On 9/3/20 12:32 PM, Christoph Hellwig wrote:
We could trivially do something like this to allow the file system to call iomap_dio_complete without i_rwsem:
This is sort of how I envisioned it worked, which is why I was surprised when it bit us. This would be the cleanest stop-gap right now, but as it stands I've already sent a patch to work around the problem in btrfs for now, and we plan to take up the locking rework next go around. Thanks,
Josef
