Merge bk-internal.mysql.com:/home/bk/mysql-5.1

into serg.mylan:/usr/home/serg/Abk/m51


sql/log.cc:
  Auto merged

mysql-test/r/csv.result

@@ -4944,10 +4944,10 @@ val
 UPDATE  bug13894 SET val=6 WHERE val=10;
 SELECT * FROM bug13894;
 val
-5
-11
 6
 6
+5
+11
 DROP TABLE bug13894;
 DROP TABLE IF EXISTS bug14672;
 CREATE TABLE bug14672 (c1 integer) engine = CSV;

sql/ha_ndbcluster_binlog.cc

@@ -453,7 +453,7 @@ static void ndbcluster_binlog_wait(THD *thd)
 }
 
 /*
- Called from MYSQL_LOG::reset_logs in log.cc when binlog is emptied
+ Called from MYSQL_BIN_LOG::reset_logs in log.cc when binlog is emptied
 */
 static int ndbcluster_reset_logs(THD *thd)
 {
@@ -477,7 +477,7 @@ static int ndbcluster_reset_logs(THD *thd)
 }
 
 /*
-  Called from MYSQL_LOG::purge_logs in log.cc when the binlog "file"
+  Called from MYSQL_BIN_LOG::purge_logs in log.cc when the binlog "file"
   is removed
 */
 

sql/log.cc

@@ -42,7 +42,7 @@ extern char *opt_logname, *opt_slow_logname;
 
 LOGGER logger;
 
-MYSQL_LOG mysql_bin_log;
+MYSQL_BIN_LOG mysql_bin_log;
 ulong sync_binlog_counter= 0;
 
 static bool test_if_number(const char *str,
@@ -86,7 +86,7 @@ handlerton binlog_hton;
   SYNOPSIS
     open_log_table()
 
-    log_type   type of the log table to open: QUERY_LOG_GENERAL
+    log_table_type   type of the log table to open: QUERY_LOG_GENERAL
                      or QUERY_LOG_SLOW
 
   DESCRIPTION
@@ -102,14 +102,14 @@ handlerton binlog_hton;
     TRUE - error occured
 */
 
-bool Log_to_csv_event_handler::open_log_table(uint log_type)
+bool Log_to_csv_event_handler::open_log_table(uint log_table_type)
 {
   THD *log_thd, *curr= current_thd;
   TABLE_LIST *table;
   bool error= FALSE;
   DBUG_ENTER("open_log_table");
 
-  switch (log_type) {
+  switch (log_table_type) {
   case QUERY_LOG_GENERAL:
     log_thd= general_log_thd;
     table= &general_log;
@@ -222,7 +222,7 @@ Log_to_csv_event_handler::~Log_to_csv_event_handler()
   SYNOPSIS
     reopen_log_table()
 
-    log_type   type of the log table to open: QUERY_LOG_GENERAL
+    log_table_type   type of the log table to open: QUERY_LOG_GENERAL
                      or QUERY_LOG_SLOW
 
   DESCRIPTION
@@ -240,12 +240,12 @@ Log_to_csv_event_handler::~Log_to_csv_event_handler()
     TRUE - open_log_table() returned an error
 */
 
-bool Log_to_csv_event_handler::reopen_log_table(uint log_type)
+bool Log_to_csv_event_handler::reopen_log_table(uint log_table_type)
 {
   /* don't open the log table, if it wasn't enabled during startup */
   if (!logger.is_log_tables_initialized)
     return FALSE;
-  return open_log_table(log_type);
+  return open_log_table(log_table_type);
 }
 
 void Log_to_csv_event_handler::cleanup()
@@ -525,9 +525,8 @@ void Log_to_file_event_handler::cleanup()
 
 void Log_to_file_event_handler::flush()
 {
-  /* reopen log files */
-  mysql_log.new_file(1);
-  mysql_slow_log.new_file(1);
+  mysql_log.reopen_file();
+  mysql_slow_log.reopen_file();
 }
 
 /*
@@ -582,9 +581,9 @@ void LOGGER::cleanup_end()
 }
 
 
-void LOGGER::close_log_table(uint log_type, bool lock_in_use)
+void LOGGER::close_log_table(uint log_table_type, bool lock_in_use)
 {
-  table_log_handler->close_log_table(log_type, lock_in_use);
+  table_log_handler->close_log_table(log_table_type, lock_in_use);
 }
 
 
@@ -624,9 +623,9 @@ void LOGGER::init_log_tables()
 }
 
 
-bool LOGGER::reopen_log_table(uint log_type)
+bool LOGGER::reopen_log_table(uint log_table_type)
 {
-  return table_log_handler->reopen_log_table(log_type);
+  return table_log_handler->reopen_log_table(log_table_type);
 }
 
 
@@ -961,7 +960,7 @@ int LOGGER::set_handlers(uint error_log_printer,
   SYNOPSIS
     close_log_table()
 
-    log_type       type of the log table to close: QUERY_LOG_GENERAL
+    log_table_type   type of the log table to close: QUERY_LOG_GENERAL
                      or QUERY_LOG_SLOW
     lock_in_use      Set to TRUE if the caller owns LOCK_open. FALSE otherwise.
 
@@ -973,7 +972,7 @@ int LOGGER::set_handlers(uint error_log_printer,
 */
 
 void Log_to_csv_event_handler::
-  close_log_table(uint log_type, bool lock_in_use)
+  close_log_table(uint log_table_type, bool lock_in_use)
 {
   THD *log_thd, *curr= current_thd;
   TABLE_LIST *table;
@@ -981,7 +980,7 @@ void Log_to_csv_event_handler::
   if (!logger.is_log_tables_initialized)
     return;                                     /* do nothing */
 
-  switch (log_type) {
+  switch (log_table_type) {
   case QUERY_LOG_GENERAL:
     log_thd= general_log_thd;
     table= &general_log;
@@ -1022,7 +1021,7 @@ void Log_to_csv_event_handler::
 
 /*
   this function is mostly a placeholder.
-  conceptually, binlog initialization (now mostly done in MYSQL_LOG::open)
+  conceptually, binlog initialization (now mostly done in MYSQL_BIN_LOG::open)
   should be moved here.
 */
 
@@ -1111,7 +1110,7 @@ static int binlog_prepare(THD *thd, bool all)
     do nothing.
     just pretend we can do 2pc, so that MySQL won't
     switch to 1pc.
-    real work will be done in MYSQL_LOG::log()
+    real work will be done in MYSQL_BIN_LOG::log()
   */
   return 0;
 }
@@ -1127,7 +1126,7 @@ static int binlog_commit(THD *thd, bool all)
 
   if (trx_data->empty())
   {
-    // we're here because trans_log was flushed in MYSQL_LOG::log()
+    // we're here because trans_log was flushed in MYSQL_BIN_LOG::log()
     DBUG_RETURN(0);
   }
   Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), TRUE, FALSE);
@@ -1367,12 +1366,119 @@ static int find_uniq_filename(char *name)
 }
 
 
+void MYSQL_LOG::init(enum_log_type log_type_arg,
+                     enum cache_type io_cache_type_arg)
+{
+  DBUG_ENTER("MYSQL_LOG::init");
+  log_type= log_type_arg;
+  io_cache_type= io_cache_type_arg;
+  DBUG_PRINT("info",("log_type: %d", log_type));
+  DBUG_VOID_RETURN;
+}
+
+
+/*
+  Open a (new) log file.
+
+  SYNOPSIS
+    open()
+
+    log_name            The name of the log to open
+    log_type_arg        The type of the log. E.g. LOG_NORMAL
+    new_name            The new name for the logfile. This is only needed
+                        when the method is used to open the binlog file.
+    io_cache_type_arg   The type of the IO_CACHE to use for this log file
+
+  DESCRIPTION
+    Open the logfile, init IO_CACHE and write startup messages
+    (in case of general and slow query logs).
+
+  RETURN VALUES
+    0   ok
+    1   error
+*/
+
+bool MYSQL_LOG::open(const char *log_name, enum_log_type log_type_arg,
+                     const char *new_name, enum cache_type io_cache_type_arg)
+{
+  char buff[FN_REFLEN];
+  File file= -1;
+  int open_flags= O_CREAT | O_BINARY;
+  DBUG_ENTER("MYSQL_LOG::open");
+  DBUG_PRINT("enter", ("log_type: %d", (int) log_type_arg));
+
+  write_error= 0;
+
+  init(log_type_arg, io_cache_type_arg);
+
+  if (!(name= my_strdup(log_name, MYF(MY_WME))))
+  {
+    name= (char *)log_name; // for the error message
+    goto err;
+  }
+
+  if (new_name)
+    strmov(log_file_name, new_name);
+  else if (generate_new_name(log_file_name, name))
+    goto err;
+
+  if (io_cache_type == SEQ_READ_APPEND)
+    open_flags |= O_RDWR | O_APPEND;
+  else
+    open_flags |= O_WRONLY | (log_type == LOG_BIN ? 0 : O_APPEND);
+
+  db[0]= 0;
+
+  if ((file= my_open(log_file_name, open_flags,
+                     MYF(MY_WME | ME_WAITTANG))) < 0 ||
+      init_io_cache(&log_file, file, IO_SIZE, io_cache_type,
+                    my_tell(file, MYF(MY_WME)), 0,
+                    MYF(MY_WME | MY_NABP |
+                        ((log_type == LOG_BIN) ? MY_WAIT_IF_FULL : 0))))
+    goto err;
+
+  if (log_type == LOG_NORMAL)
+  {
+    char *end;
+    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s. "
+#ifdef EMBEDDED_LIBRARY
+		        "embedded library\n", my_progname, server_version
+#elif __NT__
+			"started with:\nTCP Port: %d, Named Pipe: %s\n",
+			my_progname, server_version, mysqld_port,
+                        mysqld_unix_port
+#else
+			"started with:\nTcp port: %d  Unix socket: %s\n",
+			my_progname, server_version, mysqld_port,
+                        mysqld_unix_port
+#endif
+                       );
+    end= strnmov(buff + len, "Time                 Id Command    Argument\n",
+                 sizeof(buff) - len);
+    if (my_b_write(&log_file, (byte*) buff, (uint) (end-buff)) ||
+	flush_io_cache(&log_file))
+      goto err;
+  }
+
+  log_state= LOG_OPENED;
+  DBUG_RETURN(0);
+
+err:
+  sql_print_error("Could not use %s for logging (error %d). \
+Turning logging off for the whole duration of the MySQL server process. \
+To turn it on again: fix the cause, \
+shutdown the MySQL server and restart it.", name, errno);
+  if (file >= 0)
+    my_close(file, MYF(0));
+  end_io_cache(&log_file);
+  safeFree(name);
+  log_state= LOG_CLOSED;
+  DBUG_RETURN(1);
+}
+
 MYSQL_LOG::MYSQL_LOG()
-  :bytes_written(0), last_time(0), query_start(0), name(0),
-   prepared_xids(0), log_type(LOG_CLOSED), file_id(1), open_count(1),
-   write_error(FALSE), inited(FALSE), need_start_event(TRUE),
-   m_table_map_version(0),
-   description_event_for_exec(0), description_event_for_queue(0)
+  : name(0),  log_type(LOG_UNKNOWN), log_state(LOG_CLOSED), write_error(FALSE),
+    inited(FALSE)
 {
   /*
     We don't want to initialize LOCK_Log here as such initialization depends on
@@ -1380,9 +1486,54 @@ MYSQL_LOG::MYSQL_LOG()
     called only in main(). Doing initialization here would make it happen
     before main().
   */
-  index_file_name[0] = 0;
-  bzero((char*) &log_file,sizeof(log_file));
-  bzero((char*) &index_file, sizeof(index_file));
+  bzero((char*) &log_file, sizeof(log_file));
+}
+
+void MYSQL_LOG::init_pthread_objects()
+{
+  DBUG_ASSERT(inited == 0);
+  inited= 1;
+  (void) pthread_mutex_init(&LOCK_log, MY_MUTEX_INIT_SLOW);
+}
+
+/*
+  Close the log file
+
+  SYNOPSIS
+    close()
+    exiting     Bitmask. For the slow and general logs the only used bit is
+                LOG_CLOSE_TO_BE_OPENED. This is used if we intend to call
+                open at once after close.
+
+  NOTES
+    One can do an open on the object at once after doing a close.
+    The internal structures are not freed until cleanup() is called
+*/
+
+void MYSQL_LOG::close(uint exiting)
+{					// One can't set log_type here!
+  DBUG_ENTER("MYSQL_LOG::close");
+  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
+  if (log_state == LOG_OPENED)
+  {
+    end_io_cache(&log_file);
+
+    if (my_sync(log_file.file, MYF(MY_WME)) && ! write_error)
+    {
+      write_error= 1;
+      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
+    }
+
+    if (my_close(log_file.file, MYF(MY_WME)) && ! write_error)
+    {
+      write_error= 1;
+      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
+    }
+  }
+
+  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
+  safeFree(name);
+  DBUG_VOID_RETURN;
 }
 
