strip crashes on MSVC generated object files

strip crashes on MSVC generated object files

[Christian Groessler]

Hi,

current CVS strip targetted for i686-mingw32 crashes when stripping
MSVC generated object files.
It happens for example when using the --strip-debug option.

I've put a demo object file on
ftp://ftp.groessler.org/pub/chris/tmp/strip-crash.zip (blavc.obj).

These MSVC object files have an additional .drectve section which
might trigger the problem.

What happens is that coffgen.c(bfd_coff_get_comdat_section) gets
called with an *ABS* section and then crashes when dereferencing
used_by_bfd (in coff_section_data).

I've attached a patch to workaround the issue, but it's probably more
curing the symptoms instead of the disease?

regards,
chris



2005-02-18  Christian Groessler  <chris@groessler.org>

        * coffgen.c (bfd_coff_get_comdat_section): Do not call
        coff_section_data for *ABS* sections.



Index: bfd/coffgen.c
===================================================================
RCS file: /cvs/src/src/bfd/coffgen.c,v
retrieving revision 1.46
diff -p -u -r1.46 coffgen.c
--- bfd/coffgen.c       31 Jan 2005 23:13:17 -0000      1.46
+++ bfd/coffgen.c       18 Feb 2005 13:23:15 -0000
@@ -2427,6 +2427,8 @@ bfd_coff_set_symbol_class (abfd, symbol,
 struct coff_comdat_info *
 bfd_coff_get_comdat_section (bfd *abfd, struct bfd_section *sec)
 {
+  if (strcmp (sec->name, BFD_ABS_SECTION_NAME) == 0)
+    return NULL;
   if (bfd_get_flavour (abfd) == bfd_target_coff_flavour)
     return coff_section_data (abfd, sec)->comdat;
   else

Re: strip crashes on MSVC generated object files

[Alan Modra]

On Fri, Feb 18, 2005 at 02:47:52PM +0100, Christian Groessler wrote:
> I've attached a patch to workaround the issue, but it's probably more
> curing the symptoms instead of the disease?

It's a reasonable thing to do.  The disease in this case is that the
absolute input section isn't attached to each bfd that defines absolute
symbols.  However, fixing that properly requires quite a large patch.

I'm committing a more direct approach than your patch.  There are other
sections besides the absolute section that won't have coff_section_data.

	* coffgen.c (bfd_coff_get_comdat_section): Check that
	coff_section_data isn't NULL.

Index: bfd/coffgen.c
===================================================================
RCS file: /cvs/src/src/bfd/coffgen.c,v
retrieving revision 1.47
diff -u -p -r1.47 coffgen.c
--- bfd/coffgen.c	20 Feb 2005 14:59:06 -0000	1.47
+++ bfd/coffgen.c	21 Feb 2005 11:18:11 -0000
@@ -2427,7 +2427,8 @@ bfd_coff_set_symbol_class (abfd, symbol,
 struct coff_comdat_info *
 bfd_coff_get_comdat_section (bfd *abfd, struct bfd_section *sec)
 {
-  if (bfd_get_flavour (abfd) == bfd_target_coff_flavour)
+  if (bfd_get_flavour (abfd) == bfd_target_coff_flavour
+      && coff_section_data (abfd, sec) != NULL)
     return coff_section_data (abfd, sec)->comdat;
   else
     return NULL;

-- 
Alan Modra
IBM OzLabs - Linux Technology Centre