kern-sandbox 0.1.13 → 0.1.15

package/README.md

@@ -33,10 +33,11 @@ import { runCode, withSandbox, Sandbox } from "kern-sandbox";
 npm install kern-sandbox
 ```
 
-You also need the `kern` binary on `PATH` (or point `$KERN_BIN` at it). One line on Linux:
+You also need the `kern` binary on `PATH` (or point `$KERN_BIN` at it). Build it on Linux (no
+prebuilt binaries are published yet):
 
 ```sh
-curl -fsSL https://raw.githubusercontent.com/getkern/kern/main/install.sh | sh
+cargo install --git https://github.com/getkern/kern getkern --locked
 ```
 
 kern needs a Linux kernel with unprivileged user namespaces + cgroup v2. On Windows it runs under WSL2.
@@ -134,11 +135,20 @@ new Sandbox({
   pids,            // default 256
   timeoutS,        // default 30, MANDATORY per-call deadline
   network,         // default false (RELAXES ISOLATION)
+  capDrop,         // default ["ALL"]: capabilities dropped from every box. kern always drops
+                   // 14 dangerous ones; this drops the rest, which were held over the box's own
+                   // user namespace. Pass [] to keep them (needed only if the workload binds a
+                   // port below 1024 INSIDE the box).
   mounts,          // { hostSrc: boxTarget } or { src: [target, "ro"] }
   profiles,        // reusable kern.toml profiles: ["vcpu:heavy", "vgpio:leds", "vdisk:scratch"]
   env,             // { KEY: "value" }
   maxOutputBytes,  // default 64 MiB
   enforceLimits,   // default true; false is best-effort and NO faster (see the Python README)
+  securityProfile, // "untrusted" = seccomp allowlist + cap-drop ALL + read-only root, one opt-in bundle
+  apparmor,        // a PRE-LOADED AppArmor profile the box enters on exec (Docker's --security-opt
+                   // apparmor=), an LSM layer over seccomp; kern fails the box CLOSED if it isn't loaded.
+  requireLimits,   // default false; true = FAIL-CLOSED (refuse to start unless caps enforced). NOT
+                   // enforceLimits (that picks the cap PATH); mutually exclusive with KERN_ALLOW_UNCAPPED env.
   depsReadonly,    // default false
   trackFiles,      // default true: diff the workspace each call for result.files (O(files)); false = [], O(1)
   onStdout,        // (chunk: Buffer) => void, live stdout streaming (result.stdout still captured)
@@ -214,8 +224,9 @@ clear error otherwise). The Python binding uses the stdlib `tarfile` and has no
 kern is a **kernel-boundary** sandbox for **your own or semi-trusted** code (CI, dev, edge, your
 agents' code). Its seccomp filter is a **denylist**: right for semi-trusted agent code, **not** a hard
 boundary against deliberately hostile multi-tenant code. For that, reach for a microVM (Firecracker /
-Kata) or gVisor. A deny-by-default allowlist mode is on the roadmap. See the project's
-[SECURITY.md](https://github.com/getkern/kern/blob/main/SECURITY.md).
+Kata) or gVisor. A deny-by-default seccomp **allowlist** ships as opt-in today: pass
+`securityProfile: "untrusted"` (or the `KERN_SECCOMP=allowlist` env); making it the default is future
+work. See the project's [SECURITY.md](https://github.com/getkern/kern/blob/main/SECURITY.md).
