summaryrefslogtreecommitdiff
path: root/toolkit/components/terminator/nsTerminator.cpp
blob: 007885b8d4090bd6d5f21315cf4515b70ec2cbed (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*-
 * vim: sw=2 ts=2 et lcs=trail\:.,tab\:>~ :
 * 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 "nsTerminator.h"

#include "prthread.h"
#include "prmon.h"
#include "plstr.h"
#include "prio.h"

#include "nsString.h"
#include "nsServiceManagerUtils.h"
#include "nsDirectoryServiceUtils.h"
#include "nsAppDirectoryServiceDefs.h"

#include "nsIObserverService.h"
#include "nsIPrefService.h"

#if defined(XP_WIN)
#include <windows.h>
#else
#include <unistd.h>
#endif

#include "mozilla/ArrayUtils.h"
#include "mozilla/Attributes.h"
#include "mozilla/DebugOnly.h"
#include "mozilla/MemoryChecking.h"
#include "mozilla/Preferences.h"
#include "mozilla/Services.h"
#include "mozilla/UniquePtr.h"
#include "mozilla/Unused.h"
#include "mozilla/Telemetry.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

namespace mozilla {

namespace {

// 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;
};

/**
 * Entry point for the watchdog thread
 */
void
RunWatchdog(void* arg)
{
  PR_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(1000 /* ms */);
#else
    usleep(1000000 /* usec */);
#endif

    if (gHeartbeat++ < timeToLive) {
      continue;
    }

    // Shutdown is apparently dead. Crash the process.
    MOZ_CRASH("Shutdown too long, probably frozen, causing a crash.");
  }
}

////////////////////////////////////////////
//
// Writer thread
//
// This nspr thread is in charge of writing to disk statistics produced by the
// watchdog thread and collected by the main thread. Note that we use a nspr
// thread rather than usual XPCOM I/O simply because we outlive XPCOM and its
// threads.
//

// Utility class, used by UniquePtr<> to close nspr files.
class PR_CloseDelete
{
public:
  constexpr PR_CloseDelete() {}

  PR_CloseDelete(const PR_CloseDelete& aOther)
  {}

  void operator()(PRFileDesc* aPtr) const
  {
    PR_Close(aPtr);
  }
};

//
// Communication between the main thread and the writer thread.
//
// Main thread:
//
// * Whenever a shutdown step has been completed, the main thread
// obtains the number of ticks from the watchdog threads, builds
// a string representing all the data gathered so far, places
// this string in `gWriteData`, and wakes up the writer thread
// using `gWriteReady`. If `gWriteData` already contained a non-null
// pointer, this means that the writer thread is lagging behind the
// main thread, and the main thread cleans up the memory.
//
// Writer thread:
//
// * When awake, the writer thread swaps `gWriteData` to nullptr. If
// `gWriteData` contained data to write, the . If so, the writer
// thread writes the data to a file named "ShutdownDuration.json.tmp",
// then moves that file to "ShutdownDuration.json" and cleans up the
// data. If `gWriteData` contains a nullptr, the writer goes to sleep
// until it is awkened using `gWriteReady`.
//
//
// The data written by the writer thread will be read by another
// module upon the next restart and fed to Telemetry.
//
Atomic<nsCString*> gWriteData(nullptr);
PRMonitor* gWriteReady = nullptr;

void RunWriter(void* arg)
{
  PR_SetCurrentThreadName("Shutdown Statistics Writer");

  MOZ_LSAN_INTENTIONALLY_LEAK_OBJECT(arg);
  // Shutdown will generally complete before we have a chance to
  // deallocate. This is not a leak.

  // Setup destinationPath and tmpFilePath

  nsCString destinationPath(static_cast<char*>(arg));
  nsAutoCString tmpFilePath;
  tmpFilePath.Append(destinationPath);
  tmpFilePath.AppendLiteral(".tmp");

  // Cleanup any file leftover from a previous run
  Unused << PR_Delete(tmpFilePath.get());
  Unused << PR_Delete(destinationPath.get());

  while (true) {
    //
    // Check whether we have received data from the main thread.
    //
    // We perform the check before waiting on `gWriteReady` as we may
    // have received data while we were busy writing.
    //
    // Also note that gWriteData may have been modified several times
    // since we last checked. That's ok, we are not losing any important
    // data (since we keep adding data), and we are not leaking memory
    // (since the main thread deallocates any data that hasn't been
    // consumed by the writer thread).
    //
    UniquePtr<nsCString> data(gWriteData.exchange(nullptr));
    if (!data) {
      // Data is not available yet.
      // Wait until the main thread provides it.
      PR_EnterMonitor(gWriteReady);
      PR_Wait(gWriteReady, PR_INTERVAL_NO_TIMEOUT);
      PR_ExitMonitor(gWriteReady);
      continue;
    }

    MOZ_LSAN_INTENTIONALLY_LEAK_OBJECT(data.get());
    // Shutdown may complete before we have a chance to deallocate.
    // This is not a leak.

