gecko-dev/toolkit/components/extensions/ExtensionStorageSync.jsm

1252 строки
44 KiB
JavaScript
Исходник Обычный вид История

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/. */
// TODO:
// * find out how the Chrome implementation deals with conflicts
"use strict";
/* exported extensionIdToCollectionId */
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
this.EXPORTED_SYMBOLS = ["ExtensionStorageSync", "extensionStorageSync"];
const Ci = Components.interfaces;
const Cc = Components.classes;
const Cu = Components.utils;
const Cr = Components.results;
const global = this;
Cu.importGlobalProperties(["atob", "btoa"]);
Cu.import("resource://gre/modules/AppConstants.jsm");
const KINTO_PROD_SERVER_URL = "https://webextensions.settings.services.mozilla.com/v1";
const KINTO_DEFAULT_SERVER_URL = KINTO_PROD_SERVER_URL;
const STORAGE_SYNC_ENABLED_PREF = "webextensions.storage.sync.enabled";
const STORAGE_SYNC_SERVER_URL_PREF = "webextensions.storage.sync.serverURL";
const STORAGE_SYNC_SCOPE = "sync:addon_storage";
const STORAGE_SYNC_CRYPTO_COLLECTION_NAME = "storage-sync-crypto";
const STORAGE_SYNC_CRYPTO_KEYRING_RECORD_ID = "keys";
const STORAGE_SYNC_CRYPTO_SALT_LENGTH_BYTES = 32;
const FXA_OAUTH_OPTIONS = {
scope: STORAGE_SYNC_SCOPE,
};
const HISTOGRAM_GET_OPS_SIZE = "STORAGE_SYNC_GET_OPS_SIZE";
const HISTOGRAM_SET_OPS_SIZE = "STORAGE_SYNC_SET_OPS_SIZE";
const HISTOGRAM_REMOVE_OPS = "STORAGE_SYNC_REMOVE_OPS";
const SCALAR_EXTENSIONS_USING = "storage.sync.api.usage.extensions_using";
const SCALAR_ITEMS_STORED = "storage.sync.api.usage.items_stored";
const SCALAR_STORAGE_CONSUMED = "storage.sync.api.usage.storage_consumed";
// Default is 5sec, which seems a bit aggressive on the open internet
const KINTO_REQUEST_TIMEOUT = 30000;
Cu.import("resource://gre/modules/XPCOMUtils.jsm");
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
Cu.import("resource://gre/modules/ExtensionUtils.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "AsyncShutdown",
"resource://gre/modules/AsyncShutdown.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "BulkKeyBundle",
"resource://services-sync/keys.js");
XPCOMUtils.defineLazyModuleGetter(this, "CollectionKeyManager",
"resource://services-sync/record.js");
XPCOMUtils.defineLazyModuleGetter(this, "CommonUtils",
"resource://services-common/utils.js");
XPCOMUtils.defineLazyModuleGetter(this, "CryptoUtils",
"resource://services-crypto/utils.js");
XPCOMUtils.defineLazyModuleGetter(this, "ExtensionStorage",
"resource://gre/modules/ExtensionStorage.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "fxAccounts",
"resource://gre/modules/FxAccounts.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "KintoHttpClient",
"resource://services-common/kinto-http-client.js");
XPCOMUtils.defineLazyModuleGetter(this, "Kinto",
"resource://services-common/kinto-offline-client.js");
XPCOMUtils.defineLazyModuleGetter(this, "FirefoxAdapter",
"resource://services-common/kinto-storage-adapter.js");
XPCOMUtils.defineLazyModuleGetter(this, "Log",
"resource://gre/modules/Log.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "Observers",
"resource://services-common/observers.js");
XPCOMUtils.defineLazyModuleGetter(this, "Services",
"resource://gre/modules/Services.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "Sqlite",
"resource://gre/modules/Sqlite.jsm");
XPCOMUtils.defineLazyModuleGetter(this, "Svc",
"resource://services-sync/util.js");
XPCOMUtils.defineLazyModuleGetter(this, "Utils",
"resource://services-sync/util.js");
XPCOMUtils.defineLazyPreferenceGetter(this, "prefPermitsStorageSync",
STORAGE_SYNC_ENABLED_PREF, true);
XPCOMUtils.defineLazyPreferenceGetter(this, "prefStorageSyncServerURL",
STORAGE_SYNC_SERVER_URL_PREF,
KINTO_DEFAULT_SERVER_URL);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const {
runSafeSyncWithoutClone,
} = ExtensionUtils;
// Map of Extensions to Set<Contexts> to track contexts that are still
// "live" and use storage.sync.
const extensionContexts = new Map();
// Borrow logger from Sync.
const log = Log.repository.getLogger("Sync.Engine.Extension-Storage");
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
// A global that is fxAccounts, or null if (as on android) fxAccounts
// isn't available.
let _fxaService = null;
if (AppConstants.platform != "android") {
_fxaService = fxAccounts;
}
class ServerKeyringDeleted extends Error {
constructor() {
super("server keyring appears to have disappeared; we were called to decrypt null");
}
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
/**
* Check for FXA and throw an exception if we don't have access.
*
* @param {Object} fxAccounts The reference we were hoping to use to
* access FxA
* @param {string} action The thing we were doing when we decided to
* see if we had access to FxA
*/
function throwIfNoFxA(fxAccounts, action) {
if (!fxAccounts) {
throw new Error(`${action} is impossible because FXAccounts is not available; are you on Android?`);
}
}
// Global ExtensionStorageSync instance that extensions and Fx Sync use.
