When using scsi-block, any errors returned in host_status are unconditionally reported to the guest instead of considering the error policy set with rerror/werror and e.g. stopping the VM instead.

This issue shows with the reproducer from RHEL-65852 where scsi-block is backed by an iscsi block device on the host and the connection is interrupted by dropping all packets from the iscsi target (simulating a network disconnect) if /sys/block/sda/device/scsi_disk/*/max_retries is set to 0.

This problem was introduced in upstream commit f3126d65b393, first contained in QEMU 6.0, so the first affected RHEL version should be 8.6/9.0.