Re: ELF sections vs. segments

[Valery Ushakov <uwe%stderr.spb.ru@localhost>]

On Thu, Jan 19, 2017 at 19:02:15 +0300, Valery Ushakov wrote:

> The question I have can probably be classified as a platonic
> essentialism problem, but it has some practical conseqeunces too.
> 
> I've noticed that netbsd native elf signature check looks for a netbsd
> note _section_.
> 
>   sys/kern/exec_elf.c
>   - netbsd_elf_signature
> 
> 
> OTOH, linunx compat looks for a linux note _segment_.
> 
>   sys/compat/linux/common/linux_exec_elf32.c
>   - ELFNAME2(linux,signature)
> 
> 
> Normally, it doesn't matter, as there's a note segment for each note
> section.  However I'm playing with a forth system that cross-compiles
> and writes out its own ELF image and that standalone image only has
> segments, but no sections.
> 
> I'm inclined to think that a standalone ELF binary should be able to
> get away with this kind of arrangement.  Sections are for linkers,
> segments are for loaders.  The binary is intended to be loaded, so it
> provides info about its segments and that should be enough.
> 
> Comments?  Shall we fix the check to look for note segments?

Studying the history of exec_elf.c I can see:

revision 1.15
date: 2010-03-20 01:08:13 +0300;  author: christos;  state: Exp;  lines: +74 -50;
[...]
- Re-do note parsing code to read the section headers instead of the program
  headers because the new binutils merge all the note sections in one program
  header. This fixes all the pax note parsing which has been broken for all
  binaries built with the new binutils.

I guess that's still a concern since x86 and e.g. sh3 do have single
note segments for two note sections (hmm, OTOH on my -6 macppc there
are two segments).

-uwe