syncable.h revision dc0f95d653279beabeb9817299e2902918ba123e
1// Copyright (c) 2010 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#ifndef CHROME_BROWSER_SYNC_SYNCABLE_SYNCABLE_H_
6#define CHROME_BROWSER_SYNC_SYNCABLE_SYNCABLE_H_
7#pragma once
8
9#include <algorithm>
10#include <bitset>
11#include <iosfwd>
12#include <limits>
13#include <set>
14#include <string>
15#include <vector>
16
17#include "base/atomicops.h"
18#include "base/basictypes.h"
19#include "base/file_path.h"
20#include "base/gtest_prod_util.h"
21#include "base/synchronization/lock.h"
22#include "base/time.h"
23#include "chrome/browser/sync/protocol/sync.pb.h"
24#include "chrome/browser/sync/syncable/autofill_migration.h"
25#include "chrome/browser/sync/syncable/blob.h"
26#include "chrome/browser/sync/syncable/dir_open_result.h"
27#include "chrome/browser/sync/syncable/directory_event.h"
28#include "chrome/browser/sync/syncable/syncable_id.h"
29#include "chrome/browser/sync/syncable/model_type.h"
30#include "chrome/browser/sync/util/channel.h"
31#include "chrome/browser/sync/util/dbgq.h"
32#include "chrome/common/deprecated/event_sys.h"
33
34struct PurgeInfo;
35
36namespace sync_api {
37class ReadTransaction;
38class WriteNode;
39class ReadNode;
40}
41
42namespace syncable {
43class Entry;
44
45std::ostream& operator<<(std::ostream& s, const Entry& e);
46
47class DirectoryBackingStore;
48
49static const int64 kInvalidMetaHandle = 0;
50
51enum {
52  BEGIN_FIELDS = 0,
53  INT64_FIELDS_BEGIN = BEGIN_FIELDS
54};
55
56enum MetahandleField {
57  // Primary key into the table.  Keep this as a handle to the meta entry
58  // across transactions.
59  META_HANDLE = INT64_FIELDS_BEGIN
60};
61
62enum BaseVersion {
63  // After initial upload, the version is controlled by the server, and is
64  // increased whenever the data or metadata changes on the server.
65  BASE_VERSION = META_HANDLE + 1,
66};
67
68enum Int64Field {
69  SERVER_VERSION = BASE_VERSION + 1,
70  MTIME,
71  SERVER_MTIME,
72  CTIME,
73  SERVER_CTIME,
74
75  // A numeric position value that indicates the relative ordering of
76  // this object among its siblings.
77  SERVER_POSITION_IN_PARENT,
78
79  LOCAL_EXTERNAL_ID,  // ID of an item in the external local storage that this
80                      // entry is associated with. (such as bookmarks.js)
81
82  INT64_FIELDS_END
83};
84
85enum {
86  INT64_FIELDS_COUNT = INT64_FIELDS_END,
87  ID_FIELDS_BEGIN = INT64_FIELDS_END,
88};
89
90enum IdField {
91  // Code in InitializeTables relies on ID being the first IdField value.
92  ID = ID_FIELDS_BEGIN,
93  PARENT_ID,
94  SERVER_PARENT_ID,
95
96  PREV_ID,
97  NEXT_ID,
98  ID_FIELDS_END
99};
100
101enum {
102  ID_FIELDS_COUNT = ID_FIELDS_END - ID_FIELDS_BEGIN,
103  BIT_FIELDS_BEGIN = ID_FIELDS_END
104};
105
106enum IndexedBitField {
107  IS_UNSYNCED = BIT_FIELDS_BEGIN,
108  IS_UNAPPLIED_UPDATE,
109  INDEXED_BIT_FIELDS_END,
110};
111
112enum IsDelField {
113  IS_DEL = INDEXED_BIT_FIELDS_END,
114};
115
116enum BitField {
117  IS_DIR = IS_DEL + 1,
118  SERVER_IS_DIR,
119  SERVER_IS_DEL,
120  BIT_FIELDS_END
121};
122
123enum {
124  BIT_FIELDS_COUNT = BIT_FIELDS_END - BIT_FIELDS_BEGIN,
125  STRING_FIELDS_BEGIN = BIT_FIELDS_END
126};
127
128enum StringField {
129  // Name, will be truncated by server. Can be duplicated in a folder.
130  NON_UNIQUE_NAME = STRING_FIELDS_BEGIN,
131  // The server version of |NON_UNIQUE_NAME|.
132  SERVER_NON_UNIQUE_NAME,
133
134  // A tag string which identifies this node as a particular top-level
135  // permanent object.  The tag can be thought of as a unique key that
136  // identifies a singleton instance.
137  UNIQUE_SERVER_TAG,  // Tagged by the server
138  UNIQUE_CLIENT_TAG,  // Tagged by the client
139  STRING_FIELDS_END,
140};
141
142enum {
143  STRING_FIELDS_COUNT = STRING_FIELDS_END - STRING_FIELDS_BEGIN,
144  PROTO_FIELDS_BEGIN = STRING_FIELDS_END
145};
146
147// From looking at the sqlite3 docs, it's not directly stated, but it
148// seems the overhead for storing a NULL blob is very small.
149enum ProtoField {
150  SPECIFICS = PROTO_FIELDS_BEGIN,
151  SERVER_SPECIFICS,
152  PROTO_FIELDS_END,
153};
154
155enum {
156  PROTO_FIELDS_COUNT = PROTO_FIELDS_END - PROTO_FIELDS_BEGIN
157};
158
159enum {
160  FIELD_COUNT = PROTO_FIELDS_END,
161  // Past this point we have temporaries, stored in memory only.
