stevesu@azure.UUCP (Steve Summit) (11/22/83)
It appears that it is impossible to nohup a process which was
invoked by a remote shell using the 4.1(a, maybe) rlogin command.
The rlogind program (which is the daemon that runs on the remote
host and is the parent of the remote shell) invokes a routine
called cleanup() when the remote shell terminates. Among other
things, cleanup() does a kill(0, SIGKILL). Since SIGKILL "cannot
be caught or ignored," all of your remote processes get
terminated at that time, even if you nohup'ed them and put them
in the background. If the last thing you did before you logged
out was an lpr, the lpr daemon, which had put itself in the
background and ignored SIGHUP, would die, too. Furthermore, it
wouldn't have a chance to remove its lock file, so the lpr would
be hung until the lock file was manually removed and the daemon
restarted.
Under 4.2, cleanup() does a vhangup() and a shutdown() before the
kill(). vhangup() and shutdown() apparently hang up and shut
down the login and socket connection (I haven't finished reading
the 4.2 documentation yet). vhanghup() sends everybody a SIGHUP,
which is sensible, but unless vhangup() or shutdown() also
diddles with the process group, the kill() will still kill
everything.
My question is this: does anyone know why cleanup() does a
kill(0, SIGKILL) instead of a kill(0, SIGHUP)? We haven't
converted to 4.2 yet, so I don't know if the change is needed
there, but am I likely to break something in the 4.1 rlogind if I
change the SIGKILL to a SIGHUP?
Steve Summit
Tektronix, Inc.
tektronix!tekmdp!stevesu
P.S. Please DON'T remind me that if I was using csh my
background processes would be in a separate process group and
hence immune to any kill (0, ...). Some of us still like, and
use, the Bourne shell.