On Thu, 24 May 2001, Lars Hecking wrote:

> 
>  OS: SunOS 5.8 Generic_108528-07
>  Compiler: gcc 2.95.3
>  Software: mm-1.1.3, openssl-0.9.6a,
>            apache_1.3.19 and mod_ssl-2.8.3-1.3.19,
>            apache_1.3.20 and mod_ssl-2.8.4-1.3.20
> 
>  The new shmcb option in SSLSessionCache is broken, at least on my platform.
>  More or less all httpd's die a horrible death if a https connection is
>  initiated:

Yup. There has been a report of this on HPUX 11 (or maybe 10.20, can't recall)
but I was unable to get anyone to reproduce it with debugging turned on. Could
you please recompile your server with debugging and get a back-trace on one of
these? (Contact me if you need any help doing this). It would help us all a lot
(and would prevent anyone else encountering the same problem).

I suspect the problem is in one of the "safe" functions in shmcb. The cache's
internal structure keeps the data packed byte-by-byte. As such, these "safe"
functions exist to allow data-types that'd normally need to be aligned on
word-boundaries to be read/written to arbitrary addresses - it's achieved by
using memcpy() to read/write the data as a sequence of bytes.

Anyhow - one or two compilers were "optimising" these safe functions whereby
they'd realise what the trick was doing and replace the byte-by-byte copy with a
single read (ie. "wait, you're copying 4 bytes into an 'unsigned int' - why
don't I just convert that to a single read?"). In other words, defeating the
whole point of the function. :-( This could be what's happening in your case -
if so, we just need to make the "safe" functions safer. The other possibility is
that some part of the code should be using a "safe" function and it has been
skipped - if so, that's even easier to fix.

What CPU are you running BTW? It's obviously not a pentium variant :-)

> [Tue May 22 18:39:59 2001] [notice] child pid 11584 exit signal Bus Error (10)

Word-type operations that don't sit on the required byte-alignment generate a
SIGBUS error - so this really indicates that it's simply a "safe" function
issue. This should be easy to correct - we just need to get a backtrace to find
out which line(s) of code the crash is occuring in.

Cheers,
Geoff


______________________________________________________________________
Apache Interface to OpenSSL (mod_ssl)                   www.modssl.org
User Support Mailing List                      [EMAIL PROTECTED]
Automated List Manager                            [EMAIL PROTECTED]

Reply via email to