redo/undo support functions and cleanups.

src/backend/access/nbtree/nbtree.c

@@ -12,7 +12,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtree.c,v 1.65 2000/10/13 12:05:20 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/nbtree/nbtree.c,v 1.66 2000/10/20 11:01:03 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -1261,44 +1261,40 @@ _bt_del_item(Relation reln, Buffer buffer, BTItem btitem, bool insert,
 	}
 
 	lp = PageGetItemId(page, offno);
-	if (ItemIdDeleted(lp))	/* marked for deletion */
-	{
-		if (!InRecovery)
-			elog(STOP, "btree_%s_undo: deleted target tuple in rollback",
-				(insert) ? "insert" : "split");
-	}
-	else if (InRecovery)	/* check heap tuple */
+
+	if (InRecovery)					/* check heap tuple */
 	{
-		int			result;
-		CommandId	cid;
-		RelFileNode	hnode;
-		Size		hsize = (insert) ? SizeOfBtreeInsert : SizeOfBtreeSplit;
-
-		memcpy(&cid, (char*)xlrec + hsize, sizeof(CommandId));
-		memcpy(&hnode, (char*)xlrec + hsize + sizeof(CommandId), sizeof(RelFileNode));
-		result = XLogIsOwnerOfTuple(hnode, &(btitem->bti_itup.t_tid),
-					record->xl_xid, cid);
-		if (result < 0)	/* not owner */
+		if (!ItemIdDeleted(lp))
 		{
-			UnlockAndReleaseBuffer(buffer);
-			return;
-		}
-	}
-	else if (! BufferIsUpdatable(buffer))	/* normal rollback */
-	{
-		lp->lp_flags |= LP_DELETE;
-		MarkBufferForCleanup(buffer, IndexPageCleanup);
-		return;
-	}
+			int			result;
+			CommandId	cid;
+			RelFileNode	hnode;
+			Size		hsize = (insert) ? SizeOfBtreeInsert : SizeOfBtreeSplit;
 
-	PageIndexTupleDelete(page, offno);
-	if (InRecovery)
-	{
+			memcpy(&cid, (char*)xlrec + hsize, sizeof(CommandId));
+			memcpy(&hnode, (char*)xlrec + hsize + sizeof(CommandId), sizeof(RelFileNode));
+			result = XLogIsOwnerOfTuple(hnode, &(btitem->bti_itup.t_tid),
+						record->xl_xid, cid);
+			if (result < 0)	/* not owner */
+			{
+				UnlockAndReleaseBuffer(buffer);
+				return;
+			}
+		}
+		PageIndexTupleDelete(page, offno);
 		pageop = (BTPageOpaque) PageGetSpecialPointer(page);
 		pageop->btpo_flags |= BTP_REORDER;
+		UnlockAndWriteBuffer(buffer);
+		return;
 	}
-	UnlockAndWriteBuffer(buffer);
 
+	/* normal rollback */
+	if (ItemIdDeleted(lp))	/* marked for deletion ?! */
+		elog(STOP, "btree_%s_undo: deleted target tuple in rollback",
+			(insert) ? "insert" : "split");
+
+	lp->lp_flags |= LP_DELETE;
+	MarkBufferForCleanup(buffer, IndexPageCleanup);
 	return;
 }
 

src/backend/access/transam/xact.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/access/transam/xact.c,v 1.72 2000/10/11 21:28:17 momjian Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/access/transam/xact.c,v 1.73 2000/10/20 11:01:04 vadim Exp $
  *
  * NOTES
  *		Transaction aborts can now occur two ways:
@@ -154,6 +154,8 @@
  */
 #include "postgres.h"
 
+#include <sys/time.h>
+
 #include "access/nbtree.h"
 #include "catalog/heap.h"
 #include "catalog/index.h"
@@ -215,6 +217,19 @@ TransactionState CurrentTransactionState = &CurrentTransactionStateData;
 int			DefaultXactIsoLevel = XACT_READ_COMMITTED;
 int			XactIsoLevel;
 
