened |= (POLLIN | POLLRDNORM) & sought;
+
+      /* Distinguish hung-up sockets from other errors.  */
+      else if (socket_errno == ESHUTDOWN || socket_errno == ECONNRESET
+	       || socket_errno == ECONNABORTED || socket_errno == ENETRESET)
+	happened |= POLLHUP;
+
+      else
+	happened |= POLLERR;
+    }
+
+  if (FD_ISSET (fd, wfds))
+    happened |= (POLLOUT | POLLWRNORM | POLLWRBAND) & sought;
+
+  if (FD_ISSET (fd, efds))
+    happened |= (POLLPRI | POLLRDBAND) & sought;
+
+  return happened;
+}
+#endif /* !MinGW */
+
 int
 poll (pfd, nfd, timeout)
      struct pollfd *pfd;
      nfds_t nfd;
      int timeout;
 {
+#ifndef WIN32_NATIVE
   fd_set rfds, wfds, efds;
   struct timeval tv;
   struct timeval *ptv;
@@ -137,16 +365,11 @@ poll (pfd, nfd, timeout)
 			       | POLLWRNORM | POLLWRBAND)))
 	{
 	  maxfd = pfd[i].fd;
-
-	  /* Windows use a linear array of sockets (of size FD_SETSIZE). The
-	     descriptor value is not used to address the array.  */
-#if defined __CYGWIN__ || (!defined _WIN32 && !defined __WIN32__)
 	  if (maxfd > FD_SETSIZE)
 	    {
 	      errno = EOVERFLOW;
 	      return -1;
 	    }
-#endif
 	}
     }
 
@@ -162,61 +385,179 @@ poll (pfd, nfd, timeout)
       pfd[i].revents = 0;
     else
       {
-	int happened = 0, sought = pfd[i].events;
-	if (FD_ISSET (pfd[i].fd, &rfds))
+        int happened = compute_revents (pfd[i].fd, pfd[i].events,
+                                        &rfds, &wfds, &efds);
+	if (happened)
 	  {
-	    int r;
-	    int socket_errno;
+	    pfd[i].revents = happened;
+	    rc++;
+	  }
+      }
 
-#if defined __MACH__ && defined __APPLE__
-	    /* There is a bug in Mac OS X that causes it to ignore MSG_PEEK
-	       for some kinds of descriptors.  Detect if this descriptor is a
-	       connected socket, a server socket, or something else using a
-	       0-byte recv, and use ioctl(2) to detect POLLHUP.  */
-	    r = recv (pfd[i].fd, NULL, 0, MSG_PEEK);
-	    socket_errno = (r < 0) ? errno : 0;
-	    if (r == 0 || socket_errno == ENOTSOCK)
-	      ioctl(pfd[i].fd, FIONREAD, &r);
+  return rc;
 #else
-	    char data[64];
-	    r = recv (pfd[i].fd, data, sizeof (data), MSG_PEEK);
-            
-# ifdef WIN32
-	    if (r < 0 && GetLastError() == 10057) /* server socket */
-              socket_errno = ENOTCONN;
-	    else
-# endif
-	    socket_errno = (r < 0) ? errno : 0;
-#endif
-	    if (r == 0)
-	      happened |= POLLHUP;
+  static struct timeval tv0;
+  static HANDLE hEvent;
+  WSANETWORKEVENTS ev;
+  HANDLE h, handle_array[FD_SETSIZE + 2];
+  DWORD ret, wait_timeout, nhandles;
+  fd_set rfds, wfds, xfds;
+  BOOL poll_again;
+  MSG msg;
+  char sockbuf[256];
+  int rc;
+  nfds_t i;
 
-	    /* If the event happened on an unconnected server socket,
-	       that's fine. */
-	    else if (r > 0 || ( /* (r == -1) && */ socket_errno == ENOTCONN))
-	      happened |= (POLLIN | POLLRDNORM) & sought;
+  if (nfd < 0 || timeout < -1)
+    {
+      errno = EINVAL;
+      return -1;
+    }
 
-	    /* Distinguish hung-up sockets from other errors.  */
-	    else if (socket_errno == ESHUTDOWN || socket_errno == ECONNRESET
-		     || socket_errno == ECONNABORTED || socket_errno == ENETRESET)
-	      happened |= POLLHUP;
+  if (!hEvent)
+    hEvent = CreateEvent (NULL, FALSE, FALSE, NULL);
 
-	    else
-	      happened |= POLLERR;
-	  }
+  handle_array[0] = hEvent;
+  nhandles = 1;
+  FD_ZERO (&rfds);
+  FD_ZERO (&wfds);
+  FD_ZERO (&xfds);
 
-	if (FD_ISSET (pfd[i].fd, &wfds))
-	  happened |= (POLLOUT | POLLWRNORM | POLLWRBAND) & sought;
+  /* Classify socket handles and create fd sets. */
+  for (i = 0; i < nfd; i++)
+    {
+      size_t optlen = sizeof(sockbuf);
+      pfd[i].revents = 0;
+      if (pfd[i].fd < 0)
+        continue;
+      if (!(pfd[i].events & (POLLIN | POLLRDNORM |
+                             POLLOUT | POLLWRNORM | POLLWRBAND)))
+	continue;
 
