Hypothesis,

Test (1) - The error is NEVER propagated to upper layers:

# xfs and ext4 mounted automatically

inaddy@iscsihang:~$ mount | grep _netde
/dev/sda1 on /ext4 type ext4 (rw,relatime,stripe=32,data=ordered,_netdev)
/dev/sdb1 on /xfs type xfs (rw,relatime,attr2,inode64,noquota,_netdev)

# no error propagation

inaddy@iscsihang:~$ sudo iscsiadm -m node -o show | grep timeo.replace
node.session.timeo.replacement_timeout = -1
node.session.timeo.replacement_timeout = -1

# target server can't give any more packets to guest:

inaddy@machete:~$ sudo iptables -A INPUT -s 192.168.49.8 -p tcp --destination-port 3260 -j DROP

# reboot can't succeed

inaddy@iscsihang:~$ sudo reboot

[   27.596135]  connection2:0: ping timeout of 5 secs expired, recv timeout 5, last rx 4294896692, last ping 4294897944, now 4294899196
[   27.628109]  connection1:0: ping timeout of 5 secs expired, recv timeout 5, last rx 4294896700, last ping 4294897952, now 4294899204

Systemd hangs forever: 

[  OK  ] Stopped target Remote File Systems.
         Unmounting /ext4...
         Unmounting /xfs...

OBS: There is a tight relationship in between connection disappearing before the umount service runs and the capability of systemd to shutdown the machine entirely. I would say that, in case of no error propagation, is even worse since kernel would be locked up forever:

[  240.132208] INFO: task systemd:1094 blocked for more than 120 seconds.
[  240.133499]       Not tainted 4.4.0-93-generic #116-Ubuntu
[  240.134544] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  240.136092] INFO: task umount:1199 blocked for more than 120 seconds.
[  240.137262]       Not tainted 4.4.0-93-generic #116-Ubuntu
[  240.138302] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  240.139742] INFO: task umount:1201 blocked for more than 120 seconds.
[  240.140898]       Not tainted 4.4.0-93-generic #116-Ubuntu
[  240.141953] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.

Systemd is still trying...

[  OK  ] Unmounted /ext4.
[  OK  ] Unmounted /xfs.
[  OK  ] Stopped File System Check on /dev/disk/by-label/XFS.
[  OK  ] Stopped File System Check on /dev/disk/by-label/EXT4.
[  OK  ] Removed slice system-systemd\x2dfsck.slice.
[  OK  ] Stopped target Remote File Systems (Pre).
         Stopping Login to default iSCSI targets...

[  360.140109] INFO: task systemd:1094 blocked for more than 120 seconds.
[  360.141219]       Not tainted 4.4.0-93-generic #116-Ubuntu
[  360.142100] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  360.143377] INFO: task umount:1199 blocked for more than 120 seconds.
[  360.144451]       Not tainted 4.4.0-93-generic #116-Ubuntu
[  360.145333] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
[  360.146576] INFO: task umount:1201 blocked for more than 120 seconds.
[  360.147586]       Not tainted 4.4.0-93-generic #116-Ubuntu
[  360.148472] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.

This will happen forever. I still have to find a way of causing systemd to shutdown network and cause this hang because error, likely, is propagated after the umount service gives up its logic (or something like it) <-- theory.