Install infrastructure for shared-memory free space map. Doesn't actually

do anything yet, but it has the necessary connections to initialization and so forth. Make some gestures towards allowing number of blocks in a relation to be BlockNumber, ie, unsigned int, rather than signed int. (I doubt I got all the places that are sloppy about it, yet.) On the way, replace the hardwired NLOCKS_PER_XACT fudge factor with a GUC variable.
2025-11-12 05:01:15 +03:00 · 2001-06-27 23:31:40 +00:00
parent b559382134
commit e0c9301c87
26 changed files with 572 additions and 316 deletions
--- a/src/backend/storage/Makefile
+++ b/src/backend/storage/Makefile
@@ -1,14 +1,14 @@
 #
 # Makefile for the storage manager subsystem
 #
-# $Header: /cvsroot/pgsql/src/backend/storage/Makefile,v 1.8 2000/08/31 16:10:30 petere Exp $
+# $Header: /cvsroot/pgsql/src/backend/storage/Makefile,v 1.9 2001/06/27 23:31:39 tgl Exp $
 #

 subdir = src/backend/storage
 top_builddir = ../../..
 include $(top_builddir)/src/Makefile.global

-SUBDIRS     := buffer file ipc large_object lmgr page smgr
+SUBDIRS     := buffer file freespace ipc large_object lmgr page smgr
 SUBDIROBJS  := $(SUBDIRS:%=%/SUBSYS.o)