    //
    // Write to a temporary file
    //
    // In case of any error, we simply give up. Since the data is
    // hardly critical, we don't want to spend too much effort
    // salvaging it.
    //
    UniquePtr<PRFileDesc, PR_CloseDelete>
      tmpFileDesc(PR_Open(tmpFilePath.get(),
                          PR_WRONLY | PR_TRUNCATE | PR_CREATE_FILE,
                          00600));

    // Shutdown may complete before we have a chance to close the file.
    // This is not a leak.
    MOZ_LSAN_INTENTIONALLY_LEAK_OBJECT(tmpFileDesc.get());

    if (tmpFileDesc == nullptr) {
      break;
    }
    if (PR_Write(tmpFileDesc.get(), data->get(), data->Length()) == -1) {
      break;
    }
    tmpFileDesc.reset();

    //
    // Rename on top of destination file.
    //
    // This is not sufficient to guarantee that the destination file
    // will be written correctly, but, again, we don't care enough
    // about the data to make more efforts.
    //
    if (PR_Rename(tmpFilePath.get(), destinationPath.get()) != PR_SUCCESS) {
      break;
    }
  }
}

/**
 * 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
{
  char const* const mTopic;
  int mTicks;

  constexpr explicit ShutdownStep(const char *const topic)
    : mTopic(topic)
    , mTicks(-1)
  {}

};

static ShutdownStep sShutdownSteps[] = {
  ShutdownStep("quit-application"),
  ShutdownStep("profile-change-teardown"),
  ShutdownStep("profile-before-change"),
  ShutdownStep("xpcom-will-shutdown"),
  ShutdownStep("xpcom-shutdown"),
};

} // namespace

NS_IMPL_ISUPPORTS(nsTerminator, nsIObserver)

nsTerminator::nsTerminator()
  : mInitialized(false)
  , mCurrentStep(-1)
{
}

// During startup, register as an observer for all interesting topics.
nsresult
nsTerminator::SelfInit()
{
  nsCOMPtr<nsIObserverService> os = mozilla::services::GetObserverService();
  if (!os) {
    return NS_ERROR_UNEXPECTED;
  }

  for (size_t i = 0; i < ArrayLength(sShutdownSteps); ++i) {
    DebugOnly<nsresult> rv = os->AddObserver(this, sShutdownSteps[i].mTopic, false);
    NS_WARNING_ASSERTION(NS_SUCCEEDED(rv), "AddObserver failed");
  }

  return NS_OK;
}

// Actually launch these threads. This takes place at the first sign of shutdown.
void
nsTerminator::Start()
{
  MOZ_ASSERT(!mInitialized);
  StartWatchdog();
#if !defined(DEBUG)
  // Only allow nsTerminator to write on non-debug builds so we don't get leak warnings on
  // shutdown for intentional leaks (see bug 1242084). This will be enabled again by bug
  // 1255484 when 1255478 lands.
  StartWriter();
#endif // !defined(DEBUG)
  mInitialized = true;
}

// 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);
  // Ignore negative values
  if (crashAfterMS <= 0) {
    crashAfterMS = FALLBACK_ASYNCSHUTDOWN_CRASH_AFTER_MS;
  }

  // Add a little padding, to ensure that we do not crash before
  // AsyncShutdown.
  if (crashAfterMS > INT32_MAX - ADDITIONAL_WAIT_BEFORE_CRASH_MS) {
    // Defend against overflow
    crashAfterMS = INT32_MAX;
  } else {
    crashAfterMS += ADDITIONAL_WAIT_BEFORE_CRASH_MS;
  }

  UniquePtr<Options> options(new Options());
#if !defined(XP_SOLARIS)
  const PRIntervalTime ticksDuration = PR_MillisecondsToInterval(1000);
#else
  const PRIntervalTime ticksDuration = 1000;
#endif  
  options->crashAfterTicks = crashAfterMS / ticksDuration;

