officerdev/platform
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

per-user linux accounts, stage 1: the account and the privilege drop

Browse Source 
A member gets a real Linux account whose home is the directory the platform already
provisions for them. Nothing uses it yet — this is the mechanism plus the account,
deliberately with no behaviour change, so the file browser and terminal can be moved
onto something already proven.

Bun.spawn silently ignores uid/gid. Verified on 1.3.10: from uid 1000,
Bun.spawn(['id','-u'], {uid: 65534}) exits 0 and prints 1000. No throw, no warning.
Bun's types don't declare the option so typed code can't reach it by accident, but the
runtime accepts it, and a silently absent isolation boundary is the worst outcome this
feature could have. So privilege drops go through sudo -n setpriv, and a test pins Bun's
behaviour — if it's ever implemented, that test tells us we may simplify.

sudo is required for the drop and not because of the uid: --init-groups fails with
"Operation not permitted" for an unprivileged caller even when reuid'ing to its own
account, because setgroups(2) is root-only. --reset-env is what stops the platform's
environment crossing; verified POSTGRES_URL is unset on the far side and HOME arrives
from the target's passwd entry.

Three bugs that only a real run with a real useradd could find:

- chmod after chown fails forever, because chmod needs ownership. Both orderings fail
  unprivileged. Both operations now go through sudo, which is what makes it re-runnable.
- a member could read ANOTHER member's home: provisionUserDirs created at the default
  umask (755) and only the account being created got confined. An unlistable parent is
  no protection when the child is world-readable and emails are guessable. The skeleton
  is now created closed, 711 on the account dir and 700 inside.
- platform/.env was 664 and a member's shell printed JWT_SECRET, which is enough to mint
  an owner token and bypass every capability check. Now a boot check that refuses to
  start with OFFICER_OS_USERS on while any .env in the project root is group- or
  world-readable.

Design, the measured results and the staging plan: docs/per-user-linux-accounts.md.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
pastilhas
2026-08-11 15:38:32 +00:00
co-authored by Claude Opus 5
parent 69a31051ac
commit 5c7ceb2283
7 changed files with 779 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/per-user-linux-accounts.md
+259
View File
@@ -0,0 +1,259 @@
# Per-user Linux accounts
**Status: in progress.** Stage 1 (the account and the privilege-drop mechanism) is being built now.
Agents are explicitly out of scope for the first pass.
## What this is for
Today every `execution` capability — terminal, chat, files, tasks, items, desktop, browser — runs as the
**owner's OS user in the owner's home**. That is why `capabilities/registry.ts` declares them
`kind: 'execution'` and why `authorize.ts` strips them from a grant even if a row somehow contains one.
The registry says so out loud: *"revisit only if per-user home confinement is ever solved — and that is a
project, not a checkbox."*
This is that project. A member gets a real Linux account whose home is the directory the platform already
provisions for them, and the surfaces that execute code run **as that account**. The payoff is three
things at once:
- **Isolation** — a member cannot read another member's files, because the kernel says so rather than
because a path check happened to be right.
- **Permissions** — "may they see this" becomes a mode bit, checked by the OS on every syscall, instead
of a predicate the platform has to remember to apply on every route.
- **Separable agents** — `claude` and `opencode` run as the member, with their own `~/.claude`, their own
transcripts and their own session state, because the CLI groups by HOME and cwd.
## The target for the first test
A member signs in and:
- the **file browser** shows their home as the root and cannot navigate above it;
- the **terminal** lands in their home and has no permission to see anything above it.
Nothing else changes. Agents stay owner-only until this much is solid.
## The layout, and what each mode bit is for
```
data/ 711 service user traverse only — a member cannot enumerate the members
└── <email>/ 711 service user traverse only — a member cannot see their OWN siblings
├── home/ 700 the member their real Linux home
├── attachments/ 700 service user platform-written; unreachable even by name
├── email_accounts/ 700 service user "
├── dashboards/ 700 service user "
└── … 700 service user "
```
The important line is the second one. `data/<email>/` is traverse-only **to its own member**: they need
`x` to reach `home/`, and they must not have `r`, or they could list the platform's private tree beside
it. And because every sibling is `700 service user`, knowing a name does not help — traversal without
read gets you exactly one place, which is where they are going anyway.
This is also what resolves the two-sided ownership problem. The platform runs as the service user and
writes attachments, email databases and dashboards into `data/<email>/`; the member owns only `home/`.
Nobody needs a shared group, a setgid bit or an ACL, and neither side can write where the other lives.
**Members' homes stay under `DATA_PATH`** rather than moving to `/home/<user>`. They are the platform's
data, they belong with the rest of that account's data, and the directory is already provisioned there by
`provisionUserDirs`. A move would also break `getOwnerHomeDir`'s fallback, which is the only shape the
code has ever had for a non-owner home.
## Hard prerequisite: the secrets a shell can currently read
**This must be fixed before any member gets a shell, and it is not optional.**
On this machine, verified 2026-08-11:
| path | mode | consequence |
| --- | --- | --- |
| `/home/pastilhas` | 751 | traversable by anyone (no listing) |
| `…/officer.dev` | 775 | listable by anyone |
| `…/platform/.env` | **664** | **world-readable** |
`platform/.env` holds `POSTGRES_URL`, the JWT signing secret and every service credential. A member with
a real shell could read it and mint themselves an owner token, which makes the whole exercise worse than
not doing it — the capability model would be intact and completely bypassed.
