[5/10] RFC: introduce gdb_bfd_section_index

[Yao Qi]

Tom,
This patch is related to a GDB crash for tic6x-uclinux target.  I am not 
familiar with objfile/section stuff, so correct me if my thoughts below 
are incorrect.

On 02/19/2013 04:27 AM, Tom Tromey wrote:
>   void
>   build_objfile_section_table (struct objfile *objfile)
>   {
> -  objfile->sections_end = 0;
> +  int count = gdb_bfd_count_sections (objfile->obfd);
> +
> +  objfile->sections = OBSTACK_CALLOC (&objfile->objfile_obstack,
> +				      count,
> +				      struct obj_section);
> +  objfile->sections_end = (objfile->sections + count);

We start allocate 'gdb_bfd_count_sections' (which is bfd_count_sections 
+ 4) entries of sections.

>     bfd_map_over_sections (objfile->obfd,
>   			 add_to_objfile_sections, (void *) objfile);
> -  objfile->sections = obstack_finish (&objfile->objfile_obstack);
> -  objfile->sections_end = objfile->sections + (size_t) objfile->sections_end;
> +
> +  /* See gdb_bfd_section_index.  */
> +  add_to_objfile_sections_full (objfile->obfd, bfd_com_section_ptr, objfile, 1);
> +  add_to_objfile_sections_full (objfile->obfd, bfd_und_section_ptr, objfile, 1);
> +  add_to_objfile_sections_full (objfile->obfd, bfd_abs_section_ptr, objfile, 1);
> +  add_to_objfile_sections_full (objfile->obfd, bfd_ind_section_ptr, objfile, 1);
>   }

>
> diff --git a/gdb/symtab.c b/gdb/symtab.c
> index df9caef..0f1cd68 100644
> --- a/gdb/symtab.c
> +++ b/gdb/symtab.c
> @@ -1060,7 +1060,7 @@ fixup_section (struct general_symbol_info *ginfo,
>
>         ALL_OBJFILE_OSECTIONS (objfile, s)
>   	{
> -	  int idx = s->the_bfd_section->index;
> +	  int idx = s - objfile->sections;

The range of IDX will be [0, bfd_count_sections () + 4),

>   	  CORE_ADDR offset = ANOFFSET (objfile->section_offsets, idx);

the 'objfile->section_offsets' is allocated with objfile->num_sections 
entries, which is bfd_count_sections (), IIUC.  I can't reproduce any 
crash caused by this potential overflow here.

 > diff --git a/gdb/solib-dsbt.c b/gdb/solib-dsbt.c
 > index c41326b..586ab8e 100644
 > --- a/gdb/solib-dsbt.c
 > +++ b/gdb/solib-dsbt.c
 > @@ -1054,7 +1054,7 @@ dsbt_relocate_main_executable (void)
 >         int osect_idx;
 >         int seg;
 >
 > -      osect_idx = osect->the_bfd_section->index;
 > +      osect_idx = osect - symfile_objfile->sections;
 >
 >         /* Current address of section.  */
 >         addr = obj_section_addr (osect);

The code here can abbreviated like this,

   new_offsets = xcalloc (symfile_objfile->num_sections,
			 sizeof (struct section_offsets));

   ALL_OBJFILE_OSECTIONS (symfile_objfile, osect)
     {
       osect_idx = osect - symfile_objfile->sections;

               new_offsets->offsets[osect_idx] =

I can see a crash here.  We allocate NEW_OFFSETS for 
'symfile_objfile->num_sections' entries, but ALL_OBJFILE_OSECTIONS will 
iterate 'symfile_objfile->num_sections + 4'.  These four special 
sections are not used in tic6x, so probably I can skip them when setting 
'new_offsets'.  Before I post a patch for the crash, I'd like to know 
how do you think about this problem overall.

-- 
Yao (齐尧)