gecko-dev/browser/actors/PluginParent.jsm

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

742 lines
23 KiB
JavaScript
Raw Normal View History

Bug 1505913 - make plugin click-to-play and crash handling fission-compatible, r=mconley At a high level, this change does the following: - move the pluginchild actor to be a JSWindowActorChild - move the parent handling from browser-plugins into a JSWindowActorParent - move the crash handling from ContentCrashHandlers.jsm to the parent actor, using a `PluginManager` object. It needs to talk to the actors (and vice versa), so this seemed a better fit than spreading actor implementation details to other JSMs. - switch to using plugin IDs to identify plugins cross-process, instead of combinations of names or other properties of the plugin tag. As part of that, ensured plugin IDs are unique between "fake" plugins and the other ones. - drop support for having a notification for more than 1 plugin. We only support Flash, in practice, so there didn't seem to be much point in the added complexity of trying to support more than 1 thing. Some notes: - the previous implementation mixes runIDs (for NPAPI plugin process "runs") and GMP pluginIDs when doing crashreporting. AFAICT there is no guarantee these don't conflict, so I've split them out to avoid issues. There's a pluginCrashID object I pass around instead that has either a runID or pluginID. Happy to rename some more for clarity. - the previous implementation used `pluginInfo` and `plugin` for a bunch of different types of variables. I've tried to be consistent, where: * `pluginElement` is a DOM element for a plugin * `activationInfo` is a JS object used to track click to play state for a plugin * `plugin` is a plugintag as returned by the pluginhost service * `pluginCrashID` is an identifier for a crashed plugin (see previous point). - I'm still using broadcastAsyncMessage to tell the content processes about gmp plugin crashes and plugin crash submission updates, because there's no guarantee the actors are instantiated (for gmp plugins) nor can the parent easily find out which actors to talk to (for either gmp or npapi plugins). Open to suggestions there, too. I think our best bet might be moving that to IPDL-based IPC within the GMP code, but that feels like a separate bug. Differential Revision: https://phabricator.services.mozilla.com/D37665 --HG-- rename : browser/base/content/browser-plugins.js => browser/actors/PluginParent.jsm extra : moz-landing-system : lando
2019-07-23 22:04:40 +00:00
/* -*- indent-tabs-mode: nil; js-indent-level: 2 -*-
* 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/. */
"use strict";
var EXPORTED_SYMBOLS = ["PluginParent", "PluginManager"];
const { AppConstants } = ChromeUtils.import(
"resource://gre/modules/AppConstants.jsm"
);
const { Services } = ChromeUtils.import("resource://gre/modules/Services.jsm");
const { XPCOMUtils } = ChromeUtils.import(
"resource://gre/modules/XPCOMUtils.jsm"
);
XPCOMUtils.defineLazyServiceGetter(
this,
"gPluginHost",
"@mozilla.org/plugin/host;1",
"nsIPluginHost"
);
ChromeUtils.defineModuleGetter(
this,
"BrowserUtils",
"resource://gre/modules/BrowserUtils.jsm"
);
ChromeUtils.defineModuleGetter(
this,
"CrashSubmit",
"resource://gre/modules/CrashSubmit.jsm"
);
XPCOMUtils.defineLazyGetter(this, "gNavigatorBundle", function() {
const url = "chrome://browser/locale/browser.properties";
return Services.strings.createBundle(url);
});
const kNotificationId = "click-to-play-plugins";
const {
PLUGIN_ACTIVE,
PLUGIN_VULNERABLE_NO_UPDATE,
PLUGIN_VULNERABLE_UPDATABLE,
PLUGIN_CLICK_TO_PLAY_QUIET,
} = Ci.nsIObjectLoadingContent;
const PluginManager = {
_initialized: false,
pluginMap: new Map(),
crashReports: new Map(),
gmpCrashes: new Map(),
_pendingCrashQueries: new Map(),
// BrowserGlue.jsm ensures we catch all plugin crashes.