 /* this is called only once */
@@ -1393,12 +1544,8 @@ void MYSQL_LOG::cleanup()
   if (inited)
   {
     inited= 0;
-    close(LOG_CLOSE_INDEX|LOG_CLOSE_STOP_EVENT);
-    delete description_event_for_queue;
-    delete description_event_for_exec;
     (void) pthread_mutex_destroy(&LOCK_log);
-    (void) pthread_mutex_destroy(&LOCK_index);
-    (void) pthread_cond_destroy(&update_cond);
+    close(0);
   }
   DBUG_VOID_RETURN;
 }
@@ -1406,8 +1553,8 @@ void MYSQL_LOG::cleanup()
 
 int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
 {
-  fn_format(new_name,log_name,mysql_data_home,"",4);
-  if (log_type != LOG_NORMAL)
+  fn_format(new_name, log_name, mysql_data_home, "", 4);
+  if (log_type == LOG_BIN)
   {
     if (!fn_ext(log_name)[0])
     {
@@ -1422,30 +1569,284 @@ int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
 }
 
 
-void MYSQL_LOG::init(enum_log_type log_type_arg,
-		     enum cache_type io_cache_type_arg,
-		     bool no_auto_events_arg,
-                     ulong max_size_arg)
+/*
+  Reopen the log file
+
+  SYNOPSIS
+    reopen_file()
+
+  DESCRIPTION
+    Reopen the log file. The method is used during FLUSH LOGS
+    and locks LOCK_log mutex
+*/
+
+
+void MYSQL_QUERY_LOG::reopen_file()
 {
-  DBUG_ENTER("MYSQL_LOG::init");
-  log_type = log_type_arg;
-  io_cache_type = io_cache_type_arg;
-  no_auto_events = no_auto_events_arg;
-  max_size=max_size_arg;
-  DBUG_PRINT("info",("log_type: %d max_size: %lu", log_type, max_size));
+  char *save_name;
+
+  DBUG_ENTER("MYSQL_LOG::reopen_file");
+  if (!is_open())
+  {
+    DBUG_PRINT("info",("log is closed"));
+    DBUG_VOID_RETURN;
+  }
+
+  pthread_mutex_lock(&LOCK_log);
+
+  save_name= name;
+  name= 0;				// Don't free name
+  close(LOG_CLOSE_TO_BE_OPENED);
+
+  /*
+     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
+  */
+
+  open(save_name, log_type, 0, io_cache_type);
+  my_free(save_name, MYF(0));
+
+  pthread_mutex_unlock(&LOCK_log);
+
   DBUG_VOID_RETURN;
 }
 
 
-void MYSQL_LOG::init_pthread_objects()
+/*
+  Write a command to traditional general log file
+
+  SYNOPSIS
+    write()
+
+    event_time        command start timestamp
+    user_host         the pointer to the string with user@host info
+    user_host_len     length of the user_host string. this is computed once
+                      and passed to all general log  event handlers
+    thread_id         Id of the thread, issued a query
+    command_type      the type of the command being logged
+    command_type_len  the length of the string above
+    sql_text          the very text of the query being executed
+    sql_text_len      the length of sql_text string
+
+  DESCRIPTION
+
+   Log given command to to normal (not rotable) log file
+
+  RETURN
+    FASE - OK
+    TRUE - error occured
+*/
+
+bool MYSQL_QUERY_LOG::write(time_t event_time, const char *user_host,
+                            uint user_host_len, int thread_id,
+                            const char *command_type, uint command_type_len,
+                            const char *sql_text, uint sql_text_len)
 {
-  DBUG_ASSERT(inited == 0);
-  inited= 1;
-  (void) pthread_mutex_init(&LOCK_log,MY_MUTEX_INIT_SLOW);
-  (void) pthread_mutex_init(&LOCK_index, MY_MUTEX_INIT_SLOW);
-  (void) pthread_cond_init(&update_cond, 0);
+  char buff[32];
+  uint length= 0;
+  char time_buff[MAX_TIME_SIZE];
+  struct tm start;
+  uint time_buff_len= 0;
+
+  /* Test if someone closed between the is_open test and lock */
+  if (is_open())
+  {
+    /* Note that my_b_write() assumes it knows the length for this */
+      if (event_time != last_time)
+      {
+        last_time= event_time;
+
+        localtime_r(&event_time, &start);
+
+        time_buff_len= my_snprintf(time_buff, MAX_TIME_SIZE,
+                                   "%02d%02d%02d %2d:%02d:%02d",
+                                   start.tm_year % 100, start.tm_mon + 1,
+                                   start.tm_mday, start.tm_hour,
+                                   start.tm_min, start.tm_sec);
+
+        if (my_b_write(&log_file, (byte*) &time_buff, time_buff_len))
+          goto err;
+      }
+      else
+        if (my_b_write(&log_file, (byte*) "\t\t" ,2) < 0)
+          goto err;
+
+    /* command_type, thread_id */
+    length= my_snprintf(buff, 32, "%5ld ", thread_id);
+
+    if (my_b_write(&log_file, (byte*) buff, length))
+      goto err;
+
+    if (my_b_write(&log_file, (byte*) command_type, command_type_len))
+      goto err;
+
+    if (my_b_write(&log_file, (byte*) "\t", 1))
+      goto err;
+
+    /* sql_text */
+    if (my_b_write(&log_file, (byte*) sql_text, sql_text_len))
+      goto err;
+
+    if (my_b_write(&log_file, (byte*) "\n", 1) ||
+        flush_io_cache(&log_file))
+      goto err;
+  }
+
+  return FALSE;
+err:
+
+  if (!write_error)
+  {
+    write_error= 1;
+    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
+  }
+  return TRUE;
+}
+
+
+/*
+  Log a query to the traditional slow log file
+
+  SYNOPSIS
+    write()
+
+    thd               THD of the query
+    current_time      current timestamp
+    query_start_arg   command start timestamp
+    user_host         the pointer to the string with user@host info
+    user_host_len     length of the user_host string. this is computed once
+                      and passed to all general log event handlers
+    query_time        Amount of time the query took to execute (in seconds)
+    lock_time         Amount of time the query was locked (in seconds)
+    is_command        The flag, which determines, whether the sql_text is a
+                      query or an administrator command.
+    sql_text          the very text of the query or administrator command
+                      processed
+    sql_text_len      the length of sql_text string
+
+  DESCRIPTION
+
+   Log a query to the slow log file.
+
+  RETURN
+    FALSE - OK
+    TRUE - error occured
+*/
+
+bool MYSQL_QUERY_LOG::write(THD *thd, time_t current_time,
+                            time_t query_start_arg, const char *user_host,
+                            uint user_host_len, longlong query_time,
+                            longlong lock_time, bool is_command,
+                            const char *sql_text, uint sql_text_len)
+{
+  bool error= 0;
+  DBUG_ENTER("MYSQL_QUERY_LOG::write");
+
+  if (!is_open())
+    DBUG_RETURN(0);
+
+  if (is_open())
+  {						// Safety agains reopen
+    int tmp_errno= 0;
+    char buff[80], *end;
+    uint buff_len;
+    end= buff;
+
+    if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
+    {
+      Security_context *sctx= thd->security_ctx;
+      if (current_time != last_time)
+      {
+        last_time= current_time;
+        struct tm start;
+        localtime_r(&current_time, &start);
+
+        buff_len= my_snprintf(buff, sizeof buff,
+                              "# Time: %02d%02d%02d %2d:%02d:%02d\n",
+                              start.tm_year % 100, start.tm_mon + 1,
+                              start.tm_mday, start.tm_hour,
+                              start.tm_min, start.tm_sec);
+
+        /* Note that my_b_write() assumes it knows the length for this */
+        if (my_b_write(&log_file, (byte*) buff, buff_len))
+          tmp_errno= errno;
+      }
+      if (my_b_printf(&log_file, "# User@Host: ", sizeof("# User@Host: ") - 1)
+          != sizeof("# User@Host: ") - 1)
+        tmp_errno= errno;
+      if (my_b_printf(&log_file, user_host, user_host_len) != user_host_len)
+        tmp_errno= errno;
+      if (my_b_write(&log_file, (byte*) "\n", 1))
+        tmp_errno= errno;
+    }
+    /* For slow query log */
+    if (my_b_printf(&log_file,
+                    "# Query_time: %lu  Lock_time: %lu"
+                    " Rows_sent: %lu  Rows_examined: %lu\n",
+                    (ulong) query_time, (ulong) lock_time,
+                    (ulong) thd->sent_row_count,
+                    (ulong) thd->examined_row_count) == (uint) -1)
+      tmp_errno= errno;
+    if (thd->db && strcmp(thd->db, db))
+    {						// Database changed
+      if (my_b_printf(&log_file,"use %s;\n",thd->db) == (uint) -1)
+        tmp_errno= errno;
+      strmov(db,thd->db);
+    }
+    if (thd->last_insert_id_used)
+    {
+      end=strmov(end, ",last_insert_id=");
+      end=longlong10_to_str((longlong) thd->current_insert_id, end, -10);
+    }
+    // Save value if we do an insert.
+    if (thd->insert_id_used)
+    {
+      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
+      {
+        end=strmov(end,",insert_id=");
+        end=longlong10_to_str((longlong) thd->last_insert_id, end, -10);
+      }
+    }
+
+    /*
+      This info used to show up randomly, depending on whether the query
+      checked the query start time or not. now we always write current
+      timestamp to the slow log
+    */
+    end= strmov(end, ",timestamp=");
+    end= int10_to_str((long) current_time, end, 10);
+
+    if (end != buff)
+    {
+      *end++=';';
+      *end='\n';
+      if (my_b_write(&log_file, (byte*) "SET ", 4) ||
+          my_b_write(&log_file, (byte*) buff + 1, (uint) (end-buff)))
+        tmp_errno= errno;
+    }
+    if (is_command)
+    {
+      end= strxmov(buff, "# administrator command: ", NullS);
+      buff_len= (ulong) (end - buff);
+      my_b_write(&log_file, (byte*) buff, buff_len);
+    }
+    if (my_b_write(&log_file, (byte*) sql_text, sql_text_len) ||
+        my_b_write(&log_file, (byte*) ";\n",2) ||
+        flush_io_cache(&log_file))
+      tmp_errno= errno;
+    if (tmp_errno)
+    {
+      error= 1;
+      if (! write_error)
+      {
+        write_error= 1;
+        sql_print_error(ER(ER_ERROR_ON_WRITE), name, error);
+      }
+    }
+  }
+  DBUG_RETURN(error);
 }
 
+
 const char *MYSQL_LOG::generate_name(const char *log_name,
                                       const char *suffix,
                                       bool strip_ext, char *buff)
@@ -1456,23 +1857,79 @@ const char *MYSQL_LOG::generate_name(const char *log_name,
       TODO: The following should be using fn_format();  We just need to
       first change fn_format() to cut the file name if it's too long.
     */
-    strmake(buff,glob_hostname,FN_REFLEN-5);
-    strmov(fn_ext(buff),suffix);
+    strmake(buff, glob_hostname, FN_REFLEN - 5);
+    strmov(fn_ext(buff), suffix);
     return (const char *)buff;
   }
   // get rid of extension if the log is binary to avoid problems
   if (strip_ext)
   {
-    char *p = fn_ext(log_name);
-    uint length=(uint) (p-log_name);
-    strmake(buff,log_name,min(length,FN_REFLEN));
+    char *p= fn_ext(log_name);
+    uint length= (uint) (p - log_name);
+    strmake(buff, log_name, min(length, FN_REFLEN));
     return (const char*)buff;
   }
   return log_name;
 }
 
 
-bool MYSQL_LOG::open_index_file(const char *index_file_name_arg,
+
+MYSQL_BIN_LOG::MYSQL_BIN_LOG()
+  :bytes_written(0), prepared_xids(0), file_id(1), open_count(1),
+   need_start_event(TRUE), m_table_map_version(0),
+   description_event_for_exec(0), description_event_for_queue(0)
+{
+  /*
+    We don't want to initialize locks here as such initialization depends on
+    safe_mutex (when using safe_mutex) which depends on MY_INIT(), which is
+    called only in main(). Doing initialization here would make it happen
+    before main().
+  */
+  index_file_name[0] = 0;
+  bzero((char*) &index_file, sizeof(index_file));
+}
+
+/* this is called only once */
+
+void MYSQL_BIN_LOG::cleanup()
+{
+  DBUG_ENTER("cleanup");
+  if (inited)
+  {
+    inited= 0;
+    close(LOG_CLOSE_INDEX|LOG_CLOSE_STOP_EVENT);
+    delete description_event_for_queue;
+    delete description_event_for_exec;
+    (void) pthread_mutex_destroy(&LOCK_log);
+    (void) pthread_mutex_destroy(&LOCK_index);
+    (void) pthread_cond_destroy(&update_cond);
+  }
+  DBUG_VOID_RETURN;
+}
+
+
+/* Init binlog-specific vars */
+void MYSQL_BIN_LOG::init(bool no_auto_events_arg, ulong max_size_arg)
+{
+  DBUG_ENTER("MYSQL_BIN_LOG::init");
+  no_auto_events= no_auto_events_arg;
+  max_size= max_size_arg;
+  DBUG_PRINT("info",("max_size: %lu", max_size));
+  DBUG_VOID_RETURN;
+}
+
+
+void MYSQL_BIN_LOG::init_pthread_objects()
+{
+  DBUG_ASSERT(inited == 0);
+  inited= 1;
+  (void) pthread_mutex_init(&LOCK_log, MY_MUTEX_INIT_SLOW);
+  (void) pthread_mutex_init(&LOCK_index, MY_MUTEX_INIT_SLOW);
+  (void) pthread_cond_init(&update_cond, 0);
+}
+
+
+bool MYSQL_BIN_LOG::open_index_file(const char *index_file_name_arg,
                                 const char *log_name)
 {
   File index_file_nr= -1;
@@ -1509,10 +1966,10 @@ bool MYSQL_LOG::open_index_file(const char *index_file_name_arg,
 
 
 /*
-  Open a (new) log file.
+  Open a (new) binlog file.
 