162  BEGIN_TEMPS = PROTO_FIELDS_END,
163  BIT_TEMPS_BEGIN = BEGIN_TEMPS,
164};
165
166enum BitTemp {
167  SYNCING = BIT_TEMPS_BEGIN,
168  BIT_TEMPS_END,
169};
170
171enum {
172  BIT_TEMPS_COUNT = BIT_TEMPS_END - BIT_TEMPS_BEGIN
173};
174
175class BaseTransaction;
176class WriteTransaction;
177class ReadTransaction;
178class Directory;
179class ScopedDirLookup;
180
181// Instead of:
182//   Entry e = transaction.GetById(id);
183// use:
184//   Entry e(transaction, GET_BY_ID, id);
185//
186// Why?  The former would require a copy constructor, and it would be difficult
187// to enforce that an entry never outlived its transaction if there were a copy
188// constructor.
189enum GetById {
190  GET_BY_ID
191};
192
193enum GetByClientTag {
194  GET_BY_CLIENT_TAG
195};
196
197enum GetByServerTag {
198  GET_BY_SERVER_TAG
199};
200
201enum GetByHandle {
202  GET_BY_HANDLE
203};
204
205enum Create {
206  CREATE
207};
208
209enum CreateNewUpdateItem {
210  CREATE_NEW_UPDATE_ITEM
211};
212
213typedef std::set<int64> MetahandleSet;
214
215// Why the singular enums?  So the code compile-time dispatches instead of
216// runtime dispatches as it would with a single enum and an if() statement.
217
218// The EntryKernel class contains the actual data for an entry.
219struct EntryKernel {
220 private:
221  std::string string_fields[STRING_FIELDS_COUNT];
222  sync_pb::EntitySpecifics specifics_fields[PROTO_FIELDS_COUNT];
223  int64 int64_fields[INT64_FIELDS_COUNT];
224  Id id_fields[ID_FIELDS_COUNT];
225  std::bitset<BIT_FIELDS_COUNT> bit_fields;
226  std::bitset<BIT_TEMPS_COUNT> bit_temps;
227
228 public:
229  EntryKernel();
230  ~EntryKernel();
231
232  // Set the dirty bit, and optionally add this entry's metahandle to
233  // a provided index on dirty bits in |dirty_index|. Parameter may be null,
234  // and will result only in setting the dirty bit of this entry.
235  inline void mark_dirty(syncable::MetahandleSet* dirty_index) {
236    if (!dirty_ && dirty_index) {
237      DCHECK_NE(0, ref(META_HANDLE));
238      dirty_index->insert(ref(META_HANDLE));
239    }
240    dirty_ = true;
241  }
242
243  // Clear the dirty bit, and optionally remove this entry's metahandle from
244  // a provided index on dirty bits in |dirty_index|. Parameter may be null,
245  // and will result only in clearing dirty bit of this entry.
246  inline void clear_dirty(syncable::MetahandleSet* dirty_index) {
247    if (dirty_ && dirty_index) {
248      DCHECK_NE(0, ref(META_HANDLE));
249      dirty_index->erase(ref(META_HANDLE));
250    }
251    dirty_ = false;
252  }
253
254  inline bool is_dirty() const {
255    return dirty_;
256  }
257
258  // Setters.
259  inline void put(MetahandleField field, int64 value) {
260    int64_fields[field - INT64_FIELDS_BEGIN] = value;
261  }
262  inline void put(Int64Field field, int64 value) {
263    int64_fields[field - INT64_FIELDS_BEGIN] = value;
264  }
265  inline void put(IdField field, const Id& value) {
266    id_fields[field - ID_FIELDS_BEGIN] = value;
267  }
268  inline void put(BaseVersion field, int64 value) {
269    int64_fields[field - INT64_FIELDS_BEGIN] = value;
270  }
271  inline void put(IndexedBitField field, bool value) {
272    bit_fields[field - BIT_FIELDS_BEGIN] = value;
273  }
274  inline void put(IsDelField field, bool value) {
275    bit_fields[field - BIT_FIELDS_BEGIN] = value;
276  }
277  inline void put(BitField field, bool value) {
278    bit_fields[field - BIT_FIELDS_BEGIN] = value;
279  }
280  inline void put(StringField field, const std::string& value) {
281    string_fields[field - STRING_FIELDS_BEGIN] = value;
282  }
283  inline void put(ProtoField field, const sync_pb::EntitySpecifics& value) {
284    specifics_fields[field - PROTO_FIELDS_BEGIN].CopyFrom(value);
285  }
286  inline void put(BitTemp field, bool value) {
287    bit_temps[field - BIT_TEMPS_BEGIN] = value;
288  }
289
290  // Const ref getters.
291  inline int64 ref(MetahandleField field) const {
292    return int64_fields[field - INT64_FIELDS_BEGIN];
293  }
294  inline int64 ref(Int64Field field) const {
295    return int64_fields[field - INT64_FIELDS_BEGIN];
296  }
297  inline const Id& ref(IdField field) const {
298    return id_fields[field - ID_FIELDS_BEGIN];
299  }
300  inline int64 ref(BaseVersion field) const {
301    return int64_fields[field - INT64_FIELDS_BEGIN];
302  }
303  inline bool ref(IndexedBitField field) const {
304    return bit_fields[field - BIT_FIELDS_BEGIN];
305  }
306  inline bool ref(IsDelField field) const {
307    return bit_fields[field - BIT_FIELDS_BEGIN];
308  }
309  inline bool ref(BitField field) const {
310    return bit_fields[field - BIT_FIELDS_BEGIN];
311  }
312  inline const std::string& ref(StringField field) const {
313    return string_fields[field - STRING_FIELDS_BEGIN];
314  }
315  inline const sync_pb::EntitySpecifics& ref(ProtoField field) const {
316    return specifics_fields[field - PROTO_FIELDS_BEGIN];
317  }
318  inline bool ref(BitTemp field) const {
319    return bit_temps[field - BIT_TEMPS_BEGIN];
320  }
321
322  // Non-const, mutable ref getters for object types only.