So stage 1 includes: `chmod 600` on every `.env`, `chmod 751` on the project root so the tree is
traversable but not listable, and a **boot-time check that refuses to enable OS users while any `.env`
under the project root is group- or world-readable.** A prerequisite that is merely written down is a
prerequisite that gets skipped.
The same applies to `capabilities/` (775 today) and to the repo checkout itself: a member can read the
platform source. That is acceptable — it is not secret — but anything credential-shaped inside it is not.
## The mechanism, and the trap in it
### `Bun.spawn` silently ignores `uid` and `gid`
Verified on bun 1.3.10, 2026-08-11. From uid 1000:
```js
Bun.spawn(['id', '-u'], { uid: 65534, gid: 65534 }) // exit 0, prints "1000"
```
It does not throw. It does not warn. It accepts the option and runs as the parent. Every agent, task and
script spawn in this codebase goes through `Bun.spawn`.
Two honest qualifications, because the danger is narrower than it first looks:
- **Bun's own types do not declare `uid`**, so `bunx tsgo` rejects it. Typed code cannot reach this by
accident — confirmed while writing the test, which needs a cast to reproduce the behaviour at all.
- What *can* reach it is a spread of untyped config, an `as any`, or a plain-JS sidecar. Two of the four
sidecars are `.mjs`.
So the exposure is real but bounded, and the mitigation is the same either way: privilege drops go through
an external wrapper, and a test pins Bun's runtime behaviour. If Bun ever implements the option, that test
fails and tells us we may simplify. **A silently absent isolation boundary is the worst possible outcome of
this project**, so it is worth a test that exists only to observe something staying broken.
### `sudo -n setpriv`, and why both words are needed
`runAs` builds:
```
sudo -n setpriv --reuid=<user> --regid=<user> --init-groups --reset-env -- <argv…>
```
- `--reuid`/`--regid` set the real ids, not just effective — there is nothing to switch back to.
- `--init-groups` applies the account's supplementary groups. Without it the process keeps the *owner's*
groups, which is a quiet way to retain access we just took away.
- `--reset-env` clears the inherited environment and then sets `HOME`, `SHELL`, `USER`, `LOGNAME` and
`PATH` from the target's passwd entry. Both halves matter: the parent's env contains the owner's `HOME`,
and on a process started by PM2 in the platform directory it contains everything Bun auto-loaded from
`.env`.
**`sudo` is not optional, and the reason is not the uid.** Measured 2026-08-11: `--init-groups` fails with
`initgroups failed: Operation not permitted` for an unprivileged caller *even when reuid'ing to its own
account* — `setgroups(2)` is root-only, unconditionally. So there is no unprivileged form of this. `-n`
makes a missing sudoers entry an immediate error rather than a process hanging on a password prompt no
user will ever see.
Verified end to end, dropping to the current account:
```
$ sudo -n setpriv --reuid=pastilhas --regid=pastilhas --init-groups --reset-env -- \
sh -c 'id -u; id -G; echo HOME=$HOME; echo SECRET=${POSTGRES_URL:-unset}'
1000
1000 4 24 27 30 46 101 988 1001 ← supplementary groups from the account, not inherited
HOME=/home/pastilhas ← from passwd, after the reset
SECRET=unset ← the platform's .env did NOT cross
```
That last line is the whole security property, demonstrated rather than asserted, and it is pinned by a
test (`os-user.test.ts` → "does not pass the platform environment through").
`sudo -u <user>` alone would also work and be shorter. It is not used because its environment handling is
sudoers *policy* — `env_reset`, `env_keep`, `always_set_home` — and "which variables cross into a member's
shell" must not depend on a config file someone may have edited.
Root is available: `scripts/setup.sh` §4 installs `/etc/sudoers.d/officer-service` granting the service
user `NOPASSWD: ALL` on the full profile. The light profile deliberately skips it, so a light install that
wants OS users needs a **narrow** entry — `useradd`, `chown`, `setpriv` — which is better than the blanket
rule anyway.
### The terminal is the easy one
`sidecar/pty/sessions.mjs` uses **node-pty** under **node**, and node-pty's `spawn` genuinely honours
`uid`/`gid` (it is a native binding, not Bun's spawn). Two options; we take the second:
1. Run the pty sidecar as root and pass `uid`/`gid` per session.
2. Keep the sidecar unprivileged and make the command `setpriv … <shell> -i`.
(2) means no root daemon and one mechanism shared with everything else. A root daemon accepting session
requests over a socket is a bigger promise than this feature needs to make.
## Naming
`officer_<shellname>`, where `<shellname>` is `toShellUsername(username, email)` — the existing
sanitiser, which already lowercases, strips `@…`, replaces illegal characters and truncates to 32. The
combined name is truncated to 32 again.
The prefix earns its ugliness three times: it cannot collide with a system account, it makes every
account this feature created greppable in `/etc/passwd`, and it means a member cannot pick a username
that shadows something real.
The resolved name is **stored** on the user row (`users.os_user`) rather than re-derived. `useradd` can
adjust or refuse a name, and re-deriving would mean the platform's idea of who a member is could drift
from what is actually in `/etc/passwd`.
