0001-Fix-rare-race-in-SSI-interaction-with-empty-btrees.patch

From 69bc0e964cfa0081058305c46f54f11b684919b1 Mon Sep 17 00:00:00 2001
From: Thomas Munro <thomas.munro@gmail.com>
Date: Mon, 19 Jun 2023 16:28:02 +1200
Subject: [PATCH] Fix rare race in SSI interaction with empty btrees.

When locking key-space gaps in btrees with a page-level SIREAD locks, we
have a special case for completely empty btrees, since there is no page
to lock.  This was racy, because a matching key could be inserted
between the _bt_search() and the PredicateLockRelation() calls, with
unluckly scheduling.

Fix, by rechecking _bt_search() after taking the relation-level SIREAD
lock, if using SERIALIZABLE isolation.

XXX WIP, not the end of the story

Bug #17949.

Reported-by: Artem Anisimov <artem.anisimov.255@gmail.com>
Discussion: https://postgr.es/m/17949-a0f17035294a55e2%40postgresql.org

diff --git a/src/backend/access/nbtree/nbtsearch.c b/src/backend/access/nbtree/nbtsearch.c
index 7e05e58676..162a06741d 100644
--- a/src/backend/access/nbtree/nbtsearch.c
+++ b/src/backend/access/nbtree/nbtsearch.c
@@ -1381,23 +1381,33 @@ _bt_first(IndexScanDesc scan, ScanDirection dir)
 	if (!BufferIsValid(buf))
 	{
 		/*
-		 * We only get here if the index is completely empty. Lock relation
-		 * because nothing finer to lock exists.
+		 * Since we have no pages locked, it's possible for another transaction
+		 * to insert data between _bt_search() and PredicateLockRelation().  We
+		 * have to try again after taking a relation-level predicate lock, to
+		 * close a narrow window where we wouldn't scan concurrently inserted
+		 * tuples, but the writer wouldn't see our predicate lock.
 		 */
-		PredicateLockRelation(rel, scan->xs_snapshot);
-
-		/*
-		 * mark parallel scan as done, so that all the workers can finish
-		 * their scan
-		 */
-		_bt_parallel_done(scan);
-		BTScanPosInvalidate(so->currPos);
+		if (IsolationIsSerializable())
+		{
+			PredicateLockRelation(rel, scan->xs_snapshot);
+			stack = _bt_search(rel, NULL, &inskey, &buf, BT_READ,
+							   scan->xs_snapshot);
+			_bt_freestack(stack);
+		}
 
-		return false;
+		if (!BufferIsValid(buf))
+		{
+			/*
+			 * Mark parallel scan as done, so that all the workers can finish
+			 * their scan.
+			 */
+			_bt_parallel_done(scan);
+			BTScanPosInvalidate(so->currPos);
+			return false;
+		}
 	}
-	else
-		PredicateLockPage(rel, BufferGetBlockNumber(buf),
-						  scan->xs_snapshot);
+
+	PredicateLockPage(rel, BufferGetBlockNumber(buf), scan->xs_snapshot);
 
 	_bt_initialize_more_data(so, dir);
 
-- 
2.39.2 (Apple Git-143)