 
 ## License
 

package/index.js

@@ -36,7 +36,7 @@ const crypto = require("crypto");
 const zlib = require("zlib");
 const { spawn, spawnSync } = require("child_process");
 
-const VERSION = "0.1.13";
+const VERSION = "0.1.15";
 
 const DEFAULT_IMAGE = "python:3.12-slim";
 const WORKSPACE = "/workspace"; // where the persistent workspace is mounted inside every box
@@ -552,6 +552,12 @@ function validateProfile(token) {
 // re-validates and SSRF-checks the resolved IPs; this is the binding's first gate.
 const DOMAIN_RE = /^(?=.{1,253}$)(?:[A-Za-z0-9](?:[A-Za-z0-9-]{0,61}[A-Za-z0-9])?\.)+[A-Za-z]{2,63}$/;
 
+// A Linux capability name for `kern box --cap-drop`, with or without the CAP_ prefix, or the literal
+// ALL. Underscore-JOINED uppercase segments rather than "any of [A-Z0-9_]": the looser form accepts
+// "CAP_", because the optional prefix does not have to consume it. Not a way to smuggle a flag, but
+// a name kern rejects at box start, and validating here exists to fail at construction instead.
+const CAP_RE = /^(?=.{1,32}$)(?:CAP_)?[A-Z][A-Z0-9]*(?:_[A-Z0-9]+)*$/;
+
 /** Validate one egress-allowlist domain (an FQDN like "pypi.org") before it reaches the argv. */
 function validateDomain(domain) {
   if (typeof domain !== "string" || !DOMAIN_RE.test(domain))
@@ -562,6 +568,34 @@ function validateDomain(domain) {
   return domain;
 }
 
+/** Validate one capability name for `--cap-drop` before it reaches the argv. */
+function validateCap(name) {
+  if (typeof name !== "string" || !CAP_RE.test(name))
+    throw new SandboxError(
+      `invalid capability ${JSON.stringify(name)}: expected 'ALL' or an uppercase capability name ` +
+        "such as 'NET_BIND_SERVICE' or 'CAP_NET_BIND_SERVICE'",
+    );
+  return name;
+}
+
+// An AppArmor profile name for `kern box --apparmor`. Same discipline as validateCap: handed to kern as
+// its own argv element, so it must not start with a dash (-> another flag) or carry a space. Letters,
+// digits and ._- cover ordinary profile names (docker-default, unconfined, kern-box); kern fails closed
+// if the profile is not loaded. Namespaced names with / or : are intentionally not accepted here (use
+// the CLI). Compared byte-for-byte with the Python binding's _APPARMOR_RE by a parity test - keep them
+// identical and free of chars that need escaping in a regex literal (e.g. /).
+const APPARMOR_RE = /^[A-Za-z0-9_.][A-Za-z0-9_.-]{0,127}$/;
+
+/** Validate an AppArmor profile name for `--apparmor` before it reaches the argv. */
+function validateApparmor(name) {
+  if (typeof name !== "string" || !APPARMOR_RE.test(name))
+    throw new SandboxError(
+      `invalid AppArmor profile ${JSON.stringify(name)}: expected a loaded profile name like ` +
+        "'docker-default' or 'unconfined' (letters, digits and ._-, not starting with a dash)",
+    );
+  return name;
+}
+
 /** Map a Node close event {code, signal} to a unix-style rc (128 + signum for a signal). */
 function toRc(code, signal) {
   if (typeof code === "number") return code;
@@ -764,7 +798,27 @@ class Sandbox {
     this.onStdout = opts.onStdout ?? null;
     this.onStderr = opts.onStderr ?? null;
     this.enforceLimits = opts.enforceLimits ?? true;
+    // `--require-limits`: refuse to start unless the memory/pids caps are ACTUALLY enforced (read back
+    // from the cgroup), rather than running best-effort uncapped - the fail-closed OOM / fork-bomb
+    // backstop. Distinct from `enforceLimits` (systemd-scope vs best-effort PATH); this makes an
+    // unenforceable cap fatal.
+    this.requireLimits = opts.requireLimits ?? false;
+    // `--security-profile "untrusted"`: an opt-in hardening BUNDLE (seccomp allowlist + cap-drop ALL +
+    // read-only root) for code nobody has read. The root goes read-only but a bound `mounts` path stays
+    // writable, so it composes with this SDK. null (default) leaves kern's normal posture.
