Re: ISAPI Crash Debugging Notes

--- Zeev Suraski <zeev@zend.com> wrote: > At 09:53 27/08/2002, Michael Sisolak wrote: > > > >i've been doing some debugging of the crashes when running php > under > > > >the isapi sapi. is anybody else currently looking at this? if > you > > > >are let's talk - here are my notes so far. > > > > > > first off, can you try using > > > http://www.php.net/~zeev/php-4.2.3rc1.tar.gz > > > and see if you experience the same problems? i've made some > > > thread-safety related fixes there, even though none should lead > > > to a hard crash... > > > >zeev, > > > >i believe that i may have tracked down a major isapi filter > >thread-safety hole. if i understand what zend_startup() is doing > >correctly, it assumes that the thread that it runs under will not > also > >be used to serve http requests. > > No, the startup thread can be used as a regular HTTP thread, nothing > in the > code should prevent that. > The globals code in the startup sequence is very tricky, but it > should work. This is a sample of the debug output that I'm seeing (this is PHP 4.2.1 running under Windows 2000 Professional, single CPU): [1948] the startup thread is 1956 [1948] GLOBAL_FUNCTION_TABLE is 4226120 [1948] in HttpExtensionProc thread 1956, CG(function_table) is 4226120 [1948] in HttpExtensionProc thread 1520, CG(function_table) is 18834536 [1948] in HttpExtensionProc thread 920, CG(function_table) is 18838912 [1948] in HttpExtensionProc thread 1956, CG(function_table) is 4226120 [1948] in HttpExtensionProc thread 1256, CG(function_table) is 19250808 [1948] in HttpExtensionProc thread 920, CG(function_table) is 18838912 [1948] in HttpExtensionProc thread 1956, CG(function_table) is 4226120 Notice that the initial thread (1956) appears to continue to use GLOBAL_FUNCTION_TABLE as it's own CG(function_table). I'm reporting the GLOBAL_FUNCTION_TABLE value from within zend_startup() - is there some point later in the code that makes it okay that this is the same value that I get for a CG(function_table) later? > Either way - starting threads from an ISAPI filter/extension is > completely > forbidden (or at least strongly discouraged) so even if there is some > bug > in there, starting our own thread is not an option. I don't believe that this is true. In this sample on MSDN: http://support.microsoft.com/default.aspx?scid=KB;EN-US;Q197728 their code for GetExtensionVersion calls AfxBeginThread. Also in this document: http://msdn.microsoft.com/library/en-us/vccore/html/vcconatlserverandcom.asp point 1 recommends "Create your own thread in GetExtensionVersion on which you can initialize COM..." > Can you give me the data I mentioned in the previous letter? This is PHP 4.2.1 running under Windows 2000 Professional (SP3) on a single CPU system. I can't really send the code as I'm currently testing with a large application that I have been running under CGI. I'll see if I can come up with a simpler example that causes the crash, but the debug output above would be the same either way. Please let me know if there is anything else I can provide to help in tracking this down. Michael Sisolak msisolak@gvpi.com __________________________________________________ Do You Yahoo!? Yahoo! Finance - Get real-time stock quotes http://finance.yahoo.com