diff --git a/src/backend/executor/nodeHash.c b/src/backend/executor/nodeHash.c
index 214cc7f06964dbf8c3af866a5b27949bdb37663e..c6295acf7237a23947202302a1298ce135065f2c 100644
--- a/src/backend/executor/nodeHash.c
+++ b/src/backend/executor/nodeHash.c
@@ -6,7 +6,7 @@
  * Copyright (c) 1994, Regents of the University of California
  *
  *
- *	$Id: nodeHash.c,v 1.38 1999/07/17 20:16:58 momjian Exp $
+ *	$Id: nodeHash.c,v 1.39 1999/10/13 15:02:25 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -75,12 +75,7 @@ ExecHash(Hash *node)
 		 * ----------------
 		 */
 		for (i = 0; i < nbatch; i++)
-		{
-			File		tfile = OpenTemporaryFile();
-
-			Assert(tfile >= 0);
-			hashtable->innerBatchFile[i] = BufFileCreate(tfile);
-		}
+			hashtable->innerBatchFile[i] = BufFileCreateTemp();
 	}
 
 	/* ----------------
diff --git a/src/backend/executor/nodeHashjoin.c b/src/backend/executor/nodeHashjoin.c
index 439b8634cda1eb81a772129394634903f84cd729..ffda9723182a7b04594b223f642cd2c497f3911f 100644
--- a/src/backend/executor/nodeHashjoin.c
+++ b/src/backend/executor/nodeHashjoin.c
@@ -7,7 +7,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/executor/nodeHashjoin.c,v 1.26 1999/07/17 20:16:58 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/executor/nodeHashjoin.c,v 1.27 1999/10/13 15:02:25 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -129,12 +129,7 @@ ExecHashJoin(HashJoin *node)
 		 * ----------------
 		 */
 		for (i = 0; i < hashtable->nbatch; i++)
-		{
-			File		tfile = OpenTemporaryFile();
-
-			Assert(tfile >= 0);
-			hashtable->outerBatchFile[i] = BufFileCreate(tfile);
-		}
+			hashtable->outerBatchFile[i] = BufFileCreateTemp();
 	}
 	else if (hashtable == NULL)
 		return NULL;
@@ -551,13 +546,12 @@ ExecHashJoinNewBatch(HashJoinState *hjstate)
 	 * Rewind inner and outer batch files for this batch, so that we can
 	 * start reading them.
 	 */
-	if (BufFileSeek(hashtable->outerBatchFile[newbatch - 1], 0L,
-					SEEK_SET) != 0L)
+	if (BufFileSeek(hashtable->outerBatchFile[newbatch - 1], 0, 0L, SEEK_SET))
 		elog(ERROR, "Failed to rewind hash temp file");
 
 	innerFile = hashtable->innerBatchFile[newbatch - 1];
 
-	if (BufFileSeek(innerFile, 0L, SEEK_SET) != 0L)
+	if (BufFileSeek(innerFile, 0, 0L, SEEK_SET))
 		elog(ERROR, "Failed to rewind hash temp file");
 
 	/*
diff --git a/src/backend/storage/file/Makefile b/src/backend/storage/file/Makefile
index 9f321ff57fb6f8d0d14372fba92af17532436a5b..766a0c1d1c39763c2e15379a8779605fd88b654b 100644
--- a/src/backend/storage/file/Makefile
+++ b/src/backend/storage/file/Makefile
@@ -4,7 +4,7 @@
 #    Makefile for storage/file
 #
 # IDENTIFICATION
-#    $Header: /cvsroot/pgsql/src/backend/storage/file/Makefile,v 1.5 1998/04/06 00:25:05 momjian Exp $
+#    $Header: /cvsroot/pgsql/src/backend/storage/file/Makefile,v 1.6 1999/10/13 15:02:29 tgl Exp $
 #
 #-------------------------------------------------------------------------
 
@@ -13,7 +13,7 @@ include ../../../Makefile.global
 
 CFLAGS += -I../..
 
