log.cc 146 KB
Newer Older
1
/* Copyright (C) 2000-2003 MySQL AB
serg@serg.mylan's avatar
serg@serg.mylan committed
2

bk@work.mysql.com's avatar
bk@work.mysql.com committed
3 4
   This program is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
5
   the Free Software Foundation; version 2 of the License.
serg@serg.mylan's avatar
serg@serg.mylan committed
6

bk@work.mysql.com's avatar
bk@work.mysql.com committed
7 8 9 10
   This program is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.
serg@serg.mylan's avatar
serg@serg.mylan committed
11

bk@work.mysql.com's avatar
bk@work.mysql.com committed
12 13 14 15 16
   You should have received a copy of the GNU General Public License
   along with this program; if not, write to the Free Software
   Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA */


17 18 19 20 21 22 23 24 25
/**
  @file

  @brief
  logging of commands

  @todo
    Abort logging when we get an error in reading or writing log files
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
26 27

#include "mysql_priv.h"
28
#include "sql_repl.h"
lars@mysql.com's avatar
lars@mysql.com committed
29
#include "rpl_filter.h"
30
#include "rpl_rli.h"
bk@work.mysql.com's avatar
bk@work.mysql.com committed
31 32 33 34 35

#include <my_dir.h>
#include <stdarg.h>
#include <m_ctype.h>				// For test_if_number

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
36 37 38 39
#ifdef __NT__
#include "message.h"
#endif

acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
40 41
#include <mysql/plugin.h>

42 43 44 45
/* max size of the log message */
#define MAX_LOG_BUFFER_SIZE 1024
#define MAX_USER_HOST_SIZE 512
#define MAX_TIME_SIZE 32
46
#define MY_OFF_T_UNDEF (~(my_off_t)0UL)
47

48
#define FLAGSTR(V,F) ((V)&(F)?#F" ":"")
49 50 51

LOGGER logger;

52
MYSQL_BIN_LOG mysql_bin_log;
53
ulong sync_binlog_counter= 0;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
54

55 56
static Muted_query_log_event invisible_commit;

bk@work.mysql.com's avatar
bk@work.mysql.com committed
57 58
static bool test_if_number(const char *str,
			   long *res, bool allow_wildcards);
59 60 61 62 63 64 65
static int binlog_init(void *p);
static int binlog_close_connection(handlerton *hton, THD *thd);
static int binlog_savepoint_set(handlerton *hton, THD *thd, void *sv);
static int binlog_savepoint_rollback(handlerton *hton, THD *thd, void *sv);
static int binlog_commit(handlerton *hton, THD *thd, bool all);
static int binlog_rollback(handlerton *hton, THD *thd, bool all);
static int binlog_prepare(handlerton *hton, THD *thd, bool all);
66

67 68 69 70 71 72 73 74 75
/**
  Silence all errors and warnings reported when performing a write
  to a log table.
  Errors and warnings are not reported to the client or SQL exception
  handlers, so that the presence of logging does not interfere and affect
  the logic of an application.
*/
class Silence_log_table_errors : public Internal_error_handler
{
76
  char m_message[MYSQL_ERRMSG_SIZE];
77 78
public:
  Silence_log_table_errors()
79 80 81
  {
    m_message[0]= '\0';
  }
82 83 84

  virtual ~Silence_log_table_errors() {}

85
  virtual bool handle_error(uint sql_errno, const char *message,
86 87
                            MYSQL_ERROR::enum_warning_level level,
                            THD *thd);
88
  const char *message() const { return m_message; }
89 90 91 92
};

bool
Silence_log_table_errors::handle_error(uint /* sql_errno */,
93
                                       const char *message_arg,
94 95 96
                                       MYSQL_ERROR::enum_warning_level /* level */,
                                       THD * /* thd */)
{
97
  strmake(m_message, message_arg, sizeof(m_message)-1);
98 99 100 101
  return TRUE;
}


102 103 104 105 106 107 108 109
sql_print_message_func sql_print_message_handlers[3] =
{
  sql_print_information,
  sql_print_warning,
  sql_print_error
};


110 111
char *make_default_log_name(char *buff,const char* log_ext)
{
112
  strmake(buff, pidfile_name, FN_REFLEN-5);
113
  return fn_format(buff, buff, mysql_data_home, log_ext,
114
                   MYF(MY_UNPACK_FILENAME|MY_REPLACE_EXT));
115 116
}

117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151
/*
  Helper class to hold a mutex for the duration of the
  block.