323  inline std::string& mutable_ref(StringField field) {
324    return string_fields[field - STRING_FIELDS_BEGIN];
325  }
326  inline sync_pb::EntitySpecifics& mutable_ref(ProtoField field) {
327    return specifics_fields[field - PROTO_FIELDS_BEGIN];
328  }
329  inline Id& mutable_ref(IdField field) {
330    return id_fields[field - ID_FIELDS_BEGIN];
331  }
332
333 private:
334  // Tracks whether this entry needs to be saved to the database.
335  bool dirty_;
336};
337
338// A read-only meta entry.
339class Entry {
340  friend class Directory;
341  friend std::ostream& operator << (std::ostream& s, const Entry& e);
342
343 public:
344  // After constructing, you must check good() to test whether the Get
345  // succeeded.
346  Entry(BaseTransaction* trans, GetByHandle, int64 handle);
347  Entry(BaseTransaction* trans, GetById, const Id& id);
348  Entry(BaseTransaction* trans, GetByServerTag, const std::string& tag);
349  Entry(BaseTransaction* trans, GetByClientTag, const std::string& tag);
350
351  bool good() const { return 0 != kernel_; }
352
353  BaseTransaction* trans() const { return basetrans_; }
354
355  // Field accessors.
356  inline int64 Get(MetahandleField field) const {
357    DCHECK(kernel_);
358    return kernel_->ref(field);
359  }
360  inline Id Get(IdField field) const {
361    DCHECK(kernel_);
362    return kernel_->ref(field);
363  }
364  inline int64 Get(Int64Field field) const {
365    DCHECK(kernel_);
366    return kernel_->ref(field);
367  }
368  inline int64 Get(BaseVersion field) const {
369    DCHECK(kernel_);
370    return kernel_->ref(field);
371  }
372  inline bool Get(IndexedBitField field) const {
373    DCHECK(kernel_);
374    return kernel_->ref(field);
375  }
376  inline bool Get(IsDelField field) const {
377    DCHECK(kernel_);
378    return kernel_->ref(field);
379  }
380  inline bool Get(BitField field) const {
381    DCHECK(kernel_);
382    return kernel_->ref(field);
383  }
384  const std::string& Get(StringField field) const;
385  inline const sync_pb::EntitySpecifics& Get(ProtoField field) const {
386    DCHECK(kernel_);
387    return kernel_->ref(field);
388  }
389  inline bool Get(BitTemp field) const {
390    DCHECK(kernel_);
391    return kernel_->ref(field);
392  }
393
394  ModelType GetServerModelType() const;
395  ModelType GetModelType() const;
396
397  // If this returns false, we shouldn't bother maintaining
398  // a position value (sibling ordering) for this item.
399  bool ShouldMaintainPosition() const {
400    return GetModelType() == BOOKMARKS;
401  }
402
403  inline bool ExistsOnClientBecauseNameIsNonEmpty() const {
404    DCHECK(kernel_);
405    return !kernel_->ref(NON_UNIQUE_NAME).empty();
406  }
407
408  inline bool IsRoot() const {
409    DCHECK(kernel_);
410    return kernel_->ref(ID).IsRoot();
411  }
412
413  Directory* dir() const;
414
415  const EntryKernel GetKernelCopy() const {
416    return *kernel_;
417  }
418
419  // Compute a local predecessor position for |update_item|, based on its
420  // absolute server position.  The returned ID will be a valid predecessor
421  // under SERVER_PARENT_ID that is consistent with the
422  // SERVER_POSITION_IN_PARENT ordering.
423  Id ComputePrevIdFromServerPosition(const Id& parent_id) const;
424
425 protected:  // Don't allow creation on heap, except by sync API wrappers.
426  friend class sync_api::ReadNode;
427  void* operator new(size_t size) { return (::operator new)(size); }
428
429  inline Entry(BaseTransaction* trans)
430      : basetrans_(trans),
431        kernel_(NULL) { }
432
433 protected:
434
435  BaseTransaction* const basetrans_;
436
437  EntryKernel* kernel_;
438
439  DISALLOW_COPY_AND_ASSIGN(Entry);
440};
441
442// A mutable meta entry.  Changes get committed to the database when the
443// WriteTransaction is destroyed.
444class MutableEntry : public Entry {
445  friend class WriteTransaction;
446  friend class Directory;
447  void Init(WriteTransaction* trans, const Id& parent_id,
448      const std::string& name);
449 public:
450  MutableEntry(WriteTransaction* trans, Create, const Id& parent_id,
451               const std::string& name);
452  MutableEntry(WriteTransaction* trans, CreateNewUpdateItem, const Id& id);
453  MutableEntry(WriteTransaction* trans, GetByHandle, int64);
454  MutableEntry(WriteTransaction* trans, GetById, const Id&);
455  MutableEntry(WriteTransaction* trans, GetByClientTag, const std::string& tag);
456  MutableEntry(WriteTransaction* trans, GetByServerTag, const std::string& tag);
457
458  inline WriteTransaction* write_transaction() const {
459    return write_transaction_;
460  }
461
462  // Field Accessors.  Some of them trigger the re-indexing of the entry.
463  // Return true on success, return false on failure, which means
464  // that putting the value would have caused a duplicate in the index.
465  // TODO(chron): Remove some of these unecessary return values.
466  bool Put(Int64Field field, const int64& value);
467  bool Put(IdField field, const Id& value);
468
469  // Do a simple property-only update if the PARENT_ID field.  Use with caution.
470  //
471  // The normal Put(IS_PARENT) call will move the item to the front of the
472  // sibling order to maintain the linked list invariants when the parent
473  // changes.  That's usually what you want to do, but it's inappropriate
474  // when the caller is trying to change the parent ID of a the whole set
475  // of children (e.g. because the ID changed during a commit).  For those
476  // cases, there's this function.  It will corrupt the sibling ordering
477  // if you're not careful.