-OBJS = fd.o
+OBJS = fd.o buffile.o
 
 all: SUBSYS.o
 
diff --git a/src/backend/storage/file/buffile.c b/src/backend/storage/file/buffile.c
new file mode 100644
index 0000000000000000000000000000000000000000..cd7da900b4f08afcdcf6b71c08325dceaad6a203
--- /dev/null
+++ b/src/backend/storage/file/buffile.c
@@ -0,0 +1,556 @@
+/*-------------------------------------------------------------------------
+ *
+ * buffile.c
+ *	  Management of large buffered files, primarily temporary files.
+ *
+ * Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/file/buffile.c,v 1.1 1999/10/13 15:02:29 tgl Exp $
+ *
+ * NOTES:
+ *
+ * BufFiles provide a very incomplete emulation of stdio atop virtual Files
+ * (as managed by fd.c).  Currently, we only support the buffered-I/O
+ * aspect of stdio: a read or write of the low-level File occurs only
+ * when the buffer is filled or emptied.  This is an even bigger win
+ * for virtual Files than for ordinary kernel files, since reducing the
+ * frequency with which a virtual File is touched reduces "thrashing"
+ * of opening/closing file descriptors.
+ *
+ * Note that BufFile structs are allocated with palloc(), and therefore
+ * will go away automatically at transaction end.  If the underlying
+ * virtual File is made with OpenTemporaryFile, then all resources for
+ * the file are certain to be cleaned up even if processing is aborted
+ * by elog(ERROR).  To avoid confusion, the caller should take care that
+ * all calls for a single BufFile are made in the same palloc context.
+ *
+ * BufFile also supports temporary files that exceed the OS file size limit
+ * (by opening multiple fd.c temporary files).  This is an essential feature
+ * for sorts and hashjoins on large amounts of data.  It is possible to have
+ * more than one BufFile reading/writing the same temp file, although the
+ * caller is responsible for avoiding ill effects from buffer overlap when
+ * this is done.
+ *-------------------------------------------------------------------------
+ */
+
+#include <errno.h>
+
+#include "postgres.h"
+
+#include "storage/buffile.h"
+
+/*
+ * The maximum safe file size is presumed to be RELSEG_SIZE * BLCKSZ.
+ * Note we adhere to this limit whether or not LET_OS_MANAGE_FILESIZE
+ * is defined, although md.c ignores it when that symbol is defined.
+ */
+#define MAX_PHYSICAL_FILESIZE  (RELSEG_SIZE * BLCKSZ)
+
+/*
+ * To handle multiple BufFiles on a single logical temp file, we use this
+ * data structure representing a logical file (which can be made up of
+ * multiple physical files to get around the OS file size limit).
+ */
+typedef struct LogicalFile
+{
+	int			refCount;		/* number of BufFiles using me */
+	bool		isTemp;			/* can only add files if this is TRUE */
+	int			numFiles;		/* number of physical files in set */
+	/* all files except the last have length exactly MAX_PHYSICAL_FILESIZE */
+
+	File	   *files;			/* palloc'd array with numFiles entries */
+	long	   *offsets;		/* palloc'd array with numFiles entries */
+	/* offsets[i] is the current seek position of files[i].  We use this
+	 * to avoid making redundant FileSeek calls.
+	 */
+} LogicalFile;
+
+/*
+ * A single file buffer looks like this.
+ */
+struct BufFile
+{
+	LogicalFile *logFile;		/* the underlying LogicalFile */
+	bool		dirty;			/* does buffer need to be written? */
+	/*
+	 * "current pos" is position of start of buffer within LogicalFile.
+	 * Position as seen by user of BufFile is (curFile, curOffset + pos).
+	 */
+	int			curFile;		/* file index (0..n) part of current pos */
+	int			curOffset;		/* offset part of current pos */
+	int			pos;			/* next read/write position in buffer */
+	int			nbytes;			/* total # of valid bytes in buffer */
+	char		buffer[BLCKSZ];
+};
+
+static LogicalFile *makeLogicalFile(File firstfile);
+static void extendLogicalFile(LogicalFile *file);
+static void deleteLogicalFile(LogicalFile *file);
+static void BufFileLoadBuffer(BufFile *file);
+static void BufFileDumpBuffer(BufFile *file);
+static int	BufFileFlush(BufFile *file);
+
+
+/*
+ * Create a LogicalFile with one component file and refcount 1.
+ * NOTE: caller must set isTemp true if appropriate.
+ */
+static LogicalFile *
+makeLogicalFile(File firstfile)
+{
+	LogicalFile *file = (LogicalFile *) palloc(sizeof(LogicalFile));
+
+	file->refCount = 1;
+	file->isTemp = false;
+	file->numFiles = 1;
+	file->files = (File *) palloc(sizeof(File));
+	file->files[0] = firstfile;
+	file->offsets = (long *) palloc(sizeof(long));
+	file->offsets[0] = 0L;
+
+	return file;
+}
+
+/*
+ * Add another component temp file.
+ */
+static void
+extendLogicalFile(LogicalFile *file)
+{
+	File		pfile;
+
+	Assert(file->isTemp);
+	pfile = OpenTemporaryFile();
+	Assert(pfile >= 0);
+
+	file->files = (File *) repalloc(file->files,
+									(file->numFiles+1) * sizeof(File));
+	file->offsets = (long *) repalloc(file->offsets,
+									  (file->numFiles+1) * sizeof(long));
+	file->files[file->numFiles] = pfile;
+	file->offsets[file->numFiles] = 0L;
+	file->numFiles++;
+}
+
+/*
+ * Close and delete a LogicalFile when its refCount has gone to zero.
+ */
+static void
+deleteLogicalFile(LogicalFile *file)
+{
+	int i;
+
+	for (i = 0; i < file->numFiles; i++)
+		FileClose(file->files[i]);
+	pfree(file->files);
+	pfree(file->offsets);
+	pfree(file);
+}
+
+/*
+ * Create a BufFile for a new temporary file (which will expand to become
+ * multiple temporary files if more than MAX_PHYSICAL_FILESIZE bytes are
+ * written to it).
+ */
+BufFile *
+BufFileCreateTemp(void)
+{
+	BufFile    *bfile = (BufFile *) palloc(sizeof(BufFile));
+	File		pfile;
+	LogicalFile *lfile;
+
+	pfile = OpenTemporaryFile();
+	Assert(pfile >= 0);
+
+	lfile = makeLogicalFile(pfile);
+	lfile->isTemp = true;
+
+	bfile->logFile = lfile;
+	bfile->dirty = false;
+	bfile->curFile = 0;
+	bfile->curOffset = 0L;
+	bfile->pos = 0;
+	bfile->nbytes = 0;
+
+	return bfile;
+}
+
+/*
+ * Create a BufFile and attach it to an already-opened virtual File.
+ *
+ * This is comparable to fdopen() in stdio.  This is the only way at present
+ * to attach a BufFile to a non-temporary file.  Note that BufFiles created
+ * in this way CANNOT be expanded into multiple files.
+ */
+BufFile *
+BufFileCreate(File file)
+{
+	BufFile    *bfile = (BufFile *) palloc(sizeof(BufFile));
+	LogicalFile *lfile;
+
+	lfile = makeLogicalFile(file);
+
+	bfile->logFile = lfile;
+	bfile->dirty = false;
+	bfile->curFile = 0;
+	bfile->curOffset = 0L;
+	bfile->pos = 0;
+	bfile->nbytes = 0;
+
+	return bfile;
+}
+
+/*
+ * Create an additional BufFile accessing the same underlying file as an
+ * existing BufFile.  This is useful for having multiple read/write access
+ * positions in a single temporary file.  Note the caller is responsible
+ * for avoiding trouble due to overlapping buffer positions!  (Caller may
+ * assume that buffer size is BLCKSZ...)
+ */
+BufFile *
+BufFileReaccess(BufFile *file)
+{
+	BufFile    *bfile = (BufFile *) palloc(sizeof(BufFile));
+
+	bfile->logFile = file->logFile;
+	bfile->logFile->refCount++;
+	bfile->dirty = false;
+	bfile->curFile = 0;
+	bfile->curOffset = 0L;
+	bfile->pos = 0;
+	bfile->nbytes = 0;
+
+	return bfile;
+}
+
+/*
+ * Close a BufFile
+ *
+ * Like fclose(), this also implicitly FileCloses the underlying File.
+ */
+void
+BufFileClose(BufFile *file)
+{
+	/* flush any unwritten data */
+	BufFileFlush(file);
+	/* close the underlying (with delete if it's a temp file) */
+	if (--(file->logFile->refCount) <= 0)
+		deleteLogicalFile(file->logFile);
+	/* release the buffer space */
+	pfree(file);
+}
+
+/* BufFileLoadBuffer
+ *
+ * Load some data into buffer, if possible, starting from curOffset.
+ * At call, must have dirty = false, pos and nbytes = 0.
+ * On exit, nbytes is number of bytes loaded.
+ */
+static void
+BufFileLoadBuffer(BufFile *file)
+{
+	LogicalFile *lfile = file->logFile;
+	File	thisfile;
+
+	/*
+	 * Advance to next component file if necessary and possible.
