fix: drive background refresh and phishing update from alarms (closes #158)

src/background/index.js

@@ -12,13 +12,20 @@ const {
     currentNetwork,
 } = require("../shared/state");
 const { refreshBalances, getProvider } = require("../shared/balances");
-const { debugFetch } = require("../shared/log");
+const { debugFetch, log } = require("../shared/log");
 const { verifySignedTx, verifySignature } = require("../shared/approvalVerify");
 const {
     isPhishingDomain,
-    updatePhishingList,
-    startPeriodicRefresh,
+    refreshPhishingListOnSchedule,
+    initPhishingList,
 } = require("../shared/phishingDomains");
+const {
+    BALANCE_REFRESH_ALARM,
+    PHISHING_REFRESH_ALARM,
+    BALANCE_REFRESH_PERIOD_MINUTES,
+    ensureRecurringAlarms,
+    registerAlarmHandlers,
+} = require("../shared/alarms");
 
 const storageApi =
     typeof browser !== "undefined"
@@ -591,12 +598,22 @@ async function broadcastAccountsChanged() {
 // Background balance refresh: every 60 seconds when the popup isn't open.
 // When the popup IS open, its 10-second interval keeps lastBalanceRefresh
 // fresh, so this naturally skips.
-const BACKGROUND_REFRESH_INTERVAL = 60000;
+//
+// The alarm period alone sets the cadence; this guard only suppresses a
+// refresh something else has just done, so it must stay strictly shorter than
+// the period. Timed to the period it would veto every tick it gates —
+// lastBalanceRefresh is stamped after the refresh runs, so a tick one period
+// after the last one always lands inside a guard of equal length and the real
+// cadence becomes two periods. Half the period keeps it comfortably above the
+// popup's 10-second refresh, so an open popup still suppresses the background
+// job, and comfortably below the alarm period, so the schedule always wins.
+const BALANCE_REFRESH_PERIOD_MS = BALANCE_REFRESH_PERIOD_MINUTES * 60 * 1000;
+const RECENT_BALANCE_REFRESH_MS = Math.floor(BALANCE_REFRESH_PERIOD_MS / 2);
 
 async function backgroundRefresh() {
     await loadState();
     const now = Date.now();
-    if (now - (state.lastBalanceRefresh || 0) < BACKGROUND_REFRESH_INTERVAL)
+    if (now - (state.lastBalanceRefresh || 0) < RECENT_BALANCE_REFRESH_MS)
         return;
     if (state.wallets.length === 0) return;
     await refreshBalances(
@@ -609,12 +626,58 @@ async function backgroundRefresh() {
     await saveState();
 }
 
-setInterval(backgroundRefresh, BACKGROUND_REFRESH_INTERVAL);
+// Both recurring jobs run off alarms, not timers. On Chrome MV3 this file is
+// a service worker that the browser terminates after about 30 seconds idle,
+// so a setInterval would only ever survive until the first idle period and
+// module-level state does not outlive it. Alarms are held by the browser and
+// wake the worker to deliver them.
+registerAlarmHandlers({
+    [BALANCE_REFRESH_ALARM]: backgroundRefresh,
+    // The scheduled refresh, which restores persisted state on a freshly
+    // revived worker and then fetches unconditionally. The freshness guards
+    // belong to the startup path; applying them here would make the tick skip
+    // itself.
+    [PHISHING_REFRESH_ALARM]: refreshPhishingListOnSchedule,
+});
 
