On Tue, 9 Nov 2021 at 20:42, Peter Zijlstra <[email protected]> wrote:
>
> On Tue, Nov 09, 2021 at 07:54:23PM +0100, Ard Biesheuvel wrote:
> > On Tue, 9 Nov 2021 at 19:53, Peter Zijlstra <[email protected]> wrote:
> > >
> > > On Tue, Nov 09, 2021 at 05:45:45PM +0100, Ard Biesheuvel wrote:
> > > > @@ -196,13 +190,21 @@ extern long __static_call_return0(void);
> > > > EXPORT_SYMBOL_GPL(STATIC_CALL_KEY(name)); \
> > > > EXPORT_SYMBOL_GPL(STATIC_CALL_TRAMP(name))
> > > >
> > > > +#define EXPORT_STATIC_CALL_GETKEY_HELPER(name) \
> > > > + struct static_call_key *STATIC_CALL_GETKEY(name)(void) { \
> > > > + BUG_ON(!core_kernel_text( \
> > > > + (unsigned long)__builtin_return_address(0))); \
> > > > + return &STATIC_CALL_KEY(name); \
> > > > + } \
> > > > + EXPORT_SYMBOL_GPL(STATIC_CALL_GETKEY(name))
> > >
> > > So if I were a nevarious module, I would look up the above symbol from
> > > kallsyms (it is exported and easily obtainable) and then simply
> > > read the text to discover the key address and we're in business.
> > >
> >
> > Yeah I realised that. So would you prefer to have a
> > .static_call_tramp_key section in each module and look up the keys in
> > the module loader?
>
> Think so... do you actually have a need for this?
Well, not exactly. But the XOR patch I sent earlier today does cover
this use case, and would be broken if we implemented inline static
calls on arm64.
So the easy fix is to allow EXPORT_STATIC_CALL_RO() only from the core
kernel, and require EXPORT_STATIC_CALL_RW() otherwise. That way, this
can never end up being used in a way that breaks at runtime. Would you
prefer doing that instead?