[musl] _chk Variants

[musl] _chk Variants

[Mike Granby]

I have been trying to run a GCC cross-compiler within Alpine, and hitting an issue with musl not providing the _chk variants of various APIs. For example, glibc provides __printf_chk which takes an additional argument and optionally performs various security checks before proceeding. I know the official way forward is to rebuild the compiler with the option to disable these calls, but rebuilding the GCC toolchain is not for the faint of heart, and I have thus been loath to go down that path. Instead, I developed a patch for the Alpine's musl APK that implemented the _chk variants as mere wrappers for the underlying APIs, thus allowing GCC to operate, albeit without the extra security that a full implementation would provide. This met my immediate need, but I am interested as to the view of the community in either implementing the _chk variants for real, or in providing the wrappers to enable better binary compatibility.

Re: [musl] _chk Variants

[alice]

On Mon Jul 11, 2022 at 10:11 PM CEST, Mike Granby wrote:
> I have been trying to run a GCC cross-compiler within Alpine, and
> hitting an issue with musl not providing the _chk variants of various
> APIs. For example, glibc provides __printf_chk which takes an
> additional argument and optionally performs various security checks
> before proceeding. I know the official way forward is to rebuild the
> compiler with the option to disable these calls, but rebuilding the
> GCC toolchain is not for the faint of heart, and I have thus been
> loath to go down that path.
have you tried using a cross toolchain from https://musl.cc ? the
underlying musl-cross-make[0] system used to make them is very easy to
use, and can generate you a cross toolchain after copying and editing a
config.mak with roughly 3 lines of text (TARGET, some cflags, etc..).
[0]: https://github.com/richfelker/musl-cross-make
> Instead, I developed a patch for the
> Alpine's musl APK that implemented the _chk variants as mere wrappers
> for the underlying APIs, thus allowing GCC to operate, albeit without
> the extra security that a full implementation would provide. This met
> my immediate need, but I am interested as to the view of the community
> in either implementing the _chk variants for real, or in providing the
> wrappers to enable better binary compatibility.