// On Android, because there's no FXAccounts instance, any syncing
// operations will fail.
this.extensionStorageSync = null;
/**
* Utility function to enforce an order of fields when computing an HMAC.
*
* @param {KeyBundle} keyBundle The key bundle to use to compute the HMAC
* @param {string} id The record ID to use when computing the HMAC
* @param {string} IV The IV to use when computing the HMAC
* @param {string} ciphertext The ciphertext over which to compute the HMAC
* @returns {string} The computed HMAC
*/
function ciphertextHMAC(keyBundle, id, IV, ciphertext) {
const hasher = keyBundle.sha256HMACHasher;
return Utils.bytesAsHex(Utils.digestUTF8(id + IV + ciphertext, hasher));
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
/**
* Get the current user's hashed kB.
*
* @param {FXAccounts} fxaService The service to use to get the
* current user.
* @returns {string} sha256 of the user's kB as a hex string
*/
const getKBHash = async function(fxaService) {
const signedInUser = await fxaService.getSignedInUser();
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
if (!signedInUser) {
throw new Error("User isn't signed in!");
}
if (!signedInUser.kB) {
throw new Error("User doesn't have kB??");
}
let kBbytes = CommonUtils.hexToBytes(signedInUser.kB);
let hasher = Cc["@mozilla.org/security/hash;1"]
.createInstance(Ci.nsICryptoHash);
hasher.init(hasher.SHA256);
return CommonUtils.bytesAsHex(CryptoUtils.digestBytes(signedInUser.uid + kBbytes, hasher));
};
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
/**
* A "remote transformer" that the Kinto library will use to
* encrypt/decrypt records when syncing.
*
* This is an "abstract base class". Subclass this and override
* getKeys() to use it.
*/
class EncryptionRemoteTransformer {
async encode(record) {
const keyBundle = await this.getKeys();
if (record.ciphertext) {
throw new Error("Attempt to reencrypt??");
}
let id = await this.getEncodedRecordId(record);
if (!id) {
throw new Error("Record ID is missing or invalid");
}
let IV = Svc.Crypto.generateRandomIV();
let ciphertext = Svc.Crypto.encrypt(JSON.stringify(record),
keyBundle.encryptionKeyB64, IV);
let hmac = ciphertextHMAC(keyBundle, id, IV, ciphertext);
const encryptedResult = {ciphertext, IV, hmac, id};
// Copy over the _status field, so that we handle concurrency
// headers (If-Match, If-None-Match) correctly.
// DON'T copy over "deleted" status, because then we'd leak
// plaintext deletes.
encryptedResult._status = record._status == "deleted" ? "updated" : record._status;
if (record.hasOwnProperty("last_modified")) {
encryptedResult.last_modified = record.last_modified;
}
return encryptedResult;
}
async decode(record) {
if (!record.ciphertext) {
// This can happen for tombstones if a record is deleted.
if (record.deleted) {
return record;
}
throw new Error("No ciphertext: nothing to decrypt?");
}
const keyBundle = await this.getKeys();
// Authenticate the encrypted blob with the expected HMAC
let computedHMAC = ciphertextHMAC(keyBundle, record.id, record.IV, record.ciphertext);
if (computedHMAC != record.hmac) {
Utils.throwHMACMismatch(record.hmac, computedHMAC);
}
// Handle invalid data here. Elsewhere we assume that cleartext is an object.
let cleartext = Svc.Crypto.decrypt(record.ciphertext,
keyBundle.encryptionKeyB64, record.IV);
let jsonResult = JSON.parse(cleartext);
if (!jsonResult || typeof jsonResult !== "object") {
throw new Error("Decryption failed: result is <" + jsonResult + ">, not an object.");
}
if (record.hasOwnProperty("last_modified")) {
jsonResult.last_modified = record.last_modified;
}
// _status: deleted records were deleted on a client, but
// uploaded as an encrypted blob so we don't leak deletions.
// If we get such a record, flag it as deleted.
if (jsonResult._status == "deleted") {
jsonResult.deleted = true;
}
return jsonResult;
}
/**
* Retrieve keys to use during encryption.
*
* Returns a Promise<KeyBundle>.
*/
getKeys() {
throw new Error("override getKeys in a subclass");
}
/**
* Compute the record ID to use for the encoded version of the
* record.
*
* The default version just re-uses the record's ID.
*
* @param {Object} record The record being encoded.
* @returns {Promise<string>} The ID to use.
*/
getEncodedRecordId(record) {
return Promise.resolve(record.id);
}
}
global.EncryptionRemoteTransformer = EncryptionRemoteTransformer;
/**
* An EncryptionRemoteTransformer that provides a keybundle derived
* from the user's kB, suitable for encrypting a keyring.
*/
class KeyRingEncryptionRemoteTransformer extends EncryptionRemoteTransformer {
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
constructor(fxaService) {
super();
this._fxaService = fxaService;
}
getKeys() {
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
throwIfNoFxA(this._fxaService, "encrypting chrome.storage.sync records");
const self = this;
return (async function() {
const user = await self._fxaService.getSignedInUser();
// FIXME: we should permit this if the user is self-hosting
// their storage
if (!user) {
throw new Error("user isn't signed in to FxA; can't sync");
}
if (!user.kB) {
throw new Error("user doesn't have kB");
}
let kB = Utils.hexToBytes(user.kB);
let keyMaterial = CryptoUtils.hkdf(kB, undefined,
"identity.mozilla.com/picl/v1/chrome.storage.sync", 2 * 32);
let bundle = new BulkKeyBundle();
// [encryptionKey, hmacKey]
bundle.keyPair = [keyMaterial.slice(0, 32), keyMaterial.slice(32, 64)];
return bundle;
})();
}
// Pass through the kbHash field from the unencrypted record. If
// encryption fails, we can use this to try to detect whether we are
// being compromised or if the record here was encoded with a
// different kB.
async encode(record) {
const encoded = await super.encode(record);
encoded.kbHash = record.kbHash;
return encoded;
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async decode(record) {
if (record === null) {
// XXX: This is a hack that detects a situation that should
// never happen by using a technique that shouldn't actually
// work. See
// https://bugzilla.mozilla.org/show_bug.cgi?id=1359879 for
// the whole gory story.
