Fix incorrect accessing of pfree'd memory in Memoize

For pass-by-reference types, the code added in 0b053e7, which aimed to
resolve a memory leak, was overly aggressive in resetting the per-tuple
memory context which could result in pfree'd memory being accessed
resulting in failing to find previously cached results in the hash
table.

What was happening was prepare_probe_slot() was switching to the
per-tuple memory context and calling ExecEvalExpr().  ExecEvalExpr() may
have required a memory allocation.  Both MemoizeHash_hash() and
MemoizeHash_equal() were aggressively resetting the per-tuple context
and after determining the hash value, the context would have gotten reset
before MemoizeHash_equal() was called.  This could have resulted in
MemoizeHash_equal() looking at pfree'd memory.

This is less likely to have caused issues on a production build as some
other allocation would have had to have reused the pfree'd memory to
overwrite it.  Otherwise, the original contents would have been intact.
However, this clearly caused issues on MEMORY_CONTEXT_CHECKING builds.

Author: Tender Wang, Andrei Lepikhov
Reported-by: Tender Wang (using SQLancer)
Reviewed-by: Andrei Lepikhov, Richard Guo, David Rowley
Discussion: https://postgr.es/m/CAHewXNnT6N6UJkya0z-jLFzVxcwGfeRQSfhiwA+NyLg-x8iGew@mail.gmail.com
Backpatch-through: 14, where Memoize was added

Author: david-rowley

src/backend/executor/nodeMemoize.c

@@ -13,7 +13,7 @@
  * Memoize nodes are intended to sit above parameterized nodes in the plan
  * tree in order to cache results from them.  The intention here is that a
  * repeat scan with a parameter value that has already been seen by the node
- * can fetch tuples from the cache rather than having to re-scan the outer
+ * can fetch tuples from the cache rather than having to re-scan the inner
  * node all over again.  The query planner may choose to make use of one of
  * these when it thinks rescans for previously seen values are likely enough
  * to warrant adding the additional node.
@@ -207,7 +207,6 @@ MemoizeHash_hash(struct memoize_hash *tb, const MemoizeKey *key)
 		}
 	}
 
-	ResetExprContext(econtext);
 	MemoryContextSwitchTo(oldcontext);
 	return murmurhash32(hashkey);
 }
@@ -265,15 +264,14 @@ MemoizeHash_equal(struct memoize_hash *tb, const MemoizeKey *key1,
 			}
 		}
 
-		ResetExprContext(econtext);
 		MemoryContextSwitchTo(oldcontext);
 		return match;
 	}
 	else
 	{
 		econtext->ecxt_innertuple = tslot;
 		econtext->ecxt_outertuple = pslot;
-		return ExecQualAndReset(mstate->cache_eq_expr, econtext);
+		return ExecQual(mstate->cache_eq_expr, econtext);
 	}
 }
 
@@ -699,9 +697,18 @@ static TupleTableSlot *
 ExecMemoize(PlanState *pstate)
 {
 	MemoizeState *node = castNode(MemoizeState, pstate);
+	ExprContext *econtext = node->ss.ps.ps_ExprContext;
 	PlanState  *outerNode;
 	TupleTableSlot *slot;
 
+	CHECK_FOR_INTERRUPTS();
+
+	/*
+	 * Reset per-tuple memory context to free any expression evaluation
+	 * storage allocated in the previous tuple cycle.
+	 */
+	ResetExprContext(econtext);
+
 	switch (node->mstatus)
 	{
 		case MEMO_CACHE_LOOKUP:

src/test/regress/expected/memoize.out

@@ -129,10 +129,39 @@ WHERE t1.unique1 < 10;
     20 | 0.50000000000000000000
 (1 row)
 
+SET enable_mergejoin TO off;
+-- Test for varlena datatype with expr evaluation
+CREATE TABLE expr_key (x numeric, t text);
+INSERT INTO expr_key (x, t)
+SELECT d1::numeric, d1::text FROM (
+    SELECT round((d / pi())::numeric, 7) AS d1 FROM generate_series(1, 20) AS d
+) t;
+-- duplicate rows so we get some cache hits
+INSERT INTO expr_key SELECT * FROM expr_key;
+CREATE INDEX expr_key_idx_x_t ON expr_key (x, t);
+VACUUM ANALYZE expr_key;
+-- Ensure we get we get a cache miss and hit for each of the 20 distinct values
+SELECT explain_memoize('
+SELECT * FROM expr_key t1 INNER JOIN expr_key t2
+ON t1.x = t2.t::numeric AND t1.t::numeric = t2.x;', false);
+                                      explain_memoize                                      
+-------------------------------------------------------------------------------------------
+ Nested Loop (actual rows=80 loops=N)
+   ->  Seq Scan on expr_key t1 (actual rows=40 loops=N)
+   ->  Memoize (actual rows=2 loops=N)
+         Cache Key: t1.x, (t1.t)::numeric
+         Cache Mode: logical
+         Hits: 20  Misses: 20  Evictions: Zero  Overflows: 0  Memory Usage: NkB
+         ->  Index Only Scan using expr_key_idx_x_t on expr_key t2 (actual rows=2 loops=N)
+               Index Cond: (x = (t1.t)::numeric)
+               Filter: (t1.x = (t)::numeric)
+               Heap Fetches: N
+(10 rows)
+
+DROP TABLE expr_key;
 -- Reduce work_mem and hash_mem_multiplier so that we see some cache evictions
 SET work_mem TO '64kB';
 SET hash_mem_multiplier TO 1.0;
-SET enable_mergejoin TO off;
 -- Ensure we get some evictions.  We're unable to validate the hits and misses
 -- here as the number of entries that fit in the cache at once will vary
 -- between different machines.

src/test/regress/sql/memoize.sql

@@ -74,10 +74,31 @@ LATERAL (
 ON t1.two = t2.two
 WHERE t1.unique1 < 10;
 
+SET enable_mergejoin TO off;
+
+-- Test for varlena datatype with expr evaluation
+CREATE TABLE expr_key (x numeric, t text);
+INSERT INTO expr_key (x, t)
+SELECT d1::numeric, d1::text FROM (
+    SELECT round((d / pi())::numeric, 7) AS d1 FROM generate_series(1, 20) AS d
+) t;
+
+-- duplicate rows so we get some cache hits
+INSERT INTO expr_key SELECT * FROM expr_key;
+
+CREATE INDEX expr_key_idx_x_t ON expr_key (x, t);
+VACUUM ANALYZE expr_key;
+
+-- Ensure we get we get a cache miss and hit for each of the 20 distinct values
+SELECT explain_memoize('
+SELECT * FROM expr_key t1 INNER JOIN expr_key t2
+ON t1.x = t2.t::numeric AND t1.t::numeric = t2.x;', false);
+
+DROP TABLE expr_key;
+
 -- Reduce work_mem and hash_mem_multiplier so that we see some cache evictions
 SET work_mem TO '64kB';
 SET hash_mem_multiplier TO 1.0;
-SET enable_mergejoin TO off;
 -- Ensure we get some evictions.  We're unable to validate the hits and misses
 -- here as the number of entries that fit in the cache at once will vary
 -- between different machines.