 all: SUBSYS.o
--- a/src/backend/storage/freespace/Makefile
+++ b/src/backend/storage/freespace/Makefile
@@ -0,0 +1,30 @@
+#-------------------------------------------------------------------------
+#
+# Makefile--
+#    Makefile for storage/freespace
+#
+# IDENTIFICATION
+#    $Header: /cvsroot/pgsql/src/backend/storage/freespace/Makefile,v 1.1 2001/06/27 23:31:39 tgl Exp $
+#
+#-------------------------------------------------------------------------
+
+subdir = src/backend/storage/freespace
+top_builddir = ../../../..
+include $(top_builddir)/src/Makefile.global
+
+OBJS = freespace.o
+
+all: SUBSYS.o
+
+SUBSYS.o: $(OBJS)
+	$(LD) $(LDREL) $(LDOUT) SUBSYS.o $(OBJS)
+
+depend dep:
+	$(CC) -MM $(CFLAGS) *.c >depend
+
+clean: 
+	rm -f SUBSYS.o $(OBJS)
+
+ifeq (depend,$(wildcard depend))
+include depend
+endif
--- a/src/backend/storage/freespace/freespace.c
+++ b/src/backend/storage/freespace/freespace.c
@@ -0,0 +1,183 @@
+/*-------------------------------------------------------------------------
+ *
+ * freespace.c
+ *	  POSTGRES free space map for quickly finding free space in relations
+ *
+ *
+ * Portions Copyright (c) 1996-2001, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/freespace/freespace.c,v 1.1 2001/06/27 23:31:39 tgl Exp $
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "storage/freespace.h"
+#include "storage/itemid.h"
+#include "storage/shmem.h"
+
+
+/*
+ * Shared free-space-map objects
+ *
+ * Note: we handle pointers to these items as pointers, not as SHMEM_OFFSETs.
+ * This assumes that all processes accessing the map will have the shared
+ * memory segment mapped at the same place in their address space.
+ */
+typedef struct FSMHeader FSMHeader;
+typedef struct FSMRelation FSMRelation;
+typedef struct FSMChunk FSMChunk;
+
+/* Header for whole map */
+struct FSMHeader
+{
+	HTAB	   *relationHash;	/* hashtable of FSMRelation entries */
+	FSMRelation *relationList;	/* FSMRelations in order by recency of use */
+	int			numRelations;	/* number of FSMRelations now in use */
+	FSMChunk   *freeChunks;		/* linked list of currently-free chunks */
+};
+
+/*
+ * Per-relation struct --- this is an entry in the shared hash table.
+ * The hash key is the RelFileNode value (hence, we look at the physical
+ * relation ID, not the logical ID, which is appropriate).
+ */
+struct FSMRelation
+{
+	RelFileNode	key;			/* hash key (must be first) */
+	FSMRelation *nextRel;		/* next rel in order by recency of use */
+	FSMRelation *priorRel;		/* prior rel in order by recency of use */
+	FSMChunk   *relChunks;		/* linked list of page info chunks */
+};
+
+#define SHMEM_FSMHASH_KEYSIZE  sizeof(RelFileNode)
+#define SHMEM_FSMHASH_DATASIZE (sizeof(FSMRelation) - SHMEM_FSMHASH_KEYSIZE)
+
+#define CHUNKPAGES  32			/* each chunk can store this many pages */
+
+struct FSMChunk
+{
+	FSMChunk   *next;			/* linked-list link */
+	int			numPages;		/* number of pages described here */
+	BlockNumber	pages[CHUNKPAGES]; /* page numbers within relation */
+	ItemLength	bytes[CHUNKPAGES]; /* free space available on each page */
+};
+
+
+SPINLOCK	FreeSpaceLock;		/* in Shmem or created in
+								 * CreateSpinlocks() */
+
+int		MaxFSMRelations;		/* these are set by guc.c */
+int		MaxFSMPages;
+
+static FSMHeader *FreeSpaceMap;	/* points to FSMHeader in shared memory */
+
+
+/*
+ * InitFreeSpaceMap -- Initialize the freespace module.
+ *
+ * This must be called once during shared memory initialization.
+ * It builds the empty free space map table.  FreeSpaceLock must also be
+ * initialized at some point, but is not touched here --- we assume there is
+ * no need for locking, since only the calling process can be accessing shared
+ * memory as yet.  FreeSpaceShmemSize() was called previously while computing
+ * the space needed for shared memory.
+ */
+void
+InitFreeSpaceMap(void)
+{
+	HASHCTL		info;
+	FSMChunk   *chunks,
+			   *prevchunk;
+	int			nchunks;
+
+	/* Create table header */
