Load the table of visited links from database file asynchronously.

components/visitedlink/browser/visitedlink_master.h

 // Copyright (c) 2012 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.
 
 #ifndef COMPONENTS_VISITEDLINK_BROWSER_VISITEDLINK_MASTER_H_
 #define COMPONENTS_VISITEDLINK_BROWSER_VISITEDLINK_MASTER_H_
 
 #if defined(OS_WIN)
 #include <windows.h>
 #endif
 #include <set>
 #include <vector>
 
 #include "base/callback.h"
 #include "base/callback_forward.h"
 #include "base/files/file_path.h"
 #include "base/gtest_prod_util.h"
+#include "base/memory/ref_counted.h"
 #include "base/memory/shared_memory.h"
+#include "base/memory/weak_ptr.h"
 #include "base/threading/sequenced_worker_pool.h"
 #include "components/visitedlink/common/visitedlink_common.h"
 
 #if defined(UNIT_TEST) || defined(PERF_TEST) || !defined(NDEBUG)
 #include "base/logging.h"
 #endif
 
 class GURL;
 
 namespace content {
@@ skipping 20 matching lines @@
 
     // Called when link coloring database has been created or replaced. The
     // argument is the new table handle.
     virtual void NewTable(base::SharedMemory*) = 0;
 
     // Called when new link has been added. The argument is the fingerprint
     // (hash) of the link.
     virtual void Add(Fingerprint fingerprint) = 0;
 
     // Called when link coloring state has been reset. This may occur when
-    // entire or parts of history were deleted.
-    virtual void Reset() = 0;
+    // entire or parts of history were deleted. Also this may occur when
+    // the table was rebuilt or loaded. The salt is stored in the database file.
+    // As a result the salt will change after loading the table from the
+    // database file. In this case we use |invalidate_hashes| to inform that
+    // all cached visitedlink hashes need to be recalculated.
+    virtual void Reset(bool invalidate_hashes) = 0;
   };
 
   VisitedLinkMaster(content::BrowserContext* browser_context,
                     VisitedLinkDelegate* delegate,
                     bool persist_to_disk);
 
   // In unit test mode, we allow the caller to optionally specify the database
   // filename so that it can be run from a unit test. The directory where this
   // file resides must exist in this mode. You can also specify the default
   // table size to test table resizing. If this parameter is 0, we will use the
@@ skipping 20 matching lines @@
   // until this is called. Returns true on success, false means that this
   // object won't work.
   bool Init();
 
   base::SharedMemory* shared_memory() { return shared_memory_; }
 
   // Adds a URL to the table.
   void AddURL(const GURL& url);
 
   // Adds a set of URLs to the table.
-  void AddURLs(const std::vector<GURL>& url);
+  void AddURLs(const std::vector<GURL>& urls);
 
   // See DeleteURLs.
   class URLIterator {
    public:
     // HasNextURL must return true when this is called. Returns the next URL
     // then advances the iterator. Note that the returned reference is only
     // valid until the next call of NextURL.
     virtual const GURL& NextURL() = 0;
 
     // Returns true if still has URLs to be iterated.
@@ skipping 41 matching lines @@
   void RewriteFile() {
     WriteFullTable();
   }
 #endif
 
  private:
   FRIEND_TEST_ALL_PREFIXES(VisitedLinkTest, Delete);
   FRIEND_TEST_ALL_PREFIXES(VisitedLinkTest, BigDelete);
   FRIEND_TEST_ALL_PREFIXES(VisitedLinkTest, BigImport);
 
+  // Keeps the result of loading the table from the database file to the UI
+  // thread.
+  struct LoadFromFileResult;
+
+  using TableLoadCompleteCallback = base::Callback<void(
+      bool success,
+      scoped_refptr<LoadFromFileResult> load_from_file_result)>;
+
   // Object to rebuild the table on the history thread (see the .cc file).
   class TableBuilder;
 
   // Byte offsets of values in the header.
   static const int32 kFileHeaderSignatureOffset;
   static const int32 kFileHeaderVersionOffset;
   static const int32 kFileHeaderLengthOffset;
   static const int32 kFileHeaderUsedOffset;
   static const int32 kFileHeaderSaltOffset;
 
@@ skipping 27 matching lines @@
   // These functions are only called if |persist_to_disk_| is true.
 