// Barring crashes, we don't need to do anything until/unless an
// actor gets instantiated, in which case we also care about the
// plugin list changing.
ensureInitialized() {
if (this._initialized) {
return;
}
this._initialized = true;
this._updatePluginMap();
Services.obs.addObserver(this, "plugins-list-updated");
Services.obs.addObserver(this, "profile-after-change");
},
destroy() {
if (!this._initialized) {
return;
}
Services.obs.removeObserver(this, "plugins-list-updated");
Services.obs.removeObserver(this, "profile-after-change");
this.crashReports = new Map();
this.gmpCrashes = new Map();
this.pluginMap = new Map();
},
observe(subject, topic, data) {
switch (topic) {
case "plugins-list-updated":
this._updatePluginMap();
break;
case "plugin-crashed":
this.ensureInitialized();
this._registerNPAPICrash(subject);
break;
case "gmp-plugin-crash":
this.ensureInitialized();
this._registerGMPCrash(subject);
break;
case "profile-after-change":
this.destroy();
break;
}
},
getPluginTagById(id) {
return this.pluginMap.get(id);
},
_updatePluginMap() {
this.pluginMap = new Map();
let plugins = gPluginHost.getPluginTags();
for (let plugin of plugins) {
this.pluginMap.set(plugin.id, plugin);
}
},
// Crashed-plugin observer. Notified once per plugin crash, before events
// are dispatched to individual plugin instances. However, because of IPC,
// the event and the observer notification may still race.
_registerNPAPICrash(subject) {
let propertyBag = subject;
if (
!(propertyBag instanceof Ci.nsIWritablePropertyBag2) ||
!propertyBag.hasKey("runID") ||
!propertyBag.hasKey("pluginName")
) {
Cu.reportError(
"A NPAPI plugin crashed, but the notification is incomplete."
);
return;
}
let runID = propertyBag.getPropertyAsUint32("runID");
let uglyPluginName = propertyBag.getPropertyAsAString("pluginName");
let pluginName = BrowserUtils.makeNicePluginName(uglyPluginName);
let pluginDumpID = propertyBag.getPropertyAsAString("pluginDumpID");
let state;
let crashReporter = Services.appinfo.QueryInterface(Ci.nsICrashReporter);
if (!AppConstants.MOZ_CRASHREPORTER || !crashReporter.enabled) {
// This state tells the user that crash reporting is disabled, so we
// cannot send a report.
state = "noSubmit";
} else if (!pluginDumpID) {
// If we don't have a minidumpID, we can't submit anything.
// This can happen if the plugin is killed from the task manager.
// This state tells the user that this is the case.
state = "noReport";
} else {
// This state asks the user to submit a crash report.
state = "please";
}
let crashInfo = { runID, state, pluginName, pluginDumpID };
Bug 1505913 - make plugin click-to-play and crash handling fission-compatible, r=mconley At a high level, this change does the following: - move the pluginchild actor to be a JSWindowActorChild - move the parent handling from browser-plugins into a JSWindowActorParent - move the crash handling from ContentCrashHandlers.jsm to the parent actor, using a `PluginManager` object. It needs to talk to the actors (and vice versa), so this seemed a better fit than spreading actor implementation details to other JSMs. - switch to using plugin IDs to identify plugins cross-process, instead of combinations of names or other properties of the plugin tag. As part of that, ensured plugin IDs are unique between "fake" plugins and the other ones. - drop support for having a notification for more than 1 plugin. We only support Flash, in practice, so there didn't seem to be much point in the added complexity of trying to support more than 1 thing. Some notes: - the previous implementation mixes runIDs (for NPAPI plugin process "runs") and GMP pluginIDs when doing crashreporting. AFAICT there is no guarantee these don't conflict, so I've split them out to avoid issues. There's a pluginCrashID object I pass around instead that has either a runID or pluginID. Happy to rename some more for clarity. - the previous implementation used `pluginInfo` and `plugin` for a bunch of different types of variables. I've tried to be consistent, where: * `pluginElement` is a DOM element for a plugin * `activationInfo` is a JS object used to track click to play state for a plugin * `plugin` is a plugintag as returned by the pluginhost service * `pluginCrashID` is an identifier for a crashed plugin (see previous point). - I'm still using broadcastAsyncMessage to tell the content processes about gmp plugin crashes and plugin crash submission updates, because there's no guarantee the actors are instantiated (for gmp plugins) nor can the parent easily find out which actors to talk to (for either gmp or npapi plugins). Open to suggestions there, too. I think our best bet might be moving that to IPDL-based IPC within the GMP code, but that feels like a separate bug. Differential Revision: https://phabricator.services.mozilla.com/D37665 --HG-- rename : browser/base/content/browser-plugins.js => browser/actors/PluginParent.jsm extra : moz-landing-system : lando
2019-07-23 22:04:40 +00:00
this.crashReports.set(runID, crashInfo);
let listeners = this._pendingCrashQueries.get(runID) || [];
for (let listener of listeners) {
listener(crashInfo);
}
this._pendingCrashQueries.delete(runID);
},
_registerGMPCrash(subject) {
let propertyBag = subject;
if (
!(propertyBag instanceof Ci.nsIWritablePropertyBag2) ||
!propertyBag.hasKey("pluginID") ||
!propertyBag.hasKey("pluginDumpID") ||
!propertyBag.hasKey("pluginName")
) {
Cu.reportError("PluginManager can not read plugin information.");
return;
}
let pluginID = propertyBag.getPropertyAsUint32("pluginID");
let pluginDumpID = propertyBag.getPropertyAsAString("pluginDumpID");
if (pluginDumpID) {
this.gmpCrashes.set(pluginID, { pluginDumpID, pluginID });
}
// Only the parent process gets the gmp-plugin-crash observer
// notification, so we need to inform any content processes that
// the GMP has crashed. This then fires PluginCrashed events in
// all the relevant windows, which will trigger child actors being
// created, which will contact us again, when we'll use the
// gmpCrashes collection to respond.