//
// For reasons that aren't clear yet,
// sometimes the server-side keyring is deleted. When we try
// to sync our copy of the keyring, we get a conflict with the
// deleted version. Due to a bug in kinto.js, we are called to
// decode the deleted version, which is represented as
// null. For now, try to handle this by throwing a specific
// kind of exception which we can catch and recover from the
// same way we would do with any other kind of undecipherable
// keyring -- wiping the bucket and reuploading everything.
//
// Eventually we will probably fix the bug in kinto.js, and
// this will have to move somewhere else, probably in the code
// that detects a resolved conflict.
throw new ServerKeyringDeleted();
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
try {
return await super.decode(record);
} catch (e) {
if (Utils.isHMACMismatch(e)) {
const currentKBHash = await getKBHash(this._fxaService);
if (record.kbHash != currentKBHash) {
// Some other client encoded this with a kB that we don't
// have access to.
KeyRingEncryptionRemoteTransformer.throwOutdatedKB(currentKBHash, record.kbHash);
}
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
throw e;
}
}
// Generator and discriminator for KB-is-outdated exceptions.
static throwOutdatedKB(shouldBe, is) {
throw new Error(`kB hash on record is outdated: should be ${shouldBe}, is ${is}`);
}
static isOutdatedKB(exc) {
const kbMessage = "kB hash on record is outdated: ";
return exc && exc.message && exc.message.indexOf &&
(exc.message.indexOf(kbMessage) == 0);
}
}
global.KeyRingEncryptionRemoteTransformer = KeyRingEncryptionRemoteTransformer;
/**
* A Promise that centralizes initialization of ExtensionStorageSync.
*
* This centralizes the use of the Sqlite database, to which there is
* only one connection which is shared by all threads.
*
* Fields in the object returned by this Promise:
*
* - connection: a Sqlite connection. Meant for internal use only.
* - kinto: a KintoBase object, suitable for using in Firefox. All
* collections in this database will use the same Sqlite connection.
*/
const storageSyncInit = (async function() {
const path = "storage-sync.sqlite";
const opts = {path, sharedMemoryCache: false};
const connection = await Sqlite.openConnection(opts);
await FirefoxAdapter._init(connection);
return {
connection,
kinto: new Kinto({
adapter: FirefoxAdapter,
adapterOptions: {sqliteHandle: connection},
timeout: KINTO_REQUEST_TIMEOUT,
}),
};
})();
AsyncShutdown.profileBeforeChange.addBlocker(
"ExtensionStorageSync: close Sqlite handle",
async function() {
const ret = await storageSyncInit;
const {connection} = ret;
await connection.close();
}
);
// Kinto record IDs have two condtions:
//
// - They must contain only ASCII alphanumerics plus - and _. To fix
// this, we encode all non-letters using _C_, where C is the
// percent-encoded character, so space becomes _20_
// and underscore becomes _5F_.
//
// - They must start with an ASCII letter. To ensure this, we prefix
// all keys with "key-".
function keyToId(key) {
function escapeChar(match) {
return "_" + match.codePointAt(0).toString(16).toUpperCase() + "_";
}
return "key-" + key.replace(/[^a-zA-Z0-9]/g, escapeChar);
}
// Convert a Kinto ID back into a chrome.storage key.
// Returns null if a key couldn't be parsed.
function idToKey(id) {
function unescapeNumber(match, group1) {
return String.fromCodePoint(parseInt(group1, 16));
}
// An escaped ID should match this regex.
// An escaped ID should consist of only letters and numbers, plus
// code points escaped as _[0-9a-f]+_.
const ESCAPED_ID_FORMAT = /^(?:[a-zA-Z0-9]|_[0-9A-F]+_)*$/;
if (!id.startsWith("key-")) {
return null;
}
const unprefixed = id.slice(4);
// Verify that the ID is the correct format.
if (!ESCAPED_ID_FORMAT.test(unprefixed)) {
return null;
}
return unprefixed.replace(/_([0-9A-F]+)_/g, unescapeNumber);
}
// An "id schema" used to validate Kinto IDs and generate new ones.
const storageSyncIdSchema = {
// We should never generate IDs; chrome.storage only acts as a
// key-value store, so we should always have a key.
generate() {
throw new Error("cannot generate IDs");
},
// See keyToId and idToKey for more details.
validate(id) {
return idToKey(id) !== null;
},
};
// An "id schema" used for the system collection, which doesn't
// require validation or generation of IDs.
const cryptoCollectionIdSchema = {
generate() {
throw new Error("cannot generate IDs for system collection");
},
validate(id) {
return true;
},
};
/**
* Wrapper around the crypto collection providing some handy utilities.
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
class CryptoCollection {
constructor(fxaService) {
this._fxaService = fxaService;
}
async getCollection() {
throwIfNoFxA(this._fxaService, "tried to access cryptoCollection");
const {kinto} = await storageSyncInit;
return kinto.collection(STORAGE_SYNC_CRYPTO_COLLECTION_NAME, {
idSchema: cryptoCollectionIdSchema,
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
remoteTransformers: [new KeyRingEncryptionRemoteTransformer(this._fxaService)],
});
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Generate a new salt for use in hashing extension and record
* IDs.
