On Thu, Nov 03, 2022 at 08:39:27AM +1100, Dave Chinner wrote: > My concerns with putting it into the iomap is that different > filesystems will have different mechanisms for detecting stale > iomaps. THe way we do it with a generation counter is pretty coarse > as any change to the extent map will invalidate the iomap regardless > of whether they overlap or not. OTOH it is a good way to nudge users to at least implement this simple but working scheme, and it has no real overhead if people later want to do something more fancy. > If, in future, we want something more complex and finer grained > (e.g. an iext tree cursor) to allow us to determine if the > change to the extent tree actually modified the extent backing the > iomap, then we are going to need an opaque cookie of some kind, not > a u32 or a u32*. Yes, but for that it actually has to be worth it and be implemented.
