If no hw-handler is loaded, the dm_scsi_error_handler is never called. Wouldn't it be sane to always call as the default error decoder it if no explicit hw-handler is in place? It would only change the default behaviour in case SCSI sense data was available, so it'd be pretty safe. Given that probably >90% of all multipath deployments are going to be based on SCSI, I think that would make a lot of sense.
Created attachment 113662 [details] Proposed patch How about this one.
Seems a reasonable thing to do.
If it ever causes problems in the future, we can simply move it into a SCSI hw_handler and expect userspace to choose it if there isn't a more-specific one to use.
Adding to next -udm.
Old bz, superceded by hardware handler restructuring