[PATCH v2] ld: Turn on --error-execstack/--error-rwx-segments

[H.J. Lu]

On Mon, Jan 29, 2024 at 6:03 AM Jan Beulich <jbeulich@suse.com> wrote:
>
> On 29.01.2024 14:03, H.J. Lu wrote:
> > On Sun, Jan 28, 2024 at 11:30 PM Jan Beulich <jbeulich@suse.com> wrote:
> >>
> >> On 26.01.2024 22:45, H.J. Lu wrote:
> >>> --- a/ld/lexsup.c
> >>> +++ b/ld/lexsup.c
> >>> @@ -1947,6 +1947,17 @@ parse_args (unsigned argc, char **argv)
> >>>        && command_line.check_section_addresses < 0)
> >>>      command_line.check_section_addresses = 0;
> >>>
> >>> +  /* Override --no-error-execstack and --no-warn-execstack and turn on
> >>> +     --error-execstack for --warn-execstack and --error-rwx-segments for
> >>> +     --warn-rwx-segments if --fatal-warnings is used.  */
> >>> +  if (config.fatal_warnings)
> >>> +    {
> >>> +      if (link_info.warn_execstack)
> >>> +     link_info.error_execstack = 1;
> >>> +      if (!link_info.no_warn_rwx_segments)
> >>> +     link_info.warn_is_error_for_rwx_segments = 1;
> >>> +    }
> >>> +
> >>>    if (export_list)
> >>>      {
> >>>        struct bfd_elf_version_expr *head = export_list->head.list;
> >>
> >> If I'm not mistaken and if the comment is properly describing things,
> >> this placement of the addition means --no-* last on the command line
> >> wouldn't be honored anymore, when later arguments ought to override
> >> earlier ones.
> >>
> >
> > Before my change, we got
> >
> > [hjl@gnu-cfl-3 ld]$ touch x.s
> > [hjl@gnu-cfl-3 ld]$ gcc -c x.s
> > [hjl@gnu-cfl-3 ld]$ ld -shared -z stack-size=0x123400 --fatal-warnings
> > --warn-execstack --no-error-execstack  x.o
> > ld: warning: x.o: missing .note.GNU-stack section implies executable stack
> > ld: NOTE: This behaviour is deprecated and will be removed in a future
> > version of the linker
> > [hjl@gnu-cfl-3 ld]$ echo $?
> > 1
> >
> > The warning is misleading since --fatal-warnings IS position independent
> > and is always fatal for ANY warning.  My patch changes it to
> >
> > [hjl@gnu-cfl-3 ld]$ ./ld-new -shared -z stack-size=0x123400
> > --fatal-warnings --warn-execstack --no-error-execstack  x.o
> > ./ld-new: error: x.o: is triggering the generation of an executable
> > stack because it does not have a .note.GNU-stack section
> > ./ld-new: failed to set dynamic section sizes: file format not recognized
> > [hjl@gnu-cfl-3 ld]$
> >
> > Any other comments to my patch?
>
> Well. To me, it is ambiguous what "--fatal-warnings --warn-execstack
> --no-error-execstack" actually is intended to mean (and that's regardless
> of --fatal-warnings position on the command line). One way of
> interpreting is your way. The other is to say that --no-error-execstack
> means "no error for anything related to executable stacks", i.e.
> --fatal-warnings not affecting the overall result (when there are no
> other warnings).

My interpretation is what --fatal-warnings implements.


--
H.J.