cloud-hypervisor: container with more than one block device fails to start (repeated --disk flags)

[CapThunder19]

Description

A Linux container on cloud-hypervisor never starts if it has more than one block device.

CloudHypervisor.BuildExecCmd adds a separate --disk flag per block device:
https://github.com/urunc-dev/urunc/blob/main/pkg/unikontainers/hypervisors/cloud_hypervisor.go#L111-L121

But cloud-hypervisor expects all disks after a single --disk. Repeating the flag is a
hard error, not an override:

error: the argument '--disk <disk>...' cannot be used multiple times

So a block rootfs plus one block-backed volume produces two flags and the VM is never
created. Nothing filters the block list per monitor, so N devices always give N flags
(blockRootfs.getBlockDevs() -> monitorResources.BlockArgs -> Linux.MonitorBlockCli()
cloud-hypervisor case).

A single disk works fine, which is probably why this hasn't come up: all four
cloud-hypervisor e2e cases use one disk.

There the two --initramfs appends could never both fire.
Here they always do

QEMU and firecracker aren't affected (qemu concatenates into one string, firecracker
builds a JSON drives array)

System info

• Urunc version: 3395292 (main)
• Arch: x86_64, kernel 6.6.87.2-microsoft-standard-WSL2
• VMM: cloud-hypervisor v50.0.0 (same version pinned in ci.yml)
• Unikernel: linux

Steps to reproduce

No containerd, docker or devmapper needed. As root:

1. CGO_ENABLED=1 go build -o /root/urunc ./cmd/urunc, install cloud-hypervisor v50.0

2. Loop-backed ext4 filesystem, which becomes the second disk:

   truncate -s 64M data.img && mkfs.ext4 -q -F data.img
   mount -o loop data.img /root/uruncdemo/datamnt
   

3. Bundle with rootfs.img and kernel in its rootfs, plus:

   "mounts": [{"destination":"/data","type":"bind",
               "source":"/root/uruncdemo/datamnt","options":["rbind","rw"]}],
   "annotations": {
     "com.urunc.unikernel.unikernelType": "linux",
     "com.urunc.unikernel.hypervisor": "cloud-hypervisor",
     "com.urunc.unikernel.binary": "/kernel",
     "com.urunc.unikernel.block": "/rootfs.img",
     "com.urunc.unikernel.blkMntPoint": "/"
   }
   

4. urunc --debug --root /run/urunc create -b <bundle> demoA, then urunc --root /run/urunc start demoA

5. Repeat with "mounts": [] as a control

I used the block annotation instead of devmapper to get a block rootfs. Both end up in
the same blockRootfs builder and generate the same argv

Logs / output

Two disks:

command="[/usr/local/bin/cloud-hypervisor --memory size=268M --cpus boot=1
 --kernel /kernel --console off --serial tty --seccomp false
 --disk path=/rootfs.img,id=rootfs --disk path=/dev/loop0,id=vol0
 --cmdline panic=-1 console=ttyS0 root=/dev/vda rw ... init=/sbin/init -- ]"

error: the argument '--disk <disk>...' cannot be used multiple times
Usage: cloud-hypervisor [OPTIONS]

Same container with the volume removed, one disk:

Fatal error: VmBoot(VmBoot(KernelLoad(Bzimage(InvalidBzImage))))
  0: Error booting VM
  1: The VM could not boot
  2: Cannot load the kernel into memory
  3: Kernel Loader: failed to load bzImage kernel image

With one disk it clears argument parsing and only stops on my placeholder /kernel
The argv is built and rejected before the kernel is
read, so the placeholder isn't what breaks the two-disk case. Full logs attached.

Possible fix

Collect the specs and emit one flag:

var disks []string
for _, blockArg := range blockArgs { ...; disks = append(disks, diskArg) }
if len(disks) > 0 {
    exArgs = append(exArgs, "--disk")
    exArgs = append(exArgs, disks...)
}

Happy to send a patch if you want it.
There's no cloud_hypervisor_test.go right now, so
I could add one asserting a single --disk for N devices.

Related to #160.

[cmainas]

Hello @CapThunder19 ,

it seems cloud-hypervisor expects a single --disk cli argument. We should update the respective code block.

[CapThunder19]

Thanks for conforming should i work on this issue now??

[cmainas]

Yes, if an issue is verified, there is no reason to ask to work on it. Feel free to open a PR.