478  void PutParentIdPropertyOnly(const Id& parent_id);
479
480  bool Put(StringField field, const std::string& value);
481  bool Put(BaseVersion field, int64 value);
482
483  bool Put(ProtoField field, const sync_pb::EntitySpecifics& value);
484  inline bool Put(BitField field, bool value) {
485    return PutField(field, value);
486  }
487  inline bool Put(IsDelField field, bool value) {
488    return PutIsDel(value);
489  }
490  bool Put(IndexedBitField field, bool value);
491
492  // Sets the position of this item, and updates the entry kernels of the
493  // adjacent siblings so that list invariants are maintained.  Returns false
494  // and fails if |predecessor_id| does not identify a sibling.  Pass the root
495  // ID to put the node in first position.
496  bool PutPredecessor(const Id& predecessor_id);
497
498  inline bool Put(BitTemp field, bool value) {
499    return PutTemp(field, value);
500  }
501
502 protected:
503  syncable::MetahandleSet* GetDirtyIndexHelper();
504
505  template <typename FieldType, typename ValueType>
506  inline bool PutField(FieldType field, const ValueType& value) {
507    DCHECK(kernel_);
508    if (kernel_->ref(field) != value) {
509      kernel_->put(field, value);
510      kernel_->mark_dirty(GetDirtyIndexHelper());
511    }
512    return true;
513  }
514
515  template <typename TempType, typename ValueType>
516  inline bool PutTemp(TempType field, const ValueType& value) {
517    DCHECK(kernel_);
518    kernel_->put(field, value);
519    return true;
520  }
521
522  bool PutIsDel(bool value);
523
524 private:  // Don't allow creation on heap, except by sync API wrappers.
525  friend class sync_api::WriteNode;
526  void* operator new(size_t size) { return (::operator new)(size); }
527
528  bool PutImpl(StringField field, const std::string& value);
529  bool PutUniqueClientTag(const std::string& value);
530
531  // Adjusts the successor and predecessor entries so that they no longer
532  // refer to this entry.
533  void UnlinkFromOrder();
534
535  // Kind of redundant. We should reduce the number of pointers
536  // floating around if at all possible. Could we store this in Directory?
537  // Scope: Set on construction, never changed after that.
538  WriteTransaction* const write_transaction_;
539
540 protected:
541  MutableEntry();
542
543  DISALLOW_COPY_AND_ASSIGN(MutableEntry);
544};
545
546class LessParentIdAndHandle;
547template <typename FieldType, FieldType field_index>
548class LessField;
549class LessEntryMetaHandles {
550 public:
551  inline bool operator()(const syncable::EntryKernel& a,
552                         const syncable::EntryKernel& b) const {
553    return a.ref(META_HANDLE) < b.ref(META_HANDLE);
554  }
555};
556typedef std::set<EntryKernel, LessEntryMetaHandles> OriginalEntries;
557
558// How syncable indices & Indexers work.
559//
560// The syncable Directory maintains several indices on the Entries it tracks.
561// The indices follow a common pattern:
562//   (a) The index allows efficient lookup of an Entry* with particular
563//       field values.  This is done by use of a std::set<> and a custom
564//       comparator.
565//   (b) There may be conditions for inclusion in the index -- for example,
566//       deleted items might not be indexed.
567//   (c) Because the index set contains only Entry*, one must be careful
568//       to remove Entries from the set before updating the value of
569//       an indexed field.
570// The traits of an index are a Comparator (to define the set ordering) and a
571// ShouldInclude function (to define the conditions for inclusion).  For each
572// index, the traits are grouped into a class called an Indexer which
573// can be used as a template type parameter.
574
575// Traits type for metahandle index.
576struct MetahandleIndexer {
577  // This index is of the metahandle field values.
578  typedef LessField<MetahandleField, META_HANDLE> Comparator;
579
580  // This index includes all entries.
581  inline static bool ShouldInclude(const EntryKernel* a) {
582    return true;
583  }
584};
585
586// Traits type for ID field index.
587struct IdIndexer {
588  // This index is of the ID field values.
589  typedef LessField<IdField, ID> Comparator;
590
591  // This index includes all entries.
592  inline static bool ShouldInclude(const EntryKernel* a) {
593    return true;
594  }
595};
596
597// Traits type for unique client tag index.
598struct ClientTagIndexer {
599  // This index is of the client-tag values.
600  typedef LessField<StringField, UNIQUE_CLIENT_TAG> Comparator;
601
602  // Items are only in this index if they have a non-empty client tag value.
603  static bool ShouldInclude(const EntryKernel* a);
604};
605
606// This index contains EntryKernels ordered by parent ID and metahandle.
607// It allows efficient lookup of the children of a given parent.
608struct ParentIdAndHandleIndexer {
609  // This index is of the parent ID and metahandle.  We use a custom
610  // comparator.
611  class Comparator {
612   public:
613    bool operator() (const syncable::EntryKernel* a,
614                     const syncable::EntryKernel* b) const;
615  };
616
617  // This index does not include deleted items.
618  static bool ShouldInclude(const EntryKernel* a);
619};
620
621// Given an Indexer providing the semantics of an index, defines the
622// set type used to actually contain the index.
623template <typename Indexer>
624struct Index {
625  typedef std::set<EntryKernel*, typename Indexer::Comparator> Set;
626};
627
628// a WriteTransaction has a writer tag describing which body of code is doing
629// the write. This is defined up here since DirectoryChangeEvent also contains
630// one.
631enum WriterTag {
632  INVALID,
633  SYNCER,
634  AUTHWATCHER,
635  UNITTEST,
636  VACUUM_AFTER_SAVE,
637  PURGE_ENTRIES,
638  SYNCAPI
639};
640
641// A separate Event type and channel for very frequent changes, caused
642// by anything, not just the user.
643struct DirectoryChangeEvent {
644  enum {
645    // Means listener should go through list of original entries and
646    // calculate what it needs to notify.  It should *not* call any
647    // callbacks or attempt to lock anything because a
648    // WriteTransaction is being held until the listener returns.
