Subject: unkillable mt(1) process
To: None <email@example.com>
From: Geert Hendrickx <firstname.lastname@example.org>
Date: 05/14/2007 09:41:59
I've found references to this problem in the archives, and discussions on
how this should theoretically be solves, but no practical solution or
workaround except "reboot".
I have a cron script that performs nightly backups to tape with dump(8).
It worked fine until our filesystems exceeded the size of a single tape.
The backup is aborted because there's no tty, but some process seems to
keep /dev/nrst0 open (fstat doesn't show any, but "mt status" says
"/dev/nrst0: Device busy". The next morning, an operator ejects the
tape, inserts a new one, but from that moment on, a hanging "mt" process
is unkillable, and the (new) tape is not accessible until I reboot the
machine. The only other useful suggestion I found is "scsictl scsibus0
reset"  but that doesn't help either.
Any other ideas? The system is NetBSD 3.0.2.