  Eliminates the need for explicit unlocking of mutexes on, e.g.,
  error returns.  On passing a null pointer, the sentry will not do
  anything.
 */
class Mutex_sentry
{
public:
  Mutex_sentry(pthread_mutex_t *mutex)
    : m_mutex(mutex)
  {
    if (m_mutex)
      pthread_mutex_lock(mutex);
  }

  ~Mutex_sentry()
  {
    if (m_mutex)
      pthread_mutex_unlock(m_mutex);
#ifndef DBUG_OFF
    m_mutex= 0;
#endif
  }

private:
  pthread_mutex_t *m_mutex;

  // It's not allowed to copy this object in any way
  Mutex_sentry(Mutex_sentry const&);
  void operator=(Mutex_sentry const&);
};

152
/*
153
  Helper class to store binary log transaction data.
154
*/
155 156 157
class binlog_trx_data {
public:
  binlog_trx_data()
158
    : m_pending(0), before_stmt_pos(MY_OFF_T_UNDEF)
159 160 161 162 163 164 165 166 167 168 169 170 171 172
  {
    trans_log.end_of_file= max_binlog_cache_size;
  }

  ~binlog_trx_data()
  {
    DBUG_ASSERT(pending() == NULL);
    close_cached_file(&trans_log);
  }

  my_off_t position() const {
    return my_b_tell(&trans_log);
  }

173 174
  bool empty() const
  {
175
    return pending() == NULL && my_b_tell(&trans_log) == 0;
176
  }
177 178 179 180 181 182 183

  /*
    Truncate the transaction cache to a certain position. This
    includes deleting the pending event.
   */
  void truncate(my_off_t pos)
  {
184 185
    DBUG_PRINT("info", ("truncating to position %lu", (ulong) pos));
    DBUG_PRINT("info", ("before_stmt_pos=%lu", (ulong) pos));
186 187 188
    delete pending();
    set_pending(0);
    reinit_io_cache(&trans_log, WRITE_CACHE, pos, 0, 0);
189 190
    if (pos < before_stmt_pos)
      before_stmt_pos= MY_OFF_T_UNDEF;
191 192 193 194 195 196 197 198 199
  }

  /*
    Reset the entire contents of the transaction cache, emptying it
    completely.
   */
  void reset() {
    if (!empty())
      truncate(0);
200
    before_stmt_pos= MY_OFF_T_UNDEF;
201 202 203 204 205 206 207 208 209 210 211 212 213
    trans_log.end_of_file= max_binlog_cache_size;
  }

  Rows_log_event *pending() const
  {
    return m_pending;
  }

  void set_pending(Rows_log_event *const pending)
  {
    m_pending= pending;
  }

214
  IO_CACHE trans_log;                         // The transaction cache
215 216 217 218 219 220 221 222 223 224 225 226 227

private:
  /*
    Pending binrows event. This event is the event where the rows are
    currently written.
   */
  Rows_log_event *m_pending;

public:
  /*
    Binlog position before the start of the current statement.
  */
  my_off_t before_stmt_pos;
228 229
};

230
handlerton *binlog_hton;
231

232 233 234 235 236 237 238 239 240 241 242 243 244
bool LOGGER::is_log_table_enabled(uint log_table_type)
{
  switch (log_table_type) {
  case QUERY_LOG_SLOW:
    return (table_log_handler != NULL) && opt_slow_log;
  case QUERY_LOG_GENERAL:
    return (table_log_handler != NULL) && opt_log ;
  default:
    DBUG_ASSERT(0);
    return FALSE;                             /* make compiler happy */
  }
}

