Files
OmniRoute/src/server/authz/routeGuard.ts
diegosouzapw 5f29bcf26a fix(authz): gate the cli-tools status and skills execution routes to LOCAL_ONLY
GHSA-35fw-cv32-2373 and GHSA-jx89-f37j-pq89 — the same defect class as
/api/acp/agents (GHSA-hf57): a route whose handler chain spawns a host process
was classified Tier 3 MANAGEMENT only, and requireManagementAuth() waives auth
when requireLogin=false. Hard Rules #15/#17 require the LOCAL_ONLY gate, which
runs on the stamped real peer before any auth check.

cli-tools (GHSA-35fw): 14 routes reach
getCliRuntimeStatus() -> locateCommand() -> runProcess("sh", ["-c",
'command -v -- "$1"']) -> spawn(), exactly like their six gated siblings
(forge/grok-build/jcode/qwen/omp/letta-settings):
all-statuses, status, and the claude/cline/codewhale/codex/crush/deepseek-tui/
droid/kilo/openclaw/pi/smelt-settings routes. The advisory counted 13; it
missed /api/cli-tools/detect, which is heavier — detectAllTools() runs
execFile(binary, ["--version"]) and execFile("which") per tool.

skills (GHSA-jx89): POST /api/skills/install stores the request's handlerCode
verbatim as the skill handler with no allowlist, so a value equal to a built-in
name (execute_command / eval_code) aliases the real sandboxed built-in;
POST /api/skills/executions then runs it. The sandbox is a real container, but
the spawn is transitive, which is why the 6A.8 source scan never flagged it.

Entries are exact paths, not a /api/cli-tools/ blanket prefix: apply, backups,
config, guide-settings, hermes-agent-settings, keys, logs, openclaw/auto-order
and codex-profiles do not spawn and remote dashboards use them. All 16 are
mirrored into SPAWN_CAPABLE_PREFIXES (no manage-scope bypass) and added to the
route-guard-membership roots so the gate enforces them from now on.

Functional trade-off, same one already accepted for grok/forge/jcode/qwen: a
dashboard served through a tunnel no longer shows the CLI Tools status badges.

Tests are red-first. Two existing negative controls pointed at routes that turn
out to spawn (/api/cli-tools/all-statuses, /api/skills/install); they now point
at routes that genuinely do not (/api/cli-tools/config, /api/skills/marketplace,
/api/skills/skillssh/install), so the non-over-gating assertions are kept.
2026-09-15 12:28:20 -03:00

362 lines
25 KiB
TypeScript

/**
* 3-tier route guard constants and helpers.
*
* Tier 1 — LOCAL_ONLY: accessible only from loopback. These routes spawn
* child processes; exposing them to non-local traffic is a known CVE class
* (GHSA-fhh6-4qxv-rpqj). Blocked unconditionally regardless of auth state.
*
* Carve-out: paths matching the live manage-scope bypass list (DB-stored,
* read via `getAuthzBypassSnapshot()`) MAY also be accessed from
* non-loopback if and only if the request carries an API key with the
* `manage` scope (or an authenticated dashboard session — see
* `policies/management.ts`). The bypass is opt-in per prefix and can be
* killed globally via the `localOnlyManageScopeBypassEnabled` setting.
* Unauthenticated requests to bypassable paths are still rejected with
* 403 LOCAL_ONLY.
*
* Tier 2 — ALWAYS_PROTECTED: auth is always required, even when
* requireLogin=false. Covers destructive / irreversible operations.
*
* Tier 3 — MANAGEMENT (default): auth required, but bypassed when
* requireLogin=false (existing behaviour).
