[systemd-devel] BTI interaction between seccomp filters in systemd and glibc mprotect calls, causing service failures

[Szabolcs Nagy]

The 10/22/2020 10:31, Catalin Marinas wrote:
> IIUC, the problem is with the main executable which is mapped by the
> kernel without PROT_BTI. The dynamic loader wants to set PROT_BTI but
> does not have the original file descriptor to be able to remap. Its only
> choice is mprotect() and this fails because of the MDWX policy.
> 
> Not sure whether the kernel has the right information but could it map
> the main executable with PROT_BTI if the corresponding PT_GNU_PROPERTY
> is found? The current ABI states it only sets PROT_BTI for the
> interpreter who'd be responsible for setting the PROT_BTI on the main
> executable. I can't tell whether it would break anything but it's worth
> a try:

i think it would work, but now i can't easily
tell from the libc if i have to do the mprotect
on the main exe or not.

i guess i can just always mprotect and ignore
the failure?

> 
> diff --git a/arch/arm64/kernel/process.c b/arch/arm64/kernel/process.c
> index 4784011cecac..0a08fb9133e8 100644
> --- a/arch/arm64/kernel/process.c
> +++ b/arch/arm64/kernel/process.c
> @@ -730,14 +730,6 @@ asmlinkage void __sched arm64_preempt_schedule_irq(void)
>  int arch_elf_adjust_prot(int prot, const struct arch_elf_state *state,
>  			 bool has_interp, bool is_interp)
>  {
> -	/*
> -	 * For dynamically linked executables the interpreter is
> -	 * responsible for setting PROT_BTI on everything except
> -	 * itself.
> -	 */
> -	if (is_interp != has_interp)
> -		return prot;
> -
>  	if (!(state->flags & ARM64_ELF_BTI))
>  		return prot;
>  
> 
> -- 
> Catalin

--