   DESCRIPTION
-  - If binary logs, also open the index file and register the new
+  - Open the log file and the index file. Register the new
     file name in it
   - When calling this when the file is in use, you must have a locks
     on LOCK_log and LOCK_index.
@@ -1522,94 +1979,32 @@ bool MYSQL_LOG::open_index_file(const char *index_file_name_arg,
     1	error
 */
 
-bool MYSQL_LOG::open(const char *log_name,
+bool MYSQL_BIN_LOG::open(const char *log_name,
                          enum_log_type log_type_arg,
                          const char *new_name,
                          enum cache_type io_cache_type_arg,
                          bool no_auto_events_arg,
                          ulong max_size_arg,
                          bool null_created_arg)
-{
-  char buff[FN_REFLEN];
-  File file= -1;
-  int open_flags = O_CREAT | O_BINARY;
-  DBUG_ENTER("MYSQL_LOG::open");
-  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
-
-  last_time=query_start=0;
-  write_error=0;
-
-  init(log_type_arg,io_cache_type_arg,no_auto_events_arg,max_size_arg);
-
-  if (!(name=my_strdup(log_name,MYF(MY_WME))))
-  {
-    name= (char *)log_name; // for the error message
-    goto err;
-  }
-  if (new_name)
-    strmov(log_file_name,new_name);
-  else if (generate_new_name(log_file_name, name))
-    goto err;
-
-  if (io_cache_type == SEQ_READ_APPEND)
-    open_flags |= O_RDWR | O_APPEND;
-  else
-    open_flags |= O_WRONLY | (log_type == LOG_BIN ? 0 : O_APPEND);
-
-  db[0]=0;
-  open_count++;
-  if ((file=my_open(log_file_name,open_flags,
-		    MYF(MY_WME | ME_WAITTANG))) < 0 ||
-      init_io_cache(&log_file, file, IO_SIZE, io_cache_type,
-		    my_tell(file,MYF(MY_WME)), 0, 
-                    MYF(MY_WME | MY_NABP |
-                        ((log_type == LOG_BIN) ? MY_WAIT_IF_FULL : 0))))
-    goto err;
-
-  switch (log_type) {
-  case LOG_NORMAL:
-  {
-    char *end;
-    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s. "
-#ifdef EMBEDDED_LIBRARY
-		        "embedded library\n", my_progname, server_version
-#elif __NT__
-			"started with:\nTCP Port: %d, Named Pipe: %s\n",
-			my_progname, server_version, mysqld_port, mysqld_unix_port
-#else
-			"started with:\nTcp port: %d  Unix socket: %s\n",
-			my_progname,server_version,mysqld_port,mysqld_unix_port
-#endif
-                       );
-    end=strnmov(buff+len,"Time                 Id Command    Argument\n",
-                sizeof(buff)-len);
-    if (my_b_write(&log_file, (byte*) buff,(uint) (end-buff)) ||
-	flush_io_cache(&log_file))
-      goto err;
-    break;
-  }
-  case LOG_NEW:
-  {
-    uint len;
-    time_t skr=time(NULL);
-    struct tm tm_tmp;
+{
+  char buff[FN_REFLEN];
+  File file= -1;
+  int open_flags = O_CREAT | O_BINARY;
+  DBUG_ENTER("MYSQL_BIN_LOG::open");
+  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
+
+  write_error=0;
+
+  /* open the main log file */
+  if (MYSQL_LOG::open(log_name, log_type_arg, new_name, io_cache_type_arg))
+    DBUG_RETURN(1);                            /* all warnings issued */
+
+  init(no_auto_events_arg, max_size_arg);
+
+  open_count++;
+
+  DBUG_ASSERT(log_type == LOG_BIN);
 
-    localtime_r(&skr,&tm_tmp);
-    len= my_snprintf(buff,sizeof(buff),
-		     "# %s, Version: %s at %02d%02d%02d %2d:%02d:%02d\n",
-		     my_progname,server_version,
-		     tm_tmp.tm_year % 100,
-		     tm_tmp.tm_mon+1,
-		     tm_tmp.tm_mday,
-		     tm_tmp.tm_hour,
-		     tm_tmp.tm_min,
-		     tm_tmp.tm_sec);
-    if (my_b_write(&log_file, (byte*) buff, len) ||
-	flush_io_cache(&log_file))
-      goto err;
-    break;
-  }
-  case LOG_BIN:
   {
     bool write_file_name_to_index_file=0;
 
@@ -1700,13 +2095,9 @@ bool MYSQL_LOG::open(const char *log_name,
           my_sync(index_file.file, MYF(MY_WME)))
 	goto err;
     }
-    break;
-  }
-  case LOG_CLOSED:				// Impossible
-  case LOG_TO_BE_OPENED:
-    DBUG_ASSERT(1);
-    break;
   }
+  log_state= LOG_OPENED;
+
   DBUG_RETURN(0);
 
 err:
@@ -1719,12 +2110,12 @@ shutdown the MySQL server and restart it.", name, errno);
   end_io_cache(&log_file);
   end_io_cache(&index_file);
   safeFree(name);
-  log_type= LOG_CLOSED;
+  log_state= LOG_CLOSED;
   DBUG_RETURN(1);
 }
 
 
-int MYSQL_LOG::get_current_log(LOG_INFO* linfo)
+int MYSQL_BIN_LOG::get_current_log(LOG_INFO* linfo)
 {
   pthread_mutex_lock(&LOCK_log);
   strmake(linfo->log_file_name, log_file_name, sizeof(linfo->log_file_name)-1);
@@ -1811,7 +2202,7 @@ static bool copy_up_file_and_fill(IO_CACHE *index_file, my_off_t offset)
     LOG_INFO_IO		Got IO error while reading file
 */
 
-int MYSQL_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
+int MYSQL_BIN_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
 			    bool need_lock)
 {
   int error= 0;
@@ -1885,7 +2276,7 @@ int MYSQL_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
     LOG_INFO_IO		Got IO error while reading file
 */
 
-int MYSQL_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
+int MYSQL_BIN_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
 {
   int error= 0;
   uint length;
@@ -1933,12 +2324,11 @@ int MYSQL_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
     1   error
 */
 
-bool MYSQL_LOG::reset_logs(THD* thd)
+bool MYSQL_BIN_LOG::reset_logs(THD* thd)
 {
   LOG_INFO linfo;
   bool error=0;
   const char* save_name;
-  enum_log_type save_log_type;
   DBUG_ENTER("reset_logs");
 
   ha_reset_logs(thd);
@@ -1960,7 +2350,6 @@ bool MYSQL_LOG::reset_logs(THD* thd)
   /* Save variables so that we can reopen the log */
   save_name=name;
   name=0;					// Protect against free
-  save_log_type=log_type;
   close(LOG_CLOSE_TO_BE_OPENED);
 
   /* First delete all old log files */
@@ -1984,8 +2373,7 @@ bool MYSQL_LOG::reset_logs(THD* thd)
   if (!thd->slave_thread)
     need_start_event=1;
   if (!open_index_file(index_file_name, 0))
-    open(save_name, save_log_type, 0,
-         io_cache_type, no_auto_events, max_size, 0);
+    open(save_name, log_type, 0, io_cache_type, no_auto_events, max_size, 0);
   my_free((gptr) save_name, MYF(0));
 
 err:
@@ -2033,7 +2421,7 @@ bool MYSQL_LOG::reset_logs(THD* thd)
 
 #ifdef HAVE_REPLICATION
 
-int MYSQL_LOG::purge_first_log(struct st_relay_log_info* rli, bool included) 
+int MYSQL_BIN_LOG::purge_first_log(struct st_relay_log_info* rli, bool included) 
 {
   int error;
   DBUG_ENTER("purge_first_log");
@@ -2109,7 +2497,7 @@ int MYSQL_LOG::purge_first_log(struct st_relay_log_info* rli, bool included)
   Update log index_file
 */
 
-int MYSQL_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
+int MYSQL_BIN_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
 {
   if (copy_up_file_and_fill(&index_file, log_info->index_file_start_offset))
     return LOG_INFO_IO;
@@ -2142,7 +2530,7 @@ int MYSQL_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
     LOG_INFO_EOF		to_log not found
 */
 
-int MYSQL_LOG::purge_logs(const char *to_log, 
+int MYSQL_BIN_LOG::purge_logs(const char *to_log, 
                           bool included,
                           bool need_mutex, 
                           bool need_update_threads, 
@@ -2228,7 +2616,7 @@ int MYSQL_LOG::purge_logs(const char *to_log,
     LOG_INFO_PURGE_NO_ROTATE	Binary file that can't be rotated
 */
 
-int MYSQL_LOG::purge_logs_before_date(time_t purge_time)
+int MYSQL_BIN_LOG::purge_logs_before_date(time_t purge_time)
 {
   int error;
   LOG_INFO log_info;
@@ -2287,7 +2675,7 @@ int MYSQL_LOG::purge_logs_before_date(time_t purge_time)
     If file name will be longer then FN_REFLEN it will be truncated
 */
 
-void MYSQL_LOG::make_log_name(char* buf, const char* log_ident)
+void MYSQL_BIN_LOG::make_log_name(char* buf, const char* log_ident)
 {
   uint dir_len = dirname_length(log_file_name); 
   if (dir_len > FN_REFLEN)
@@ -2301,29 +2689,48 @@ void MYSQL_LOG::make_log_name(char* buf, const char* log_ident)
   Check if we are writing/reading to the given log file
 */
 
-bool MYSQL_LOG::is_active(const char *log_file_name_arg)
+bool MYSQL_BIN_LOG::is_active(const char *log_file_name_arg)
 {
   return !strcmp(log_file_name, log_file_name_arg);
 }
 
 
+/*
+  Wrappers around new_file_impl to avoid using argument
+  to control locking. The argument 1) less readable 2) breaks
+  incapsulation 3) allows external access to the class without
+  a lock (which is not possible with private new_file_without_locking
+  method).
+*/
+
+void MYSQL_BIN_LOG::new_file()
+{
+  new_file_impl(1);
+}
+
+
+void MYSQL_BIN_LOG::new_file_without_locking()
+{
+  new_file_impl(0);
+}
+
+
 /*
   Start writing to a new log file or reopen the old file
 
   SYNOPSIS
-    new_file()
+    new_file_impl()
     need_lock		Set to 1 if caller has not locked LOCK_log
 
   NOTE
     The new file name is stored last in the index file
 */
 
-void MYSQL_LOG::new_file(bool need_lock)
+void MYSQL_BIN_LOG::new_file_impl(bool need_lock)
 {
   char new_name[FN_REFLEN], *new_name_ptr, *old_name;
-  enum_log_type save_log_type;
 
-  DBUG_ENTER("MYSQL_LOG::new_file");
+  DBUG_ENTER("MYSQL_BIN_LOG::new_file_impl");
   if (!is_open())
   {
     DBUG_PRINT("info",("log is closed"));
@@ -2389,12 +2796,11 @@ void MYSQL_LOG::new_file(bool need_lock)
     signal_update();
   }
   old_name=name;
-  save_log_type=log_type;
   name=0;				// Don't free name
   close(LOG_CLOSE_TO_BE_OPENED);
 
   /*
-     Note that at this point, log_type != LOG_CLOSED (important for is_open()).
+     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
   */
 
   /*
@@ -2406,7 +2812,7 @@ void MYSQL_LOG::new_file(bool need_lock)
      trigger temp tables deletion on slaves.
   */
 
-  open(old_name, save_log_type, new_name_ptr,
+  open(old_name, log_type, new_name_ptr,
        io_cache_type, no_auto_events, max_size, 1);
   my_free(old_name,MYF(0));
 
@@ -2419,11 +2825,11 @@ void MYSQL_LOG::new_file(bool need_lock)
 }
 
 
-bool MYSQL_LOG::append(Log_event* ev)
+bool MYSQL_BIN_LOG::append(Log_event* ev)
 {
   bool error = 0;
   pthread_mutex_lock(&LOCK_log);
-  DBUG_ENTER("MYSQL_LOG::append");
+  DBUG_ENTER("MYSQL_BIN_LOG::append");
 
   DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
   /*
@@ -2438,7 +2844,7 @@ bool MYSQL_LOG::append(Log_event* ev)
   bytes_written+= ev->data_written;
   DBUG_PRINT("info",("max_size: %lu",max_size));
   if ((uint) my_b_append_tell(&log_file) > max_size)
-    new_file(0);
+    new_file_without_locking();
 
 err:
   pthread_mutex_unlock(&LOCK_log);
@@ -2447,10 +2853,10 @@ bool MYSQL_LOG::append(Log_event* ev)
 }
 
