umount: cannot unmount

[Steve Faidley]

This is more of a solution, rather then a question. But I thought it worth putting on the forums.
I had SG package that failed halt as it could not unmount one of it's filesystems(LVM, not CD or NFS) tonight. I tried what I think are all the normal things; ps, lsof, fuser. None of them returned any info indicating there was a process holding up the unmount.
I was running a constant loop of fuser -ku .. ;unmount .. and got nothing.
I had started to call together a bridgeline to reboot the box when I thought of trying one more thing.
You all remember;
export UNIX95=yes;ps -efo pid,comm,args

I'll be darned, the offender's popped up like they were on Viagra.

I had saved off lsof output, but not ps -ef so I can only detail lsof output.

export UNIX95=yes;ps -efo pid,comm,args | grep ops;
14764 cat cat /plops/u90/export/pplops/DAILY/20080415/p_0415_pplops.3.np
14774 cat cat /plops/u90/export/pplops/DAILY/20080415/p_0415_pplops.5.np
4689 grep grep ops
14757 cat cat /plops/u90/export/pplops/DAILY/20080415/p_0415_pplops.2.np
4687 fuser fuser -uk /dev/vgops50/lvol4
14769 cat cat /plops/u90/export/pplops/DAILY/20080415/p_0415_pplops.4.np
14779 cat cat /plops/u90/export/pplops/DAILY/20080415/p_0415_pplops.6.np
14752 cat cat /plops/u90/export/pplops/DAILY/20080415/p_0415_pplops.1.np


lsof grep on one of the PID's;
cat 14764 oracle cwd DIR 64,0x20001 16384 6 /u00/home/oracle
cat 14764 oracle txt REG 64,0x6 20480 328 /usr/bin/cat
cat 14764 oracle mem REG 64,0x6 24576 36714 /usr/lib/libdld.2
cat 14764 oracle mem REG 64,0x6 1826816 36697 /usr/lib/libc.2
cat 14764 oracle mem REG 64,0x6 278528 33693 /usr/lib/dld.sl
cat 14764 oracle 0u CHR 3,0x2 0t0 66 /dev/null
cat 14764 oracle 1w FIFO 0x39574fe08 0t0 30479180
cat 14764 oracle 2u REG 64,0x5 446019 903 /tmp (/dev/vg00/lvol5)

It seems that fuser and lsof, (and basic ps for that matter) do NOT pull the long arg string.

Once I killed off the processes indicated by the ps -efo I was able to unmount the filesystem, vgchange the VG and let the batch process continue.

Has anyone else ever seen/tried this?

Hope this saves someone a reboot!
Let me know.

[piyush mathiya]

Steve,
Thanks, it is really good solution of that type of problem, because i have also faced this problem before some times,
Thanks again.

[Rasheed Tamton]

Did you check with ipcs (ipcs -mab). Were there any shared memory stuff not cleared. I had long back some issues with one SG package which had not cleared shared memory correctly and had problems of halting packages.

Also, did you use the umount -f (Forcibly unmount a file system).

[Jeeshan]

HI Steve

Yeah, good way. but i think

fuser -ku to lvol path rather than this long task is fruitful in very much critical situation.

[Steve Faidley]

Unfortunately this is on 11i v1, no -f option for umount.
Wow, ipcs to do an unmount, thats a new one to me. How do you translate ipcs output to a mount point?