Files
Termix/src/backend/operations/gate.ts
T

892 lines
26 KiB
TypeScript

/**
* The one path every action takes, from any surface.
*
* In order: plugin up, grant still held, input valid, role permission, host
* access, key scope, destructive confirm, approval, limits, run row, audit.
* Handlers run in-process, so this is an access check for callers, not a
* sandbox around plugins.
*
* Handler failures stay here. They never reach the plugin error budget, or five
* failed hosts in a rollout would switch the plugin off.
*/
import { randomUUID } from "node:crypto";
import {
capText,
matchesWhen,
parseOperationInput,
redactFields,
riskRank,
type OperationAction,
type OperationRisk,
} from "@termix-ssh/plugin-sdk/operations";
import { MAX_EVENT_DEPTH } from "./constants.js";
import { getAction, type OperationRegistration } from "./registry.js";
import {
canSelfApprove,
evaluateApproval,
type ApprovalDecision,
} from "./approvals.js";
import { withLimits } from "./limits.js";
import { emitOperationEvent } from "./events.js";
import { dropResourceCache } from "./resources.js";
import { loadOperationHost, type OperationHost } from "./hosts.js";
import { runAsActor, type ActorOrigin } from "../plugins/actor.js";
import { hasCapability } from "../plugins/permissions.js";
import { qualifyPermission } from "../plugins/manifest.js";
import { safeModeOn } from "../plugins/safe-mode.js";
import { PermissionManager } from "../utils/permission-manager.js";
import { logAudit } from "../utils/audit-logger.js";
import {
decryptSystemSecret,
encryptSystemSecret,
isSystemEncrypted,
} from "../utils/system-secret-crypto.js";
import { pluginLogger } from "../utils/logger.js";
import {
createCurrentOperationRepository,
createCurrentUserRepository,
} from "../database/repositories/factory.js";
import type {
OperationApprovalRecord,
OperationRunRecord,
} from "../database/repositories/operation-repository.js";
const MAX_OUTPUT = 32_000;
const DEFAULT_TIMEOUT_MS = 300_000;
const READ_TIMEOUT_MS = 60_000;
const DEFAULT_APPROVAL_MINUTES = 24 * 60;
const DRAIN_WAIT_MS = 10_000;
export type OperationErrorCode =
| "NOT_FOUND"
| "UNAVAILABLE"
| "INVALID_INPUT"
| "FORBIDDEN"
| "SCOPE"
| "NEEDS_CONFIRM"
| "DENIED"
| "TOO_DEEP"
| "STALE"
| "CONFLICT"
| "SAFE_MODE";
const STATUS: Record<OperationErrorCode, number> = {
NOT_FOUND: 404,
UNAVAILABLE: 503,
INVALID_INPUT: 400,
FORBIDDEN: 403,
SCOPE: 403,
NEEDS_CONFIRM: 409,
DENIED: 403,
TOO_DEEP: 409,
STALE: 409,
CONFLICT: 409,
SAFE_MODE: 409,
};
/** Refused before anything ran. Carries a status so routes can pass it on. */
export class OperationError extends Error {
readonly status: number;
constructor(
readonly code: OperationErrorCode,
message: string,
readonly details?: unknown,
) {
super(message);
this.name = "OperationError";
this.status = STATUS[code];
}
}
/** A handler throws this when a host needs a person to sign in first. */
export class OperationNeedsInteraction extends Error {
readonly code = "NEEDS_INTERACTION";
}
export interface OperationRequest {
actionId: string;
hostId?: number;
resourceKey?: string;
input?: unknown;
dryRun?: boolean;
/** A person agreed to a destructive action. */
confirmed?: boolean;
idempotencyKey?: string;
batchId?: string;
undoOf?: string;
/** Background reads for tables and tiles. Writes no run row. */
internal?: boolean;
userId: string;
origin: ActorOrigin;
apiKeyScopes?: string[] | null;
depth?: number;
}
export interface OperationOutcome {
runId: string | null;
status: string;
summary?: string;
changed?: boolean;
data?: unknown;
error?: string;