245

246 247 248 249 250 251 252 253 254 255 256 257
/* Check if a given table is opened log table */
int check_if_log_table(uint db_len, const char *db, uint table_name_len,
                       const char *table_name, uint check_if_opened)
{
  if (db_len == 5 &&
      !(lower_case_table_names ?
        my_strcasecmp(system_charset_info, db, "mysql") :
        strcmp(db, "mysql")))
  {
    if (table_name_len == 11 && !(lower_case_table_names ?
                                  my_strcasecmp(system_charset_info,
                                                table_name, "general_log") :
258 259 260 261 262 263 264 265 266 267 268 269
                                  strcmp(table_name, "general_log")))
    {
      if (!check_if_opened || logger.is_log_table_enabled(QUERY_LOG_GENERAL))
        return QUERY_LOG_GENERAL;
      return 0;
    }

    if (table_name_len == 8 && !(lower_case_table_names ?
      my_strcasecmp(system_charset_info, table_name, "slow_log") :
      strcmp(table_name, "slow_log")))
    {
      if (!check_if_opened || logger.is_log_table_enabled(QUERY_LOG_SLOW))
270
        return QUERY_LOG_SLOW;
271 272
      return 0;
    }
273 274 275 276
  }
  return 0;
}

277

278 279 280 281 282 283 284 285 286
Log_to_csv_event_handler::Log_to_csv_event_handler()
{
}


Log_to_csv_event_handler::~Log_to_csv_event_handler()
{
}

287

288 289 290 291 292 293 294
void Log_to_csv_event_handler::cleanup()
{
  logger.is_log_tables_initialized= FALSE;
}

/* log event handlers */

295
/**
296 297
  Log command to the general log table

298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322
  Log given command to the general log table.

  @param  event_time        command start timestamp
  @param  user_host         the pointer to the string with user@host info
  @param  user_host_len     length of the user_host string. this is computed
                            once and passed to all general log event handlers
  @param  thread_id         Id of the thread, issued a query
  @param  command_type      the type of the command being logged
  @param  command_type_len  the length of the string above
  @param  sql_text          the very text of the query being executed
  @param  sql_text_len      the length of sql_text string


  @return This function attempts to never call my_error(). This is
  necessary, because general logging happens already after a statement
  status has been sent to the client, so the client can not see the
  error anyway. Besides, the error is not related to the statement
  being executed and is internal, and thus should be handled
  internally (@todo: how?).
  If a write to the table has failed, the function attempts to
  write to a short error message to the file. The failure is also
  indicated in the return value. 

  @retval  FALSE   OK
  @retval  TRUE    error occured
323 324 325
*/

bool Log_to_csv_event_handler::
326
  log_general(THD *thd, time_t event_time, const char *user_host,
327 328
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
329 330
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
331
{
332 333 334 335 336 337
  TABLE_LIST table_list;
  TABLE *table;
  bool result= TRUE;
  bool need_close= FALSE;
  bool need_pop= FALSE;
  bool need_rnd_end= FALSE;
malff/marcsql@weblab.(none)'s avatar
malff/marcsql@weblab.(none) committed
338
  uint field_index;
339 340 341 342 343
  Silence_log_table_errors error_handler;
  Open_tables_state open_tables_backup;
  ulonglong save_thd_options;
  bool save_time_zone_used;

344 345 346 347 348
  /*
    CSV uses TIME_to_timestamp() internally if table needs to be repaired
    which will set thd->time_zone_used
  */
  save_time_zone_used= thd->time_zone_used;
349 350 351 352 353 354 355 356 357 358 359 360 361

  save_thd_options= thd->options;
  thd->options&= ~OPTION_BIN_LOG;

  bzero(& table_list, sizeof(TABLE_LIST));
  table_list.alias= table_list.table_name= GENERAL_LOG_NAME.str;
  table_list.table_name_length= GENERAL_LOG_NAME.length;

  table_list.lock_type= TL_WRITE_CONCURRENT_INSERT;

  table_list.db= MYSQL_SCHEMA_NAME.str;
  table_list.db_length= MYSQL_SCHEMA_NAME.length;

362 363 364 365 366 367 368 369 370 371 372 373 374 375
  /*
    1) open_performance_schema_table generates an error of the
    table can not be opened or is corrupted.
    2) "INSERT INTO general_log" can generate warning sometimes.

    Suppress these warnings and errors, they can't be dealt with
    properly anyway.

