On Wed, Aug 14, 2019 at 01:14:35PM +0200, Christoph Hellwig wrote: > Just curious from looking this again - shoudn't the 0 block be turned > into an error by the bmap() function? At least for the legacy ->bmap > case so that we don't have to carry this cruft around. I don't think so, this patch does not change the bmap interface on the filesystems itself, and at this point, we don't have a way to distinguish between a hole and an error from the filesystem's bmap() interface. The main idea of enabling bmap() to return errors, was to take advantage of the fiemap errors (in another patch in this series). Although, I do agree that this also opens the possibility to change the interface itself, and make ->bmap calls to really return errors, but this will require a bigger change in all users of ->bmap(). I can work on that, but I'd prefer to change this interface in another patchset, after this one is merged. So, we can work only on those filesystems where ->bmap() will still be a thing. Cheers. -- Carlos