## Out of scope, and honest about it
- **This is not a sandbox.** A member with a shell is on the machine. They cannot read the owner's files
or another member's, and `sudo` is not theirs — but they can run code, see process names, and reach the
network. It isolates members from each other and from accidents, not from the host.
- **Agents come later.** `claude-manager.ts` drives turns through `query()` from
`@anthropic-ai/claude-agent-sdk`, which spawns `claude` itself and takes `env`/`cwd` but has nowhere to
put a uid. Dropping privileges has to happen *outside* the SDK, which makes a member's turn its own
process — a change of shape rather than a flag. The credential is not the problem:
`officer-anthropic-proxy` already holds it, so a member's `claude` needs only `ANTHROPIC_BASE_URL`
pointed at the proxy and no key of its own.
- **`pty`, `vault` and `opencode` receive no identity at all** (`TODO.md` → Multi-user). pty keys purely
on a `sessionId` from the query string, and its `/_officer/sessions` endpoints list and kill *every*
session on the box. Safe today only because terminal is owner-only. **The moment a member has a shell
that is a cross-user kill switch**, so it is fixed in the same stage as the terminal, not after.
- **Email change orphans a home.** The on-disk layout is keyed on email everywhere. Renaming an account
would leave its home behind under the old address. Pre-existing, unfixed, worth knowing.
## What the first real run proved, and what it corrected
Stage 1 was exercised end to end against a throwaway `DATA_PATH` with a real `useradd`. Every property
below was **observed**, not reasoned about:
| attempted, as the member | result |
| --- | --- |
| write in own home | OK |
| read `…/<email>/attachments/private.txt` | Permission denied |
| `ls …/<email>/` (their own account dir) | Permission denied |
| `ls $DATA_PATH` (enumerate the members) | Permission denied |
| `ls …/other-member@example.com/home` | Permission denied |
| `cd $HOME/..` | **succeeds** — see below |
Three bugs surfaced only by running it:
1. **`chmod` after `chown` fails forever.** `chmod` requires ownership, so once the home belongs to the
member the service user cannot set its mode. Both orderings fail unprivileged — the first on the second
run, the second immediately. Both operations now go through sudo, which is what makes the function
re-runnable.
2. **A member could read another member's home.** `provisionUserDirs` created directories at the default
umask (`755`), and the confinement pass only ever ran for the account being created. `DATA_PATH` being
unlistable is not protection when the child is world-readable and the attacker knows an email address.
The skeleton is now created closed — `711` on the account directory, `700` inside — so *unconfined* is
also *unreachable*.
3. **`platform/.env` was readable, and printing `JWT_SECRET` from a member's shell was confirmed.** This is
the prerequisite above, demonstrated. It is now a boot check (`assertSecretsClosed`) that refuses to
start with `OFFICER_OS_USERS` on while any `.env` in the project root is group- or world-readable.
**`cd $HOME/..` succeeding is correct and worth being precise about.** `711` grants traversal, so `cd`
works while `ls` does not — they can stand in the directory and see nothing in it. Beyond that, a real
shell can reach `/etc`, `/usr` and anything else the system leaves world-readable, because that is what a
shell is. So:
- the **file browser** genuinely cannot go above the home — that is path containment in `resolveUserPath`,
enforced by the platform;
- the **terminal** cannot *read* anything above the home, but is not confined to it. Confining it would
mean a namespace or a chroot, which is a different and much larger feature.
Say "cannot see behind it", not "cannot leave it".
## Follow-ups this creates
- **Deleting a member no longer removes their home.** It belongs to their uid, so the platform cannot
remove it — `rm -rf` fails with EPERM, which is how it was noticed. `deleteUserHandler` does not touch
disk today so nothing is broken, but account deletion will need `userdel` and a sudo `rm` to stop
leaving an orphaned, unremovable directory behind.
- **`confineUserTree` sets `DATA_PATH` itself to 711.** If a container ever bind-mounts a path under
`DATA_PATH` and runs as another uid, it will traverse but not list. Nothing does today.
## Stages
1. **The account and the mechanism.** `users.os_user`; `ensureOsUser` (useradd + chown + the mode bits
above); `runAs`; the `.env` permission gate; tests including the Bun-ignores-uid pin. **No behaviour
change** — accounts are created and nothing uses them yet.
2. **`getOwnerHomeDir` honours its email argument.** It takes an email and throws it away whenever
`HOME_DIR` is set, which is always on a real install. Seven call sites; this one change repoints the
file browser, chat, tasks, agents and the VNC password file per account.
3. **The file browser**, rooted at the member's home. Containment already exists — `resolveUserPath` +
`isInside`, which has the `..`-escape fix in it — so this is a root-resolution change, not new
security code.
4. **The terminal**, via `setpriv`, plus pty identity. One `execution` capability reopened.
5. **Agents.** Separately, later, with the SDK problem solved first.
src/databases/officer_db/src/schema/auth.ts
+12
View File
@@ -35,6 +35,18 @@ export const users = pgTable(
role: text('role', { enum: USER_ROLES }).notNull().default('Member'),
name: text('name'),
username: text('username').unique(),
/**
* The Linux account this platform account runs as, when per-user OS accounts are enabled.
