screen: Crashes the kernel!

[Gregory N. Bond]

Environment: Sun 3/260, SunOs 3.5Export, Screen v 2.0a 19-oct-88

The screen program recently posted to comp.sources.unix uses unix
domain sockets to communicate.  This latest version also has a feature 
that allows you to detatch a login session and re-attatch it at
another terminal.

However, if you exit from screen by exiting the shells in turn,
rather than by using ^A^\, then there is a screen process left
running and connected to the socket $HOME/.screen/`hostname`.`tty`.
The next time you log in (I think it has to be straight away), you
get a message saying there is a screen program running, use "screen
-r" to reattach.  If you the do a screen -r, it hangs.  If you then
attempt to kill the screen, using ^\, the kernel panics with a kernel
mode bus error.  This happens frequently (i.e. twice in 10 minutes of
my playing around), but not exactly every time.  Naturally, as this is
a live system, I am loathe to do too much experimenting!

This appears to be the old berkeley "kill process in unix domain
accept" bug, about which I have heard.  Can anyone point out the exact
circumstances this bug will bite,  if it is known to exist in SunOs
3.5Export, and what the standard workaround is.  I would DEARLY love to
hear from anyone who has fixed this bug, as some not-quite
sophisticated users are bound to find this bug one day!

BTW, I *LOVE* this program, and use it all the time.  And the new
version is much more robust than the old version (1.1i), particularly
in the handling of complex curses-based programs (where the old
version lost characters and gave garbage screens after a few minutes
of heavy use).

Greg.
-- 
Gregory Bond, Burdett Buckeridge & Young Ltd, Melbourne, Australia
Internet: gnb at melba.bby.oz.au    non-MX: gnb%melba.bby.oz at uunet.uu.net
Uucp: {uunet,mnetor,pyramid,ubc-vision,ukc,mcvax,...}!munnari!melba.bby.oz!gnb