   // Posts the given task to the blocking worker pool with our options.
   void PostIOTask(const tracked_objects::Location& from_here,
                   const base::Closure& task);
 
   // Writes the entire table to disk. It will leave the table file open and
   // the handle to it will be stored in file_.
   void WriteFullTable();
 
-  // Try to load the table from the database file. If the file doesn't exist or
-  // is corrupt, this will return failure.
+  // Tries to load asynchronously the table from the database file.
   bool InitFromFile();
 
+  // Load the table from the database file. Calls |callback| when completed. It
+  // is called from the background thread. It must be first in the sequence of
+  // background operations with the database file.
+  static void LoadFromFile(const base::FilePath& filename,
+                           const TableLoadCompleteCallback& callback);
+
+  // Load the table from the database file. Returns true on success.
+  // Fills parameter |load_from_file_result| on success. It is called from
+  // the background thread.
+  static bool LoadApartFromFile(
+      const base::FilePath& filename,
+      scoped_refptr<LoadFromFileResult>* load_from_file_result);
+
+  // It is called from the background thread and executed on the UI
+  // thread.
+  void OnTableLoadComplete(
+      bool success,
+      scoped_refptr<LoadFromFileResult> load_from_file_result);
+
   // Reads the header of the link coloring database from disk. Assumes the
-  // file pointer is at the beginning of the file and that there are no pending
-  // asynchronous I/O operations.
+  // file pointer is at the beginning of the file and that it is the first
+  // asynchronous I/O operation on the background thread.
   //
   // Returns true on success and places the size of the table in num_entries
   // and the number of nonzero fingerprints in used_count. This will fail if
   // the version of the file is not the current version of the database.
-  bool ReadFileHeader(FILE* hfile, int32* num_entries, int32* used_count,
-                      uint8 salt[LINK_SALT_LENGTH]);
+  static bool ReadFileHeader(FILE* hfile,
+                             int32* num_entries,
+                             int32* used_count,
+                             uint8 salt[LINK_SALT_LENGTH]);
 
   // Fills *filename with the name of the link database filename
   bool GetDatabaseFileName(base::FilePath* filename);
 
   // Wrapper around Window's WriteFile using asynchronous I/O. This will proxy
   // the write to a background thread.
   void WriteToFile(FILE** hfile, off_t offset, void* data, int32 data_size);
 
   // Helper function to schedule and asynchronous write of the used count to
   // disk (this is a common operation).
   void WriteUsedItemCountToFile();
 
   // Helper function to schedule an asynchronous write of the given range of
   // hash functions to disk. The range is inclusive on both ends. The range can
   // wrap around at 0 and this function will handle it.
   void WriteHashRangeToFile(Hash first_hash, Hash last_hash);
 
-  // Synchronous read from the file. Assumes there are no pending asynchronous
-  // I/O functions. Returns true if the entire buffer was successfully filled.
-  bool ReadFromFile(FILE* hfile, off_t offset, void* data, size_t data_size);
+  // Synchronous read from the file. Assumes that it is the first asynchronous
+  // I/O operation in the background thread. Returns true if the entire buffer
+  // was successfully filled.
+  static bool ReadFromFile(FILE* hfile,
+                           off_t offset,
+                           void* data,
+                           size_t data_size);
 
   // General table handling
   // ----------------------
 
   // Called to add a fingerprint to the table. If |send_notifications| is true
   // and the item is added successfully, Listener::Add will be invoked.
   // Returns the index of the inserted fingerprint or null_hash_ if there was a
   // duplicate and this item was skippped.
   Hash AddFingerprint(Fingerprint fingerprint, bool send_notifications);
 
   // Deletes all fingerprints from the given vector from the current hash table
   // and syncs it to disk if there are changes. This does not update the
   // deleted_since_rebuild_ list, the caller must update this itself if there
   // is an update pending.
   void DeleteFingerprintsFromCurrentTable(
       const std::set<Fingerprint>& fingerprints);
 