*
* Stored rather than re-derived from `username`. `useradd` can adjust or refuse a name, and a derived
* value would let the platform's idea of who someone is drift from what is actually in `/etc/passwd`
* — which, for a field that decides whose uid executes a shell, is not a drift to discover later.
*
* NULL means no OS account: every account created before the feature, every account on a host where
* it is switched off, and the owner (who runs as the service user itself).
* See docs/per-user-linux-accounts.md.
*/
osUser: text('os_user').unique(),
avatar: text('avatar'),
passwordChangedAt: timestamp('password_changed_at', { withTimezone: true }),
createdAt: timestamp('created_at', { withTimezone: true }).notNull().defaultNow(),
src/server.tsx
+6
View File
@@ -3,6 +3,7 @@ import type { ServerWebSocket } from 'bun';
import { serve } from 'bun';
import { honoServer, PROTECTED_API_PREFIXES, UNPROTECTED_API_PREFIXES } from './servers/hono';
import { assertCapabilityTotality } from './servers/capabilities/totality';
import { assertSecretsClosed } from './servers/os-user';
import { resolveAuthToken } from './servers/auth-token';
import { isWsProviderAllowed } from './servers/capabilities/authorize';
import { isTokenBlacklisted } from 'officerdb';
@@ -149,6 +150,11 @@ assertCapabilityTotality({
wsProviders: Object.keys(handlers),
});
// And, when members have real Linux accounts, that they cannot read the credentials that would make those
// accounts pointless. Also before serve(), also throws: a shell handed out next to a world-readable
// JWT_SECRET is worse than no isolation, because the model looks intact. No-op while the feature is off.
await assertSecretsClosed(process.cwd());
async function upgradeWs(
req: Request,
server: any,
src/servers/api/users/create-user.ts
+23 -2
View File
@@ -1,9 +1,10 @@
import type { Handler } from 'hono';
import { createUser, getUserByEmail, getUserByUsername, USER_ROLES } from 'officerdb';
import { createUser, updateUser, getUserByEmail, getUserByUsername, USER_ROLES } from 'officerdb';
import type { UserRole } from 'officerdb';
import argon2 from 'argon2';
import * as errors from '@@/custom-errors';
import { provisionUserDirs } from '@@/data-path';
import { OS_USERS_ENABLED, ensureOsUser } from '@@/os-user';
import { validatePassword } from '../auth/validate-password';
import { validateUsername } from '../auth/validate-username';
import { toPublicUser } from './manage-users';
@@ -85,5 +86,25 @@ export const createUserHandler: Handler = async function (ctx) {
console.warn(`[users] created ${email} but could not provision its data directories`, ex);
}
return ctx.json({ user: toPublicUser(user) }, 201);
// The Linux account, when the host is set up for it. Same posture as the directories and for the same
// reason: this is a side effect of creating a platform account, and a failed `useradd` must not undo an
// account that otherwise exists and can sign in. The row simply keeps `osUser: null`, which every
// consumer already has to handle — that is what an account made before this feature looks like.
//
// Reported back in the response rather than only logged, so the owner sees "created, but no OS account"
// at the moment they click rather than discovering it when a terminal opens in the wrong home.
let osUser: string | null = null;
let osUserError: string | null = null;
if (OS_USERS_ENABLED) {
const result = await ensureOsUser({ email, username });
if (result.ok) {
osUser = result.osUser;
await updateUser(user.id, { osUser: result.osUser });
} else {
osUserError = result.error;
console.warn(`[users] created ${email} but could not create its Linux account: ${result.error}`);
}
}
return ctx.json({ user: { ...toPublicUser(user), osUser }, osUserError }, 201);
};
src/servers/data-path.ts
+19 -3
View File
@@ -1,5 +1,5 @@
import { join, resolve } from 'node:path';
import { mkdirSync } from 'node:fs';
import { chmodSync, mkdirSync } from 'node:fs';
import { homedir } from 'node:os';
export const DATA_PATH = process.env.DATA_PATH ?? join(process.cwd(), 'data');
@@ -64,14 +64,30 @@ export const USER_DIRS = [
] as const;
/**
* Create an account's root and its skeleton. Idempotent — an existing directory is left exactly as it is.
* Create an account's root and its skeleton, closed by default.
*
* Keyed on email because that is what the on-disk layout uses everywhere else (`DATA_PATH/<email>/…`).
* Renaming an account's email would orphan its directory; that is pre-existing and not this function's
* problem, but it is the reason nothing here derives a path from the id.
*
* ── Why the modes are set here and not only by os-user.ts ──
*
* `711` on the account directory, `700` on everything inside it. Measured while testing per-user Linux
* accounts: at the default umask these came out `755`, and a member with a shell could read ANOTHER
* member's home directory just by naming it — the parent being unlistable is not protection when the
* child itself is world-readable. "Locked unless something opens it" has to be the resting state, so it
* belongs at creation rather than in the confinement pass, which only ever runs for accounts that have an
* OS user.
*
* `chmod` explicitly rather than mkdir's `mode`, which is masked by the umask and does nothing at all for
* a directory that already exists.
