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 void CompileAsserts() {
  22   // Process ID APIs on Windows talk in DWORDs, whereas for string formatting
  23   // and parsing, this code uses int. In practice there are no process IDs with
  24   // the high bit set on Windows, so there's no danger of overflow if this is
  25   // done consistently.
  26   static_assert(sizeof(DWORD) == sizeof(int),
  27                 "process ids are expected to be no larger than int");
  28 }
  29
  30 // This function does soft matching on the PID recorded in the key only.
  31 // Due to PID reuse, the possibility exists that the process that's now live
  32 // with the given PID is not the same process the data was recorded for.
  33 // This doesn't matter for the purpose, as eventually the data will be
  34 // scavenged and reported.
  35 bool IsDeadProcess(base::StringPiece16 key_or_value_name) {
  36   // Truncate the input string to the first occurrence of '-', if one exists.
  37   size_t num_end = key_or_value_name.find(L'-');
  38   if (num_end != base::StringPiece16::npos)
  39     key_or_value_name = key_or_value_name.substr(0, num_end);
  40
  41   // Convert to the numeric PID.
  42   int pid = 0;
  43   if (!base::StringToInt(key_or_value_name, &pid) || pid == 0)
  44     return true;
  45
  46   // This is a very inexpensive check for the common case of our own PID.
  47   if (static_cast<base::ProcessId>(pid) == base::GetCurrentProcId())
  48     return false;
  49
  50   // The process is not our own - see whether a process with this PID exists.
  51   // This is more expensive than the above check, but should also be very rare,
  52   // as this only happens more than once for a given PID if a user is running
  53   // multiple Chrome instances concurrently.
  54   base::Process process =
  55       base::Process::Open(static_cast<base::ProcessId>(pid));
  56   if (process.IsValid()) {
  57     // The fact that it was possible to open the process says it's live.
  58     return false;
  59   }
  60
  61   return true;
  62 }
  63
  64 void RecordExitCodes(const base::string16& registry_path) {
  65   base::win::RegKey regkey(HKEY_CURRENT_USER,
  66                            registry_path.c_str(),
  67                            KEY_QUERY_VALUE | KEY_SET_VALUE);
  68   if (!regkey.Valid())
  69     return;
  70
  71   size_t num = regkey.GetValueCount();
  72   if (num == 0)
  73     return;
  74   std::vector<base::string16> to_delete;
  75
  76   // Record the exit codes in a sparse stability histogram, as the range of
  77   // values used to report failures is large.
  78   base::HistogramBase* exit_code_histogram =
  79       base::SparseHistogram::FactoryGet(
  80           WatcherMetricsProviderWin::kBrowserExitCodeHistogramName,
  81           base::HistogramBase::kUmaStabilityHistogramFlag);
  82
  83   for (size_t i = 0; i < num; ++i) {
  84     base::string16 name;
  85     if (regkey.GetValueNameAt(static_cast<int>(i), &name) == ERROR_SUCCESS) {
  86       DWORD exit_code = 0;
  87       if (regkey.ReadValueDW(name.c_str(), &exit_code) == ERROR_SUCCESS) {
  88         // Do not report exit codes for processes that are still live,
  89         // notably for our own process.
  90         if (exit_code != STILL_ACTIVE || IsDeadProcess(name)) {
  91           to_delete.push_back(name);
  92           exit_code_histogram->Add(exit_code);
  93         }
  94       }
  95     }
  96   }
  97
  98   // Delete the values reported above.
  99   for (size_t i = 0; i < to_delete.size(); ++i)
 100     regkey.DeleteValue(to_delete[i].c_str());
 101 }
 102
 103 void ReadSingleExitFunnel(
 104     base::win::RegKey* parent_key, const base::char16* name,
 105     std::vector<std::pair<base::string16, int64>>* events_out) {
 106   DCHECK(parent_key);
 107   DCHECK(name);
 108   DCHECK(events_out);
 109
 110   base::win::RegKey regkey(parent_key->Handle(), name, KEY_READ | KEY_WRITE);
 111   if (!regkey.Valid())
 112     return;
 113
 114   // Exit early if no work to do.
 115   size_t num = regkey.GetValueCount();
 116   if (num == 0)
 117     return;
 118
 119   // Enumerate the recorded events for this process for processing.
 120   std::vector<std::pair<base::string16, int64>> events;
 121   for (size_t i = 0; i < num; ++i) {
 122     base::string16 event_name;
 123     LONG res = regkey.GetValueNameAt(static_cast<int>(i), &event_name);
 124     if (res == ERROR_SUCCESS) {
 125       int64 event_time = 0;
 126       res = regkey.ReadInt64(event_name.c_str(), &event_time);
 127       if (res == ERROR_SUCCESS)
 128         events.push_back(std::make_pair(event_name, event_time));
 129     }
 130   }
 131
 132   // Attempt to delete the values before reporting anything.
