On 02/20/2017 03:52 PM, Nicholas A. Bellinger wrote: > As-is just ignoring CMD_T_COMPLETE in core_tmr_drain_state_list() is not > enough to address this bug and not cause other issues, because once a > se_cmd descriptor is handed back to the fabric driver after > transport_cmd_check_stop_to_fabric() is called, > __target_check_io_state() must not attempt to abort the descriptor. Please note that my patch is not ignoring CMD_T_COMPLETE - what it does is to postpone sending back the LUN RESET response to the initiator until the responses for all commands affected by the LUN RESET have been sent. I can move this patch after the patch that makes TMF handling synchronous because that patch makes it safe to set the CMD_T_ABORTED flag at any time. > That said, here is how I'd like to address this particular bug. > > 1) Allow CMD_T_COMPLETE to occur, but still ignore se_cmds that have > already called transport_cmd_check_stop_to_fabric(). Eg: CMD_T_ACTIVE > is not set, but CMD_T_SENT is set. My understanding of your patch is that it will cause the LUN RESET implementation to ignore those commands for which CMD_T_FABRIC_STOP has been set and that commands for which CMD_T_ACTIVE has been set but CMD_T_SENT not will also be ignored. Sorry but I don't think this approach is sufficient to fix the data corruption issue I observed. Bart. -- To unsubscribe from this list: send the line "unsubscribe target-devel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