*
* @returns {string} A base64-encoded string of the salt
*/
getNewSalt() {
return btoa(CryptoUtils.generateRandomBytes(STORAGE_SYNC_CRYPTO_SALT_LENGTH_BYTES));
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Retrieve the keyring record from the crypto collection.
*
* You can use this if you want to check metadata on the keyring
* record rather than use the keyring itself.
*
* The keyring record, if present, should have the structure:
*
* - kbHash: a hash of the user's kB. When this changes, we will
* try to sync the collection.
* - uuid: a record identifier. This will only change when we wipe
* the collection (due to kB getting reset).
* - keys: a "WBO" form of a CollectionKeyManager.
* - salts: a normal JS Object with keys being collection IDs and
* values being base64-encoded salts to use when hashing IDs
* for that collection.
* @returns {Promise<Object>}
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async getKeyRingRecord() {
const collection = await this.getCollection();
const cryptoKeyRecord = await collection.getAny(STORAGE_SYNC_CRYPTO_KEYRING_RECORD_ID);
let data = cryptoKeyRecord.data;
if (!data) {
// This is a new keyring. Invent an ID for this record. If this
// changes, it means a client replaced the keyring, so we need to
// reupload everything.
const uuidgen = Cc["@mozilla.org/uuid-generator;1"].getService(Ci.nsIUUIDGenerator);
const uuid = uuidgen.generateUUID().toString();
data = {uuid, id: STORAGE_SYNC_CRYPTO_KEYRING_RECORD_ID};
}
return data;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async getSalts() {
const cryptoKeyRecord = await this.getKeyRingRecord();
return cryptoKeyRecord && cryptoKeyRecord.salts;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Used for testing with a known salt.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
*
* @param {string} extensionId The extension ID for which to set a
* salt.
* @param {string} salt The salt to use for this extension, as a
* base64-encoded salt.
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async _setSalt(extensionId, salt) {
const cryptoKeyRecord = await this.getKeyRingRecord();
cryptoKeyRecord.salts = cryptoKeyRecord.salts || {};
cryptoKeyRecord.salts[extensionId] = salt;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await this.upsert(cryptoKeyRecord);
}
/**
* Hash an extension ID for a given user so that an attacker can't
* identify the extensions a user has installed.
*
* The extension ID is assumed to be a string (i.e. series of
* code points), and its UTF8 encoding is prefixed with the salt
* for that collection and hashed.
*
* The returned hash must conform to the syntax for Kinto
* identifiers, which (as of this writing) must match
* [a-zA-Z0-9][a-zA-Z0-9_-]*. We thus encode the hash using
* "base64-url" without padding (so that we don't get any equals
* signs (=)). For fear that a hash could start with a hyphen
* (-) or an underscore (_), prefix it with "ext-".
*
* @param {string} extensionId The extension ID to obfuscate.
* @returns {Promise<bytestring>} A collection ID suitable for use to sync to.
*/
extensionIdToCollectionId(extensionId) {
return this.hashWithExtensionSalt(CommonUtils.encodeUTF8(extensionId), extensionId)
.then(hash => `ext-${hash}`);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Hash some value with the salt for the given extension.
*
* The value should be a "bytestring", i.e. a string whose
* "characters" are values, each within [0, 255]. You can produce
* such a bytestring using e.g. CommonUtils.encodeUTF8.
*
* The returned value is a base64url-encoded string of the hash.
*
* @param {bytestring} value The value to be hashed.
* @param {string} extensionId The ID of the extension whose salt
* we should use.
* @returns {Promise<bytestring>} The hashed value.
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async hashWithExtensionSalt(value, extensionId) {
const salts = await this.getSalts();
const saltBase64 = salts && salts[extensionId];
if (!saltBase64) {
// This should never happen; salts should be populated before
// we need them by ensureCanSync.
throw new Error(`no salt available for ${extensionId}; how did this happen?`);
}
const hasher = Cc["@mozilla.org/security/hash;1"]
.createInstance(Ci.nsICryptoHash);
hasher.init(hasher.SHA256);
const salt = atob(saltBase64);
const message = `${salt}\x00${value}`;
const hash = CryptoUtils.digestBytes(message, hasher);
return CommonUtils.encodeBase64URL(hash, false);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Retrieve the actual keyring from the crypto collection.
*
* @returns {Promise<CollectionKeyManager>}
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async getKeyRing() {
const cryptoKeyRecord = await this.getKeyRingRecord();
const collectionKeys = new CollectionKeyManager();
if (cryptoKeyRecord.keys) {
collectionKeys.setContents(cryptoKeyRecord.keys, cryptoKeyRecord.last_modified);
} else {
// We never actually use the default key, so it's OK if we
// generate one multiple times.
collectionKeys.generateDefaultKey();
}
// Pass through uuid field so that we can save it if we need to.
collectionKeys.uuid = cryptoKeyRecord.uuid;
return collectionKeys;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async updateKBHash(kbHash) {
const coll = await this.getCollection();
await coll.update({id: STORAGE_SYNC_CRYPTO_KEYRING_RECORD_ID,
kbHash: kbHash},
{patch: true});
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async upsert(record) {
const collection = await this.getCollection();
await collection.upsert(record);
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async sync(extensionStorageSync) {
const collection = await this.getCollection();
return await extensionStorageSync._syncCollection(collection, {
strategy: "server_wins",
});
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Reset sync status for ALL collections by directly
* accessing the FirefoxAdapter.
