Neil Carlson wrote:

> In a further test, I copied a directory tree (~50MB) into a brand new
> jfs partition, and then removed it. Prior to the copy, df showed 196KB
> used in the partition (500MB), presumably for the journal/log. After
> the cp/rm/sync it showed 332KB used. Unmounting the partition, and
> running "fsck -fv" showed nothing unusual, but after remounting, df
> showed ~210KB used.  Is this hysteretic behavior normal?  I'd expect
> an empty partition to show the same "used" amount whether it was
> newly created or emptied with a rm -rf.


Something is wrong here.  It's possible that the "used" space may be 
larger than it originally was.  This is due to the addressing structures 
that are allocated as new inodes are allocated.  The space used by the 
inodes are freed, but the addressing structures do not become free 
space, but are reused as needed.  However, running fsck against the 
partition should not affect the "used" space unless something was wrong 
before running fsck.  Running "fsck -nv" before fixing it with "-fv" 
should report some discrepencies.

> I also monitored the syslog
> file during this test, and occasionally an ls on the jfs partition
> after the rm -rf seemed to trigger an "XTREE_GETPAGE: xtree page
> corrupt" message that others have reported.  I wasn't able to reliably
> reproduce this though; I'll continue to do some testing.


This is definitely a problem.  I have not been able to reproduce this 
myself, but would be interested if anyone has a reliable way of 
reproducing it.


> 
> Thanks for your effort in bringing JFS to linux!


Thanks for your help!


> Neil


-- 
David Kleikamp
IBM Linux Technology Center

_______________________________________________
Jfs-discussion mailing list
[EMAIL PROTECTED]
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/jfs-discussion

Reply via email to