+#ifdef XLOG
+#include "access/xlogutils.h"
+
+int			CommitDelay;
+
+void		xact_redo(XLogRecPtr lsn, XLogRecord *record);
+void		xact_undo(XLogRecPtr lsn, XLogRecord *record);
+
+static void (*_RollbackFunc)(void*) = NULL;
+static void *_RollbackData = NULL;
+
+#endif
+
 /* ----------------
  *		info returned when the system is disabled
  *
@@ -676,6 +691,28 @@ RecordTransactionCommit()
 		 */
 		TransactionIdCommit(xid);
 
+#ifdef XLOG
+		{
+			xl_xact_commit	xlrec;
+			struct timeval	delay;
+			XLogRecPtr		recptr;
+
+			xlrec.xtime = time(NULL);
+			/*
+			 * MUST SAVE ARRAY OF RELFILENODE-s TO DROP
+			 */
+			recptr = XLogInsert(RM_XACT_ID, XLOG_XACT_COMMIT,
+				(char*) &xlrec, SizeOfXactCommit, NULL, 0);
+
+			/* 
+			 * Sleep before commit! So we can flush more than one
+			 * commit records per single fsync.
+			 */
+			delay.tv_sec = 0;
+			delay.tv_usec = CommitDelay;
+			(void) select(0, NULL, NULL, NULL, &delay);
+		}
+#endif
 		/*
 		 * Now write the log info to the disk too.
 		 */
@@ -785,6 +822,18 @@ RecordTransactionAbort()
 	if (SharedBufferChanged && !TransactionIdDidCommit(xid))
 		TransactionIdAbort(xid);
 
+#ifdef XLOG
+	if (SharedBufferChanged)
+	{
+		xl_xact_abort	xlrec;
+		XLogRecPtr		recptr;
+
+		xlrec.xtime = time(NULL);
+		recptr = XLogInsert(RM_XACT_ID, XLOG_XACT_ABORT,
+			(char*) &xlrec, SizeOfXactAbort, NULL, 0);
+	}
+#endif
+
 	/*
 	 * Tell bufmgr and smgr to release resources.
 	 */
@@ -1123,10 +1172,13 @@ AbortTransaction()
 	AtEOXact_SPI();
 	AtEOXact_nbtree();
 	AtAbort_Cache();
-	AtAbort_Locks();
 	AtAbort_Memory();
 	AtEOXact_Files();
 
+	/* Here we'll rollback xaction changes */
+
+	AtAbort_Locks();
+
 	SharedBufferChanged = false; /* safest place to do it */
 
 	/* ----------------
@@ -1663,3 +1715,54 @@ IsTransactionBlock()
 
 	return false;
 }
+
+#ifdef XLOG
+
+void
+xact_redo(XLogRecPtr lsn, XLogRecord *record)
+{
+	uint8	info = record->xl_info & ~XLR_INFO_MASK;
+
+	if (info == XLOG_XACT_COMMIT)
+	{
+		xl_xact_commit	*xlrec = (xl_xact_commit*) XLogRecGetData(record);
+
+		XLogMarkCommitted(record->xl_xid);
+		/* MUST REMOVE FILES OF ALL DROPPED RELATIONS */
+	}
+	else if (info == XLOG_XACT_ABORT)
+	{
+		XLogMarkAborted(record->xl_xid);
+	}
+	else
+		elog(STOP, "xact_redo: unknown op code %u", info);
+}
+
+void
+xact_undo(XLogRecPtr lsn, XLogRecord *record)
+{
+	uint8	info = record->xl_info & ~XLR_INFO_MASK;
+
+	if (info == XLOG_XACT_COMMIT)	/* shouldn't be called by XLOG */
+		elog(STOP, "xact_undo: can't undo committed xaction");
+	else if (info != XLOG_XACT_ABORT)
+		elog(STOP, "xact_redo: unknown op code %u", info);
+}
+
+void
+XactPushRollback(void (*func) (void *), void* data)
+{
+	if (_RollbackFunc != NULL)
+		elog(STOP, "XactPushRollback: already installed");
+
+	_RollbackFunc = func;
+	_RollbackData = data;
+}
+
+void
+XactPopRollback(void)
+{
+	_RollbackFunc = NULL;
+}
+
+#endif

