On Thu, Jun 25, 2026 at 07:16:47PM +0800, Zhang Yi wrote:

I've thought about this more and feel that we need to add one more
scenario here:

3) If the blocks at the boundaries are in dirty unwritten or in delay
   allocated state, it should be handled the same way as scenario 2.
   Additionally, before returning, we need to flush these two boundary
   blocks that have been partially zeroed, to ensure that after
   FALLOC_FL_WRITE_ZEROES, all ranges are in the written state.

What do you think?

Yes.  Can you send an xfstest addition the exercises these corner
cases?