649    CALCULATE_CHANGES,
650    // Means the WriteTransaction is ending, and this is the absolute
651    // last chance to perform any read operations in the current transaction.
652    // It is not recommended that the listener perform any writes.
653    TRANSACTION_ENDING,
654    // Means the WriteTransaction has been released and the listener
655    // can now take action on the changes it calculated.
656    TRANSACTION_COMPLETE,
657    // Channel is closing.
658    SHUTDOWN
659  } todo;
660  // These members are only valid for CALCULATE_CHANGES.
661  const OriginalEntries* originals;
662  BaseTransaction* trans;  // This is valid also for TRANSACTION_ENDING
663  WriterTag writer;
664  typedef DirectoryChangeEvent EventType;
665  static inline bool IsChannelShutdownEvent(const EventType& e) {
666    return SHUTDOWN == e.todo;
667  }
668};
669
670// The name Directory in this case means the entire directory
671// structure within a single user account.
672//
673// Sqlite is a little goofy, in that each thread must access a database
674// via its own handle.  So, a Directory object should only be accessed
675// from a single thread.  Use DirectoryManager's Open() method to
676// always get a directory that has been properly initialized on the
677// current thread.
678//
679// The db is protected against concurrent modification by a reader/
680// writer lock, negotiated by the ReadTransaction and WriteTransaction
681// friend classes.  The in-memory indices are protected against
682// concurrent modification by the kernel lock.
683//
684// All methods which require the reader/writer lock to be held either
685//   are protected and only called from friends in a transaction
686//   or are public and take a Transaction* argument.
687//
688// All methods which require the kernel lock to be already held take a
689// ScopeKernelLock* argument.
690//
691// To prevent deadlock, the reader writer transaction lock must always
692// be held before acquiring the kernel lock.
693class ScopedKernelLock;
694class IdFilter;
695class DirectoryManager;
696
697class Directory {
698  friend class BaseTransaction;
699  friend class Entry;
700  friend class MutableEntry;
701  friend class ReadTransaction;
702  friend class ReadTransactionWithoutDB;
703  friend class ScopedKernelLock;
704  friend class ScopedKernelUnlock;
705  friend class WriteTransaction;
706  friend class SyncableDirectoryTest;
707  FRIEND_TEST_ALL_PREFIXES(SyncableDirectoryTest,
708                           TakeSnapshotGetsAllDirtyHandlesTest);
709  FRIEND_TEST_ALL_PREFIXES(SyncableDirectoryTest,
710                           TakeSnapshotGetsOnlyDirtyHandlesTest);
711  FRIEND_TEST_ALL_PREFIXES(SyncableDirectoryTest, TestPurgeEntriesWithTypeIn);
712  FRIEND_TEST_ALL_PREFIXES(SyncableDirectoryTest,
713                           TakeSnapshotGetsMetahandlesToPurge);
714
715 public:
716  class EventListenerHookup;
717
718  // Various data that the Directory::Kernel we are backing (persisting data
719  // for) needs saved across runs of the application.
720  struct PersistedKernelInfo {
721    AutofillMigrationDebugInfo autofill_migration_debug_info;
722
723    PersistedKernelInfo();
724    ~PersistedKernelInfo();
725
726    // Set the |download_progress| entry for the given model to a
727    // "first sync" start point.  When such a value is sent to the server,
728    // a full download of all objects of the model will be initiated.
729    void reset_download_progress(ModelType model_type);
730
731    // Last sync timestamp fetched from the server.
732    sync_pb::DataTypeProgressMarker download_progress[MODEL_TYPE_COUNT];
733    // true iff we ever reached the end of the changelog.
734    ModelTypeBitSet initial_sync_ended;
735    // The store birthday we were given by the server. Contents are opaque to
736    // the client.
737    std::string store_birthday;
738    // The next local ID that has not been used with this cache-GUID.
739    int64 next_id;
740    // The persisted notification state.
741    std::string notification_state;
742
743    AutofillMigrationState autofill_migration_state;
744  };
745
746  // What the Directory needs on initialization to create itself and its Kernel.
747  // Filled by DirectoryBackingStore::Load.
748  struct KernelLoadInfo {
749    PersistedKernelInfo kernel_info;
750    std::string cache_guid;  // Created on first initialization, never changes.
751    int64 max_metahandle;    // Computed (using sql MAX aggregate) on init.
752    KernelLoadInfo() : max_metahandle(0) {
753    }
754  };
755
756  // The dirty/clean state of kernel fields backed by the share_info table.
757  // This is public so it can be used in SaveChangesSnapshot for persistence.
758  enum KernelShareInfoStatus {
759    KERNEL_SHARE_INFO_INVALID,
760    KERNEL_SHARE_INFO_VALID,
761    KERNEL_SHARE_INFO_DIRTY
762  };
763
764  // When the Directory is told to SaveChanges, a SaveChangesSnapshot is
765  // constructed and forms a consistent snapshot of what needs to be sent to
766  // the backing store.
767  struct SaveChangesSnapshot {
768    SaveChangesSnapshot();
769    ~SaveChangesSnapshot();
770
771    KernelShareInfoStatus kernel_info_status;
772    PersistedKernelInfo kernel_info;
773    OriginalEntries dirty_metas;
774    MetahandleSet metahandles_to_purge;
775  };
776
777  Directory();
778  virtual ~Directory();
779
780  DirOpenResult Open(const FilePath& file_path, const std::string& name);
781
782  void Close();
783
784  int64 NextMetahandle();
785  // Always returns a negative id.  Positive client ids are generated
786  // by the server only.
787  Id NextId();
788
789  const FilePath& file_path() const { return kernel_->db_path; }
790  bool good() const { return NULL != store_; }
791
792  // The download progress is an opaque token provided by the sync server
793  // to indicate the continuation state of the next GetUpdates operation.