if (Services.ppmm) {
let pluginName = propertyBag.getPropertyAsAString("pluginName");
Services.ppmm.broadcastAsyncMessage("gmp-plugin-crash", {
pluginName,
pluginID,
});
}
},
/**
* Submit a crash report for a crashed NPAPI plugin.
*
* @param pluginCrashID
* An object with either a runID (for NPAPI crashes) or a pluginID
* property (for GMP plugin crashes).
* A run ID is a unique identifier for a particular run of a plugin
* process - and is analogous to a process ID (though it is managed
* by Gecko instead of the operating system).
* @param keyVals
* An object whose key-value pairs will be merged
* with the ".extra" file submitted with the report.
* The properties of htis object will override properties
* of the same name in the .extra file.
*/
submitCrashReport(pluginCrashID, keyVals = {}) {
let report = this.getCrashReport(pluginCrashID);
if (!report) {
Cu.reportError(
`Could not find plugin dump IDs for ${JSON.stringify(pluginCrashID)}.` +
`It is possible that a report was already submitted.`
);
return;
}
let { pluginDumpID } = report;
Bug 1505913 - make plugin click-to-play and crash handling fission-compatible, r=mconley At a high level, this change does the following: - move the pluginchild actor to be a JSWindowActorChild - move the parent handling from browser-plugins into a JSWindowActorParent - move the crash handling from ContentCrashHandlers.jsm to the parent actor, using a `PluginManager` object. It needs to talk to the actors (and vice versa), so this seemed a better fit than spreading actor implementation details to other JSMs. - switch to using plugin IDs to identify plugins cross-process, instead of combinations of names or other properties of the plugin tag. As part of that, ensured plugin IDs are unique between "fake" plugins and the other ones. - drop support for having a notification for more than 1 plugin. We only support Flash, in practice, so there didn't seem to be much point in the added complexity of trying to support more than 1 thing. Some notes: - the previous implementation mixes runIDs (for NPAPI plugin process "runs") and GMP pluginIDs when doing crashreporting. AFAICT there is no guarantee these don't conflict, so I've split them out to avoid issues. There's a pluginCrashID object I pass around instead that has either a runID or pluginID. Happy to rename some more for clarity. - the previous implementation used `pluginInfo` and `plugin` for a bunch of different types of variables. I've tried to be consistent, where: * `pluginElement` is a DOM element for a plugin * `activationInfo` is a JS object used to track click to play state for a plugin * `plugin` is a plugintag as returned by the pluginhost service * `pluginCrashID` is an identifier for a crashed plugin (see previous point). - I'm still using broadcastAsyncMessage to tell the content processes about gmp plugin crashes and plugin crash submission updates, because there's no guarantee the actors are instantiated (for gmp plugins) nor can the parent easily find out which actors to talk to (for either gmp or npapi plugins). Open to suggestions there, too. I think our best bet might be moving that to IPDL-based IPC within the GMP code, but that feels like a separate bug. Differential Revision: https://phabricator.services.mozilla.com/D37665 --HG-- rename : browser/base/content/browser-plugins.js => browser/actors/PluginParent.jsm extra : moz-landing-system : lando
2019-07-23 22:04:40 +00:00
let submissionPromise = CrashSubmit.submit(pluginDumpID, {
recordSubmission: true,
extraExtraKeyVals: keyVals,
});
this.broadcastState(pluginCrashID, "submitting");
submissionPromise.then(
() => {
this.broadcastState(pluginCrashID, "success");
},
() => {
this.broadcastState(pluginCrashID, "failed");
}
);
if (pluginCrashID.hasOwnProperty("runID")) {
this.crashReports.delete(pluginCrashID.runID);
} else {
this.gmpCrashes.delete(pluginCrashID.pluginID);
}
},
broadcastState(pluginCrashID, state) {
if (!pluginCrashID.hasOwnProperty("runID")) {
return;
}
let { runID } = pluginCrashID;
Services.ppmm.broadcastAsyncMessage(
"PluginParent:NPAPIPluginCrashReportSubmitted",
{ runID, state }
);
},
getCrashReport(pluginCrashID) {
if (pluginCrashID.hasOwnProperty("pluginID")) {
return this.gmpCrashes.get(pluginCrashID.pluginID);
}
return this.crashReports.get(pluginCrashID.runID);
},
/**
* Called by actors when they want crash info on behalf of the child.