-// Fetch the phishing domain blocklist delta on startup and refresh every 24h.
-// The vendored blocklist is bundled at build time; this fetches only new entries.
-updatePhishingList();
-startPeriodicRefresh();
+// Everything the background context needs re-established on start. This runs
+// on a fresh install, on browser startup, and on every revival of a
+// terminated worker, so it must be idempotent: ensureRecurringAlarms() only
+// creates alarms that are missing or carrying a stale period, and
+// initPhishingList() fetches only when the persisted timestamps say the list
+// is stale.
+//
+// On a fresh install the top-level call and the onInstalled listener both run,
+// close enough together that both could see an alarm missing and create it.
+// Sharing one in-flight run makes the "create only when missing" check
+// race-free; the memo is dropped once it settles so a later onStartup runs
+// again.
+let backgroundJobsRun = null;
+
+function startBackgroundJobs() {
+    if (backgroundJobsRun) return backgroundJobsRun;
+    backgroundJobsRun = Promise.all([
+        ensureRecurringAlarms(),
+        initPhishingList(),
+    ])
+        .catch((err) => {
+            // An alarm that failed to schedule means a recurring job silently
+            // never runs again; it must not be an unhandled rejection.
+            log.errorf("background job startup failed:", err);
+        })
+        .finally(() => {
+            backgroundJobsRun = null;
+        });
+    return backgroundJobsRun;
+}
+
+if (runtime.onInstalled) {
+    runtime.onInstalled.addListener(startBackgroundJobs);
+}
+if (runtime.onStartup) {
+    runtime.onStartup.addListener(startBackgroundJobs);
+}
+startBackgroundJobs();
 
 // When approval window is closed without a response, treat as rejection
 if (windowsApi && windowsApi.onRemoved) {

src/shared/alarms.js

@@ -0,0 +1,114 @@
+// Periodic scheduling for the background context.
+//
+// The Chrome MV3 service worker is terminated after roughly 30 seconds idle,
+// which takes every setInterval/setTimeout with it. The extension alarms API
+// is the mechanism that survives: the browser holds the schedule and wakes
+// the worker to deliver onAlarm. Firefox MV2 runs a persistent background
+// page where timers would survive, but alarms behave identically there, so
+// both targets share this path and both manifests declare the "alarms"
+// permission.
+//
+// Periods are whole minutes at or above the browser-enforced one-minute
+// minimum, so nothing here is silently clamped to a slower cadence.
+//
+// Trap for anyone changing a period: each job also carries a freshness guard
+// that can veto its own scheduled tick. A guard timed to the alarm period
+// halves the real cadence, because the guard is measured from when the last
+// run finished and the alarm fires one run-duration earlier than that. Every
+// guard must therefore either be strictly shorter than the period it gates or
+// be bypassed on the scheduled tick — see backgroundRefresh() in
+// src/background/index.js and updatePhishingList() in shared/phishingDomains.js.
+
+const BALANCE_REFRESH_ALARM = "autistmask-balance-refresh";
+const PHISHING_REFRESH_ALARM = "autistmask-phishing-refresh";
+
+const MIN_ALARM_PERIOD_MINUTES = 1;
+const BALANCE_REFRESH_PERIOD_MINUTES = 1;
+const PHISHING_REFRESH_PERIOD_MINUTES = 24 * 60;
+
+// Resolved on use rather than captured at module load: the worker is torn
+// down and re-evaluated repeatedly, and tests install a stub after requiring
+// the module.
+function alarmsApi() {
+    if (typeof browser !== "undefined" && browser.alarms) return browser.alarms;
+    if (typeof chrome !== "undefined" && chrome.alarms) return chrome.alarms;
+    return null;
+}
+
+/**
+ * Create an alarm unless one with the requested period already exists.
+ *
+ * The existence check is load-bearing: creating an alarm resets its schedule,
+ * and this runs on every worker wake. Creating unconditionally would push the
+ * next fire time out on every incoming message, so a busy extension would
+ * never see the alarm fire at all.
+ *
+ * The period comparison is equally load-bearing in the other direction: an
+ * alarm created by an older version keeps its old period forever unless a
+ * changed constant re-creates it, so a period edit would never reach an
+ * existing install. Re-creating on a period change happens once and then
+ * settles into the existence check above.
+ *
+ * @param {string} name
+ * @param {number} periodInMinutes
+ * @returns {Promise<boolean>} true if the alarm was created by this call.
+ */
+async function ensureAlarm(name, periodInMinutes) {
+    const api = alarmsApi();
+    if (!api) return false;
+    const period = Math.max(periodInMinutes, MIN_ALARM_PERIOD_MINUTES);
+    const existing = await api.get(name);
+    if (existing && existing.periodInMinutes === period) return false;
+    api.create(name, {
+        periodInMinutes: period,
+        delayInMinutes: period,
+    });
+    return true;
+}
+
+/**
+ * Ensure both recurring background jobs are scheduled. Safe to call on every
+ * worker start, on onInstalled and on onStartup.
+ *
+ * @returns {Promise<{balance: boolean, phishing: boolean}>} which alarms this
+ * call had to create.
+ */
+async function ensureRecurringAlarms() {
+    const balance = await ensureAlarm(
+        BALANCE_REFRESH_ALARM,
+        BALANCE_REFRESH_PERIOD_MINUTES,
+    );
+    const phishing = await ensureAlarm(
+        PHISHING_REFRESH_ALARM,
+        PHISHING_REFRESH_PERIOD_MINUTES,
+    );
+    return { balance, phishing };
+}
+
+/**
+ * Register per-alarm handlers. One listener dispatches by alarm name so the
+ * worker only ever installs a single onAlarm listener.
+ *
+ * @param {Object<string, function>} handlers
+ * @returns {boolean} true if the listener was installed.
+ */
+function registerAlarmHandlers(handlers) {
+    const api = alarmsApi();
+    if (!api || !api.onAlarm) return false;
+    api.onAlarm.addListener((alarm) => {
+        const handler = handlers[alarm && alarm.name];
+        if (handler) handler();
+    });
+    return true;
+}
+
+module.exports = {
+    BALANCE_REFRESH_ALARM,
+    PHISHING_REFRESH_ALARM,
+    MIN_ALARM_PERIOD_MINUTES,
+    BALANCE_REFRESH_PERIOD_MINUTES,
+    PHISHING_REFRESH_PERIOD_MINUTES,
+    ensureAlarm,
+    ensureRecurringAlarms,
+    registerAlarmHandlers,
+};