+	FreeSpaceMap = (FSMHeader *) ShmemAlloc(sizeof(FSMHeader));
+	if (FreeSpaceMap == NULL)
+		elog(FATAL, "Insufficient shared memory for free space map");
+	MemSet(FreeSpaceMap, 0, sizeof(FSMHeader));
+
+	/* Create hashtable for FSMRelations */
+	info.keysize = SHMEM_FSMHASH_KEYSIZE;
+	info.datasize = SHMEM_FSMHASH_DATASIZE;
+	info.hash = tag_hash;
+
+	FreeSpaceMap->relationHash = ShmemInitHash("Free Space Map Hash",
+											   MaxFSMRelations / 10,
+											   MaxFSMRelations,
+											   &info,
+											   (HASH_ELEM | HASH_FUNCTION));
+
+	if (!FreeSpaceMap->relationHash)
+		elog(FATAL, "Insufficient shared memory for free space map");
+
+	/* Allocate FSMChunks and fill up the free-chunks list */
+	nchunks = (MaxFSMPages - 1) / CHUNKPAGES + 1;
+
+	chunks = (FSMChunk *) ShmemAlloc(nchunks * sizeof(FSMChunk));
+	if (chunks == NULL)
+		elog(FATAL, "Insufficient shared memory for free space map");
+
+	prevchunk = NULL;
+	while (nchunks-- > 0)
+	{
+		chunks->next = prevchunk;
+		prevchunk = chunks;
+		chunks++;
+	}
+	FreeSpaceMap->freeChunks = prevchunk;
+}
+
+
+int
+FreeSpaceShmemSize(void)
+{
+	int			size;
+	int			nchunks;
+
+	/*
+	 * There is no point in allowing less than one "chunk" per relation,
+	 * so force MaxFSMPages to be at least CHUNKPAGES * MaxFSMRelations.
+	 */
+	Assert(MaxFSMRelations > 0);
+	if (MaxFSMPages < CHUNKPAGES * MaxFSMRelations)
+		MaxFSMPages = CHUNKPAGES * MaxFSMRelations;
+
+	/* table header */
+	size = MAXALIGN(sizeof(FSMHeader));
+
+	/* hash table, including the FSMRelation objects */
+	size += hash_estimate_size(MaxFSMRelations,
+							   SHMEM_FSMHASH_KEYSIZE,
+							   SHMEM_FSMHASH_DATASIZE);
+
+	/* FSMChunk objects */
+	nchunks = (MaxFSMPages - 1) / CHUNKPAGES + 1;
+
+	size += MAXALIGN(nchunks * sizeof(FSMChunk));
+
+	return size;
+}
+
+
+void
+FreeSpaceMapForgetRel(RelFileNode *rel)
+{
+}
+
+
+#ifdef FREESPACE_DEBUG
+/*
+ * Dump contents of freespace map for debugging.
+ *
+ * We assume caller holds the FreeSpaceLock, or is otherwise unconcerned
+ * about other processes.
+ */
+void
+DumpFreeSpace(void)
+{
+}
+
+#endif	 /* FREESPACE_DEBUG */
--- a/src/backend/storage/ipc/ipci.c
+++ b/src/backend/storage/ipc/ipci.c
@@ -8,7 +8,7 @@
 *
 *
 * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/ipci.c,v 1.40 2001/03/22 03:59:45 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/ipci.c,v 1.41 2001/06/27 23:31:39 tgl Exp $
 *
 *-------------------------------------------------------------------------
 */
@@ -19,6 +19,7 @@
 #include "miscadmin.h"
 #include "access/xlog.h"
 #include "storage/bufmgr.h"
+#include "storage/freespace.h"
 #include "storage/lmgr.h"
 #include "storage/proc.h"
 #include "storage/sinval.h"
@@ -47,8 +48,12 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int maxBackends)
 	 * moderately-accurate estimates for the big hogs, plus 100K for the
 	 * stuff that's too small to bother with estimating.
 	 */
-	size = BufferShmemSize() + LockShmemSize(maxBackends) +
-		XLOGShmemSize() + SLockShmemSize() + SInvalShmemSize(maxBackends);
+	size = BufferShmemSize();
+	size += LockShmemSize(maxBackends);
+	size += XLOGShmemSize();
+	size += SLockShmemSize();
+	size += SInvalShmemSize(maxBackends);
+	size += FreeSpaceShmemSize();
 #ifdef STABLE_MEMORY_STORAGE
 	size += MMShmemSize();
 #endif
@@ -96,4 +101,9 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int maxBackends)
 	 * Set up shared-inval messaging
 	 */
 	CreateSharedInvalidationState(maxBackends);
+
+	/*
+	 * Set up free-space map
+	 */
+	InitFreeSpaceMap();
 }
--- a/src/backend/storage/ipc/spin.c
+++ b/src/backend/storage/ipc/spin.c
@@ -14,7 +14,7 @@
 *
 *
 * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/Attic/spin.c,v 1.32 2001/03/22 03:59:45 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/ipc/Attic/spin.c,v 1.33 2001/06/27 23:31:39 tgl Exp $
 *
 *-------------------------------------------------------------------------
 */
@@ -31,19 +31,18 @@


