Multi-Process Tracking Support

base/debug/activity_tracker.h

 // Copyright 2016 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 // Activity tracking provides a low-overhead method of collecting information
 // about the state of the application for analysis both while it is running
 // and after it has terminated unexpectedly. Its primary purpose is to help
 // locate reasons the browser becomes unresponsive by providing insight into
 // what all the various threads and processes are (or were) doing.
 
 #ifndef BASE_DEBUG_ACTIVITY_TRACKER_H_
 #define BASE_DEBUG_ACTIVITY_TRACKER_H_
 
 // std::atomic is undesired due to performance issues when used as global
 // variables. There are no such instances here. This module uses the
 // PersistentMemoryAllocator which also uses std::atomic and is written
 // by the same author.
 #include <atomic>
 #include <map>
 #include <memory>
 #include <string>
 #include <vector>
 
 #include "base/atomicops.h"
 #include "base/base_export.h"
+#include "base/callback.h"
 #include "base/compiler_specific.h"
 #include "base/gtest_prod_util.h"
 #include "base/location.h"
 #include "base/metrics/persistent_memory_allocator.h"
+#include "base/process/process_handle.h"
 #include "base/strings/string_piece.h"
 #include "base/strings/utf_string_conversions.h"
+#include "base/task_runner.h"
 #include "base/threading/platform_thread.h"
 #include "base/threading/thread_checker.h"
 #include "base/threading/thread_local_storage.h"
 
 namespace base {
 
 struct PendingTask;
 
 class FilePath;
 class Lock;
 class PlatformThreadHandle;
 class Process;
-class StaticAtomicSequenceNumber;
 class WaitableEvent;
 
 namespace debug {
 
 class ThreadActivityTracker;
 
 
 enum : int {
   // The maximum number of call-stack addresses stored per activity. This
   // cannot be changed without also changing the version number of the
   // structure. See kTypeIdActivityTracker in GlobalActivityTracker.
   kActivityCallStackSize = 10,
 };
 
+// A class for keeping all information needed to verify that a structure is
+// associated with a given process.
+struct OwningProcess {
+  OwningProcess();
+  ~OwningProcess();
+
+  // Initializes structure with the current process id and the current time.
+  // These can uniquely identify a process. A unique non-zero data_id will be
+  // set making it possible to tell using atomic reads if the data has changed.
+  void Release_Initialize();
+
+  // Explicitly sets the process ID.
+  void SetOwningProcessIdForTesting(ProcessId pid, int64_t stamp);
+
+  // Gets the associated process ID, in native form, and the creation timestamp
+  // from memory without loading the entire structure for analysis. This will
+  // return false if no valid process ID is available.
+  static bool GetOwningProcessId(const void* memory,
+                                 ProcessId* out_id,
+                                 int64_t* out_stamp);
+
+  // SHA1(base::debug::OwningProcess): Increment this if structure changes!
+  static constexpr uint32_t kPersistentTypeId = 0xB1179672 + 1;
+
+  // Expected size for 32/64-bit check by PersistentMemoryAllocator.
+  static constexpr size_t kExpectedInstanceSize = 24;
+
+  std::atomic<uint32_t> data_id;
+  uint32_t padding;
+  int64_t process_id;
+  int64_t create_stamp;
+};
+
 // The data associated with an activity is dependent upon the activity type.
 // This union defines all of the various fields. All fields must be explicitly
 // sized types to ensure no interoperability problems between 32-bit and
 // 64-bit systems.
 union ActivityData {
   // Generic activities don't have any defined structure.
   struct {
     uint32_t id;   // An arbitrary identifier used for association.
     int32_t info;  // An arbitrary value used for information purposes.
   } generic;
@@ skipping 217 matching lines @@
   static void FillFrom(Activity* activity,
                        const void* program_counter,
                        const void* origin,
                        Type type,
                        const ActivityData& data);
 };
 
