Re: Intermittent "*** buffer overflow detected ***" errors with-D_FORTIFY_SOURCE=2 (?)

On Sat, Dec 05, 2009 at 06:17:23PM +0100, Cosimo Streppone wrote:
> Hi perl5-porters,
> 
> I'm trying to work on an "interesting" problem.
> It regards an XS module I wrote some time ago.
> This module used to work perfectly under various Linux distributions,
> including RHEL4, and Win32.
> 
> Now, with RHEL5, it's been reported to crash with "*** buffer overflow
> detected ***" messages,
> presumably when using `call_sv()' from XS code.
> 
> After digging a bit into the problem, it seems that the use of
> -D_FORTIFY_SOURCE=2
> in conjunction with and -O2 in $CFLAGS is causing these (fake?) buffer
> overflows.
> 
> Actually the problems are intermittent: that is, if I run the test
> suite of the module,
> it fails maybe 40%-50% of the times. Horrible.
> 
> Reading briefly about _FORTIFY_SOURCE=2 from people having the same problem,
> it would seem to me that it is just a really bad idea to have that enabled
> for a perl (interpreter and modules) installation. Unfortunately, this
> is the default in $CFLAGS for RHEL5.
> 
> So, coming to the question, has any of you dealt with this before?
> Any advice?

Assuming both perl and your module are built with debugging
information, what errors/backtraces do you see when you run your test
case under valgrind.

Of course, valgrind typically won't help if the overflowing buffers
are on the stack instead of malloced().

Tony

• Intermittent "*** buffer overflow detected ***" errors with -D_FORTIFY_SOURCE=2 (?) by Cosimo Streppone
• Re: Intermittent "*** buffer overflow detected ***" errors with-D_FORTIFY_SOURCE=2 (?) by Tony Cook