errorCode?: string;
approvalId?: string;
}
interface Prepared {
registration: OperationRegistration;
action: OperationAction;
pluginVersion: string;
input: Record<string, unknown>;
host: OperationHost | null;
effectiveRisk: OperationRisk;
}
/** The sealed copy an approval is bound to. */
interface ApprovalSnapshot {
runId: string;
actionId: string;
actionVersion: number;
pluginVersion: string;
hostId: number | null;
hostUpdatedAt: string | null;
resourceKey: string | null;
input: Record<string, unknown>;
}
const active = new Map<
string,
{ pluginId: string; controller: AbortController; done: Promise<unknown> }
>();
const draining = new Set<string>();
function now(): string {
return new Date().toISOString();
}
function errorText(error: unknown): string {
return error instanceof Error ? error.message : String(error);
}
async function loadedPlugin(pluginId: string) {
const { getPluginRuntime } = await import("../plugins/index.js");
return getPluginRuntime().loader.get(pluginId);
}
async function usernameOf(userId: string): Promise<string> {
try {
const user = await createCurrentUserRepository().findById(userId);
return user?.username ?? userId;
} catch {
return userId;
}
}
/** Every check that does not depend on who approved it. */
async function prepare(request: OperationRequest): Promise<Prepared> {
if (safeModeOn()) {
throw new OperationError("SAFE_MODE", "Actions are off in safe mode");
}
if ((request.depth ?? 0) > MAX_EVENT_DEPTH) {
throw new OperationError("TOO_DEEP", "Too many actions led to this one");
}
const registration = getAction(request.actionId);
if (!registration) {
throw new OperationError(
"NOT_FOUND",
`No action "${request.actionId}" is available`,
);
}
const { action, pluginId } = registration;
const plugin = await loadedPlugin(pluginId);
if (plugin?.state !== "active" || draining.has(pluginId)) {
throw new OperationError("UNAVAILABLE", `${pluginId} is not running`);
}
// Rechecked at use, so revoking the grant stops it on the next call.
if (
!(await hasCapability(
pluginId,
"operations:provide",
plugin.manifest.capabilities,
))
) {
throw new OperationError("UNAVAILABLE", `${pluginId} may not run actions`);
}
const parsed = parseOperationInput(action.inputs, request.input);
if (parsed.errors.length > 0) {
throw new OperationError(
"INVALID_INPUT",
parsed.errors.join("; "),
parsed.errors,
);
}
const permissions = PermissionManager.getInstance();
const permission = qualifyPermission(pluginId, action.permission);
if (!(await permissions.hasPermission(request.userId, permission))) {
throw new OperationError(
"FORBIDDEN",
`This needs the ${permission} permission`,
);
}
let host: OperationHost | null = null;
if (action.scope !== "server") {
if (request.hostId === undefined) {
throw new OperationError("INVALID_INPUT", "This action needs a host");
}
if (action.scope === "resource" && !request.resourceKey) {
throw new OperationError("INVALID_INPUT", "This action needs a target");
}
const access = await permissions.canAccessHost(
request.userId,
request.hostId,
action.hostAccess ?? "edit",
);
if (!access.hasAccess) {
throw new OperationError(
"FORBIDDEN",
`You need ${action.hostAccess ?? "edit"} access to this host`,
);
}
host = await loadOperationHost(request.hostId, plugin.manifest);
if (!host) throw new OperationError("NOT_FOUND", "Host not found");
if (!matchesWhen(action.when, host)) {
throw new OperationError(
"UNAVAILABLE",
"This action does not apply to this host",
);
}
}
// A claim by the plugin, so only signed official plugins are taken at
// their word when rules decide what needs a person.
