Re: [PATCH v10 00/13] arm64: Branch Target Identification support

From: Catalin Marinas
Date: Mon Mar 23 2020 - 08:21:52 EST


On Fri, Mar 20, 2020 at 05:39:46PM +0000, Szabolcs Nagy wrote:
> The 03/16/2020 16:50, Mark Brown wrote:
> > This patch series implements support for ARMv8.5-A Branch Target
> > Identification (BTI), which is a control flow integrity protection
> > feature introduced as part of the ARMv8.5-A extensions.
>
> i was playing with this and it seems the kernel does not add
> PROT_BTI to non-static executables (i.e. there is an interpreter).
>
> i thought any elf that the kernel maps would get PROT_BTI from the
> kernel. (i want to remove the mprotect in glibc when not necessary)

I haven't followed the early discussions but I think this makes sense.

> i tested by linking a hello world exe with -Wl,-z,force-bti (and
> verified that the property note is there) and expected it to crash
> (with SIGILL) when the dynamic linker jumps to _start in the exe,
> but it executed without errors (if i do the mprotect in glibc then
> i get SIGILL as expected).
>
> is this deliberate? does the kernel map static exe and dynamic
> linked exe differently?

I think the logic is in patch 5:

+int arch_elf_adjust_prot(int prot, const struct arch_elf_state *state,
+ bool has_interp, bool is_interp)
+{
+ if (is_interp != has_interp)
+ return prot;
+
+ if (!(state->flags & ARM64_ELF_BTI))
+ return prot;
+
+ if (prot & PROT_EXEC)
+ prot |= PROT_BTI;
+
+ return prot;
+}

At a quick look, for dynamic binaries we have has_interp == true and
is_interp == false. I don't know why but, either way, the above code
needs a comment with some justification.

--
Catalin