Check for STATUS_DELETE_PENDING on Windows.

1.  Update our open() wrapper to check for NT's STATUS_DELETE_PENDING
and translate it to Unix-like errors.  This is done with
RtlGetLastNtStatus(), which is dynamically loaded from ntdll.  A new
file win32ntdll.c centralizes lookup of NT functions, in case we decide
to add more in the future.

2.  Remove non-working code that was trying to do something similar for
stat(), and just reuse the open() wrapper code.  As a side effect,
stat() also gains resilience against "sharing violation" errors.

3.  Since stat() is used very early in process startup, remove the
requirement that the Win32 signal event has been created before
pgwin32_open_handle() is reached.  Instead, teach pg_usleep() to fall
back to a non-interruptible sleep if reached before the signal event is
available.

This could be back-patched, but for now it's in master only.  The
problem has apparently been with us for a long time and generated only a
few complaints.  Proposed patches trigger it more often, which led to
this investigation and fix.

Reviewed-by: Andres Freund <andres@anarazel.de>
Reviewed-by: Alexander Lakhin <exclusion@gmail.com>
Reviewed-by: Juan José Santamaría Flecha <juanjo.santamaria@gmail.com>
Discussion: https://postgr.es/m/CA%2BhUKGJz_pZTF9mckn6XgSv69%2BjGwdgLkxZ6b3NWGLBCVjqUZA%40mail.gmail.com

configure

@@ -16738,6 +16738,12 @@ esac
  ;;
 esac
 
+  case " $LIBOBJS " in
+  *" win32ntdll.$ac_objext "* ) ;;
+  *) LIBOBJS="$LIBOBJS win32ntdll.$ac_objext"
+ ;;
+esac
+
   case " $LIBOBJS " in
   *" win32security.$ac_objext "* ) ;;
   *) LIBOBJS="$LIBOBJS win32security.$ac_objext"

configure.ac

@@ -1932,6 +1932,7 @@ if test "$PORTNAME" = "win32"; then
   AC_LIBOBJ(system)
   AC_LIBOBJ(win32env)
   AC_LIBOBJ(win32error)
+  AC_LIBOBJ(win32ntdll)
   AC_LIBOBJ(win32security)
   AC_LIBOBJ(win32setlocale)
   AC_LIBOBJ(win32stat)

src/backend/port/win32/signal.c

@@ -52,7 +52,17 @@ static BOOL WINAPI pg_console_handler(DWORD dwCtrlType);
 void
 pg_usleep(long microsec)
 {
-	Assert(pgwin32_signal_event != NULL);
+	if (unlikely(pgwin32_signal_event == NULL))
+	{
+		/*
+		 * If we're reached by pgwin32_open_handle() early in startup before
+		 * the signal event is set up, just fall back to a regular
+		 * non-interruptible sleep.
+		 */
+		SleepEx((microsec < 500 ? 1 : (microsec + 500) / 1000), FALSE);
+		return;
+	}
+
 	if (WaitForSingleObject(pgwin32_signal_event,
 							(microsec < 500 ? 1 : (microsec + 500) / 1000))
 		== WAIT_OBJECT_0)

src/include/port.h

@@ -296,6 +296,7 @@ extern bool rmtree(const char *path, bool rmtopdir);
  * passing of other special options.
  */
 #define		O_DIRECT	0x80000000
+extern HANDLE pgwin32_open_handle(const char *, int, bool);
 extern int	pgwin32_open(const char *, int,...);
 extern FILE *pgwin32_fopen(const char *, const char *);
 #define		open(a,b,c) pgwin32_open(a,b,c)

src/include/port/win32ntdll.h

@@ -0,0 +1,27 @@
+/*-------------------------------------------------------------------------
+ *
+ * win32ntdll.h
+ *	  Dynamically loaded Windows NT functions.
+ *
+ * Portions Copyright (c) 2021, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/port/win32ntdll.h
+ *
+ *-------------------------------------------------------------------------
+ */
+
+/*
+ * Because this includes NT headers that normally conflict with Win32 headers,
+ * any translation unit that includes it should #define UMDF_USING_NTSTATUS
+ * before including <windows.h>.
+ */
+
+#include <ntstatus.h>
+#include <winternl.h>
+
+typedef NTSTATUS (__stdcall *RtlGetLastNtStatus_t) (void);
+
+extern RtlGetLastNtStatus_t pg_RtlGetLastNtStatus;
+
+extern int	initialize_ntdll(void);