src/shared/ens.js

@@ -1,6 +1,11 @@
 // Cached ENS reverse resolution.
 // Resolves addresses to ENS names via ethers provider.lookupAddress(),
 // caching results in localStorage with a 12-hour TTL.
+//
+// POPUP ONLY. localStorage does not exist in the Chrome MV3 service worker,
+// so this module must not be pulled into src/background/. Anything the
+// background context needs to cache goes in extension storage instead (see
+// shared/phishingDomains.js).
 
 const { getProvider } = require("./balances");
 const { log } = require("./log");

src/shared/phishingDomains.js

@@ -8,8 +8,14 @@
 // The domain-checker checks the in-memory delta first (fresh/recent scam
 // sites), then falls back to the vendored list.
 //
-// If the delta is under 256 KiB it is persisted to localStorage so it
-// survives extension/service-worker restarts.
+// If the delta and its fetch timestamp fit in 256 KiB they are persisted to
+// extension storage, so they survive termination of the MV3 service worker.
+// Extension storage, not localStorage: localStorage does not exist in a
+// service worker, so the previous persistence never ran on Chrome at all.
+// The stored timestamps are what keep a restarted worker from re-fetching on
+// every wake while still noticing an overdue update. Those guards apply to the
+// startup path only; the 24-hour alarm tick bypasses them, or it would veto
+// its own refresh — see updatePhishingList().
 
 const vendoredConfig = require("./phishingBlocklist.json");
 