 /* Probably should move these to an appropriate header file */
-extern SPINLOCK ShmemLock;
-extern SPINLOCK ShmemIndexLock;
 extern SPINLOCK BufMgrLock;
-extern SPINLOCK LockMgrLock;
-extern SPINLOCK ProcStructLock;
-extern SPINLOCK SInvalLock;
 extern SPINLOCK OidGenLockId;
 extern SPINLOCK XidGenLockId;
 extern SPINLOCK ControlFileLockId;
-
+extern SPINLOCK ShmemLock;
+extern SPINLOCK ShmemIndexLock;
+extern SPINLOCK LockMgrLock;
+extern SPINLOCK SInvalLock;
+extern SPINLOCK ProcStructLock;
+extern SPINLOCK FreeSpaceLock;
 #ifdef STABLE_MEMORY_STORAGE
 extern SPINLOCK MMCacheLock;
-
 #endif


@@ -57,16 +56,16 @@ extern SPINLOCK MMCacheLock;
 static void
 InitSpinLockIDs(void)
 {
-	ShmemLock = (SPINLOCK) SHMEMLOCKID;
-	ShmemIndexLock = (SPINLOCK) SHMEMINDEXLOCKID;
 	BufMgrLock = (SPINLOCK) BUFMGRLOCKID;
-	LockMgrLock = (SPINLOCK) LOCKMGRLOCKID;
-	ProcStructLock = (SPINLOCK) PROCSTRUCTLOCKID;
-	SInvalLock = (SPINLOCK) SINVALLOCKID;
 	OidGenLockId = (SPINLOCK) OIDGENLOCKID;
 	XidGenLockId = (SPINLOCK) XIDGENLOCKID;
 	ControlFileLockId = (SPINLOCK) CNTLFILELOCKID;
-
+	ShmemLock = (SPINLOCK) SHMEMLOCKID;
+	ShmemIndexLock = (SPINLOCK) SHMEMINDEXLOCKID;
+	LockMgrLock = (SPINLOCK) LOCKMGRLOCKID;
+	SInvalLock = (SPINLOCK) SINVALLOCKID;
+	ProcStructLock = (SPINLOCK) PROCSTRUCTLOCKID;
+	FreeSpaceLock = (SPINLOCK) FREESPACELOCKID;
 #ifdef STABLE_MEMORY_STORAGE
 	MMCacheLock = (SPINLOCK) MMCACHELOCKID;
 #endif
--- a/src/backend/storage/lmgr/lock.c
+++ b/src/backend/storage/lmgr/lock.c
@@ -8,7 +8,7 @@
 *
 *
 * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/lock.c,v 1.89 2001/06/22 00:04:59 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/lmgr/lock.c,v 1.90 2001/06/27 23:31:39 tgl Exp $
 *
 * NOTES
 *	  Outside modules can create a lock table and acquire/release
@@ -40,6 +40,13 @@
 #include "utils/memutils.h"
 #include "utils/ps_status.h"

+
+/* This configuration variable is used to set the lock table size */
+int		max_locks_per_xact;		/* set by guc.c */
+
+#define NLOCKENTS(maxBackends)	(max_locks_per_xact * (maxBackends))
+
+
 static int WaitOnLock(LOCKMETHOD lockmethod, LOCKMODE lockmode,
 		   LOCK *lock, HOLDER *holder);
 static void LockCountMyLocks(SHMEM_OFFSET lockOffset, PROC *proc,
@@ -1388,6 +1395,7 @@ int
 LockShmemSize(int maxBackends)
 {
 	int			size = 0;
+	long		max_table_size = NLOCKENTS(maxBackends);

 	size += MAXALIGN(sizeof(PROC_HDR)); /* ProcGlobal */
 	size += maxBackends * MAXALIGN(sizeof(PROC));		/* each MyProc */
@@ -1395,12 +1403,12 @@ LockShmemSize(int maxBackends)
 																 * lockMethodTable->ctl */

 	/* lockHash table */
-	size += hash_estimate_size(NLOCKENTS(maxBackends),
+	size += hash_estimate_size(max_table_size,
 							   SHMEM_LOCKTAB_KEYSIZE,
 							   SHMEM_LOCKTAB_DATASIZE);

 	/* holderHash table */
-	size += hash_estimate_size(NLOCKENTS(maxBackends),
+	size += hash_estimate_size(max_table_size,
 							   SHMEM_HOLDERTAB_KEYSIZE,
 							   SHMEM_HOLDERTAB_DATASIZE);

--- a/src/backend/storage/smgr/md.c
+++ b/src/backend/storage/smgr/md.c
@@ -8,7 +8,7 @@
 *
 *
 * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/md.c,v 1.85 2001/06/06 17:07:46 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/md.c,v 1.86 2001/06/27 23:31:39 tgl Exp $
 *
 *-------------------------------------------------------------------------
 */
@@ -72,10 +72,10 @@ static MemoryContext MdCxt;		/* context for all my allocations */
 /* routines declared here */
 static void mdclose_fd(int fd);
 static int	_mdfd_getrelnfd(Relation reln);
-static MdfdVec *_mdfd_openseg(Relation reln, int segno, int oflags);
-static MdfdVec *_mdfd_getseg(Relation reln, int blkno);
+static MdfdVec *_mdfd_openseg(Relation reln, BlockNumber segno, int oflags);
+static MdfdVec *_mdfd_getseg(Relation reln, BlockNumber blkno);

-static int	_mdfd_blind_getseg(RelFileNode rnode, int blkno);
+static int	_mdfd_blind_getseg(RelFileNode rnode, BlockNumber blkno);

 static int	_fdvec_alloc(void);
 static void _fdvec_free(int);
@@ -93,7 +93,7 @@ static BlockNumber _mdnblocks(File file, Size blcksz);
 *		Returns SM_SUCCESS or SM_FAIL with errno set as appropriate.
 */
 int
-mdinit()
+mdinit(void)
 {
 	int			i;