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async resetSyncStatus() {
const coll = await this.getCollection();
await coll.db.resetSyncStatus();
}
// Used only for testing.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async _clear() {
const collection = await this.getCollection();
await collection.clear();
}
}
this.CryptoCollection = CryptoCollection;
/**
* An EncryptionRemoteTransformer for extension records.
*
* It uses the special "keys" record to find a key for a given
* extension, thus its name
* CollectionKeyEncryptionRemoteTransformer.
*
* Also, during encryption, it will replace the ID of the new record
* with a hashed ID, using the salt for this collection.
*
* @param {string} extensionId The extension ID for which to find a key.
*/
let CollectionKeyEncryptionRemoteTransformer = class extends EncryptionRemoteTransformer {
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
constructor(cryptoCollection, extensionId) {
super();
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
this.cryptoCollection = cryptoCollection;
this.extensionId = extensionId;
}
async getKeys() {
// FIXME: cache the crypto record for the duration of a sync cycle?
const collectionKeys = await this.cryptoCollection.getKeyRing();
if (!collectionKeys.hasKeysFor([this.extensionId])) {
// This should never happen. Keys should be created (and
// synced) at the beginning of the sync cycle.
throw new Error(`tried to encrypt records for ${this.extensionId}, but key is not present`);
}
return collectionKeys.keyForCollection(this.extensionId);
}
getEncodedRecordId(record) {
// It isn't really clear whether kinto.js record IDs are
// bytestrings or strings that happen to only contain ASCII
// characters, so encode them to be sure.
const id = CommonUtils.encodeUTF8(record.id);
// Like extensionIdToCollectionId, the rules about Kinto record
// IDs preclude equals signs or strings starting with a
// non-alphanumeric, so prefix all IDs with a constant "id-".
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return this.cryptoCollection.hashWithExtensionSalt(id, this.extensionId)
.then(hash => `id-${hash}`);
}
};
global.CollectionKeyEncryptionRemoteTransformer = CollectionKeyEncryptionRemoteTransformer;
/**
* Clean up now that one context is no longer using this extension's collection.
*
* @param {Extension} extension
* The extension whose context just ended.
* @param {Context} context
* The context that just ended.
*/
function cleanUpForContext(extension, context) {
const contexts = extensionContexts.get(extension);
if (!contexts) {
Cu.reportError(new Error(`Internal error: cannot find any contexts for extension ${extension.id}`));
}
contexts.delete(context);
if (contexts.size === 0) {
// Nobody else is using this collection. Clean up.
extensionContexts.delete(extension);
}
}
/**
* Generate a promise that produces the Collection for an extension.
*
* @param {CryptoCollection} cryptoCollection
* @param {Extension} extension
* The extension whose collection needs to
* be opened.
* @param {Context} context
* The context for this extension. The Collection
* will shut down automatically when all contexts
* close.
* @returns {Promise<Collection>}
*/
const openCollection = async function(cryptoCollection, extension, context) {
let collectionId = extension.id;
const {kinto} = await storageSyncInit;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const remoteTransformers = [new CollectionKeyEncryptionRemoteTransformer(cryptoCollection, extension.id)];
const coll = kinto.collection(collectionId, {
idSchema: storageSyncIdSchema,
remoteTransformers,
});
return coll;
};
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
class ExtensionStorageSync {
/**
* @param {FXAccounts} fxaService (Optional) If not
* present, trying to sync will fail.
* @param {nsITelemetry} telemetry Telemetry service to use to
* report sync usage.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
*/
constructor(fxaService, telemetry) {
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
this._fxaService = fxaService;
this._telemetry = telemetry;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
this.cryptoCollection = new CryptoCollection(fxaService);
this.listeners = new WeakMap();
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async syncAll() {
const extensions = extensionContexts.keys();
const extIds = Array.from(extensions, extension => extension.id);
log.debug(`Syncing extension settings for ${JSON.stringify(extIds)}`);
if (extIds.length == 0) {
// No extensions to sync. Get out.
return;
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await this.ensureCanSync(extIds);
await this.checkSyncKeyRing();
const promises = Array.from(extensionContexts.keys(), extension => {
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return openCollection(this.cryptoCollection, extension).then(coll => {
return this.sync(extension, coll);
});
});
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await Promise.all(promises);
// This needs access to an adapter, but any adapter will do
const collection = await this.cryptoCollection.getCollection();
const storage = await collection.db.calculateStorage();
this._telemetry.scalarSet(SCALAR_EXTENSIONS_USING, storage.length);
for (let {collectionName, size, numRecords} of storage) {
this._telemetry.keyedScalarSet(SCALAR_ITEMS_STORED, collectionName, numRecords);
this._telemetry.keyedScalarSet(SCALAR_STORAGE_CONSUMED, collectionName, size);
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async sync(extension, collection) {
throwIfNoFxA(this._fxaService, "syncing chrome.storage.sync");
const signedInUser = await this._fxaService.getSignedInUser();
if (!signedInUser) {
// FIXME: this should support syncing to self-hosted
log.info("User was not signed into FxA; cannot sync");
throw new Error("Not signed in to FxA");
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const collectionId = await this.cryptoCollection.extensionIdToCollectionId(extension.id);
let syncResults;
try {
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
syncResults = await this._syncCollection(collection, {
strategy: "client_wins",
collection: collectionId,
});
} catch (err) {
log.warn("Syncing failed", err);
throw err;
}
let changes = {};
for (const record of syncResults.created) {
changes[record.key] = {
newValue: record.data,
};
}
for (const record of syncResults.updated) {
// N.B. It's safe to just pick old.key because it's not
// possible to "rename" a record in the storage.sync API.