+    this.securityProfile = opts.securityProfile ?? null;
+    // `--apparmor "<profile>"`: enter a pre-loaded AppArmor profile on the box's exec (Docker's
+    // `--security-opt apparmor=`), a kernel-enforced LSM layer over namespaces + seccomp. The profile
+    // must be loaded on the host; kern fails the box CLOSED if it is not. null (default) applies none.
+    this.apparmor = opts.apparmor ?? null;
     this.depsReadonly = opts.depsReadonly ?? false;
+    // Capabilities dropped from every box this sandbox starts, as kern's own `--cap-drop` takes them.
+    // The default drops the lot: kern already drops 14 dangerous capabilities unconditionally, but the
+    // rest were still held over the box's own user namespace, on the one code path whose purpose is
+    // running code nobody has read. Defence in depth rather than the boundary itself, and measured to
+    // cost nothing. It is NOT behaviour-free: a workload binding a port below 1024 INSIDE the box
+    // needs CAP_NET_BIND_SERVICE. Pass `capDrop: []` for the previous behaviour.
+    this.capDrop = opts.capDrop ?? ["ALL"];
     // trackFiles=true populates result.files by walking the workspace before AND after each call (O(N)
     // in file count); a long session that accretes files slows every runCode. false = result.files [], O(1).
     this.trackFiles = opts.trackFiles ?? true;
@@ -790,8 +844,20 @@ class Sandbox {
         this._mountArgs.push("-v", ro ? `${real}:${tgt}:ro` : `${real}:${tgt}`);
       }
     }
+    // A bare string has a .map-less shape here, but Array.from("ALL") would yield ["A","L","L"] and
+    // three bogus flags, so refuse the string by name and say what to write instead.
+    if (typeof this.capDrop === "string")
+      throw new SandboxError(
+        `capDrop must be an array of names, not a bare string: write capDrop: [${JSON.stringify(
+          this.capDrop,
+        )}] for one, or capDrop: [] to drop none`,
+      );
+    if (!Array.isArray(this.capDrop))
+      throw new SandboxError("capDrop must be an array of capability names");
+    this._capDropArgs = this.capDrop.flatMap((c) => ["--cap-drop", validateCap(c)]);
     this._profileArgs = (this.profiles || []).map(validateProfile);
     this._egressAllow = (this.egressAllow || []).map(validateDomain);
+    if (this.apparmor !== null) validateApparmor(this.apparmor);
     if (this._egressAllow.length && this.network)
       throw new SandboxError(
         "egressAllow and network:true are mutually exclusive: egressAllow gives a restricted domain " +
@@ -881,10 +947,14 @@ class Sandbox {
       }
     }
     // kern's own --timeout is a tight BACKSTOP just beyond our deadline; OUR wait is the authority.
+    argv.push(...this._capDropArgs);
     argv.push("--timeout", String(Math.floor(timeoutS) + 5));
     if (this.memoryMb !== null) argv.push("--memory", `${this.memoryMb}m`);
     if (this.cpus !== null) argv.push("--cpus", String(this.cpus));
     if (this.pids !== null) argv.push("--pids-limit", String(this.pids));
+    if (this.requireLimits) argv.push("--require-limits");
+    if (this.securityProfile !== null) argv.push("--security-profile", this.securityProfile);
+    if (this.apparmor !== null) argv.push("--apparmor", this.apparmor);
     // Network mode: egressAllow (a domain allowlist via an isolated netns + kern's filtering proxy)
     // governs the untrusted runCode/run boxes; the setup box keeps the full network it needs to install
     // deps. egressAllow and network are mutually exclusive (checked at construction).

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "kern-sandbox",
-  "version": "0.1.13",
+  "version": "0.1.15",
   "description": "kern is a fast, rootless sandbox and virtual resource runtime for any workload, including untrusted and AI-generated code; kern-sandbox is its Node/TypeScript binding. Run untrusted or agent-generated code (Python/JS/Bash) in a real, kernel-enforced box in single-digit milliseconds, with no cloud, no account and no VM.",
   "keywords": [
     "sandbox",