+	 *
+	 * This path can only be taken if there is more than one component,
+	 * so it won't interfere with reading a non-temp file that is over
+	 * MAX_PHYSICAL_FILESIZE.
+	 */
+	if (file->curOffset >= MAX_PHYSICAL_FILESIZE &&
+		file->curFile+1 < lfile->numFiles)
+	{
+		file->curFile++;
+		file->curOffset = 0L;
+	}
+	thisfile = lfile->files[file->curFile];
+	/*
+	 * May need to reposition physical file, if more than one BufFile
+	 * is using it.
+	 */
+	if (file->curOffset != lfile->offsets[file->curFile])
+	{
+		if (FileSeek(thisfile, file->curOffset, SEEK_SET) != file->curOffset)
+			return;				/* seek failed, read nothing */
+		lfile->offsets[file->curFile] = file->curOffset;
+	}
+	file->nbytes = FileRead(thisfile, file->buffer, sizeof(file->buffer));
+	if (file->nbytes < 0)
+		file->nbytes = 0;
+	lfile->offsets[file->curFile] += file->nbytes;
+	/* we choose not to advance curOffset here */
+}
+
+/* BufFileDumpBuffer
+ *
+ * Dump buffer contents starting at curOffset.
+ * At call, should have dirty = true, nbytes > 0.
+ * On exit, dirty is cleared if successful write, and curOffset is advanced.
+ */
+static void
+BufFileDumpBuffer(BufFile *file)
+{
+	LogicalFile *lfile = file->logFile;
+	int			wpos = 0;
+	int			bytestowrite;
+	File		thisfile;
+
+	/*
+	 * Unlike BufFileLoadBuffer, we must dump the whole buffer even if
+	 * it crosses a component-file boundary; so we need a loop.
+	 */
+	while (wpos < file->nbytes)
+	{
+		/*
+		 * Advance to next component file if necessary and possible.
+		 */
+		if (file->curOffset >= MAX_PHYSICAL_FILESIZE && lfile->isTemp)
+		{
+			while (file->curFile+1 >= lfile->numFiles)
+				extendLogicalFile(lfile);
+			file->curFile++;
+			file->curOffset = 0L;
+		}
+		/*
+		 * Enforce per-file size limit only for temp files, else just try
+		 * to write as much as asked...
+		 */
+		bytestowrite = file->nbytes - wpos;
+		if (lfile->isTemp)
+		{
+			long	availbytes = MAX_PHYSICAL_FILESIZE - file->curOffset;
+
+			if ((long) bytestowrite > availbytes)
+				bytestowrite = (int) availbytes;
+		}
+		thisfile = lfile->files[file->curFile];
+		/*
+		 * May need to reposition physical file, if more than one BufFile
+		 * is using it.
+		 */
+		if (file->curOffset != lfile->offsets[file->curFile])
+		{
+			if (FileSeek(thisfile, file->curOffset, SEEK_SET) != file->curOffset)
+				return;			/* seek failed, give up */
+			lfile->offsets[file->curFile] = file->curOffset;
+		}
+		bytestowrite = FileWrite(thisfile, file->buffer, bytestowrite);
+		if (bytestowrite <= 0)
+			return;				/* failed to write */
+		lfile->offsets[file->curFile] += bytestowrite;
+		file->curOffset += bytestowrite;
+		wpos += bytestowrite;
+	}
+	file->dirty = false;
+	/*
+	 * At this point, curOffset has been advanced to the end of the buffer,
+	 * ie, its original value + nbytes.  We need to make it point to the
+	 * logical file position, ie, original value + pos, in case that is less
+	 * (as could happen due to a small backwards seek in a dirty buffer!)
+	 */
+	file->curOffset -= (file->nbytes - file->pos);
+	if (file->curOffset < 0)	/* handle possible segment crossing */
+	{
+		file->curFile--;
+		Assert(file->curFile >= 0);
+		file->curOffset += MAX_PHYSICAL_FILESIZE;
+	}
+	/* Now we can set the buffer empty without changing the logical position */
+	file->pos = 0;
+	file->nbytes = 0;
+}
+
+/* BufFileRead
+ *
+ * Like fread() except we assume 1-byte element size.
+ */
+size_t
+BufFileRead(BufFile *file, void *ptr, size_t size)
+{
+	size_t		nread = 0;
+	size_t		nthistime;
+
+	if (file->dirty)
+	{
+		if (BufFileFlush(file) != 0)
+			return 0;			/* could not flush... */
+		Assert(! file->dirty);
+	}
+
+	while (size > 0)
+	{
+		if (file->pos >= file->nbytes)
+		{
+			/* Try to load more data into buffer. */
+			file->curOffset += file->pos;
+			file->pos = 0;
+			file->nbytes = 0;
+			BufFileLoadBuffer(file);
+			if (file->nbytes <= 0)
+				break;			/* no more data available */
+		}
+
+		nthistime = file->nbytes - file->pos;
+		if (nthistime > size)
+			nthistime = size;
+		Assert(nthistime > 0);
+
+		memcpy(ptr, file->buffer + file->pos, nthistime);
+
+		file->pos += nthistime;
+		ptr = (void *) ((char *) ptr + nthistime);
+		size -= nthistime;
+		nread += nthistime;
+	}
+
+	return nread;
+}
+
+/* BufFileWrite
+ *
+ * Like fwrite() except we assume 1-byte element size.
+ */
+size_t
+BufFileWrite(BufFile *file, void *ptr, size_t size)
+{
+	size_t		nwritten = 0;
+	size_t		nthistime;
+
+	while (size > 0)
+	{
+		if (file->pos >= BLCKSZ)
+		{
+			/* Buffer full, dump it out */
+			if (file->dirty)
+			{
+				BufFileDumpBuffer(file);
+				if (file->dirty)
+					break;		/* I/O error */
+			}
+			else
+			{
+				/* Hmm, went directly from reading to writing? */
+				file->curOffset += file->pos;
+				file->pos = 0;
+				file->nbytes = 0;
+			}
+		}
+
+		nthistime = BLCKSZ - file->pos;
+		if (nthistime > size)
+			nthistime = size;
+		Assert(nthistime > 0);
+
+		memcpy(file->buffer + file->pos, ptr, nthistime);
+
+		file->dirty = true;
+		file->pos += nthistime;
+		if (file->nbytes < file->pos)
+			file->nbytes = file->pos;
+		ptr = (void *) ((char *) ptr + nthistime);
+		size -= nthistime;
+		nwritten += nthistime;
+	}
+
+	return nwritten;
+}
+
+/* BufFileFlush
+ *
+ * Like fflush()
+ */
+static int
+BufFileFlush(BufFile *file)
+{
+	if (file->dirty)
+	{
+		BufFileDumpBuffer(file);
+		if (file->dirty)
+			return EOF;
+	}
+
+	return 0;
+}
+
+/* BufFileSeek
+ *
+ * Like fseek().  Result is 0 if OK, EOF if not.
+ */
+int
+BufFileSeek(BufFile *file, int fileno, long offset, int whence)
+{
+	int newFile;
+	long newOffset;
+	switch (whence)
+	{
+		case SEEK_SET:
+			if (fileno < 0 || fileno >= file->logFile->numFiles ||
+				offset < 0)
+				return EOF;
+			newFile = fileno;
+			newOffset = offset;
+			break;
+		case SEEK_CUR:
+			/*
+			 * Relative seek considers only the signed offset, ignoring fileno.
+			 * Note that large offsets (> 1 gig) risk overflow.
+			 */
+			newFile = file->curFile;
+			newOffset = (file->curOffset + file->pos) + offset;
+			break;
+#ifdef NOT_USED
+		case SEEK_END:
+			/* could be implemented, not needed currently */
+			break;
+#endif
+		default:
+			elog(ERROR, "BufFileSeek: invalid whence: %d", whence);
+			return EOF;
+	}
+	while (newOffset < 0)
+	{
+		if (--newFile < 0)
+			return EOF;
+		newOffset += MAX_PHYSICAL_FILESIZE;
+	}
+	if (file->logFile->isTemp)
+	{
+		while (newOffset > MAX_PHYSICAL_FILESIZE)
+		{
+			if (++newFile >= file->logFile->numFiles)
+				return EOF;
+			newOffset -= MAX_PHYSICAL_FILESIZE;
+		}
+	}
+	if (newFile == file->curFile &&
+		newOffset >= file->curOffset &&
+		newOffset <= file->curOffset + file->nbytes)
+	{
+		/*
+		 * Seek is to a point within existing buffer; we can just adjust
+		 * pos-within-buffer, without flushing buffer.  Note this is OK
+		 * whether reading or writing, but buffer remains dirty if we
+		 * were writing.
+		 */
+		file->pos = (int) (newOffset - file->curOffset);
+		return 0;
+	}
+	/* Otherwise, must reposition buffer, so flush any dirty data */
+	if (BufFileFlush(file) != 0)
+		return EOF;
+	file->curFile = newFile;
+	file->curOffset = newOffset;
+	file->pos = 0;
+	file->nbytes = 0;
+	return 0;
+}
+
+extern void
+BufFileTell(BufFile *file, int *fileno, long *offset)
+{
+	*fileno = file->curFile;
+	*offset = file->curOffset + file->pos;
+}
diff --git a/src/backend/storage/file/fd.c b/src/backend/storage/file/fd.c
index 4cdb638819e81b2be731b4d61b1da56b0dc241d8..2fce82ecfd6aaaff658ffd89392fdcaaa9af3234 100644
--- a/src/backend/storage/file/fd.c
+++ b/src/backend/storage/file/fd.c
@@ -6,7 +6,7 @@
  * Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $Id: fd.c,v 1.48 1999/09/27 15:47:49 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/file/fd.c,v 1.49 1999/10/13 15:02:29 tgl Exp $
  *
  * NOTES:
  *
@@ -49,7 +49,6 @@
 #include "miscadmin.h"
 #include "storage/fd.h"
 
