Re: [PATCH v1 1/1] iomap: don't mark folio uptodate if read IO has bytes pending

[Christoph Hellwig <hch@infradead.org>]

On Mon, Feb 23, 2026 at 03:53:15PM -0800, Joanne Koong wrote:
> > There are three ways that iomap can be reading into the pagecache:
> > a) async ->readahead,
> > b) synchronous ->read_folio (page faults), and
> 
> b) is async as well. The code for b) and a) are exactly the same (the
> logic in iomap_read_folio_iter())

Yes.

> > This is confusing to me.  It would be more straightforward (I think) if
> > we just did it for all cases instead of adding more conditionals.  IOWs,
> > how hard would it be to consolidate the read code so that there's one
> > function that iomap calls when it has filled out part of a folio.  Is
> > that possible, even though we shouldn't be calling folio_end_read during
> > a pagecache write?
> 
> imo, I don't think the synchronous ->read_folio_range() for buffered
> writes should be consolidated with the async read logic.

Yes.  I've been thinking about that on and off, but unfortunately so far
I've not come up with a good idea how to merge the code.  Doing so would
be very useful for many reasons.

The problem with that isn't really async vs sync; ->read_folio clearly
shows you you turn underlying asynchronous logic into a synchronous call.
It's really about the range logic, where the writer preparation might
want to only read the head and the tail segments of a folio.

But if we can merge that into the main implementation and have a single
core implementation we'd be much better off.

Anyone looking for a "little" project? :)