Re: [PATCH] ld: provide __ehdr_start magic symbol

[Alan Modra <amodra at gmail dot com>]

On Tue, Jun 19, 2012 at 02:28:45PM -0700, Roland McGrath wrote:
> +  if (hdrs_segment != NULL && link_info != NULL)
> +    {
> +      /* There is a segment that contains both the file headers and the
> +	 program headers, so provide a symbol __ehdr_start pointing there.
> +	 A program can use this to examine itself robustly.  */
> +
> +      struct elf_link_hash_entry *hash
> +	= elf_link_hash_lookup (elf_hash_table (link_info), "__ehdr_start",
> +				FALSE, FALSE, TRUE);
> +      if (hash != NULL
> +	  && (hash->root.type == bfd_link_hash_new
> +	      || hash->root.type == bfd_link_hash_undefined
> +	      || hash->root.type == bfd_link_hash_undefweak
> +	      || hash->root.type == bfd_link_hash_common))

For other linker PROVIDEd symbols, we don't test for undefweak.  ie. a
weak reference stays undefined.

> +	{
> +	  hash->root.type = bfd_link_hash_defined;
> +	  hash->root.u.def.value = filehdr_vaddr;
> +	  hash->root.u.def.section = bfd_abs_section_ptr;

I've just recently changed a lot of places where the linker defines
symbols that hold a virtual address to *not* use absolute symbols.
Absolute syms shouldn't be relocated by ld.so.  They are currently,
but if this is ever fixed your __ehdr_start sym won't relocate when a
shared lib or pie is loaded at some address other than the one it was
linked at.

Maybe
	  if (hdrs_segment->count != 0)
	    {
	      asection *s = hdrs_segment->sections[0];
	      hash->root.u.def.value = filehdr_vaddr - s->vma;
	      hash->root.u.def.section = s;
	    }
and look through other elf_segment_map entries for the lowest vma
section if you happen to have ehdr in a segment by itself, falling
back to bfd_abs_section_ptr if there are no sections.

> +	  hash->def_regular = 1;
> +	  hash->non_elf = 0;
> +	}
> +    }
> +

-- 
Alan Modra
Australia Development Lab, IBM