src/backend/access/transam/xlog.c

@@ -6,7 +6,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Header: /cvsroot/pgsql/src/backend/access/transam/xlog.c,v 1.17 2000/07/04 01:49:43 vadim Exp $
+ * $Header: /cvsroot/pgsql/src/backend/access/transam/xlog.c,v 1.18 2000/10/20 11:01:04 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -40,6 +40,7 @@ char		ControlFilePath[MAXPGPATH];
 uint32		XLOGbuffers = 0;
 XLogRecPtr	MyLastRecPtr = {0, 0};
 bool		StopIfError = false;
+bool		InRecovery = false;
 
 SPINLOCK	ControlFileLockId;
 SPINLOCK	XidGenLockId;
@@ -163,17 +164,6 @@ typedef struct CheckPoint
 #define NextBufIdx(curridx)		\
 		((curridx == XLogCtl->XLogCacheBlck) ? 0 : (curridx + 1))
 
-#define XLByteLT(left, right)		\
-			(right.xlogid > left.xlogid || \
-			(right.xlogid == left.xlogid && right.xrecoff > left.xrecoff))
-
-#define XLByteLE(left, right)		\
-			(right.xlogid > left.xlogid || \
-			(right.xlogid == left.xlogid && right.xrecoff >=  left.xrecoff))
-
-#define XLByteEQ(left, right)		\
-			(right.xlogid == left.xlogid && right.xrecoff ==  left.xrecoff)
-
 #define InitXLBuffer(curridx)	(\
 				XLogCtl->xlblocks[curridx].xrecoff = \
 				(XLogCtl->xlblocks[Insert->curridx].xrecoff == XLogFileSize) ? \

src/backend/access/transam/xlogutils.c

 /*-------------------------------------------------------------------------
  *
- * xlog.c
+ * xlogutils.c
  *
  *
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
@@ -9,12 +9,26 @@
  *-------------------------------------------------------------------------
  */
 
+#ifdef XLOG
+
 #include "postgres.h"
 
 #include "access/xlog.h"
 #include "access/xact.h"
+#include "storage/bufpage.h"
+#include "storage/bufmgr.h"
+#include "storage/smgr.h"
+#include "access/htup.h"
+#include "access/xlogutils.h"
+#include "catalog/pg_database.h"
 
-#ifdef XLOG
+/*
+ * ---------------------------------------------------------------
+ *
+ * Index support functions
+ *
+ *----------------------------------------------------------------
+ */
 
 /*
  * Check if specified heap tuple was inserted by given
@@ -59,7 +73,8 @@ XLogIsOwnerOfTuple(RelFileNode hnode, ItemPointer iptr,
 
 	htup = (HeapTupleHeader) PageGetItem(page, lp);
 
-	if (PageGetSUI(page) != ThisStartUpID || htup->t_xmin != xid || htup->t_cmin != cid)
+	Assert(PageGetSUI(page) == ThisStartUpID);
+	if (htup->t_xmin != xid || htup->t_cmin != cid)
 	{
 		UnlockAndReleaseBuffer(buffer);
 		return(-1);
@@ -70,6 +85,8 @@ XLogIsOwnerOfTuple(RelFileNode hnode, ItemPointer iptr,
 }
 
 /*
+ * MUST BE CALLED ONLY ON RECOVERY.
+ *
  * Check if exists valid (inserted by not aborted xaction) heap tuple
  * for given item pointer
  */
@@ -98,6 +115,14 @@ XLogIsValidTuple(RelFileNode hnode, ItemPointer iptr)
 		UnlockAndReleaseBuffer(buffer);
 		return(false);
 	}
+
+	if (PageGetSUI(page) != ThisStartUpID)
+	{
+		Assert(PageGetSUI(page) < ThisStartUpID);
+		UnlockAndReleaseBuffer(buffer);
+		return(true);
+	}
+
 	lp = PageGetItemId(page, ItemPointerGetOffsetNumber(iptr));
 	if (!ItemIdIsUsed(lp) || ItemIdDeleted(lp))
 	{
@@ -107,7 +132,9 @@ XLogIsValidTuple(RelFileNode hnode, ItemPointer iptr)
 
 	htup = (HeapTupleHeader) PageGetItem(page, lp);
 