src/port/open.c

@@ -13,17 +13,20 @@
 
 #ifdef WIN32
 
+#define UMDF_USING_NTSTATUS
+
 #ifndef FRONTEND
 #include "postgres.h"
 #else
 #include "postgres_fe.h"
 #endif
 
+#include "port/win32ntdll.h"
+
 #include <fcntl.h>
 #include <assert.h>
 #include <sys/stat.h>
 
-
 static int
 openFlagsToCreateFileFlags(int openFlags)
 {
@@ -56,38 +59,25 @@ openFlagsToCreateFileFlags(int openFlags)
 }
 
 /*
- *	 - file attribute setting, based on fileMode?
+ * Internal function used by pgwin32_open() and _pgstat64().  When
+ * backup_semantics is true, directories may be opened (for limited uses).  On
+ * failure, INVALID_HANDLE_VALUE is returned and errno is set.
  */
-int
-pgwin32_open(const char *fileName, int fileFlags,...)
+HANDLE
+pgwin32_open_handle(const char *fileName, int fileFlags, bool backup_semantics)
 {
-	int			fd;
-	HANDLE		h = INVALID_HANDLE_VALUE;
+	HANDLE		h;
 	SECURITY_ATTRIBUTES sa;
 	int			loops = 0;
 
+	if (initialize_ntdll() < 0)
+		return INVALID_HANDLE_VALUE;
+
 	/* Check that we can handle the request */
 	assert((fileFlags & ((O_RDONLY | O_WRONLY | O_RDWR) | O_APPEND |
 						 (O_RANDOM | O_SEQUENTIAL | O_TEMPORARY) |
 						 _O_SHORT_LIVED | O_DSYNC | O_DIRECT |
 						 (O_CREAT | O_TRUNC | O_EXCL) | (O_TEXT | O_BINARY))) == fileFlags);
-#ifndef FRONTEND
-	Assert(pgwin32_signal_event != NULL);	/* small chance of pg_usleep() */
-#endif
-
-#ifdef FRONTEND
-
-	/*
-	 * Since PostgreSQL 12, those concurrent-safe versions of open() and
-	 * fopen() can be used by frontends, having as side-effect to switch the
-	 * file-translation mode from O_TEXT to O_BINARY if none is specified.
-	 * Caller may want to enforce the binary or text mode, but if nothing is
-	 * defined make sure that the default mode maps with what versions older
-	 * than 12 have been doing.
-	 */
-	if ((fileFlags & O_BINARY) == 0)
-		fileFlags |= O_TEXT;
-#endif
 