*/
export const provisionUserDirs = (email: string): void => {
for (const dir of USER_DIRS) mkdirSync(join(DATA_PATH, email, dir), { recursive: true });
const accountDir = join(DATA_PATH, email);
for (const dir of USER_DIRS) mkdirSync(join(accountDir, dir), { recursive: true });
// Traversable, not listable: reaching `home` must not mean enumerating the platform's tree beside it.
chmodSync(accountDir, 0o711);
for (const dir of USER_DIRS) chmodSync(join(accountDir, dir), 0o700);
};
export const getTmpAttachmentsDir = (email: string) => join(DATA_PATH, email, 'attachments', 'tmp');
src/servers/os-user.test.ts
+168
View File
@@ -0,0 +1,168 @@
import { describe, expect, test } from 'bun:test';
import { chmod, mkdtemp, writeFile } from 'node:fs/promises';
import { tmpdir } from 'node:os';
import { join } from 'node:path';
import { findReadableSecrets, osUserNameFor, runAsArgv, OS_USER_PREFIX } from './os-user';
// The tests that matter here are the two that prove the MECHANISM rather than the plumbing: that Bun
// ignores `uid`, and that `setpriv` does not. Everything else in os-user.ts touches the passwd database
// and is exercised by hand — see docs/per-user-linux-accounts.md.
describe('Bun.spawn uid', () => {
// THE pin. `Bun.spawn` accepts `uid`/`gid` at RUNTIME and silently ignores them, so a privilege drop
// written the obvious way runs as the parent while looking correct. This test exists so that:
//
// - nobody "simplifies" runAs into a uid option, and
// - if Bun ever implements it, this fails and tells us we may.
//
// Bun's own types do not declare `uid`, so typed code cannot reach this by accident — hence the cast,
// which stands in for the ways you WOULD reach it: a spread of untyped config, or an `as any` in a
// hurry. The runtime is what silently accepts it, and the runtime is what this pins.
//
// Skipped when running as root, where setuid would actually be permitted and the observation changes.
test.skipIf(process.getuid?.() === 0)('is ignored at runtime — this is why runAs exists', async () => {
const nobody = 65534;
expect(process.getuid?.()).not.toBe(nobody);
const options = { uid: nobody, gid: nobody, stdout: 'pipe', stderr: 'pipe' } as unknown as {
stdout: 'pipe';
stderr: 'pipe';
};
const proc = Bun.spawn(['id', '-u'], options);
const seen = (await new Response(proc.stdout).text()).trim();
const code = await proc.exited;
// If this ever fails, read the assertion rather than fixing it: either the child ran as `nobody`
// (Bun now honours the option) or it refused with EPERM (Bun now tries). Both are good news.
expect(code).toBe(0);
expect(seen).toBe(String(process.getuid?.()));
expect(seen).not.toBe(String(nobody));
});
});
/** Whether this machine can actually drop privileges. See the sudo note in runAsArgv. */
const canSudo = await (async () => {
const proc = Bun.spawn(['sudo', '-n', 'true'], { stdout: 'ignore', stderr: 'ignore' });
return (await proc.exited) === 0;
})();
describe('runAsArgv', () => {
test('wraps the command in sudo + setpriv with real ids, groups and a reset environment', () => {
expect(runAsArgv('officer_ana', ['zsh', '-i'])).toEqual([
'sudo',
'-n',
'setpriv',
'--reuid=officer_ana',
'--regid=officer_ana',
'--init-groups',
'--reset-env',
'--',
'zsh',
'-i',
]);
});
// --init-groups and --reset-env are not decoration: without the first the process keeps the owner's
// supplementary groups, and without the second it inherits everything Bun loaded from .env.
test('never omits --init-groups or --reset-env', () => {
const argv = runAsArgv('officer_ana', ['true']);
expect(argv).toContain('--init-groups');
expect(argv).toContain('--reset-env');
});
test('refuses an empty user or command rather than running as the owner', () => {
expect(() => runAsArgv('', ['true'])).toThrow();
expect(() => runAsArgv('officer_ana', [])).toThrow();
});
// Proves the argv composes and runs end to end. Targets the CURRENT account so no test user has to be
// created, which still exercises sudo, setpriv, initgroups and the env reset.
//
// Skipped where passwordless sudo is unavailable — that is a machine that cannot run this feature at
// all, and a red test there would say "the code is broken" instead of "this host is not set up".
test.skipIf(!canSudo)('runs the command as the requested account', async () => {
const me = process.getuid?.() ?? 0;
const proc = Bun.spawn(runAsArgv(String(me), ['id', '-u']), { stdout: 'pipe', stderr: 'pipe' });
const out = (await new Response(proc.stdout).text()).trim();
const err = (await new Response(proc.stderr).text()).trim();
expect(await proc.exited, `setpriv failed: ${err}`).toBe(0);
expect(out).toBe(String(me));
});
// The property the whole feature rests on: the platform's environment does not cross the boundary. This
// process is started by PM2 in the platform directory, so Bun has auto-loaded `.env` into it — the JWT
// secret and POSTGRES_URL are in `process.env` right now. A member's shell must not see them.