* Will either return such info immediately if we have it, or return
* a promise, which resolves when we do have it. The promise resolution
* function is kept around for when we get the `plugin-crashed` observer
* notification.
*/
awaitPluginCrashInfo(runID) {
if (this.crashReports.has(runID)) {
return this.crashReports.get(runID);
}
let listeners = this._pendingCrashQueries.get(runID);
if (!listeners) {
listeners = [];
this._pendingCrashQueries.set(runID, listeners);
}
return new Promise(resolve => listeners.push(resolve));
},
/**
* This allows dependency injection, where an automated test can
* dictate how and when we respond to a child's inquiry about a crash.
* This is helpful when testing different orderings for plugin crash
* notifications (ie race conditions).
*
* Concretely, for the toplevel browsingContext of the `browser` we're
* passed, call the passed `handler` function the next time the child
* asks for crash data (using PluginContent:GetCrashData). We'll return
* the result of the function to the child. The message in question
* uses the actor query API, so promises and/or async functions will
* Just Work.
*/
mockResponse(browser, handler) {
let { currentWindowGlobal } = browser.frameLoader.browsingContext;
currentWindowGlobal.getActor("Plugin")._mockedResponder = handler;
},
};
class PluginParent extends JSWindowActorParent {
constructor() {
super();
PluginManager.ensureInitialized();
}
receiveMessage(msg) {
let browser = this.manager.rootFrameLoader.ownerElement;
let win = browser.ownerGlobal;
switch (msg.name) {
case "PluginContent:ShowClickToPlayNotification":
this.showClickToPlayNotification(
browser,
msg.data.plugin,
msg.data.showNow
);
break;
case "PluginContent:RemoveNotification":
this.removeNotification(browser);
break;
case "PluginContent:ShowPluginCrashedNotification":
this.showPluginCrashedNotification(browser, msg.data.pluginCrashID);
break;
case "PluginContent:SubmitReport":
if (AppConstants.MOZ_CRASHREPORTER) {
this.submitReport(
msg.data.runID,
msg.data.keyVals,
msg.data.submitURLOptIn
);
}
break;
case "PluginContent:LinkClickCallback":
switch (msg.data.name) {
case "managePlugins":
case "openHelpPage":
this[msg.data.name](win);
break;
case "openPluginUpdatePage":
this.openPluginUpdatePage(win, msg.data.pluginId);
break;
}
break;
case "PluginContent:GetCrashData":
if (this._mockedResponder) {
let rv = this._mockedResponder(msg.data);
delete this._mockedResponder;
return rv;
}
return PluginManager.awaitPluginCrashInfo(msg.data.runID);
default:
Cu.reportError(
"PluginParent did not expect to handle message " + msg.name
);
break;
}
return null;
}
// Callback for user clicking on a disabled plugin
managePlugins(window) {
window.BrowserOpenAddonsMgr("addons://list/plugin");
}
// Callback for user clicking on the link in a click-to-play plugin
// (where the plugin has an update)
async openPluginUpdatePage(window, pluginId) {
let pluginTag = PluginManager.getPluginTagById(pluginId);
if (!pluginTag) {
return;
}
let { Blocklist } = ChromeUtils.import(
"resource://gre/modules/Blocklist.jsm"
);
let url = await Blocklist.getPluginBlockURL(pluginTag);
window.openTrustedLinkIn(url, "tab");
}
submitReport(runID, keyVals, submitURLOptIn) {
if (!AppConstants.MOZ_CRASHREPORTER) {
return;
}
Services.prefs.setBoolPref(
"dom.ipc.plugins.reportCrashURL",
!!submitURLOptIn
);
PluginManager.submitCrashReport({ runID }, keyVals);
}
// Callback for user clicking a "reload page" link
reloadPage(browser) {
browser.reload();
}
// Callback for user clicking the help icon
openHelpPage(window) {
window.openHelpLink("plugin-crashed", false);
}
_clickToPlayNotificationEventCallback(event) {
if (event == "showing") {
Services.telemetry
.getHistogramById("PLUGINS_NOTIFICATION_SHOWN")
.add(!this.options.showNow);
} else if (event == "dismissed") {
// Once the popup is dismissed, clicking the icon should show the full
// list again
this.options.showNow = false;
}
}
/**
* Called from the plugin doorhanger to set the new permissions for a plugin
* and activate plugins if necessary.
