Re: XFS regression?

[David Chinner <dgc@xxxxxxx>]

On Mon, Oct 15, 2007 at 03:28:34PM +0530, Bhagi rathi wrote:
> Thanks Dave for the response.   Thinking futher, why is that xfs_iunpin has
> to mark the inode dirty?

Because the inode has been modified, and instead of sprinkling
mark_inode_dirty_sync() all over the code, we can do it in a single
spot that catches all inode modifications. We don't have to think about
it by doing this - inodes in transactions get marked dirty for free....

> All transactions generally modify one time or other, xfs_ichgtime takes care
> of marking inode as
> dirty.

Sure, but there's plenty of other transactions that don't have such
a convenient hook.

Cheers,

Dave.
-- 
Dave Chinner
Principal Engineer
SGI Australian Software Group