*/
import { getAuthzBypassSnapshot } from "@/lib/config/runtimeSettings";
import {
SPAWN_CAPABLE_PREFIXES,
SPAWN_CAPABLE_PATTERNS,
} from "@/shared/constants/spawnCapablePrefixes";
import { VNC_ROUTE_PREFIX } from "@/lib/vncSession/manifest";
const LOOPBACK_HOSTS = new Set(["localhost", "127.0.0.1", "::1"]);
export const LOCAL_ONLY_API_PREFIXES: ReadonlyArray<string> = [
"/api/mcp/",
"/api/cli-tools/runtime/",
"/api/cli-tools/omp-settings", // spawns `which omp` to detect the CLI install (Hard Rules #15 + #17, #6318)
"/api/cli-tools/letta-settings", // spawns `which letta` to detect the CLI install (Hard Rules #15 + #17, #6318)
"/api/cli-tools/grok-build-settings", // GET calls getCliRuntimeStatus("grok-build"), which spawns a child process to locate + healthcheck the `grok` binary — same transitive-spawn surface that classified /api/skills/collect/ (Hard Rules #15 + #17). Writing ~/.grok/config.toml is inherently a local-machine operation, so loopback-only costs no real capability.
"/api/cli-tools/forge-settings", // spawns via getCliRuntimeStatus() to detect the `forge` CLI install (Hard Rules #15 + #17, #7263)
"/api/cli-tools/jcode-settings", // spawns via getCliRuntimeStatus() to detect the `jcode` CLI install (Hard Rules #15 + #17, #7263)
"/api/cli-tools/qwen-settings", // GET probes the local `qwen` binary; writes target ~/.qwen config files (Hard Rules #15 + #17)
// GHSA-35fw-cv32-2373: the 14 cli-tools routes below reach the SAME spawn as their six
// gated siblings above — getCliRuntimeStatus() -> locateCommand() -> runProcess("sh", -c
// 'command -v -- "$1"') -> spawn() — but sat on Tier 3 MANAGEMENT only, which
// requireManagementAuth() waives under requireLogin=false (incl. the fresh-install window).
// Exact entries on purpose: a blanket "/api/cli-tools/" prefix would also lock the
// non-spawning apply/backups/config/guide-settings/hermes-agent-settings/keys/logs/
// openclaw/auto-order routes that tunnel-served dashboards legitimately use.
"/api/cli-tools/all-statuses", // GET calls getCliRuntimeStatus() per CLI_TOOL_IDS entry (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/claude-settings", // spawns via getCliRuntimeStatus() to detect the `claude` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/cline-settings", // spawns via getCliRuntimeStatus() to detect the `cline` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/codewhale-settings", // spawns via getCliRuntimeStatus() to detect the `codewhale` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/codex-settings", // spawns via getCliRuntimeStatus() to detect the `codex` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/crush-settings", // spawns via getCliRuntimeStatus() to detect the `crush` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/deepseek-tui-settings", // spawns via getCliRuntimeStatus() to detect the `deepseek-tui` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/detect", // GET calls detectAllTools() -> execFile(binary, --version) + execFile("which") per tool (src/lib/cli-helper/tool-detector.ts) (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/droid-settings", // spawns via getCliRuntimeStatus() to detect the `droid` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/kilo-settings", // spawns via getCliRuntimeStatus() to detect the `kilo` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/openclaw-settings", // spawns via getCliRuntimeStatus() to detect the `openclaw` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373). Does NOT cover the non-spawning sibling /api/cli-tools/openclaw/auto-order (different segment).
"/api/cli-tools/pi-settings", // spawns via getCliRuntimeStatus() to detect the `pi` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/smelt-settings", // spawns via getCliRuntimeStatus() to detect the `smelt` CLI install (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/cli-tools/status", // GET calls getCliRuntimeStatus() per CLI_TOOL_IDS entry (Hard Rules #15 + #17, GHSA-35fw-cv32-2373)
"/api/services/", // T-10: embedded service lifecycle (spawn child processes)
"/api/tunnels/cloudflared", // POST installs/starts/stops cloudflared; safe methods are exempted below
"/api/tunnels/tailscale/disable", // stops Funnel and may stop tailscaled/Tailscale service
"/api/tunnels/tailscale/enable", // starts tailscaled/login/funnel subprocesses
"/api/tunnels/tailscale/install", // downloads/installs Tailscale and starts its daemon
"/api/tunnels/tailscale/login", // spawns `tailscale up`
"/api/tunnels/tailscale/start-daemon", // starts tailscaled/Tailscale service
"/dashboard/providers/services/", // T-07: reverse proxy to embedded service UIs
"/api/copilot/", // unauthenticated LLM driver — CLI-only by default; admins can opt-in to remote access via manage-scope bypass
"/api/tools/agent-bridge/", // AgentBridge: spawns MITM server + DNS edits (Hard Rules #15 + #17)
"/api/settings/mitm", // "Enable MITM" flow: installs a system-wide trusted root CA (security add-trusted-cert / certutil / update-ca-certificates) and writes /etc/hosts DNS overrides via src/mitm/* — host-level TLS interception. Was MANAGEMENT-only, so requireLogin=false left it remotely reachable (GHSA-x7vm-hp44-9p79, Hard Rules #15 + #17). Same tier as /api/tools/agent-bridge/.
