Re: [Libreoffice-commits] core.git: passing a NULL pointer to fileno is not allowed

Hey Stephan,

On Fri, Jun 17, 2016 at 9:38 AM, Stephan Bergmann <sbergman@redhat.com>
wrote:

> On 06/15/2016 09:52 PM, Markus Mohrhard wrote:
>
> > commit b5876bfcb69a65c87d602bae687b3c0634c0a1e7
> > Author: Markus Mohrhard <markus.mohrhard@googlemail.com>
> > Date:   Wed Jun 15 20:15:20 2016 +0200
> >
> >     passing a NULL pointer to fileno is not allowed
> >
> >     See crash reports at
> >     http://crashreport.libreoffice.org/stats/signature/do_msvcr_magic+0x7
> >     and documentation at
> >     https://msdn.microsoft.com/en-us/library/zs6wbdhx.aspx
> >
> >     Change-Id: Ia9166d3b9fa10b87585821504e39cdfecbd22eda
> >     Reviewed-on: https://gerrit.libreoffice.org/26317
> >     Reviewed-by: Michael Stahl <mstahl@redhat.com>
> >     Tested-by: Markus Mohrhard <markus.mohrhard@googlemail.com>
> >
> > diff --git a/jvmfwk/plugins/sunmajor/pluginlib/sunjavaplugin.cxx
> > b/jvmfwk/plugins/sunmajor/pluginlib/sunjavaplugin.cxx
> > index fad3c0a9..43ddc82 100644
> > --- a/jvmfwk/plugins/sunmajor/pluginlib/sunjavaplugin.cxx
> > +++ b/jvmfwk/plugins/sunmajor/pluginlib/sunjavaplugin.cxx
> > @@ -548,6 +548,9 @@ static void do_msvcr_magic(rtl_uString *jvm_dll)
> >
> >      FILE *f = _wfopen(reinterpret_cast<LPCWSTR>(Module->buffer), L"rb");
> >
> > +    if (!f)
> > +        return;
> > +
> >      if (fstat(fileno(f), &st) == -1)
> >      {
> >          fclose(f);
>
> What I don't quite understand is: <
> https://cgit.freedesktop.org/libreoffice/core/commit/?id=a82e532ce006c54b2740de74d1da5d11307da7c1>
> "fdo#38913: Prevent invalid parameter handler crashes" establishes a
> _set_invalid_parameter_handler (that outputs to std::wcerr and then
> returns), so one would assume that calling fileno(nullptr) would not abort
> via MSVCR's _invalid_parameter_noinfo (see <
> http://crashreport.libreoffice.org/stats/crash_details/8e1e262d-3281-431c-89b5-a5642033fe94>)
> but rather return EINVAL.


If breakpad is enabled it overwrites the invalid parameter handler. We can
of course change that behavior (just a different flag to the breakpad init
code) but at least I think that it is a good thing to not ignore these
errors.

Regards,
Markus