Back-patch fix to grab read lock on a buffer while it is written out.

1 files changed, 47 insertions, 33 deletions

diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index aa93a04b3c1..a1edc3a3db1 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -8,7 +8,7 @@
  *
  *
  * IDENTIFICATION
- *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.81 2000/05/19 03:22:28 tgl Exp $
+ *	  $Header: /cvsroot/pgsql/src/backend/storage/buffer/bufmgr.c,v 1.81.2.1 2000/09/25 04:34:10 tgl Exp $
  *
  *-------------------------------------------------------------------------
  */
@@ -75,7 +75,6 @@ static void WaitIO(BufferDesc *buf, SPINLOCK spinlock);
 static void StartBufferIO(BufferDesc *buf, bool forInput);
 static void TerminateBufferIO(BufferDesc *buf);
 static void ContinueBufferIO(BufferDesc *buf, bool forInput);
-extern void InitBufferIO(void);
 extern void AbortBufferIO(void);
 
 /*
@@ -430,17 +429,10 @@ BufferAlloc(Relation reln,
 	inProgress = FALSE;
 	for (buf = (BufferDesc *) NULL; buf == (BufferDesc *) NULL;)
 	{
-
-		/* GetFreeBuffer will abort if it can't find a free buffer */
 		buf = GetFreeBuffer();
 
-		/*
-		 * But it can return buf == NULL if we are in aborting transaction
-		 * now and so elog(ERROR,...) in GetFreeBuffer will not abort
-		 * again.
-		 */
-		if (buf == NULL)
-			return NULL;
+		/* GetFreeBuffer will abort if it can't find a free buffer */
+		Assert(buf);
 
 		/*
 		 * There should be exactly one pin on the buffer after it is
@@ -790,11 +782,21 @@ FlushBuffer(Buffer buffer, bool release)
 	WaitIO(bufHdr, BufMgrLock); /* confirm end of IO */
 	bufHdr->flags &= ~BM_JUST_DIRTIED;
 	StartBufferIO(bufHdr, false);		/* output IO start */
+
 	SpinRelease(BufMgrLock);
 
+	/*
+	 * Grab a read lock on the buffer to ensure that no
+	 * other backend changes its contents while we write it;
+	 * see comments in BufferSync().
+	 */
+	LockBuffer(BufferDescriptorGetBuffer(bufHdr), BUFFER_LOCK_SHARE);
+
 	status = smgrflush(DEFAULT_SMGR, bufrel, bufHdr->tag.blockNum,
 					   (char *) MAKE_PTR(bufHdr->data));
 
+	LockBuffer(BufferDescriptorGetBuffer(bufHdr), BUFFER_LOCK_UNLOCK);
+
 	/* drop relcache refcnt incremented by RelationIdCacheGetRelation */
 	RelationDecrementReferenceCount(bufrel);
 
@@ -1018,19 +1020,6 @@ ClearBufferDirtiedByMe(Buffer buffer, BufferDesc *bufHdr)
  *		that have been dirtied by the current xact and flush them to disk.
  *		We do *not* flush dirty buffers that have been dirtied by other xacts.
  *		(This is a substantial change from pre-7.0 behavior.)
- *
- *	OLD COMMENTS (do these still apply?)
- *
- *		Also, we need to be sure that no other transaction is
- *		modifying the page as we flush it.	This is only a problem for objects
- *		that use a non-two-phase locking protocol, like btree indices.	For
- *		those objects, we would like to set a write lock for the duration of
- *		our IO.  Another possibility is to code updates to btree pages
- *		carefully, so that writing them out out of order cannot cause
- *		any unrecoverable errors.
- *
- *		I don't want to think hard about this right now, so I will try
- *		to come back to it later.
  */
 static void
 BufferSync()
@@ -1113,15 +1102,28 @@ BufferSync()
 					bufHdr->flags &= ~BM_JUST_DIRTIED;
 					StartBufferIO(bufHdr, false);		/* output IO start */
 
+					SpinRelease(BufMgrLock);
+
+					/*
+					 * Grab a read lock on the buffer to ensure that no
+					 * other backend changes its contents while we write it;
+					 * otherwise we could write a non-self-consistent page
+					 * image to disk, which'd be bad news if the other
+					 * transaction aborts before writing its changes.
+					 *
+					 * Note that we still need the BM_JUST_DIRTIED mechanism
+					 * in case someone dirties the buffer just before we
+					 * grab this lock or just after we release it.
+					 */
+					LockBuffer(BufferDescriptorGetBuffer(bufHdr),
+							   BUFFER_LOCK_SHARE);
+
 					/*
 					 * If we didn't have the reldesc in our local cache,
 					 * write this page out using the 'blind write' storage
 					 * manager routine.  If we did find it, use the
 					 * standard interface.
 					 */
-#ifndef OPTIMIZE_SINGLE
-					SpinRelease(BufMgrLock);
-#endif	 /* OPTIMIZE_SINGLE */
 					if (reln == (Relation) NULL)
 					{
 						status = smgrblindwrt(DEFAULT_SMGR,
@@ -1138,9 +1140,14 @@ BufferSync()
 										   bufHdr->tag.blockNum,
 										(char *) MAKE_PTR(bufHdr->data));
 					}
-#ifndef OPTIMIZE_SINGLE
+
+					/*
+					 * Release the per-buffer readlock, reacquire BufMgrLock.
+					 */
+					LockBuffer(BufferDescriptorGetBuffer(bufHdr),
+							   BUFFER_LOCK_UNLOCK);
+
 					SpinAcquire(BufMgrLock);
-#endif	 /* OPTIMIZE_SINGLE */
 
 					UnpinBuffer(bufHdr);
 					if (status == SM_FAIL)
@@ -1523,9 +1530,14 @@ BufferReplace(BufferDesc *bufHdr)
 	/* To check if block content changed while flushing. - vadim 01/17/97 */
 	bufHdr->flags &= ~BM_JUST_DIRTIED;
 
-#ifndef OPTIMIZE_SINGLE
 	SpinRelease(BufMgrLock);
-#endif	 /* OPTIMIZE_SINGLE */
+
+	/*
+	 * Grab a read lock on the buffer to ensure that no
+	 * other backend changes its contents while we write it;
+	 * see comments in BufferSync().
+	 */
+	LockBuffer(BufferDescriptorGetBuffer(bufHdr), BUFFER_LOCK_SHARE);
 
 	if (reln != (Relation) NULL)
 	{
@@ -1541,9 +1553,9 @@ BufferReplace(BufferDesc *bufHdr)
 							  false);	/* no fsync */
 	}
 
-#ifndef OPTIMIZE_SINGLE
+	LockBuffer(BufferDescriptorGetBuffer(bufHdr), BUFFER_LOCK_UNLOCK);
+
 	SpinAcquire(BufMgrLock);
-#endif	 /* OPTIMIZE_SINGLE */
 
 	/* drop relcache refcnt incremented by RelationIdCacheGetRelation */
 	if (reln != (Relation) NULL)
@@ -2488,11 +2500,13 @@ ContinueBufferIO(BufferDesc *buf, bool forInput)
 	IsForInput = forInput;
 }
 
+#ifdef NOT_USED
 void
 InitBufferIO(void)
 {
 	InProgressBuf = (BufferDesc *) 0;
 }
+#endif
 
 /*
  *	This function is called from ProcReleaseSpins().