test.skipIf(!canSudo)('does not pass the platform environment through', async () => {
const me = process.getuid?.() ?? 0;
const proc = Bun.spawn(runAsArgv(String(me), ['sh', '-c', 'echo "[${OFFICER_LEAK_PROBE:-unset}]"']), {
env: { ...process.env, OFFICER_LEAK_PROBE: 'this-must-not-cross' },
stdout: 'pipe',
stderr: 'pipe',
});
const out = (await new Response(proc.stdout).text()).trim();
expect(await proc.exited).toBe(0);
expect(out).toBe('[unset]');
});
// …and HOME is the target account's, not the caller's. This is what makes a member's shell and their
// agent's config land in their own directory rather than the owner's.
test.skipIf(!canSudo)('sets HOME from the target account, not the caller', async () => {
const me = process.getuid?.() ?? 0;
const proc = Bun.spawn(runAsArgv(String(me), ['sh', '-c', 'echo "$HOME"']), { stdout: 'pipe', stderr: 'pipe' });
const out = (await new Response(proc.stdout).text()).trim();
expect(await proc.exited).toBe(0);
expect(out).toBeTruthy();
// Read from passwd rather than inherited: `--reset-env` cleared the caller's HOME before setting it.
const passwd = Bun.spawn(['sh', '-c', `getent passwd ${me} | cut -d: -f6`], { stdout: 'pipe' });
expect(out).toBe((await new Response(passwd.stdout).text()).trim());
});
});
describe('osUserNameFor', () => {
test('prefixes so it cannot collide with a system account', () => {
expect(osUserNameFor({ username: 'ana', email: 'ana@example.com' })).toBe(`${OS_USER_PREFIX}ana`);
});
test('falls back to the email local part when there is no username', () => {
expect(osUserNameFor({ username: null, email: 'Ana.Silva@example.com' })).toBe(`${OS_USER_PREFIX}ana.silva`);
});
test('sanitises what useradd would refuse', () => {
expect(osUserNameFor({ username: 'Ana Silva!', email: 'a@b.com' })).toBe(`${OS_USER_PREFIX}ana_silva_`);
});
// The prefix can push an already-32-char sanitised name over the limit, and useradd rejects the whole
// name rather than truncating it.
test('stays within the 32-character limit', () => {
const name = osUserNameFor({ username: 'a'.repeat(40), email: 'a@b.com' });
expect(name.length).toBe(32);
expect(name.startsWith(OS_USER_PREFIX)).toBe(true);
});
test('a username that tries to shadow root is still prefixed', () => {
expect(osUserNameFor({ username: 'root', email: 'r@b.com' })).toBe(`${OS_USER_PREFIX}root`);
});
});
describe('findReadableSecrets', () => {
test('reports a group- or world-readable .env, and nothing when it is 600', async () => {
const dir = await mkdtemp(join(tmpdir(), 'officer-secrets-'));
const env = join(dir, '.env');
await writeFile(env, 'JWT_SECRET=x\n');
await chmod(env, 0o644);
expect(await findReadableSecrets(dir)).toEqual([env]);
// Group-only still counts: a member's supplementary groups are not ours to predict.
await chmod(env, 0o640);
expect(await findReadableSecrets(dir)).toEqual([env]);
await chmod(env, 0o600);
expect(await findReadableSecrets(dir)).toEqual([]);
});
test('an absent .env is not a finding', async () => {
const dir = await mkdtemp(join(tmpdir(), 'officer-secrets-'));
expect(await findReadableSecrets(dir)).toEqual([]);
});
});
src/servers/os-user.ts
+292
View File
@@ -0,0 +1,292 @@
import { chmod, mkdir, readdir, stat } from 'node:fs/promises';
import { existsSync } from 'node:fs';
import { join } from 'node:path';
import { DATA_PATH, USER_DIRS, toShellUsername } from './data-path';
// Real Linux accounts for members, so the surfaces that execute code can run as them.
//
// Design, prerequisites and the staging plan: docs/per-user-linux-accounts.md. Read it before changing
// anything here — several of the choices below look arbitrary and are not.
//
// ── The one thing to know ──
//
// `Bun.spawn` SILENTLY IGNORES `uid` and `gid`. Verified on bun 1.3.10: from uid 1000,
// `Bun.spawn(['id','-u'], { uid: 65534 })` exits 0 and prints 1000. No throw, no warning. So a privilege
// drop written the obvious way would look like it worked while every member's process ran as the owner —
// an isolation boundary that is silently absent, which is worse than none at all because it is believed.
//
// Everything here goes through `setpriv`. os-user.test.ts pins Bun's behaviour so that if it is ever
// fixed, a test tells us we may simplify, rather than someone assuming it and being wrong.
/** Off unless explicitly enabled: this needs root, and a light install has no sudoers entry. */
export const OS_USERS_ENABLED = process.env.OFFICER_OS_USERS === 'true' || process.env.OFFICER_OS_USERS === '1';
/** Prefixed so it cannot collide with a system account, and so `/etc/passwd` shows what we created. */
export const OS_USER_PREFIX = 'officer_';
const MAX_USERNAME = 32;
/**
* The Linux account name for a platform account.
*
* Built on `toShellUsername`, which already lowercases, strips anything from `@` on, replaces illegal
* characters and truncates. Truncated again after the prefix, because the prefix can push a 32-char
* result over the limit and `useradd` would refuse the whole thing.
