On Tue, Sep 10, 2019 at 12:39 AM Eric Biggers [off-list ref] wrote:

quoted

Core file operations (open, close, sync, msync, truncate) are
now allowed to update the measurement immediately. In order
to maintain sufficient write performance for writes, add a
latency tunable delayed work workqueue for computing the
measurements.

This still doesn't make it crash-safe.  So why is it okay?

If Android is the load, this makes it crash safe 99% of the time and
that is considerably better than 0% of the time.

That said, we have now a patch draft forming up that pushes the update
to the ext4 journal. With this patch on top we should reach the
magical 100% given data=journal mount. One step at a time.


--
Janne