"/api/cli-tools/antigravity-mitm", // Antigravity MITM enable flow: same privileged CA-trust + DNS surface as /api/settings/mitm (GHSA-x7vm-hp44-9p79, Hard Rules #15 + #17). Covers the /alias child route by prefix.
"/api/tools/traffic-inspector/", // Traffic Inspector: http-proxy listener + system proxy (Hard Rules #15 + #17)
"/api/issue-agent/", // Issue Agent: recorded/local triage executor surface; keep loopback/LAN until sandbox + audit hardening is complete
"/api/plugins/", // plugins: load/execute via worker_threads + child_process (Hard Rules #15 + #17)
"/api/plugins", // bare path: GET list + POST install also trigger plugin loading
"/api/middleware/", // SECURITY_AUDIT M8: middleware hooks compile+run arbitrary JS via new vm.Script (src/lib/middleware/registry.ts) on the request hot path — same code-exec class as /api/plugins/, so loopback-gate it for parity (Hard Rules #15 + #17)
"/api/system/version", // auto-update: spawns git checkout + npm install — RCE-via-tunnel surface (Hard Rules #15 + #17, found by 6A.8 route-guard gate)
"/api/db-backups/exportAll", // spawns tar for export archive (Hard Rules #15 + #17, found by 6A.8 route-guard gate)
"/api/local/", // T-12: 1-click local service launchers (Redis today; spawns podman/docker) — loopback-enforced by isLocalRequestAllowed() in src/lib/security/localEndpoints.ts (Hard Rules #15 + #17)
"/api/headroom/start", // Headroom token-saver proxy lifecycle: spawns headroom-ai python CLI (Hard Rules #15 + #17)
"/api/headroom/stop", // Headroom token-saver proxy lifecycle: sends SIGTERM/SIGKILL to managed PID (Hard Rules #15 + #17)
"/api/jobs", // JobRegistry control (enable/disable/run-now) + run history - runtime job administration, loopback-only (Hard Rules #15 + #17)
"/api/jobs/", // sub-paths: /api/jobs/:id/{runs,enable,disable,run-now} (the bare `/api/jobs` above matches the list route; this matches children)
"/api/oauth/cursor/auto-import", // spawns execFile("which", argv-array-of-one-arg "cursor") to verify a local Cursor install before importing creds — RCE-via-tunnel surface (Hard Rules #15 + #17, found by 6A.8 route-guard gate). Specific path only: the rest of /api/oauth/ (browser redirect/callback flows) must stay remote-reachable. Note: this comment intentionally avoids a literal closing square bracket character — check-openapi-security-tiers.mjs's naive regex parser for this array stops at the first one it finds, silently truncating its view of every entry after this one.
"/api/oauth/kiro/auto-import", // reads host-local Kiro credential files (homedir kiro-cli data) — must reach the loopback-only gate, not the PUBLIC /api/oauth/ prefix (GHSA-wgwc-crjm-pmwv, GHSA-gxv4-955v-v6cm). Excluded from PUBLIC in publicApiRoutes.ts.
"/api/skills/collect/", // Skill Collector CLI detection: GET .../detect probes getCliRuntimeStatus() per CLI_TOOL_IDS entry, which spawns a child process to check each tool — RCE-via-tunnel surface (Hard Rules #15 + #17, PR #6294 review).