@@ -194,11 +194,11 @@ mdunlink(RelFileNode rnode)
 	if (status == SM_SUCCESS)
 	{
 		char	   *segpath = (char *) palloc(strlen(path) + 12);
-		int			segno;
+		BlockNumber	segno;

 		for (segno = 1;; segno++)
 		{
-			sprintf(segpath, "%s.%d", path, segno);
+			sprintf(segpath, "%s.%u", path, segno);
 			if (unlink(segpath) < 0)
 			{
 				/* ENOENT is expected after the last segment... */
@@ -246,7 +246,7 @@ mdextend(Relation reln, BlockNumber blocknum, char *buffer)
 	v = _mdfd_getseg(reln, blocknum);

 #ifndef LET_OS_MANAGE_FILESIZE
-	seekpos = (long) (BLCKSZ * (blocknum % RELSEG_SIZE));
+	seekpos = (long) (BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE)));
 #ifdef DIAGNOSTIC
 	if (seekpos >= BLCKSZ * RELSEG_SIZE)
 		elog(FATAL, "seekpos too big!");
@@ -283,7 +283,7 @@ mdextend(Relation reln, BlockNumber blocknum, char *buffer)

 #ifndef LET_OS_MANAGE_FILESIZE
 #ifdef DIAGNOSTIC
-	if (_mdnblocks(v->mdfd_vfd, BLCKSZ) > RELSEG_SIZE)
+	if (_mdnblocks(v->mdfd_vfd, BLCKSZ) > ((BlockNumber) RELSEG_SIZE))
 		elog(FATAL, "segment too big!");
 #endif
 #endif
@@ -338,7 +338,7 @@ mdopen(Relation reln)
 	Md_fdvec[vfd].mdfd_chain = (MdfdVec *) NULL;

 #ifdef DIAGNOSTIC
-	if (_mdnblocks(fd, BLCKSZ) > RELSEG_SIZE)
+	if (_mdnblocks(fd, BLCKSZ) > ((BlockNumber) RELSEG_SIZE))
 		elog(FATAL, "segment too big on relopen!");
 #endif
 #endif
@@ -438,7 +438,7 @@ mdread(Relation reln, BlockNumber blocknum, char *buffer)
 	v = _mdfd_getseg(reln, blocknum);

 #ifndef LET_OS_MANAGE_FILESIZE
-	seekpos = (long) (BLCKSZ * (blocknum % RELSEG_SIZE));
+	seekpos = (long) (BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE)));

 #ifdef DIAGNOSTIC
 	if (seekpos >= BLCKSZ * RELSEG_SIZE)
@@ -482,7 +482,7 @@ mdwrite(Relation reln, BlockNumber blocknum, char *buffer)
 	v = _mdfd_getseg(reln, blocknum);

 #ifndef LET_OS_MANAGE_FILESIZE
-	seekpos = (long) (BLCKSZ * (blocknum % RELSEG_SIZE));
+	seekpos = (long) (BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE)));
 #ifdef DIAGNOSTIC
 	if (seekpos >= BLCKSZ * RELSEG_SIZE)
 		elog(FATAL, "seekpos too big!");
@@ -516,7 +516,7 @@ mdflush(Relation reln, BlockNumber blocknum, char *buffer)
 	v = _mdfd_getseg(reln, blocknum);

 #ifndef LET_OS_MANAGE_FILESIZE
-	seekpos = (long) (BLCKSZ * (blocknum % RELSEG_SIZE));
+	seekpos = (long) (BLCKSZ * (blocknum % ((BlockNumber) RELSEG_SIZE)));
 #ifdef DIAGNOSTIC
 	if (seekpos >= BLCKSZ * RELSEG_SIZE)
 		elog(FATAL, "seekpos too big!");
@@ -561,7 +561,7 @@ mdblindwrt(RelFileNode rnode,
 		return SM_FAIL;

 #ifndef LET_OS_MANAGE_FILESIZE
-	seekpos = (long) (BLCKSZ * (blkno % RELSEG_SIZE));
+	seekpos = (long) (BLCKSZ * (blkno % ((BlockNumber) RELSEG_SIZE)));
 #ifdef DIAGNOSTIC
 	if (seekpos >= BLCKSZ * RELSEG_SIZE)
 		elog(FATAL, "seekpos too big!");
@@ -659,16 +659,14 @@ mdblindmarkdirty(RelFileNode rnode,
 *
 *		Returns # of blocks, elog's on error.
 */
-int
+BlockNumber
 mdnblocks(Relation reln)
 {
 	int			fd;
 	MdfdVec    *v;
-
 #ifndef LET_OS_MANAGE_FILESIZE
-	int			nblocks;
-	int			segno;
-
+	BlockNumber	nblocks;
+	BlockNumber	segno;
 #endif