 // This class manages arbitrary user data that can be associated with activities
 // done by a thread by supporting key/value pairs of any type. This can provide
 // additional information during debugging. It is also used to store arbitrary
-// global data. All updates must be done from the same thread.
+// global data. All updates must be done from the same thread though other
+// threads can read it concurrently if they create new objects using the same
+// memory.
 class BASE_EXPORT ActivityUserData {
  public:
   // List of known value type. REFERENCE types must immediately follow the non-
   // external types.
   enum ValueType : uint8_t {
     END_OF_VALUES = 0,
     RAW_VALUE,
     RAW_VALUE_REFERENCE,
     STRING_VALUE,
     STRING_VALUE_REFERENCE,
@@ skipping 41 matching lines @@
 
   using Snapshot = std::map<std::string, TypedValue>;
 
   ActivityUserData(void* memory, size_t size);
   virtual ~ActivityUserData();
 
   // Gets the unique ID number for this user data. If this changes then the
   // contents have been overwritten by another thread. The return value is
   // always non-zero unless it's actually just a data "sink".
   uint32_t id() const {
-    return memory_ ? id_->load(std::memory_order_relaxed) : 0;
+    return header_ ? header_->owner.data_id.load(std::memory_order_relaxed) : 0;
   }
 
   // Writes a |value| (as part of a key/value pair) that will be included with
   // the activity in any reports. The same |name| can be written multiple times
   // with each successive call overwriting the previously stored |value|. For
   // raw and string values, the maximum size of successive writes is limited by
   // the first call. The length of "name" is limited to 255 characters.
   //
   // This information is stored on a "best effort" basis. It may be dropped if
   // the memory buffer is full or the associated activity is beyond the maximum
@@ skipping 33 matching lines @@
   }
 
   // Creates a snapshot of the key/value pairs contained within. The returned
   // data will be fixed, independent of whatever changes afterward. There is
   // protection against concurrent modification of the values but no protection
   // against a complete overwrite of the contents; the caller must ensure that
   // the memory segment is not going to be re-initialized while this runs.
   bool CreateSnapshot(Snapshot* output_snapshot) const;
 
   // Gets the base memory address used for storing data.
-  const void* GetBaseAddress();
+  const void* GetBaseAddress() const;
+
+  // Explicitly sets the process ID.
+  void SetOwningProcessIdForTesting(ProcessId pid, int64_t stamp);
+
+  // Gets the associated process ID, in native form, and the creation timestamp
+  // from tracker memory without loading the entire structure for analysis. This
+  // will return false if no valid process ID is available.
+  static bool GetOwningProcessId(const void* memory,
+                                 ProcessId* out_id,
+                                 int64_t* out_stamp);
 
  protected:
   virtual void Set(StringPiece name,
                    ValueType type,
                    const void* memory,
                    size_t size);
 
  private:
   FRIEND_TEST_ALL_PREFIXES(ActivityTrackerTest, UserDataTest);
 
   enum : size_t { kMemoryAlignment = sizeof(uint64_t) };
 
+  // A structure that defines the structure header in memory.
+  struct MemoryHeader {
+    MemoryHeader();
+    ~MemoryHeader();
+
+    OwningProcess owner;  // Information about the creating process.
+  };
+
+  // Header to a key/value record held in persistent memory.
+  struct FieldHeader {
+    FieldHeader();
+    ~FieldHeader();
+
+    std::atomic<uint8_t> type;         // Encoded ValueType
+    uint8_t name_size;                 // Length of "name" key.
+    std::atomic<uint16_t> value_size;  // Actual size of of the stored value.
+    uint16_t record_size;              // Total storage of name, value, header.
+  };
+
   // A structure used to reference data held outside of persistent memory.
   struct ReferenceRecord {
     uint64_t address;
     uint64_t size;
   };
 
-  // Header to a key/value record held in persistent memory.
-  struct Header {
-    std::atomic<uint8_t> type;         // Encoded ValueType
-    uint8_t name_size;                 // Length of "name" key.
-    std::atomic<uint16_t> value_size;  // Actual size of of the stored value.
-    uint16_t record_size;              // Total storage of name, value, header.
-  };
-
   // This record is used to hold known value is a map so that they can be
   // found and overwritten later.
   struct ValueInfo {
     ValueInfo();
     ValueInfo(ValueInfo&&);
     ~ValueInfo();
 
