On Thu, May 26, 2011 at 04:23:37PM +0200, Jan Kara wrote:

quoted

This doesn't look quite correct me.  One of the major points of the
new truncate sequence is to to *_truncate_page before updating i_size,
so that we can properly handle an error there.  With your patch it's
still called too late.

  OK, I missed this point (frankly, handling of error from
ext3_block_truncate_page() does not seem like a huge win when we cannot
handle error anywhere else during truncate but it's an improvement I
agree). I'll move ext3_block_truncate_page() before the setting of i_size.
Thanks for having a look at the patch.

Btw, I think ext4 has the same issue.