Files
sousa-gecko/toolkit/components/terminator/nsTerminator.cpp
T

476 lines
16 KiB
C++

/* 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/. */
/**
* A watchdog designed to terminate shutdown if it lasts too long.
*
* This watchdog is designed as a worst-case problem container for the
* common case in which Firefox just won't shutdown.
*
* We spawn a thread during quit-application. If any of the shutdown
* steps takes more than n milliseconds (63000 by default), kill the
* process as fast as possible, without any cleanup.
*/
#include "mozilla/ShutdownPhase.h"
#include "nsTerminator.h"
#include "prthread.h"
#include "prmon.h"
#include "prio.h"
#include "prenv.h"
#include "nsString.h"
#include "nsDirectoryServiceUtils.h"
#include "nsAppDirectoryServiceDefs.h"
#include "nsExceptionHandler.h"
#include "GeckoProfiler.h"
#include "nsThreadUtils.h"
#include "nsXULAppAPI.h"
#if defined(XP_WIN)
# include <windows.h>
#else
# include <unistd.h>
#endif
#include "mozilla/AppShutdown.h"
#include "mozilla/Atomics.h"
#include "mozilla/DebugOnly.h"
#include "mozilla/IntentionalCrash.h"
#include "mozilla/MemoryChecking.h"
#include "mozilla/Preferences.h"
#include "mozilla/SpinEventLoopUntil.h"
#include "mozilla/UniquePtr.h"
#include "mozilla/dom/workerinternals/RuntimeService.h"
// Normally, the number of milliseconds that AsyncShutdown waits until
// it decides to crash is specified as a preference. We use the
// following value as a fallback if for some reason the preference is
// absent.
#define FALLBACK_ASYNCSHUTDOWN_CRASH_AFTER_MS 60000
// Additional number of milliseconds to wait until we decide to exit
// forcefully.
#define ADDITIONAL_WAIT_BEFORE_CRASH_MS 3000
#define HEARTBEAT_INTERVAL_MS 100
namespace mozilla {
namespace {
/**
* A step during shutdown.
*
* Shutdown is divided in steps, which all map to an observer
* notification. The duration of a step is defined as the number of
* ticks between the time we receive a notification and the next one.
*/
struct ShutdownStep {
mozilla::ShutdownPhase mPhase;
Atomic<int> mTicks;
constexpr explicit ShutdownStep(mozilla::ShutdownPhase aPhase)
: mPhase(aPhase), mTicks(-1) {}
};
static ShutdownStep sShutdownSteps[] = {
ShutdownStep(mozilla::ShutdownPhase::AppShutdownConfirmed),
ShutdownStep(mozilla::ShutdownPhase::AppShutdownNetTeardown),
ShutdownStep(mozilla::ShutdownPhase::AppShutdownTeardown),
ShutdownStep(mozilla::ShutdownPhase::AppShutdown),
ShutdownStep(mozilla::ShutdownPhase::AppShutdownQM),
ShutdownStep(mozilla::ShutdownPhase::XPCOMWillShutdown),
ShutdownStep(mozilla::ShutdownPhase::XPCOMShutdown),
ShutdownStep(mozilla::ShutdownPhase::XPCOMShutdownThreads),
ShutdownStep(mozilla::ShutdownPhase::XPCOMShutdownFinal),
ShutdownStep(mozilla::ShutdownPhase::CCPostLastCycleCollection),
};
int GetStepForPhase(mozilla::ShutdownPhase aPhase) {
for (size_t i = 0; i < std::size(sShutdownSteps); i++) {
if (sShutdownSteps[i].mPhase >= aPhase) {
return (int)i;
}
}
return -1;
}
// Utility function: create a thread that is non-joinable,
// does not prevent the process from terminating, is never
// cooperatively scheduled, and uses a default stack size.
PRThread* CreateSystemThread(void (*start)(void* arg), void* arg) {
PRThread* thread =
PR_CreateThread(PR_SYSTEM_THREAD, /* This thread will not prevent the
process from terminating */
start, arg, PR_PRIORITY_LOW,
PR_GLOBAL_THREAD, /* Make sure that the thread is never
cooperatively scheduled */
PR_UNJOINABLE_THREAD, 0 /* Use default stack size */
);
MOZ_LSAN_INTENTIONALLY_LEAK_OBJECT(
thread); // This pointer will never be deallocated.
return thread;
}
////////////////////////////////////////////
//
// The watchdog
//
// This nspr thread is in charge of crashing the process if any stage of
// shutdown lasts more than some predefined duration. As a side-effect, it
// measures the duration of each stage of shutdown.
