Re: [PATCH] elf: Try not pointing empty PT_LOAD segment's offset past EOF

[Bálint Réczey <balint at balintreczey dot hu>]

Hi,

Florian Weimer <fweimer@redhat.com> ezt írta (időpont: 2019. dec. 10.,
K, 14:23):
>
> * Alan Modra:
>
> > On Sat, Dec 07, 2019 at 05:22:13PM +0100, Bálint Réczey wrote:
> >> diff --git a/bfd/elf.c b/bfd/elf.c
> >> index 1aa2603ee8..e1a9a02eec 100644
> >> --- a/bfd/elf.c
> >> +++ b/bfd/elf.c
> >> @@ -5752,7 +5752,12 @@ assign_file_positions_for_load_sections (bfd *abfd,
> >>        || (p->p_type == PT_NOTE && bfd_get_format (abfd) == bfd_core))
> >>      {
> >>        if (!m->includes_filehdr && !m->includes_phdrs)
> >> -        p->p_offset = off;
> >> +        if (no_contents)
> >> +          /* Try avoiding pointing past the EOF with this empty segment's
> >> +             p_offset. */
> >> +          p->p_offset = p->p_offset % maxpagesize;
> >> +        else
> >> +          p->p_offset = off;
> >>        else
> >>          {
> >>            file_ptr adjust;
> >
> > How did you test this patch?  I suspect you are just leaving p_offset
> > at zero and therefore will cause failures on glibc systems.
>
> I think glibc requires ordering by increasing virtual address, per the
> specification.  That does not seem to change here.
>
> Testing on glibc is of course recommended, though.  It also needs to be
> tested with various kernel versions.

For the record I've tested the rebuilt gzip binary on Ubuntu's
5.0.0-37-generic and on 5.3.0-24-generic.

Cheers,
Balint