Comment 79 for bug 1569925

2 things for previous comment #75:

1)
I pasted wrong pastebin URL. I'll describe the fix: its a service unit file that runs at the end of shutdown, right before it, re-using /run (tmpfs). Service has a preparation script that copies binaries from initrd image - so there is a minimum execution environment available - and runs the second script. Second script initializes iscsid daemon and waits in a loop for all the portals to be ONLINE and existing sessions to be LOGGED. After that it does the logout on all existing sessions (this case this would be leftovers from my previous comment AND / root disks).

2)
I tested this approach and it works good for iscsi root disks because there is a network interface that stays connected all the time. In the case of leftovers from previous unsuccessful umount attempts, the network.target would be already gone (with networkd / ifupdown interfaces already shutdown). IF I keep my interfaces configured, then this approach works for root disks AND for leftovers. For this approach to be successful for this case, this "cleanup" service would have to raise existing interfaces up again for the iscsid to login and logout to work (OR the kernel hang be resolved, then leftovers would be just left there, without logouts).