*/
export function osUserNameFor(params: { username: string | null; email: string }): string {
const base = toShellUsername(params.username ?? '', params.email);
return `${OS_USER_PREFIX}${base}`.slice(0, MAX_USERNAME);
}
export type RunAsOptions = {
/** Passed through to the wrapped command. `setpriv --reset-env` means nothing else survives. */
env?: Record<string, string>;
cwd?: string;
};
/**
* The argv that runs `command` as `osUser`. Pure, so the shape is testable without spawning anything.
*
* sudo -n REQUIRED, and not merely for the uid. Measured 2026-08-11: `--init-groups` fails
* with "initgroups failed: Operation not permitted" for an unprivileged caller even
* when reuid'ing to its OWN account — setgroups(2) is root-only, full stop. So this
* cannot be done without privilege, and `-n` makes a missing sudoers entry an
* immediate error instead of a process blocking on a password prompt nobody will see.
* --reuid/--regid the REAL ids, not merely effective — there is nothing to switch back to.
* --init-groups apply the account's supplementary groups. Without it the process keeps the OWNER'S
* groups, which quietly retains access we just took away.
* --reset-env drop the inherited environment, then set HOME/SHELL/USER/LOGNAME/PATH from the
* target's passwd entry. Both halves matter: the parent's env carries the owner's HOME
* and — in a PM2 process started in the platform directory — everything Bun auto-loaded
* from `.env`. Verified: `POSTGRES_URL` is unset on the far side, and HOME arrives as
* the member's own.
*
* `sudo -u <user>` alone would also work and would be shorter. It is not used because its environment
* handling is sudoers policy (`env_reset`, `env_keep`, `always_set_home`) rather than something this file
* states — and "which variables cross into a member's shell" is exactly the question that must not depend
* on a config file somebody may have edited.
*/
export function runAsArgv(osUser: string, command: string[]): string[] {
if (!osUser) throw new Error('runAsArgv: no OS user');
if (!command.length) throw new Error('runAsArgv: empty command');
return [
'sudo',
'-n',
'setpriv',
`--reuid=${osUser}`,
`--regid=${osUser}`,
'--init-groups',
'--reset-env',
'--',
...command,
];
}
/**
* Run a command as another Linux account.
*
* Deliberately does NOT accept a `uid` option. The only supported way to change user in this codebase is
* this function, precisely because the option that looks like it would work does nothing.
*/
export function runAs(osUser: string, command: string[], options: RunAsOptions = {}) {
return Bun.spawn(runAsArgv(osUser, command), {
cwd: options.cwd,
// Reaches sudo and setpriv, NOT the command — `--reset-env` clears it on the way through. Anything
// the command needs beyond the passwd-derived HOME/SHELL/USER/LOGNAME/PATH has to be stated inside
// `command` itself (`env FOO=bar cmd …`). That asymmetry is deliberate: it means a variable can only
// cross into a member's process because someone wrote it there.
env: options.env,
stdout: 'pipe',
stderr: 'pipe',
});
}
async function run(command: string[]): Promise<{ ok: boolean; out: string }> {
const proc = Bun.spawn(command, { stdout: 'pipe', stderr: 'pipe' });
const [out, err] = await Promise.all([new Response(proc.stdout).text(), new Response(proc.stderr).text()]);
const code = await proc.exited;
return { ok: code === 0, out: `${out}${err}`.trim() };
}
/** uid/gid from the passwd database, or null when the account does not exist. */
export async function lookupOsUser(osUser: string): Promise<{ uid: number; gid: number } | null> {
const uid = await run(['id', '-u', osUser]);
if (!uid.ok) return null;
const gid = await run(['id', '-g', osUser]);
if (!gid.ok) return null;
return { uid: Number(uid.out), gid: Number(gid.out) };
}
/** The member's home: `DATA_PATH/<email>/home`, where `provisionUserDirs` already put it. */
export const osUserHome = (email: string): string => join(DATA_PATH, email, 'home');
export type EnsureOsUserResult =
| { ok: true; osUser: string; uid: number; gid: number; created: boolean }
| { ok: false; error: string };
/**
* Create the Linux account if it does not exist, then place the ownership and mode bits.
*
* Idempotent in both halves: an existing account is adopted rather than recreated, and the modes are
* re-applied every time, so a directory the platform added later is confined without needing a
* migration.
*
* Never throws. Account creation is a side effect of creating a platform account, and a `useradd` that
* failed must not leave a half-made user behind — the caller records the error and the platform account
* simply has no OS account yet.
*/
export async function ensureOsUser(params: { email: string; username: string | null }): Promise<EnsureOsUserResult> {
const osUser = osUserNameFor(params);
const home = osUserHome(params.email);
let created = false;
let ids = await lookupOsUser(osUser);
if (!ids) {
// `-M` because provisionUserDirs already made the directory, and letting useradd create it would
// copy /etc/skel in as root-owned. `-s` explicitly: /etc/default/useradd here says /bin/sh, and a
// member opening a terminal should get the same shell everyone else gets.
const create = await run([
'sudo',
'-n',
'useradd',
'--home-dir',
home,
'-M',
'--shell',
process.env.SHELL ?? '/bin/bash',
osUser,
]);
if (!create.ok) return { ok: false, error: `useradd failed: ${create.out}` };
created = true;
ids = await lookupOsUser(osUser);
if (!ids) return { ok: false, error: `useradd reported success but ${osUser} is not in passwd` };
}
const confined = await confineUserTree({ email: params.email, uid: ids.uid, gid: ids.gid });
if (!confined.ok) return { ok: false, error: confined.error };
return { ok: true, osUser, uid: ids.uid, gid: ids.gid, created };
}
/**
* Place the mode bits described in docs/per-user-linux-accounts.md § "The layout".