-	if (XLogIsAborted(PageGetSUI(page), htup->t_xmin))
+	/* MUST CHECK WASN'T TUPLE INSERTED IN PREV STARTUP */
+
+	if (XLogIsAborted(htup->t_xmin))
 	{
 		UnlockAndReleaseBuffer(buffer);
 		return(false);
@@ -117,4 +144,257 @@ XLogIsValidTuple(RelFileNode hnode, ItemPointer iptr)
 	return(true);
 }
 
+/*
+ * ---------------------------------------------------------------
+ *
+ * Transaction support functions for recovery
+ *
+ * On recovery we create tmp file to know what xactions were
+ * committed/aborted (2 bits per xaction).
+ *
+ *----------------------------------------------------------------
+ */
+
+bool
+XLogIsAborted(TransactionId xid)
+{
+	return(false);
+}
+
+bool
+XLogIsCommitted(TransactionId xid)
+{
+	return(true);
+}
+
+void
+XLogMarkAborted(TransactionId xid)
+{
+	return;
+}
+
+void
+XLogMarkCommitted(TransactionId xid)
+{
+	return;
+}
+
+
+/*
+ * ---------------------------------------------------------------
+ *
+ * Storage related support functions
+ *
+ *----------------------------------------------------------------
+ */
+
+Buffer
+XLogReadBuffer(bool extend, Relation reln, BlockNumber blkno)
+{
+	BlockNumber	lastblock = RelationGetNumberOfBlocks(reln);
+	Buffer		buffer;
+
+	if (blkno >= lastblock)
+	{
+		buffer = InvalidBuffer;
+		if (extend)		/* we do this in recovery only - no locks */
+		{
+			Assert(InRecovery);
+			while (lastblock <= blkno)
+			{
+				buffer = ReadBuffer(reln, P_NEW);
+				lastblock++;
+			}
+		}
+		if (buffer != InvalidBuffer)
+			LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
+		return(buffer);
+	}
+
+	buffer = ReadBuffer(reln, blkno);
+	if (buffer != InvalidBuffer)
+		LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
+	return(buffer);
+}
+
+/*
+ * "Relation" cache
+ */
+
+typedef struct XLogRelDesc
+{
+	RelationData			reldata;
+	struct XLogRelDesc	   *lessRecently;
+	struct XLogRelDesc	   *moreRecently;
+} XLogRelDesc;
+
+typedef struct XLogRelCacheEntry
+{
+	RelFileNode		rnode;
+	XLogRelDesc	   *rdesc;
+} XLogRelCacheEntry;
+
+static HTAB				   *_xlrelcache;
+static XLogRelDesc		   *_xlrelarr = NULL;
+static Form_pg_class		_xlpgcarr = NULL;
+static int					_xlast = 0;
+static int					_xlcnt = 0;
+#define	_XLOG_INITRELCACHESIZE	32
+#define	_XLOG_MAXRELCACHESIZE	512
+
+void
+XLogCloseRelationCache(void)
+{
+	int i;
+
+	if (!_xlrelarr)
+		return;
+
+	for (i = 1; i < _xlast; i++)
+	{
+		Relation	reln = &(_xlrelarr[i].reldata);
+		if (reln->rd_fd >= 0)
+			smgrclose(DEFAULT_SMGR, reln);
+	}
+
+	free(_xlrelarr);
+	free(_xlpgcarr);
+
+	hash_destroy(_xlrelcache);
+	_xlrelarr = NULL;
+}
+
+static void
+_xl_init_rel_cache(void)
+{
+	HASHCTL	ctl;
+
+	_xlcnt = _XLOG_INITRELCACHESIZE;
+	_xlast = 0;
+	_xlrelarr = (XLogRelDesc*) malloc(sizeof(XLogRelDesc) * _xlcnt);
+	memset(_xlrelarr, 0, sizeof(XLogRelDesc) * _xlcnt);
+	_xlpgcarr = (Form_pg_class) malloc(sizeof(FormData_pg_class) * _xlcnt);
+	memset(_xlpgcarr, 0, sizeof(XLogRelDesc) * _xlcnt);
+
+	memset(&ctl, 0, (int) sizeof(ctl));
+	ctl.keysize = sizeof(RelFileNode);
+	ctl.datasize = sizeof(XLogRelDesc*);
+	ctl.hash = tag_hash;
+
