RE: [PHP-DEV] Re: [PHP-CVS] com php-src: Fix possible overflow in openssl_pbkdf2: ext/openssl/openssl.c

Hi Matt, > -----Original Message----- > From: Matt Wilmas [mailto:php_lists@realplain.com] > Sent: Wednesday, August 19, 2015 12:40 AM > To: Anatol Belski <anatol.php@belski.net>; 'Jakub Zelenka' > <bukka@php.net> > Cc: php-cvs@lists.php.net; 'PHP internals list' <internals@lists.php.net> > Subject: Re: [PHP-DEV] Re: [PHP-CVS] com php-src: Fix possible overflow in > openssl_pbkdf2: ext/openssl/openssl.c > > Hi Anatol, > > ----- Original Message ----- > From: "Anatol Belski" > Sent: Tuesday, August 18, 2015 > > > Ok, got the idea now. But sizeof(size_t) is always > > > sizeof(zend_long). And for signed, still it is simpler and safer to > > just use an #ifdef around. > > No, == like I said. :-) OK, just saw your correction (max instead of > sizeof); but we're never comparing a size_t with zend_long, so it doesn't matter. > Yeah, I'm just too tired now, too much sizeof :) > I don't know where you're getting zend_long from anyway. [U]INT_MAX is an > int, not zend_long... > > The check in the macro, that the compiler can remove, should be fine in all cases > (even if *int* was 64-bit). > > I don't see how it's simpler or safer to use #if. > > But actually, if #if was used, should really be checking SIZE_MAX > UINT_MAX I > guess, and ZEND_LONG_MAX > INT_MAX. > > But using sizeof() in the macro definition gives the same result... > On ILP64 platform integer is 64-bit, yes. But that won't differ as that is 64-bit and unsigned (size_t) is always bigger than signed(int, zend_long, whatever signed). Now, what I initially was proposing - on 32-bit, #ifdef away the check about INT_MAX < some_int, nothing else. Nothing else can be improved in those checks (size_t > signed (int, zend_long, etc) on the same platform). It is not a big case over all the core, but some check in the openssl ext, no global thing here. So a simple practical solution for a concrete case :) With your function optimization example - yeah, it will most likely be that. However disputable in a concrete program/build, as it depends on the optimization algo, on the data source, on the compiler itself. IMHO it is good we have gcc or vc++, but still #ifdef excludes any interpretation and is simple. Good night Anatol