    QQ: this problem needs to be studied in more detail.
    Comment this 2 lines and run "cast.test" to see what's happening.
  */
  thd->push_internal_handler(& error_handler);
  need_pop= TRUE;

376 377 378 379
  if (!(table= open_performance_schema_table(thd, & table_list,
                                             & open_tables_backup)))
    goto err;

380 381
  need_close= TRUE;

382
  if (table->file->extra(HA_EXTRA_MARK_AS_LOG_TABLE) ||
383 384 385 386 387 388 389
      table->file->ha_rnd_init(0))
    goto err;

  need_rnd_end= TRUE;

  /* Honor next number columns if present */
  table->next_number_field= table->found_next_number_field;
390

391 392 393 394 395 396 397 398 399
  /*
    NOTE: we do not call restore_record() here, as all fields are
    filled by the Logger (=> no need to load default ones).
  */

  /*
    We do not set a value for table->field[0], as it will use
    default value (which is CURRENT_TIMESTAMP).
  */
400

401
  /* check that all columns exist */
402
  if (table->s->fields < 6)
403 404
    goto err;

405 406
  DBUG_ASSERT(table->field[0]->type() == MYSQL_TYPE_TIMESTAMP);

407 408
  ((Field_timestamp*) table->field[0])->store_timestamp((my_time_t)
                                                        event_time);
409

410 411 412 413
  /* do a write */
  if (table->field[1]->store(user_host, user_host_len, client_cs) ||
      table->field[2]->store((longlong) thread_id, TRUE) ||
      table->field[3]->store((longlong) server_id, TRUE) ||
414
      table->field[4]->store(command_type, command_type_len, client_cs))
415 416
    goto err;

417 418 419 420 421 422
  /*
    A positive return value in store() means truncation.
    Still logging a message in the log in this case.
  */
  if (table->field[5]->store(sql_text, sql_text_len, client_cs) < 0)
    goto err;
423 424

  /* mark all fields as not null */
425 426 427 428 429
  table->field[1]->set_notnull();
  table->field[2]->set_notnull();
  table->field[3]->set_notnull();
  table->field[4]->set_notnull();
  table->field[5]->set_notnull();
430

431 432 433 434 435 436
  /* Set any extra columns to their default values */
  for (field_index= 6 ; field_index < table->s->fields ; field_index++)
  {
    table->field[field_index]->set_default();
  }

437
  /* log table entries are not replicated */
438
  if (table->file->ha_write_row(table->record[0]))
439
    goto err;
440

441
  result= FALSE;
442

443
err:
444
  if (result && !thd->killed)
445 446
    sql_print_error("Failed to write to mysql.general_log: %s",
                    error_handler.message());
447

448 449 450 451 452 453 454 455 456 457 458 459 460
  if (need_rnd_end)
  {
    table->file->ha_rnd_end();
    table->file->ha_release_auto_increment();
  }
  if (need_pop)
    thd->pop_internal_handler();
  if (need_close)
    close_performance_schema_table(thd, & open_tables_backup);

  thd->options= save_thd_options;
  thd->time_zone_used= save_time_zone_used;
  return result;
461 462 463 464 465 466 467
}


/*
  Log a query to the slow log table

  SYNOPSIS
468
    log_slow()
469 470 471 472 473 474
    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
475 476
    query_time        Amount of time the query took to execute (in microseconds)
    lock_time         Amount of time the query was locked (in microseconds)
477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495
    is_command        The flag, which determines, whether the sql_text is a
                      query or an administrator command (these are treated
                      differently by the old logging routines)
    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 table

  RETURN
    FALSE - OK
    TRUE - error occured
*/

bool Log_to_csv_event_handler::
  log_slow(THD *thd, time_t current_time, time_t query_start_arg,
           const char *user_host, uint user_host_len,
496
           ulonglong query_utime, ulonglong lock_utime, bool is_command,
497 498
           const char *sql_text, uint sql_text_len)
{
499 500 501 502 503
  TABLE_LIST table_list;
  TABLE *table;
  bool result= TRUE;
  bool need_close= FALSE;
  bool need_rnd_end= FALSE;
504
  Silence_log_table_errors error_handler;
505
  Open_tables_state open_tables_backup;
506
  CHARSET_INFO *client_cs= thd->variables.character_set_client;
507
  bool save_time_zone_used;
508
  DBUG_ENTER("Log_to_csv_event_handler::log_slow");
509

