From f51b34b3eddbc501063f7b8ac470d26ce4e18a48 Mon Sep 17 00:00:00 2001
From: Noah Misch <noah@leadboat.com>
Date: Tue, 24 Sep 2024 15:25:18 -0700
Subject: For inplace update durability, make heap_update() callers wait.

The previous commit fixed some ways of losing an inplace update.  It
remained possible to lose one when a backend working toward a
heap_update() copied a tuple into memory just before inplace update of
that tuple.  In catalogs eligible for inplace update, use LOCKTAG_TUPLE
to govern admission to the steps of copying an old tuple, modifying it,
and issuing heap_update().  This includes MERGE commands.  To avoid
changing most of the pg_class DDL, don't require LOCKTAG_TUPLE when
holding a relation lock sufficient to exclude inplace updaters.
Back-patch to v12 (all supported versions).  In v13 and v12, "UPDATE
pg_class" or "UPDATE pg_database" can still lose an inplace update.  The
v14+ UPDATE fix needs commit 86dc90056dfdbd9d1b891718d2e5614e3e432f35,
and it wasn't worth reimplementing that fix without such infrastructure.

Reviewed by Nitin Motiani and (in earlier versions) Heikki Linnakangas.

Discussion: https://postgr.es/m/20231027214946.79.nmisch@google.com
---
 src/backend/executor/nodeModifyTable.c | 30 ++++++++++++++++++++++++++++++
 1 file changed, 30 insertions(+)

(limited to 'src/backend/executor/nodeModifyTable.c')

diff --git a/src/backend/executor/nodeModifyTable.c b/src/backend/executor/nodeModifyTable.c
index 1006f450258..214482d193d 100644
--- a/src/backend/executor/nodeModifyTable.c
+++ b/src/backend/executor/nodeModifyTable.c
@@ -1825,6 +1825,7 @@ ExecUpdate(ModifyTableState *mtstate,
 	}
 	else
 	{
+		ItemPointerData lockedtid;
 		LockTupleMode lockmode;
 		bool		partition_constraint_failed;
 		bool		update_indexes;
@@ -1837,6 +1838,7 @@ ExecUpdate(ModifyTableState *mtstate,
 		 * to do them again.)
 		 */
 lreplace:
+		lockedtid = *tupleid;
 
 		/*
 		 * Constraints and GENERATED expressions might reference the tableoid
@@ -2014,6 +2016,14 @@ lreplace:
 							if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
 								ExecInitUpdateProjection(mtstate, resultRelInfo);
 
+							if (resultRelInfo->ri_needLockTagTuple)
+							{
+								UnlockTuple(resultRelationDesc,
+											&lockedtid, InplaceUpdateTupleLock);
+								LockTuple(resultRelationDesc,
+										  tupleid, InplaceUpdateTupleLock);
+							}
+
 							/* Fetch the most recent version of old tuple. */
 							oldSlot = resultRelInfo->ri_oldTupleSlot;
 							if (!table_tuple_fetch_row_version(resultRelationDesc,
@@ -2143,6 +2153,14 @@ ExecOnConflictUpdate(ModifyTableState *mtstate,
 	TransactionId xmin;
 	bool		isnull;
 
+	/*
+	 * Parse analysis should have blocked ON CONFLICT for all system
+	 * relations, which includes these.  There's no fundamental obstacle to
+	 * supporting this; we'd just need to handle LOCKTAG_TUPLE like the other
+	 * ExecUpdate() caller.
+	 */
+	Assert(!resultRelInfo->ri_needLockTagTuple);
+
 	/* Determine lock mode to use */
 	lockmode = ExecUpdateLockMode(estate, resultRelInfo);
 
@@ -2502,6 +2520,7 @@ ExecModifyTable(PlanState *pstate)
 	ItemPointerData tuple_ctid;
 	HeapTupleData oldtupdata;
 	HeapTuple	oldtuple;
+	bool		tuplock;
 
 	CHECK_FOR_INTERRUPTS();
 
@@ -2701,6 +2720,8 @@ ExecModifyTable(PlanState *pstate)
 								  estate, node->canSetTag);
 				break;
 			case CMD_UPDATE:
+				tuplock = false;
+
 				/* Initialize projection info if first time for this table */
 				if (unlikely(!resultRelInfo->ri_projectNewInfoValid))
 					ExecInitUpdateProjection(node, resultRelInfo);
@@ -2712,6 +2733,7 @@ ExecModifyTable(PlanState *pstate)
 				oldSlot = resultRelInfo->ri_oldTupleSlot;
 				if (oldtuple != NULL)
 				{
+					Assert(!resultRelInfo->ri_needLockTagTuple);
 					/* Use the wholerow junk attr as the old tuple. */
 					ExecForceStoreHeapTuple(oldtuple, oldSlot, false);
 				}
@@ -2721,6 +2743,11 @@ ExecModifyTable(PlanState *pstate)
 					Relation	relation = resultRelInfo->ri_RelationDesc;
 
 					Assert(tupleid != NULL);
+					if (resultRelInfo->ri_needLockTagTuple)
+					{
+						LockTuple(relation, tupleid, InplaceUpdateTupleLock);
+						tuplock = true;
+					}
 					if (!table_tuple_fetch_row_version(relation, tupleid,
 													   SnapshotAny,
 													   oldSlot))
@@ -2733,6 +2760,9 @@ ExecModifyTable(PlanState *pstate)
 				slot = ExecUpdate(node, resultRelInfo, tupleid, oldtuple, slot,
 								  planSlot, &node->mt_epqstate, estate,
 								  node->canSetTag);
+				if (tuplock)
+					UnlockTuple(resultRelInfo->ri_RelationDesc, tupleid,
+								InplaceUpdateTupleLock);
 				break;
 			case CMD_DELETE:
 				slot = ExecDelete(node, resultRelInfo, tupleid, oldtuple,
-- 
cgit v1.2.3