Re: are quotas journaled?

["Stephen C. Tweedie" <sct redhat com>]

Hi,

On Thu, Sep 13, 2001 at 10:12:44AM -0600, Andreas Dilger wrote:
> On Sep 13, 2001  15:14 +0100, Matt Bernstein wrote:
> > quotacheck takes longer than fsck on our ext2 fileservers. Is this
> > redundant in ext3? Do I need data=journal?
> 
> It is possible to data journal specific files (such as quota files)
> even if the rest of the filesystem is using data=ordered.  In the 2.2
> patches, the ext3 code would specifically set the data-journaling
> flag on quota files, but I don't know if this happens in 2.4.  If not,
> you would need a one-time "chattr +j <quota-file[s]>" on the ext3
> quota files, and it would then start to be data journaled.
> 
> Having said that, I'm not 100% sure if this is working correctly for
> 2.4 kernels.  There were reports (not with quota) that setting the
> journal attribute on a file didn't work, but this _should_ have been
> fixed in the most recent release of ext3 for 2.4.

The attribute handling has been fixed in 2.4, but there's still one
problem: the 2.4 quota code is not currently flushing quotas to disk
on transaction commit, so the updates aren't atomic.  I added a
writethrough mechanism for quota updates in 2.2 to deal with
this, but that seems missing from the 2.4 port.  I will be fixing
this by adding a mechanism to forcibly flush fs metadata out to the
buffer cache when we commit a transaction --- we want that mechanism
for inode updates anyway, as it will significantly improve the
performance of inode flushing for ext3.

Cheers, 
 Stephen