 
-bool MYSQL_LOG::appendv(const char* buf, uint len,...)
+bool MYSQL_BIN_LOG::appendv(const char* buf, uint len,...)
 {
   bool error= 0;
-  DBUG_ENTER("MYSQL_LOG::appendv");
+  DBUG_ENTER("MYSQL_BIN_LOG::appendv");
   va_list(args);
   va_start(args,len);
 
@@ -2468,7 +2874,7 @@ bool MYSQL_LOG::appendv(const char* buf, uint len,...)
   } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
   DBUG_PRINT("info",("max_size: %lu",max_size));
   if ((uint) my_b_append_tell(&log_file) > max_size)
-    new_file(0);
+    new_file_without_locking();
 
 err:
   if (!error)
@@ -2477,99 +2883,7 @@ bool MYSQL_LOG::appendv(const char* buf, uint len,...)
 }
 
 
-/*
-  Write a command to traditional general log file
-
-  SYNOPSIS
-    write()
-
-    event_time        command start timestamp
-    user_host         the pointer to the string with user@host info
-    user_host_len     length of the user_host string. this is computed once
-                      and passed to all general log  event handlers
-    thread_id         Id of the thread, issued a query
-    command_type      the type of the command being logged
-    command_type_len  the length of the string above
-    sql_text          the very text of the query being executed
-    sql_text_len      the length of sql_text string
-
-  DESCRIPTION
-
-   Log given command to to normal (not rotable) log file
-
-  RETURN
-    FASE - OK
-    TRUE - error occured
-*/
-
-bool MYSQL_LOG::write(time_t event_time, const char *user_host,
-                      uint user_host_len, int thread_id,
-                      const char *command_type, uint command_type_len,
-                      const char *sql_text, uint sql_text_len)
-{
-  char buff[32];
-  uint length= 0;
-  char time_buff[MAX_TIME_SIZE];
-  struct tm start;
-  uint time_buff_len= 0;
-
-  /* Test if someone closed between the is_open test and lock */
-  if (is_open())
-  {
-    /* Note that my_b_write() assumes it knows the length for this */
-      if (event_time != last_time)
-      {
-        last_time= event_time;
-
-        localtime_r(&event_time, &start);
-
-        time_buff_len= my_snprintf(time_buff, MAX_TIME_SIZE,
-                                   "%02d%02d%02d %2d:%02d:%02d",
-                                   start.tm_year % 100, start.tm_mon + 1,
-                                   start.tm_mday, start.tm_hour,
-                                   start.tm_min, start.tm_sec);
-
-        if (my_b_write(&log_file, (byte*) &time_buff, time_buff_len))
-          goto err;
-      }
-      else
-        if (my_b_write(&log_file, (byte*) "\t\t" ,2) < 0)
-          goto err;
-
-    /* command_type, thread_id */
-    length= my_snprintf(buff, 32, "%5ld ", thread_id);
-
-    if (my_b_write(&log_file, (byte*) buff, length))
-      goto err;
-
-    if (my_b_write(&log_file, (byte*) command_type, command_type_len))
-      goto err;
-
-    if (my_b_write(&log_file, (byte*) "\t", 1))
-      goto err;
-
-    /* sql_text */
-    if (my_b_write(&log_file, (byte*) sql_text, sql_text_len))
-      goto err;
-
-    if (my_b_write(&log_file, (byte*) "\n", 1) ||
-        flush_io_cache(&log_file))
-      goto err;
-  }
-
-  return FALSE;
-err:
-
-  if (!write_error)
-  {
-    write_error= 1;
-    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
-  }
-  return TRUE;
-}
-
-
-bool MYSQL_LOG::flush_and_sync()
+bool MYSQL_BIN_LOG::flush_and_sync()
 {
   int err=0, fd=log_file.file;
   safe_mutex_assert_owner(&LOCK_log);
@@ -2583,7 +2897,7 @@ bool MYSQL_LOG::flush_and_sync()
   return err;
 }
 
-void MYSQL_LOG::start_union_events(THD *thd)
+void MYSQL_BIN_LOG::start_union_events(THD *thd)
 {
   DBUG_ASSERT(!thd->binlog_evt_union.do_union);
   thd->binlog_evt_union.do_union= TRUE;
@@ -2592,13 +2906,13 @@ void MYSQL_LOG::start_union_events(THD *thd)
   thd->binlog_evt_union.first_query_id= thd->query_id;
 }
 
-void MYSQL_LOG::stop_union_events(THD *thd)
+void MYSQL_BIN_LOG::stop_union_events(THD *thd)
 {
   DBUG_ASSERT(thd->binlog_evt_union.do_union);
   thd->binlog_evt_union.do_union= FALSE;
 }
 
-bool MYSQL_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
+bool MYSQL_BIN_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
 {
   return (thd->binlog_evt_union.do_union && 
           query_id_param >= thd->binlog_evt_union.first_query_id);
@@ -2705,9 +3019,10 @@ THD::binlog_set_pending_rows_event(Rows_log_event* ev)
   (either cached binlog if transaction, or disk binlog). Sets a new pending
   event.
 */
-int MYSQL_LOG::flush_and_set_pending_rows_event(THD *thd, Rows_log_event* event)
+int MYSQL_BIN_LOG::
+  flush_and_set_pending_rows_event(THD *thd, Rows_log_event* event)
 {
-  DBUG_ENTER("MYSQL_LOG::flush_and_set_pending_rows_event(event)");
+  DBUG_ENTER("MYSQL_BIN_LOG::flush_and_set_pending_rows_event(event)");
   DBUG_ASSERT(thd->current_stmt_binlog_row_based && mysql_bin_log.is_open());
   DBUG_PRINT("enter", ("event=%p", event));
 
@@ -2791,11 +3106,11 @@ int MYSQL_LOG::flush_and_set_pending_rows_event(THD *thd, Rows_log_event* event)
   Write an event to the binary log
 */
 
-bool MYSQL_LOG::write(Log_event *event_info)
+bool MYSQL_BIN_LOG::write(Log_event *event_info)
 {
   THD *thd= event_info->thd;
   bool error= 1;
-  DBUG_ENTER("MYSQL_LOG::write(Log_event *)");
+  DBUG_ENTER("MYSQL_BIN_LOG::write(Log_event *)");
 
   if (thd->binlog_evt_union.do_union)
   {
@@ -3015,14 +3330,14 @@ bool general_log_print(THD *thd, enum enum_server_command command,
   return error;
 }
 
-void MYSQL_LOG::rotate_and_purge(uint flags)
+void MYSQL_BIN_LOG::rotate_and_purge(uint flags)
 {
   if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
     pthread_mutex_lock(&LOCK_log);
   if ((flags & RP_FORCE_ROTATE) ||
       (my_b_tell(&log_file) >= (my_off_t) max_size))
   {
-    new_file(0);
+    new_file_without_locking();
 #ifdef HAVE_REPLICATION
     if (expire_logs_days)
     {
@@ -3036,7 +3351,7 @@ void MYSQL_LOG::rotate_and_purge(uint flags)
     pthread_mutex_unlock(&LOCK_log);
 }
 
-uint MYSQL_LOG::next_file_id()
+uint MYSQL_BIN_LOG::next_file_id()
 {
   uint res;
   pthread_mutex_lock(&LOCK_log);
@@ -3067,9 +3382,9 @@ uint MYSQL_LOG::next_file_id()
       that the same updates are run on the slave.
 */
 
-bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
+bool MYSQL_BIN_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
 {
-  DBUG_ENTER("MYSQL_LOG::write(THD *, IO_CACHE *, Log_event *)");
+  DBUG_ENTER("MYSQL_BIN_LOG::write(THD *, IO_CACHE *, Log_event *)");
   VOID(pthread_mutex_lock(&LOCK_log));
 
   if (likely(is_open()))                       // Should always be true
@@ -3164,148 +3479,6 @@ bool MYSQL_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
 }
 
 
-/*
-  Log a query to the traditional slow log file
-
-  SYNOPSIS
-    write()
-
-    thd               THD of the query
-    current_time      current timestamp
-    query_start_arg   command start timestamp
-    user_host         the pointer to the string with user@host info
-    user_host_len     length of the user_host string. this is computed once
-                      and passed to all general log event handlers
-    query_time        Amount of time the query took to execute (in seconds)
-    lock_time         Amount of time the query was locked (in seconds)
-    is_command        The flag, which determines, whether the sql_text is a
-                      query or an administrator command.
-    sql_text          the very text of the query or administrator command
-                      processed
-    sql_text_len      the length of sql_text string
-
-  DESCRIPTION
-
-   Log a query to the slow log file.
-
-  RETURN
-    FALSE - OK
-    TRUE - error occured
-*/
-
-bool MYSQL_LOG::write(THD *thd, time_t current_time, time_t query_start_arg,
-                      const char *user_host, uint user_host_len,
-                      longlong query_time, longlong lock_time, bool is_command,
-                      const char *sql_text, uint sql_text_len)
-{
-  bool error= 0;
-  DBUG_ENTER("MYSQL_LOG::write");
-
-  if (!is_open())
-    DBUG_RETURN(0);
-
-  if (is_open())
-  {						// Safety agains reopen
-    int tmp_errno= 0;
-    char buff[80], *end;
-    uint buff_len;
-    end= buff;
-
-    if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
-    {
-      Security_context *sctx= thd->security_ctx;
-      if (current_time != last_time)
-      {
-        last_time= current_time;
-        struct tm start;
-        localtime_r(&current_time, &start);
-
-        buff_len= my_snprintf(buff, sizeof buff,
-                              "# Time: %02d%02d%02d %2d:%02d:%02d\n",
-                              start.tm_year % 100, start.tm_mon + 1,
-                              start.tm_mday, start.tm_hour,
-                              start.tm_min, start.tm_sec);
-
-        /* Note that my_b_write() assumes it knows the length for this */
-        if (my_b_write(&log_file, (byte*) buff, buff_len))
-          tmp_errno=errno;
-      }
-      if (my_b_printf(&log_file, "# User@Host: ", sizeof("# User@Host: ") - 1))
-        tmp_errno=errno;
-      if (my_b_printf(&log_file, user_host, user_host_len))
-        tmp_errno=errno;
-      if (my_b_write(&log_file, (byte*) "\n", 1))
-        tmp_errno=errno;
-    }
-    /* For slow query log */
-    if (my_b_printf(&log_file,
-                    "# Query_time: %lu  Lock_time: %lu"
-                    " Rows_sent: %lu  Rows_examined: %lu\n",
-                    (ulong) query_time, (ulong) lock_time,
-                    (ulong) thd->sent_row_count,
-                    (ulong) thd->examined_row_count) == (uint) -1)
-      tmp_errno=errno;
-    if (thd->db && strcmp(thd->db,db))
-    {						// Database changed
-      if (my_b_printf(&log_file,"use %s;\n",thd->db) == (uint) -1)
-        tmp_errno=errno;
-      strmov(db,thd->db);
-    }
-    if (thd->last_insert_id_used)
-    {
-      end=strmov(end,",last_insert_id=");
-      end=longlong10_to_str((longlong) thd->current_insert_id,end,-10);
-    }
-    // Save value if we do an insert.
-    if (thd->insert_id_used)
-    {
-      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
-      {
-        end=strmov(end,",insert_id=");
-        end=longlong10_to_str((longlong) thd->last_insert_id,end,-10);
-      }
-    }
-
-    /*
-      This info used to show up randomly, depending on whether the query
-      checked the query start time or not. now we always write current
-      timestamp to the slow log
-    */
-    end= strmov(end, ",timestamp=");
-    end= int10_to_str((long) current_time, end, 10);
-
-    if (end != buff)
-    {
-      *end++=';';
-      *end='\n';
-      if (my_b_write(&log_file, (byte*) "SET ",4) ||
-          my_b_write(&log_file, (byte*) buff+1,(uint) (end-buff)))
-        tmp_errno=errno;
-    }
-    if (is_command)
-    {
-      end= strxmov(buff, "# administrator command: ", NullS);
-      buff_len= (ulong) (end - buff);
-      my_b_write(&log_file, (byte*) buff, buff_len);
-    }
-    if (my_b_write(&log_file, (byte*) sql_text, sql_text_len) ||
-        my_b_write(&log_file, (byte*) ";\n",2) ||
-        flush_io_cache(&log_file))
-      tmp_errno=errno;
-    if (tmp_errno)
-    {
-      error=1;
-      if (! write_error)
-      {
-        write_error=1;
-        sql_print_error(ER(ER_ERROR_ON_WRITE),name,error);
-      }
-    }
-  }
-  DBUG_RETURN(error);
-}
-
-
 /*
   Wait until we get a signal that the binary log has been updated
 
@@ -3322,7 +3495,7 @@ bool MYSQL_LOG::write(THD *thd, time_t current_time, time_t query_start_arg,
     THD::enter_cond() (see NOTES in sql_class.h).
 */
 
