On Fri, 11 Oct 2024 at 19:57, Amir Goldstein <amir73il@xxxxxxxxx> wrote: > But why do we want to avoid copying attributes from the underlying inode? Because that's just a special case. The general case is that backing data is mapped into fuse file data, possibly using more than one extent and not necessarily starting at zero offset. In this case using the backing file's size doesn't make sense generally. And because it's easy to avoid, I don't see why we'd need to force using the backing inode attributes at this point. Your work on directory tree passthrough is related, but I think it's separate enough to not mix their traits. When that is finalized we can possibly add back mirroring of i_size on write, but I think the general case shouldn't have that. Thanks, Miklos