@@ -17,7 +23,14 @@ const BLOCKLIST_URL =
     "https://raw.githubusercontent.com/MetaMask/eth-phishing-detect/main/src/config.json";
 
 const CACHE_TTL_MS = 24 * 60 * 60 * 1000; // 24 hours
-const REFRESH_INTERVAL_MS = 24 * 60 * 60 * 1000; // 24 hours
+
+// Floor on how often an unscheduled path may hit the network. The worker is
+// revived every ~30 seconds while the browser is busy, and every revival runs
+// the startup path; without a persisted record of the last attempt, any state
+// that leaves lastFetchTime unset — a fetch that failed, or a delta too large
+// to store — would download the full list on every single wake.
+const MIN_FETCH_ATTEMPT_INTERVAL_MS = 60 * 60 * 1000; // 1 hour
+
 const DELTA_STORAGE_KEY = "phishing-delta";
 const MAX_DELTA_BYTES = 256 * 1024; // 256 KiB
 
@@ -29,45 +42,104 @@ const vendoredBlacklist = new Set(
 // Delta set — only entries from live list that are NOT in vendored.
 let deltaBlacklist = new Set();
 let lastFetchTime = 0;
+let lastAttemptTime = 0;
 let fetchPromise = null;
-let refreshTimer = null;
+let loadPromise = null;
+
+// Resolved on use rather than captured at module load, so a test can install
+// a stub after requiring the module and so the popup — which has no reason to
+// touch the delta — does not fail to load where the API is absent.
+function storageApi() {
+    if (typeof browser !== "undefined" && browser.storage) {
+        return browser.storage.local;
+    }
+    if (typeof chrome !== "undefined" && chrome.storage) {
+        return chrome.storage.local;
+    }
+    return null;
+}
 
 /**
- * Load delta entries from localStorage on startup.
- * Called once during module initialization in the background script.
+ * Sanitise a timestamp read back from storage.
+ *
+ * A value in the future is permanent poison: every guard here measures elapsed
+ * time as `Date.now() - stamp` and tests only the lower bound, so a stamp a
+ * year ahead suppresses updates for a year with no path that ever clears it.
+ * Clock skew and a restored profile backup both produce one. Since these
+ * timestamps only ever gate work, discarding an impossible one is safe: it
+ * costs at most a single extra fetch and restores a sane value immediately.
+ *
+ * @param {unknown} value
+ * @returns {number} the timestamp, or 0 if it is unusable.
  */
-function loadDeltaFromStorage() {
+function sanitizeTimestamp(value) {
+    if (typeof value !== "number" || !Number.isFinite(value)) return 0;
+    if (value <= 0 || value > Date.now()) return 0;
+    return value;
+}
+
+/**
+ * Load the persisted delta and its timestamps from extension storage.
+ * Runs once per worker lifetime; every entry point funnels through
+ * ensureDeltaLoaded() so a wake from termination restores state exactly once.
+ *
+ * @returns {Promise<void>}
+ */
+async function loadDeltaFromStorage() {
+    const storage = storageApi();
+    if (!storage) return;
     try {
-        const raw = localStorage.getItem(DELTA_STORAGE_KEY);
-        if (!raw) return;
-        const data = JSON.parse(raw);
-        if (data.blacklist && Array.isArray(data.blacklist)) {
+        const result = await storage.get(DELTA_STORAGE_KEY);
+        const data = result && result[DELTA_STORAGE_KEY];
+        if (!data) return;
+        if (Array.isArray(data.blacklist)) {
             deltaBlacklist = new Set(
                 data.blacklist.map((d) => d.toLowerCase()),
             );
         }
+        lastFetchTime = sanitizeTimestamp(data.lastFetchTime);
+        lastAttemptTime = sanitizeTimestamp(data.lastAttemptTime);
     } catch {
-        // localStorage unavailable or corrupt — start empty
+        // Storage unavailable or corrupt — start empty and re-fetch.
     }
 }
 
