Reproducer for kubernetes/kubernetes#78987

Usage:

1. Create 10 IQNs and run 10 loops that just attach / mount / (write or check) / detach / unmount each IQN in endless loop:

   $ for i in {1..10}; do ./load.sh $i &>log-$i & done
   

   This should work without any issue (and error in dmesg).

2. Create a new IQN while the previous endless loops are still running:

   $ ./poke.sh 666
   

   This will only add a new IQN to the target and delete it. Nobody uses the IQN!

3. Check that the other IQNs were not happy about 2.

   $ dmesg | grep error
   [77922.686359] Buffer I/O error on dev sdc, logical block 123896, async page read
   [77922.706699] Buffer I/O error on dev sdc, logical block 123897, async page read
   [77922.724978] Buffer I/O error on dev sdc, logical block 123898, async page read
   [77922.730593] Buffer I/O error on dev sdc, logical block 123899, async page read
   

4. (cleanup). Stop the endless loops from 1.

   $ touch finish
   

   This should stop and clean up everything. Use clean.sh if not.

System load may be important factor in reproducing this bug. Some iscsiadm timeouts are expected, they're retried shortly. 5 parallel loops are enough for single core VM, try with higher numbers if you can't reproduce it.