"/api/skills/install", // POST stores the request's handlerCode verbatim as the skill handler with no allowlist; a value equal to the built-in `execute_command` / `eval_code` name aliases the real sandboxed built-in (src/lib/skills/executor.ts -> builtins.ts -> sandbox.ts childProcess.spawn). Transitive spawn the 6A.8 source-scan cannot see. Same class as /api/acp/agents (Hard Rules #15 + #17, GHSA-jx89-f37j-pq89)
"/api/skills/executions", // POST runs skillExecutor.execute() on any global/system skill with caller-chosen input — reaches the container spawn in src/lib/skills/sandbox.ts; only isAuthenticated()-gated, which requireLogin=false waives (Hard Rules #15 + #17, GHSA-jx89-f37j-pq89). Registry list/delete, marketplace and skillssh stay remote-reachable.
"/api/discovery/", // Discovery tool (opt-in provider scanner): the scan route makes outbound probes to provider endpoints (SSRF-adjacent) and the whole surface is an admin research tool — strict-loopback only, no manage-scope bypass (NOT in LOCAL_ONLY_MANAGE_SCOPE_BYPASS_PREFIXES). See _tasks/features-v3.8.42/gaps/DISCOVERY_TOOL_DESIGN.md.
VNC_ROUTE_PREFIX, // #7892: /api/vnc-session/* spawns Docker containers via child_process.spawn (src/lib/vncSession/service.ts) — RCE-via-tunnel surface (Hard Rules #15 + #17), same CVE class (GHSA-fhh6-4qxv-rpqj).
"/api/acp/agents", // ACP custom-agent registry: POST registers a client-chosen `binary`; GET / POST {action:"refresh"} runs detectInstalledAgents() -> execFileSync(probe.command, probe.args, { shell }) transitively (src/lib/acp/registry.ts) — RCE-via-tunnel surface (Hard Rules #15 + #17, #7948)
"/api/resilience/connections", // Per-account resilience state. NOTE: prefix matching also gates future /api/resilience/connections-* paths.
"/dashboard/resilience/connections", // Per-account resilience state. NOTE: this endpoint is READ-ONLY (no child process spawn, unlike every other entry in this list); gated because it exposes per-account operational state (cooldown/breaker/lockout). Do not treat as precedent for non-spawning routes.
"/api/providers/cursor/agent-availability", // credential-free dashboard-nudge check: spawns `cursor-agent status --format json` via checkCursorAgentAvailability()/getCachedCursorAgentAvailability() (src/lib/cursor/renewal.ts) — RCE-via-tunnel surface (Hard Rules #15 + #17). Narrow-scoped like /login and /refresh-cursor, not the whole /api/providers/ tree. Placed under /api/providers/ rather than /api/oauth/ because /api/oauth/ is PUBLIC-classified and never reaches this LOCAL_ONLY gate.
"/api/modality-bridge/video/", // Video Bridge status + extraction broker; fixed ffmpeg/ffprobe subprocesses, strict loopback only (Hard Rules #15 + #17)
];
/**
* LOCAL_ONLY routes whose spawn-capable segment sits AFTER a dynamic path
* parameter, so a flat prefix in `LOCAL_ONLY_API_PREFIXES` cannot target them
* without over-broadening (e.g. locking the entire `/api/providers/` subtree,
* which remote dashboards legitimately use for provider CRUD). These are matched
* by regex instead against the concrete resolved path — which is already
* `request.nextUrl.pathname` (see `runAuthzPipeline`/`classifyRoute`), the
* SAME string Next.js's own file-based router uses to resolve the `[id]`
* dynamic segment, so there is no decode/normalization mismatch between what
* this regex sees and what actually gets dispatched to the route handler.
*
* - `POST /api/providers/{id}/login` launches a headful Playwright Chromium
* (a child process) to drive a web-cookie login. Loopback enforcement must
* happen unconditionally before any auth check (Hard Rules #15 + #17), so a
* leaked JWT via tunnel cannot trigger a browser spawn.
* - `POST /api/providers/{id}/refresh-cursor` nudges `cursor-agent`
* (`--list-models`/`status`, via `src/lib/cursor/renewal.ts`) as part of
* a manual Cursor session renewal attempt — the same RCE-via-tunnel
* surface (Hard Rules #15 + #17). The rest of `/api/providers/`,
* including the generic `/refresh` route, intentionally stays
* remote-reachable — only this Cursor-specific spawn-capable path is
* gated, matching the `/login` precedent's narrow-scoping rationale.
