iSCSI-attached disks of size 0 or which cannot be read lead to VBoxSVC going into a 100% CPU loop

[wonko1953]

Scenario 1:

• FreeBSD 12.0 as iSCSI server (machine A), exporting several block devices
• Misconfiguration results in these devices being exported with size 0 (reason in this case: the blocksize argument in /dev/ctl.conf was missing and needed to reflect the underlying device block size of 4096)
• FreeBSD 12.0 (machine C) hosts various VirtualBox clients

Scenario 2:

• Linux openSUSE Leap 15.1 as iSCSI server (machine B), exporting a block device
• Firewall misconfiguration leads to read failures when accessing the exported device
• FreeBSD 12.0 (machine C) hosts various VirtualBox clients

Result (both scenarios):

• On machine C, VBoxSVC goes into a tight loop retrying accesses to these devices
• All running VirtualBox client instances hang, including those not accessing the misconfigured iSCSI exports
• On a multiprocessor system, VBoxSVC uses n x 100% CPU (don't know how n happens, probably the number of concurrently running clients)

Expected result:

• The error should be treated reasonably, probably forwarded to the client (as a read error of the device)

Note:

• The same issue also happened with earlier releases.