     StringPiece name;                 // The "key" of the record.
     ValueType type;                   // The type of the value.
     void* memory;                     // Where the "value" is held.
@@ skipping 14 matching lines @@
   // updates of the values. This is "mutable" because it changes on "const"
   // objects even when the actual data values can't change.
   mutable std::map<StringPiece, ValueInfo> values_;
 
   // Information about the memory block in which new data can be stored. These
   // are "mutable" because they change even on "const" objects that are just
   // skipping already set values.
   mutable char* memory_;
   mutable size_t available_;
 
-  // A pointer to the unique ID for this instance.
-  std::atomic<uint32_t>* const id_;
-
-  // This ID is used to create unique indentifiers for user data so that it's
-  // possible to tell if the information has been overwritten.
-  static StaticAtomicSequenceNumber next_id_;
+  // A pointer to the memory header for this instance.
+  MemoryHeader* const header_;
 
   DISALLOW_COPY_AND_ASSIGN(ActivityUserData);
 };
 
 // This class manages tracking a stack of activities for a single thread in
 // a persistent manner, implementing a bounded-size stack in a fixed-size
 // memory allocation. In order to support an operational mode where another
 // thread is analyzing this data in real-time, atomic operations are used
 // where necessary to guarantee a consistent view from the outside.
 //
@@ skipping 122 matching lines @@
   // Returns whether the current data is valid or not. It is not valid if
   // corruption has been detected in the header or other data structures.
   bool IsValid() const;
 
   // Gets a copy of the tracker contents for analysis. Returns false if a
   // snapshot was not possible, perhaps because the data is not valid; the
   // contents of |output_snapshot| are undefined in that case. The current
   // implementation does not support concurrent snapshot operations.
   bool CreateSnapshot(Snapshot* output_snapshot) const;
 
+  // Gets the base memory address used for storing data.
+  const void* GetBaseAddress();
+
+  // Explicitly sets the process ID.
+  void SetOwningProcessIdForTesting(ProcessId pid, int64_t stamp);
+
+  // Gets the associated process ID, in native form, and the creation timestamp
+  // from tracker memory without loading the entire structure for analysis. This
+  // will return false if no valid process ID is available.
+  static bool GetOwningProcessId(const void* memory,
+                                 ProcessId* out_id,
+                                 int64_t* out_stamp);
+
   // Calculates the memory size required for a given stack depth, including
   // the internal header structure for the stack.
   static size_t SizeForStackDepth(int stack_depth);
 
  private:
   friend class ActivityTrackerTest;
 
   Header* const header_;        // Pointer to the Header structure.
   Activity* const stack_;       // The stack of activities.
   const uint32_t stack_slots_;  // The total number of stack slots.
@@ skipping 11 matching lines @@
 // for the data to be analyzed by a parallel process or even post-mortem.
 class BASE_EXPORT GlobalActivityTracker {
  public:
   // Type identifiers used when storing in persistent memory so they can be
   // identified during extraction; the first 4 bytes of the SHA1 of the name
   // is used as a unique integer. A "version number" is added to the base
   // so that, if the structure of that object changes, stored older versions
   // will be safely ignored. These are public so that an external process
   // can recognize records of this type within an allocator.
   enum : uint32_t {
-    kTypeIdActivityTracker = 0x5D7381AF + 3,   // SHA1(ActivityTracker) v3
-    kTypeIdUserDataRecord = 0x615EDDD7 + 2,    // SHA1(UserDataRecord) v2
+    kTypeIdActivityTracker = 0x5D7381AF + 4,   // SHA1(ActivityTracker) v4
+    kTypeIdUserDataRecord = 0x615EDDD7 + 3,    // SHA1(UserDataRecord) v3
     kTypeIdGlobalLogMessage = 0x4CF434F9 + 1,  // SHA1(GlobalLogMessage) v1
-    kTypeIdGlobalDataRecord = kTypeIdUserDataRecord + 1000,
+    kTypeIdProcessDataRecord = kTypeIdUserDataRecord + 0x100,
+    kTypeIdGlobalDataRecord = kTypeIdUserDataRecord + 0x200,
 
     kTypeIdActivityTrackerFree = ~kTypeIdActivityTracker,
     kTypeIdUserDataRecordFree = ~kTypeIdUserDataRecord,
+    kTypeIdProcessDataRecordFree = ~kTypeIdProcessDataRecord,
   };
 