+	_xlrelcache = hash_create(_XLOG_INITRELCACHESIZE, &ctl,
+								HASH_ELEM | HASH_FUNCTION);
+}
+
+static XLogRelDesc*
+_xl_new_reldesc(void)
+{
+	_xlast++;
+	if (_xlast < _xlcnt)
+	{
+		_xlrelarr[_xlast].reldata.rd_rel = &(_xlpgcarr[_xlast]);
+		return(&(_xlrelarr[_xlast]));
+	}
+
+	if ( 2 * _xlcnt <= _XLOG_MAXRELCACHESIZE)
+	{
+		_xlrelarr = (XLogRelDesc*) realloc(_xlrelarr, 
+						2 * sizeof(XLogRelDesc) * _xlcnt);
+		memset(&(_xlrelarr[_xlcnt]), 0, sizeof(XLogRelDesc) * _xlcnt);
+		_xlpgcarr = (Form_pg_class) realloc(_xlpgcarr, 
+						2 * sizeof(FormData_pg_class) * _xlcnt);
+		memset(&(_xlpgcarr[_xlcnt]), 0, sizeof(FormData_pg_class) * _xlcnt);
+		_xlcnt += _xlcnt;
+		_xlrelarr[_xlast].reldata.rd_rel = &(_xlpgcarr[_xlast]);
+		return(&(_xlrelarr[_xlast]));
+	}
+	else /* reuse */
+	{
+		XLogRelCacheEntry	   *hentry;
+		bool					found;
+		XLogRelDesc			   *res = _xlrelarr[0].moreRecently;
+		Form_pg_class			tpgc = res->reldata.rd_rel;
+
+		res->lessRecently->moreRecently = res->moreRecently;
+		res->moreRecently->lessRecently = res->lessRecently;
+
+		hentry = (XLogRelCacheEntry*) hash_search(_xlrelcache, 
+			(char*)&(res->reldata.rd_node), HASH_REMOVE, &found);
+
+		if (hentry == NULL)
+			elog(STOP, "XLogOpenRelation: can't delete from cache");
+		if (!found)
+			elog(STOP, "XLogOpenRelation: file was not found in cache");
+
+		if (res->reldata.rd_fd >= 0)
+			smgrclose(DEFAULT_SMGR, &(res->reldata));
+
+		memset(res, 0, sizeof(XLogRelDesc));
+		memset(tpgc, 0, sizeof(FormData_pg_class));
+		res->reldata.rd_rel = tpgc;
+
+		_xlast--;
+		return(res);
+	}
+}
+
+Relation
+XLogOpenRelation(bool redo, RmgrId rmid, RelFileNode rnode)
+{
+	XLogRelDesc			   *res;
+	XLogRelCacheEntry	   *hentry;
+	bool					found;
+
+	if (!_xlrelarr)
+		_xl_init_rel_cache();
+
+	hentry = (XLogRelCacheEntry*) 
+			hash_search(_xlrelcache, (char*)&rnode, HASH_FIND, &found);
+
+	if (hentry == NULL)
+		elog(STOP, "XLogOpenRelation: error in cache");
+
+	if (found)
+	{
+		res = hentry->rdesc;
+
+		res->lessRecently->moreRecently = res->moreRecently;
+		res->moreRecently->lessRecently = res->lessRecently;
+	}
+	else
+	{
+		res = _xl_new_reldesc();
+
+		sprintf(RelationGetPhysicalRelationName(&(res->reldata)), "%u", rnode.relNode);
+
+		/* unexisting DB id */
+		res->reldata.rd_lockInfo.lockRelId.dbId = RecoveryDb;
+		res->reldata.rd_lockInfo.lockRelId.relId = rnode.relNode;
+		res->reldata.rd_node = rnode;
+
+		hentry = (XLogRelCacheEntry*) 
+			hash_search(_xlrelcache, (char*)&rnode, HASH_ENTER, &found);
+
+		if (hentry == NULL)
+			elog(STOP, "XLogOpenRelation: can't insert into cache");
+
+		if (found)
+			elog(STOP, "XLogOpenRelation: file found on insert into cache");
+
+		hentry->rdesc = res;
+
+		res->reldata.rd_unlinked = true;	/* look smgropen */
+		res->reldata.rd_fd = smgropen(DEFAULT_SMGR, &(res->reldata));
+	}
+
+	res->moreRecently = &(_xlrelarr[0]);
+	res->lessRecently = _xlrelarr[0].lessRecently;
+	_xlrelarr[0].lessRecently = res;
+	res->lessRecently->moreRecently = res;
+
+	if (res->reldata.rd_fd < 0)		/* file doesn't exist */
+		return(NULL);
+
+	return(&(res->reldata));
+}
+
 #endif

