ELF NT_GNU_PROPERTY_TYPE_0 Questions

ELF NT_GNU_PROPERTY_TYPE_0 Questions

[Dave Martin]

Hi there,

Can you clarify a couple of points about the SysV ABI Linux
Extensions [1] for me?

1) Can there be more than one NT_GNU_PROPERTY_TYPE_0 note in a valid
ELF file?  I think the answer should be "no".

2) Is is permissible for an ELF ET_EXEC or ET_DYN file that contains
an NT_GNU_PROPERTY_TYPE_0 property not to have a PT_GNU_PROPERTY phdrs
entry mapping it?  Except for historical usage by RedHat (which
apparently can be worked round in userspace) it seems reasonable for
the answer to be "no", at least for Linux.

3) Is it permissible for the PT_GNU_PROPERTY phdr (if present) to
map anything other than precisely one NT_GNU_PROPERTY_TYPE_0
note?  I think the answer should be "no".

4) Is an NT_GNU_PROPERTY_TYPE_0 note allowed to contain two or more
properties with the same pr_type?  I think the answer should be "no".

5) What's the rationale for sorting the properties by pr_type?  I can
see this would make it easier for the linker to merge
NT_GNU_PROPERTY_TYPE_0 notes from different files, but I'm wondering
whether the kernel really needs to enforce the ordering when loading
an ELF.  The kernel doesn't need to merge property lists together.

6) Do you have a view on the best way to define the Elf_Prop type in
headers?  bfd elf-bfd.h seems to have elf_property, but this doesn't
follow the style of the public ELF headers.

Maybe Linux should keep its definition internal rather than publishing
it in include/uapi/elf.h...  I'm not sure.

Thanks
---Dave

[1] Linux Extensions to gABI
https://github.com/hjl-tools/linux-abi/wiki/Linux-Extensions-to-gABI

Re: ELF NT_GNU_PROPERTY_TYPE_0 Questions

[Florian Weimer]

* Dave Martin:

> Hi there,
>
> Can you clarify a couple of points about the SysV ABI Linux
> Extensions [1] for me?
>
> 1) Can there be more than one NT_GNU_PROPERTY_TYPE_0 note in a valid
> ELF file?  I think the answer should be "no".

Yes, if it has been produced by a link editors which does not about
property notes.  The ELF file still needs to be treated as valid, but
the note should be ignored.

> 2) Is is permissible for an ELF ET_EXEC or ET_DYN file that contains
> an NT_GNU_PROPERTY_TYPE_0 property not to have a PT_GNU_PROPERTY phdrs
> entry mapping it?  Except for historical usage by RedHat (which
> apparently can be worked round in userspace) it seems reasonable for
> the answer to be "no", at least for Linux.

Using an older link editor on a CET-enabled distribution will produce
such binaries, too.  The ELF file still needs to be treated as valid,
but the property date should be ignored.

> 3) Is it permissible for the PT_GNU_PROPERTY phdr (if present) to
> map anything other than precisely one NT_GNU_PROPERTY_TYPE_0
> note?  I think the answer should be "no".

Correct.  Additional processing logic in the link editor is needed.

> 4) Is an NT_GNU_PROPERTY_TYPE_0 note allowed to contain two or more
> properties with the same pr_type?  I think the answer should be "no".

H.J. needs to answer that.

> 5) What's the rationale for sorting the properties by pr_type?  I can
> see this would make it easier for the linker to merge
> NT_GNU_PROPERTY_TYPE_0 notes from different files, but I'm wondering
> whether the kernel really needs to enforce the ordering when loading
> an ELF.  The kernel doesn't need to merge property lists together.

Likewise.

> 6) Do you have a view on the best way to define the Elf_Prop type in
> headers?  bfd elf-bfd.h seems to have elf_property, but this doesn't
> follow the style of the public ELF headers.

We should put it into <elf.h> in glibc.  We don't want to rely on UAPI
headers there because this version of <elf.h> is used in many places.

Thanks,
Florian