 	sa.nLength = sizeof(sa);
 	sa.bInheritHandle = TRUE;
@@ -102,6 +92,7 @@ pgwin32_open(const char *fileName, int fileFlags,...)
 						   &sa,
 						   openFlagsToCreateFileFlags(fileFlags),
 						   FILE_ATTRIBUTE_NORMAL |
+						   (backup_semantics ? FILE_FLAG_BACKUP_SEMANTICS : 0) |
 						   ((fileFlags & O_RANDOM) ? FILE_FLAG_RANDOM_ACCESS : 0) |
 						   ((fileFlags & O_SEQUENTIAL) ? FILE_FLAG_SEQUENTIAL_SCAN : 0) |
 						   ((fileFlags & _O_SHORT_LIVED) ? FILE_ATTRIBUTE_TEMPORARY : 0) |
@@ -140,38 +131,55 @@ pgwin32_open(const char *fileName, int fileFlags,...)
 		/*
 		 * ERROR_ACCESS_DENIED is returned if the file is deleted but not yet
 		 * gone (Windows NT status code is STATUS_DELETE_PENDING).  In that
-		 * case we want to wait a bit and try again, giving up after 1 second
-		 * (since this condition should never persist very long).  However,
-		 * there are other commonly-hit cases that return ERROR_ACCESS_DENIED,
-		 * so care is needed.  In particular that happens if we try to open a
-		 * directory, or of course if there's an actual file-permissions
-		 * problem.  To distinguish these cases, try a stat().  In the
-		 * delete-pending case, it will either also get STATUS_DELETE_PENDING,
-		 * or it will see the file as gone and fail with ENOENT.  In other
-		 * cases it will usually succeed.  The only somewhat-likely case where
-		 * this coding will uselessly wait is if there's a permissions problem
-		 * with a containing directory, which we hope will never happen in any
-		 * performance-critical code paths.
+		 * case, we'd better ask for the NT status too so we can translate it
+		 * to a more Unix-like error.  We hope that nothing clobbers the NT
+		 * status in between the internal NtCreateFile() call and CreateFile()
+		 * returning.
+		 *
+		 * If there's no O_CREAT flag, then we'll pretend the file is
+		 * invisible.  With O_CREAT, we have no choice but to report that
+		 * there's a file in the way (which wouldn't happen on Unix).
 		 */
-		if (err == ERROR_ACCESS_DENIED)
+		if (err == ERROR_ACCESS_DENIED &&
+			pg_RtlGetLastNtStatus() == STATUS_DELETE_PENDING)
 		{
-			if (loops < 10)
-			{
-				struct stat st;
-
-				if (stat(fileName, &st) != 0)
-				{
-					pg_usleep(100000);
-					loops++;
-					continue;
-				}
-			}
+			if (fileFlags & O_CREAT)
+				err = ERROR_FILE_EXISTS;
+			else
+				err = ERROR_FILE_NOT_FOUND;
 		}
 
 		_dosmaperr(err);
-		return -1;
+		return INVALID_HANDLE_VALUE;
 	}
 
+	return h;
+}
+
+int
+pgwin32_open(const char *fileName, int fileFlags,...)
+{
+	HANDLE		h;
+	int			fd;
+
+	h = pgwin32_open_handle(fileName, fileFlags, false);
+	if (h == INVALID_HANDLE_VALUE)
+		return -1;
+
+#ifdef FRONTEND
+
+	/*
+	 * Since PostgreSQL 12, those concurrent-safe versions of open() and
+	 * fopen() can be used by frontends, having as side-effect to switch the
+	 * file-translation mode from O_TEXT to O_BINARY if none is specified.
+	 * Caller may want to enforce the binary or text mode, but if nothing is
+	 * defined make sure that the default mode maps with what versions older
+	 * than 12 have been doing.
+	 */
+	if ((fileFlags & O_BINARY) == 0)
+		fileFlags |= O_TEXT;
+#endif
+
 	/* _open_osfhandle will, on error, set errno accordingly */
 	if ((fd = _open_osfhandle((intptr_t) h, fileFlags & O_APPEND)) < 0)
 		CloseHandle(h);			/* will not affect errno */

src/port/win32ntdll.c

@@ -0,0 +1,69 @@
+/*-------------------------------------------------------------------------
+ *
+ * win32ntdll.c
+ *	  Dynamically loaded Windows NT functions.
+ *
+ * Portions Copyright (c) 2021, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ *
+ * IDENTIFICATION
+ *	  src/port/win32ntdll.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#define UMDF_USING_NTSTATUS
+
+#include "c.h"
+
+#include "port/win32ntdll.h"
+
+RtlGetLastNtStatus_t pg_RtlGetLastNtStatus;
+
+typedef struct NtDllRoutine
+{
+	const char *name;
+	pg_funcptr_t *address;
+} NtDllRoutine;
+
+static const NtDllRoutine routines[] = {
+	{"RtlGetLastNtStatus", (pg_funcptr_t *) &pg_RtlGetLastNtStatus}
+};
+
+static bool initialized;
+
+int
+initialize_ntdll(void)
+{
+	HMODULE		module;
+
+	if (initialized)
+		return 0;
+
+	if (!(module = LoadLibraryEx("ntdll.dll", NULL, 0)))
+	{
+		_dosmaperr(GetLastError());
+		return -1;
+	}
+
+	for (int i = 0; i < lengthof(routines); ++i)
+	{
+		pg_funcptr_t address;
+
+		address = (pg_funcptr_t) GetProcAddress(module, routines[i].name);
+		if (!address)
+		{
+			_dosmaperr(GetLastError());
+			FreeLibrary(module);
+
+			return -1;
+		}
+
+		*(pg_funcptr_t *) routines[i].address = address;
+	}
+
+	initialized = true;
+
+	return 0;
+}

