Re: [musl] Protect pthreads' mutexes against use-after-destroy

[Rich Felker <dalias@libc.org>]

On Sun, Jan 21, 2024 at 12:06:14PM +0000, julien.voisin@dustri.org wrote:
> > Draft attached in case anyone wants to play with it. This could
> > probably be something we could consider to adopt.
> 
> Couldn't a macro like `#define mutex_is_destroyed (!(m->_m_type & 8) && (m->_m_lock == 0x3fffffff)` be
> used instead? Or at least named constants instead of `8` and `0x3fffffff`..

Maybe something like that, but that's a change I'd like to make in a
consistent uniform way for all of the uses of magic numbers in the
mutex implementation. Just introducing it in a single place like this
doesn't really help readability; in some ways it makes it less
readable since you can't see how it's interacting with the other
tests.

If doing it, I think it would probably make more sense not to have
that predicate macro, but instead something like:

	if (own == M_UNRECOVERABLE && !(m->_m_type & MT_ROBUST))

because seeing the individual parts is relevant to understanding:

> Also, the code-style seems inconsistent:
> 
> ```
> +       if (own == 0x3fffffff) {
> +               /* Catch use-after-destroy */
> +               if (!(type & 8)) a_crash();                                                                                                                                                      
> +               return ENOTRECOVERABLE;
> +       }
> ```
> 
> vs 
> 
> ```
> +                       /* Catch use-after-destroy */
> +                       if (own == 0x3fffffff && !(type & 8)) a_crash();
>                         return EPERM;
> ```
> 
> Both are the same check, yet only one has both conditions in a single `if`.

These are decision trees on what to do in exceptional cases. The aim
is not to present a consistent "style" between two functions that do
different things and have different decision trees for how to act.

Rich