*/
export const LOCAL_ONLY_API_PATTERNS: ReadonlyArray<RegExp> = [
/^\/api\/providers\/[^/]+\/login\/?$/,
/^\/api\/providers\/volcengine-plan\/connect(\/.*)?$/, // manual headful flow + session-based phone/SMS auto-login (both spawn Playwright)
/^\/api\/providers\/[^/]+\/refresh-cursor\/?$/,
/^\/api\/providers\/[^/]+\/chatgpt-web-codex-doctor\/?$/,
];
// `SPAWN_CAPABLE_PREFIXES` / `SPAWN_CAPABLE_PATTERNS` (the spawn-capable
// deny-lists) now live in the server-free leaf module
// `@/shared/constants/spawnCapablePrefixes` so that client-reachable
// validation schemas can import them without pulling this module's server
// runtime (runtimeSettings → localDb → ioredis) into the browser bundle.
// Imported above for the runtime check in `isLocalOnlyBypassableByManageScope`;
// re-exported here so existing `@/server/authz/routeGuard` importers keep working.
export { SPAWN_CAPABLE_PREFIXES, SPAWN_CAPABLE_PATTERNS };
/**
* Compile-time default of the manage-scope bypass list. Kept as an exported
* constant so the Settings inventory page (and audit code) can render the
* "available bypassable prefixes" choices independent of current DB state.
*
* The RUNTIME decision in `isLocalOnlyBypassableByManageScope` does NOT
* consult this constant — it reads `getAuthzBypassSnapshot().prefixes`,
* which is hot-reloaded on every settings PATCH.
*/
export const LOCAL_ONLY_MANAGE_SCOPE_BYPASS_PREFIXES: ReadonlyArray<string> = ["/api/mcp/"];
export const ALWAYS_PROTECTED_API_PATHS: ReadonlyArray<string> = [
"/api/shutdown",
"/api/providers/health-autopilot/actions",
"/api/settings/database",
// Full-database export/import: a credential dump and an irreversible replace.
// Must stay authenticated even under requireLogin=false, for the same reason
// /api/settings/database already does. isAlwaysProtectedPath matches on a path
// boundary, so this covers export, exportAll and import. (GHSA-mghq-58h3-qcqj)
"/api/db-backups",
// Legacy siblings of /api/db-backups left out of the mghq fix: export-json
// dumps every stored credential and import-json irreversibly replaces
// settings/connections, and both handlers only gate on isAuthRequired() —
// which is false under requireLogin=false. (GHSA-v7g9-7f55-5g46)
"/api/settings/export-json",
"/api/settings/import-json",
// Bulk log export: call_logs carries prompts and responses, proxy_logs carries
// client/public IPs, and the handler only calls requireManagementAuth() with no
// alwaysRequireAuth. Found sweeping the GHSA-5926-2w35-7h4q class.
"/api/logs/export",
// Codex CLI profile store. GET leaks the operator's account label; PUT writes
// attacker-supplied auth.json + config.toml straight into the operator's Codex
// CLI config (ensureCliConfigWriteAllowed() only checks CLI_ALLOW_CONFIG_WRITES,
// which defaults to true), so a POST+PUT pair repoints the CLI at attacker
// credentials or an attacker base URL. Found sweeping the same class.
"/api/cli-tools/codex-profiles",
// Writes into ~/.gemini/antigravity-cli/antigravity-oauth-token. Same family
// as the {claude,codex}-auth/apply-local pattern below; a plain path because
// it carries no dynamic segment.
"/api/providers/agy-auth/apply-local",
];
/**
* ALWAYS_PROTECTED routes whose path carries a dynamic segment, so the plain
* exact/prefix list above cannot express them: a `/api/providers/` prefix would
* hard-gate the entire provider surface and break every keyless local-first
* install. Mirrors LOCAL_ONLY_API_PATTERNS.
*
* The Claude/Codex OAuth export routes return the connection's raw
* access_token / refresh_token (and the Codex id_token) and gate only on
* `requireManagementAuth(request)` with no `alwaysRequireAuth`, which fails open
* under requireLogin=false (GHSA-5926-2w35-7h4q). They are the siblings that
* both GHSA-mghq-58h3-qcqj and GHSA-v7g9-7f55-5g46 missed.