-void MYSQL_LOG::wait_for_update(THD* thd, bool is_slave)
+void MYSQL_BIN_LOG::wait_for_update(THD* thd, bool is_slave)
 {
   const char *old_msg;
   DBUG_ENTER("wait_for_update");
@@ -3355,11 +3528,11 @@ void MYSQL_LOG::wait_for_update(THD* thd, bool is_slave)
     The internal structures are not freed until cleanup() is called
 */
 
-void MYSQL_LOG::close(uint exiting)
+void MYSQL_BIN_LOG::close(uint exiting)
 {					// One can't set log_type here!
-  DBUG_ENTER("MYSQL_LOG::close");
+  DBUG_ENTER("MYSQL_BIN_LOG::close");
   DBUG_PRINT("enter",("exiting: %d", (int) exiting));
-  if (log_type != LOG_CLOSED && log_type != LOG_TO_BE_OPENED)
+  if (log_state == LOG_OPENED)
   {
 #ifdef HAVE_REPLICATION
     if (log_type == LOG_BIN && !no_auto_events &&
@@ -3371,7 +3544,6 @@ void MYSQL_LOG::close(uint exiting)
       signal_update();
     }
 #endif /* HAVE_REPLICATION */
-    end_io_cache(&log_file);
 
     /* don't pwrite in a file opened with O_APPEND - it doesn't work */
     if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
@@ -3381,16 +3553,8 @@ void MYSQL_LOG::close(uint exiting)
       my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
     }
 
-    if (my_sync(log_file.file,MYF(MY_WME)) && ! write_error)
-    {
-      write_error=1;
-      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
-    }
-    if (my_close(log_file.file,MYF(MY_WME)) && ! write_error)
-    {
-      write_error=1;
-      sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
-    }
+    /* this will cleanup IO_CACHE, sync and close the file */
+    MYSQL_LOG::close(exiting);
   }
 
   /*
@@ -3407,13 +3571,13 @@ void MYSQL_LOG::close(uint exiting)
       sql_print_error(ER(ER_ERROR_ON_WRITE), index_file_name, errno);
     }
   }
-  log_type= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
+  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
   safeFree(name);
   DBUG_VOID_RETURN;
 }
 
 
-void MYSQL_LOG::set_max_size(ulong max_size_arg)
+void MYSQL_BIN_LOG::set_max_size(ulong max_size_arg)
 {
   /*
     We need to take locks, otherwise this may happen:
@@ -3422,7 +3586,7 @@ void MYSQL_LOG::set_max_size(ulong max_size_arg)
     uses the old_max_size argument, so max_size_arg has been overwritten and
     it's like if the SET command was never run.
   */
-  DBUG_ENTER("MYSQL_LOG::set_max_size");
+  DBUG_ENTER("MYSQL_BIN_LOG::set_max_size");
   pthread_mutex_lock(&LOCK_log);
   if (is_open())
     max_size= max_size_arg;
@@ -3571,9 +3735,9 @@ bool flush_error_log()
    return result;
 }
 
-void MYSQL_LOG::signal_update()
+void MYSQL_BIN_LOG::signal_update()
 {
-  DBUG_ENTER("MYSQL_LOG::signal_update");
+  DBUG_ENTER("MYSQL_BIN_LOG::signal_update");
   pthread_cond_broadcast(&update_cond);
   DBUG_VOID_RETURN;
 }
@@ -4176,7 +4340,7 @@ int TC_LOG::using_heuristic_recover()
 }
 
 /****** transaction coordinator log for 2pc - binlog() based solution ******/
-#define TC_LOG_BINLOG MYSQL_LOG
+#define TC_LOG_BINLOG MYSQL_BIN_LOG
 
 /*
   TODO keep in-memory list of prepared transactions

sql/log.h

@@ -147,33 +147,85 @@ typedef struct st_log_info
 class Log_event;
 class Rows_log_event;
 
-enum enum_log_type { LOG_CLOSED, LOG_TO_BE_OPENED, LOG_NORMAL, LOG_NEW, LOG_BIN};
+enum enum_log_type { LOG_UNKNOWN, LOG_NORMAL, LOG_BIN };
+enum enum_log_state { LOG_OPENED, LOG_CLOSED, LOG_TO_BE_OPENED };
 
 /*
-  TODO split MYSQL_LOG into base MYSQL_LOG and
-  MYSQL_QUERY_LOG, MYSQL_SLOW_LOG, MYSQL_BIN_LOG
-  most of the code from MYSQL_LOG should be in the MYSQL_BIN_LOG
-  only (TC_LOG included)
-
   TODO use mmap instead of IO_CACHE for binlog
   (mmap+fsync is two times faster than write+fsync)
 */
 
-class MYSQL_LOG: public TC_LOG
+class MYSQL_LOG
+{
+public:
+  MYSQL_LOG();
+  void init_pthread_objects();
+  void cleanup();
+  bool open(const char *log_name,
+            enum_log_type log_type,
+            const char *new_name,
+            enum cache_type io_cache_type_arg);
+  void init(enum_log_type log_type_arg,
+            enum cache_type io_cache_type_arg);
+  void close(uint exiting);
+  inline bool is_open() { return log_state != LOG_CLOSED; }
+  const char *generate_name(const char *log_name, const char *suffix,
+                            bool strip_ext, char *buff);
+  int generate_new_name(char *new_name, const char *log_name);
+ protected:
+  /* LOCK_log is inited by init_pthread_objects() */
+  pthread_mutex_t LOCK_log;
+  char *name;
+  char log_file_name[FN_REFLEN];
+  char time_buff[20], db[NAME_LEN + 1];
+  bool write_error, inited;
+  IO_CACHE log_file;
+  enum_log_type log_type;
+  volatile enum_log_state log_state;
+  enum cache_type io_cache_type;
+  friend class Log_event;
+};
+
+class MYSQL_QUERY_LOG: public MYSQL_LOG
+{
+public:
+  MYSQL_QUERY_LOG() : last_time(0) {}
+  void reopen_file();
+  bool write(time_t event_time, const char *user_host,
+             uint user_host_len, int thread_id,
+             const char *command_type, uint command_type_len,
+             const char *sql_text, uint sql_text_len);
+  bool write(THD *thd, time_t current_time, time_t query_start_arg,
+             const char *user_host, uint user_host_len,
+             longlong query_time, longlong lock_time, bool is_command,
+             const char *sql_text, uint sql_text_len);
+  bool open_slow_log(const char *log_name)
+  {
+    char buf[FN_REFLEN];
+    return open(generate_name(log_name, "-slow.log", 0, buf), LOG_NORMAL, 0,
+                WRITE_CACHE);
+  }
+  bool open_query_log(const char *log_name)
+  {
+    char buf[FN_REFLEN];
+    return open(generate_name(log_name, ".log", 0, buf), LOG_NORMAL, 0,
+                WRITE_CACHE);
+  }
+private:
+  time_t last_time;
+};
+
+class MYSQL_BIN_LOG: public TC_LOG, private MYSQL_LOG
 {
  private:
   /* LOCK_log and LOCK_index are inited by init_pthread_objects() */
-  pthread_mutex_t LOCK_log, LOCK_index;
+  pthread_mutex_t LOCK_index;
   pthread_mutex_t LOCK_prep_xids;
   pthread_cond_t  COND_prep_xids;
   pthread_cond_t update_cond;
   ulonglong bytes_written;
-  time_t last_time,query_start;
-  IO_CACHE log_file;
   IO_CACHE index_file;
-  char *name;
-  char time_buff[20],db[NAME_LEN+1];
-  char log_file_name[FN_REFLEN],index_file_name[FN_REFLEN];
+  char index_file_name[FN_REFLEN];
   /*
      The max size before rotation (usable only if log_type == LOG_BIN: binary
      logs and relay logs).
@@ -186,13 +238,10 @@ class MYSQL_LOG: public TC_LOG
   */
   ulong max_size;
   ulong prepared_xids; /* for tc log - number of xids to remember */
-  volatile enum_log_type log_type;
-  enum cache_type io_cache_type;
   // current file sequence number for load data infile binary logging
   uint file_id;
   uint open_count;				// For replication
   int readers_count;
-  bool write_error, inited;
   bool need_start_event;
   /*
     no_auto_events means we don't want any of these automatic events :
@@ -202,13 +251,21 @@ class MYSQL_LOG: public TC_LOG
     In 5.0 it's 0 for relay logs too!
   */
   bool no_auto_events;
-  friend class Log_event;
 
   ulonglong m_table_map_version;
 
   int write_to_file(IO_CACHE *cache);
+  /*
+    This is used to start writing to a new log file. The difference from
+    new_file() is locking. new_file_without_locking() does not acquire
+    LOCK_log.
+  */
+  void new_file_without_locking();
+  void new_file_impl(bool need_lock);
 
 public:
+  MYSQL_LOG::generate_name;
+  MYSQL_LOG::is_open;
   /*
     These describe the log's format. This is used only for relay logs.
     _for_exec is used by the SQL thread, _for_queue by the I/O thread. It's
@@ -220,9 +277,9 @@ class MYSQL_LOG: public TC_LOG
   Format_description_log_event *description_event_for_exec,
     *description_event_for_queue;
 
-  MYSQL_LOG();
+  MYSQL_BIN_LOG();
   /*
-    note that there's no destructor ~MYSQL_LOG() !
+    note that there's no destructor ~MYSQL_BIN_LOG() !
     The reason is that we don't want it to be automatically called
     on exit() - but only during the correct shutdown process
   */
@@ -264,9 +321,7 @@ class MYSQL_LOG: public TC_LOG
   void signal_update();
   void wait_for_update(THD* thd, bool master_or_slave);
   void set_need_start_event() { need_start_event = 1; }
-  void init(enum_log_type log_type_arg,
-	    enum cache_type io_cache_type_arg,
-	    bool no_auto_events_arg, ulong max_size);
+  void init(bool no_auto_events_arg, ulong max_size);
   void init_pthread_objects();
   void cleanup();
   bool open(const char *log_name,
@@ -275,35 +330,10 @@ class MYSQL_LOG: public TC_LOG
 	    enum cache_type io_cache_type_arg,
 	    bool no_auto_events_arg, ulong max_size,
             bool null_created);
-  const char *generate_name(const char *log_name, const char *suffix,
-                            bool strip_ext, char *buff);
-  /* simplified open_xxx wrappers for the gigantic open above */
-  bool open_query_log(const char *log_name)
-  {
-    char buf[FN_REFLEN];
-    return open(generate_name(log_name, ".log", 0, buf),
-                LOG_NORMAL, 0, WRITE_CACHE, 0, 0, 0);
-  }
-  bool open_slow_log(const char *log_name)
-  {
-    char buf[FN_REFLEN];
-    return open(generate_name(log_name, "-slow.log", 0, buf),
-                LOG_NORMAL, 0, WRITE_CACHE, 0, 0, 0);
-  }
   bool open_index_file(const char *index_file_name_arg,
                        const char *log_name);
-  void new_file(bool need_lock);
-  /* log a command to the old-fashioned general log */
-  bool write(time_t event_time, const char *user_host,
-             uint user_host_len, int thread_id,
-             const char *command_type, uint command_type_len,
-             const char *sql_text, uint sql_text_len);
-
-  /* log a query to the old-fashioned slow query log */
-  bool write(THD *thd, time_t current_time, time_t query_start_arg,
-             const char *user_host, uint user_host_len,
-             longlong query_time, longlong lock_time, bool is_command,
-             const char *sql_text, uint sql_text_len);
+  /* Use this to start writing a new log file */
+  void new_file();
 
   bool write(Log_event* event_info); // binary log write
   bool write(THD *thd, IO_CACHE *cache, Log_event *commit_event);
@@ -319,7 +349,6 @@ class MYSQL_LOG: public TC_LOG
   bool appendv(const char* buf,uint len,...);
   bool append(Log_event* ev);
 
-  int generate_new_name(char *new_name,const char *old_name);
   void make_log_name(char* buf, const char* log_ident);
   bool is_active(const char* log_file_name);
   int update_log_index(LOG_INFO* linfo, bool need_update_threads);
@@ -339,7 +368,6 @@ class MYSQL_LOG: public TC_LOG
   int find_next_log(LOG_INFO* linfo, bool need_mutex);
   int get_current_log(LOG_INFO* linfo);
   uint next_file_id();
-  inline bool is_open() { return log_type != LOG_CLOSED; }
   inline char* get_index_fname() { return index_file_name;}
   inline char* get_log_fname() { return log_file_name; }
   inline char* get_name() { return name; }
@@ -416,7 +444,8 @@ class Log_to_csv_event_handler: public Log_event_handler
 
 class Log_to_file_event_handler: public Log_event_handler
 {
-  MYSQL_LOG mysql_log, mysql_slow_log;
+  MYSQL_QUERY_LOG mysql_log;
+  MYSQL_QUERY_LOG mysql_slow_log;
   bool is_initialized;
 public:
   Log_to_file_event_handler(): is_initialized(FALSE)

sql/log_event.h

@@ -470,7 +470,7 @@ enum Int_event_type
 
 #ifndef MYSQL_CLIENT
 class String;
-class MYSQL_LOG;
+class MYSQL_BIN_LOG;
 class THD;
 #endif
 

sql/mysql_priv.h

@@ -1537,7 +1537,7 @@ extern char *default_tz_name;
 extern my_bool opt_large_pages;
 extern uint opt_large_page_size;
 