+  // An enumeration of common process life stages. All entries are given an
+  // explicit number so they are known and remain constant; this allows for
+  // cross-version analysis either locally or on a server.
+  enum ProcessPhase : int {
+    // The phases are generic and may have meaning to the tracker.
+    PROCESS_PHASE_UNKNOWN = 0,
+    PROCESS_LAUNCHED = 1,
+    PROCESS_LAUNCH_FAILED = 2,
+    PROCESS_EXITED_CLEANLY = 10,
+    PROCESS_EXITED_WITH_CODE = 11,
+
+    // Add here whatever is useful for analysis.
+    PROCESS_SHUTDOWN_STARTED = 100,
+    PROCESS_MAIN_LOOP_STARTED = 101,
+  };
+
+  // A callback made when a process exits to allow immediate analysis of its
+  // data. Note that the system may reuse the |process_id| so when fetching
+  // records it's important to ensure that what is returned was created before
+  // the |exit_stamp|. Movement of |process_data| information is allowed.
+  using ProcessExitCallback =
+      Callback<void(int64_t process_id,
+                    int64_t exit_stamp,
+                    int exit_code,
+                    ProcessPhase exit_phase,
+                    std::string&& command_line,
+                    ActivityUserData::Snapshot&& process_data)>;
+
   // This structure contains information about a loaded module, as shown to
   // users of the tracker.
   struct BASE_EXPORT ModuleInfo {
     ModuleInfo();
     ModuleInfo(ModuleInfo&& rhs);
     ModuleInfo(const ModuleInfo& rhs);
     ~ModuleInfo();
 
     ModuleInfo& operator=(ModuleInfo&& rhs);
     ModuleInfo& operator=(const ModuleInfo& rhs);
@@ skipping 111 matching lines @@
       return tracker;
     return CreateTrackerForCurrentThread();
   }
 
   // Creates an activity-tracker for the current thread.
   ThreadActivityTracker* CreateTrackerForCurrentThread();
 
   // Releases the activity-tracker for the current thread (for testing only).
   void ReleaseTrackerForCurrentThreadForTesting();
 
+  // Sets a task-runner that can be used for background work.
+  void SetBackgroundTaskRunner(const scoped_refptr<TaskRunner>& runner);
+
+  // Sets an optional callback to be called when a process exits.
+  void SetProcessExitCallback(ProcessExitCallback callback);
+
+  // Manages process lifetimes. These are called by the process that launched
+  // and reaped the subprocess, not the subprocess itself. If it is expensive
+  // to generate the parameters, Get() the global tracker and call these
+  // conditionally rather than using the static versions.
+  void RecordProcessLaunch(ProcessId process_id,
+                           const FilePath::StringType& cmd);
+  void RecordProcessLaunch(ProcessId process_id,
+                           const FilePath::StringType& exe,
+                           const FilePath::StringType& args);
+  void RecordProcessExit(ProcessId process_id, int exit_code);
+  static void RecordProcessLaunchIfEnabled(ProcessId process_id,
+                                           const FilePath::StringType& cmd) {
+    GlobalActivityTracker* tracker = Get();
+    if (tracker)
+      tracker->RecordProcessLaunch(process_id, cmd);
+  }
+  static void RecordProcessLaunchIfEnabled(ProcessId process_id,
+                                           const FilePath::StringType& exe,
+                                           const FilePath::StringType& args) {
+    GlobalActivityTracker* tracker = Get();
+    if (tracker)
+      tracker->RecordProcessLaunch(process_id, exe, args);
+  }
+  static void RecordProcessExitIfEnabled(ProcessId process_id, int exit_code) {
+    GlobalActivityTracker* tracker = Get();
+    if (tracker)
+      tracker->RecordProcessExit(process_id, exit_code);
+  }
+  // Sets the "phase" of the current process, useful for knowing what it was
+  // doing when it last reported.
+  void SetProcessPhase(ProcessPhase phase);
+  static void SetProcessPhaseIfEnabled(ProcessPhase phase) {
+    GlobalActivityTracker* tracker = Get();
+    if (tracker)
+      tracker->SetProcessPhase(phase);
+  }
+
   // Records a log message. The current implementation does NOT recycle these
   // only store critical messages such as FATAL ones.
   void RecordLogMessage(StringPiece message);
   static void RecordLogMessageIfEnabled(StringPiece message) {
     GlobalActivityTracker* tracker = Get();
     if (tracker)
       tracker->RecordLogMessage(message);
   }
 