* aNewState should be one of:
* - "allownow"
* - "block"
* - "continue"
* - "continueblocking"
*/
_updatePluginPermission(aBrowser, aActivationInfo, aNewState) {
let permission;
let histogram = Services.telemetry.getHistogramById(
"PLUGINS_NOTIFICATION_USER_ACTION_2"
);
let window = aBrowser.ownerGlobal;
let notification = window.PopupNotifications.getNotification(
kNotificationId,
aBrowser
);
// Update the permission manager.
// Also update the current state of activationInfo.fallbackType so that
// subsequent opening of the notification shows the current state.
switch (aNewState) {
case "allownow":
permission = Ci.nsIPermissionManager.ALLOW_ACTION;
histogram.add(0);
aActivationInfo.fallbackType = PLUGIN_ACTIVE;
notification.options.extraAttr = "active";
break;
case "block":
permission = Ci.nsIPermissionManager.PROMPT_ACTION;
histogram.add(2);
let pluginTag = PluginManager.getPluginTagById(aActivationInfo.id);
switch (pluginTag.blocklistState) {
case Ci.nsIBlocklistService.STATE_VULNERABLE_UPDATE_AVAILABLE:
aActivationInfo.fallbackType = PLUGIN_VULNERABLE_UPDATABLE;
break;
case Ci.nsIBlocklistService.STATE_VULNERABLE_NO_UPDATE:
aActivationInfo.fallbackType = PLUGIN_VULNERABLE_NO_UPDATE;
break;
default:
// PLUGIN_CLICK_TO_PLAY_QUIET will only last until they reload the page, at
// which point it will be PLUGIN_CLICK_TO_PLAY (the overlays will appear)
aActivationInfo.fallbackType = PLUGIN_CLICK_TO_PLAY_QUIET;
}
notification.options.extraAttr = "inactive";
break;
// In case a plugin has already been allowed/disallowed in another tab, the
// buttons matching the existing block state shouldn't change any permissions
// but should run the plugin-enablement code below.
case "continue":
aActivationInfo.fallbackType = PLUGIN_ACTIVE;
notification.options.extraAttr = "active";
break;
case "continueblocking":
aActivationInfo.fallbackType = PLUGIN_CLICK_TO_PLAY_QUIET;
notification.options.extraAttr = "inactive";
break;
default:
Cu.reportError(Error("Unexpected plugin state: " + aNewState));
return;
}
if (aNewState != "continue" && aNewState != "continueblocking") {
let { principal } = notification.options;
Services.perms.addFromPrincipal(
principal,
aActivationInfo.permissionString,
permission,
Ci.nsIPermissionManager.EXPIRE_SESSION,
0 // do not expire (only expire at the end of the session)
Bug 1505913 - make plugin click-to-play and crash handling fission-compatible, r=mconley At a high level, this change does the following: - move the pluginchild actor to be a JSWindowActorChild - move the parent handling from browser-plugins into a JSWindowActorParent - move the crash handling from ContentCrashHandlers.jsm to the parent actor, using a `PluginManager` object. It needs to talk to the actors (and vice versa), so this seemed a better fit than spreading actor implementation details to other JSMs. - switch to using plugin IDs to identify plugins cross-process, instead of combinations of names or other properties of the plugin tag. As part of that, ensured plugin IDs are unique between "fake" plugins and the other ones. - drop support for having a notification for more than 1 plugin. We only support Flash, in practice, so there didn't seem to be much point in the added complexity of trying to support more than 1 thing. Some notes: - the previous implementation mixes runIDs (for NPAPI plugin process "runs") and GMP pluginIDs when doing crashreporting. AFAICT there is no guarantee these don't conflict, so I've split them out to avoid issues. There's a pluginCrashID object I pass around instead that has either a runID or pluginID. Happy to rename some more for clarity. - the previous implementation used `pluginInfo` and `plugin` for a bunch of different types of variables. I've tried to be consistent, where: * `pluginElement` is