src/port/win32stat.c

@@ -18,53 +18,6 @@
 #include "c.h"
 #include <windows.h>
 
-/*
- * In order to support MinGW and MSVC2013 we use NtQueryInformationFile as an
- * alternative for GetFileInformationByHandleEx. It is loaded from the ntdll
- * library.
- */
-#if _WIN32_WINNT < 0x0600
-#include <winternl.h>
-
-#if !defined(__MINGW32__) && !defined(__MINGW64__)
-/* MinGW includes this in <winternl.h>, but it is missing in MSVC */
-typedef struct _FILE_STANDARD_INFORMATION
-{
-	LARGE_INTEGER AllocationSize;
-	LARGE_INTEGER EndOfFile;
-	ULONG		NumberOfLinks;
-	BOOLEAN		DeletePending;
-	BOOLEAN		Directory;
-} FILE_STANDARD_INFORMATION;
-#define FileStandardInformation 5
-#endif							/* !defined(__MINGW32__) &&
-								 * !defined(__MINGW64__) */
-
-typedef NTSTATUS (NTAPI * PFN_NTQUERYINFORMATIONFILE)
-			(IN HANDLE FileHandle,
-			 OUT PIO_STATUS_BLOCK IoStatusBlock,
-			 OUT PVOID FileInformation,
-			 IN ULONG Length,
-			 IN FILE_INFORMATION_CLASS FileInformationClass);
-
-static PFN_NTQUERYINFORMATIONFILE _NtQueryInformationFile = NULL;
-
-static HMODULE ntdll = NULL;
-
-/*
- * Load DLL file just once regardless of how many functions we load/call in it.
- */
-static void
-LoadNtdll(void)
-{
-	if (ntdll != NULL)
-		return;
-	ntdll = LoadLibraryEx("ntdll.dll", NULL, 0);
-}
-
-#endif							/* _WIN32_WINNT < 0x0600 */
-
-
 /*
  * Convert a FILETIME struct into a 64 bit time_t.
  */
@@ -162,120 +115,18 @@ int
 _pgstat64(const char *name, struct stat *buf)
 {
 	/*
-	 * We must use a handle so lstat() returns the information of the target
-	 * file.  To have a reliable test for ERROR_DELETE_PENDING, we use
-	 * NtQueryInformationFile from Windows 2000 or
-	 * GetFileInformationByHandleEx from Server 2008 / Vista.
+	 * Our open wrapper will report STATUS_DELETE_PENDING as ENOENT.  We
+	 * request FILE_FLAG_BACKUP_SEMANTICS so that we can open directories too,
+	 * for limited purposes.  We use the private handle-based version, so we
+	 * don't risk running out of fds.
 	 */
-	SECURITY_ATTRIBUTES sa;
 	HANDLE		hFile;
 	int			ret;
-#if _WIN32_WINNT < 0x0600
-	IO_STATUS_BLOCK ioStatus;
-	FILE_STANDARD_INFORMATION standardInfo;
-#else
-	FILE_STANDARD_INFO standardInfo;
-#endif
-
-	if (name == NULL || buf == NULL)
-	{
-		errno = EINVAL;
-		return -1;
-	}
 
