On Wed, Sep 03, 2025 at 11:09:32PM -0700, Christoph Hellwig wrote: > On Fri, Aug 29, 2025 at 04:56:16PM -0700, Joanne Koong wrote: > > Propagate any error encountered in iomap_read_folio() back up to its > > caller (otherwise a default -EIO will be passed up by > > filemap_read_folio() to callers). This is standard behavior for how > > other filesystems handle their ->read_folio() errors as well. > > Is it? As far as I remember we, or willy in particular has been > trying to kill this error return - it isn't very hepful when the > actually interesting real errors only happen on async completion > anyway. I killed the error return from ->readahead (formerly readpages). By definition, nobody is interested in the error of readahead since nobody asked for the data in those pages. I designed an error reporting mechanism a while back that allowed the errno to propagate from completion context to whoever was waiting on the folio(s) that were part of a read request. I can dig that patchset up again if there's interest.