 	fd = _mdfd_getrelnfd(reln);
@@ -679,10 +677,10 @@ mdnblocks(Relation reln)
 	for (;;)
 	{
 		nblocks = _mdnblocks(v->mdfd_vfd, BLCKSZ);
-		if (nblocks > RELSEG_SIZE)
+		if (nblocks > ((BlockNumber) RELSEG_SIZE))
 			elog(FATAL, "segment too big in mdnblocks!");
-		if (nblocks < RELSEG_SIZE)
-			return (segno * RELSEG_SIZE) + nblocks;
+		if (nblocks < ((BlockNumber) RELSEG_SIZE))
+			return (segno * ((BlockNumber) RELSEG_SIZE)) + nblocks;
 		/*
 		 * If segment is exactly RELSEG_SIZE, advance to next one.
 		 */
@@ -713,18 +711,16 @@ mdnblocks(Relation reln)
 /*
 *	mdtruncate() -- Truncate relation to specified number of blocks.
 *
- *		Returns # of blocks or -1 on error.
+ *		Returns # of blocks or InvalidBlockNumber on error.
 */
-int
-mdtruncate(Relation reln, int nblocks)
+BlockNumber
+mdtruncate(Relation reln, BlockNumber nblocks)
 {
-	int			curnblk;
 	int			fd;
 	MdfdVec    *v;
-
+	BlockNumber	curnblk;
 #ifndef LET_OS_MANAGE_FILESIZE
-	int			priorblocks;
-
+	BlockNumber	priorblocks;
 #endif

 	/*
@@ -732,8 +728,8 @@ mdtruncate(Relation reln, int nblocks)
 	 * that truncate/delete loop will get them all!
 	 */
 	curnblk = mdnblocks(reln);
-	if (nblocks < 0 || nblocks > curnblk)
-		return -1;				/* bogus request */
+	if (nblocks > curnblk)
+		return InvalidBlockNumber; /* bogus request */
 	if (nblocks == curnblk)
 		return nblocks;			/* no work */

@@ -748,7 +744,6 @@ mdtruncate(Relation reln, int nblocks)

 		if (priorblocks > nblocks)
 		{
-
 			/*
 			 * This segment is no longer wanted at all (and has already
 			 * been unlinked from the mdfd_chain). We truncate the file
@@ -763,27 +758,25 @@ mdtruncate(Relation reln, int nblocks)
 												 * segment */
 			pfree(ov);
 		}
-		else if (priorblocks + RELSEG_SIZE > nblocks)
+		else if (priorblocks + ((BlockNumber) RELSEG_SIZE) > nblocks)
 		{
-
 			/*
 			 * This is the last segment we want to keep. Truncate the file
 			 * to the right length, and clear chain link that points to
 			 * any remaining segments (which we shall zap). NOTE: if
 			 * nblocks is exactly a multiple K of RELSEG_SIZE, we will
 			 * truncate the K+1st segment to 0 length but keep it. This is
-			 * mainly so that the right thing happens if nblocks=0.
+			 * mainly so that the right thing happens if nblocks==0.
 			 */
-			int			lastsegblocks = nblocks - priorblocks;
+			BlockNumber		lastsegblocks = nblocks - priorblocks;

 			if (FileTruncate(v->mdfd_vfd, lastsegblocks * BLCKSZ) < 0)
-				return -1;
+				return InvalidBlockNumber;
 			v = v->mdfd_chain;
 			ov->mdfd_chain = (MdfdVec *) NULL;
 		}
 		else
 		{
-
 			/*
 			 * We still need this segment and 0 or more blocks beyond it,
 			 * so nothing to do here.
@@ -794,7 +787,7 @@ mdtruncate(Relation reln, int nblocks)
 	}
 #else
 	if (FileTruncate(v->mdfd_vfd, nblocks * BLCKSZ) < 0)
-		return -1;
+		return InvalidBlockNumber;
 #endif

 	return nblocks;
@@ -940,7 +933,7 @@ _fdvec_free(int fdvec)
 }

 static MdfdVec *
-_mdfd_openseg(Relation reln, int segno, int oflags)
+_mdfd_openseg(Relation reln, BlockNumber segno, int oflags)
 {
 	MdfdVec    *v;
 	int			fd;
@@ -953,7 +946,7 @@ _mdfd_openseg(Relation reln, int segno, int oflags)
 	if (segno > 0)
 	{
 		fullpath = (char *) palloc(strlen(path) + 12);
-		sprintf(fullpath, "%s.%d", path, segno);
+		sprintf(fullpath, "%s.%u", path, segno);
 		pfree(path);
 	}
 	else
@@ -977,7 +970,7 @@ _mdfd_openseg(Relation reln, int segno, int oflags)
 	v->mdfd_chain = (MdfdVec *) NULL;

