Re: [PATCH 3/7] ptrace: move SIGTRAP on exec(2) logic to ptrace_event()

[Oleg Nesterov <oleg@redhat.com>]

On 06/17, Tejun Heo wrote:
>
> Move SIGTRAP on exec(2) logic from tracehook_report_exec() to
> ptrace_event().  This is part of changes to make ptrace_event()
> smarter and handle ptrace event related details in one place.

I am going to apply all patches except 6/7, but to be honest this
one looks a bit ugly to me.

> -static inline int ptrace_event(int event, unsigned long message)
> +static inline void ptrace_event(int event, unsigned long message)
>  {
> -	if (likely(!ptrace_event_enabled(current, event)))
> -		return false;
> -	current->ptrace_message = message;
> -	ptrace_notify((event << 8) | SIGTRAP);
> -	return true;
> +	if (unlikely(ptrace_event_enabled(current, event))) {
> +		current->ptrace_message = message;
> +		ptrace_notify((event << 8) | SIGTRAP);
> +	} else if (event == PTRACE_EVENT_EXEC && unlikely(current->ptrace)) {
> +		/* legacy EXEC report via SIGTRAP */
> +		send_sig(SIGTRAP, current, 0);
> +	}

Why does this make ptrace_event() smarter?

OK, tracehooks should die. But why should we move this special case
into ptrace_event? Say, a simple

	static inline void ptrace_exec_event(...)
	{
		if (!ptrace_event_enabled(PTRACE_EVENT_EXEC))
			send_sig(SIGTRAP, current, 0);
		else
			ptrace_event(PTRACE_EVENT_EXEC);
	}

in fs/exec.c looks a bit better to me.

Oleg.