Improving halt package interaction...

[Douglas McClendon]

Thomas M Steenholdt wrote:
> Hi all...
> 
> I'd like to bring up a suggestion that would make it easier for 
> packagers to modify the halt process, without resorting to changing 
> files owned by other packages.

...

> I think it would be very valuable to extend this functionality of 
> /etc/init.d/halt to support a /etc/halt.d/ directory in addition to the 
> halt.local script 

...

> What do you guys think about this?

I for one could immediately use this for my LiveUSB 'persistence' 
feature.  I.e. instead of booting a livecd/usb with copy-on-write rootfs 
changes going to ram, the changes go to a file on a usbstick filesystem. 
  The reason that I need* to muck with halt.local is to cleanly 
readonly-remount the usbstick fs.  (*) actually I need to do this after 
the rootfs gets readonly-remounted, which is technically the very next 
thing after the current halt.local call.  But I can envision if 
/etc/halt.d/ existed, I could do a 99(last) script, which basically 
replicates the very short amount of code that happens after the current 
halt.local call.

Unfortunately the one other thing that the feature needs to change in 
halt, is to somehow prevent the usbstick filesystem from being unmounted 
before the rootfs unmounted(/remount-ro).  Currently I do this hackishly 
by patching halt and functions (with the assumption that the protected 
filesystem is mounted on /mnt/overlayfs).  Anybody have any advice for a 
proper clean way I should submit a patch for that might be accepted?

-dmc