 #ifdef DIAGNOSTIC
-	if (_mdnblocks(fd, BLCKSZ) > RELSEG_SIZE)
+	if (_mdnblocks(fd, BLCKSZ) > ((BlockNumber) RELSEG_SIZE))
 		elog(FATAL, "segment too big on openseg!");
 #endif
 #endif
@@ -1007,17 +1000,19 @@ _mdfd_getrelnfd(Relation reln)
 /* Find the segment of the relation holding the specified block */

 static MdfdVec *
-_mdfd_getseg(Relation reln, int blkno)
+_mdfd_getseg(Relation reln, BlockNumber blkno)
 {
 	MdfdVec    *v;
-	int			segno;
 	int			fd;
-	int			i;
+#ifndef LET_OS_MANAGE_FILESIZE
+	BlockNumber	segno;
+	BlockNumber	i;
+#endif

 	fd = _mdfd_getrelnfd(reln);

 #ifndef LET_OS_MANAGE_FILESIZE
-	for (v = &Md_fdvec[fd], segno = blkno / RELSEG_SIZE, i = 1;
+	for (v = &Md_fdvec[fd], segno = blkno / ((BlockNumber) RELSEG_SIZE), i = 1;
 		 segno > 0;
 		 i++, segno--)
 	{
@@ -1038,7 +1033,7 @@ _mdfd_getseg(Relation reln, int blkno)
 			v->mdfd_chain = _mdfd_openseg(reln, i, (segno == 1) ? O_CREAT : 0);

 			if (v->mdfd_chain == (MdfdVec *) NULL)
-				elog(ERROR, "cannot open segment %d of relation %s (target block %d): %m",
+				elog(ERROR, "cannot open segment %u of relation %s (target block %u): %m",
 					 i, RelationGetRelationName(reln), blkno);
 		}
 		v = v->mdfd_chain;
@@ -1064,26 +1059,24 @@ _mdfd_getseg(Relation reln, int blkno)
 */

 static int
-_mdfd_blind_getseg(RelFileNode rnode, int blkno)
+_mdfd_blind_getseg(RelFileNode rnode, BlockNumber blkno)
 {
 	char	   *path;
 	int			fd;
-
 #ifndef LET_OS_MANAGE_FILESIZE
-	int			segno;
-
+	BlockNumber	segno;
 #endif

 	path = relpath(rnode);

 #ifndef LET_OS_MANAGE_FILESIZE
 	/* append the '.segno', if needed */
-	segno = blkno / RELSEG_SIZE;
+	segno = blkno / ((BlockNumber) RELSEG_SIZE);
 	if (segno > 0)
 	{
 		char	   *segpath = (char *) palloc(strlen(path) + 12);

-		sprintf(segpath, "%s.%d", path, segno);
+		sprintf(segpath, "%s.%u", path, segno);
 		pfree(path);
 		path = segpath;
 	}
--- a/src/backend/storage/smgr/mm.c
+++ b/src/backend/storage/smgr/mm.c
@@ -11,7 +11,7 @@
 *
 *
 * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/Attic/mm.c,v 1.23 2001/05/10 20:38:49 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/Attic/mm.c,v 1.24 2001/06/27 23:31:39 tgl Exp $
 *
 *-------------------------------------------------------------------------
 */
@@ -489,15 +489,15 @@ mmblindwrt(char *dbstr,
 /*
 *	mmnblocks() -- Get the number of blocks stored in a relation.
 *
- *		Returns # of blocks or -1 on error.
+ *		Returns # of blocks or InvalidBlockNumber on error.
 */
-int
+BlockNumber
 mmnblocks(Relation reln)
 {
 	MMRelTag	rtag;
 	MMRelHashEntry *rentry;
 	bool		found;
-	int			nblocks;
+	BlockNumber	nblocks;

 	if (reln->rd_rel->relisshared)
 		rtag.mmrt_dbid = (Oid) 0;
@@ -520,7 +520,7 @@ mmnblocks(Relation reln)
 	if (found)
 		nblocks = rentry->mmrhe_nblocks;
 	else
-		nblocks = -1;
+		nblocks = InvalidBlockNumber;