const official =
plugin.source === "bundled" || plugin.signedTier === "official";
const effectiveRisk: OperationRisk =
!official && action.risk === "read" ? "change" : action.risk;
return {
registration,
action,
pluginVersion: plugin.manifest.version,
input: parsed.values,
host,
effectiveRisk,
};
}
function checkKeyScope(request: OperationRequest, action: OperationAction) {
const scopes = request.apiKeyScopes;
if (scopes === undefined || scopes === null) return;
if (scopes.includes("operations:run")) return;
if (scopes.includes("operations:read") && action.risk === "read") return;
throw new OperationError("SCOPE", "This API key may not run this action");
}
function originText(origin: ActorOrigin): string {
return JSON.stringify(origin);
}
function auditRun(
request: OperationRequest,
username: string,
action: OperationAction,
host: OperationHost | null,
success: boolean,
errorMessage?: string,
verb = "operation_run",
): void {
void logAudit({
userId: request.userId,
username,
action: verb,
resourceType: "operation",
resourceId: action.id,
resourceName: host?.name ?? undefined,
details: originText(request.origin),
success,
errorMessage,
}).catch(() => {});
}
async function execute(
prepared: Prepared,
request: OperationRequest,
runId: string | null,
username: string,
): Promise<OperationOutcome> {
const { action, registration, input, host } = prepared;
const repository = createCurrentOperationRepository();
const controller = new AbortController();
const dryRun = request.dryRun === true;
const timeoutMs =
action.timeoutMs ??
(action.risk === "read" ? READ_TIMEOUT_MS : DEFAULT_TIMEOUT_MS);
const lines: string[] = [];
const context = {
hostId: request.hostId,
resourceKey: request.resourceKey,
runId,
dryRun,
signal: controller.signal,
log: (line: string) => {
if (lines.length < 200) lines.push(line);
},
};
const work = async () => {
if (dryRun && action.dryRun !== "native") {
const summary = registration.handler.describe
? await registration.handler.describe(input, context)
: `Would run ${action.id}`;
return { summary, changed: false };
}
return registration.handler.run(input, context);
};
const timer = setTimeout(() => controller.abort(), timeoutMs);
timer.unref?.();
const aborted = new Promise<never>((_, reject) => {
controller.signal.addEventListener("abort", () =>
reject(new Error("The action was stopped or timed out")),
);
});
aborted.catch(() => {});
// Settles when the handler really stops, which can be after the run gives
// up on it. A plugin going down waits on this, not on the run.
let handlerStopped: () => void = () => {};
const stopped = new Promise<void>((resolve) => {
handlerStopped = resolve;
});
const startWork = () => {
// Cancelled while waiting for a slot, so it never starts.
if (controller.signal.aborted) {
handlerStopped();
return aborted;
}
const handling = Promise.resolve().then(work);
void handling.finally(handlerStopped).catch(() => {});
return Promise.race([handling, aborted]);
};
const running = withLimits(
{ hostId: request.hostId, exclusive: action.exclusive },
() =>
runAsActor(request.userId, "service", startWork, undefined, {
origin: request.origin,
apiKeyScopes: request.apiKeyScopes,
runId: runId ?? undefined,
depth: (request.depth ?? 0) + 1,
}),
);
if (runId) {
active.set(runId, {
pluginId: registration.pluginId,
controller,
done: stopped,
});
}
let outcome: OperationOutcome;
try {
const result = await running;
const data = redactFields(action.outputs, result.data);
outcome = {
runId,
status: "success",
summary: result.summary,
changed: result.changed,
data,
};
if (runId) {
await repository.updateRunFrom(runId, "running", {
status: "success",
summary: capText(result.summary ?? "", 2000),
changed: result.changed ?? null,
output:
data === undefined ? null : capText(JSON.stringify(data), MAX_OUTPUT),
undoInput: result.undoInput ? JSON.stringify(result.undoInput) : null,
sealedInput: null,
finishedAt: now(),
});
}
} catch (error) {
const needsPerson =
(error as { code?: unknown } | null)?.code === "NEEDS_INTERACTION";
const stopping = draining.has(registration.pluginId);
const status = needsPerson
? "needs_interaction"
: stopping
? "interrupted"
: "failed";
const message = stopping
? "The plugin stopped during this run. Check the host."
