On Mon, Jan 22, 2024 at 01:37:45AM -0500, Kent Overstreet wrote: > > Without this patch as in current mainline you will get -EOPNOTSUPP > > because sending REQ_OP_FLUSH and finally check for that to catch bugs > > like the one fixed with this patch. > > Then why did the user report -EOPNOTSUPP with the patch, which went away > when reverted?
I have no idea, as we never return -EOPNOTSUPP for REQ_OP_WRITE | REQ_PREFLUSH this would be odd. According to your report the users runs a Fedora kernel and this commit never went upstream or even into linux-next, which doesn't quite add a up to me either. But in the end just try it yourself. On current mainline with the buggy REQ_OP_FLUSH you should be seeing -EOPNOTSUPP every single time you try a data integrity operation, without it you should not. Even the most basic test should be catching this.
