Re: RFC: Can static executables contain relocations against symbols ?

[Alan Modra <amodra@gmail.com>]

On Wed, Mar 29, 2023 at 03:39:46PM +0100, Nick Clifton via Binutils wrote:
>   The issue appears to be that static-pie executables are being created
>   with a .rela.got section that has an sh_link field set to point to the
>   .symtab section.

Nick, the patch you posted is for code with this comment:
	case SHT_REL:
	case SHT_RELA:
	  /* A reloc section which we are treating as a normal BFD
	     section.  sh_link is the section index of the symbol
	     table.  sh_info is the section index of the section to
	     which the relocation entries apply.  We assume that an
	     allocated reloc section uses the dynamic symbol table
	     if there is one.  Otherwise we guess the normal symbol
	     table.  FIXME: How can we be sure?  */

ie. we are way off into the weeds already.  However, I think the
comment is wrongly placed, because it looks to me like we handle the
usual SHT_REL/SHT_RELA sections here too.

So is the .rela.got (or .rela.plt) section one of the reloc sections
created in bfd_section_from_shdr elf.c:2382?  If we need to handle
this sort of reloc section better (and I doubt we do need to) then we
should do something about tracking the original sh_link and sh_info
fields when creating them,  Hmm, maybe they are still available, and
can be mapped to output sections?

Anyway, does the following fix the particular problem you are seeing?

diff --git a/bfd/elf.c b/bfd/elf.c
index 45e53640e8f..027d0143735 100644
--- a/bfd/elf.c
+++ b/bfd/elf.c
@@ -3870,21 +3870,23 @@ assign_section_numbers (bfd *abfd, struct bfd_link_info *link_info)
 	{
 	case SHT_REL:
 	case SHT_RELA:
-	  /* A reloc section which we are treating as a normal BFD
-	     section.  sh_link is the section index of the symbol
-	     table.  sh_info is the section index of the section to
-	     which the relocation entries apply.  We assume that an
-	     allocated reloc section uses the dynamic symbol table
-	     if there is one.  Otherwise we guess the normal symbol
-	     table.  FIXME: How can we be sure?  */
-	  if (d->this_hdr.sh_link == 0 && (sec->flags & SEC_ALLOC) != 0)
-	    {
-	      s = bfd_get_section_by_name (abfd, ".dynsym");
-	      if (s != NULL)
-		d->this_hdr.sh_link = elf_section_data (s)->this_idx;
-	    }
+	  /* sh_link is the section index of the symbol table.
+	     sh_info is the section index of the section to which the
+	     relocation entries apply.  */
 	  if (d->this_hdr.sh_link == 0)
-	    d->this_hdr.sh_link = elf_onesymtab (abfd);
+	    {
+	      /* FIXME maybe: If this is a reloc section which we are
+		 treating as a normal section then we likely should
+		 not be assuming its sh_link is .dynsym or .symtab.  */
+	      if ((sec->flags & SEC_ALLOC) != 0)
+		{
+		  s = bfd_get_section_by_name (abfd, ".dynsym");
+		  if (s != NULL)
+		    d->this_hdr.sh_link = elf_section_data (s)->this_idx;
+		}
+	      else
+		d->this_hdr.sh_link = elf_onesymtab (abfd);
+	    }
 
 	  s = elf_get_reloc_section (sec);
 	  if (s != NULL)

-- 
Alan Modra
Australia Development Lab, IBM