non-killable processes

[Markus Wild]

I'm having a problem with a process that keeps sleeping in a non-killable
state.  This happens roughly once every 50 times the program is run, and
it's quite a problem, because since the process can't be killed, the
filesystem on which it is running can't be unmounted, and so the system
can't be shut down as cleanly as necessary (work for fsck on next run).
 
Examining the pcb with 'crash' showed that the process was hanging in
the close() system call. Now my release-notes say, that ufs is buggy, but
they only say that from time to time it panics the kernel (which it
really did on a few rare occasions;-)), they don't mention a problem like this.
What I'd like to know is:  is there a way around this problem?  Can I do
something before calling close(2) to get around this bug?  Does it have
something to do with mmap()s? Does an fsync() help?
 
If it can help, this is the relevant information I gathered together with 
/bin/ps and /usr/ucb/ps:
 
 F S PID PPID CP PRI NI     ADDR  SZ  RSS    WCHAN S TT        TIME COMMAND
18 S 361    1  0  98 20 40170e50 302  129 4005638c S term/con5 0:02 gcc-as
 
This is quite a nasty problem, and I wouldn't mind if somehow I could get
rid of it.  (No, don't tell me to use the s5 fs, I can't stand the 14
character limit, and it's really ways slower than ufs.)
 
-Markus


-- 
Markus M. Wild    - mwild at iiic.ethz.ch  |  wild at nessie.cs.id.ethz.ch
--
Still looking for a REAL debugger for System V Release 4 ...