794  void GetDownloadProgress(
795      ModelType type,
796      sync_pb::DataTypeProgressMarker* value_out) const;
797  void GetDownloadProgressAsString(
798      ModelType type,
799      std::string* value_out) const;
800  void SetDownloadProgress(
801      ModelType type,
802      const sync_pb::DataTypeProgressMarker& value);
803
804  bool initial_sync_ended_for_type(ModelType type) const;
805  void set_initial_sync_ended_for_type(ModelType type, bool value);
806  AutofillMigrationState get_autofill_migration_state() const;
807
808  AutofillMigrationDebugInfo get_autofill_migration_debug_info() const;
809
810  void set_autofill_migration_state(AutofillMigrationState state);
811
812  void set_autofill_migration_state_debug_info(
813      syncable::AutofillMigrationDebugInfo::PropertyToSet property_to_set,
814      const syncable::AutofillMigrationDebugInfo& info);
815
816  const std::string& name() const { return kernel_->name; }
817
818  // (Account) Store birthday is opaque to the client,
819  // so we keep it in the format it is in the proto buffer
820  // in case we switch to a binary birthday later.
821  std::string store_birthday() const;
822  void set_store_birthday(std::string store_birthday);
823
824  std::string GetAndClearNotificationState();
825  void SetNotificationState(const std::string& notification_state);
826
827  // Unique to each account / client pair.
828  std::string cache_guid() const;
829
830  browser_sync::ChannelHookup<DirectoryChangeEvent>* AddChangeObserver(
831      browser_sync::ChannelEventHandler<DirectoryChangeEvent>* observer);
832
833 protected:  // for friends, mainly used by Entry constructors
834  virtual EntryKernel* GetEntryByHandle(int64 handle);
835  virtual EntryKernel* GetEntryByHandle(int64 metahandle,
836      ScopedKernelLock* lock);
837  virtual EntryKernel* GetEntryById(const Id& id);
838  EntryKernel* GetEntryByServerTag(const std::string& tag);
839  virtual EntryKernel* GetEntryByClientTag(const std::string& tag);
840  EntryKernel* GetRootEntry();
841  bool ReindexId(EntryKernel* const entry, const Id& new_id);
842  void ReindexParentId(EntryKernel* const entry, const Id& new_parent_id);
843  void ClearDirtyMetahandles();
844
845  // These don't do semantic checking.
846  // The semantic checking is implemented higher up.
847  void UnlinkEntryFromOrder(EntryKernel* entry,
848                            WriteTransaction* trans,
849                            ScopedKernelLock* lock);
850
851  // Overridden by tests.
852  virtual DirectoryBackingStore* CreateBackingStore(
853      const std::string& dir_name,
854      const FilePath& backing_filepath);
855
856 private:
857  // These private versions expect the kernel lock to already be held
858  // before calling.
859  EntryKernel* GetEntryById(const Id& id, ScopedKernelLock* const lock);
860
861  DirOpenResult OpenImpl(const FilePath& file_path, const std::string& name);
862
863  template <class T> void TestAndSet(T* kernel_data, const T* data_to_set);
864
865  struct DirectoryEventTraits {
866    typedef DirectoryEvent EventType;
867    static inline bool IsChannelShutdownEvent(const DirectoryEvent& event) {
868      return DIRECTORY_DESTROYED == event;
869    }
870  };
871 public:
872  typedef EventChannel<DirectoryEventTraits, base::Lock> Channel;
873  typedef std::vector<int64> ChildHandles;
874
875  // Returns the child meta handles for given parent id.
876  void GetChildHandles(BaseTransaction*, const Id& parent_id,
877      ChildHandles* result);
878
879  // Find the first or last child in the positional ordering under a parent,
880  // and return its id.  Returns a root Id if parent has no children.
881  virtual Id GetFirstChildId(BaseTransaction* trans, const Id& parent_id);
882  Id GetLastChildId(BaseTransaction* trans, const Id& parent_id);
883
884  // Compute a local predecessor position for |update_item|.  The position
885  // is determined by the SERVER_POSITION_IN_PARENT value of |update_item|,
886  // as well as the SERVER_POSITION_IN_PARENT values of any up-to-date
887  // children of |parent_id|.
888  Id ComputePrevIdFromServerPosition(
889      const EntryKernel* update_item,
890      const syncable::Id& parent_id);
891
892  // SaveChanges works by taking a consistent snapshot of the current Directory
893  // state and indices (by deep copy) under a ReadTransaction, passing this
894  // snapshot to the backing store under no transaction, and finally cleaning
895  // up by either purging entries no longer needed (this part done under a
896  // WriteTransaction) or rolling back the dirty bits.  It also uses
897  // internal locking to enforce SaveChanges operations are mutually exclusive.
898  //
899  // WARNING: THIS METHOD PERFORMS SYNCHRONOUS I/O VIA SQLITE.
900  bool SaveChanges();
901
902  // Returns the number of entities with the unsynced bit set.
903  int64 unsynced_entity_count() const;
904
905  // Get GetUnsyncedMetaHandles should only be called after SaveChanges and
906  // before any new entries have been created. The intention is that the
907  // syncer should call it from its PerformSyncQueries member.
908  typedef std::vector<int64> UnsyncedMetaHandles;
909  void GetUnsyncedMetaHandles(BaseTransaction* trans,
910                              UnsyncedMetaHandles* result);
911
912  // Get all the metahandles for unapplied updates
913  typedef std::vector<int64> UnappliedUpdateMetaHandles;
914  void GetUnappliedUpdateMetaHandles(BaseTransaction* trans,
915                                     UnappliedUpdateMetaHandles* result);
916
917  // Get the channel for post save notification, used by the syncer.
918  inline Channel* channel() const {
919    return kernel_->channel;
920  }
921
922  // Checks tree metadata consistency.
923  // If full_scan is false, the function will avoid pulling any entries from the
924  // db and scan entries currently in ram.