+function ensureDeltaLoaded() {
+    if (!loadPromise) loadPromise = loadDeltaFromStorage();
+    return loadPromise;
+}
+
 /**
- * Persist delta to localStorage if it fits within MAX_DELTA_BYTES.
+ * Persist the delta and its timestamps if they fit within MAX_DELTA_BYTES.
+ *
+ * The 256 KiB cap covers the delta and its freshness claim: when the delta is
+ * too large to keep, lastFetchTime goes with it, so the next start re-fetches
+ * rather than trusting a freshness claim for a delta it no longer holds.
+ * lastAttemptTime is written either way — it records that the network was
+ * contacted, which stays true whatever became of the response, and it is what
+ * stops a permanently oversized list from downloading on every worker wake.
+ *
+ * @returns {Promise<void>}
  */
-function saveDeltaToStorage() {
+async function saveDeltaToStorage() {
+    const storage = storageApi();
+    if (!storage) return;
     try {
         const data = {
             blacklist: Array.from(deltaBlacklist),
+            lastFetchTime,
+            lastAttemptTime,
         };
         const json = JSON.stringify(data);
         if (json.length < MAX_DELTA_BYTES) {
-            localStorage.setItem(DELTA_STORAGE_KEY, json);
+            await storage.set({ [DELTA_STORAGE_KEY]: data });
+        } else if (lastAttemptTime > 0) {
+            await storage.set({ [DELTA_STORAGE_KEY]: { lastAttemptTime } });
         } else {
-            // Too large — remove stale key if present
-            localStorage.removeItem(DELTA_STORAGE_KEY);
+            await storage.remove(DELTA_STORAGE_KEY);
         }
     } catch {
-        // localStorage unavailable — skip silently
+        // Storage unavailable — skip silently
     }
 }
 
@@ -76,6 +148,7 @@ function saveDeltaToStorage() {
  * Used for both live fetches and testing.
  *
  * @param {{ blacklist?: string[] }} config
+ * @returns {Promise<void>} resolves once the delta has been persisted.
  */
 function loadConfig(config) {
     const liveBlacklist = (config.blacklist || []).map((d) => d.toLowerCase());
@@ -86,7 +159,7 @@ function loadConfig(config) {
     );
 
     lastFetchTime = Date.now();
-    saveDeltaToStorage();
+    return saveDeltaToStorage();
 }
 
 /**
@@ -111,6 +184,11 @@ function hostnameVariants(hostname) {
  * Check if a hostname is on the phishing blocklist.
  * Checks delta first (fresh/recent scam sites), then vendored list.
  *
+ * Synchronous by design — callers answer an approval prompt with it. On a
+ * worker that has just woken, the persisted delta may still be loading; the
+ * vendored list, which is bundled and always present, carries the check until
+ * it lands.
+ *
  * @param {string} hostname - The hostname to check.
  * @returns {boolean}
  */
@@ -127,28 +205,59 @@ function isPhishingDomain(hostname) {
 
 /**
  * Fetch the latest blocklist and compute delta against vendored data.
- * De-duplicates concurrent fetches. Results are cached for CACHE_TTL_MS.
+ * De-duplicates concurrent fetches. Results are cached for CACHE_TTL_MS,
+ * counted from the persisted timestamp so the cache outlives the worker.
  *
+ * `force` is what makes the 24-hour alarm actually refresh every 24 hours.
+ * The alarm fires one period after the previous alarm, but lastFetchTime is
+ * stamped when that fetch *completed*, so an unforced tick lands one fetch
+ * latency inside its own TTL, skips, and turns the real cadence into 48 hours.
+ * Shortening the TTL instead would not fix it: the worker wakes every ~30
+ * seconds and the startup path re-checks the TTL each time, so a shortened TTL
+ * simply becomes the real cadence. The TTL is there to stop redundant fetches
+ * on wake, and the scheduled tick is not redundant, so it bypasses it.
+ *
+ * @param {{force?: boolean}} [opts] force: fetch unless one is already in
+ * flight, ignoring both the freshness and the retry guard. For the scheduled
+ * alarm tick only.
  * @returns {Promise<void>}
  */
-async function updatePhishingList() {
-    // Skip if recently fetched
-    if (Date.now() - lastFetchTime < CACHE_TTL_MS && lastFetchTime > 0) {
-        return;
+async function updatePhishingList({ force = false } = {}) {
+    // A worker that has just been revived knows nothing until the persisted
+    // record is back in memory; without this the freshness check below would
+    // always see 0 and re-fetch on every wake.
+    await ensureDeltaLoaded();
+
+    if (!force) {
+        const now = Date.now();
+        // Skip if recently fetched.
+        if (lastFetchTime > 0 && now - lastFetchTime < CACHE_TTL_MS) return;
+        // Skip if the network was contacted recently and the result was not
+        // usable — a failed fetch or an oversized delta leaves lastFetchTime
+        // unset, and without this every wake would retry.
+        if (
+            lastAttemptTime > 0 &&
+            now - lastAttemptTime < MIN_FETCH_ATTEMPT_INTERVAL_MS
+        ) {
+            return;
+        }
     }
 
