Re: [Comments Needed] scan vs remove_target deadlock

[Mike Christie <michaelc@cs.wisc.edu>]

Mike Christie wrote:
> James Smart wrote:
>>> Actually, maybe, I should not have brought this up as it could just be
>>> more of a workaround of the core problem. For FC in fc_user_scan() do
>>> you need some sort of lock around the rport loop?
>> Yes, I had noticed this as well. However, I don't think this is influencing
>> the deadlock.
>>
> 
> iscsi needed a lock too. And so we ended up just adding a semaphore
> around the addition and deletion and scanning of sessions. We also do

Do what can happen is that we go from

iscs_user_scan()
	grab iscsi lock around sessions
		scsi-ml scan()
			grab scsi scan lock


delete session
	grab iscsi lock
	remove session from sessions list

	scsi ml host/decice deletion
		grab scsi scan lock

and I am just saying I think we are duplicating some of the locking in
the transport class (due to some weirdness with the userspace
workarounds this is more or less true).