Re: strange crash in SSL code while opening some documents

On 06/08/2013 03:01 PM, Markus Mohrhard wrote:
> 2013/6/8 Markus Mohrhard <markus.mohrhard@googlemail.com>:
> > so it seems that on my computer with gcc 4.8 and linux binutils
> > 2.23.51 I have some problems with our code for SSL handling.
> >
> > With each document trying to connect to a https website I get a crash
> > with the following backtrace:
> >
> > #0  0x00007ffff75cb207 in __strncmp_sse42 () from /lib64/libc.so.6
> > #1  0x00007fffba1b0f90 in ssl_cipher_process_rulestr
> > (rule_str=0x7fffba1f27b8 "AES:ALL:!aNULL:!eNULL:+RC4:@STRENGTH",
> > co_list=0x1dfd950, head_p=0x7fffffff3640, tail_p=0x7fffffff3638,
> > ca_list=0x1dfe240)
> >      at ssl_ciph.c:876
> > #2  0x00007fffba1b157e in ssl_create_cipher_list
> > (ssl_method=0x7fffba4447e0 <SSLv23_client_method_data.13759>,
> > cipher_list=0x1dfd318, cipher_list_by_id=0x1dfd320, rule_str=
> >      0x7fffba1f27b8 "AES:ALL:!aNULL:!eNULL:+RC4:@STRENGTH") at ssl_ciph.c:1040
> > #3  0x00007fffba1aa965 in SSL_CTX_new (meth=0x7fffba4447e0
> > <SSLv23_client_method_data.13759>) at ssl_lib.c:1534
> > #4  0x00007fffba0dd9bd in ne_ssl_context_create (mode=0) at
> > /devel/libo/libo3/workdir/unxlngx6/UnpackedTarball/neon/src/ne_openssl.c:568
> > #5  0x00007fffba0d4621 in ne_session_create (scheme=0x7fffcc169618
> > "https", hostname=0x7fffcc159350 "bankieren.rabobank.nl", port=443)
> >      at /devel/libo/libo3/workdir/unxlngx6/UnpackedTarball/neon/src/ne_session.c:176
> > #6  0x00007fffba48eef7 in webdav_ucp::NeonSession::Init
> > (this=0x1de7b70) at
> > /devel/libo/libo3/ucb/source/ucp/webdav-neon/NeonSession.cxx:711
> > #7  0x00007fffba48f40e in webdav_ucp::NeonSession::UsesProxy
> > (this=0x1de7b70) at
> > /devel/libo/libo3/ucb/source/ucp/webdav-neon/NeonSession.cxx:823
> > #8  0x00007fffba476a31 in webdav_ucp::DAVResourceAccess::getRequestURI
> > (this=0x1de7830) at
> > /devel/libo/libo3/ucb/source/ucp/webdav-neon/DAVResourceAccess.cxx:1107
> > #9  0x00007fffba472f28 in webdav_ucp::DAVResourceAccess::PROPFIND
> > (this=0x1de7830, nDepth=webdav_ucp::DAVZERO,
> > rPropertyNames=std::__debug::vector of length 4, capacity 4 = {...},
> >      rResources=std::__debug::vector of length 0, capacity 0,
> > xEnv=uno::Reference to (ucbhelper::CommandEnvironment *)
> > 0x7fffcc158678) at
> > /devel/libo/libo3/ucb/source/ucp/webdav-neon/DAVResourceAccess.cxx:239
> >
> >
> > So I debugged this and it looks like this might be a compile or
> > linking problem because stepping through the code does not show
> > anything going wrong but at some point the values seem to no longer
> > make sense.
> >
> > Can someone else with gcc 4.8 and linux current binutils check if he
> > sees the same problem? I think the new Fedora 19 uses the same
> > configuration so anyone running this and having a current master build
> > should be able to check it. A possible test document for this is
> > ooo121280 or lp990868.
>
>
> I'm sorry lp#990868 is an issue for calc developers. But the following
> documents should all show the issue (at least they do for me):
>
> Crash:/devel/files/ods/fdo42684-53636.ods through disposed

I could reproduce it with 
<https://bugs.freedesktop.org/attachment.cgi?id=53636> and my 
--without-system-libs build.  A regression related to symbol 
visibility/resolution, fixed now with 
<http://cgit.freedesktop.org/libreoffice/core/commit/?id=d257209a5443e27cefc8872151d776767605cbae> 
"Revert 'fix gold warning about hidden symbols.'"

Stephan