   // Removes the indicated fingerprint from the table. If the update_file flag
   // is set, the changes will also be written to disk. Returns true if the
   // fingerprint was deleted, false if it was not in the table to delete.
   bool DeleteFingerprint(Fingerprint fingerprint, bool update_file);
 
   // Creates a new empty table, call if InitFromFile() fails. Normally, when
   // |suppress_rebuild| is false, the table will be rebuilt from history,
   // keeping us in sync. When |suppress_rebuild| is true, the new table will be
   // empty and we will not consult history. This is used when clearing the
   // database and for unit tests.
   bool InitFromScratch(bool suppress_rebuild);
 
-  // Allocates the Fingerprint structure and length. When init_to_empty is set,
-  // the table will be filled with 0s and used_items_ will be set to 0 as well.
-  // If the flag is not set, these things are untouched and it is the
-  // responsibility of the caller to fill them (like when we are reading from
-  // a file).
-  bool CreateURLTable(int32 num_entries, bool init_to_empty);
+  // Allocates the Fingerprint structure and length. Structure is filled with 0s
+  // and shared header with salt and used_items_ is set to 0.
+  bool CreateURLTable(int32 num_entries);
+
+  // Allocates the Fingerprint structure and length. Returns true on success.
+  // Structure is filled with 0s and shared header with salt. The result of
+  // allocation is saved into |shared_memory| and |hash_table| points to the
+  // beginning of Fingerprint table in |shared_memory|.
+  static bool CreateApartURLTable(int32 num_entries,
+                                  const uint8 salt[LINK_SALT_LENGTH],
+                                  scoped_ptr<base::SharedMemory>* shared_memory,
+                                  VisitedLinkCommon::Fingerprint** hash_table);
 
   // A wrapper for CreateURLTable, this will allocate a new table, initialized
   // to empty. The caller is responsible for saving the shared memory pointer
   // and handles before this call (they will be replaced with new ones) and
   // releasing them later. This is designed for callers that make a new table
   // and then copy values from the old table to the new one, then release the
   // old table.
   //
   // Returns true on success. On failure, the old table will be restored. The
   // caller should not attemp to release the pointer/handle in this case.
   bool BeginReplaceURLTable(int32 num_entries);
 
   // unallocates the Fingerprint table
   void FreeURLTable();
 
   // For growing the table. ResizeTableIfNecessary will check to see if the
   // table should be resized and calls ResizeTable if needed. Returns true if
   // we decided to resize the table.
   bool ResizeTableIfNecessary();
 
   // Resizes the table (growing or shrinking) as necessary to accomodate the
   // current count.
   void ResizeTable(int32 new_size);
 
+  // Returns the default table size. It can be overrided in unit tests.
+  uint32 DefaultTableSize() const;
+
   // Returns the desired table size for |item_count| URLs.
   uint32 NewTableSizeForCount(int32 item_count) const;
 
   // Computes the table load as fraction. For example, if 1/4 of the entries are
   // full, this value will be 0.25
   float ComputeTableLoad() const {
     return static_cast<float>(used_items_) / static_cast<float>(table_length_);
   }
 
   // Initializes a rebuild of the visited link database based on the browser
@@ skipping 18 matching lines @@
     if (hash >= table_length_ - 1)
       return 0;  // Wrap around.
     return hash + 1;
   }
   inline Hash DecrementHash(Hash hash) {
     if (hash <= 0)
       return table_length_ - 1;  // Wrap around.
     return hash - 1;
   }
 
-#ifndef NDEBUG
-  // Indicates whether any asynchronous operation has ever been completed.
-  // We do some synchronous reads that require that no asynchronous operations
-  // are pending, yet we don't track whether they have been completed. This
-  // flag is a sanity check that these reads only happen before any
-  // asynchronous writes have been fired.
-  bool posted_asynchronous_operation_;
-#endif
-
   // Reference to the browser context that this object belongs to
   // (it knows the path to where the data is stored)
   content::BrowserContext* browser_context_;
 