-	if (FD_ISSET (pfd[i].fd, &efds))
-	  happened |= (POLLPRI | POLLRDBAND) & sought;
+      h = (HANDLE) _get_osfhandle (pfd[i].fd);
+      assert (h != NULL);
 
-	if (happened)
-	  {
-	    pfd[i].revents = happened;
-	    rc++;
-	  }
-      }
+      /* Under Wine, it seems that getsockopt returns 0 for pipes too.
+	 WSAEnumNetworkEvents instead distinguishes the two correctly.  */
+      ev.lNetworkEvents = 0xDEADBEEF;
+      WSAEnumNetworkEvents ((SOCKET) h, NULL, &ev);
+      if (ev.lNetworkEvents != 0xDEADBEEF)
+        {
+          int requested = FD_CLOSE;
+
+          /* see above; socket handles are mapped onto select.  */
+          if (pfd[i].events & (POLLIN | POLLRDNORM))
+	    {
+              requested |= FD_READ | FD_ACCEPT;
+	      FD_SET ((SOCKET) h, &rfds);
+	    }
+          if (pfd[i].events & (POLLOUT | POLLWRNORM | POLLWRBAND))
+	    {
+              requested |= FD_WRITE | FD_CONNECT;
+	      FD_SET ((SOCKET) h, &wfds);
+	    }
+          if (pfd[i].events & (POLLPRI | POLLRDBAND))
+	    {
+              requested |= FD_OOB;
+	      FD_SET ((SOCKET) h, &xfds);
+	    }
+
+          if (requested)
+            WSAEventSelect ((SOCKET) h, hEvent, requested);
+        }
+      else
+        {
+          handle_array[nhandles++] = h;
+
+	  /* Poll now.  If we get an event, do not poll again.  */
+          pfd[i].revents = win32_compute_revents (h, pfd[i].events);
+          if (pfd[i].revents)
+	    wait_timeout = 0;
+        }
+    }
+
+  if (select (0, &rfds, &wfds, &xfds, &tv0) > 0)
+    {
+      /* Do MsgWaitForMultipleObjects anyway to dispatch messages, but
+	 no need to call select again.  */
+      poll_again = FALSE;
+      wait_timeout = 0;
+    }
+  else
+    {
+      poll_again = TRUE;
+      if (timeout == INFTIM)
+        wait_timeout = INFINITE;
+      else
+        wait_timeout = timeout;
+    }
+
+  for (;;)
+    {
+      ret = MsgWaitForMultipleObjects (nhandles, handle_array, FALSE,
+				       wait_timeout, QS_ALLINPUT);
+
+      if (ret == WAIT_OBJECT_0 + nhandles)
+	{
+          /* new input of some other kind */
+	  BOOL bRet;
+          while ((bRet = PeekMessage (&msg, NULL, 0, 0, PM_REMOVE)) != 0)
+            {
+              TranslateMessage (&msg);
+              DispatchMessage (&msg);
+            }
+	}
+      else
+	break;
+    }
+
+  if (poll_again)
+    select (0, &rfds, &wfds, &xfds, &tv0);
+
+  /* Place a sentinel at the end of the array.  */
+  handle_array[nhandles] = NULL;
+  nhandles = 1;
+  for (i = 0; i < nfd; i++)
+    {
+      int happened;
+
+      if (pfd[i].fd < 0)
+        continue;
+      if (!(pfd[i].events & (POLLIN | POLLRDNORM |
+                             POLLOUT | POLLWRNORM | POLLWRBAND)))
+	continue;
+
+      h = (HANDLE) _get_osfhandle (pfd[i].fd);
+      if (h != handle_array[nhandles])
+        {
+          /* It's a socket.  */
+          WSAEnumNetworkEvents ((SOCKET) h, NULL, &ev);
+	  WSAEventSelect ((SOCKET) h, 0, 0);
+
+	  /* If we're lucky, WSAEnumNetworkEvents already provided a way
+	     to distinguish FD_READ and FD_ACCEPT; this saves a recv later.  */
+	  if (FD_ISSET ((SOCKET) h, &rfds)
+	      && !(ev.lNetworkEvents & (FD_READ | FD_ACCEPT)))
+	    ev.lNetworkEvents |= FD_READ | FD_ACCEPT;
+	  if (FD_ISSET ((SOCKET) h, &wfds))
+	    ev.lNetworkEvents |= FD_WRITE | FD_CONNECT;
+	  if (FD_ISSET ((SOCKET) h, &xfds))
+	    ev.lNetworkEvents |= FD_OOB;
+	     
+          happened = win32_compute_revents_socket ((SOCKET) h, pfd[i].events,
+						   ev.lNetworkEvents);
+        }
+      else
+        {
+          /* Not a socket.  */
+          nhandles++;
+          happened = win32_compute_revents (h, pfd[i].events);
+        }
+
+       if ((pfd[i].revents |= happened) != 0)
+        rc++;
+    }
 
   return rc;
+#endif
 }