*/
export const ALWAYS_PROTECTED_API_PATTERNS: ReadonlyArray<RegExp> = [
// `export` hands the caller the raw token; `apply-local` writes it into the
// host's CLI config (~/.codex/auth.json and the Claude equivalent). The second
// does not disclose the credential, but "anonymous" is still the wrong
// audience for it. ALWAYS_PROTECTED rather than LOCAL_ONLY on purpose: it
// closes the anonymous hole without breaking an operator driving the dashboard
// through a tunnel.
/^\/api\/providers\/[^/]+\/(claude|codex)-auth\/(export|apply-local)\/?$/,
];
export function isLoopbackHost(hostHeader: string | null): boolean {
if (!hostHeader) return false;
let host = hostHeader.trim();
if (host.startsWith("[")) {
// IPv6 literal: [::1] or [::1]:port
const bracketEnd = host.indexOf("]");
host = bracketEnd >= 0 ? host.slice(1, bracketEnd) : host.slice(1);
} else if ((host.match(/:/g) || []).length === 1) {
// IPv4 / hostname with a single :port — strip it. A bare IPv6 address
// ("::1", "::ffff:127.0.0.1") has multiple colons and must stay intact
// (splitting on ":" would mangle it to "" and miss the loopback match).
host = host.split(":")[0];
}
host = host.replace(/^::ffff:/i, "");
return LOOPBACK_HOSTS.has(host.toLowerCase());
}
/**
* Classify a resolved peer IP into the locality tiers the authz layer cares
* about. `null`/unknown → "remote" (fail closed). Used by the pipeline to stamp
* a trusted locality marker that route handlers read without re-deriving it
* from the spoofable Host header.
*/
export function classifyHostLocality(ip: string | null): "loopback" | "lan" | "remote" {
if (!ip) return "remote";
if (isLoopbackHost(ip)) return "loopback";
if (isPrivateLanHost(ip)) return "lan";
return "remote";
}
/**
* Private-LAN ranges (RFC 1918 IPv4 + IPv6 ULA/link-local). Matched against the
* real socket peer address (NOT the spoofable Host header), so a public-internet
* client — which presents a public source IP — never matches.
*/
const PRIVATE_LAN_PATTERNS: ReadonlyArray<RegExp> = [
/^10\.\d{1,3}\.\d{1,3}\.\d{1,3}$/,
/^100\.(6[4-9]|[78]\d|9\d|1[01]\d|12[0-7])\.\d{1,3}\.\d{1,3}$/,
/^192\.168\.\d{1,3}\.\d{1,3}$/,
/^172\.(1[6-9]|2\d|3[01])\.\d{1,3}\.\d{1,3}$/,
/^f[cd][0-9a-f]{2}:/i, // IPv6 ULA fc00::/7
/^fe80:/i, // IPv6 link-local
];
/**
* True when the peer address is a private-LAN address. Used to widen the
* LOCAL_ONLY tier to a trusted private network (owner-authorized 2026-05-30 for
* a LAN-deployed instance). Loopback-only surfaces that do NOT use this (e.g.
* the CLI-token path) remain strictly loopback.
*/
export function isPrivateLanHost(hostHeader: string | null): boolean {
if (!hostHeader) return false;
let host = hostHeader.trim();
if (host.startsWith("[")) {
const bracketEnd = host.indexOf("]");
host = bracketEnd >= 0 ? host.slice(1, bracketEnd) : host.slice(1);
}
host = host.replace(/^::ffff:/i, "");
// Strip :port only for IPv4 / hostname (a lone colon); leave IPv6 intact.
if ((host.match(/:/g) || []).length === 1) host = host.split(":")[0];
host = host.toLowerCase();
return PRIVATE_LAN_PATTERNS.some((re) => re.test(host));
}
/**
* Paths that are LOCAL_ONLY for all write methods but may be accessed from
* non-loopback clients when the request method is GET, HEAD, or OPTIONS.
*
* Rule: a path belongs here only when the read methods perform NO child-process
* spawn and expose NO privileged mutation — only the write methods do.
