DEREF_AFTER_NULL: src/common/jsonapi.c:2529 - Mailing list pgsql-hackers

From Галкин Сергей
Subject DEREF_AFTER_NULL: src/common/jsonapi.c:2529
Date
Msg-id 0b32e30f2fb94ae3b7f4ee15bbb072c0@rutoken.ru
Whole thread Raw
Responses Re: DEREF_AFTER_NULL: src/common/jsonapi.c:2529
List pgsql-hackers

Hello, a static analyzer pointed out a possible NULL dereference at the end of json_errdetail() (src/common/jsonapi.c):

return lex->errormsg->data;

That seemed plausible to me, since there is a comment just above saying that lex->errormsg can be NULL in shlib code. I also checked PQExpBufferBroken(), and it does handle NULL, but that call is under #ifdef, while the final access to lex->errormsg->data is unconditional.

I may be missing some invariant here, but it seems worth adding an explicit NULL check. I prepared a corresponding patch and am attaching it below in case you agree that this is a real issue.

diff --git a/src/common/jsonapi.c b/src/common/jsonapi.c
index 1145d93945f..192040b5443 100644
--- a/src/common/jsonapi.c
+++ b/src/common/jsonapi.c
@@ -2525,6 +2525,9 @@ json_errdetail(JsonParseErrorType error, JsonLexContext *lex)
     if (PQExpBufferBroken(lex->errormsg))
         return _("out of memory while constructing error description");
 #endif
+    
+    if (!lex->errormsg)
+        return _("out of memory while constructing error description");
 
     return lex->errormsg->data;
 }

Best regards, Galkin Sergey

Attachment

pgsql-hackers by date:

Previous
From: Michael Paquier
Date:
Subject: Re: BM_IO_ERROR flag is lost in TerminateBufferIO due to order of operations in UnlockBufHdrExt
Next
From: Alexandre Felipe
Date:
Subject: Re: SLOPE - Planner optimizations on monotonic expressions.