925  // If full_scan is true, all entries will be pulled from the database.
926  // No return value, CHECKs will be triggered if we're given bad
927  // information.
928  void CheckTreeInvariants(syncable::BaseTransaction* trans,
929                           bool full_scan);
930
931  void CheckTreeInvariants(syncable::BaseTransaction* trans,
932                           const OriginalEntries* originals);
933
934  void CheckTreeInvariants(syncable::BaseTransaction* trans,
935                           const MetahandleSet& handles,
936                           const IdFilter& idfilter);
937
938  // Purges all data associated with any entries whose ModelType or
939  // ServerModelType is found in |types|, from _both_ memory and disk.
940  // Only  valid, "real" model types are allowed in |types| (see model_type.h
941  // for definitions).  "Purge" is just meant to distinguish from "deleting"
942  // entries, which means something different in the syncable namespace.
943  // WARNING! This can be real slow, as it iterates over all entries.
944  // WARNING! Performs synchronous I/O.
945  virtual void PurgeEntriesWithTypeIn(const std::set<ModelType>& types);
946
947 private:
948  // Helper to prime ids_index, parent_id_and_names_index, unsynced_metahandles
949  // and unapplied_metahandles from metahandles_index.
950  void InitializeIndices();
951
952  // Constructs a consistent snapshot of the current Directory state and
953  // indices (by deep copy) under a ReadTransaction for use in |snapshot|.
954  // See SaveChanges() for more information.
955  void TakeSnapshotForSaveChanges(SaveChangesSnapshot* snapshot);
956
957  // Purges from memory any unused, safe to remove entries that were
958  // successfully deleted on disk as a result of the SaveChanges that processed
959  // |snapshot|.  See SaveChanges() for more information.
960  void VacuumAfterSaveChanges(const SaveChangesSnapshot& snapshot);
961
962  // Rolls back dirty bits in the event that the SaveChanges that
963  // processed |snapshot| failed, for example, due to no disk space.
964  void HandleSaveChangesFailure(const SaveChangesSnapshot& snapshot);
965
966  // For new entry creation only
967  void InsertEntry(EntryKernel* entry, ScopedKernelLock* lock);
968  void InsertEntry(EntryKernel* entry);
969
970  // Used by CheckTreeInvariants
971  void GetAllMetaHandles(BaseTransaction* trans, MetahandleSet* result);
972  bool SafeToPurgeFromMemory(const EntryKernel* const entry) const;
973
974  // Internal setters that do not acquire a lock internally.  These are unsafe
975  // on their own; caller must guarantee exclusive access manually by holding
976  // a ScopedKernelLock.
977  void set_initial_sync_ended_for_type_unsafe(ModelType type, bool x);
978  void SetNotificationStateUnsafe(const std::string& notification_state);
979
980  Directory& operator = (const Directory&);
981
982 public:
983  typedef Index<MetahandleIndexer>::Set MetahandlesIndex;
984  typedef Index<IdIndexer>::Set IdsIndex;
985  // All entries in memory must be in both the MetahandlesIndex and
986  // the IdsIndex, but only non-deleted entries will be the
987  // ParentIdChildIndex.
988  typedef Index<ParentIdAndHandleIndexer>::Set ParentIdChildIndex;
989
990  // Contains both deleted and existing entries with tags.
991  // We can't store only existing tags because the client would create
992  // items that had a duplicated ID in the end, resulting in a DB key
993  // violation. ID reassociation would fail after an attempted commit.
994  typedef Index<ClientTagIndexer>::Set ClientTagIndex;
995
996 protected:
997  // Used by tests.
998  void init_kernel(const std::string& name);
999
1000 private:
1001
1002  struct Kernel {
1003    Kernel(const FilePath& db_path, const std::string& name,
1004           const KernelLoadInfo& info);
1005
1006    ~Kernel();
1007
1008    void AddRef();  // For convenience.
1009    void Release();
1010
1011    FilePath const db_path;
1012    // TODO(timsteele): audit use of the member and remove if possible
1013    volatile base::subtle::AtomicWord refcount;
1014
1015    // Implements ReadTransaction / WriteTransaction using a simple lock.
1016    base::Lock transaction_mutex;
1017
1018    // The name of this directory.
1019    std::string const name;
1020
1021    // Protects all members below.
1022    // The mutex effectively protects all the indices, but not the
1023    // entries themselves.  So once a pointer to an entry is pulled
1024    // from the index, the mutex can be unlocked and entry read or written.
1025    //
1026    // Never hold the mutex and do anything with the database or any
1027    // other buffered IO.  Violating this rule will result in deadlock.
1028    base::Lock mutex;
1029    // Entries indexed by metahandle
1030    MetahandlesIndex* metahandles_index;
1031    // Entries indexed by id
1032    IdsIndex* ids_index;
1033    ParentIdChildIndex* parent_id_child_index;
1034    ClientTagIndex* client_tag_index;
1035    // So we don't have to create an EntryKernel every time we want to
1036    // look something up in an index.  Needle in haystack metaphor.
1037    EntryKernel needle;
1038
1039    // 3 in-memory indices on bits used extremely frequently by the syncer.
1040    MetahandleSet* const unapplied_update_metahandles;
1041    MetahandleSet* const unsynced_metahandles;
1042    // Contains metahandles that are most likely dirty (though not
1043    // necessarily).  Dirtyness is confirmed in TakeSnapshotForSaveChanges().
1044    MetahandleSet* const dirty_metahandles;
1045
1046    // When a purge takes place, we remove items from all our indices and stash
1047    // them in here so that SaveChanges can persist their permanent deletion.
1048    MetahandleSet* const metahandles_to_purge;
1049
1050    // TODO(ncarter): Figure out what the hell this is, and comment it.
1051    Channel* const channel;
1052
1053    // The changes channel mutex is explicit because it must be locked
1054    // while holding the transaction mutex and released after
1055    // releasing the transaction mutex.
