Re: How to crash postgres using savepoints - Mailing list pgsql-hackers

On Tue, 2004-08-03 at 03:41, Gavin Sherry wrote:
> Attached is a patch fixing this.
> 
> One question I do have:
> 
>     if (target->savepointLevel != s->savepointLevel)
> 
> Will this ever be true in the current code? I cannot see anything setting
> savepointLevel explicitly.

>From reading the lists, it seems like that's allowing for some
functionality that was talked about but wasn't part of the semantics
agreed upon by the end of the discussion.

I have a question for you also. I just posted a patch at about the same
time you did (I sent it to pgsql-patches, but I haven't seen it appear
yet). Mine was a one-liner (appended to end of this email) and all it
did was add a check into the aforementioned line for a non-null target
pointer. My patch seemed to work, so I'd like to know why you changed
the structure around more. I did notice some things were a little
cleaner, so was it just clean-up or does my patch fail in some way?

Regards,Jeff

--- xact.c.old  2004-08-03 03:18:12.000000000 -0700
+++ xact.c      2004-08-03 03:19:05.000000000 -0700
@@ -2529,7 +2529,7 @@               target = target->parent;
               /* we don't cross savepoint level boundaries */
-               if (target->savepointLevel != s->savepointLevel)
+               if (PointerIsValid(target) && (target->savepointLevel !=
s->savepointLevel))                       ereport(ERROR,                                      
(errcode(ERRCODE_S_E_INVALID_SPECIFICATION),                                        errmsg("no such savepoint")));