From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S263628AbTDGUEQ (for ); Mon, 7 Apr 2003 16:04:16 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S263636AbTDGUEQ (for ); Mon, 7 Apr 2003 16:04:16 -0400 Received: from chaos.analogic.com ([204.178.40.224]:53123 "EHLO chaos.analogic.com") by vger.kernel.org with ESMTP id S263628AbTDGUEO (for ); Mon, 7 Apr 2003 16:04:14 -0400 Date: Mon, 7 Apr 2003 16:28:29 -0400 (EDT) From: "Richard B. Johnson" X-X-Sender: root@chaos Reply-To: root@chaos.analogic.com To: Robert White cc: linux-kernel@vger.kernel.org Subject: Re: Stupid API Question. In-Reply-To: Message-ID: References: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Mon, 7 Apr 2003, Robert White wrote: > I have looked, but I have not found... > > Is there a (safe) Kernel API for setting a processes appearance on the PS > listing? (e.g. it's name etc.) > > I have seen the nonsense about clearing out your various argv[] elements and > then using strncpy to replace argv[0]. That seems neither safe nor elegant. > It also seems thread unsafe, or at least thread un-interesting, since all > the threads would be tagged with the same/last such call. > > If it exists, where might it be found? > > > If it doesn't exist... > > An API that could change the text for everything after argv[0] and filled a > pointer in the process structure for the process, would be interesting. > > I say *after* argv[0] because I think having *any* means to change the > presentation of argv[0] is a far-too-gaping security hole. > > I have no idea what the order of magnitude for this effort might be. > Managing a normally-null pointer that overrides the various facts if it is > set, and is freed back into the kernel memory pool when the process exits, > wouldn't be that tough. Intercepting whatever ps(1) (etc.) does to build > the output might be non-trivial. > > Anyway, just a thought... > > Rob. Well `ps` does the wrong thing so users need to do the wrong thing to be compatible. `ps` reads the command-line from the proc file-system. Obviously the command-line can't be changed after a program starts executing because it's something that was passed to the program before execution. That's like making somebody un-pregnant. So, hacks from hackers `setproctitle` comes to mind, replace some of the command-line, usually overwriting the environment strings, with some alien text. With the current API, (Unix) the "'correct'" way to set a process title is for everybody to agree upon an environment string like "TITLE=". Then, if you want to change the process title, you change that string. `ps` would then look for the "TITLE=" token and use the string after that. However, it's unlikely that anybody will ever do that because of: (1) Inertia. (2) NIH. PS it's trivial to modify `ps` to search for "TITLE=" and display that, as well as the command-line strings IFF "TITLE=" is not found. A new `ps` would make new/old compatibility work, but again, it's unlikely that this 20 year-old problem will ever be fixed. Cheers, Dick Johnson Penguin : Linux version 2.4.20 on an i686 machine (797.90 BogoMips). Why is the government concerned about the lunatic fringe? Think about it.