510
  thd->push_internal_handler(& error_handler);
511 512 513 514 515 516
  /*
    CSV uses TIME_to_timestamp() internally if table needs to be repaired
    which will set thd->time_zone_used
  */
  save_time_zone_used= thd->time_zone_used;

517 518 519 520 521 522 523 524 525
  bzero(& table_list, sizeof(TABLE_LIST));
  table_list.alias= table_list.table_name= SLOW_LOG_NAME.str;
  table_list.table_name_length= SLOW_LOG_NAME.length;

  table_list.lock_type= TL_WRITE_CONCURRENT_INSERT;

  table_list.db= MYSQL_SCHEMA_NAME.str;
  table_list.db_length= MYSQL_SCHEMA_NAME.length;

526 527 528
  if (!(table= open_performance_schema_table(thd, & table_list,
                                             & open_tables_backup)))
    goto err;
529 530 531

  need_close= TRUE;

532
  if (table->file->extra(HA_EXTRA_MARK_AS_LOG_TABLE) ||
533 534 535 536 537 538 539
      table->file->ha_rnd_init(0))
    goto err;

  need_rnd_end= TRUE;

  /* Honor next number columns if present */
  table->next_number_field= table->found_next_number_field;
540

541
  restore_record(table, s->default_values);    // Get empty record
542

543 544 545 546
  /* check that all columns exist */
  if (table->s->fields < 11)
    goto err;

547 548
  /* store the time and user values */
  DBUG_ASSERT(table->field[0]->type() == MYSQL_TYPE_TIMESTAMP);
549 550
  ((Field_timestamp*) table->field[0])->store_timestamp((my_time_t)
                                                        current_time);
551 552
  if (table->field[1]->store(user_host, user_host_len, client_cs))
    goto err;
553 554 555

  if (query_start_arg)
  {
556 557
    longlong query_time= (longlong) (query_utime/1000000);
    longlong lock_time=  (longlong) (lock_utime/1000000);
558 559 560 561 562
    /*
      A TIME field can not hold the full longlong range; query_time or
      lock_time may be truncated without warning here, if greater than
      839 hours (~35 days)
    */
563
    MYSQL_TIME t;
564 565
    t.neg= 0;

566
    /* fill in query_time field */
567 568
    calc_time_from_sec(&t, (long) min(query_time, (longlong) TIME_MAX_VALUE_SECONDS), 0);
    if (table->field[2]->store_time(&t, MYSQL_TIMESTAMP_TIME))
569
      goto err;
570
    /* lock_time */
571 572
    calc_time_from_sec(&t, (long) min(lock_time, (longlong) TIME_MAX_VALUE_SECONDS), 0);
    if (table->field[3]->store_time(&t, MYSQL_TIMESTAMP_TIME))
573
      goto err;
574
    /* rows_sent */
575 576
    if (table->field[4]->store((longlong) thd->sent_row_count, TRUE))
      goto err;
577
    /* rows_examined */
578 579
    if (table->field[5]->store((longlong) thd->examined_row_count, TRUE))
      goto err;
580 581 582 583 584 585 586 587
  }
  else
  {
    table->field[2]->set_null();
    table->field[3]->set_null();
    table->field[4]->set_null();
    table->field[5]->set_null();
  }
588
  /* fill database field */
589
  if (thd->db)
590
  {
591 592
    if (table->field[6]->store(thd->db, thd->db_length, client_cs))
      goto err;
593 594
    table->field[6]->set_notnull();
  }
595

596
  if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
597
  {
598 599 600 601 602
    if (table->
        field[7]->store((longlong)
                        thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                        TRUE))
      goto err;
603 604
    table->field[7]->set_notnull();
  }
605

606 607 608 609 610 611 612
  /*
    Set value if we do an insert on autoincrement column. Note that for
    some engines (those for which get_auto_increment() does not leave a
    table lock until the statement ends), this is just the first value and
    the next ones used may not be contiguous to it.
  */
  if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
613
  {
614 615 616 617
    if (table->
        field[8]->store((longlong)
          thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(), TRUE))
      goto err;
618 619
    table->field[8]->set_notnull();
  }
620

621 622
  if (table->field[9]->store((longlong) server_id, TRUE))
    goto err;
623
  table->field[9]->set_notnull();
624