   // Records a module load/unload event. This is safe to call multiple times
   // even with the same information.
   void RecordModuleInfo(const ModuleInfo& info);
   static void RecordModuleInfoIfEnabled(const ModuleInfo& info) {
     GlobalActivityTracker* tracker = Get();
     if (tracker)
       tracker->RecordModuleInfo(info);
   }
 
   // Record field trial information. This call is thread-safe. In addition to
   // this, construction of a GlobalActivityTracker will cause all existing
   // active field trials to be fetched and recorded.
   void RecordFieldTrial(const std::string& trial_name, StringPiece group_name);
   static void RecordFieldTrialIfEnabled(const std::string& trial_name,
                                         StringPiece group_name) {
     GlobalActivityTracker* tracker = Get();
     if (tracker)
       tracker->RecordFieldTrial(trial_name, group_name);
   }
 
+  // Accesses the process data record for storing arbitrary key/value pairs.
+  // Updates to this are thread-safe.
+  ActivityUserData& process_data() { return process_data_; }
+
   // Accesses the global data record for storing arbitrary key/value pairs.
+  // Updates to this are thread-safe.
   ActivityUserData& global_data() { return global_data_; }
 
  private:
   friend class GlobalActivityAnalyzer;
   friend class ScopedThreadActivity;
   friend class ActivityTrackerTest;
 
   enum : int {
     // The maximum number of threads that can be tracked within a process. If
     // more than this number run concurrently, tracking of new ones may cease.
     kMaxThreadCount = 100,
     kCachedThreadMemories = 10,
     kCachedUserDataMemories = 10,
   };
 
   // A wrapper around ActivityUserData that is thread-safe and thus can be used
   // in the global scope without the requirement of being called from only one
   // thread.
-  class GlobalUserData : public ActivityUserData {
+  class ThreadSafeUserData : public ActivityUserData {
    public:
-    GlobalUserData(void* memory, size_t size);
-    ~GlobalUserData() override;
+    ThreadSafeUserData(void* memory, size_t size);
+    ~ThreadSafeUserData() override;
 
    private:
     void Set(StringPiece name,
              ValueType type,
              const void* memory,
              size_t size) override;
 
     Lock data_lock_;
 
-    DISALLOW_COPY_AND_ASSIGN(GlobalUserData);
+    DISALLOW_COPY_AND_ASSIGN(ThreadSafeUserData);
   };
 
   // State of a module as stored in persistent memory. This supports a single
   // loading of a module only. If modules are loaded multiple times at
   // different addresses, only the last will be recorded and an unload will
   // not revert to the information of any other addresses.
   struct BASE_EXPORT ModuleInfoRecord {
     // SHA1(ModuleInfoRecord): Increment this if structure changes!
     static constexpr uint32_t kPersistentTypeId = 0x05DB5F41 + 1;
 
     // Expected size for 32/64-bit check by PersistentMemoryAllocator.
-    static constexpr size_t kExpectedInstanceSize = 56;
+    static constexpr size_t kExpectedInstanceSize =
+        OwningProcess::kExpectedInstanceSize + 56;
 
     // The atomic unfortunately makes this a "complex" class on some compilers
     // and thus requires an out-of-line constructor & destructor even though
     // they do nothing.
     ModuleInfoRecord();
     ~ModuleInfoRecord();
 