a DOM element for a plugin * `activationInfo` is a JS object used to track click to play state for a plugin * `plugin` is a plugintag as returned by the pluginhost service * `pluginCrashID` is an identifier for a crashed plugin (see previous point). - I'm still using broadcastAsyncMessage to tell the content processes about gmp plugin crashes and plugin crash submission updates, because there's no guarantee the actors are instantiated (for gmp plugins) nor can the parent easily find out which actors to talk to (for either gmp or npapi plugins). Open to suggestions there, too. I think our best bet might be moving that to IPDL-based IPC within the GMP code, but that feels like a separate bug. Differential Revision: https://phabricator.services.mozilla.com/D37665 --HG-- rename : browser/base/content/browser-plugins.js => browser/actors/PluginParent.jsm extra : moz-landing-system : lando
2019-07-23 22:04:40 +00:00
);
}
this.sendAsyncMessage("PluginParent:ActivatePlugins", {
activationInfo: aActivationInfo,
newState: aNewState,
});
}
showClickToPlayNotification(browser, plugin, showNow) {
let window = browser.ownerGlobal;
if (!window.PopupNotifications) {
return;
}
Bug 1505913 - make plugin click-to-play and crash handling fission-compatible, r=mconley At a high level, this change does the following: - move the pluginchild actor to be a JSWindowActorChild - move the parent handling from browser-plugins into a JSWindowActorParent - move the crash handling from ContentCrashHandlers.jsm to the parent actor, using a `PluginManager` object. It needs to talk to the actors (and vice versa), so this seemed a better fit than spreading actor implementation details to other JSMs. - switch to using plugin IDs to identify plugins cross-process, instead of combinations of names or other properties of the plugin tag. As part of that, ensured plugin IDs are unique between "fake" plugins and the other ones. - drop support for having a notification for more than 1 plugin. We only support Flash, in practice, so there didn't seem to be much point in the added complexity of trying to support more than 1 thing. Some notes: - the previous implementation mixes runIDs (for NPAPI plugin process "runs") and GMP pluginIDs when doing crashreporting. AFAICT there is no guarantee these don't conflict, so I've split them out to avoid issues. There's a pluginCrashID object I pass around instead that has either a runID or pluginID. Happy to rename some more for clarity. - the previous implementation used `pluginInfo` and `plugin` for a bunch of different types of variables. I've tried to be consistent, where: * `pluginElement` is a DOM element for a plugin * `activationInfo` is a JS object used to track click to play state for a plugin * `plugin` is a plugintag as returned by the pluginhost service * `pluginCrashID` is an identifier for a crashed plugin (see previous point). - I'm still using broadcastAsyncMessage to tell the content processes about gmp plugin crashes and plugin crash submission updates, because there's no guarantee the actors are instantiated (for gmp plugins) nor can the parent easily find out which actors to talk to (for either gmp or npapi plugins). Open to suggestions there, too. I think our best bet might be moving that to IPDL-based IPC within the GMP code, but that feels like a separate bug. Differential Revision: https://phabricator.services.mozilla.com/D37665 --HG-- rename : browser/base/content/browser-plugins.js => browser/actors/PluginParent.jsm extra : moz-landing-system : lando
2019-07-23 22:04:40 +00:00
let notification = window.PopupNotifications.getNotification(
kNotificationId,
browser
);
if (!plugin) {
this.removeNotification(browser);
return;
}
// We assume that we can only have 1 notification at a time anyway.