: capText(errorText(error), 2000);
outcome = { runId, status, error: message };
if (needsPerson) outcome.errorCode = "NEEDS_INTERACTION";
if (runId) {
await repository.updateRunFrom(runId, "running", {
status,
error: message,
errorCode: outcome.errorCode ?? null,
sealedInput: null,
finishedAt: now(),
});
}
} finally {
clearTimeout(timer);
if (runId && !draining.has(registration.pluginId)) active.delete(runId);
}
if (action.risk !== "read" && !dryRun) {
if (request.hostId !== undefined && outcome.status === "success") {
dropResourceCache(request.hostId);
}
auditRun(
request,
username,
action,
host,
outcome.status === "success",
outcome.error,
);
void emitOperationEvent({
eventId:
outcome.status === "success"
? "operation.succeeded"
: "operation.failed",
pluginId: registration.pluginId,
hostId: request.hostId,
severity: outcome.status === "success" ? "info" : "warning",
payload: {
actionId: action.id,
summary: outcome.summary,
error: outcome.error,
origin: request.origin,
},
runId: runId ?? undefined,
depth: (request.depth ?? 0) + 1,
});
}
return outcome;
}
function outcomeOf(
run: OperationRunRecord,
approvalId?: string,
): OperationOutcome {
return {
runId: run.id,
status: run.status,
summary: run.summary ?? undefined,
changed: run.changed ?? undefined,
error: run.error ?? undefined,
errorCode: run.errorCode ?? undefined,
approvalId,
};
}
async function requestApproval(
prepared: Prepared,
request: OperationRequest,
decision: ApprovalDecision,
run: { id: string; username: string; createdAt: string },
): Promise<OperationOutcome> {
const { action, host, input } = prepared;
const repository = createCurrentOperationRepository();
const snapshot: ApprovalSnapshot = {
runId: run.id,
actionId: action.id,
actionVersion: action.version ?? 1,
pluginVersion: prepared.pluginVersion,
hostId: request.hostId ?? null,
hostUpdatedAt: host?.updatedAt ?? null,
resourceKey: request.resourceKey ?? null,
input,
};
await repository.updateRunFrom(run.id, "created", {
status: "pending_approval",
summary: decision.reason ?? null,
sealedInput: await encryptSystemSecret(JSON.stringify(snapshot)),
});
const minutes = decision.expiresInMinutes ?? DEFAULT_APPROVAL_MINUTES;
const approvalId = randomUUID();
await repository.insertApproval({
id: approvalId,
runId: run.id,
requestedBy: request.userId,
reason: decision.reason ?? null,
allowSelf: decision.allowSelf === true || canSelfApprove(request.origin),
status: "pending",
createdAt: run.createdAt,
expiresAt: new Date(Date.now() + minutes * 60_000).toISOString(),
});
void emitOperationEvent({
eventId: "operation.approval-requested",
pluginId: prepared.registration.pluginId,
hostId: request.hostId,
severity: "warning",
payload: { actionId: action.id, approvalId },
runId: run.id,
});
return { runId: run.id, status: "pending_approval", approvalId };
}
/** Runs an action, or parks it for approval. Throws OperationError if refused. */
export async function runOperation(
request: OperationRequest,
): Promise<OperationOutcome> {
const registration = getAction(request.actionId);
if (registration) checkKeyScope(request, registration.action);
const repository = createCurrentOperationRepository();
if (request.idempotencyKey) {
const earlier = await repository.findByIdempotencyKey(
request.userId,
request.idempotencyKey,
);
if (earlier) return outcomeOf(earlier);
}
const prepared = await prepare(request);
const { action } = prepared;
const dryRun = request.dryRun === true;
const isRead = action.risk === "read";
if (action.risk === "destructive" && !dryRun && !request.confirmed) {
throw new OperationError("NEEDS_CONFIRM", "This action needs a confirm");
}
let decision: ApprovalDecision = { decision: "allow" };
if (!dryRun && !request.internal) {
decision = await evaluateApproval({
actionId: action.id,
pluginId: prepared.registration.pluginId,
risk: action.risk,
effectiveRisk: prepared.effectiveRisk,
origin: request.origin,
userId: request.userId,
hostId: request.hostId,
hostTags: prepared.host?.tags,
batchId: request.batchId,
});
}
const username = await usernameOf(request.userId);
if (decision.decision === "deny") {
auditRun(
request,
username,
action,
prepared.host,
false,
decision.reason,
"operation_denied",
);
throw new OperationError(
"DENIED",
decision.reason ?? "A rule refused this action",
);
}
// Internal reads poll constantly and would flood history and the save loop.
