robm@ataraxia.Berkeley.EDU (Rob McNicholas) (05/08/91)
We do backups by rsh'ing rdump commands to our workstations from a DECstation 3100 with an Exabyte 8mm drive. However, I find that these backups hang occasionally. If I kill off the rsh process, all processes appear to die, except netstat shows that there is still a connection from the remote machine. (In this example, bsim is the name of the remote machine, and ataraxia is the name of the machine with the tape drive). From ataraxia, netstat shows: tcp 0 0 ataraxia.shell bsim.1019 CLOSE_WAIT From bsim, it shows: tcp 0 0 bsim.1019 ataraxia.shell FIN_WAIT_2 ataraxia has a continuous load of 1, and /etc/rmt is still running. Any attempt to access the tape drive reports "/dev/nrmt0h: device busy". Attempts to kill the /etc/rmt process are fruitless. /etc/rmt shows as: root 25629 0.0 0.2 128 40 ? D 0:01 /etc/rmt Can I tell rmt to close the tape device without writing a custom program? Any other suggestions? Many thanks in advance, -Rob -- Rob McNicholas Computer Systems Support Group, U.C. Berkeley robm@janus.berkeley.edu ....!ucbvax!janus!robm Home: 415/339-1514 Work: 415/642-8633
litwack@dccs.upenn.edu (Mark Litwack) (05/25/91)
There is no way to get /etc/rmt out of the I/O wait once it is stuck (or dump for that matter, if you happen to be using it locally). A patch is available from DEC to fix processes that are hung waiting for I/O on SCSI tape drives. The patch contains mscp_tape.o, scsi.o, and scsi_asc.o. mscp_tape.o fixes the hang problem, scsi.o fixes a kernel panic, and scsi_asc.o improves SCSI tape performance for the DS5000. Overall, a good set of patches to have. -mark