const key = record.old.key;
changes[key] = {
oldValue: record.old.data,
newValue: record.new.data,
};
}
for (const record of syncResults.deleted) {
changes[record.key] = {
oldValue: record.data,
};
}
for (const conflict of syncResults.resolved) {
// FIXME: We can't send a "changed" notification because
// kinto.js only provides the newly-resolved value. But should
// we even send a notification? We use CLIENT_WINS so nothing
// has really "changed" on this end. (The change will come on
// the other end when it pulls down the update, which is handled
// by the "updated" case above.) If we are going to send a
// notification, what best values for "old" and "new"? This
// might violate client code's assumptions, since from their
// perspective, we were in state L, but this diff is from R ->
// L.
changes[conflict.key] = {
newValue: conflict.data,
};
}
if (Object.keys(changes).length > 0) {
this.notifyListeners(extension, changes);
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Utility function that handles the common stuff about syncing all
* Kinto collections (including "meta" collections like the crypto
* one).
*
* @param {Collection} collection
* @param {Object} options
* Additional options to be passed to sync().
* @returns {Promise<SyncResultObject>}
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async _syncCollection(collection, options) {
// FIXME: this should support syncing to self-hosted
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return await this._requestWithToken(`Syncing ${collection.name}`, async function(token) {
const allOptions = Object.assign({}, {
remote: prefStorageSyncServerURL,
headers: {
Authorization: "Bearer " + token,
},
}, options);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return await collection.sync(allOptions);
});
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
// Make a Kinto request with a current FxA token.
// If the response indicates that the token might have expired,
// retry the request.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async _requestWithToken(description, f) {
throwIfNoFxA(this._fxaService, "making remote requests from chrome.storage.sync");
const fxaToken = await this._fxaService.getOAuthToken(FXA_OAUTH_OPTIONS);
try {
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return await f(fxaToken);
} catch (e) {
log.error(`${description}: request failed`, e);
if (e && e.response && e.response.status == 401) {
// Our token might have expired. Refresh and retry.
log.info("Token might have expired");
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await this._fxaService.removeCachedOAuthToken({token: fxaToken});
const newToken = await this._fxaService.getOAuthToken(FXA_OAUTH_OPTIONS);
// If this fails too, let it go.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return await f(newToken);
}
// Otherwise, we don't know how to handle this error, so just reraise.
throw e;
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Helper similar to _syncCollection, but for deleting the user's bucket.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
*
* @returns {Promise<void>}
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async _deleteBucket() {
log.error("Deleting default bucket and everything in it");
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return await this._requestWithToken("Clearing server", async function(token) {
const headers = {Authorization: "Bearer " + token};
const kintoHttp = new KintoHttpClient(prefStorageSyncServerURL, {
headers: headers,
timeout: KINTO_REQUEST_TIMEOUT,
});
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return await kintoHttp.deleteBucket("default");
});
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async ensureSaltsFor(keysRecord, extIds) {
const newSalts = Object.assign({}, keysRecord.salts);
for (let collectionId of extIds) {
if (newSalts[collectionId]) {
continue;
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
newSalts[collectionId] = this.cryptoCollection.getNewSalt();
}
return newSalts;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Check whether the keys record (provided) already has salts for
* all the extensions given in extIds.
*
* @param {Object} keysRecord A previously-retrieved keys record.
* @param {Array<string>} extIds The IDs of the extensions which
* need salts.
* @returns {boolean}
*/
hasSaltsFor(keysRecord, extIds) {
if (!keysRecord.salts) {
return false;
}
for (let collectionId of extIds) {
if (!keysRecord.salts[collectionId]) {
return false;
}
}
return true;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Recursive promise that terminates when our local collectionKeys,
* as well as that on the server, have keys for all the extensions
* in extIds.
*
* @param {Array<string>} extIds
* The IDs of the extensions which need keys.
* @returns {Promise<CollectionKeyManager>}
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async ensureCanSync(extIds) {
const keysRecord = await this.cryptoCollection.getKeyRingRecord();
const collectionKeys = await this.cryptoCollection.getKeyRing();
if (collectionKeys.hasKeysFor(extIds) && this.hasSaltsFor(keysRecord, extIds)) {
return collectionKeys;
}
log.info(`Need to create keys and/or salts for ${JSON.stringify(extIds)}`);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const kbHash = await getKBHash(this._fxaService);
const newKeys = await collectionKeys.ensureKeysFor(extIds);
const newSalts = await this.ensureSaltsFor(keysRecord, extIds);
const newRecord = {
id: STORAGE_SYNC_CRYPTO_KEYRING_RECORD_ID,
keys: newKeys.asWBO().cleartext,
salts: newSalts,
uuid: collectionKeys.uuid,
// Add a field for the current kB hash.
kbHash: kbHash,
};
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await this.cryptoCollection.upsert(newRecord);
const result = await this._syncKeyRing(newRecord);
if (result.resolved.length != 0) {
// We had a conflict which was automatically resolved. We now
// have a new keyring which might have keys for the
// collections. Recurse.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return await this.ensureCanSync(extIds);
}
// No conflicts. We're good.
return newKeys;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Update the kB in the crypto record.
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async updateKeyRingKB() {
throwIfNoFxA(this._fxaService, "use of chrome.storage.sync \"keyring\"");
const signedInUser = await this._fxaService.getSignedInUser();
if (!signedInUser) {
// Although this function is meant to be called on login,
// it's not unreasonable to check any time, even if we aren't
// logged in.
//
// If we aren't logged in, we don't have any information about
// the user's kB, so we can't be sure that the user changed
// their kB, so just return.
return;
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const thisKBHash = await getKBHash(this._fxaService);
await this.cryptoCollection.updateKBHash(thisKBHash);
}
/**
* Make sure the keyring is up to date and synced.