625 626 627 628 629 630
  /*
    Column sql_text.
    A positive return value in store() means truncation.
    Still logging a message in the log in this case.
  */
  if (table->field[10]->store(sql_text, sql_text_len, client_cs) < 0)
631 632
    goto err;

633
  /* log table entries are not replicated */
634
  if (table->file->ha_write_row(table->record[0]))
635
    goto err;
636

637
  result= FALSE;
638

639
err:
640 641
  thd->pop_internal_handler();

642
  if (result && !thd->killed)
643 644
    sql_print_error("Failed to write to mysql.slow_log: %s",
                    error_handler.message());
645

646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687
  if (need_rnd_end)
  {
    table->file->ha_rnd_end();
    table->file->ha_release_auto_increment();
  }
  if (need_close)
    close_performance_schema_table(thd, & open_tables_backup);
  thd->time_zone_used= save_time_zone_used;
  DBUG_RETURN(result);
}

int Log_to_csv_event_handler::
  activate_log(THD *thd, uint log_table_type)
{
  TABLE_LIST table_list;
  TABLE *table;
  int result;
  Open_tables_state open_tables_backup;

  DBUG_ENTER("Log_to_csv_event_handler::activate_log");

  bzero(& table_list, sizeof(TABLE_LIST));

  if (log_table_type == QUERY_LOG_GENERAL)
  {
    table_list.alias= table_list.table_name= GENERAL_LOG_NAME.str;
    table_list.table_name_length= GENERAL_LOG_NAME.length;
  }
  else
  {
    DBUG_ASSERT(log_table_type == QUERY_LOG_SLOW);
    table_list.alias= table_list.table_name= SLOW_LOG_NAME.str;
    table_list.table_name_length= SLOW_LOG_NAME.length;
  }

  table_list.lock_type= TL_WRITE_CONCURRENT_INSERT;

  table_list.db= MYSQL_SCHEMA_NAME.str;
  table_list.db_length= MYSQL_SCHEMA_NAME.length;

  table= open_performance_schema_table(thd, & table_list,
                                       & open_tables_backup);
688 689 690 691 692 693 694 695
  if (table)
  {
    result= 0;
    close_performance_schema_table(thd, & open_tables_backup);
  }
  else
    result= 1;

696
  DBUG_RETURN(result);
697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720
}

bool Log_to_csv_event_handler::
  log_error(enum loglevel level, const char *format, va_list args)
{
  /* No log table is implemented */
  DBUG_ASSERT(0);
  return FALSE;
}

bool Log_to_file_event_handler::
  log_error(enum loglevel level, const char *format,
            va_list args)
{
  return vprint_msg_to_log(level, format, args);
}

void Log_to_file_event_handler::init_pthread_objects()
{
  mysql_log.init_pthread_objects();
  mysql_slow_log.init_pthread_objects();
}


721
/** Wrapper around MYSQL_LOG::write() for slow log. */
722 723 724 725

bool Log_to_file_event_handler::
  log_slow(THD *thd, time_t current_time, time_t query_start_arg,
           const char *user_host, uint user_host_len,
726
           ulonglong query_utime, ulonglong lock_utime, bool is_command,
727 728
           const char *sql_text, uint sql_text_len)
{
729 730
  return mysql_slow_log.write(thd, current_time, query_start_arg,
                              user_host, user_host_len,
731
                              query_utime, lock_utime, is_command,
732
                              sql_text, sql_text_len);
733 734 735
}


736
/**
737 738 739 740 741
   Wrapper around MYSQL_LOG::write() for general log. We need it since we
   want all log event handlers to have the same signature.
*/

bool Log_to_file_event_handler::
742
  log_general(THD *thd, time_t event_time, const char *user_host,
743 744
              uint user_host_len, int thread_id,
              const char *command_type, uint command_type_len,
745 746
              const char *sql_text, uint sql_text_len,
              CHARSET_INFO *client_cs)
747
{
748 749 750
  return mysql_log.write(event_time, user_host, user_host_len,
                         thread_id, command_type, command_type_len,
                         sql_text, sql_text_len);
751 752 753 754 755 756 757 758
}


