0

I know it isn't advised to run Qubes-Whonix from Virtualbox due to performance issues, but other than that, is it safer? I.e. is it harder or easier to hack/infect Qubes-Whonix running from Virtualbox than running as usual?

  • Can Qubes even run in Virtualbox? – forest Mar 30 at 0:13
  • And am I right in assuming that this question is basically asking if nested hypervisors are more secure? – forest Mar 30 at 0:21
2

Your security posture is likely somewhat weaker when running it in a VM. I recommend that you read the Qubes-Whonix security page, as it has some useful advice here.

Running it in a VM means you increase your attack surface. Qubes-Whonix itself is designed to offer increased security, so running it inside a VM in a regular OS doesn't make much sense. You're basically running a hardened OS and then giving an OS with a weaker security posture complete control over it. You also shouldn't use a multi-boot environment for exactly this reason.

Additionally, from a forensics perspective, you can't make as many safety guarantees about memory volatility when using a VM. The virtualisation solution may flush some guest pages to disk, and the host OS itself may swap out guest pages too. Data written to the virtual disk may remain in slack space in the virtual disk file even if you overwrite blocks from within the guest, because virtualisation solutions don't always use simplistic block storage models for virtual media.

If your threat model is such that all of your sensitive work is isolated within the Qubes VM, and you're not concerned about someone targeting you personally (as opposed to a drive-by malware campaign), then Qubes in a VM is fine for convenience. Otherwise I'd run it bare-metal.

  • VT-d is not an "emulated IOMMU". It is the name of Intel's hardware-provided IOMMU implementation. There's no fundamental difference between VT-d using DMAR to restrict where PCIe devices can write on the host and VT-d doing the exact same for a guest. The technology is identical as are the security guarantees. Not to mention, no modern hypervisor attempts to emulate an IOMMU and, if it did, the performance would be so bad that it would be utterly unusable. I'd recommend removing that paragraph since it is completely incorrect. – forest Mar 30 at 0:12
  • @forest I wasn't intending to say that VT-d was emulated IOMMU; I copy-pasted the term and messed up which one the bracket came after. Does seem like you're correct on the IOMMU/DMAR front though; I think I must've confused it with some other virtualisation security thing. Offending paragraph removed. – Polynomial Mar 30 at 0:19
  • You may have been thinking of non-accelerated virtualization in hypervisors that don't support nested VT-x. – forest Mar 30 at 0:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.