branderson
d2d00dba90
Two claims in docs/threat-model.md were false rather than loose. An escape would be a downgrade, and the account already holds more than the escape would grant: root on the VM is strictly more capability than a process inside a container with a trimmed capability set. Root on this VM is already a row in the blast radius table, which is the case the model is drawn around, and the hypervisor firewall bounds it identically however root was reached. The escape section names what an escape reaches that the container did not, and it is not one thing. Every other project on the box becomes readable, and uncommitted work in those exists nowhere else. Beyond that, the model API credential: dejima keeps it out of the container deliberately, so a session can use the API while it runs and cannot carry the token away, and the proxy holding it runs on the devbox. That makes an escape onto the devbox the step that turns temporary use into a token someone keeps. It is not the only loss that reaches past the machine, and no longer says so. Source and uncommitted work leave just as easily. What puts the credential first in the queue is that it is a standing capability rather than a copy of data, and that it sits at a known path under a known name, which makes it among the easiest things on the box to find. That also corrects the rootless rejection, which claimed the set it would protect was close to empty. It is small but not empty, and the control covers neither half of it: the proxy runs as whichever account runs the session, and one account owns every project, so moving the session to a dedicated account moves the projects with it. The threat model no longer rests on the adversary being unsophisticated. Real supply-chain malware fingerprints its environment, stages, beacons and can hand a session to a person, and a prompt injection borrows the agent as its adaptive component. What the model rests on instead is that every one of those moves needs egress, which the node firewall bounds whether the thing being bounded is a script or a person driving it. Being undirected changes what is likely, not what is possible. One person's work per devbox is stated as a requirement of the threat model rather than an observation, and cross-project context poisoning is stated as handled in dejima, in the present tense.
Devbox Hardener
Bounds what a compromised development VM can reach on your network, and tells you what that bound actually is.
Early development. Not ready to be depended on. Nothing here has been through outside review, an external audit, or adversarial testing by anyone but the person who wrote it. Every security property this describes is a claim awaiting scrutiny, not one that has survived it. Do not put this anywhere that being wrong about it would matter.
One maintainer, no support. Nothing about this is promised: not a reply, not a fix, not a release. SECURITY.md has the detail, and is where to report a vulnerability rather than in a public issue.
One person's work per devbox. A condition of the tool rather than an observation about how devboxes get used: the blast radius here is stated in terms of what is on the box, so a machine holding several people's work is out of spec and the reasoning in docs/threat-model.md does not hold for it. If you need separation between people, use a devbox each.
What a devbox is, and why it is a different problem
A devbox is a long-lived VM that runs dejima sessions. It differs from a CI runner in one way that drives everything else: it holds uncommitted work, so it cannot be rolled back on a schedule.
That matters because rollback is the entire containment story of its sister project, dejima-ci-hardener. A runner is built to be lost. Anything that gets into one has until the next nightly rebuild, and nothing on it was worth keeping anyway. Take that away and what is left has to come from somewhere else: the hypervisor boundary, and knowing precisely what sits inside it.
What it does not claim
It does not prevent root on the devbox. Nothing here tries to.
What it does is make root on the devbox reach less, and make that reach something you can check rather than assume. The honest version of what each layer grants is docs/threat-model.md, including the blast radius by layer and the options that were rejected, with reasons.
Container escape is deliberately deprioritised. An escape gains root, which is more than the container has, and root on this VM is already a row in that table, bounded by the same hypervisor firewall however it was reached. So patching is what is left worth doing, and separating escape from root inside the guest is not.
Two halves
| proxmox/README.md | The node side: create the VM, apply the firewall. The primary control. Everything is done in the web console. |
| docs/building-a-devbox.md | The guest side: install, bootstrap, verify. The same on any hypervisor. |
Do the node side first, through the end of its §3.
The split is deliberate. Nothing guest-side knows it is on Proxmox, so another hypervisor can be added later without touching a single step on the other page.
Use
sudo ./bin/dejima-devbox-bootstrap # once, in the guest
sudo dejima-devbox-audit # both halves
sudo dejima-devbox-audit reach # containment only, what the timer runs
sudo dejima-devbox-audit radius # what root here would get
All of these need root. The config is 0640 root, and the radius report stats
files in root's own home.
How it fails
The firewall is on the hypervisor, so the guest cannot verify it by reading it. That is what makes it worth having and what makes checking it awkward. The audit tests behaviour instead, by probing the default gateway: already in the routing table so it reveals nothing, guaranteed alive where an arbitrary address is not, and inside the ranges the firewall drops.
A timeout has two causes, and telling them apart is the whole design:
PASS |
Nothing answered and the network demonstrably works |
FAIL |
Something answered. Rules missing, or an exception above them |
INDETERMINATE |
Nothing answered, including things that should. Not a pass. |
A check that reports success when it does not know is worse than no check, because it gets believed.
A rule that REJECTs rather than DROPs reports FAIL even though it is
working. bash's /dev/tcp cannot tell a rejection from a live host
refusing. That is the safe direction to be wrong in, and it is why the template
specifies DROP.
Nothing is stored about what this VM must not reach. Such a list would be a map of valuable targets, kept on the machine this project assumes will be owned. The audit derives its targets from the routing table and its controls from your forge, so there is nothing to type and nothing to leak.
Tests
bats test/
test/devbox.bats asserts against a real devbox and skips everywhere else. It
is gated on DEJIMA_DEVBOX=1 rather than on detecting a devbox, because a
detector that guesses wrong skips silently, and a suite that passes by asserting
nothing is worse than one that fails.
Further reading
| docs/ | Index of everything below |
| docs/threat-model.md | What each layer grants, why escape is deprioritised, what was rejected |
| docs/reference.md | The three verdicts, the three addressing modes, the one config key |
| SECURITY.md | What this claims, and where to report a vulnerability |
| dejima-ci-hardener | The sister project, for hosts that can be rolled back |
License
GPL-3.0-or-later. See LICENSE.
The dejima wordmark and logo are not code and are not covered by that licence. They are trademarks; who holds them and what may be done with them is stated once for the organisation, at dejima's ATTRIBUTION.md.