//
// The heartbeat of the operation.
//
// Main thread:
//
// * Whenever a shutdown step has been completed, the main thread
// swaps gHeartbeat to 0 to mark that the shutdown process is still
// progressing. The value swapped away indicates the number of ticks
// it took for the shutdown step to advance.
//
// Watchdog thread:
//
// * Every tick, the watchdog thread increments gHearbeat atomically.
//
// A note about precision:
// Since gHeartbeat is generally reset to 0 between two ticks, this means
// that gHeartbeat stays at 0 less than one tick. Consequently, values
// extracted from gHeartbeat must be considered rounded up.
Atomic<uint32_t> gHeartbeat(0);
struct Options {
/**
* How many ticks before we should crash the process.
*/
uint32_t crashAfterTicks;
};
/**
* Save a profile of this process before we crash on a shutdown hang, so a
* profiled run keeps the data leading up to the hang. We gather it here on the
* watchdog thread rather than dispatching to the (blocked) main thread:
* profiler_save_profile_to_file only needs the profiler state lock, and the
* sampled main-thread stack it captures shows what is blocking shutdown.
*/
void MaybeSaveShutdownHangProfile() {
if (!profiler_is_active()) {
return;
}
// An explicit --profiler run names a shutdown profile file via
// MOZ_PROFILER_SHUTDOWN and expects the profile there; honor it first, as the
// test harnesses do, since the normal shutdown-time save never runs once we
// crash here. It is emptied in child processes, so only the parent uses it.
if (const char* shutdownFile = PR_GetEnv("MOZ_PROFILER_SHUTDOWN");
shutdownFile && *shutdownFile) {
profiler_save_profile_to_file(shutdownFile);
printf_stderr("RunWatchdog: saved a profile of the shutdown hang to %s\n",
shutdownFile);
return;
}
// Otherwise, in CI, anything written to MOZ_UPLOAD_DIR is uploaded as an
// artifact. Name it after the pid, since the parent and content processes
// each run their own terminator. Report it as a TEST-UNEXPECTED-FAIL line so
// it surfaces in Treeherder's failure summary (the main thread is blocked, so
// we can't route this through the test harness's structured log), naming the
// artifact in the "profile uploaded in <file>" form the dashboards recognize.
const char* uploadDir = PR_GetEnv("MOZ_UPLOAD_DIR");
if (!uploadDir || !*uploadDir) {
return;
}
#if defined(XP_WIN)
uint32_t pid = GetCurrentProcessId();
#else
uint32_t pid = getpid();
#endif
nsAutoCString filename;
filename.AppendPrintf("profile_shutdown_hang_%u.json", pid);
nsAutoCString path(uploadDir);
path.AppendLiteral("/");
path.Append(filename);
profiler_save_profile_to_file(path.get());
printf_stderr(
"TEST-UNEXPECTED-FAIL | shutdown hang | profile uploaded in %s\n",
filename.get());
}
/**
* Entry point for the watchdog thread
*/
void RunWatchdog(void* arg) {
NS_SetCurrentThreadName("Shutdown Hang Terminator");
// Let's copy and deallocate options, that's one less leak to worry
// about.
UniquePtr<Options> options((Options*)arg);
uint32_t crashAfterTicks = options->crashAfterTicks;
options = nullptr;
const uint32_t timeToLive = crashAfterTicks;
while (true) {
//
// We do not want to sleep for the entire duration,
// as putting the computer to sleep would suddenly
// cause us to timeout on wakeup.
//
// Rather, we prefer sleeping for at most 1 second
// at a time. If the computer sleeps then wakes up,
// we have lost at most one second, which is much
// more reasonable.
//
#if defined(XP_WIN)
Sleep(HEARTBEAT_INTERVAL_MS /* ms */);
#else
usleep(HEARTBEAT_INTERVAL_MS * 1000 /* usec */);
#endif
if (gHeartbeat++ < timeToLive) {
continue;
}
// Arrived here we know we will crash in a way or another.
// If the sampler thread is mid-write on a scheduled profile dump, let it
// finish (and, since it exits the process on completion, we never reach the
// crash below) rather than crashing over a half-written profile.
profiler_wait_for_scheduled_dump();
NoteIntentionalCrash(XRE_GetProcessTypeString());
MaybeSaveShutdownHangProfile();
// Until we have general log output for crash annotations in treeherder
// (bug 1728721) we manually spit out our nested event loop stack.