*
* DATA_PATH 711 service user traverse only — a member cannot enumerate the members
* DATA_PATH/<email> 711 service user traverse only — a member cannot list their OWN siblings
* …/home 700 the member their home
* every sibling 700 service user platform-written, unreachable even by name
*
* 711 on the account directory is the load-bearing one. The member needs `x` to reach `home`, and must
* not have `r`, or `ls` would show them the platform's private tree beside it. Because every sibling is
* 700 and owned by the service user, guessing a name gains nothing either.
*
* `chown` on the home is done with sudo: the service user owns the directory but cannot give it away —
* `chown` to another user is a root-only operation on Linux regardless of who owns the file.
*/
export async function confineUserTree(params: {
email: string;
uid: number;
gid: number;
}): Promise<{ ok: true } | { ok: false; error: string }> {
const accountDir = join(DATA_PATH, params.email);
const home = join(accountDir, 'home');
try {
if (!existsSync(home)) await mkdir(home, { recursive: true });
// Traversable, not listable. Applied to DATA_PATH itself too: without it a member can read the
// directory and learn every other member's email address.
await chmod(DATA_PATH, 0o711);
await chmod(accountDir, 0o711);
// Every sibling of `home` is the platform's. 700 means traversal alone does not open them.
const entries = await readdir(accountDir, { withFileTypes: true });
for (const entry of entries) {
if (entry.name === 'home') continue;
if (!entry.isDirectory()) continue;
await chmod(join(accountDir, entry.name), 0o700);
}
// And any of the standard set that does not exist yet, so a directory created later starts confined
// rather than at the process umask.
for (const dir of USER_DIRS) {
if (dir === 'home') continue;
const path = join(accountDir, dir);
if (!existsSync(path)) await mkdir(path, { recursive: true, mode: 0o700 });
}
// The home goes through sudo for BOTH operations, and that is the only form that is idempotent.
// `chmod` requires ownership, so:
// - chmod then chown, unprivileged: works once, then fails EPERM forever after, because the home now
// belongs to the member. Re-running an install would report failure on a correct tree.
// - chown then chmod, unprivileged: fails immediately, for the same reason.
// Both were observed. Root does not care about either ordering, so both go through sudo and the
// function can be run any number of times.
const give = await run(['sudo', '-n', 'chown', '-R', `${params.uid}:${params.gid}`, home]);
if (!give.ok) return { ok: false, error: `chown of ${home} failed: ${give.out}` };
const close = await run(['sudo', '-n', 'chmod', '700', home]);
if (!close.ok) return { ok: false, error: `chmod of ${home} failed: ${close.out}` };
return { ok: true };
} catch (ex) {
return { ok: false, error: ex instanceof Error ? ex.message : String(ex) };
}
}
/**
* Refuse to enable OS users while a secret in the project tree is readable by them.
*
* `platform/.env` was 664 on this machine when this was written — world-readable, holding the JWT signing
* secret and `POSTGRES_URL`. A member with a shell could read it and mint an owner token, which would
* leave the capability model intact and entirely bypassed.
*
* Checked at boot rather than documented, because a prerequisite that is only written down is one that
* gets skipped. Returns the offending paths; the caller decides whether that is fatal.
*/
export async function findReadableSecrets(projectDir: string): Promise<string[]> {
const candidates = ['.env', '.env.local', '.env.production'];
const bad: string[] = [];
for (const name of candidates) {
const path = join(projectDir, name);
if (!existsSync(path)) continue;
try {
const info = await stat(path);
// Anything readable by group or other. 0o044 covers both read bits.
if (info.mode & 0o044) bad.push(path);
} catch {
// Unreadable to us is not a leak to them; nothing to report.
}
}
return bad;
}
/**
* Refuse to boot with OS users enabled while a secret in the project tree is readable by them.
*
* Same posture as `assertCapabilityTotality`, and for the same reason: this is a prerequisite that
* silently not holding would make the whole feature theatre. Confirmed exploitable while testing — a
* member's shell read `platform/.env` and printed `JWT_SECRET`, which is enough to mint an owner token and
* bypass every capability check in the codebase.
*
* A no-op when the feature is off, so an existing install is unaffected until the owner opts in.
*/
export async function assertSecretsClosed(projectDir: string): Promise<void> {
if (!OS_USERS_ENABLED) return;
const readable = await findReadableSecrets(projectDir);
if (!readable.length) return;
throw new Error(
[
'OFFICER_OS_USERS is enabled, but these files are readable by other accounts on this machine:',
'',
...readable.map((p) => ` • ${p}`),
'',
'A member with a shell can read them. JWT_SECRET alone is enough to mint an owner token, which',
'bypasses every capability check. Fix with:',
'',
...readable.map((p) => ` chmod 600 ${p}`),
'',
'Then restart. See docs/per-user-linux-accounts.md → "Hard prerequisite".',
].join('\n'),
);
}