bool Log_to_file_event_handler::init()
{
  if (!is_initialized)
  {
    if (opt_slow_log)
759
      mysql_slow_log.open_slow_log(sys_var_slow_log_path.value);
760 761

    if (opt_log)
762
      mysql_log.open_query_log(sys_var_general_log_path.value);
763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779

    is_initialized= TRUE;
  }

  return FALSE;
}


void Log_to_file_event_handler::cleanup()
{
  mysql_log.cleanup();
  mysql_slow_log.cleanup();
}

void Log_to_file_event_handler::flush()
{
  /* reopen log files */
780
  if (opt_log)
781
    mysql_log.reopen_file();
782
  if (opt_slow_log)
783
    mysql_slow_log.reopen_file();
784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805
}

/*
  Log error with all enabled log event handlers

  SYNOPSIS
    error_log_print()

    level             The level of the error significance: NOTE,
                      WARNING or ERROR.
    format            format string for the error message
    args              list of arguments for the format string

  RETURN
    FALSE - OK
    TRUE - error occured
*/

bool LOGGER::error_log_print(enum loglevel level, const char *format,
                             va_list args)
{
  bool error= FALSE;
806
  Log_event_handler **current_handler;
807 808

  /* currently we don't need locking here as there is no error_log table */
809
  for (current_handler= error_log_handler_list ; *current_handler ;)
810 811 812 813 814 815
    error= (*current_handler++)->log_error(level, format, args) || error;

  return error;
}


petr@mysql.com's avatar
petr@mysql.com committed
816
void LOGGER::cleanup_base()
817 818
{
  DBUG_ASSERT(inited == 1);
819
  rwlock_destroy(&LOCK_logger);
820
  if (table_log_handler)
petr@mysql.com's avatar
petr@mysql.com committed
821
  {
822
    table_log_handler->cleanup();
petr@mysql.com's avatar
petr@mysql.com committed
823 824
    delete table_log_handler;
  }
825 826 827 828 829
  if (file_log_handler)
    file_log_handler->cleanup();
}


petr@mysql.com's avatar
petr@mysql.com committed
830 831 832 833 834 835 836 837
void LOGGER::cleanup_end()
{
  DBUG_ASSERT(inited == 1);
  if (file_log_handler)
    delete file_log_handler;
}


838
/**
839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855
  Perform basic log initialization: create file-based log handler and
  init error log.
*/
void LOGGER::init_base()
{
  DBUG_ASSERT(inited == 0);
  inited= 1;

  /*
    Here we create file log handler. We don't do it for the table log handler
    here as it cannot be created so early. The reason is THD initialization,
    which depends on the system variables (parsed later).
  */
  if (!file_log_handler)
    file_log_handler= new Log_to_file_event_handler;

  /* by default we use traditional error log */
856
  init_error_log(LOG_FILE);
857 858

  file_log_handler->init_pthread_objects();
859
  my_rwlock_init(&LOCK_logger, NULL);
860 861 862 863 864 865 866 867 868 869 870 871 872 873
}


void LOGGER::init_log_tables()
{
  if (!table_log_handler)
    table_log_handler= new Log_to_csv_event_handler;

  if (!is_log_tables_initialized &&
      !table_log_handler->init() && !file_log_handler->init())
    is_log_tables_initialized= TRUE;
}


874 875 876
bool LOGGER::flush_logs(THD *thd)
{
  int rc= 0;
877

878
  /*
879 880
    Now we lock logger, as nobody should be able to use logging routines while
    log tables are closed
881
  */
882
  logger.lock_exclusive();
883 884 885

  /* reopen log files */
  file_log_handler->flush();
886

887 888
  /* end of log flush */
  logger.unlock();
889
  return rc;
890 891 892 893 894 895 896 897 898
}


/*
  Log slow query with all enabled log event handlers

  SYNOPSIS
    slow_log_print()

899 900 901 902
    thd                 THD of the query being logged
    query               The query being logged
    query_length        The length of the query string
    current_utime       Current time in microseconds (from undefined start)
903 904

  RETURN
905 906
    FALSE   OK
    TRUE    error occured
907 908 909
*/

bool LOGGER::slow_log_print(THD *thd, const char *query, uint query_length,
910 911
                            ulonglong current_utime)

