Repair very-low-probability race condition between relation extension

and VACUUM: in the interval between adding a new page to the relation
and formatting it, it was possible for VACUUM to come along and decide
it should format the page too.  Though not harmful in itself, this would
cause data loss if a third transaction were able to insert tuples into
the vacuumed page before the original extender got control back.

src/backend/access/heap/hio.c

@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/heap/hio.c,v 1.55 2005/04/29 22:28:23 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/heap/hio.c,v 1.56 2005/05/07 21:32:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -241,13 +241,6 @@ RelationGetBufferForTuple(Relation relation, Size len,
 	 */
 	buffer = ReadBuffer(relation, P_NEW);
 
-	/*
-	 * Release the file-extension lock; it's now OK for someone else to
-	 * extend the relation some more.
-	 */
-	if (needLock)
-		UnlockRelationForExtension(relation, ExclusiveLock);
-
 	/*
 	 * We can be certain that locking the otherBuffer first is OK, since
 	 * it must have a lower page number.
@@ -256,9 +249,22 @@ RelationGetBufferForTuple(Relation relation, Size len,
 		LockBuffer(otherBuffer, BUFFER_LOCK_EXCLUSIVE);
 
 	/*
-	 * We need to initialize the empty new page.
+	 * Now acquire lock on the new page.
 	 */
 	LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
+
+	/*
+	 * Release the file-extension lock; it's now OK for someone else to
+	 * extend the relation some more.  Note that we cannot release this
+	 * lock before we have buffer lock on the new page, or we risk a
+	 * race condition against vacuumlazy.c --- see comments therein.
+	 */
+	if (needLock)
+		UnlockRelationForExtension(relation, ExclusiveLock);
+
+	/*
+	 * We need to initialize the empty new page.
+	 */
 	pageHeader = (Page) BufferGetPage(buffer);
 	Assert(PageIsNew((PageHeader) pageHeader));
 	PageInit(pageHeader, BufferGetPageSize(buffer), 0);

src/backend/access/nbtree/nbtpage.c

@@ -9,7 +9,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtpage.c,v 1.83 2005/04/29 22:28:24 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtpage.c,v 1.84 2005/05/07 21:32:23 tgl Exp $
  *
  *	NOTES
  *	   Postgres btree pages look like ordinary relation pages.	The opaque
@@ -491,18 +491,21 @@ _bt_getbuf(Relation rel, BlockNumber blkno, int access)
 
 		buf = ReadBuffer(rel, P_NEW);
 
+		/* Acquire buffer lock on new page */
+		LockBuffer(buf, BT_WRITE);
+
 		/*
-		 * Release the file-extension lock; it's now OK for someone else
-		 * to extend the relation some more.
+		 * Release the file-extension lock; it's now OK for someone else to
+		 * extend the relation some more.  Note that we cannot release this
+		 * lock before we have buffer lock on the new page, or we risk a
+		 * race condition against btvacuumcleanup --- see comments therein.
 		 */
 		if (needLock)
 			UnlockRelationForExtension(rel, ExclusiveLock);
 
-		/* Acquire appropriate buffer lock on new page */
-		LockBuffer(buf, access);
-
 		/* Initialize the new page before returning it */
 		page = BufferGetPage(buf);
+		Assert(PageIsNew((PageHeader) page));
 		_bt_pageinit(page, BufferGetPageSize(buf));
 	}
 

src/backend/access/nbtree/nbtree.c

@@ -12,7 +12,7 @@
  * Portions Copyright (c) 1994, Regents of the University of California
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtree.c,v 1.128 2005/05/06 17:24:52 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/access/nbtree/nbtree.c,v 1.129 2005/05/07 21:32:23 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -739,11 +739,35 @@ btvacuumcleanup(PG_FUNCTION_ARGS)
 	BlockNumber pages_deleted = 0;
 	MemoryContext mycontext;
 	MemoryContext oldcontext;
+	bool		needLock;
 
 	Assert(stats != NULL);
 
+	/*
+	 * First find out the number of pages in the index.  We must acquire
+	 * the relation-extension lock while doing this to avoid a race
+	 * condition: if someone else is extending the relation, there is
+	 * a window where bufmgr/smgr have created a new all-zero page but
+	 * it hasn't yet been write-locked by _bt_getbuf().  If we manage to
+	 * scan such a page here, we'll improperly assume it can be recycled.
+	 * Taking the lock synchronizes things enough to prevent a problem:
+	 * either num_pages won't include the new page, or _bt_getbuf already
+	 * has write lock on the buffer and it will be fully initialized before
+	 * we can examine it.  (See also vacuumlazy.c, which has the same issue.)
+	 *
+	 * We can skip locking for new or temp relations,
+	 * however, since no one else could be accessing them.
+	 */
+	needLock = !RELATION_IS_LOCAL(rel);
+
+	if (needLock)
+		LockRelationForExtension(rel, ExclusiveLock);
+
 	num_pages = RelationGetNumberOfBlocks(rel);
 
+	if (needLock)
+		UnlockRelationForExtension(rel, ExclusiveLock);
+
 	/* No point in remembering more than MaxFSMPages pages */
 	maxFreePages = MaxFSMPages;
 	if ((BlockNumber) maxFreePages > num_pages)

src/backend/commands/vacuumlazy.c

@@ -31,7 +31,7 @@
  *
  *
  * IDENTIFICATION
- *	  $PostgreSQL: pgsql/src/backend/commands/vacuumlazy.c,v 1.52 2005/03/25 22:51:31 tgl Exp $
+ *	  $PostgreSQL: pgsql/src/backend/commands/vacuumlazy.c,v 1.53 2005/05/07 21:32:24 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -280,8 +280,30 @@ lazy_scan_heap(Relation onerel, LVRelStats *vacrelstats,
 
 		if (PageIsNew(page))
 		{
-			/* Not sure we still need to handle this case, but... */
+			/*
+			 * An all-zeroes page could be left over if a backend extends
+			 * the relation but crashes before initializing the page.
+			 * Reclaim such pages for use.
+			 *
+			 * We have to be careful here because we could be looking at
+			 * a page that someone has just added to the relation and not
+			 * yet been able to initialize (see RelationGetBufferForTuple).
+			 * To interlock against that, release the buffer read lock
+			 * (which we must do anyway) and grab the relation extension
+			 * lock before re-locking in exclusive mode.  If the page is
+			 * still uninitialized by then, it must be left over from a
+			 * crashed backend, and we can initialize it.
+			 *
+			 * We don't really need the relation lock when this is a new
+			 * or temp relation, but it's probably not worth the code space
+			 * to check that, since this surely isn't a critical path.
+			 *
+			 * Note: the comparable code in vacuum.c need not do all this
+			 * because it's got exclusive lock on the whole relation.
+			 */
 			LockBuffer(buf, BUFFER_LOCK_UNLOCK);
+			LockRelationForExtension(onerel, ExclusiveLock);
+			UnlockRelationForExtension(onerel, ExclusiveLock);
 			LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
 			if (PageIsNew(page))
 			{