-bool	ReleaseDataFile(void);
 /*
  * Problem: Postgres does a system(ld...) to do dynamic loading.
  * This will open several extra files in addition to those used by
@@ -188,7 +187,6 @@ static int	FileAccess(File file);
 static File fileNameOpenFile(FileName fileName, int fileFlags, int fileMode);
 static char *filepath(char *filename);
 static long pg_nofile(void);
-static int	BufFileFlush(BufFile *file);
 
 /*
  * pg_fsync --- same as fsync except does nothing if -F switch was given
@@ -411,6 +409,9 @@ ReleaseLruFile()
 	LruDelete(VfdCache[0].lruMoreRecently);
 }
 
+/*
+ * Force one kernel file descriptor to be released (temporarily).
+ */
 bool
 ReleaseDataFile()
 {
@@ -506,8 +507,11 @@ FreeVfd(File file)
 
 /* filepath()
  * Convert given pathname to absolute.
- * (Is this actually necessary, considering that we should be cd'd
- * into the database directory??)
+ *
+ * (Generally, this isn't actually necessary, considering that we
+ * should be cd'd into the database directory.  Presently it is only
+ * necessary to do it in "bootstrap" mode.  Maybe we should change
+ * bootstrap mode to do the cd, and save a few cycles/bytes here.)
  */
 static char *
 filepath(char *filename)
@@ -851,7 +855,7 @@ FileTell(File file)
 #endif
 
 int
-FileTruncate(File file, int offset)
+FileTruncate(File file, long offset)
 {
 	int			returnCode;
 
