Re: [PATCH v2] module: Harden STRICT_MODULE_RWX

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Apr 08, 2020 at 05:32:49PM +0200, Jessica Yu wrote:
> +++ Peter Zijlstra [03/04/20 19:13 +0200]:
> > 
> > We're very close to enforcing W^X memory, refuse to load modules that
> > violate this principle per construction.
> > 
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > ---
> > kernel/module.c |   25 +++++++++++++++++++++++++
> > 1 file changed, 25 insertions(+)
> > 
> > --- a/kernel/module.c
> > +++ b/kernel/module.c
> > @@ -2044,9 +2044,29 @@ static void module_enable_x(const struct
> > 	frob_text(&mod->core_layout, set_memory_x);
> > 	frob_text(&mod->init_layout, set_memory_x);
> > }
> > +
> > +static int module_enforce_rwx_sections(Elf_Ehdr *hdr, Elf_Shdr *sechdrs,
> > +				       char *secstrings, struct module *mod)
> > +{
> > +	const unsigned long shf_wx = SHF_WRITE|SHF_EXECINSTR;
> > +	int i;
> > +
> > +	for (i = 0; i < hdr->e_shnum; i++) {
> > +		if ((sechdrs[i].sh_flags & shf_wx) == shf_wx)
> > +			return -ENOEXEC;
> > +	}
> > +
> > +	return 0;
> > +}
> 
> Just to clarify, did we want to enforce this only when
> CONFIG_STRICT_MODULE_RWX=y?  Because here it's still in the
> CONFIG_ARCH_HAS_STRICT_MODULE_RWX block.
> 
> Unfortunately, when we add module_enforce_rwx_sections() in the
> CONFIG_STRICT_MODULE_RWX block, we'll need two empty stubs, one for
> !CONFIG_ARCH_HAS_STRICT_MODULE_RWX and one for !CONFIG_STRICT_MODULE_RWX.
> 
> This is because the CONFIG_STRICT_MODULE_RWX block is currently nested
> within ARCH_HAS_STRICT_MODULE_RWX :/

Yeah, so the primary reason it's under that ARCH_HAS thing is indeed the
mess and the extra stub required (I'm a lazy sod at times).

I then rationalized this decision to myself that having it under
ARCH_HAS give a more consistent module loading behaviour.

But I really don't care too much, my most my .config's have
CONFIG_MODULE=n, and the ones that do not very much have the STRICT_RWX
set.

Put it where you think it's best.

Thanks!