// XXX: Remove once bug 1728721 is fixed.
nsCString stack;
AutoNestedEventLoopAnnotation::CopyCurrentStack(stack);
printf_stderr(
"RunWatchdog: Mainthread nested event loops during hang: \n --- %s\n",
stack.get());
// Let's find the last known shutdown phase.
mozilla::ShutdownPhase lastPhase = mozilla::ShutdownPhase::NotInShutdown;
// Looping inverse here to make the search more robust in case
// the observer that triggers UpdateHeartbeat was not called
// at all or in the expected order on some step. This should
// give us always the last known ShutdownStep.
for (int i = std::size(sShutdownSteps) - 1; i >= 0; --i) {
if (sShutdownSteps[i].mTicks > -1) {
lastPhase = sShutdownSteps[i].mPhase;
break;
}
}
if (lastPhase == mozilla::ShutdownPhase::NotInShutdown) {
// This is not something we expect to ever happen, but still.
CrashReporter::SetMinidumpAnalysisAllThreads();
MOZ_CRASH("Shutdown hanging before starting any known phase.");
}
// First check if worker shutdown started and is incomplete, in case
// report running workers.
mozilla::dom::workerinternals::RuntimeService* runtimeService =
mozilla::dom::workerinternals::RuntimeService::GetService();
if (runtimeService) {
// CrashIfHanging will check if we actually ever asked for worker
// shutdown, so calling it before is a no-op.
runtimeService->CrashIfHanging();
}
// Otherwise just report our shutdown phase.
// This string will be leaked.
nsCString msg;
msg.AppendPrintf(
"Shutdown hanging at step %s. "
"Something is blocking the main-thread.",
mozilla::AppShutdown::GetShutdownPhaseName(lastPhase));
CrashReporter::SetMinidumpAnalysisAllThreads();
MOZ_CRASH_UNSAFE(strdup(msg.get()));
}
}
} // namespace
NS_IMPL_ISUPPORTS(nsTerminator, nsIObserver, nsITerminatorTest)
nsTerminator::nsTerminator() : mInitialized(false), mCurrentStep(-1) {}
// Actually launch these threads. This takes place at the first sign of
// shutdown.
void nsTerminator::Start() {
MOZ_ASSERT(!mInitialized);
StartWatchdog();
mInitialized = true;
}
NS_IMETHODIMP
nsTerminator::Observe(nsISupports*, const char* aTopic, const char16_t*) {
// This Observe is now only used for testing purposes.
// XXX: Check if we should change our testing strategy.
if (strcmp(aTopic, "terminator-test-quit-application") == 0) {
AdvancePhase(mozilla::ShutdownPhase::AppShutdownConfirmed);
} else if (strcmp(aTopic, "terminator-test-profile-change-net-teardown") ==
0) {
AdvancePhase(mozilla::ShutdownPhase::AppShutdownNetTeardown);
} else if (strcmp(aTopic, "terminator-test-profile-change-teardown") == 0) {
AdvancePhase(mozilla::ShutdownPhase::AppShutdownTeardown);
} else if (strcmp(aTopic, "terminator-test-profile-before-change") == 0) {
AdvancePhase(mozilla::ShutdownPhase::AppShutdown);
} else if (strcmp(aTopic, "terminator-test-profile-before-change-qm") == 0) {
AdvancePhase(mozilla::ShutdownPhase::AppShutdownQM);
} else if (strcmp(aTopic,
"terminator-test-profile-before-change-telemetry") == 0) {
AdvancePhase(mozilla::ShutdownPhase::AppShutdownTelemetry);
} else if (strcmp(aTopic, "terminator-test-xpcom-will-shutdown") == 0) {
AdvancePhase(mozilla::ShutdownPhase::XPCOMWillShutdown);
} else if (strcmp(aTopic, "terminator-test-xpcom-shutdown") == 0) {
AdvancePhase(mozilla::ShutdownPhase::XPCOMShutdown);
} else if (strcmp(aTopic, "terminator-test-xpcom-shutdown-threads") == 0) {
AdvancePhase(mozilla::ShutdownPhase::XPCOMShutdownThreads);
} else if (strcmp(aTopic, "terminator-test-XPCOMShutdownFinal") == 0) {
AdvancePhase(mozilla::ShutdownPhase::XPCOMShutdownFinal);
} else if (strcmp(aTopic, "terminator-test-CCPostLastCycleCollection") == 0) {
AdvancePhase(mozilla::ShutdownPhase::CCPostLastCycleCollection);
}
return NS_OK;
}
// Prepare, allocate and start the watchdog thread.