src/backend/storage/buffer/bufmgr.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.87 2000/10/18 05:50:15 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.88 2000/10/20 11:01:07 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -2481,3 +2481,56 @@ AbortBufferIO(void)
 		SpinRelease(BufMgrLock);
 	}
 }
+
+/*
+ * Cleanup buffer or mark it for cleanup. Buffer may be cleaned
+ * up if it's pinned only once.
+ *
+ * NOTE: buffer must be excl locked.
+ */
+void
+MarkBufferForCleanup(Buffer buffer, void (*CleanupFunc)(Buffer))
+{
+	BufferDesc *bufHdr = &BufferDescriptors[buffer - 1];
+
+	Assert(PrivateRefCount[buffer - 1] > 0);
+
+	if (PrivateRefCount[buffer - 1] > 1)
+	{
+		LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
+		PrivateRefCount[buffer - 1]--;
+		SpinAcquire(BufMgrLock);
+		Assert(bufHdr->refcount > 0);
+		bufHdr->flags |= (BM_DIRTY | BM_JUST_DIRTIED);
+		bufHdr->CleanupFunc = CleanupFunc;
+		SpinRelease(BufMgrLock);
+		return;
+	}
+
+	SpinAcquire(BufMgrLock);
+	Assert(bufHdr->refcount > 0);
+	if (bufHdr->refcount == 1)
+	{
+		SpinRelease(BufMgrLock);
+		CleanupFunc(buffer);
+		CleanupFunc = NULL;
+	}
+	else
+		SpinRelease(BufMgrLock);
+
+	LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
+	PrivateRefCount[buffer - 1]--;
+
+	SpinAcquire(BufMgrLock);
+	Assert(bufHdr->refcount > 0);
+	bufHdr->flags |= (BM_DIRTY | BM_JUST_DIRTIED);
+	bufHdr->CleanupFunc = CleanupFunc;
+	bufHdr->refcount--;
+	if (bufHdr->refcount == 0)
+	{
+		AddBufferToFreelist(bufHdr);
+		bufHdr->flags |= BM_FREE;
+	}
+	SpinRelease(BufMgrLock);
+	return;
+}

src/backend/storage/smgr/md.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/md.c,v 1.75 2000/10/16 14:52:12 vadim Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/smgr/md.c,v 1.76 2000/10/20 11:01:11 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -875,12 +875,6 @@ mdtruncate(Relation reln, int nblocks)
 int
 mdcommit()
 {
-#ifdef XLOG
-	sync();
-	sleep(1);
-	sync();
-	return SM_SUCCESS;
-#else
 	int			i;
 	MdfdVec    *v;
 
@@ -902,7 +896,6 @@ mdcommit()
 	}
 
 	return SM_SUCCESS;
-#endif	 /* XLOG */
 }
 
 /*

src/include/access/htup.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: htup.h,v 1.35 2000/09/07 09:58:35 vadim Exp $
+ * $Id: htup.h,v 1.36 2000/10/20 11:01:14 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -95,7 +95,7 @@ typedef struct xl_heap_delete
 	xl_heaptid			target;		/* deleted tuple id */
 } xl_heap_delete;
 