     // De-duplicate concurrent calls
     if (fetchPromise) return fetchPromise;
 
     fetchPromise = (async () => {
+        lastAttemptTime = Date.now();
         try {
             const resp = await fetch(BLOCKLIST_URL);
             if (!resp.ok) throw new Error("HTTP " + resp.status);
             const config = await resp.json();
-            loadConfig(config);
+            await loadConfig(config);
         } catch {
-            // Silently fail — vendored list still provides coverage.
-            // We'll retry next time.
+            // Silently fail — vendored list still provides coverage. Persist
+            // the attempt so a persistently failing fetch is retried on the
+            // schedule rather than on every wake.
+            await saveDeltaToStorage();
         } finally {
             fetchPromise = null;
         }
@@ -158,12 +267,29 @@ async function updatePhishingList() {
 }
 
 /**
- * Start periodic refresh of the phishing list.
- * Should be called once from the background script on startup.
+ * Restore persisted state and fetch if the list is overdue.
+ *
+ * Called from the background script every time it starts — a fresh install,
+ * a browser start, and every revival of a terminated service worker all land
+ * here. The recurring 24-hour schedule itself is an alarm (see
+ * shared/alarms.js), not a timer, because timers die with the worker.
+ *
+ * @returns {Promise<void>}
  */
-function startPeriodicRefresh() {
-    if (refreshTimer) return;
-    refreshTimer = setInterval(updatePhishingList, REFRESH_INTERVAL_MS);
+async function initPhishingList() {
+    await ensureDeltaLoaded();
+    return updatePhishingList();
+}
+
+/**
+ * The 24-hour alarm tick. Separate from initPhishingList() because this is the
+ * scheduled refresh and must not be vetoed by the guards that exist to keep
+ * the unscheduled startup path off the network.
+ *
+ * @returns {Promise<void>}
+ */
+async function refreshPhishingListOnSchedule() {
+    return updatePhishingList({ force: true });
 }
 
 /**
@@ -190,21 +316,22 @@ function getDeltaSize() {
 function _reset() {
     deltaBlacklist = new Set();
     lastFetchTime = 0;
+    lastAttemptTime = 0;
     fetchPromise = null;
-    if (refreshTimer) {
-        clearInterval(refreshTimer);
-        refreshTimer = null;
-    }
+    loadPromise = null;
 }
 
-// Load persisted delta on module initialization
-loadDeltaFromStorage();
-
 module.exports = {
     isPhishingDomain,
     updatePhishingList,
-    startPeriodicRefresh,
+    refreshPhishingListOnSchedule,
+    initPhishingList,
+    loadDeltaFromStorage,
     loadConfig,
+    CACHE_TTL_MS,
+    MIN_FETCH_ATTEMPT_INTERVAL_MS,
+    DELTA_STORAGE_KEY,
+    MAX_DELTA_BYTES,
     getBlocklistSize,
     getDeltaSize,
     hostnameVariants,