On Wed, Sep 22, 2010 at 01:33:06PM -0400, Christoph Hellwig wrote:
> > + int (*grab)(struct xfs_inode *ip),
> I think we can do without this callback. The inode verification that
> needs to be done is always the same. The only difference is that the
> dqrele code skips the quota inodes - but this can easily be done in
> the execute callback, and given that it's a slow path the additional
> two igrab calls won't hurt either.
Seems fair - the grab callback is from the first version where the
reclaim walk was not split out so there was significant differences.
I'll clean that up.
> To be symmetic with that the IRELE call should also be moved from the
> execute callbacks into the caller. All this is stuff enabled by
> splitting out the reclaim code earlier which was pretty different in
> this area. In fact just moving the validation + igrab and IRELE into
> common code might just be done in a separate patch between the last one
> and this.
Yup, that makes it a lot cleaner. Will do.