@@ -862,7 +866,7 @@ FileTruncate(File file, int offset)
 
 	FileSync(file);
 	FileAccess(file);
-	returnCode = ftruncate(VfdCache[file].fd, offset);
+	returnCode = ftruncate(VfdCache[file].fd, (size_t) offset);
 	return returnCode;
 }
 
@@ -890,18 +894,6 @@ FileSync(File file)
 	return returnCode;
 }
 
-int
-FileNameUnlink(char *filename)
-{
-	int			retval;
-	char	   *fname;
-
-	fname = filepath(filename);
-	retval = unlink(fname);
-	pfree(fname);
-	return retval;
-}
-
 /*
  * Routines that want to use stdio (ie, FILE*) should use AllocateFile
  * rather than plain fopen().  This lets fd.c deal with freeing FDs if
@@ -1023,186 +1015,3 @@ AtEOXact_Files(void)
 	 */
 	tempFileCounter = 0;
 }
-
-
-/*
- * Operations on BufFiles --- a very incomplete emulation of stdio
- * atop virtual Files.	Currently, we only support the buffered-I/O
- * aspect of stdio: a read or write of the low-level File occurs only
- * when the buffer is filled or emptied.  This is an even bigger win
- * for virtual Files than ordinary kernel files, since reducing the
- * frequency with which a virtual File is touched reduces "thrashing"
- * of opening/closing file descriptors.
- *
- * Note that BufFile structs are allocated with palloc(), and therefore
- * will go away automatically at transaction end.  If the underlying
- * virtual File is made with OpenTemporaryFile, then all resources for
- * the file are certain to be cleaned up even if processing is aborted
- * by elog(ERROR).
- */
-
-struct BufFile
-{
-	File		file;			/* the underlying virtual File */
-	bool		dirty;			/* does buffer need to be written? */
-	int			pos;			/* next read/write position in buffer */
-	int			nbytes;			/* total # of valid bytes in buffer */
-	char		buffer[BLCKSZ];
-};
-
-
-/*
- * Create a BufFile and attach it to an (already opened) virtual File.
- *
- * This is comparable to fdopen() in stdio.
- */
-BufFile    *
-BufFileCreate(File file)
-{
-	BufFile    *bfile = (BufFile *) palloc(sizeof(BufFile));
-
-	bfile->file = file;
-	bfile->dirty = false;
-	bfile->pos = 0;
-	bfile->nbytes = 0;
-
-	return bfile;
-}
-
-/*
- * Close a BufFile
- *
- * Like fclose(), this also implicitly FileCloses the underlying File.
- */
-void
-BufFileClose(BufFile *file)
-{
-	/* flush any unwritten data */
-	BufFileFlush(file);
-	/* close the underlying (with delete if it's a temp file) */
-	FileClose(file->file);
-	/* release the buffer space */
-	pfree(file);
-}
-
-/* BufFileRead
- *
- * Like fread() except we assume 1-byte element size.
- */
-size_t
-BufFileRead(BufFile *file, void *ptr, size_t size)
-{
-	size_t		nread = 0;
-	size_t		nthistime;
-
-	if (file->dirty)
-	{
-		elog(NOTICE, "BufFileRead: should have flushed after writing");
-		BufFileFlush(file);
-	}
-
-	while (size > 0)
-	{
-		if (file->pos >= file->nbytes)
-		{
-			/* Try to load more data into buffer */
-			file->pos = 0;
-			file->nbytes = FileRead(file->file, file->buffer,
-									sizeof(file->buffer));
-			if (file->nbytes < 0)
-				file->nbytes = 0;
-			if (file->nbytes <= 0)
-				break;			/* no more data available */
-		}
-
-		nthistime = file->nbytes - file->pos;
-		if (nthistime > size)
-			nthistime = size;
-		Assert(nthistime > 0);
-
-		memcpy(ptr, file->buffer + file->pos, nthistime);
-
-		file->pos += nthistime;
-		ptr = (void *) ((char *) ptr + nthistime);
-		size -= nthistime;
-		nread += nthistime;
-	}
-
-	return nread;
-}
-
-/* BufFileWrite
- *
- * Like fwrite() except we assume 1-byte element size.
- */
-size_t
-BufFileWrite(BufFile *file, void *ptr, size_t size)
-{
-	size_t		nwritten = 0;
-	size_t		nthistime;
-
-	while (size > 0)
-	{
-		if (file->pos >= BLCKSZ)
-		{
-			/* Buffer full, dump it out */
-			if (file->dirty)
-			{
-				if (FileWrite(file->file, file->buffer, file->nbytes) < 0)
-					break;		/* I/O error */
-				file->dirty = false;
-			}
-			file->pos = 0;
-			file->nbytes = 0;
-		}
-
-		nthistime = BLCKSZ - file->pos;
-		if (nthistime > size)
-			nthistime = size;
-		Assert(nthistime > 0);
-
-		memcpy(file->buffer + file->pos, ptr, nthistime);
-
-		file->dirty = true;
-		file->pos += nthistime;
-		if (file->nbytes < file->pos)
-			file->nbytes = file->pos;
-		ptr = (void *) ((char *) ptr + nthistime);
-		size -= nthistime;
-		nwritten += nthistime;
-	}
-
-	return nwritten;
-}
-
-/* BufFileFlush
- *
- * Like fflush()
- */
-static int
-BufFileFlush(BufFile *file)
-{
-	if (file->dirty)
-	{
-		if (FileWrite(file->file, file->buffer, file->nbytes) < 0)
-			return EOF;
-		file->dirty = false;
-	}
-
-	return 0;
-}
-
-/* BufFileSeek
- *
- * Like fseek(), or really more like lseek() since the return value is
- * the new file offset (or -1 in case of error).
- */
-long
-BufFileSeek(BufFile *file, long offset, int whence)
-{
-	if (BufFileFlush(file) < 0)
-		return -1L;
-	file->pos = 0;
-	file->nbytes = 0;
-	return FileSeek(file->file, offset, whence);
-}
diff --git a/src/backend/storage/large_object/inv_api.c b/src/backend/storage/large_object/inv_api.c
index f59e99d7fbe90fd5b0156749b01ef84c7c83a053..555768a0f3e51439276b38b17d0bc5d62fb48343 100644
--- a/src/backend/storage/large_object/inv_api.c
+++ b/src/backend/storage/large_object/inv_api.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/large_object/inv_api.c,v 1.59 1999/09/18 19:07:32 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/large_object/inv_api.c,v 1.60 1999/10/13 15:02:25 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -300,6 +300,21 @@ inv_destroy(Oid lobjId)
  *		end of relations.  Once clustering works, we should fix this.
  */
 #ifdef NOT_USED
+
+struct pgstat
+{								/* just the fields we need from stat
+								 * structure */
+	int			st_ino;
+	int			st_mode;
+	unsigned int st_size;
+	unsigned int st_sizehigh;	/* high order bits */
+/* 2^64 == 1.8 x 10^20 bytes */
+	int			st_uid;
+	int			st_atime_s;		/* just the seconds */
+	int			st_mtime_s;		/* since SysV and the new BSD both have */
+	int			st_ctime_s;		/* usec fields.. */
+};
+
 int
 inv_stat(LargeObjectDesc *obj_desc, struct pgstat * stbuf)
 {
diff --git a/src/backend/utils/sort/psort.c b/src/backend/utils/sort/psort.c
index b1ac25280756552f089488c2d5778cc475bb45ff..14db10c119837607cdce1bfa043faf793697bbb9 100644
--- a/src/backend/utils/sort/psort.c
+++ b/src/backend/utils/sort/psort.c
@@ -4,7 +4,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- *	  $Id: psort.c,v 1.56 1999/07/17 20:18:16 momjian Exp $
+ *	  $Id: psort.c,v 1.57 1999/10/13 15:02:31 tgl Exp $
  *
  * NOTES
  *		Sorts the first relation into the second relation.
@@ -142,7 +142,8 @@ psort_begin(Sort *node, int nkeys, ScanKey key)
 		PS(node)->psort_grab_file = mergeruns(node);
 