1056    browser_sync::Channel<DirectoryChangeEvent> changes_channel;
1057
1058    base::Lock changes_channel_mutex;
1059    KernelShareInfoStatus info_status;
1060
1061    // These 3 members are backed in the share_info table, and
1062    // their state is marked by the flag above.
1063
1064    // A structure containing the Directory state that is written back into the
1065    // database on SaveChanges.
1066    PersistedKernelInfo persisted_info;
1067
1068    // A unique identifier for this account's cache db, used to generate
1069    // unique server IDs. No need to lock, only written at init time.
1070    std::string cache_guid;
1071
1072    // It doesn't make sense for two threads to run SaveChanges at the same
1073    // time; this mutex protects that activity.
1074    base::Lock save_changes_mutex;
1075
1076    // The next metahandle is protected by kernel mutex.
1077    int64 next_metahandle;
1078
1079    // Keep a history of recently flushed metahandles for debugging
1080    // purposes.  Protected by the save_changes_mutex.
1081    DebugQueue<int64, 1000> flushed_metahandles;
1082  };
1083
1084  // Helper method used to do searches on |parent_id_child_index|.
1085  ParentIdChildIndex::iterator LocateInParentChildIndex(
1086      const ScopedKernelLock& lock,
1087      const Id& parent_id,
1088      int64 position_in_parent,
1089      const Id& item_id_for_tiebreaking);
1090
1091  // Return an iterator to the beginning of the range of the children of
1092  // |parent_id| in the kernel's parent_id_child_index.
1093  ParentIdChildIndex::iterator GetParentChildIndexLowerBound(
1094      const ScopedKernelLock& lock,
1095      const Id& parent_id);
1096
1097  // Return an iterator to just past the end of the range of the
1098  // children of |parent_id| in the kernel's parent_id_child_index.
1099  ParentIdChildIndex::iterator GetParentChildIndexUpperBound(
1100      const ScopedKernelLock& lock,
1101      const Id& parent_id);
1102
1103  Kernel* kernel_;
1104
1105  DirectoryBackingStore* store_;
1106};
1107
1108class ScopedKernelLock {
1109 public:
1110  explicit ScopedKernelLock(const Directory*);
1111  ~ScopedKernelLock() {}
1112
1113  base::AutoLock scoped_lock_;
1114  Directory* const dir_;
1115  DISALLOW_COPY_AND_ASSIGN(ScopedKernelLock);
1116};
1117
1118// Transactions are now processed FIFO with a straight lock
1119class BaseTransaction {
1120  friend class Entry;
1121 public:
1122  inline Directory* directory() const { return directory_; }
1123  inline Id root_id() const { return Id(); }
1124
1125  virtual ~BaseTransaction();
1126
1127 protected:
1128  BaseTransaction(Directory* directory, const char* name,
1129                  const char* source_file, int line, WriterTag writer);
1130
1131  // For unit testing. Everything will be mocked out no point initializing.
1132  explicit BaseTransaction(Directory* directory);
1133
1134  void UnlockAndLog(OriginalEntries* entries);
1135  bool NotifyTransactionChangingAndEnding(OriginalEntries* entries);
1136  virtual void NotifyTransactionComplete();
1137
1138  Directory* const directory_;
1139  Directory::Kernel* const dirkernel_;  // for brevity
1140  const char* const name_;
1141  base::TimeTicks time_acquired_;
1142  const char* const source_file_;
1143  const int line_;
1144  WriterTag writer_;
1145
1146 private:
1147  void Lock();
1148
1149  DISALLOW_COPY_AND_ASSIGN(BaseTransaction);
1150};
1151
1152// Locks db in constructor, unlocks in destructor.
1153class ReadTransaction : public BaseTransaction {
1154 public:
1155  ReadTransaction(Directory* directory, const char* source_file,
1156                  int line);
1157  ReadTransaction(const ScopedDirLookup& scoped_dir,
1158                  const char* source_file, int line);
1159
1160  virtual ~ReadTransaction();
1161
1162 protected:  // Don't allow creation on heap, except by sync API wrapper.
1163  friend class sync_api::ReadTransaction;
1164  void* operator new(size_t size) { return (::operator new)(size); }
1165
1166  DISALLOW_COPY_AND_ASSIGN(ReadTransaction);
1167};
1168
1169// Locks db in constructor, unlocks in destructor.
1170class WriteTransaction : public BaseTransaction {
1171  friend class MutableEntry;
1172 public:
1173  explicit WriteTransaction(Directory* directory, WriterTag writer,
1174                            const char* source_file, int line);
1175  explicit WriteTransaction(const ScopedDirLookup& directory,
1176                            WriterTag writer, const char* source_file,
1177                            int line);
1178  virtual ~WriteTransaction();
1179
1180  void SaveOriginal(EntryKernel* entry);
1181
1182 protected:
1183  // Before an entry gets modified, we copy the original into a list
1184  // so that we can issue change notifications when the transaction
1185  // is done.
1186  OriginalEntries* const originals_;
1187
1188  explicit WriteTransaction(Directory *directory);
1189
1190  DISALLOW_COPY_AND_ASSIGN(WriteTransaction);
1191};
1192
1193bool IsLegalNewParent(BaseTransaction* trans, const Id& id, const Id& parentid);
1194
1195int64 Now();
1196
1197// This function sets only the flags needed to get this entry to sync.
1198void MarkForSyncing(syncable::MutableEntry* e);
1199
1200// This is not a reset.  It just sets the numeric fields which are not
1201// initialized by the constructor to zero.
1202void ZeroFields(EntryKernel* entry, int first_field);
1203
1204}  // namespace syncable
1205
1206std::ostream& operator <<(std::ostream&, const syncable::Blob&);
1207
1208#endif  // CHROME_BROWSER_SYNC_SYNCABLE_SYNCABLE_H_
1209