Messages in this thread | | | Date | Thu, 24 Jul 1997 16:35:29 -0400 | From | James Mastros <> | Subject | Re: Reserved PID's |
| |
At 12:17 PM 7/24/97 -0700, B. James Phillippe wrote: >Greetings, > [...] > I've considered writing a simple PID reservation patch that would >reserve a number of PID's (controlled on the fly by echo'ing an int to >/proc/sys/kernel/reserved-pids) for root use only. It sounds easy enough >to implement. If all non-reserved PID's are in use and a process attempts >to be created under a UID >0, kernel would return EGAIN. This would allow >only root-owned processes to fill the process table. In other cases, the >superuser would be able to log in and fix the problem. > But, I find that in most cases the offending programs are root owned anyway. Init and inetd normally spawn their processes as root-owned, no? (I'm not at my Linux box now, I'll check later if I remember.)
> I also considered reserving PID's for processes with a controlling >tty, but I don't think I'll be able to figure out how to do that. I don't think processes get a controlling terminal until after they completely initialize, so that wouldn't be possible. > > Anyone have thoughts on this? Perhaps a patch to SIGTERM the most recently 10 started (non-root if possible, otherwise kill off root) programs when the process table fills up? (10 so that it doesn't go into this too often). Even better, if there are more than one of any given process running, and the process table fills up, SIGTERM all but one of them.
> >-bp
-=- James Mastros
| |