-#define	SizeOfHeapDelete	(offsetof(xl_heaptid, tid) + SizeOfIptrData))
+#define	SizeOfHeapDelete	(offsetof(xl_heaptid, tid) + SizeOfIptrData)
 
 /* This is what we need to know about insert - 26 + data */
 typedef struct xl_heap_insert
@@ -111,12 +111,13 @@ typedef struct xl_heap_insert
 
 #define SizeOfHeapInsert	(offsetof(xl_heap_insert, mask) + sizeof(uint8))
 
-/* This is what we need to know about update - 28 + data */
+/* This is what we need to know about update - 32 + data */
 typedef struct xl_heap_update
 {
 	xl_heaptid			target;		/* deleted tuple id */
 	ItemPointerData		newtid;		/* new inserted tuple id */
 	/* something from header of new tuple version */
+	Oid					t_oid;
 	int16				t_natts;
 	uint8				t_hoff;
 	uint8				mask;		/* low 8 bits of t_infomask */

src/include/access/xact.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: xact.h,v 1.27 2000/07/28 01:04:40 tgl Exp $
+ * $Id: xact.h,v 1.28 2000/10/20 11:01:14 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -78,6 +78,35 @@ typedef TransactionStateData *TransactionState;
 	(*((TransactionId*) (dest)) = NullTransactionId)
 
 
+#ifdef XLOG
+
+/*
+ * XLOG allows to store some information in high 4 bits of log
+ * record xl_info field
+ */
+#define XLOG_XACT_COMMIT    0x00
+#define XLOG_XACT_ABORT     0x20
+
+typedef struct xl_xact_commit
+{
+	time_t		xtime;
+	/*
+	 * Array of RelFileNode-s to drop may follow
+	 * at the end of struct
+	 */
+} xl_xact_commit;
+
+#define SizeOfXactCommit	((offsetof(xl_xact_commit, xtime) + sizeof(time_t)))
+
+typedef struct xl_xact_abort
+{
+	time_t		xtime;
+} xl_xact_abort;
+
+#define SizeOfXactAbort	((offsetof(xl_xact_abort, xtime) + sizeof(time_t)))
+
+#endif
+
 /* ----------------
  *		extern definitions
  * ----------------
@@ -108,6 +137,9 @@ extern void AbortOutOfAnyTransaction(void);
 
 extern TransactionId DisabledTransactionId;
 
+extern void XactPushRollback(void (*func) (void *), void* data);
+extern void XactPopRollback(void);
+
 /* defined in xid.c */
 extern Datum xidin(PG_FUNCTION_ARGS);
 extern Datum xidout(PG_FUNCTION_ARGS);

src/include/access/xlog.h

@@ -72,12 +72,24 @@ typedef XLogPageHeaderData *XLogPageHeader;
 
 #define XLP_FIRST_IS_SUBRECORD	0x0001
 
+#define XLByteLT(left, right)		\
+			(right.xlogid > left.xlogid || \
+			(right.xlogid == left.xlogid && right.xrecoff > left.xrecoff))
+
+#define XLByteLE(left, right)		\
+			(right.xlogid > left.xlogid || \
+			(right.xlogid == left.xlogid && right.xrecoff >=  left.xrecoff))
+
+#define XLByteEQ(left, right)		\
+			(right.xlogid == left.xlogid && right.xrecoff ==  left.xrecoff)
+
 /*
  * StartUpID (SUI) - system startups counter.
  * It's to allow removing pg_log after shutdown.
  */
 typedef	uint32		StartUpID;
 extern	StartUpID	ThisStartUpID;
+extern	bool		InRecovery;
 
 extern XLogRecPtr XLogInsert(RmgrId rmid, uint8 info, 
 			char *hdr, uint32 hdrlen,

src/include/catalog/pg_database.h

@@ -8,7 +8,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: pg_database.h,v 1.10 2000/10/16 14:52:26 vadim Exp $
+ * $Id: pg_database.h,v 1.11 2000/10/20 11:01:17 vadim Exp $
  *
  * NOTES
  *	  the genbki.sh script reads this file and generates .bki
@@ -61,4 +61,12 @@ DESCR("");
 