if (request.internal && isRead) {
return execute(prepared, request, null, username);
}
const run = {
id: randomUUID(),
actionId: action.id,
actionVersion: action.version ?? 1,
pluginId: prepared.registration.pluginId,
pluginVersion: prepared.pluginVersion,
userId: request.userId,
username,
origin: originText(request.origin),
hostId: request.hostId ?? null,
hostName: prepared.host?.name ?? null,
batchId: request.batchId ?? null,
dryRun,
risk: action.risk,
status: decision.decision === "approve" ? "created" : "running",
input: JSON.stringify(redactFields(action.inputs, prepared.input)),
undoOf: request.undoOf ?? null,
idempotencyKey: request.idempotencyKey ?? null,
createdAt: now(),
startedAt: decision.decision === "approve" ? null : now(),
};
await repository.insertRun(run);
if (decision.decision === "approve") {
return requestApproval(prepared, request, decision, run);
}
return execute(prepared, request, run.id, username);
}
function originOf(run: OperationRunRecord): ActorOrigin {
try {
return JSON.parse(run.origin) as ActorOrigin;
} catch {
return { kind: "api" };
}
}
async function failStale(run: OperationRunRecord, why: string): Promise<void> {
await createCurrentOperationRepository().updateRunFrom(
run.id,
"pending_approval",
{
status: "rejected",
error: why,
sealedInput: null,
finishedAt: now(),
},
);
}
/**
* Records a person's decision. One shot: the row moves from pending once, so a
* second click or a replay does nothing. What runs is the sealed copy, never
* what the caller sends now.
*/
export async function decideOperationApproval(options: {
approvalId: string;
approverId: string;
approve: boolean;
comment?: string;
impersonated?: boolean;
}): Promise<OperationOutcome> {
const { approvalId, approverId, approve, comment } = options;
const repository = createCurrentOperationRepository();
const approval = await repository.findApproval(approvalId);
if (!approval) throw new OperationError("NOT_FOUND", "Approval not found");
if (approval.status !== "pending") {
throw new OperationError("CONFLICT", "This was already decided");
}
const run = await repository.findRun(approval.runId);
if (!run) throw new OperationError("NOT_FOUND", "Run not found");
if (options.impersonated) {
throw new OperationError("FORBIDDEN", "You cannot approve as another user");
}
if (approval.requestedBy === approverId && !approval.allowSelf) {
throw new OperationError(
"FORBIDDEN",
"Someone else has to approve your request",
);
}
if (new Date(approval.expiresAt).getTime() < Date.now()) {
await repository.decideApproval(approvalId, "expired", null, null, now());
await failStale(run, "The approval expired");
throw new OperationError("STALE", "This approval expired");
}
// Approving needs the same rights as running it.