-extern MYSQL_LOG mysql_bin_log;
+extern MYSQL_BIN_LOG mysql_bin_log;
 extern LOGGER logger;
 extern TABLE_LIST general_log, slow_log;
 extern FILE *bootstrap_file;

sql/mysqld.cc

@@ -2635,10 +2635,10 @@ static int init_common_variables(const char *conf_file_name, int argc,
   global_system_variables.time_zone= my_tz_SYSTEM;
   
   /*
-    Init mutexes for the global MYSQL_LOG objects.
+    Init mutexes for the global MYSQL_BIN_LOG objects.
     As safe_mutex depends on what MY_INIT() does, we can't init the mutexes of
-    global MYSQL_LOGs in their constructors, because then they would be inited
-    before MY_INIT(). So we do it here.
+    global MYSQL_BIN_LOGs in their constructors, because then they would be
+    inited before MY_INIT(). So we do it here.
   */
   mysql_bin_log.init_pthread_objects();
 

sql/rpl_injector.cc

@@ -25,7 +25,7 @@
 
 /* inline since it's called below */
 inline
-injector::transaction::transaction(MYSQL_LOG *log, THD *thd)
+injector::transaction::transaction(MYSQL_BIN_LOG *log, THD *thd)
   : m_state(START_STATE), m_thd(thd)
 {
   /* 

sql/rpl_injector.h

@@ -26,7 +26,7 @@
 
 /* Forward declarations */
 class handler;
-class MYSQL_LOG;
+class MYSQL_BIN_LOG;
 class st_table;
 
 typedef st_table TABLE;
@@ -219,7 +219,7 @@ class injector
 
     private:
       /* Only the injector may construct these object */
-      transaction(MYSQL_LOG *, THD *);
+      transaction(MYSQL_BIN_LOG *, THD *);
 
       void swap(transaction& o) {
         /* std::swap(m_start_pos, o.m_start_pos); */

sql/rpl_rli.h

@@ -69,7 +69,7 @@ typedef struct st_relay_log_info
     Protected with internal locks.
     Must get data_lock when resetting the logs.
   */
-  MYSQL_LOG relay_log;
+  MYSQL_BIN_LOG relay_log;
   LOG_INFO linfo;
   IO_CACHE cache_buf,*cur_log;
 

sql/slave.cc

@@ -4658,7 +4658,7 @@ static Log_event* next_event(RELAY_LOG_INFO* rli)
       When the relay log is created when the I/O thread starts, easy: the
       master will send the description event and we will queue it.
       But if the relay log is created by new_file(): then the solution is:
-      MYSQL_LOG::open() will write the buffered description event.
+      MYSQL_BIN_LOG::open() will write the buffered description event.
     */
     if ((ev=Log_event::read_log_event(cur_log,0,
                                       rli->relay_log.description_event_for_exec)))
@@ -4920,7 +4920,8 @@ event(errno: %d  cur_log->error: %d)",
   Rotate a relay log (this is used only by FLUSH LOGS; the automatic rotation
   because of size is simpler because when we do it we already have all relevant
   locks; here we don't, so this function is mainly taking locks). 
-  Returns nothing as we cannot catch any error (MYSQL_LOG::new_file() is void).
+  Returns nothing as we cannot catch any error (MYSQL_BIN_LOG::new_file()
+  is void).
 */
 
 void rotate_relay_log(MASTER_INFO* mi)
@@ -4942,7 +4943,7 @@ void rotate_relay_log(MASTER_INFO* mi)
   }
 
   /* If the relay log is closed, new_file() will do nothing. */
-  rli->relay_log.new_file(1);
+  rli->relay_log.new_file();
 
   /*
     We harvest now, because otherwise BIN_LOG_HEADER_SIZE will not immediately

sql/slave.h

@@ -73,7 +73,7 @@
   run_lock protects all information about the run state: slave_running, and the
   existence of the I/O thread (to stop/start it, you need this mutex).
   data_lock protects some moving members of the struct: counters (log name,
-  position) and relay log (MYSQL_LOG object).
+  position) and relay log (MYSQL_BIN_LOG object).
 
   In RELAY_LOG_INFO: run_lock, data_lock
   see MASTER_INFO
@@ -81,7 +81,7 @@
   Order of acquisition: if you want to have LOCK_active_mi and a run_lock, you
   must acquire LOCK_active_mi first.
 
-  In MYSQL_LOG: LOCK_log, LOCK_index of the binlog and the relay log
+  In MYSQL_BIN_LOG: LOCK_log, LOCK_index of the binlog and the relay log
   LOCK_log: when you write to it. LOCK_index: when you create/delete a binlog
   (so that you have to update the .index file).
 */

storage/csv/ha_tina.cc

@@ -61,7 +61,7 @@
 
 /* The file extension */
 #define CSV_EXT ".CSV"               // The data file
-#define CSN_EXT ".CSN"               // Files used during repair
+#define CSN_EXT ".CSN"               // Files used during repair and update
 #define CSM_EXT ".CSM"               // Meta file
 
 
@@ -77,6 +77,53 @@ static int tina_init= 0;
 static handler *tina_create_handler(TABLE_SHARE *table, MEM_ROOT *mem_root);
 static int tina_init_func();
 
+off_t Transparent_file::read_next()
+{
+  off_t bytes_read;
+
+  /*
+     No need to seek here, as the file managed by Transparent_file class
+     always points to upper_bound byte
+  */
+  if ((bytes_read= my_read(filedes, buff, buff_size, MYF(0))) == MY_FILE_ERROR)
+    return -1;
+
+  /* end of file */
+  if (!bytes_read)
+    return -1;
+
+  lower_bound= upper_bound;
+  upper_bound+= bytes_read;
+
+  return lower_bound;
+}
+
+
+char Transparent_file::get_value(off_t offset)
+{
+  off_t bytes_read;
+
+  /* check boundaries */
+  if ((lower_bound <= offset) && (offset < upper_bound))
+    return buff[offset - lower_bound];
+  else
+  {
+    VOID(my_seek(filedes, offset, MY_SEEK_SET, MYF(0)));
+    /* read appropriate portion of the file */
+    if ((bytes_read= my_read(filedes, buff, buff_size,
+                             MYF(0))) == MY_FILE_ERROR)
+      return 0;
+
+    lower_bound= offset;
+    upper_bound= lower_bound + bytes_read;
+
+    /* end of file */
+    if (upper_bound == offset)
+      return 0;
+
+    return buff[0];
+  }
+}
 handlerton tina_hton;
 
 /*****************************************************************************
@@ -92,7 +139,7 @@ int sort_set (tina_set *a, tina_set *b)
     We assume that intervals do not intersect. So, it is enought to compare
     any two points. Here we take start of intervals for comparison.
   */
-  return ( a->begin > b->begin ? -1 : ( a->begin < b->begin ? 1 : 0 ) );
+  return ( a->begin > b->begin ? 1 : ( a->begin < b->begin ? -1 : 0 ) );
 }
 
 static byte* tina_get_key(TINA_SHARE *share,uint *length,
@@ -102,55 +149,6 @@ static byte* tina_get_key(TINA_SHARE *share,uint *length,
   return (byte*) share->table_name;
 }
 
-/*
-  Reloads the mmap file.
-*/
-int get_mmap(TINA_SHARE *share, int write)
-{
-  DBUG_ENTER("ha_tina::get_mmap");
-#ifdef __NETWARE__
-  my_message(errno, "Sorry, no mmap() on Netware", 0);
-  DBUG_ASSERT(0);
-  DBUG_RETURN(1);
-#else
-  if (share->mapped_file && my_munmap(share->mapped_file,
-                                      share->file_stat.st_size))
-    DBUG_RETURN(1);
-
-  if (my_fstat(share->data_file, &share->file_stat, MYF(MY_WME)) == -1)
-    DBUG_RETURN(1);
-
-  if (share->file_stat.st_size)
-  {
-    if (write)
-      share->mapped_file= (byte *)my_mmap(NULL, share->file_stat.st_size,
-                                          PROT_READ|PROT_WRITE, MAP_SHARED,
-                                          share->data_file, 0);
-    else
-      share->mapped_file= (byte *)my_mmap(NULL, share->file_stat.st_size,
-                                          PROT_READ, MAP_PRIVATE,
-                                          share->data_file, 0);
-    if ((share->mapped_file == MAP_FAILED))
-    {
-      /*
-        Bad idea you think? See the problem is that nothing actually checks
-        the return value of ::rnd_init(), so tossing an error is about
-        it for us.
-        Never going to happen right? :)
-      */
-      my_message(errno, "Woops, blew up opening a mapped file", 0);
-      DBUG_ASSERT(0);
-      DBUG_RETURN(1);
-    }
-  }
-  else
-    share->mapped_file= NULL;
-
-  DBUG_RETURN(0);
-#endif /* __NETWARE__ */
-}
-
-
 static int tina_init_func()
 {
   if (!tina_init)
@@ -191,6 +189,7 @@ static TINA_SHARE *get_share(const char *table_name, TABLE *table)
 {
   TINA_SHARE *share;
   char meta_file_name[FN_REFLEN];
+  MY_STAT file_stat;                /* Stat information for the data file */
   char *tmp_name;
   uint length;
 
@@ -223,6 +222,8 @@ static TINA_SHARE *get_share(const char *table_name, TABLE *table)
     share->table_name= tmp_name;
     share->crashed= FALSE;
     share->rows_recorded= 0;
+    share->update_file_opened= FALSE;
+    share->tina_write_opened= FALSE;
     strmov(share->table_name, table_name);
     fn_format(share->data_file_name, table_name, "", CSV_EXT,
               MY_REPLACE_EXT|MY_UNPACK_FILENAME);
@@ -244,33 +245,21 @@ static TINA_SHARE *get_share(const char *table_name, TABLE *table)
       share->crashed= TRUE;
 
     /*
-      After we read, we set the file to dirty. When we close, we will do the
-      opposite. If the meta file will not open we assume it is crashed and
+      If the meta file will not open we assume it is crashed and
       mark it as such.
     */
     if (read_meta_file(share->meta_file, &share->rows_recorded))
       share->crashed= TRUE;
-    else
-      (void)write_meta_file(share->meta_file, share->rows_recorded, TRUE);
 
-    if ((share->data_file= my_open(share->data_file_name, O_RDWR|O_APPEND,
-                                   MYF(0))) == -1)
+    if (my_stat(share->data_file_name, &file_stat, MYF(MY_WME)) == NULL)
       goto error2;
-
-    share->mapped_file= NULL; // We don't know the state as we just allocated it
-    if (get_mmap(share, 0) > 0)
-      goto error3;
-
-    /* init file length value used by readers */
-    share->saved_data_file_length= share->file_stat.st_size;
+    share->saved_data_file_length= file_stat.st_size;
   }
   share->use_count++;
   pthread_mutex_unlock(&tina_mutex);
 
   return share;
 
-error3:
-  my_close(share->data_file,MYF(0));
 error2:
   thr_lock_delete(&share->lock);
   pthread_mutex_destroy(&share->mutex);
@@ -398,6 +387,30 @@ bool ha_tina::check_and_repair(THD *thd)
 }
 
 
+int ha_tina::init_tina_writer()
+{
+  DBUG_ENTER("ha_tina::init_tina_writer");
+
+  /*
+    Mark the file as crashed. We will set the flag back when we close
+    the file. In the case of the crash it will remain marked crashed,
+    which enforce recovery.
+  */
+  (void)write_meta_file(share->meta_file, share->rows_recorded, TRUE);
+
+  if ((share->tina_write_filedes=
+        my_open(share->data_file_name, O_RDWR|O_APPEND, MYF(0))) == -1)
+  {
+    DBUG_PRINT("info", ("Could not open tina file writes"));
+    share->crashed= TRUE;
+    DBUG_RETURN(1);
+  }
+  share->tina_write_opened= TRUE;
+
+  DBUG_RETURN(0);
+}
+
+
 bool ha_tina::is_crashed() const
 {
   DBUG_ENTER("ha_tina::is_crashed");
@@ -418,10 +431,13 @@ static int free_share(TINA_SHARE *share)
                           share->crashed ? TRUE :FALSE);
     if (my_close(share->meta_file, MYF(0)))
       result_code= 1;
-    if (share->mapped_file)
-      my_munmap(share->mapped_file, share->file_stat.st_size);
-    share->mapped_file= NULL;
-    result_code= my_close(share->data_file,MYF(0));
+    if (share->tina_write_opened)
+    {
+      if (my_close(share->tina_write_filedes, MYF(0)))
+        result_code= 1;
+      share->tina_write_opened= FALSE;
+    }
+
     hash_delete(&tina_open_tables, (byte*) share);
     thr_lock_delete(&share->lock);
     pthread_mutex_destroy(&share->mutex);
@@ -437,30 +453,41 @@ int tina_end(ha_panic_function type)
   return tina_done_func();
 }
 