*
* Current exemptions:
* /api/system/version — GET reads package.json + npm registry; only POST
* triggers the auto-update flow (spawns git checkout + npm install + pm2).
* Hard Rules #15/#17 still apply to POST.
* /api/tunnels/cloudflared — GET reads tunnel status only; only POST
* spawns the cloudflared process (#11531).
*/
export const LOCAL_ONLY_API_GET_EXEMPTIONS: ReadonlySet<string> = new Set([
"/api/system/version",
"/api/tunnels/cloudflared",
]);
/** Safe HTTP methods that can be exempted for read-only paths. */
const SAFE_METHODS = new Set(["GET", "HEAD", "OPTIONS"]);
/**
* Returns true when `path` is a local-only route that must be blocked from
* non-loopback / non-LAN callers.
*
* @param path Normalized request path (e.g. "/api/mcp/sse").
* @param method Optional HTTP method. When provided and the method is a safe
* read-only method (GET/HEAD/OPTIONS) AND the path exactly
* matches an entry in `LOCAL_ONLY_API_GET_EXEMPTIONS`, this
* function returns false — i.e. the path is NOT local-only for
* that specific safe method. With no method argument (e.g.
* from security-scan scripts that test paths without a method),
* the function returns true (safe default) to preserve the
* conservative classification used by `check-route-guard-membership`.
*/
export function isLocalOnlyPath(path: string, method?: string): boolean {
// Method-aware GET exemption: only exact-match paths in the exemption set
// are eligible; prefix/wildcard matching is intentionally NOT used to avoid
// accidentally opening sub-paths of a spawn-capable route.
if (method && SAFE_METHODS.has(method.toUpperCase()) && LOCAL_ONLY_API_GET_EXEMPTIONS.has(path)) {
return false;
}
return (
LOCAL_ONLY_API_PREFIXES.some((p) => path === p || path.startsWith(p)) ||
LOCAL_ONLY_API_PATTERNS.some((re) => re.test(path))
);
}
/**
* Runtime predicate consulted by the management policy on every non-loopback
* request to a LOCAL_ONLY path. Reads the live snapshot:
* - returns false if the global kill-switch is off
* (`localOnlyManageScopeBypassEnabled === false`),
* - returns true iff `path` matches one of the live bypass prefixes AND
* that prefix is not in `SPAWN_CAPABLE_PREFIXES` (defence-in-depth: the
* zod schema already rejects spawn-capable entries, but a malformed DB
* row should not be able to grant a bypass).
*
* O(1) (no I/O, no async). Hot-reload SLA: <50 ms — satisfied structurally.
*/
export function isLocalOnlyBypassableByManageScope(path: string): boolean {
// Precise, unconditional early-deny for regex-matched spawn-capable routes
// (e.g. /api/providers/{id}/login, /api/providers/{id}/refresh-cursor).
// Unlike the flat-prefix defence-in-depth check below, this has the
// concrete resolved `path` already, so it's an exact match — no
// reachability heuristics needed.
if (SPAWN_CAPABLE_PATTERNS.some((re) => re.test(path))) return false;
const snapshot = getAuthzBypassSnapshot();
if (!snapshot.enabled) return false;
return snapshot.prefixes.some((p) => {
// Defence-in-depth: reject a bypass prefix that is the same as, child of,
// OR PARENT of any spawn-capable prefix. The parent case catches e.g.
// `/api/cli-tools/` (parent of `/api/cli-tools/runtime/`) — a request to
// `/api/cli-tools/runtime/foo` would otherwise satisfy `path.startsWith(p)`
// and reach the spawn-capable surface without a loopback check.
if (
SPAWN_CAPABLE_PREFIXES.some(
(spawn) => p === spawn || p.startsWith(spawn) || spawn.startsWith(p)
)
) {
return false;
}
return path === p || path.startsWith(p);
});
}
export function isAlwaysProtectedPath(path: string): boolean {
return (
ALWAYS_PROTECTED_API_PATHS.some((p) => path === p || path.startsWith(p)) ||
ALWAYS_PROTECTED_API_PATTERNS.some((re) => re.test(path))
);
}