 	PS(node)->psort_current = 0;
-	PS(node)->psort_saved = 0;
+	PS(node)->psort_saved_fileno = 0;
+	PS(node)->psort_saved = 0L;
 
 	return true;
 }
@@ -227,7 +228,7 @@ inittapes(Sort *node)
 
 #define SETTUPLEN(TUP, LEN)		((TUP)->t_len = (LEN) - HEAPTUPLESIZE)
 
-#define rewind(FP)		BufFileSeek(FP, 0L, SEEK_SET)
+#define rewind(FP)		BufFileSeek(FP, 0, 0L, SEEK_SET)
 
  /*
   * USEMEM			- record use of memory FREEMEM		   - record
@@ -764,9 +765,6 @@ psort_grabtuple(Sort *node, bool *should_free)
 				tup = ALLOCTUP(tuplen);
 				SETTUPLEN(tup, tuplen);
 				GETTUP(node, tup, tuplen, PS(node)->psort_grab_file);
-
-				/* Update current merged sort file position */
-				PS(node)->psort_current += tuplen + sizeof(tlendummy);
 				return tup;
 			}
 			else
@@ -775,70 +773,67 @@ psort_grabtuple(Sort *node, bool *should_free)
 				return NULL;
 			}
 		}
-		/* Backward */
-		if (PS(node)->psort_current <= sizeof(tlendummy))
-			return NULL;
-
-		/*
+		/* Backward.
+		 *
 		 * if all tuples are fetched already then we return last tuple,
 		 * else - tuple before last returned.
 		 */
 		if (PS(node)->all_fetched)
 		{
-
 			/*
-			 * psort_current is pointing to the zero tuplen at the end of
-			 * file
+			 * Assume seek position is pointing just past the zero tuplen
+			 * at the end of file; back up and fetch last tuple's ending
+			 * length word.  If seek fails we must have a completely empty
+			 * file.
 			 */
-			BufFileSeek(PS(node)->psort_grab_file,
-				  PS(node)->psort_current - sizeof(tlendummy), SEEK_SET);
+			if (BufFileSeek(PS(node)->psort_grab_file, 0,
+							- (long) (2 * sizeof(tlendummy)), SEEK_CUR))
+				return NULL;
 			GETLEN(tuplen, PS(node)->psort_grab_file);
-			if (PS(node)->psort_current < tuplen)
-				elog(ERROR, "psort_grabtuple: too big last tuple len in backward scan");
 			PS(node)->all_fetched = false;
 		}
 		else
 		{
-			/* move to position of end tlen of prev tuple */
-			PS(node)->psort_current -= sizeof(tlendummy);
-			BufFileSeek(PS(node)->psort_grab_file,
-						PS(node)->psort_current, SEEK_SET);
-			GETLEN(tuplen, PS(node)->psort_grab_file);	/* get tlen of prev
-														 * tuple */
+			/*
+			 * Back up and fetch prev tuple's ending length word.
+			 * If seek fails, assume we are at start of file.
+			 */
+			if (BufFileSeek(PS(node)->psort_grab_file, 0,
+							- (long) sizeof(tlendummy), SEEK_CUR))
+				return NULL;
+			GETLEN(tuplen, PS(node)->psort_grab_file);
 			if (tuplen == 0)
 				elog(ERROR, "psort_grabtuple: tuplen is 0 in backward scan");
-			if (PS(node)->psort_current <= tuplen + sizeof(tlendummy))
-			{					/* prev tuple should be first one */
-				if (PS(node)->psort_current != tuplen)
-					elog(ERROR, "psort_grabtuple: first tuple expected in backward scan");
-				PS(node)->psort_current = 0;
-				BufFileSeek(PS(node)->psort_grab_file,
-							PS(node)->psort_current, SEEK_SET);
-				return NULL;
-			}
-
 			/*
-			 * Get position of prev tuple. This tuple becomes current
-			 * tuple now and we have to return previous one.
+			 * Back up to get ending length word of tuple before it.
 			 */
-			PS(node)->psort_current -= tuplen;
-			/* move to position of end tlen of prev tuple */
-			BufFileSeek(PS(node)->psort_grab_file,
-				  PS(node)->psort_current - sizeof(tlendummy), SEEK_SET);
+			if (BufFileSeek(PS(node)->psort_grab_file, 0,
+							- (long) (tuplen + 2*sizeof(tlendummy)), SEEK_CUR))
+			{
+				/* If fail, presumably the prev tuple is the first in the file.
+				 * Back up so that it becomes next to read in forward direction
+				 * (not obviously right, but that is what in-memory case does)
+				 */
+				if (BufFileSeek(PS(node)->psort_grab_file, 0,
+								- (long) (tuplen + sizeof(tlendummy)), SEEK_CUR))
+					elog(ERROR, "psort_grabtuple: too big last tuple len in backward scan");
+				return NULL;
+			}
 			GETLEN(tuplen, PS(node)->psort_grab_file);
-			if (PS(node)->psort_current < tuplen + sizeof(tlendummy))
-				elog(ERROR, "psort_grabtuple: too big tuple len in backward scan");
 		}
 
