Re: [PATCH 2/2] arch: Reserve map_shadow_stack() syscall number for all architectures

["Edgecombe, Rick P" <rick.p.edgecombe@xxxxxxxxx>]

On Wed, 2023-09-13 at 12:18 -0700, Sohil Mehta wrote:
> On 9/11/2023 2:10 PM, Edgecombe, Rick P wrote:
> > On Mon, 2023-09-11 at 18:02 +0000, Sohil Mehta wrote:
> > > diff --git a/arch/powerpc/kernel/syscalls/syscall.tbl
> > > b/arch/powerpc/kernel/syscalls/syscall.tbl
> > > index 20e50586e8a2..2767b8a42636 100644
> > > --- a/arch/powerpc/kernel/syscalls/syscall.tbl
> > > +++ b/arch/powerpc/kernel/syscalls/syscall.tbl
> > > @@ -539,3 +539,4 @@
> > >  450    nospu   set_mempolicy_home_node         sys_set_mempolicy
> > > _hom
> > > e_node
> > >  451    common  cachestat                       sys_cachestat
> > >  452    common  fchmodat2                       sys_fchmodat2
> > > +453    common  map_shadow_stack                sys_map_shadow_st
> > > ack
> > 
> > I noticed in powerpc, the not implemented syscalls are manually
> > mapped
> > to sys_ni_syscall. It also has some special extra sys_ni_syscall()
> > implementation bits to handle both ARCH_HAS_SYSCALL_WRAPPER and
> > !ARCH_HAS_SYSCALL_WRAPPER. So wondering if it might need special
> > treatment. Did you see those parts?
> > 
> 
> Thanks for pointing this out. Powerpc seems to be unique in their
> handling of not implemented syscalls. Maybe it's because of their
> special casing of the ARCH_HAS_SYSCALL_WRAPPER.
> 
> The code below in arch/powerpc/include/asm/syscalls.h suggests to me
> that it should be safe to map map_shadow_stack() to sys_ni_syscall()
> and
> the special handling will be taken care of.
> 
> #ifndef CONFIG_ARCH_HAS_SYSCALL_WRAPPER
> long sys_ni_syscall(void);
> #else
> long sys_ni_syscall(const struct pt_regs *regs);
> #endif
> 
> I don't quite understand the underlying reasoning for it though. Do
> you
> have any additional insight into how we should handle this?
> 
> I am thinking of doing the following in the next iteration unless
> someone chimes in and says otherwise.
> 
> --- a/arch/powerpc/kernel/syscalls/syscall.tbl
> +++ b/arch/powerpc/kernel/syscalls/syscall.tbl
> @@ -539,4 +539,4 @@
>  450    nospu   set_mempolicy_home_node        
> sys_set_mempolicy_home_node
>  451    common  cachestat                       sys_cachestat
>  452    common  fchmodat2                       sys_fchmodat2
> -453    common  map_shadow_stack                sys_map_shadow_stack
> +453    common  map_shadow_stack                sys_ni_syscall

It might have something to do with that powerpc's COND_SYSCALL()
implementation only defines the struct pt_regs variety, but TBH I get a
bit lost when I get to the inline assembly symbol definitions parts and
how it all ties together.

Doing powerpc's version as sys_ni_syscall seems to be consistent at
least, and makes sense with respect to the code you quoted.