if (notification) {
if (showNow) {
notification.options.showNow = true;
notification.reshow();
}
return;
}
// Construct a notification for the plugin:
let { id, fallbackType } = plugin;
let pluginTag = PluginManager.getPluginTagById(id);
if (!pluginTag) {
return;
}
let permissionString = gPluginHost.getPermissionStringForTag(pluginTag);
let active = fallbackType == PLUGIN_ACTIVE;
let { top } = this.browsingContext;
if (!top.currentWindowGlobal) {
return;
}
let principal = top.currentWindowGlobal.documentPrincipal;
Bug 1505913 - make plugin click-to-play and crash handling fission-compatible, r=mconley At a high level, this change does the following: - move the pluginchild actor to be a JSWindowActorChild - move the parent handling from browser-plugins into a JSWindowActorParent - move the crash handling from ContentCrashHandlers.jsm to the parent actor, using a `PluginManager` object. It needs to talk to the actors (and vice versa), so this seemed a better fit than spreading actor implementation details to other JSMs. - switch to using plugin IDs to identify plugins cross-process, instead of combinations of names or other properties of the plugin tag. As part of that, ensured plugin IDs are unique between "fake" plugins and the other ones. - drop support for having a notification for more than 1 plugin. We only support Flash, in practice, so there didn't seem to be much point in the added complexity of trying to support more than 1 thing. Some notes: - the previous implementation mixes runIDs (for NPAPI plugin process "runs") and GMP pluginIDs when doing crashreporting. AFAICT there is no guarantee these don't conflict, so I've split them out to avoid issues. There's a pluginCrashID object I pass around instead that has either a runID or pluginID. Happy to rename some more for clarity. - the previous implementation used `pluginInfo` and `plugin` for a bunch of different types of variables. I've tried to be consistent, where: * `pluginElement` is a DOM element for a plugin * `activationInfo` is a JS object used to track click to play state for a plugin * `plugin` is a plugintag as returned by the pluginhost service * `pluginCrashID` is an identifier for a crashed plugin (see previous point). - I'm still using broadcastAsyncMessage to tell the content processes about gmp plugin crashes and plugin crash submission updates, because there's no guarantee the actors are instantiated (for gmp plugins) nor can the parent easily find out which actors to talk to (for either gmp or npapi plugins). Open to suggestions there, too. I think our best bet might be moving that to IPDL-based IPC within the GMP code, but that feels like a separate bug. Differential Revision: https://phabricator.services.mozilla.com/D37665 --HG-- rename : browser/base/content/browser-plugins.js => browser/actors/PluginParent.jsm extra : moz-landing-system : lando
2019-07-23 22:04:40 +00:00
let options = {
dismissed: !showNow,
hideClose: true,
persistent: showNow,
eventCallback: this._clickToPlayNotificationEventCallback,
showNow,
popupIconClass: "plugin-icon",
extraAttr: active ? "active" : "inactive",
principal,
Bug 1505913 - make plugin click-to-play and crash handling fission-compatible, r=mconley At a high level, this change does the following: - move the pluginchild actor to be a JSWindowActorChild - move the parent handling from browser-plugins into a JSWindowActorParent - move the crash handling from ContentCrashHandlers.jsm to the parent actor, using a `PluginManager` object. It needs to talk to the actors (and vice versa), so this seemed a better fit than spreading actor implementation details to other JSMs. - switch to using plugin IDs to identify plugins cross-process, instead of combinations of names or other properties of the plugin tag. As part of that, ensured plugin IDs are unique between "fake" plugins and the other ones. - drop support for having a notification for more than 1 plugin. We only support Flash, in practice, so there didn't seem to be much point in the added complexity of trying to support more than 1 thing. Some notes: - the previous implementation mixes runIDs (for NPAPI plugin process "runs") and GMP pluginIDs when doing crashreporting. AFAICT there is no guarantee these don't conflict, so I've split them out to avoid issues. There's a pluginCrashID object I pass around instead that has either a runID or pluginID. Happy to rename some more for clarity. - the previous implementation used `pluginInfo` and `plugin` for a bunch of different types of variables. I've tried to be consistent, where: * `pluginElement` is a DOM element for a plugin * `activationInfo` is a JS object used to track click to play state for a plugin * `plugin` is a plugintag as returned by the pluginhost service * `pluginCrashID` is an identifier for a crashed plugin (see previous point). - I'm still using broadcastAsyncMessage to tell the content processes about gmp plugin crashes and plugin crash submission updates, because there's no guarantee the actors are instantiated (for gmp plugins) nor can the parent easily find out which actors to talk to (for either gmp or npapi plugins). Open to suggestions there, too. I think our best bet might be moving that to IPDL-based IPC within the GMP code, but that feels like a separate bug. Differential Revision: https://phabricator.services.mozilla.com/D37665 --HG-- rename : browser/base/content/browser-plugins.js => browser/actors/PluginParent.jsm extra : moz-landing-system : lando
2019-07-23 22:04:40 +00:00
};
let description;
if (
fallbackType == Ci.nsIObjectLoadingContent.PLUGIN_VULNERABLE_UPDATABLE
) {
description = gNavigatorBundle.GetStringFromName(
"flashActivate.outdated.message"
);
} else {
description = gNavigatorBundle.GetStringFromName("flashActivate.message");
}
let badge = window.document.getElementById("plugin-icon-badge");
badge.setAttribute("animate", "true");
badge.addEventListener("animationend", function animListener(event) {
if (
event.animationName == "blink-badge" &&
badge.hasAttribute("animate")
) {
badge.removeAttribute("animate");
badge.removeEventListener("animationend", animListener);
}
});
let weakBrowser = Cu.getWeakReference(browser);
let activationInfo = { id, fallbackType, permissionString };
// Note: in both of these action callbacks, we check the fallbackType on the
// activationInfo object, not the local variable. This is important because
// the activationInfo object is effectively read/write - the notification
// will stay up, and for blocking after allowing (or vice versa) to work, we
// need to always read the updated value.
