On Tue, Sep 17, 2024 at 02:24:27PM -0700, Darrick J. Wong wrote: > Ah, ok, so we're taking the invalidate_lock so that we can't have page > faults that might add folios (or dirty existing ones) in the mapping. > We're the only ones who can access the page cache, and we're doing that > so that we can zero the folios between the old EOF and the start of the > write region. > > Is that right? Then Yes. We might eventually also relax this a bit to only take the look if we actually zero anything, but for that we need to do the work to turn the iomap iterators inside out first.