*
* This is called on syncs to make sure that we don't sync anything
* to any collection unless the key for that collection is on the
* server.
*/
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async checkSyncKeyRing() {
await this.updateKeyRingKB();
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const cryptoKeyRecord = await this.cryptoCollection.getKeyRingRecord();
if (cryptoKeyRecord && cryptoKeyRecord._status !== "synced") {
// We haven't successfully synced the keyring since the last
// change. This could be because kB changed and we touched the
// keyring, or it could be because we failed to sync after
// adding a key. Either way, take this opportunity to sync the
// keyring.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await this._syncKeyRing(cryptoKeyRecord);
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async _syncKeyRing(cryptoKeyRecord) {
throwIfNoFxA(this._fxaService, "syncing chrome.storage.sync \"keyring\"");
try {
// Try to sync using server_wins.
//
// We use server_wins here because whatever is on the server is
// at least consistent with itself -- the crypto in the keyring
// matches the crypto on the collection records. This is because
// we generate and upload keys just before syncing data.
//
// It's possible that we can't decode the version on the server.
// This can happen if a user is locked out of their account, and
// does a "reset password" to get in on a new device. In this
// case, we are in a bind -- we can't decrypt the record on the
// server, so we can't merge keys. If this happens, we try to
// figure out if we're the one with the correct (new) kB or if
// we just got locked out because we have the old kB. If we're
// the one with the correct kB, we wipe the server and reupload
// everything, including a new keyring.
//
// If another device has wiped the server, we need to reupload
// everything we have on our end too, so we detect this by
// adding a UUID to the keyring. UUIDs are preserved throughout
// the lifetime of a keyring, so the only time a keyring UUID
// changes is when a new keyring is uploaded, which only happens
// after a server wipe. So when we get a "conflict" (resolved by
// server_wins), we check whether the server version has a new
// UUID. If so, reset our sync status, so that we'll reupload
// everything.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const result = await this.cryptoCollection.sync(this);
if (result.resolved.length > 0) {
if (result.resolved[0].uuid != cryptoKeyRecord.uuid) {
log.info(`Detected a new UUID (${result.resolved[0].uuid}, was ${cryptoKeyRecord.uuid}). Reseting sync status for everything.`);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await this.cryptoCollection.resetSyncStatus();
// Server version is now correct. Return that result.
return result;
}
}
// No conflicts, or conflict was just someone else adding keys.
return result;
} catch (e) {
if (KeyRingEncryptionRemoteTransformer.isOutdatedKB(e) ||
e instanceof ServerKeyringDeleted ||
// This is another way that ServerKeyringDeleted can
// manifest; see bug 1350088 for more details.
e.message == "Server has been flushed.") {
// Check if our token is still valid, or if we got locked out
// between starting the sync and talking to Kinto.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const isSessionValid = await this._fxaService.sessionStatus();
if (isSessionValid) {
log.error("Couldn't decipher old keyring; deleting the default bucket and resetting sync status");
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await this._deleteBucket();
await this.cryptoCollection.resetSyncStatus();
// Reupload our keyring, which is the only new keyring.
// We don't want client_wins here because another device
// could have uploaded another keyring in the meantime.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return await this.cryptoCollection.sync(this);
}
}
throw e;
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
/**
* Get the collection for an extension, and register the extension
* as being "in use".
*
* @param {Extension} extension
* The extension for which we are seeking
* a collection.
* @param {Context} context
* The context of the extension, so that we can
* stop syncing the collection when the extension ends.
* @returns {Promise<Collection>}
*/
getCollection(extension, context) {
if (prefPermitsStorageSync !== true) {
return Promise.reject({message: `Please set ${STORAGE_SYNC_ENABLED_PREF} to true in about:config`});
}
// Register that the extension and context are in use.
if (!extensionContexts.has(extension)) {
extensionContexts.set(extension, new Set());
}
const contexts = extensionContexts.get(extension);
if (!contexts.has(context)) {
// New context. Register it and make sure it cleans itself up
// when it closes.
contexts.add(context);
context.callOnClose({
close: () => cleanUpForContext(extension, context),
});
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
return openCollection(this.cryptoCollection, extension, context);
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async set(extension, items, context) {
const coll = await this.getCollection(extension, context);
const keys = Object.keys(items);
const ids = keys.map(keyToId);
const histogramSize = this._telemetry.getKeyedHistogramById(HISTOGRAM_SET_OPS_SIZE);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const changes = await coll.execute(txn => {
let changes = {};
for (let [i, key] of keys.entries()) {
const id = ids[i];
let item = items[key];
histogramSize.add(extension.id, JSON.stringify(item).length);
let {oldRecord} = txn.upsert({
id,
key,
data: item,
});
changes[key] = {
newValue: item,
};
if (oldRecord && oldRecord.data) {
// Extract the "data" field from the old record, which
// represents the value part of the key-value store
changes[key].oldValue = oldRecord.data;
}
}
return changes;
}, {preloadIds: ids});
this.notifyListeners(extension, changes);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async remove(extension, keys, context) {
const coll = await this.getCollection(extension, context);
keys = [].concat(keys);
const ids = keys.map(keyToId);
let changes = {};
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await coll.execute(txn => {
for (let [i, key] of keys.entries()) {
const id = ids[i];
const res = txn.deleteAny(id);
if (res.deleted) {
changes[key] = {
oldValue: res.data.data,
};
}
}
return changes;
}, {preloadIds: ids});
if (Object.keys(changes).length > 0) {
this.notifyListeners(extension, changes);
}
const histogram = this._telemetry.getKeyedHistogramById(HISTOGRAM_REMOVE_OPS);
histogram.add(extension.id, keys.length);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async clear(extension, context) {
// We can't call Collection#clear here, because that just clears
// the local database. We have to explicitly delete everything so
// that the deletions can be synced as well.
