Re: [PATCH 3/3] module: Improve module __ro_after_init handling

From: Josh Poimboeuf
Date: Fri Jun 14 2019 - 14:28:41 EST


On Fri, Jun 14, 2019 at 04:14:53PM +0200, Petr Mladek wrote:
> > -void __module_enable_ro(const struct module *mod, bool after_init)
> > +static void __module_enable_ro(const struct module *mod)
> > {
> > if (!rodata_enabled)
> > return;
> > @@ -1973,15 +1973,15 @@ void __module_enable_ro(const struct module *mod, bool after_init)
> >
> > frob_rodata(&mod->init_layout, set_memory_ro);
> >
> > - if (after_init)
> > + if (mod->state == MODULE_STATE_LIVE)
> > frob_ro_after_init(&mod->core_layout, set_memory_ro);
>
> This works only now because __module_enable_ro() is called only from
> three locations (klp_init_object_loaded(), complete_formation(),
> and do_init_module(). And they all are called in a well defined order
> from load_module().
>
> Only the final call in do_init_module() should touch the after_init
> section.
>
> IMHO, the most clean solutiuon would be to call frob_ro_after_init()
> from extra __module_after_init_enable_ro() or so. This should be
> called only from the single place.

Agreed, that would be better. I'll be gone for a week but I'll make
these changes when I get back. Thanks.

--
Josh