Re: [RFC/PATCH] Implement new PTRACE_EVENT_SYSCALL_{ENTER,EXIT}

[Pedro Alves <palves@redhat.com>]

On 01/07/2014 03:30 PM, Oleg Nesterov wrote:
> 
> If we add the new API, perhaps we should change ptrace_resume ?
> I mean,
> 
> 	--- x/kernel/ptrace.c
> 	+++ x/kernel/ptrace.c
> 	@@ -723,7 +723,9 @@ static int ptrace_resume(struct task_str
> 		if (!valid_signal(data))
> 			return -EIO;
> 	 
> 	-	if (request == PTRACE_SYSCALL)
> 	+	if (request == PTRACE_SYSCALL ||
> 	+	    ptrace_event_enabled(PTRACE_EVENT_SYSCALL_ENTER) ||
> 	+	    ptrace_event_enabled(PTRACE_EVENT_SYSCALL_EXIT))
> 			set_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
> 		else
> 			clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
> 
> 
> This way PTRACE_O_SYSCALL_* will work like other ptrace options which
> ask to report an event.

+10^6.  With PTRACE_SYSCALL/sysgood, we don't have a way to trace
syscalls when single-stepping, which isn't much of a problem for
strace, but of course is for GDB.  That is one of the things the
new API should definitely sort out.

-- 
Pedro Alves