From: Robert Haas <rhaas@postgresql.org>
Date: Mon, 20 Sep 2021 18:07:01 +0000 (-0400)
Subject: can_allocate_segment could end up true on the first pass through the
X-Git-Url: http://git.postgresql.org/gitweb/static/gitweb.js?a=commitdiff_plain;h=df6ea196eca8030643f0048e072c5f99bf1dda6e;p=users%2Frhaas%2Fpostgres.git

can_allocate_segment could end up true on the first pass through the
loop, when we don't hold an exclusive lock. fix.
---

diff --git a/src/backend/access/conveyor/conveyor.c b/src/backend/access/conveyor/conveyor.c
index f4c78e495a..aa56050bf6 100644
--- a/src/backend/access/conveyor/conveyor.c
+++ b/src/backend/access/conveyor/conveyor.c
@@ -254,7 +254,8 @@ ConveyorBeltGetNewPage(ConveyorBelt *cb, CBPageNo *pageno)
 		 * only true if we've verified that the size of the relation on disk
 		 * is large enough.
 		 */
-		can_allocate_segment = (free_segno != CB_INVALID_SEGMENT)
+		can_allocate_segment = (mode == BUFFER_LOCK_EXCLUSIVE)
+			&& (free_segno != CB_INVALID_SEGMENT)
 			&& (free_segno < next_segno ||
 				free_segno < possibly_not_on_disk_segno);