Make json{b}_populate_recordset() use the right tuple descriptor.

json{b}_populate_recordset() used the tuple descriptor created from the
query-level AS clause without worrying about whether it matched the actual
input record type.  If it didn't, that would usually result in a crash,
though disclosure of server memory contents seems possible as well, for a
skilled attacker capable of issuing crafted SQL commands.  Instead, use
the query-supplied descriptor only when there is no input tuple to look at,
and otherwise get a tuple descriptor based on the input tuple's own type
marking.  The core code will detect any type mismatch in the latter case.

Michael Paquier and Tom Lane, per a report from David Rowley.
Back-patch to 9.3 where this functionality was introduced.

Security: CVE-2017-15098

3 files changed, 55 insertions, 21 deletions

diff --git a/src/backend/utils/adt/jsonfuncs.c b/src/backend/utils/adt/jsonfuncs.c
index 6d1fed7c063..3258db9449f 100644
--- a/src/backend/utils/adt/jsonfuncs.c
+++ b/src/backend/utils/adt/jsonfuncs.c
@@ -1601,19 +1601,47 @@ json_populate_recordset(PG_FUNCTION_ARGS)
 				 errmsg("set-valued function called in context that "
 						"cannot accept a set")));
 
-
 	rsi->returnMode = SFRM_Materialize;
 
-	/*
-	 * get the tupdesc from the result set info - it must be a record type
-	 * because we already checked that arg1 is a record type.
-	 */
-	(void) get_call_result_type(fcinfo, NULL, &tupdesc);
+	/* if the json is null send back an empty set */
+	if (PG_ARGISNULL(1))
+		PG_RETURN_NULL();
+
+	json = PG_GETARG_TEXT_P(1);
+
+	if (PG_ARGISNULL(0))
+	{
+		rec = NULL;
+
+		/*
+		 * get the tupdesc from the result set info - it must be a record type
+		 * because we already checked that arg1 is a record type
+		 */
+		if (get_call_result_type(fcinfo, NULL, &tupdesc) != TYPEFUNC_COMPOSITE)
+			ereport(ERROR,
+					(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
+					 errmsg("function returning record called in context "
+							"that cannot accept type record")));
+	}
+	else
+	{
+		rec = PG_GETARG_HEAPTUPLEHEADER(0);
+
+		/*
+		 * use the input record's own type marking to find a tupdesc for it.
+		 */
+		tupType = HeapTupleHeaderGetTypeId(rec);
+		tupTypmod = HeapTupleHeaderGetTypMod(rec);
+		tupdesc = lookup_rowtype_tupdesc(tupType, tupTypmod);
+	}
+
+	tupType = tupdesc->tdtypeid;
+	tupTypmod = tupdesc->tdtypmod;
+	ncolumns = tupdesc->natts;
 
 	state = palloc0(sizeof(PopulateRecordsetState));
 	sem = palloc0(sizeof(JsonSemAction));
 
-
 	/* make these in a sufficiently long-lived memory context */
 	old_cxt = MemoryContextSwitchTo(rsi->econtext->ecxt_per_query_memory);
 
@@ -1625,20 +1653,8 @@ json_populate_recordset(PG_FUNCTION_ARGS)
 
 	MemoryContextSwitchTo(old_cxt);
 
-	/* if the json is null send back an empty set */
-	if (PG_ARGISNULL(1))
-		PG_RETURN_NULL();
-
-	json = PG_GETARG_TEXT_P(1);
-
-	if (PG_ARGISNULL(0))
-		rec = NULL;
-	else
-		rec = PG_GETARG_HEAPTUPLEHEADER(0);
-
-	tupType = tupdesc->tdtypeid;
-	tupTypmod = tupdesc->tdtypmod;
-	ncolumns = tupdesc->natts;
+	/* unnecessary, but harmless, if tupdesc came from get_call_result_type: */
+	ReleaseTupleDesc(tupdesc);
 
 	lex = makeJsonLexContext(json, true);
 
diff --git a/src/test/regress/expected/json.out b/src/test/regress/expected/json.out
index bdedf3a7010..6df9af29f97 100644
--- a/src/test/regress/expected/json.out
+++ b/src/test/regress/expected/json.out
@@ -907,3 +907,16 @@ select * from json_populate_recordset(row('def',99,null)::jpop,'[{"a":[100,200,3
 ERROR:  cannot call json_populate_recordset on a nested object
 select * from json_populate_recordset(row('def',99,null)::jpop,'[{"c":[100,200,300],"x":43.2},{"a":{"z":true},"b":3,"c":"2012-01-20 10:42:53"}]') q;
 ERROR:  cannot call json_populate_recordset on a nested object
+-- negative cases where the wrong record type is supplied
+select * from json_populate_recordset(row(0::int),'[{"a":"1","b":"2"},{"a":"3"}]') q (a text, b text);
+ERROR:  function return row and query-specified return row do not match
+DETAIL:  Returned row contains 1 attribute, but query expects 2.
+select * from json_populate_recordset(row(0::int,0::int),'[{"a":"1","b":"2"},{"a":"3"}]') q (a text, b text);
+ERROR:  function return row and query-specified return row do not match
+DETAIL:  Returned type integer at ordinal position 1, but query expects text.
+select * from json_populate_recordset(row(0::int,0::int,0::int),'[{"a":"1","b":"2"},{"a":"3"}]') q (a text, b text);
+ERROR:  function return row and query-specified return row do not match
+DETAIL:  Returned row contains 3 attributes, but query expects 2.
+select * from json_populate_recordset(row(1000000000::int,50::int),'[{"b":"2"},{"a":"3"}]') q (a text, b text);
+ERROR:  function return row and query-specified return row do not match
+DETAIL:  Returned type integer at ordinal position 1, but query expects text.
diff --git a/src/test/regress/sql/json.sql b/src/test/regress/sql/json.sql
index 7a8c7b138a5..c6a07ef88be 100644
--- a/src/test/regress/sql/json.sql
+++ b/src/test/regress/sql/json.sql
@@ -302,3 +302,8 @@ select * from json_populate_recordset(row('def',99,null)::jpop,'[{"a":"blurfl","
 select * from json_populate_recordset(row('def',99,null)::jpop,'[{"a":[100,200,300],"x":43.2},{"a":{"z":true},"b":3,"c":"2012-01-20 10:42:53"}]') q;
 select * from json_populate_recordset(row('def',99,null)::jpop,'[{"c":[100,200,300],"x":43.2},{"a":{"z":true},"b":3,"c":"2012-01-20 10:42:53"}]') q;
 
+-- negative cases where the wrong record type is supplied
+select * from json_populate_recordset(row(0::int),'[{"a":"1","b":"2"},{"a":"3"}]') q (a text, b text);
+select * from json_populate_recordset(row(0::int,0::int),'[{"a":"1","b":"2"},{"a":"3"}]') q (a text, b text);
+select * from json_populate_recordset(row(0::int,0::int,0::int),'[{"a":"1","b":"2"},{"a":"3"}]') q (a text, b text);
+select * from json_populate_recordset(row(1000000000::int,50::int),'[{"b":"2"},{"a":"3"}]') q (a text, b text);