// By design, it will never finish, nor be deallocated.
void nsTerminator::StartWatchdog() {
int32_t crashAfterMS =
Preferences::GetInt("toolkit.asyncshutdown.crash_timeout",
FALLBACK_ASYNCSHUTDOWN_CRASH_AFTER_MS);
int32_t additionalWaitBeforeCrashMs =
Preferences::GetInt("toolkit.asyncshutdown.crash_timeout_additional_wait",
ADDITIONAL_WAIT_BEFORE_CRASH_MS);
// Ignore negative values
if (crashAfterMS <= 0) {
crashAfterMS = FALLBACK_ASYNCSHUTDOWN_CRASH_AFTER_MS;
}
// Keep the same guarantee as before, so that crashAfterTicks > 0
if (additionalWaitBeforeCrashMs <= 0) {
additionalWaitBeforeCrashMs = ADDITIONAL_WAIT_BEFORE_CRASH_MS;
}
// Add a little padding, to ensure that we do not crash before
// AsyncShutdown.
if (crashAfterMS > INT32_MAX - additionalWaitBeforeCrashMs) {
// Defend against overflow
crashAfterMS = INT32_MAX;
} else {
crashAfterMS += additionalWaitBeforeCrashMs;
}
#ifdef MOZ_VALGRIND
// If we're running on Valgrind, we'll be making forward progress at a
// rate of somewhere between 1/25th and 1/50th of normal. This can cause
// timeouts frequently enough to be a problem for the Valgrind runs on
// automation: see bug 1296819. As an attempt to avoid the worst of this,
// scale up the presented timeout by a factor of three. For a
// non-Valgrind-enabled build, or for an enabled build which isn't running
// on Valgrind, the timeout is unchanged.
if (RUNNING_ON_VALGRIND) {
const int32_t scaleUp = 3;
if (crashAfterMS >= (INT32_MAX / scaleUp) - 1) {
// Defend against overflow
crashAfterMS = INT32_MAX;
} else {
crashAfterMS *= scaleUp;
}
}
#endif
UniquePtr<Options> options(new Options());
// Guarantee that crashAfterTicks is non-zero
options->crashAfterTicks = std::max(1, crashAfterMS / HEARTBEAT_INTERVAL_MS);
DebugOnly<PRThread*> watchdogThread =
CreateSystemThread(RunWatchdog, options.release());
MOZ_ASSERT(watchdogThread);
}
// This helper is here to preserve the existing crash reporting behavior
// based on observer topic names, using the shutdown phase name only for
// phases without associated topic.
const char* GetReadableNameForPhase(mozilla::ShutdownPhase aPhase) {
const char* readableName = mozilla::AppShutdown::GetObserverKey(aPhase);
if (!readableName) {
readableName = mozilla::AppShutdown::GetShutdownPhaseName(aPhase);
}
return readableName;
}
void nsTerminator::AdvancePhase(mozilla::ShutdownPhase aPhase) {
// If the phase is unknown, just ignore it.
auto step = GetStepForPhase(aPhase);
if (step < 0) {
return;
}
// As we have seen examples in the wild of shutdown notifications
// not being sent (or not being sent in the expected order), we do
// not assume a specific order.
if (!mInitialized) {
Start();
}
UpdateHeartbeat(step);
UpdateCrashReport(GetReadableNameForPhase(aPhase));
}
void nsTerminator::UpdateHeartbeat(int32_t aStep) {
MOZ_ASSERT(aStep >= mCurrentStep);
if (aStep > mCurrentStep) {
// Reset the clock, find out how long the current phase has lasted.
uint32_t ticks = gHeartbeat.exchange(0);
if (mCurrentStep >= 0) {
sShutdownSteps[mCurrentStep].mTicks = ticks;
}
sShutdownSteps[aStep].mTicks = 0;
mCurrentStep = aStep;
}
}
void nsTerminator::UpdateCrashReport(const char* aTopic) {
// In case of crash, we wish to know where in shutdown we are
CrashReporter::RecordAnnotationCString(
CrashReporter::Annotation::ShutdownProgress, aTopic);
}
NS_IMETHODIMP
nsTerminator::GetTicksForShutdownPhases(JSContext* aCx,
JS::MutableHandle<JS::Value> aRetval) {
JS::Rooted<JSObject*> obj(aCx, JS_NewPlainObject(aCx));
aRetval.setObject(*obj);
for (auto& shutdownStep : sShutdownSteps) {
if (shutdownStep.mTicks >= 0) {
JS_DefineProperty(aCx, obj, GetReadableNameForPhase(shutdownStep.mPhase),
shutdownStep.mTicks, JSPROP_ENUMERATE);
}
}
return NS_OK;
} // namespace mozilla
} // namespace mozilla