On Wed, Dec 20, 2000 at 05:06:37PM +0200, Wirta, Ville wrote:
> I'm not actually reusing SSL_s but just read and write on it. Every
> operation is actually surrounded by mutexes so I'm quite a bit confused
> about what's still going wrong. (If you remember) I've been talking with you
> too about this before. ...the problems are still the same --> everything
> runs fine but occasionally an SSL_s struct just says 'caboom'. It either
> gets filled with zeros or has completely crazy values. What can be that
> horrible thing that threads are causing compared to processes? Or should I
> just make new SSL_s for every request instead of trying to read and write
> multiple times --> that works fine but means a new handshake for every
> connection --> CPU has a load of 100% :-(
There should be nothing wrong with multiple read/write processes :-)
If your SSL object gets stomped over, something is really messed up with
the memory management. I would however think, that you should be able to
nail this problem down using "efence", as it protects memory locations
before/after allocated regions, so if something stomps over your SSL object,
a BUS ERROR with core dump will be initiated (efence will generate a BUS
ERROR instead of the more obvious SEGV). You may then analyze the situation
leading to the problem.
Best regards,
Lutz
PS. The coredumps generated by efence are not completely perfect. On HP-UX
they are not usable as some information is missing. Having the process in
the debugger is no problem however... There is an appropriate warning in the
efence documentation.
--
Lutz Jaenicke [EMAIL PROTECTED]
BTU Cottbus http://www.aet.TU-Cottbus.DE/personen/jaenicke/
Lehrstuhl Allgemeine Elektrotechnik Tel. +49 355 69-4129
Universitaetsplatz 3-4, D-03044 Cottbus Fax. +49 355 69-4153
______________________________________________________________________
OpenSSL Project http://www.openssl.org
User Support Mailing List [EMAIL PROTECTED]
Automated List Manager [EMAIL PROTECTED]
