Joe Thornber wrote:
dm-mpath.c: Use schedule_work() rather than queue_work(). Having our own thread for mpath seems like overkill.
When I originally looked at workqueues I also thought it was overkill becuase failures should be a rare case. However, becuase dispatch_failed (__make_request) can sleep I thought there might tbe problems if someone is doing software iSCSI and dm multipath. It looks like some network drivers are using the default workqueues to schedule work for handling timeouts, so I am not sure what is going to happen if we block, but the network connections are timing out around us.
Could this be a problem? Mike