components/browser_watcher/watcher_metrics_provider_win.cc

   1 // Copyright (c) 2014 The Chromium Authors. All rights reserved.
   2 // Use of this source code is governed by a BSD-style license that can be
   3 // found in the LICENSE file.
   4
   5 #include "components/browser_watcher/watcher_metrics_provider_win.h"
   6
   7 #include <limits>
   8 #include <vector>
   9
  10 #include "base/metrics/sparse_histogram.h"
  11 #include "base/process/process.h"
  12 #include "base/strings/string_number_conversions.h"
  13 #include "base/strings/string_piece.h"
  14 #include "base/strings/utf_string_conversions.h"
  15 #include "base/win/registry.h"
  16
  17 namespace browser_watcher {
  18
  19 namespace {
  20
  21 // Process ID APIs on Windows talk in DWORDs, whereas for string formatting
  22 // and parsing, this code uses int. In practice there are no process IDs with
  23 // the high bit set on Windows, so there's no danger of overflow if this is
  24 // done consistently.
  25 static_assert(sizeof(DWORD) == sizeof(int),
  26               "process ids are expected to be no larger than int");
  27
  28 // This function does soft matching on the PID recorded in the key only.
  29 // Due to PID reuse, the possibility exists that the process that's now live
  30 // with the given PID is not the same process the data was recorded for.
  31 // This doesn't matter for the purpose, as eventually the data will be
  32 // scavenged and reported.
  33 bool IsDeadProcess(base::StringPiece16 key_or_value_name) {
  34   // Truncate the input string to the first occurrence of '-', if one exists.
  35   size_t num_end = key_or_value_name.find(L'-');
  36   if (num_end != base::StringPiece16::npos)
  37     key_or_value_name = key_or_value_name.substr(0, num_end);
  38
  39   // Convert to the numeric PID.
  40   int pid = 0;
  41   if (!base::StringToInt(key_or_value_name, &pid) || pid == 0)
  42     return true;
  43
  44   // This is a very inexpensive check for the common case of our own PID.
  45   if (static_cast<base::ProcessId>(pid) == base::GetCurrentProcId())
  46     return false;
  47
  48   // The process is not our own - see whether a process with this PID exists.
  49   // This is more expensive than the above check, but should also be very rare,
  50   // as this only happens more than once for a given PID if a user is running
  51   // multiple Chrome instances concurrently.
  52   base::Process process =
  53       base::Process::Open(static_cast<base::ProcessId>(pid));
  54   if (process.IsValid()) {
  55     // The fact that it was possible to open the process says it's live.
  56     return false;
  57   }
  58
  59   return true;
  60 }
  61
  62 void RecordExitCodes(const base::string16& registry_path) {
  63   base::win::RegKey regkey(HKEY_CURRENT_USER,
  64                            registry_path.c_str(),
  65                            KEY_QUERY_VALUE | KEY_SET_VALUE);
  66   if (!regkey.Valid())
  67     return;
  68
  69   size_t num = regkey.GetValueCount();
  70   if (num == 0)
  71     return;
  72   std::vector<base::string16> to_delete;
  73
  74   // Record the exit codes in a sparse stability histogram, as the range of
  75   // values used to report failures is large.
  76   base::HistogramBase* exit_code_histogram =
  77       base::SparseHistogram::FactoryGet(
  78           WatcherMetricsProviderWin::kBrowserExitCodeHistogramName,
  79           base::HistogramBase::kUmaStabilityHistogramFlag);
  80
  81   for (size_t i = 0; i < num; ++i) {
  82     base::string16 name;
  83     if (regkey.GetValueNameAt(static_cast<int>(i), &name) == ERROR_SUCCESS) {
  84       DWORD exit_code = 0;
  85       if (regkey.ReadValueDW(name.c_str(), &exit_code) == ERROR_SUCCESS) {
  86         // Do not report exit codes for processes that are still live,
  87         // notably for our own process.
  88         if (exit_code != STILL_ACTIVE || IsDeadProcess(name)) {
  89           to_delete.push_back(name);
  90           exit_code_histogram->Add(exit_code);
  91         }
  92       }
  93     }
  94   }
  95
  96   // Delete the values reported above.
  97   for (size_t i = 0; i < to_delete.size(); ++i)
  98     regkey.DeleteValue(to_delete[i].c_str());
  99 }
 100
 101 void ReadSingleExitFunnel(
 102     base::win::RegKey* parent_key, const base::char16* name,
 103     std::vector<std::pair<base::string16, int64>>* events_out) {
 104   DCHECK(parent_key);
 105   DCHECK(name);
 106   DCHECK(events_out);
 107
 108   base::win::RegKey regkey(parent_key->Handle(), name, KEY_READ | KEY_WRITE);
 109   if (!regkey.Valid())
 110     return;
 111
 112   // Exit early if no work to do.
 113   size_t num = regkey.GetValueCount();
 114   if (num == 0)
 115     return;
 116
 117   // Enumerate the recorded events for this process for processing.