   // Client owns the delegate and is responsible for it being valid through
   // the life time this VisitedLinkMaster.
   VisitedLinkDelegate* delegate_;
 
   // VisitedLinkEventListener to handle incoming events.
   scoped_ptr<Listener> listener_;
 
   // Lazily initialized sequence token for posting file tasks.
   base::SequencedWorkerPool::SequenceToken sequence_token_;
 
   // When non-NULL, indicates we are in database rebuild mode and points to
   // the class collecting fingerprint information from the history system.
   // The pointer is owned by this class, but it must remain valid while the
   // history query is running. We must only delete it when the query is done.
   scoped_refptr<TableBuilder> table_builder_;
 
   // Indicates URLs added and deleted since we started rebuilding the table.
   std::set<Fingerprint> added_since_rebuild_;
   std::set<Fingerprint> deleted_since_rebuild_;
 
-  // TODO(brettw) Support deletion, we need to track whether anything was
-  // deleted during the rebuild here. Then we should delete any of these
-  // entries from the complete table later.
-  // std::vector<Fingerprint> removed_since_rebuild_;
+  // Indicates URLs added and deleted since we started loading the table.
+  // It can be only url because after loading table the salt will be changed.
+  std::set<GURL> added_since_load_;
+  std::set<GURL> deleted_since_load_;
 
   // The currently open file with the table in it. This may be NULL if we're
   // rebuilding and haven't written a new version yet or if |persist_to_disk_|
   // is false. Writing to the file may be safely ignored in this case. Also
   // |file_| may be non-NULL but point to a NULL pointer. That would mean that
   // opening of the file is already scheduled in a background thread and any
   // writing to the file can also be scheduled to the background thread as it's
   // guaranteed to be executed after the opening.
   // The class owns both the |file_| pointer and the pointer pointed
   // by |*file_|.
   FILE** file_;
 
   // If true, will try to persist the hash table to disk. Will rebuild from
   // VisitedLinkDelegate::RebuildTable if there are disk corruptions.
   bool persist_to_disk_;
 
   // Shared memory consists of a SharedHeader followed by the table.
   base::SharedMemory *shared_memory_;
 
   // When we generate new tables, we increment the serial number of the
   // shared memory object.
   int32 shared_memory_serial_;
 
   // Number of non-empty items in the table, used to compute fullness.
   int32 used_items_;
 
+  // We set this to true to avoid writing to the database file.
+  bool table_is_loading_from_file_;
+
   // Testing values -----------------------------------------------------------
   //
   // The following fields exist for testing purposes. They are not used in
   // release builds. It'd be nice to eliminate them in release builds, but we
   // don't want to change the signature of the object between the unit test and
   // regular builds. Instead, we just have "default" values that these take
   // in release builds that give "regular" behavior.
 
   // Overridden database file name for testing
   base::FilePath database_name_override_;
 
   // When nonzero, overrides the table size for new databases for testing
   int32 table_size_override_;
 
   // When set, indicates the task that should be run after the next rebuild from
   // history is complete.
   base::Closure rebuild_complete_task_;
 
   // Set to prevent us from attempting to rebuild the database from global
   // history if we have an error opening the file. This is used for testing,
   // will be false in production.
   bool suppress_rebuild_;
 
+  base::WeakPtrFactory<VisitedLinkMaster> weak_ptr_factory_;
+
   DISALLOW_COPY_AND_ASSIGN(VisitedLinkMaster);
 };
 
 // NOTE: These methods are defined inline here, so we can share the compilation
 //       of visitedlink_master.cc between the browser and the unit/perf tests.
 
 #if defined(UNIT_TEST) || defined(PERF_TEST) || !defined(NDEBUG)
 inline void VisitedLinkMaster::DebugValidate() {
   int32 used_count = 0;
   for (int32 i = 0; i < table_length_; i++) {
     if (hash_table_[i])
       used_count++;
   }
   DCHECK_EQ(used_count, used_items_);
 }
 #endif
 
 }  // namespace visitedlink
 
 #endif  // COMPONENTS_VISITEDLINK_BROWSER_VISITEDLINK_MASTER_H_