Add OPcache restart hook

[realFlowControl]

Hey folks,

adding to #15543 which allows access to OPcache shared globals, this hook will allow observing extensions to observe the actual OPcache restart using this code snippet:

void restart_hook(zend_accel_restart_reason reason)
{
    printf("IT HAPPENED\n");
}

void call_during_minit() {
    // assume you have the `opcache_handle` available
    void (**opcache_restart_hook)(zend_accel_restart_reason reason);
    opcache_restart_hook = DL_FETCH_SYMBOL(opcache_handle, "zend_accel_schedule_restart_hook");
    // do proper error handling ;-)
    *opcache_restart_hook = restart_hook;
}

CC: @arnaud-lb

[dstogov]

If this callback is really necessary:

1. you should select a good place for it. I think it shouldn't be done under the exclusive lock.
2. instead of DL_FETCH_SYMBOL() magic, it's better to declare the callback in zend.h and zend.c.

[realFlowControl]

1. you should select a good place for it. I think it shouldn't be done under the exclusive lock.

I'd need to try my prototype usage of that hook on x86 (so far only tested on aarch64), but the pointer that I can dlsym() from zend_smm_shared_globals is readable and that's all I need. I would also argue that it is a good thing, that the memory is write protected during the hook and hooked functions can not alter anything.

2. instead of DL_FETCH_SYMBOL() magic, it's better to declare the callback in zend.h and zend.c.

It would remove the dlsym() call needed for someone using the hook, but it would also leak OPcache things into Zend, where it should not be IMHO. So far, there are no source code references from /Zend to /ext/opcache and I think this is a good thing. I would argue to leave the hook where it is, just to not leak OPcache things into other code where it should not be.

[realFlowControl]

Hey there @dstogov can you have a look again? Would be great having this in 8.4