 	SpinRelease(MMCacheLock);

--- a/src/backend/storage/smgr/smgr.c
+++ b/src/backend/storage/smgr/smgr.c
@@ -11,13 +11,14 @@
 *
 *
 * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/smgr.c,v 1.49 2001/05/10 20:38:49 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/smgr.c,v 1.50 2001/06/27 23:31:39 tgl Exp $
 *
 *-------------------------------------------------------------------------
 */
 #include "postgres.h"

 #include "storage/bufmgr.h"
+#include "storage/freespace.h"
 #include "storage/smgr.h"
 #include "utils/memutils.h"

@@ -44,8 +45,8 @@ typedef struct f_smgr
 								  char *buffer, bool dofsync);
 	int			(*smgr_markdirty) (Relation reln, BlockNumber blkno);
 	int			(*smgr_blindmarkdirty) (RelFileNode, BlockNumber blkno);
-	int			(*smgr_nblocks) (Relation reln);
-	int			(*smgr_truncate) (Relation reln, int nblocks);
+	BlockNumber	(*smgr_nblocks) (Relation reln);
+	BlockNumber	(*smgr_truncate) (Relation reln, BlockNumber nblocks);
 	int			(*smgr_commit) (void);	/* may be NULL */
 	int			(*smgr_abort) (void);	/* may be NULL */
 	int			(*smgr_sync) (void);
@@ -433,16 +434,10 @@ smgrblindmarkdirty(int16 which,
 *		Returns the number of blocks on success, aborts the current
 *		transaction on failure.
 */
-int
+BlockNumber
 smgrnblocks(int16 which, Relation reln)
 {
-	int			nblocks;
-
-	if ((nblocks = (*(smgrsw[which].smgr_nblocks)) (reln)) < 0)
-		elog(ERROR, "cannot count blocks for %s: %m",
-			 RelationGetRelationName(reln));
-
-	return nblocks;
+	return (*(smgrsw[which].smgr_nblocks)) (reln);
 }

 /*
@@ -452,16 +447,24 @@ smgrnblocks(int16 which, Relation reln)
 *		Returns the number of blocks on success, aborts the current
 *		transaction on failure.
 */
-int
-smgrtruncate(int16 which, Relation reln, int nblocks)
+BlockNumber
+smgrtruncate(int16 which, Relation reln, BlockNumber nblocks)
 {
-	int			newblks;
+	BlockNumber		newblks;

 	newblks = nblocks;
 	if (smgrsw[which].smgr_truncate)
 	{
-		if ((newblks = (*(smgrsw[which].smgr_truncate)) (reln, nblocks)) < 0)
-			elog(ERROR, "cannot truncate %s to %d blocks: %m",
+		/*
+		 * Tell the free space map to forget this relation, so that it
+		 * stops caching info about the deleted blocks.  XXX perhaps
+		 * tell it to forget only info about blocks beyond nblocks?
+		 */
+		FreeSpaceMapForgetRel(&reln->rd_node);
+
+		newblks = (*(smgrsw[which].smgr_truncate)) (reln, nblocks);
+		if (newblks == InvalidBlockNumber)
+			elog(ERROR, "cannot truncate %s to %u blocks: %m",
 				 RelationGetRelationName(reln), nblocks);
 	}

@@ -481,7 +484,6 @@ smgrDoPendingDeletes(bool isCommit)
 		pendingDeletes = pending->next;
 		if (pending->atCommit == isCommit)
 		{
-
 			/*
 			 * Get rid of any leftover buffers for the rel (shouldn't be
 			 * any in the commit case, but there can be in the abort
@@ -489,6 +491,13 @@ smgrDoPendingDeletes(bool isCommit)
 			 */
 			DropRelFileNodeBuffers(pending->relnode);

+			/*
+			 * Tell the free space map to forget this relation.  It won't
+			 * be accessed any more anyway, but we may as well recycle the
+			 * map space quickly.
+			 */
+			FreeSpaceMapForgetRel(&pending->relnode);
+
 			/*
 			 * And delete the physical files.
 			 *