Re: killing a process
In message <[🔎] 873eo25ehh.fsf@tiamat.datasync.com>, writes:
...
> This is very wrong. No matter what happens, a kill -9 should
> kill the process. Are there any other signs of a kernel malfunction?
> (You may want to reboot). If it happens again, could you supply the
> versions of the kernel and of mirror?
I don't think that this is necessarily wrong.
According to Maurice J. Bach in "Design of the Unix Operating System": `The
kernel checks for receipt of a signal when a process is about to return
from kernel mode to user mode and when it enters or leaves the sleep state
at a suitably low scheduling priority. The kernel handles signals only
when a process returns from kernel mode to user mode. Thus, a signal does
not have an instant effect on a process running in kernel mode.' (p. 201)
If a process is sleeping on some low-level I/O event that never occurs, it
will never wake up and the kernel will never check for a signal on it.
Therefore it will never die, no matter how often you run `kill -9' on it.
(This assumes that Linux is implemented in the same way as "standard Unix",
whatever that is.) I have frequently had this problem with programs
that access cartridge tapes, on Interactive Unix and on SCO. The only
solution was to reboot, because the process never woke up, but meanwhile
it had the tape device open so that no other process could access it.
If this is a correct diagnosis of the problem, the solution would have to be
to develop some system call to force the process to wake up to be killed.
I don't know whether any such call exists, but altering the kill() system
call is probably a change with fairly major implications for the kernel.
--
Oliver Elphick Oliver.Elphick@lfix.co.uk
Isle of Wight http://lfix.co.uk/oliver
Make it idiot-proof, and someone will breed a better idiot.
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-user-request@lists.debian.org .
Trouble? e-mail to templin@bucknell.edu .
Reply to: