On Mon, Oct 18, 2021 at 12:08:26PM -0500, Mike Christie wrote: > Oh wait, I just remembered the bug you might be hitting. If you > are using iblock for example, then when the iscsi target calls > into LIO core to submit the cmd we can end up calling into the block > layer and blocking on a full queue (hitting the nr_requests limit). > The iscsi layer is then not able to do its normal R2T/DataOut handling > because one of the iscsi threads is stuck. > > I'll send a patch to fix this issue. Can you please send me a patch so I could test it?