 		/*
-		 * move to prev (or last) tuple start position + sizeof(t_len)
+		 * Now we have the length of the prior tuple, back up and read it.
+		 * Note: GETTUP expects we are positioned after the initial length
+		 * word of the tuple, so back up to that point.
 		 */
-		BufFileSeek(PS(node)->psort_grab_file,
-					PS(node)->psort_current - tuplen, SEEK_SET);
+		if (BufFileSeek(PS(node)->psort_grab_file, 0,
+						- (long) tuplen, SEEK_CUR))
+			elog(ERROR, "psort_grabtuple: too big tuple len in backward scan");
 		tup = ALLOCTUP(tuplen);
 		SETTUPLEN(tup, tuplen);
 		GETTUP(node, tup, tuplen, PS(node)->psort_grab_file);
-		return tup;				/* file position is equal to psort_current */
+		return tup;
 	}
 	else
 	{
@@ -875,6 +870,8 @@ psort_grabtuple(Sort *node, bool *should_free)
 
 /*
  *		psort_markpos	- saves current position in the merged sort file
+ *
+ * XXX I suspect these need to save & restore the all_fetched flag as well!
  */
 void
 psort_markpos(Sort *node)
@@ -882,7 +879,12 @@ psort_markpos(Sort *node)
 	Assert(node != (Sort *) NULL);
 	Assert(PS(node) != (Psortstate *) NULL);
 
-	PS(node)->psort_saved = PS(node)->psort_current;
+	if (PS(node)->using_tape_files == true)
+		BufFileTell(PS(node)->psort_grab_file,
+					& PS(node)->psort_saved_fileno,
+					& PS(node)->psort_saved);
+	else
+		PS(node)->psort_saved = PS(node)->psort_current;
 }
 
 /*
@@ -897,8 +899,11 @@ psort_restorepos(Sort *node)
 
 	if (PS(node)->using_tape_files == true)
 		BufFileSeek(PS(node)->psort_grab_file,
-					PS(node)->psort_saved, SEEK_SET);
-	PS(node)->psort_current = PS(node)->psort_saved;
+					PS(node)->psort_saved_fileno,
+					PS(node)->psort_saved,
+					SEEK_SET);
+	else
+		PS(node)->psort_current = PS(node)->psort_saved;
 }
 
 /*
@@ -952,7 +957,8 @@ psort_rescan(Sort *node)
 	{
 		PS(node)->all_fetched = false;
 		PS(node)->psort_current = 0;
-		PS(node)->psort_saved = 0;
+		PS(node)->psort_saved_fileno = 0;
+		PS(node)->psort_saved = 0L;
 		if (PS(node)->using_tape_files == true)
 			rewind(PS(node)->psort_grab_file);
 	}
@@ -973,11 +979,7 @@ psort_rescan(Sort *node)
 static BufFile *
 gettape()
 {
-	File		tfile;
-
-	tfile = OpenTemporaryFile();
-	Assert(tfile >= 0);
-	return BufFileCreate(tfile);
+	return BufFileCreateTemp();
 }
 
 /*
diff --git a/src/include/executor/hashjoin.h b/src/include/executor/hashjoin.h
index 724a7ddd313f3854a4e7ecc1440415b6a2efc2a3..f501cd0365417d9a5f2752d8fd862952df5954d4 100644
--- a/src/include/executor/hashjoin.h
+++ b/src/include/executor/hashjoin.h
@@ -6,7 +6,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: hashjoin.h,v 1.14 1999/07/15 15:21:08 momjian Exp $
+ * $Id: hashjoin.h,v 1.15 1999/10/13 15:02:26 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -14,6 +14,7 @@
 #define HASHJOIN_H
 
 #include "access/htup.h"
+#include "storage/buffile.h"
 
 /* ----------------------------------------------------------------
  *				hash-join hash table structures
diff --git a/src/include/executor/nodeHashjoin.h b/src/include/executor/nodeHashjoin.h
index 2061ac7bdcaf78310627ade614dc4b614b8100db..9d5390f79c690713d2f60569fb2416c2815545d4 100644
--- a/src/include/executor/nodeHashjoin.h
+++ b/src/include/executor/nodeHashjoin.h
@@ -6,7 +6,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: nodeHashjoin.h,v 1.15 1999/07/15 15:21:12 momjian Exp $
+ * $Id: nodeHashjoin.h,v 1.16 1999/10/13 15:02:26 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -14,6 +14,7 @@
 #define NODEHASHJOIN_H
 
