sbp2: fix another deadlock after disconnection If there were commands enqueued but not completed before an SBP-2 unit was unplugged (or an attempt to reconnect failed), knodemgrd or any process which tried to remove the device would sleep uninterruptibly in blk_execute_rq(). Therefore make sure that all commands are completed when sbp2 retreats. Signed-off-by: Stefan Richter --- Same as commit bf637ec3ef4159da3dd156ecf6f6987d8c8c5dae in Linus' tree. Index: linux-2.6.15.4/drivers/ieee1394/sbp2.c =================================================================== --- linux-2.6.15.4.orig/drivers/ieee1394/sbp2.c 2006-02-27 00:29:50.000000000 +0100 +++ linux-2.6.15.4/drivers/ieee1394/sbp2.c 2006-02-27 00:31:19.000000000 +0100 @@ -650,9 +650,15 @@ static int sbp2_remove(struct device *de if (!scsi_id) return 0; - /* Trigger shutdown functions in scsi's highlevel. */ - if (scsi_id->scsi_host) + if (scsi_id->scsi_host) { + /* Get rid of enqueued commands if there is no chance to + * send them. */ + if (!sbp2util_node_is_available(scsi_id)) + sbp2scsi_complete_all_commands(scsi_id, DID_NO_CONNECT); + /* scsi_remove_device() will trigger shutdown functions of SCSI + * highlevel drivers which would deadlock if blocked. */ scsi_unblock_requests(scsi_id->scsi_host); + } sdev = scsi_id->sdev; if (sdev) { scsi_id->sdev = NULL;