On Sat, Nov 09, 2024 at 03:13:53PM +0800, Long Li wrote: > > Oh, interesting one. Do you have a reproducer we could wire up > > to xfstests? > > > > Yes, I have a simple reproducer, but it would require significant > work to incorporate it into xfstestis. Can you at least shared it? We might be able to help turning it into a test. > If we only use one size record, we can remove io_size and keep only > io_end to record the tail end of valid file data in ioend. Meanwhile, > we can add a wrapper function iomep_ioend_iosize() to get the extent > size of ioend, replacing the existing ioend->io_size. Would this work? I'd probably still use offset + size to avoid churn because it feels more natural and causes less churn, but otherwise this sounds good to me.