  DebugOnly<PRThread*> watchdogThread = CreateSystemThread(RunWatchdog,
                                                options.release());
  MOZ_ASSERT(watchdogThread);
}

// Prepare, allocate and start the writer thread. By design, it will never
// finish, nor be deallocated. In case of error, we degrade
// gracefully to not writing Telemetry data.
void
nsTerminator::StartWriter()
{
  if (!Telemetry::CanRecordExtended()) {
    return;
  }
  nsCOMPtr<nsIFile> profLD;
  nsresult rv = NS_GetSpecialDirectory(NS_APP_USER_PROFILE_LOCAL_50_DIR,
                                       getter_AddRefs(profLD));
  if (NS_FAILED(rv)) {
    return;
  }

  rv = profLD->Append(NS_LITERAL_STRING("ShutdownDuration.json"));
  if (NS_FAILED(rv)) {
    return;
  }

  nsAutoString path;
  rv = profLD->GetPath(path);
  if (NS_FAILED(rv)) {
    return;
  }

  gWriteReady = PR_NewMonitor();
  MOZ_LSAN_INTENTIONALLY_LEAK_OBJECT(gWriteReady); // We will never deallocate this object
  PRThread* writerThread = CreateSystemThread(RunWriter,
                                              ToNewUTF8String(path));

  if (!writerThread) {
    return;
  }
}

NS_IMETHODIMP
nsTerminator::Observe(nsISupports *, const char *aTopic, const char16_t *)
{
  if (strcmp(aTopic, "profile-after-change") == 0) {
    return SelfInit();
  }

  // Other notifications are shutdown-related.

  // 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(aTopic);
#if !defined(DEBUG)
  // Only allow nsTerminator to write on non-debug builds so we don't get leak warnings on
  // shutdown for intentional leaks (see bug 1242084). This will be enabled again by bug
  // 1255484 when 1255478 lands.
  UpdateTelemetry();
#endif // !defined(DEBUG)
  UpdateCrashReport(aTopic);

  // Perform a little cleanup
  nsCOMPtr<nsIObserverService> os = mozilla::services::GetObserverService();
  MOZ_RELEASE_ASSERT(os);
  (void)os->RemoveObserver(this, aTopic);

  return NS_OK;
}

void
nsTerminator::UpdateHeartbeat(const char* aTopic)
{
  // 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;
  }

  // Find out where we now are in the current shutdown.
  // Don't assume that shutdown takes place in the expected order.
  int nextStep = -1;
  for (size_t i = 0; i < ArrayLength(sShutdownSteps); ++i) {
    if (strcmp(sShutdownSteps[i].mTopic, aTopic) == 0) {
      nextStep = i;
      break;
    }
  }
  MOZ_ASSERT(nextStep != -1);
  mCurrentStep = nextStep;
}

void
nsTerminator::UpdateTelemetry()
{
  if (!Telemetry::CanRecordExtended() || !gWriteReady) {
    return;
  }

  //
  // We need Telemetry data on the effective duration of each step,
  // to be able to tune the time-to-crash of each of both the
  // Terminator and AsyncShutdown. However, at this stage, it is too
  // late to record such data into Telemetry, so we write it to disk
  // and read it upon the next startup.
  //

  // Build JSON.
  UniquePtr<nsCString> telemetryData(new nsCString());
  telemetryData->AppendLiteral("{");
  size_t fields = 0;
  for (size_t i = 0; i < ArrayLength(sShutdownSteps); ++i) {
    if (sShutdownSteps[i].mTicks < 0) {
      // Ignore this field.
      continue;
    }
    if (fields++ > 0) {
      telemetryData->Append(", ");
    }
    telemetryData->AppendLiteral("\"");
    telemetryData->Append(sShutdownSteps[i].mTopic);
    telemetryData->AppendLiteral("\": ");
    telemetryData->AppendInt(sShutdownSteps[i].mTicks);
  }
  telemetryData->AppendLiteral("}");

  if (fields == 0) {
    // Nothing to write
      return;
  }

  //
  // Send data to the worker thread.
  //
  delete gWriteData.exchange(telemetryData.release()); // Clear any data that hasn't been written yet

  // In case the worker thread was sleeping, wake it up.
  PR_EnterMonitor(gWriteReady);
  PR_Notify(gWriteReady);
  PR_ExitMonitor(gWriteReady);
}

void
nsTerminator::UpdateCrashReport(const char* aTopic)
{
  /*** STUB ***/
}


} // namespace mozilla