home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.unix.ultrix:5790 comp.unix.admin:4232
- Newsgroups: comp.unix.ultrix,comp.unix.admin
- Path: sparky!uunet!darwin.sura.net!mips!mips!munnari.oz.au!mel.dit.csiro.au!mineng.dmpe.CSIRO.AU!dmssyd.syd.dms.CSIRO.AU!metro!basser.cs.su.oz.au!yar
- From: yar@cs.su.oz.au (Ray Loyzaga)
- Subject: Re: Strange behaviour of df and /tmp
- Reply-To: yar@cluster.cs.su.oz (Ray Loyzaga)
- Organization: Basser Dept of Computer Science, University of Sydney, Australia
- Date: Mon, 20 Jul 1992 23:22:43 GMT
- Message-ID: <1992Jul20.232243.28959@cs.su.oz.au>
- References: <1992Jul20.152235.20345@news.uakron.edu>
- Sender: news@cs.su.oz.au (News)
- Lines: 49
-
- In article <1992Jul20.152235.20345@news.uakron.edu> david@vax1.cc.uakron.edu (David Sumitra) writes:
- >The system is Ultrix 4.2 on vax 11/785. I have a strange behaviour of
- >'df'. It shows 65% full on /tmp. Whereas infact, it is empty according
- >to 'ls -la'. The 'du' on /tmp seems to agree with 'ls'.
-
- Are you sure that /tmp was mounted at the time? or was the df reporting
- the status of the root partition ... assuming it was mounted ...
- >
- >According to DEC whom I called for help, this could occur when a process
- >unlinks a file and just hangs or sits for a while to complete. DEC says
- >since the file is unlinked it is not easy to get its inode to remove
- >it. It seems it is the standard feature with all unix based file
-
- This is the Unix way of making sure that the file will be removed at the
- end of the process.
-
- >systems and not just the Ultrix. The suggested solution is to unmount
- >/tmp file system and to run 'newfs'. This for me, as an SA is not a very
-
- That is not the way, for one thing the /tmp will not unmount while
- there is a process referencing it. Once the process is tracked down
- (you can probably use fuser) if it is malicious or otherwise unneccessary
- you can kill it and the disk blocks will be returned to the free list.
-
- >happy situation. This means, that any malicious user can make /tmp full
- >requiring a newfs to clear it. This may possibly require a system down
- >time also, as the open file may not allow unmounting /tmp.
-
- Under normal Unix, any user can fill a filesystem that he has write
- permission and no quotas, /tmp is just one instance of this (we
- have a much fuller set of resource controls here).
- It should NEVER require a newfs to fix it, either remove any file
- that is taking up the space or kill the process that holds the
- last reference to an already unlinked file.
-
- >I have on my system the PD 'ofiles' which is supposed to give the user i.d
- >and inode of all open files. For some reason this did not show me the
- >file that was causing the /tmp full. Luckily for me, just before I was
- >getting ready to umount /tmp as suggested, the file seemed to have
- >cleared by itself after about 4 hours.
-
- ofiles is probably broken, the process that was referencing the file
- probably finished, so you got the blocks back.
-
- >Is this a normal unix file system behaviour? Is there no other way
- >(other than newfs) an SA can clear a /tmp in this situation?
-
- The behaviour is very normal.
-
-