let mainAction = {
callback: () => {
let browserRef = weakBrowser.get();
if (!browserRef) {
return;
}
let perm =
activationInfo.fallbackType == PLUGIN_ACTIVE
? "continue"
: "allownow";
this._updatePluginPermission(browserRef, activationInfo, perm);
},
label: gNavigatorBundle.GetStringFromName("flashActivate.allow"),
accessKey: gNavigatorBundle.GetStringFromName(
"flashActivate.allow.accesskey"
),
dismiss: true,
};
let secondaryActions = [
{
callback: () => {
let browserRef = weakBrowser.get();
if (!browserRef) {
return;
}
let perm =
activationInfo.fallbackType == PLUGIN_ACTIVE
? "block"
: "continueblocking";
this._updatePluginPermission(browserRef, activationInfo, perm);
},
label: gNavigatorBundle.GetStringFromName("flashActivate.noAllow"),
accessKey: gNavigatorBundle.GetStringFromName(
"flashActivate.noAllow.accesskey"
),
dismiss: true,
},
];
window.PopupNotifications.show(
browser,
kNotificationId,
description,
"plugins-notification-icon",
mainAction,
secondaryActions,
options
);
// Check if the plugin is insecure and update the notification icon accordingly.
let haveInsecure = false;
switch (fallbackType) {
// haveInsecure will trigger the red flashing icon and the infobar
// styling below
case PLUGIN_VULNERABLE_UPDATABLE:
case PLUGIN_VULNERABLE_NO_UPDATE:
haveInsecure = true;
}
window.document
.getElementById("plugins-notification-icon")
.classList.toggle("plugin-blocked", haveInsecure);
}
removeNotification(browser) {
let { PopupNotifications } = browser.ownerGlobal;
let notification = PopupNotifications.getNotification(
kNotificationId,
browser
);
if (notification) {
PopupNotifications.remove(notification);
}
}
/**
* Shows a plugin-crashed notification bar for a browser that has had an
* invisible NPAPI plugin crash, or a GMP plugin crash.
*
* @param browser
* The browser to show the notification for.
* @param pluginCrashID
* The unique-per-process identifier for the NPAPI plugin or GMP.
* This will have either a runID or pluginID property, identifying
* an npapi plugin or gmp plugin crash, respectively.
*/
showPluginCrashedNotification(browser, pluginCrashID) {
// If there's already an existing notification bar, don't do anything.
let notificationBox = browser.getTabBrowser().getNotificationBox(browser);
let notification = notificationBox.getNotificationWithValue(
"plugin-crashed"
);
let report = PluginManager.getCrashReport(pluginCrashID);
if (notification || !report) {
return;
}
// Configure the notification bar
let priority = notificationBox.PRIORITY_WARNING_MEDIUM;
let iconURL = "chrome://global/skin/plugins/pluginGeneric.svg";
let reloadLabel = gNavigatorBundle.GetStringFromName(
"crashedpluginsMessage.reloadButton.label"
);
let reloadKey = gNavigatorBundle.GetStringFromName(
"crashedpluginsMessage.reloadButton.accesskey"
);
let buttons = [
{
label: reloadLabel,
accessKey: reloadKey,
popup: null,
callback() {
browser.reload();
},
},
];
if (AppConstants.MOZ_CRASHREPORTER) {
let submitLabel = gNavigatorBundle.GetStringFromName(
"crashedpluginsMessage.submitButton.label"
);
let submitKey = gNavigatorBundle.GetStringFromName(
"crashedpluginsMessage.submitButton.accesskey"
);
let submitButton = {
label: submitLabel,
accessKey: submitKey,
popup: null,
callback: () => {
PluginManager.submitCrashReport(pluginCrashID);
},
};
buttons.push(submitButton);
}
let messageString = gNavigatorBundle.formatStringFromName(
"crashedpluginsMessage.title",
[report.pluginName]
);
notification = notificationBox.appendNotification(
messageString,
"plugin-crashed",
iconURL,
priority,
buttons
);
// Add the "learn more" link.
let link = notification.ownerDocument.createXULElement("label", {
is: "text-link",
});
link.setAttribute(
"value",
gNavigatorBundle.GetStringFromName("crashedpluginsMessage.learnMore")
);
let crashurl = Services.urlFormatter.formatURLPref("app.support.baseURL");
crashurl += "plugin-crashed-notificationbar";
link.href = crashurl;
notification.messageText.appendChild(link);
}
}