+    OwningProcess owner;            // The process that created this record.
     uint64_t address;               // The base address of the module.
     uint64_t load_time;             // Time of last load/unload.
     uint64_t size;                  // The size of the module in bytes.
     uint32_t timestamp;             // Opaque timestamp of the module.
     uint32_t age;                   // Opaque "age" associated with the module.
     uint8_t identifier[16];         // Opaque identifier for the module.
     std::atomic<uint32_t> changes;  // Number load/unload actions.
     uint16_t pickle_size;           // The size of the following pickle.
     uint8_t loaded;                 // Flag if module is loaded or not.
     char pickle[1];                 // Other strings; may allocate larger.
@@ skipping 43 matching lines @@
 
   // Returns the memory used by an activity-tracker managed by this class.
   // It is called during the destruction of a ManagedActivityTracker object.
   void ReturnTrackerMemory(ManagedActivityTracker* tracker);
 
   // Releases the activity-tracker associcated with thread. It is called
   // automatically when a thread is joined and thus there is nothing more to
   // be tracked. |value| is a pointer to a ManagedActivityTracker.
   static void OnTLSDestroy(void* value);
 
+  // Does process-exit work. This can be run on any thread.
+  void CleanupAfterProcess(ProcessId process_id,
+                           int64_t exit_stamp,
+                           int exit_code,
+                           std::string&& command_line);
+
   // The persistent-memory allocator from which the memory for all trackers
   // is taken.
   std::unique_ptr<PersistentMemoryAllocator> allocator_;
 
   // The size (in bytes) of memory required by a ThreadActivityTracker to
   // provide the stack-depth requested during construction.
   const size_t stack_memory_size_;
 
   // The activity tracker for the currently executing thread.
   base::ThreadLocalStorage::Slot this_thread_tracker_;
 
   // The number of thread trackers currently active.
   std::atomic<int> thread_tracker_count_;
 
   // A caching memory allocator for thread-tracker objects.
   ActivityTrackerMemoryAllocator thread_tracker_allocator_;
   base::Lock thread_tracker_allocator_lock_;
 
   // A caching memory allocator for user data attached to activity data.
   ActivityTrackerMemoryAllocator user_data_allocator_;
   base::Lock user_data_allocator_lock_;
 
-  // An object for holding global arbitrary key value pairs. Values must always
-  // be written from the main UI thread.
-  GlobalUserData global_data_;
+  // An object for holding arbitrary key value pairs with thread-safe access.
+  ThreadSafeUserData process_data_;
+  ThreadSafeUserData global_data_;
 
   // A map of global module information, keyed by module path.
   std::map<const std::string, ModuleInfoRecord*> modules_;
   base::Lock modules_lock_;
 
   // The active global activity tracker.
   static subtle::AtomicWord g_tracker_;
 
+  // A lock that is used to protect access to the following fields.
+  base::Lock global_tracker_lock_;
+
+  // The collection of processes being tracked and their command-lines.
+  std::map<int64_t, std::string> known_processes_;
+
+  // A task-runner that can be used for doing background processing.
+  scoped_refptr<TaskRunner> background_task_runner_;
+
+  // A callback performed when a subprocess exits, including its exit-code
+  // and the phase it was in when that occurred. This will be called via
+  // the |background_task_runner_| if one is set or whatever thread reaped
+  // the process otherwise.
+  ProcessExitCallback process_exit_callback_;
+
   DISALLOW_COPY_AND_ASSIGN(GlobalActivityTracker);
 };
 
 
 // Record entry in to and out of an arbitrary block of code.
 class BASE_EXPORT ScopedActivity
     : public GlobalActivityTracker::ScopedThreadActivity {
  public:
   // Track activity at the specified FROM_HERE location for an arbitrary
   // 4-bit |action|, an arbitrary 32-bit |id|, and 32-bits of arbitrary
@@ skipping 114 matching lines @@
   ScopedProcessWaitActivity(const void* program_counter,
                             const base::Process* process);
   DISALLOW_COPY_AND_ASSIGN(ScopedProcessWaitActivity);
 };
 #endif
 
 }  // namespace debug
 }  // namespace base
 
 #endif  // BASE_DEBUG_ACTIVITY_TRACKER_H_