On Thu, Aug 11, 2022 at 05:47:26AM -0700, Christoph Hellwig wrote: > Maybe we just simply need to set NEEDSREPAIR for any run that does > not specify the no-modify flag and avoid this magic lazy behavior? Heh, which part of the magic lazy behavior? You asked for the bwrite hook[1], and then Brian asked for a randomly triggered hook[2] in the buffer cache IO path to enable testing, which now results in periodic failures in xfs/155... [1] https://lore.kernel.org/linux-xfs/20210209091336.GG1718132@xxxxxxxxxxxxx/ [2] https://lore.kernel.org/linux-xfs/20210212133503.GA321056@bfoster/ > That being said the code looks sane if we can't fix this better by > just working differently. There are too many patches in my tree for me to be able to take on more refactoring. --D