+
 /*
-  Finds the end of a line.
-  Supports DOS, Unix, or Mac OS line endings.
+  This function finds the end of a line and returns the length
+  of the line ending.
+
+  We support three kinds of line endings:
+  '\r'     --  Old Mac OS line ending
+  '\n'     --  Traditional Unix and Mac OS X line ending
+  '\r''\n' --  DOS\Windows line ending
 */
-byte * find_eoln(byte *data, off_t begin, off_t end, int *eoln_len)
+
+off_t find_eoln_buff(Transparent_file *data_buff, off_t begin,
+                     off_t end, int *eoln_len)
 {
-  off_t dataend= begin;
   *eoln_len= 0;
 
   for (off_t x= begin; x < end; x++)
-    if (data[x] == '\r' || data[x] == '\n')
-      (*eoln_len)++;
-    else if (!(*eoln_len))
-      dataend++;
+  {
+    /* Unix (includes Mac OS X) */
+    if (data_buff->get_value(x) == '\n')
+      *eoln_len= 1;
     else
-      return data+dataend;
+      if (data_buff->get_value(x) == '\r') // Mac or Dos
+      {
+        /* old Mac line ending */
+        if (x + 1 == end || (data_buff->get_value(x + 1) != '\n'))
+          *eoln_len= 1;
+        else // DOS style ending
+          *eoln_len= 2;
+      }
+
+    if (*eoln_len)  // end of line was found
+      return x;
+  }
 
-  /*
-    if we only have one record in the file then our for loop will break
-    before we return.  we should still have seen end of line markers and
-    so we just return the line here
-  */
-  if (*eoln_len > 0)
-    return data+dataend;
   return 0;
 }
 
@@ -478,12 +505,13 @@ ha_tina::ha_tina(TABLE_SHARE *table_arg)
     They are not probably completely right.
   */
   current_position(0), next_position(0), local_saved_data_file_length(0),
-  chain_alloced(0), chain_size(DEFAULT_CHAIN_LENGTH),
+  file_buff(0), chain_alloced(0), chain_size(DEFAULT_CHAIN_LENGTH),
   records_is_known(0)
 {
   /* Set our original buffers from pre-allocated memory */
   buffer.set((char*)byte_buffer, IO_SIZE, system_charset_info);
   chain= chain_buffer;
+  file_buff= new Transparent_file();
 }
 
 
@@ -609,19 +637,17 @@ int ha_tina::chain_append()
 */
 int ha_tina::find_current_row(byte *buf)
 {
-  byte *mapped_ptr;
-  byte *end_ptr;
+  off_t end_offset, curr_offset= current_position;
   int eoln_len;
   my_bitmap_map *org_bitmap;
   DBUG_ENTER("ha_tina::find_current_row");
 
-  mapped_ptr= (byte *)share->mapped_file + current_position;
-
   /*
     We do not read further then local_saved_data_file_length in order
     not to conflict with undergoing concurrent insert.
   */
-  if ((end_ptr=  find_eoln(share->mapped_file, current_position,
+  if ((end_offset=
+        find_eoln_buff(file_buff, current_position,
                        local_saved_data_file_length, &eoln_len)) == 0)
     DBUG_RETURN(HA_ERR_END_OF_FILE);
 
@@ -631,36 +657,39 @@ int ha_tina::find_current_row(byte *buf)
   for (Field **field=table->field ; *field ; field++)
   {
     buffer.length(0);
-    if (*mapped_ptr == '"')
-      mapped_ptr++; // Increment past the first quote
+    if (file_buff->get_value(curr_offset) == '"')
+      curr_offset++; // Incrementpast the first quote
     else
     {
       dbug_tmp_restore_column_map(table->write_set, org_bitmap);
       DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
     }
-    for(;mapped_ptr != end_ptr; mapped_ptr++)
+    for(;curr_offset != end_offset; curr_offset++)
     {
       // Need to convert line feeds!
-      if (*mapped_ptr == '"' &&
-          (((mapped_ptr[1] == ',') && (mapped_ptr[2] == '"')) ||
-           (mapped_ptr == end_ptr -1 )))
+      if (file_buff->get_value(curr_offset) == '"' &&
+          (((file_buff->get_value(curr_offset + 1) == ',') &&
+            (file_buff->get_value(curr_offset + 2) == '"')) ||
+           (curr_offset == end_offset -1 )))
       {
-        mapped_ptr += 2; // Move past the , and the "
+        curr_offset+= 2; // Move past the , and the "
         break;
       }
-      if (*mapped_ptr == '\\' && mapped_ptr != (end_ptr - 1))
+      if (file_buff->get_value(curr_offset) == '\\' &&
+          curr_offset != (end_offset - 1))
       {
-        mapped_ptr++;
-        if (*mapped_ptr == 'r')
+        curr_offset++;
+        if (file_buff->get_value(curr_offset) == 'r')
           buffer.append('\r');
-        else if (*mapped_ptr == 'n' )
+        else if (file_buff->get_value(curr_offset) == 'n' )
           buffer.append('\n');
-        else if ((*mapped_ptr == '\\') || (*mapped_ptr == '"'))
-          buffer.append(*mapped_ptr);
+        else if ((file_buff->get_value(curr_offset) == '\\') ||
+                 (file_buff->get_value(curr_offset) == '"'))
+          buffer.append(file_buff->get_value(curr_offset));
         else  /* This could only happed with an externally created file */
         {
           buffer.append('\\');
-          buffer.append(*mapped_ptr);
+          buffer.append(file_buff->get_value(curr_offset));
         }
       }
       else // ordinary symbol
@@ -669,18 +698,18 @@ int ha_tina::find_current_row(byte *buf)
           We are at final symbol and no last quote was found =>
           we are working with a damaged file.
         */
-        if (mapped_ptr == end_ptr -1)
+        if (curr_offset == end_offset - 1)
         {
           dbug_tmp_restore_column_map(table->write_set, org_bitmap);
           DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
         }
-        buffer.append(*mapped_ptr);
+        buffer.append(file_buff->get_value(curr_offset));
       }
     }
     if (bitmap_is_set(table->read_set, (*field)->field_index))
       (*field)->store(buffer.ptr(), buffer.length(), system_charset_info);
   }
-  next_position= (end_ptr - share->mapped_file)+eoln_len;
+  next_position= end_offset + eoln_len;
   /* Maybe use \N for null? */
   memset(buf, 0, table->s->null_bytes); /* We do not implement nulls! */
   dbug_tmp_restore_column_map(table->write_set, org_bitmap);
@@ -779,7 +808,7 @@ void ha_tina::get_status()
 void ha_tina::update_status()
 {
   /* correct local_saved_data_file_length for writers */
-  share->saved_data_file_length= share->file_stat.st_size;
+  share->saved_data_file_length= local_saved_data_file_length;
 }
 
 
@@ -830,6 +859,9 @@ int ha_tina::open(const char *name, int mode, uint open_options)
     DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
   }
 
+  if ((data_file= my_open(share->data_file_name, O_RDONLY, MYF(0))) == -1)
+    DBUG_RETURN(0);
+
   /*
     Init locking. Pass handler object to the locking routines,
     so that they could save/update local_saved_data_file_length value
@@ -848,12 +880,14 @@ int ha_tina::open(const char *name, int mode, uint open_options)
 
 /*
   Close a database file. We remove ourselves from the shared strucutre.
-  If it is empty we destroy it and free the mapped file.
+  If it is empty we destroy it.
 */
 int ha_tina::close(void)
 {
+  int rc= 0;
   DBUG_ENTER("ha_tina::close");
-  DBUG_RETURN(free_share(share));
+  rc= my_close(data_file, MYF(0));
+  DBUG_RETURN(free_share(share) || rc);
 }
 
 /*
@@ -876,22 +910,17 @@ int ha_tina::write_row(byte * buf)
 
   size= encode_quote(buf);
 
-  if (my_write(share->data_file, (byte*)buffer.ptr(), size,
-               MYF(MY_WME | MY_NABP)))
+  if (!share->tina_write_opened)
+    if (init_tina_writer())
       DBUG_RETURN(-1);
 
-  /*
-    Ok, this is means that we will be doing potentially bad things
-    during a bulk insert on some OS'es. What we need is a cleanup
-    call for ::write_row that would let us fix up everything after the bulk
-    insert. The archive handler does this with an extra mutx call, which
-    might be a solution for this.
-  */
-  if (get_mmap(share, 0) > 0)
+   /* use pwrite, as concurrent reader could have changed the position */
+  if (my_write(share->tina_write_filedes, (byte*)buffer.ptr(), size,
+               MYF(MY_WME | MY_NABP)))
     DBUG_RETURN(-1);
 
   /* update local copy of the max position to see our own changes */
-  local_saved_data_file_length= share->file_stat.st_size;
+  local_saved_data_file_length+= size;
 
   /* update shared info */
   pthread_mutex_lock(&share->mutex);
@@ -906,6 +935,23 @@ int ha_tina::write_row(byte * buf)
 }
 
 
+int ha_tina::open_update_temp_file_if_needed()
+{
+  char updated_fname[FN_REFLEN];
+
+  if (!share->update_file_opened)
+  {
+    if ((update_temp_file=
+           my_create(fn_format(updated_fname, share->table_name,
+                               "", CSN_EXT,
+                               MY_REPLACE_EXT | MY_UNPACK_FILENAME),
+                     0, O_RDWR | O_TRUNC, MYF(MY_WME))) < 0)
+      return 1;
+    share->update_file_opened= TRUE;
+  }
+  return 0;
+}
+
 /*
   This is called for an update.
   Make sure you put in code to increment the auto increment, also
@@ -929,16 +975,16 @@ int ha_tina::update_row(const byte * old_data, byte * new_data)
   if (chain_append())
     DBUG_RETURN(-1);
 
-  if (my_write(share->data_file, (byte*)buffer.ptr(), size,
+  if (open_update_temp_file_if_needed())
+    DBUG_RETURN(-1);
+
+  if (my_write(update_temp_file, (byte*)buffer.ptr(), size,
                MYF(MY_WME | MY_NABP)))
     DBUG_RETURN(-1);
 
   /* UPDATE should never happen on the log tables */
   DBUG_ASSERT(!share->is_log_table);
 
-  /* update local copy of the max position to see our own changes */
-  local_saved_data_file_length= share->file_stat.st_size;
-
   DBUG_RETURN(0);
 }
 
@@ -1004,6 +1050,8 @@ int ha_tina::rnd_init(bool scan)
 {
   DBUG_ENTER("ha_tina::rnd_init");
 
+  /* set buffer to the beginning of the file */
+  file_buff->init_buff(data_file);
   if (share->crashed)
       DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
 
@@ -1011,11 +1059,6 @@ int ha_tina::rnd_init(bool scan)
   stats.records= 0;
   records_is_known= 0;
   chain_ptr= chain;
-#ifdef HAVE_MADVISE
-  if (scan)
-    (void) madvise(share->mapped_file, share->file_stat.st_size,
-                   MADV_SEQUENTIAL);
-#endif
 
   DBUG_RETURN(0);
 }
@@ -1045,8 +1088,11 @@ int ha_tina::rnd_next(byte *buf)
   ha_statistic_increment(&SSV::ha_read_rnd_next_count);
 
   current_position= next_position;
-  if (!share->mapped_file)
+
+  /* don't scan an empty file */
+  if (!local_saved_data_file_length)
     DBUG_RETURN(HA_ERR_END_OF_FILE);
+
   if ((rc= find_current_row(buf)))
     DBUG_RETURN(rc);
 
@@ -1115,6 +1161,22 @@ int ha_tina::extra(enum ha_extra_function operation)
   DBUG_RETURN(0);
 }
 
+/*
+  Set end_pos to the last valid byte of continuous area, closest
+  to the given "hole", stored in the buffer. "Valid" here means,
+  not listed in the chain of deleted records ("holes").
+*/
+bool ha_tina::get_write_pos(off_t *end_pos, tina_set *closest_hole)
+{
+  if (closest_hole == chain_ptr) /* no more chains */
+    *end_pos= file_buff->end();
+  else
+    *end_pos= min(file_buff->end(),
+                  closest_hole->begin);
+  return (closest_hole != chain_ptr) && (*end_pos == closest_hole->begin);
+}
+
+
 /*
   Called after each table scan. In particular after deletes,
   and updates. In the last case we employ chain of deleted
@@ -1123,53 +1185,107 @@ int ha_tina::extra(enum ha_extra_function operation)
 */
 int ha_tina::rnd_end()
 {
+  char updated_fname[FN_REFLEN];
+  off_t file_buffer_start= 0;
   DBUG_ENTER("ha_tina::rnd_end");
 
   records_is_known= 1;
 
-  /* First position will be truncate position, second will be increment */
   if ((chain_ptr - chain)  > 0)
   {
-    tina_set *ptr;
-    size_t length;
+    tina_set *ptr= chain;
 
     /*
-      Setting up writable map, this will contain all of the data after the
-      get_mmap call that we have added to the file.
+      Re-read the beginning of a file (as the buffer should point to the
+      end of file after the scan).
     */
-    if (get_mmap(share, 1) > 0)
-      DBUG_RETURN(-1);
-    length= share->file_stat.st_size;
+    file_buff->init_buff(data_file);
 