-	/* fast not-exists check */
-	if (GetFileAttributes(name) == INVALID_FILE_ATTRIBUTES)
-	{
-		_dosmaperr(GetLastError());
-		return -1;
-	}
-
-	/* get a file handle as lightweight as we can */
-	sa.nLength = sizeof(SECURITY_ATTRIBUTES);
-	sa.bInheritHandle = TRUE;
-	sa.lpSecurityDescriptor = NULL;
-	hFile = CreateFile(name,
-					   GENERIC_READ,
-					   (FILE_SHARE_READ | FILE_SHARE_WRITE | FILE_SHARE_DELETE),
-					   &sa,
-					   OPEN_EXISTING,
-					   (FILE_FLAG_NO_BUFFERING | FILE_FLAG_BACKUP_SEMANTICS |
-						FILE_FLAG_OVERLAPPED),
-					   NULL);
+	hFile = pgwin32_open_handle(name, O_RDONLY, true);
 	if (hFile == INVALID_HANDLE_VALUE)
-	{
-		DWORD		err = GetLastError();
-
-		CloseHandle(hFile);
-		_dosmaperr(err);
 		return -1;
-	}
-
-	memset(&standardInfo, 0, sizeof(standardInfo));
-
-#if _WIN32_WINNT < 0x0600
-	if (_NtQueryInformationFile == NULL)
-	{
-		/* First time through: load ntdll.dll and find NtQueryInformationFile */
-		LoadNtdll();
-		if (ntdll == NULL)
-		{
-			DWORD		err = GetLastError();
-
-			CloseHandle(hFile);
-			_dosmaperr(err);
-			return -1;
-		}
-
-		_NtQueryInformationFile = (PFN_NTQUERYINFORMATIONFILE) (pg_funcptr_t)
-			GetProcAddress(ntdll, "NtQueryInformationFile");
-		if (_NtQueryInformationFile == NULL)
-		{
-			DWORD		err = GetLastError();
 
-			CloseHandle(hFile);
-			_dosmaperr(err);
-			return -1;
-		}
-	}
-
-	if (!NT_SUCCESS(_NtQueryInformationFile(hFile, &ioStatus, &standardInfo,
-											sizeof(standardInfo),
-											FileStandardInformation)))
-	{
-		DWORD		err = GetLastError();
-
-		CloseHandle(hFile);
-		_dosmaperr(err);
-		return -1;
-	}
-#else
-	if (!GetFileInformationByHandleEx(hFile, FileStandardInfo, &standardInfo,
-									  sizeof(standardInfo)))
-	{
-		DWORD		err = GetLastError();
-
-		CloseHandle(hFile);
-		_dosmaperr(err);
-		return -1;
-	}
-#endif							/* _WIN32_WINNT < 0x0600 */
-
-	if (standardInfo.DeletePending)
-	{
-		/*
-		 * File has been deleted, but is not gone from the filesystem yet.
-		 * This can happen when some process with FILE_SHARE_DELETE has it
-		 * open, and it will be fully removed once that handle is closed.
-		 * Meanwhile, we can't open it, so indicate that the file just doesn't
-		 * exist.
-		 */
-		CloseHandle(hFile);
-		errno = ENOENT;
-		return -1;
-	}
-
-	/* At last we can invoke fileinfo_to_stat */
 	ret = fileinfo_to_stat(hFile, buf);
 
 	CloseHandle(hFile);
@@ -316,11 +167,6 @@ _pgfstat64(int fileno, struct stat *buf)
 		return 0;
 	}
 
-	/*
-	 * Since we already have a file handle there is no need to check for
-	 * ERROR_DELETE_PENDING.
-	 */
-
 	return fileinfo_to_stat(hFile, buf);
 }
 

src/tools/msvc/Mkvcbuild.pm

@@ -107,7 +107,8 @@ sub mkvcbuild
 	  pg_strong_random.c pgcheckdir.c pgmkdirp.c pgsleep.c pgstrcasecmp.c
 	  pqsignal.c mkdtemp.c qsort.c qsort_arg.c bsearch_arg.c quotes.c system.c
 	  strerror.c tar.c thread.c
-	  win32env.c win32error.c win32security.c win32setlocale.c win32stat.c);
+	  win32env.c win32error.c win32ntdll.c
+	  win32security.c win32setlocale.c win32stat.c);
 
 	push(@pgportfiles, 'strtof.c') if ($vsVersion < '14.00');