On Fri, 2021-04-30 at 16:15 -0400, Mike Snitzer wrote: > > If calling sg_io() is an issue then how does Martin's latest > patchset, > that exports and calls sg_io direct from DM, work!? It works by doing sg_io on the _path_ devices. There's no difference wrt to the current code in this respect. My code just retries on a different path when one fails. Hannes' original idea had passed the _dm_ device directly to scsi_cmd_blk_ioctl(). Martin