     /*
-      The sort handles updates/deletes with random orders.
-      It also sorts so that we move the final blocks to the
-      beginning so that we move the smallest amount of data possible.
+      The sort is needed when there were updates/deletes with random orders.
+      It sorts so that we move the firts blocks to the beginning.
     */
     qsort(chain, (size_t)(chain_ptr - chain), sizeof(tina_set),
           (qsort_cmp)sort_set);
-    for (ptr= chain; ptr < chain_ptr; ptr++)
+
+    off_t write_begin= 0, write_end;
+
+    /* create the file to write updated table if it wasn't yet created */
+    if (open_update_temp_file_if_needed())
+      DBUG_RETURN(-1);
+
+    /* write the file with updated info */
+    while ((file_buffer_start != -1))     // while not end of file
     {
-      memmove(share->mapped_file + ptr->begin, share->mapped_file + ptr->end,
-              length - (size_t)ptr->end);
-      length= length - (size_t)(ptr->end - ptr->begin);
+      bool in_hole= get_write_pos(&write_end, ptr);
+
+      /* if there is something to write, write it */
+      if ((write_end - write_begin) &&
+          (my_write(update_temp_file,
+                    (byte*)(file_buff->ptr() +
+                            (write_begin - file_buff->start())),
+                    write_end - write_begin, MYF_RW)))
+        goto error;
+
+      if (in_hole)
+      {
+        /* skip hole */
+        while (file_buff->end() <= ptr->end && file_buffer_start != -1)
+          file_buffer_start= file_buff->read_next();
+        write_begin= ptr->end;
+        ptr++;
+      }
+      else
+        write_begin= write_end;
+
+      if (write_end == file_buff->end())
+        file_buffer_start= file_buff->read_next(); /* shift the buffer */
+
     }
 
-    /* Unmap the file before the new size is set */
-    if (my_munmap(share->mapped_file, share->file_stat.st_size))
+    if (my_sync(update_temp_file, MYF(MY_WME)) ||
+        my_close(update_temp_file, MYF(0)))
       DBUG_RETURN(-1);
-    /* We set it to null so that get_mmap() won't try to unmap it */
-    share->mapped_file= NULL;
 
-    /* Set the file to the new size */
-    if (my_chsize(share->data_file, length, 0, MYF(MY_WME)))
+    share->update_file_opened= FALSE;
+
+    if (share->tina_write_opened)
+    {
+      if (my_close(share->tina_write_filedes, MYF(0)))
         DBUG_RETURN(-1);
+      /*
+        Mark that the writer fd is closed, so that init_tina_writer()
+        will reopen it later.
+      */
+      share->tina_write_opened= FALSE;
+    }
 
-    if (get_mmap(share, 0) > 0)
+    /*
+      Close opened fildes's. Then move updated file in place
+      of the old datafile.
+    */
+    if (my_close(data_file, MYF(0)) ||
+        my_rename(fn_format(updated_fname, share->table_name, "", CSN_EXT,
+                            MY_REPLACE_EXT | MY_UNPACK_FILENAME),
+                  share->data_file_name, MYF(0)))
       DBUG_RETURN(-1);
+
+    /* Open the file again */
+    if (((data_file= my_open(share->data_file_name, O_RDONLY, MYF(0))) == -1))
+      DBUG_RETURN(-1);
+    /*
+      The datafile is consistent at this point and the write filedes is
+      closed, so nothing worrying will happen to it in case of a crash.
+      Here we record this fact to the meta-file.
+    */
+    (void)write_meta_file(share->meta_file, share->rows_recorded, FALSE);
   }
 
   DBUG_RETURN(0);
+error:
+  my_close(update_temp_file, MYF(0));
+  share->update_file_opened= FALSE;
+  DBUG_RETURN(-1);
 }
 
 
@@ -1198,10 +1314,11 @@ int ha_tina::repair(THD* thd, HA_CHECK_OPT* check_opt)
   File repair_file;
   int rc;
   ha_rows rows_repaired= 0;
+  off_t write_begin= 0, write_end;
   DBUG_ENTER("ha_tina::repair");
 
   /* empty file */
-  if (!share->mapped_file)
+  if (!share->saved_data_file_length)
   {
     share->rows_recorded= 0;
     goto end;
@@ -1212,12 +1329,15 @@ int ha_tina::repair(THD* thd, HA_CHECK_OPT* check_opt)
   if (!(buf= (byte*) my_malloc(table->s->reclength, MYF(MY_WME))))
     DBUG_RETURN(HA_ERR_OUT_OF_MEM);
 
+  /* position buffer to the start of the file */
+  file_buff->init_buff(data_file);
+
   /*
     Local_saved_data_file_length is initialized during the lock phase.
     Sometimes this is not getting executed before ::repair (e.g. for
     the log tables). We set it manually here.
   */
-  local_saved_data_file_length= share->file_stat.st_size;
+  local_saved_data_file_length= share->saved_data_file_length;
   /* set current position to the beginning of the file */
   current_position= next_position= 0;
 
@@ -1232,11 +1352,10 @@ int ha_tina::repair(THD* thd, HA_CHECK_OPT* check_opt)
 
   if (rc == HA_ERR_END_OF_FILE)
   {
-    /* All rows were read ok until end of file, the file does not need repair. */
-
     /*
-      If rows_recorded != rows_repaired, we should update
-      rows_recorded value to the current amount of rows.
+      All rows were read ok until end of file, the file does not need repair.
+      If rows_recorded != rows_repaired, we should update rows_recorded value
+      to the current amount of rows.
     */
     share->rows_recorded= rows_repaired;
     goto end;
@@ -1252,36 +1371,45 @@ int ha_tina::repair(THD* thd, HA_CHECK_OPT* check_opt)
                            0, O_RDWR | O_TRUNC,MYF(MY_WME))) < 0)
     DBUG_RETURN(HA_ERR_CRASHED_ON_REPAIR);
 
-  if (my_write(repair_file, (byte*)share->mapped_file, current_position,
-               MYF(MY_NABP)))
-    DBUG_RETURN(HA_ERR_CRASHED_ON_USAGE);
-  my_close(repair_file, MYF(0));
+  file_buff->init_buff(data_file);
+
+
   /* we just truncated the file up to the first bad row. update rows count. */
   share->rows_recorded= rows_repaired;
 
-  if (my_munmap(share->mapped_file, share->file_stat.st_size))
+  /* write repaired file */
+  while (1)
+  {
+    write_end= min(file_buff->end(), current_position);
+    if ((write_end - write_begin) &&
+        (my_write(repair_file, (byte*)file_buff->ptr(),
+                  write_end - write_begin, MYF_RW)))
       DBUG_RETURN(-1);
-  /* We set it to null so that get_mmap() won't try to unmap it */
-  share->mapped_file= NULL;
+
+    write_begin= write_end;
+    if (write_end== current_position)
+      break;
+    else
+      file_buff->read_next(); /* shift the buffer */
+  }
 
   /*
-    Close the "to"-file before renaming
-    On Windows one cannot rename a file, which descriptor
-    is still open. EACCES will be returned when trying to delete 
-    the "to"-file in my_rename()
+    Close the files and rename repaired file to the datafile.
+    We have to close the files, as on Windows one cannot rename
+    a file, which descriptor is still open. EACCES will be returned
+    when trying to delete the "to"-file in my_rename().
   */
-  my_close(share->data_file,MYF(0));
-
-  if (my_rename(repaired_fname, share->data_file_name, MYF(0)))
+  if (my_close(data_file,MYF(0)) || my_close(repair_file, MYF(0)) ||
+      my_rename(repaired_fname, share->data_file_name, MYF(0)))
     DBUG_RETURN(-1);
 
   /* Open the file again, it should now be repaired */
-  if ((share->data_file= my_open(share->data_file_name, O_RDWR|O_APPEND,
+  if ((data_file= my_open(share->data_file_name, O_RDWR|O_APPEND,
                           MYF(0))) == -1)
      DBUG_RETURN(-1);
 
-  if (get_mmap(share, 0) > 0)
-    DBUG_RETURN(-1);
+  /* Set new file size. The file size will be updated by ::update_status() */
+  local_saved_data_file_length= (size_t) current_position;
 
 end:
   share->crashed= FALSE;
@@ -1300,17 +1428,12 @@ int ha_tina::delete_all_rows()
   if (!records_is_known)
     DBUG_RETURN(my_errno=HA_ERR_WRONG_COMMAND);
 
-  /* Unmap the file before the new size is set */
-  if (share->mapped_file && my_munmap(share->mapped_file,
-                                      share->file_stat.st_size))
+  if (!share->tina_write_opened)
+    if (init_tina_writer())
       DBUG_RETURN(-1);
-  share->mapped_file= NULL;
 
   /* Truncate the file to zero size */
-  rc= my_chsize(share->data_file, 0, 0, MYF(MY_WME));
-
-  if (get_mmap(share, 0) > 0)
-    DBUG_RETURN(-1);
+  rc= my_chsize(share->tina_write_filedes, 0, 0, MYF(MY_WME));
 
   stats.records=0;
   DBUG_RETURN(rc);
@@ -1372,12 +1495,15 @@ int ha_tina::check(THD* thd, HA_CHECK_OPT* check_opt)
   if (!(buf= (byte*) my_malloc(table->s->reclength, MYF(MY_WME))))
     DBUG_RETURN(HA_ERR_OUT_OF_MEM);
 
+  /* position buffer to the start of the file */
+  file_buff->init_buff(data_file);
+
   /*
     Local_saved_data_file_length is initialized during the lock phase.
     Check does not use store_lock in certain cases. So, we set it
     manually here.
   */
-  local_saved_data_file_length= share->file_stat.st_size;
+  local_saved_data_file_length= share->saved_data_file_length;
   /* set current position to the beginning of the file */
   current_position= next_position= 0;
   /* Read the file row-by-row. If everything is ok, repair is not needed. */

storage/csv/ha_tina.h

@@ -29,15 +29,12 @@
 typedef struct st_tina_share {
   char *table_name;
   char data_file_name[FN_REFLEN];
-  byte *mapped_file;                /* mapped region of file */
   uint table_name_length, use_count;
   /*
     Below flag is needed to make log tables work with concurrent insert.
     For more details see comment to ha_tina::update_status.
   */
   my_bool is_log_table;
-  MY_STAT file_stat;                /* Stat information for the data file */
-  File data_file;                   /* Current open data file */
   /*
     Here we save the length of the file for readers. This is updated by
     inserts, updates and deletes. The var is initialized along with the
@@ -46,7 +43,10 @@ typedef struct st_tina_share {
   off_t saved_data_file_length;
   pthread_mutex_t mutex;
   THR_LOCK lock;
+  bool update_file_opened;
+  bool tina_write_opened;
   File meta_file;           /* Meta file we use */
+  File tina_write_filedes;  /* File handler for readers */
   bool crashed;             /* Meta file is crashed */
   ha_rows rows_recorded;    /* Number of rows in tables */
 } TINA_SHARE;
@@ -56,6 +56,49 @@ struct tina_set {
   off_t end;
 };
 
+class Transparent_file
+{
+   File filedes;
+   byte *buff;  /* in-memory window to the file or mmaped area */
+   /* current window sizes */
+   off_t lower_bound;
+   off_t upper_bound;
+   uint buff_size;
+
+ public:
+
+   Transparent_file() : lower_bound(0), buff_size(IO_SIZE)
+   { buff= (byte *) my_malloc(buff_size*sizeof(byte),  MYF(MY_WME)); }
+
+   ~Transparent_file()
+   { my_free((gptr)buff, MYF(MY_ALLOW_ZERO_PTR)); }
+
+   void init_buff(File filedes_arg)
+   {
+     filedes= filedes_arg;
+     /* read the beginning of the file */
+     lower_bound= 0;
+     VOID(my_seek(filedes, 0, MY_SEEK_SET, MYF(0)));
+     if (filedes && buff)
+       upper_bound= my_read(filedes, buff, buff_size, MYF(0));
+   }
+
+   byte *ptr()
+   { return buff; }
+
+   off_t start()
+   { return lower_bound; }
+
+   off_t end()
+   { return upper_bound; }
+
+   /* get a char from the given position in the file */
+   char get_value (off_t offset);
+   /* shift a buffer windows to see the next part of the file */
+   off_t read_next();
+
+};
+
 class ha_tina: public handler
 {
   THR_LOCK_DATA lock;      /* MySQL lock */
@@ -64,6 +107,9 @@ class ha_tina: public handler
   off_t next_position;     /* Next position in the file scan */
   off_t local_saved_data_file_length; /* save position for reads */
   byte byte_buffer[IO_SIZE];
+  Transparent_file *file_buff;
+  File data_file;                   /* File handler for readers */
+  File update_temp_file;
   String buffer;
   /*
     The chain contains "holes" in the file, occured because of
@@ -77,12 +123,19 @@ class ha_tina: public handler
   uint32 chain_size;
   bool records_is_known;
 
+private:
+  bool get_write_pos(off_t *end_pos, tina_set *closest_hole);
+  int open_update_temp_file_if_needed();
+  int init_tina_writer();
+
 public:
   ha_tina(TABLE_SHARE *table_arg);
   ~ha_tina()
   {
     if (chain_alloced)
-      my_free((gptr)chain,0);
+      my_free((gptr)chain, 0);
+    if (file_buff)
+      delete file_buff;
   }
   const char *table_type() const { return "CSV"; }
   const char *index_type(uint inx) { return "NONE"; }