 118   std::vector<std::pair<base::string16, int64>> events;
 119   for (size_t i = 0; i < num; ++i) {
 120     base::string16 event_name;
 121     LONG res = regkey.GetValueNameAt(static_cast<int>(i), &event_name);
 122     if (res == ERROR_SUCCESS) {
 123       int64 event_time = 0;
 124       res = regkey.ReadInt64(event_name.c_str(), &event_time);
 125       if (res == ERROR_SUCCESS)
 126         events.push_back(std::make_pair(event_name, event_time));
 127     }
 128   }
 129
 130   // Attempt to delete the values before reporting anything.
 131   // Exit if this fails to make sure there is no double-reporting on e.g.
 132   // permission problems or other corruption.
 133   for (size_t i = 0; i < events.size(); ++i) {
 134     const base::string16& event_name = events[i].first;
 135     LONG res = regkey.DeleteValue(event_name.c_str());
 136     if (res != ERROR_SUCCESS) {
 137       LOG(ERROR) << "Failed to delete value " << event_name;
 138       return;
 139     }
 140   }
 141
 142   events_out->swap(events);
 143 }
 144
 145 void MaybeRecordSingleExitFunnel(base::win::RegKey* parent_key,
 146                                  const base::char16* name,
 147                                  bool report) {
 148   std::vector<std::pair<base::string16, int64>> events;
 149   ReadSingleExitFunnel(parent_key, name, &events);
 150   if (!report)
 151     return;
 152
 153   // Find the earliest event time.
 154   int64 min_time = std::numeric_limits<int64>::max();
 155   for (size_t i = 0; i < events.size(); ++i)
 156     min_time = std::min(min_time, events[i].second);
 157
 158   // Record the exit funnel event times in a sparse stability histogram.
 159   for (size_t i = 0; i < events.size(); ++i) {
 160     std::string histogram_name(
 161         WatcherMetricsProviderWin::kExitFunnelHistogramPrefix);
 162     histogram_name.append(base::WideToUTF8(events[i].first));
 163     base::TimeDelta event_time =
 164         base::Time::FromInternalValue(events[i].second) -
 165             base::Time::FromInternalValue(min_time);
 166     base::HistogramBase* histogram =
 167         base::SparseHistogram::FactoryGet(
 168             histogram_name.c_str(),
 169             base::HistogramBase::kUmaStabilityHistogramFlag);
 170
 171     // Record the time rounded up to the nearest millisecond.
 172     histogram->Add(event_time.InMillisecondsRoundedUp());
 173   }
 174 }
 175
 176 void MaybeRecordExitFunnels(const base::string16& registry_path, bool report) {
 177   base::win::RegistryKeyIterator it(HKEY_CURRENT_USER, registry_path.c_str());
 178   if (!it.Valid())
 179     return;
 180
 181   // Exit early if no work to do.
 182   if (it.SubkeyCount() == 0)
 183     return;
 184
 185   // Open the key we use for deletion preemptively to prevent reporting
 186   // multiple times on permission problems.
 187   base::win::RegKey key(HKEY_CURRENT_USER,
 188                         registry_path.c_str(),
 189                         KEY_QUERY_VALUE);
 190   if (!key.Valid()) {
 191     LOG(ERROR) << "Failed to open " << registry_path << " for writing.";
 192     return;
 193   }
 194
 195   std::vector<base::string16> to_delete;
 196   for (; it.Valid(); ++it) {
 197     // Defer reporting on still-live processes.
 198     if (IsDeadProcess(it.Name())) {
 199       MaybeRecordSingleExitFunnel(&key, it.Name(), report);
 200       to_delete.push_back(it.Name());
 201     }
 202   }
 203
 204   for (size_t i = 0; i < to_delete.size(); ++i) {
 205     LONG res = key.DeleteEmptyKey(to_delete[i].c_str());
 206     if (res != ERROR_SUCCESS)
 207       LOG(ERROR) << "Failed to delete key " << to_delete[i];
 208   }
 209 }
 210
 211 }  // namespace
 212
 213 const char WatcherMetricsProviderWin::kBrowserExitCodeHistogramName[] =
 214     "Stability.BrowserExitCodes";
 215 const char WatcherMetricsProviderWin::kExitFunnelHistogramPrefix[] =
 216     "Stability.ExitFunnel.";
 217
 218 WatcherMetricsProviderWin::WatcherMetricsProviderWin(
 219     const base::char16* registry_path, bool report_exit_funnels) :
 220         registry_path_(registry_path),
 221         report_exit_funnels_(report_exit_funnels) {
 222 }
 223
 224 WatcherMetricsProviderWin::~WatcherMetricsProviderWin() {
 225 }
 226
 227 void WatcherMetricsProviderWin::ProvideStabilityMetrics(
 228     metrics::SystemProfileProto* /* system_profile_proto */) {
 229   // Note that if there are multiple instances of Chrome running in the same
 230   // user account, there's a small race that will double-report the exit codes
 231   // from both/multiple instances. This ought to be vanishingly rare and will
 232   // only manifest as low-level "random" noise. To work around this it would be
 233   // necessary to implement some form of global locking, which is not worth it
 234   // here.
 235   RecordExitCodes(registry_path_);
 236   MaybeRecordExitFunnels(registry_path_, report_exit_funnels_);
 237 }
 238
 239 }  // namespace browser_watcher