 #define TemplateDbOid			1
 
+/* Just to mark OID as used for unused_oid script -:) */
+#define DATAMARKOID(x)
+
+DATAMARKOID( = 2)
+#define RecoveryDb	2
+
+#undef DATAMARKOID
+
 #endif	 /* PG_DATABASE_H */

src/include/storage/buf_internals.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: buf_internals.h,v 1.39 2000/10/18 05:50:16 vadim Exp $
+ * $Id: buf_internals.h,v 1.40 2000/10/20 11:01:21 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -121,10 +121,19 @@ typedef struct sbufdesc
 	 *
 	 * Why we keep relId here? To re-use file descriptors. On rollback
 	 * WAL uses dummy relId - bad (more blind writes - open/close calls),
-	 * but allowable. Obviously we should have another cache in file manager.
+	 * but allowable. Obviously we should have another cache in file manager
+	 * - fd is not relcache deal.
 	 */
 	LockRelId	relId;
 	BufferBlindId blind;		/* was used to support blind write */
+
+	/*
+	 * When we can't delete item from page (someone else has buffer pinned)
+	 * we mark buffer for cleanup by specifying appropriate for buffer
+	 * content cleanup function. Buffer will be cleaned up from release
+	 * buffer functions.
+	 */
+	void		(*CleanupFunc)(Buffer);
 } BufferDesc;
 
 /*

src/include/storage/bufmgr.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: bufmgr.h,v 1.40 2000/08/07 20:15:50 tgl Exp $
+ * $Id: bufmgr.h,v 1.41 2000/10/20 11:01:21 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -44,6 +44,17 @@ extern int	ShowPinTrace;
 #define BUFFER_LOCK_SHARE		1
 #define BUFFER_LOCK_EXCLUSIVE	2
 
+#define UnlockAndReleaseBuffer(buffer)	\
+( \
+	LockBuffer(buffer, BUFFER_LOCK_UNLOCK), \
+	ReleaseBuffer(buffer) \
+)
+
+#define UnlockAndWriteBuffer(buffer)	\
+( \
+	LockBuffer(buffer, BUFFER_LOCK_UNLOCK), \
+	WriteBuffer(buffer) \
+)
 
 /*
  * BufferIsValid
@@ -163,4 +174,7 @@ extern void UnlockBuffers(void);
 extern void LockBuffer(Buffer buffer, int mode);
 extern void AbortBufferIO(void);
 
+extern bool BufferIsUpdatable(Buffer buffer);
+extern void MarkBufferForCleanup(Buffer buffer, void (*CleanupFunc)(Buffer));
+
 #endif

src/include/storage/bufpage.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: bufpage.h,v 1.32 2000/09/07 09:58:36 vadim Exp $
+ * $Id: bufpage.h,v 1.33 2000/10/20 11:01:21 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -20,6 +20,7 @@
 #include "storage/itemid.h"
 #include "storage/off.h"
 #include "storage/page.h"
+#include "access/xlog.h"
 
 /*
  * a postgres disk page is an abstraction layered on top of a postgres

src/include/storage/itemid.h

@@ -7,7 +7,7 @@
  * Portions Copyright (c) 1996-2000, PostgreSQL, Inc
  * Portions Copyright (c) 1994, Regents of the University of California
  *
- * $Id: itemid.h,v 1.12 2000/09/07 09:58:36 vadim Exp $
+ * $Id: itemid.h,v 1.13 2000/10/20 11:01:21 vadim Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -40,6 +40,12 @@ typedef ItemIdData *ItemId;
 
 #endif
 
+/*
+ * This bit may be passed to PageAddItem together with
+ * LP_USED & LP_DELETED bits to specify overwrite mode
+ */
+#define	OverwritePageMode	0x10
+
 /*
  * Item offsets, lengths, and flags are represented by these types when
  * they're not actually stored in an ItemIdData.