Attach FPI to the first record after full_page_writes is turned on.

XLogInsert fails to attach a required FPI to the first record after
full_page_writes is turned on by the last checkpoint.  This bug got
introduced in 9.5 due to code rearrangement in commits 2c03216 and
2076db2.  Fix it by ensuring that XLogInsertRecord performs a
recomputation when the given record is generated with FPW as off but
found that the flag has been turned on while actually inserting the
record.

Reported-by: Kyotaro Horiguchi
Author: Kyotaro Horiguchi
Reviewed-by: Amit Kapila
Backpatch-through: 9.5 where this problem was introduced
Discussion: https://postgr.es/m/20180420.151043.74298611.horiguchi.kyotaro@lab.ntt.co.jp

Author: Amit Kapila

src/backend/access/transam/xlog.c

@@ -943,7 +943,7 @@ static void WALInsertLockUpdateInsertingAt(XLogRecPtr insertingAt);
  *
  * If 'fpw_lsn' is valid, it is the oldest LSN among the pages that this
  * WAL record applies to, that were not included in the record as full page
- * images.  If fpw_lsn >= RedoRecPtr, the function does not perform the
+ * images.  If fpw_lsn <= RedoRecPtr, the function does not perform the
  * insertion and returns InvalidXLogRecPtr.  The caller can then recalculate
  * which pages need a full-page image, and retry.  If fpw_lsn is invalid, the
  * record is always inserted.
@@ -976,6 +976,7 @@ XLogInsertRecord(XLogRecData *rdata,
 							   info == XLOG_SWITCH);
 	XLogRecPtr	StartPos;
 	XLogRecPtr	EndPos;
+	bool		prevDoPageWrites = doPageWrites;
 
 	/* we assume that all of the record header is in the first chunk */
 	Assert(rdata->len >= SizeOfXLogRecord);
@@ -1023,10 +1024,14 @@ XLogInsertRecord(XLogRecData *rdata,
 		WALInsertLockAcquire();
 
 	/*
-	 * Check to see if my copy of RedoRecPtr or doPageWrites is out of date.
-	 * If so, may have to go back and have the caller recompute everything.
-	 * This can only happen just after a checkpoint, so it's better to be slow
-	 * in this case and fast otherwise.
+	 * Check to see if my copy of RedoRecPtr is out of date. If so, may have
+	 * to go back and have the caller recompute everything. This can only
+	 * happen just after a checkpoint, so it's better to be slow in this case
+	 * and fast otherwise.
+	 *
+	 * Also check to see if fullPageWrites or forcePageWrites was just turned
+	 * on; if we weren't already doing full-page writes then go back and
+	 * recompute.
 	 *
 	 * If we aren't doing full-page writes then RedoRecPtr doesn't actually
 	 * affect the contents of the XLOG record, so we'll update our local copy
@@ -1041,7 +1046,9 @@ XLogInsertRecord(XLogRecData *rdata,
 	}
 	doPageWrites = (Insert->fullPageWrites || Insert->forcePageWrites);
 
-	if (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr && doPageWrites)
+	if (doPageWrites &&
+		(!prevDoPageWrites ||
+		 (fpw_lsn != InvalidXLogRecPtr && fpw_lsn <= RedoRecPtr)))
 	{
 		/*
 		 * Oops, some buffer now needs to be backed up that the caller didn't