 133   // Exit if this fails to make sure there is no double-reporting on e.g.
 134   // permission problems or other corruption.
 135   for (size_t i = 0; i < events.size(); ++i) {
 136     const base::string16& event_name = events[i].first;
 137     LONG res = regkey.DeleteValue(event_name.c_str());
 138     if (res != ERROR_SUCCESS) {
 139       LOG(ERROR) << "Failed to delete value " << event_name;
 140       return;
 141     }
 142   }
 143
 144   events_out->swap(events);
 145 }
 146
 147 void MaybeRecordSingleExitFunnel(base::win::RegKey* parent_key,
 148                                  const base::char16* name,
 149                                  bool report) {
 150   std::vector<std::pair<base::string16, int64>> events;
 151   ReadSingleExitFunnel(parent_key, name, &events);
 152   if (!report)
 153     return;
 154
 155   // Find the earliest event time.
 156   int64 min_time = std::numeric_limits<int64>::max();
 157   for (size_t i = 0; i < events.size(); ++i)
 158     min_time = std::min(min_time, events[i].second);
 159
 160   // Record the exit funnel event times in a sparse stability histogram.
 161   for (size_t i = 0; i < events.size(); ++i) {
 162     std::string histogram_name(
 163         WatcherMetricsProviderWin::kExitFunnelHistogramPrefix);
 164     histogram_name.append(base::WideToUTF8(events[i].first));
 165     base::TimeDelta event_time =
 166         base::Time::FromInternalValue(events[i].second) -
 167             base::Time::FromInternalValue(min_time);
 168     base::HistogramBase* histogram =
 169         base::SparseHistogram::FactoryGet(
 170             histogram_name.c_str(),
 171             base::HistogramBase::kUmaStabilityHistogramFlag);
 172
 173     // Record the time rounded up to the nearest millisecond.
 174     histogram->Add(event_time.InMillisecondsRoundedUp());
 175   }
 176 }
 177
 178 void MaybeRecordExitFunnels(const base::string16& registry_path, bool report) {
 179   base::win::RegistryKeyIterator it(HKEY_CURRENT_USER, registry_path.c_str());
 180   if (!it.Valid())
 181     return;
 182
 183   // Exit early if no work to do.
 184   if (it.SubkeyCount() == 0)
 185     return;
 186
 187   // Open the key we use for deletion preemptively to prevent reporting
 188   // multiple times on permission problems.
 189   base::win::RegKey key(HKEY_CURRENT_USER,
 190                         registry_path.c_str(),
 191                         KEY_QUERY_VALUE);
 192   if (!key.Valid()) {
 193     LOG(ERROR) << "Failed to open " << registry_path << " for writing.";
 194     return;
 195   }
 196
 197   std::vector<base::string16> to_delete;
 198   for (; it.Valid(); ++it) {
 199     // Defer reporting on still-live processes.
 200     if (IsDeadProcess(it.Name())) {
 201       MaybeRecordSingleExitFunnel(&key, it.Name(), report);
 202       to_delete.push_back(it.Name());
 203     }
 204   }
 205
 206   for (size_t i = 0; i < to_delete.size(); ++i) {
 207     LONG res = key.DeleteEmptyKey(to_delete[i].c_str());
 208     if (res != ERROR_SUCCESS)
 209       LOG(ERROR) << "Failed to delete key " << to_delete[i];
 210   }
 211 }
 212
 213 }  // namespace
 214
 215 const char WatcherMetricsProviderWin::kBrowserExitCodeHistogramName[] =
 216     "Stability.BrowserExitCodes";
 217 const char WatcherMetricsProviderWin::kExitFunnelHistogramPrefix[] =
 218     "Stability.ExitFunnel.";
 219
 220 WatcherMetricsProviderWin::WatcherMetricsProviderWin(
 221     const base::char16* registry_path, bool report_exit_funnels) :
 222         registry_path_(registry_path),
 223         report_exit_funnels_(report_exit_funnels) {
 224 }
 225
 226 WatcherMetricsProviderWin::~WatcherMetricsProviderWin() {
 227 }
 228
 229 void WatcherMetricsProviderWin::ProvideStabilityMetrics(
 230     metrics::SystemProfileProto* /* system_profile_proto */) {
 231   // Note that if there are multiple instances of Chrome running in the same
 232   // user account, there's a small race that will double-report the exit codes
 233   // from both/multiple instances. This ought to be vanishingly rare and will
 234   // only manifest as low-level "random" noise. To work around this it would be
 235   // necessary to implement some form of global locking, which is not worth it
 236   // here.
 237   RecordExitCodes(registry_path_);
 238   MaybeRecordExitFunnels(registry_path_, report_exit_funnels_);
 239 }
 240
 241 }  // namespace browser_watcher