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const coll = await this.getCollection(extension, context);
const res = await coll.list();
const records = res.data;
const keys = records.map(record => record.key);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
await this.remove(extension, keys, context);
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
async get(extension, spec, context) {
const coll = await this.getCollection(extension, context);
const histogramSize = this._telemetry.getKeyedHistogramById(HISTOGRAM_GET_OPS_SIZE);
let keys, records;
if (spec === null) {
records = {};
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const res = await coll.list();
for (let record of res.data) {
histogramSize.add(extension.id, JSON.stringify(record.data).length);
records[record.key] = record.data;
}
return records;
}
if (typeof spec === "string") {
keys = [spec];
records = {};
} else if (Array.isArray(spec)) {
keys = spec;
records = {};
} else {
keys = Object.keys(spec);
records = Cu.cloneInto(spec, global);
}
for (let key of keys) {
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
const res = await coll.getAny(keyToId(key));
if (res.data && res.data._status != "deleted") {
histogramSize.add(extension.id, JSON.stringify(res.data.data).length);
records[res.data.key] = res.data.data;
}
}
return records;
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
addOnChangedListener(extension, listener, context) {
let listeners = this.listeners.get(extension) || new Set();
listeners.add(listener);
this.listeners.set(extension, listeners);
// Force opening the collection so that we will sync for this extension.
return this.getCollection(extension, context);
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
removeOnChangedListener(extension, listener) {
let listeners = this.listeners.get(extension);
listeners.delete(listener);
if (listeners.size == 0) {
this.listeners.delete(extension);
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
notifyListeners(extension, changes) {
Observers.notify("ext.storage.sync-changed");
let listeners = this.listeners.get(extension) || new Set();
if (listeners) {
for (let listener of listeners) {
runSafeSyncWithoutClone(listener, changes);
}
}
Bug 1321570 - use dependency injection for fxAccounts, r=kmag Get rid of the ugly hack where test code monkeypatches a singleton to convince the rest of the ExtensionStorageSync code that a fake user is logged in. Instead, take a handle to the fxAccounts service at construction time. Test code can provide any kind of fxAccounts it wants, including one that has a hard-coded user. This provokes a bunch of changes: - ExtensionStorageSync is now no longer a singleton, but a class. You have to instantiate it in order to do anything with it. A global instance extensionStorageSync is provided for use by Sync code as well as WebExtension code. - CryptoCollection is now also a class, and each ExtensionStorageSync instance has its own CryptoCollection. This dependency should maybe also be injected, but for the time being it doesn't provide us any value to do so. - Converting singletons with asynchronous methods to classes is a pain in the neck. We convert async method foo from `foo: Task.async(....)` to `async foo() { .... }`. While we're here, convert KeyRingEncryptionRemoteTransformer#decode to async/await to eliminate a need for `let self = this`. - Update Sync code and WebExtension code to use extensionStorageSync. - There's a cyclic dependency where CryptoCollection#sync depends on ExtensionStorageSync#_syncCollection which depends on CryptoCollection#getKeyRing. As a short-term hack, we now require an ExtensionStorageSync argument to CryptoCollection#sync. - KeyRingEncryptionRemoteTransformer now takes a handle to fxAccounts on construction time as well. Because this is the only EncryptionRemoteTransformer subclass that accesses fxAccounts, we can get rid of the hack where the tests monkeypatch something in the EncryptionRemoteTransformer prototype. - CollectionKeyEncryptionRemoteTransformer now takes a handle to a CryptoCollection, rather than relying on a global one. - A bunch of methods that previously assumed access to fxAccounts now check if fxAccounts is present (i.e. if we're on Android). Strictly speaking, this isn't required by this change, but it helps make certain kinds of failure a little easier to diagnose. - Update tests, including extracting a domain-specific helper method to hide the use of CollectionKeyEncryptionRemoteTransformer. We now no longer monkeypatch in our mock fxaService, but pass it to the test so that it can do whatever it wants with it. We also provide an ExtensionStorageSync instance for convenience. Access to the global cryptoCollection is now done through an ExtensionStorageSync instance. To summarize, we've gone from a situation where a bunch of singletons had implicit dependencies on other singletons in a shared global namespace, to a situation where dependencies are provided explicitly using method/constructor arguments. This highlights some of the dependencies present: - ExtensionStorageSync depends on CryptoCollection and fxAccounts if it needs to sync - Every collection created via openCollection needs a handle to CryptoCollection so it can correctly create its remote transformers - CryptoCollection needs a handle to fxAccounts so it can create its own remote transformer for its special collection Most of this is only possible, or at least much easier, because we no longer try to juggle Sqlite connections but just keep one around forever. However, please note: - CryptoCollection needs a handle to ExtensionStorageSync to actually perform syncing logic because that's where we foolishly put the logic to make requests - There's still a backing Sqlite store which is shared by everything - There's still a singleton tracking contexts that opened extensions which we manage to try to clean contexts up correctly MozReview-Commit-ID: DGIzyRTdYZ1 --HG-- extra : rebase_source : 305a98b8e1ece3ae7a5e8a2fdb5c58d8e7c496d9
2017-01-11 00:44:01 +03:00
}
}
this.ExtensionStorageSync = ExtensionStorageSync;
this.extensionStorageSync = new ExtensionStorageSync(_fxaService, Services.telemetry);