const registration = getAction(run.actionId);
if (registration) {
const permissions = PermissionManager.getInstance();
const needed = qualifyPermission(
registration.pluginId,
registration.action.permission,
);
const allowed =
(await permissions.hasPermission(approverId, needed)) &&
(run.hostId === null ||
(
await permissions.canAccessHost(
approverId,
run.hostId,
registration.action.hostAccess ?? "edit",
)
).hasAccess);
if (!allowed) {
throw new OperationError(
"FORBIDDEN",
"You need the same access as this action to approve it",
);
}
}
const won = await repository.decideApproval(
approvalId,
approve ? "approved" : "rejected",
approverId,
comment?.slice(0, 500) ?? null,
now(),
);
if (!won) throw new OperationError("CONFLICT", "This was already decided");
const username = await usernameOf(approverId);
void logAudit({
userId: approverId,
username,
action: approve ? "operation_approved" : "operation_rejected",
resourceType: "operation",
resourceId: run.actionId,
details: run.id,
success: true,
}).catch(() => {});
if (!approve) {
await repository.updateRunFrom(run.id, "pending_approval", {
status: "rejected",
sealedInput: null,
finishedAt: now(),
});
return { runId: run.id, status: "rejected" };
}
return runApproved(run, approval);
}
async function readSnapshot(
run: OperationRunRecord,
): Promise<ApprovalSnapshot | null> {
if (!run.sealedInput || !isSystemEncrypted(run.sealedInput)) return null;
try {
const snapshot = JSON.parse(
await decryptSystemSecret(run.sealedInput),
) as ApprovalSnapshot;
return snapshot.runId === run.id ? snapshot : null;
} catch {
return null;
}
}
async function runApproved(
run: OperationRunRecord,
approval: OperationApprovalRecord,
): Promise<OperationOutcome> {
const snapshot = await readSnapshot(run);
if (!snapshot || !run.userId) {
await failStale(run, "The approved request could not be read");
throw new OperationError("STALE", "The approved request could not be read");
}
const request: OperationRequest = {
actionId: snapshot.actionId,
hostId: snapshot.hostId ?? undefined,
resourceKey: snapshot.resourceKey ?? undefined,
input: snapshot.input,
confirmed: true,
batchId: run.batchId ?? undefined,
userId: run.userId,
origin: originOf(run),
};
// The requester's own rights, as of now. They may have changed.
let prepared: Prepared;
try {
prepared = await prepare(request);
} catch (error) {
await failStale(run, errorText(error));
throw error;
}
const changedSince =
prepared.pluginVersion !== snapshot.pluginVersion ||
(prepared.action.version ?? 1) !== snapshot.actionVersion ||
(prepared.host?.updatedAt ?? null) !== snapshot.hostUpdatedAt;
if (changedSince) {
await failStale(run, "The host or plugin changed after this was requested");
throw new OperationError(
"STALE",
"The host or plugin changed after this was requested",
);
}
const repository = createCurrentOperationRepository();
const claimed = await repository.updateRunFrom(run.id, "pending_approval", {
status: "running",
startedAt: now(),
});
if (!claimed) throw new OperationError("CONFLICT", "This already ran");
const outcome = await execute(prepared, request, run.id, run.username);
return { ...outcome, approvalId: approval.id };
}
export async function cancelOperationRun(
run: OperationRunRecord,
): Promise<boolean> {
const repository = createCurrentOperationRepository();
if (run.status === "pending_approval") {
const approval = await repository.findApprovalForRun(run.id);
if (approval) {
await repository.decideApproval(
approval.id,
"rejected",
null,
"cancelled",
now(),
);
}
return repository.updateRunFrom(run.id, "pending_approval", {
status: "cancelled",
sealedInput: null,
finishedAt: now(),
});
}
const running = active.get(run.id);
if (!running) return false;
running.controller.abort();
return true;
}
/**
* Called when a plugin stops, before its SSH connections are torn down. New
* runs are refused and running ones told to stop. Anything still going after
* the wait has an unknown outcome and is never retried.
*/
export async function drainPlugin(pluginId: string): Promise<void> {
draining.add(pluginId);
try {
const runs = [...active.entries()].filter(
([, entry]) => entry.pluginId === pluginId,
);
if (runs.length === 0) return;
for (const [, entry] of runs) entry.controller.abort();
await Promise.race([
Promise.all(runs.map(([, entry]) => entry.done)),
new Promise((resolve) => setTimeout(resolve, DRAIN_WAIT_MS)),
]);
const repository = createCurrentOperationRepository();
for (const [runId] of runs) {
await repository.updateRunFrom(runId, "running", {
status: "interrupted",
error: "The plugin stopped during this run. Check the host.",
sealedInput: null,
finishedAt: now(),
});
active.delete(runId);
}
} catch (error) {
pluginLogger.warn("Could not finish stopping runs", {
operation: "operation_drain",
pluginId,
error: errorText(error),
});
} finally {
draining.delete(pluginId);
}
}
export function riskAtLeast(
risk: OperationRisk,
floor: OperationRisk,
): boolean {
return riskRank(risk) >= riskRank(floor);
}
/** Test seam. */
export function clearOperationGate(): void {
active.clear();
draining.clear();
}