 #include "nodes/plannodes.h"
+#include "storage/buffile.h"
 
 extern TupleTableSlot *ExecHashJoin(HashJoin *node);
 extern bool ExecInitHashJoin(HashJoin *node, EState *estate, Plan *parent);
diff --git a/src/include/storage/buffile.h b/src/include/storage/buffile.h
new file mode 100644
index 0000000000000000000000000000000000000000..2416d645cfe56e8b7f8eb040a4e49d0395b4cdcc
--- /dev/null
+++ b/src/include/storage/buffile.h
@@ -0,0 +1,47 @@
+/*-------------------------------------------------------------------------
+ *
+ * buffile.h
+ *	  Management of large buffered files, primarily temporary files.
+ *
+ * The BufFile routines provide a partial replacement for stdio atop
+ * virtual file descriptors managed by fd.c.  Currently they only support
+ * buffered access to a virtual file, without any of stdio's formatting
+ * features.  That's enough for immediate needs, but the set of facilities
+ * could be expanded if necessary.
+ *
+ * BufFile also supports working with temporary files that exceed the OS
+ * file size limit and/or the largest offset representable in an int.
+ * It might be better to split that out as a separately accessible module,
+ * but currently we have no need for oversize temp files without buffered
+ * access.
+ *
+ * Copyright (c) 1994, Regents of the University of California
+ *
+ * $Id: buffile.h,v 1.1 1999/10/13 15:02:32 tgl Exp $
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#ifndef BUFFILE_H
+#define BUFFILE_H
+
+#include "storage/fd.h"
+
+/* BufFile is an opaque type whose details are not known outside buffile.c. */
+
+typedef struct BufFile BufFile;
+
+/*
+ * prototypes for functions in buffile.c
+ */
+
+extern BufFile *BufFileCreateTemp(void);
+extern BufFile *BufFileCreate(File file);
+extern BufFile *BufFileReaccess(BufFile *file);
+extern void BufFileClose(BufFile *file);
+extern size_t BufFileRead(BufFile *file, void *ptr, size_t size);
+extern size_t BufFileWrite(BufFile *file, void *ptr, size_t size);
+extern int BufFileSeek(BufFile *file, int fileno, long offset, int whence);
+extern void BufFileTell(BufFile *file, int *fileno, long *offset);
+
+#endif	 /* BUFFILE_H */
diff --git a/src/include/storage/fd.h b/src/include/storage/fd.h
index beb93bf699c69d9f2817592cf1a1e6803cdb2770..42d1f46579759937fc665a11e119e7224bf3924a 100644
--- a/src/include/storage/fd.h
+++ b/src/include/storage/fd.h
@@ -6,10 +6,11 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: fd.h,v 1.17 1999/07/17 20:18:34 momjian Exp $
+ * $Id: fd.h,v 1.18 1999/10/13 15:02:32 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
+
 /*
  * calls:
  *
@@ -29,11 +30,6 @@
  * use FreeFile, not fclose, to close it.  AVOID using stdio for files
  * that you intend to hold open for any length of time, since there is
  * no way for them to share kernel file descriptors with other files.
- *
- * The BufFile routines provide a partial replacement for stdio.  Currently
- * they only support buffered access to a virtual file, without any of
- * stdio's formatting features.  That's enough for immediate needs, but
- * the set of facilities could be expanded if necessary.
  */
 #ifndef FD_H
 #define FD_H
@@ -46,25 +42,6 @@ typedef char *FileName;
 
 typedef int File;
 
-/* BufFile is an opaque type whose details are not known outside fd.c. */
-
-typedef struct BufFile BufFile;
-
-/* why is this here? fd.c doesn't want it ... */
-struct pgstat
-{								/* just the fields we need from stat
-								 * structure */
-	int			st_ino;
-	int			st_mode;
-	unsigned int st_size;
-	unsigned int st_sizehigh;	/* high order bits */
-/* 2^64 == 1.8 x 10^20 bytes */
-	int			st_uid;
-	int			st_atime_s;		/* just the seconds */
-	int			st_mtime_s;		/* since SysV and the new BSD both have */
-	int			st_ctime_s;		/* usec fields.. */
-};
-
 /*
  * prototypes for functions in fd.c
  */
@@ -78,24 +55,15 @@ extern void FileUnlink(File file);
 extern int	FileRead(File file, char *buffer, int amount);
 extern int	FileWrite(File file, char *buffer, int amount);
 extern long FileSeek(File file, long offset, int whence);
-extern int	FileTruncate(File file, int offset);
+extern int	FileTruncate(File file, long offset);
 extern int	FileSync(File file);
 
 /* Operations that allow use of regular stdio --- USE WITH CAUTION */
 extern FILE *AllocateFile(char *name, char *mode);
 extern void FreeFile(FILE *);
 
-/* Operations on BufFiles --- a very incomplete emulation of stdio
- * atop virtual Files...
- */
-extern BufFile *BufFileCreate(File file);
-extern void BufFileClose(BufFile *file);
-extern size_t BufFileRead(BufFile *file, void *ptr, size_t size);
-extern size_t BufFileWrite(BufFile *file, void *ptr, size_t size);
-extern long BufFileSeek(BufFile *file, long offset, int whence);
-
 /* Miscellaneous support routines */
-extern int	FileNameUnlink(char *filename);
+extern bool	ReleaseDataFile(void);
 extern void closeAllVfds(void);
 extern void AtEOXact_Files(void);
 extern int	pg_fsync(int fd);
diff --git a/src/include/utils/psort.h b/src/include/utils/psort.h
index 0deac024154413aa0b1799295da2e18adecf76a2..9a100bad0d8b6ae5a5c38620ebe60bb0fe39e8cb 100644
--- a/src/include/utils/psort.h
+++ b/src/include/utils/psort.h
@@ -6,7 +6,7 @@
  *
  * Copyright (c) 1994, Regents of the University of California
  *
- * $Id: psort.h,v 1.21 1999/07/16 17:07:39 momjian Exp $
+ * $Id: psort.h,v 1.22 1999/10/13 15:02:28 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -15,7 +15,7 @@
 
 #include "access/relscan.h"
 #include "nodes/plannodes.h"
-#include "storage/fd.h"
+#include "storage/buffile.h"
 #include "utils/lselect.h"
 
 #define MAXTAPES		7		/* See Knuth Fig. 70, p273 */
@@ -57,7 +57,8 @@ typedef struct Psortstate
 	struct leftist *Tuples;
 
 	BufFile    *psort_grab_file;
-	long		psort_current;	/* could be file offset, or array index */
+	long		psort_current;	/* array index (only used if not tape) */
+	int			psort_saved_fileno;	/* upper bits of psort_saved, if tape */
 	long		psort_saved;	/* could be file offset, or array index */
 	bool		using_tape_files;
 	bool		all_fetched;	/* this is for cursors */