svn commit: trunk/busybox/init

[Rob Landley]

On Saturday 21 January 2006 23:22, Rob Landley wrote:
> On Saturday 21 January 2006 21:49, Erik Andersen wrote:
> > On Sat Jan 21, 2006 at 06:41:54PM -0800, landley at busybox.net wrote:
> > > Author: landley
> > > Date: 2006-01-21 18:41:51 -0800 (Sat, 21 Jan 2006)
> > > New Revision: 13489
> > >
> > > Log:
> > > The whole "init is sometimes pid 3" thing is silly.  Init is pid 1,
> > > anything else is a kernel bug.  Both 2.4 and 2.6 should get this right
> > > now.  This
> >
> > When run within an initrd, the pid of init is most definitely
> > not going to be pid 1.  That is annoying, but not a bug, and we
> > certainly do have to handle it despite wishful thinking to the
> > contrary,
> >
> >  -Erik
>
> I know that used to be a problem, but I thought that only applied to the
> 2.2 kernel.
>
> I'll test it...
>
> Rob

Nope, you're right.  I was testing with initramfs, not initrd.  For initramfs 
forking off /init is done via execve(init_filename, argv_init, envp_init) (in 
run_init_process()), but for initrd handle_initrd() forks off 
kernel_thread(do_linuxrc, "/linuxrc", SIGCHLD); and then waits for it to 
exit.  (Still, in 2.6.15.)

Sigh.  (I need to update the initramfs docs.)

Lemme think about this.  I'll probably just revert the patch, but I want to 
see if I can think of a better solution...

Rob
-- 
Steve Ballmer: Innovation!  Inigo Montoya: You keep using that word.
I do not think it means what you think it means.