912 913
{
  bool error= FALSE;
914
  Log_event_handler **current_handler;
915 916 917
  bool is_command= FALSE;
  char user_host_buff[MAX_USER_HOST_SIZE];
  Security_context *sctx= thd->security_ctx;
918
  uint user_host_len= 0;
919
  ulonglong query_utime, lock_utime;
920 921 922 923 924 925 926

  /*
    Print the message to the buffer if we have slow log enabled
  */

  if (*slow_log_handler_list)
  {
927
    time_t current_time;
928

929
    /* do not log slow queries from replication threads */
930
    if (thd->slave_thread && !opt_log_slow_slave_statements)
931 932
      return 0;

933
    lock_shared();
934 935 936 937 938
    if (!opt_slow_log)
    {
      unlock();
      return 0;
    }
939 940

    /* fill in user_host value: the format is "%s[%s] @ %s [%s]" */
941 942 943 944 945 946 947 948 949
    user_host_len= (strxnmov(user_host_buff, MAX_USER_HOST_SIZE,
                             sctx->priv_user ? sctx->priv_user : "", "[",
                             sctx->user ? sctx->user : "", "] @ ",
                             sctx->host ? sctx->host : "", " [",
                             sctx->ip ? sctx->ip : "", "]", NullS) -
                    user_host_buff);

    current_time= my_time_possible_from_micro(current_utime);
    if (thd->start_utime)
950
    {
951 952 953 954
      query_utime= (current_utime - thd->start_utime);
      lock_utime=  (thd->utime_after_lock - thd->start_utime);
    }
    else
955
    {
956
      query_utime= lock_utime= 0;
957 958 959 960 961 962 963 964 965
    }

    if (!query)
    {
      is_command= TRUE;
      query= command_name[thd->command].str;
      query_length= command_name[thd->command].length;
    }

966 967
    for (current_handler= slow_log_handler_list; *current_handler ;)
      error= (*current_handler++)->log_slow(thd, current_time, thd->start_time,
968
                                            user_host_buff, user_host_len,
969
                                            query_utime, lock_utime, is_command,
970 971 972 973 974 975 976
                                            query, query_length) || error;

    unlock();
  }
  return error;
}

977 978
bool LOGGER::general_log_write(THD *thd, enum enum_server_command command,
                               const char *query, uint query_length)
979 980 981
{
  bool error= FALSE;
  Log_event_handler **current_handler= general_log_handler_list;
982 983 984 985 986
  char user_host_buff[MAX_USER_HOST_SIZE];
  Security_context *sctx= thd->security_ctx;
  ulong id;
  uint user_host_len= 0;
  time_t current_time;
987

988 989 990 991
  if (thd)
    id= thd->thread_id;                 /* Normal thread */
  else
    id= 0;                              /* Log from connect handler */
992

993 994 995
  lock_shared();
  if (!opt_log)
  {
996
    unlock();
997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016
    return 0;
  }
  user_host_len= strxnmov(user_host_buff, MAX_USER_HOST_SIZE,
                          sctx->priv_user ? sctx->priv_user : "", "[",
                          sctx->user ? sctx->user : "", "] @ ",
                          sctx->host ? sctx->host : "", " [",
                          sctx->ip ? sctx->ip : "", "]", NullS) -
                                                          user_host_buff;

  current_time= my_time(0);
  while (*current_handler)
    error+= (*current_handler++)->
      log_general(thd, current_time, user_host_buff,
                  user_host_len, id,
                  command_name[(uint) command].str,
                  command_name[(uint) command].length,
                  query, query_length,
                  thd->variables.character_set_client) || error;
  unlock();

1017 1018 1019
  return error;
}

1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035
bool LOGGER::general_log_print(THD *thd, enum enum_server_command command,
                               const char *format, va_list args)
{
  uint message_buff_len= 0;
  char message_buff[MAX_LOG_BUFFER_SIZE];

  /* prepare message */
  if (format)
    message_buff_len= my_vsnprintf(message_buff, sizeof(message_buff),
                                   format, args);
  else
    message_buff[0]= '\0';

  return general_log_write(thd, command, message_buff, message_buff_len);
}

1036
void LOGGER::init_error_log(uint error_log_printer)
1037
{
1038 1039
  if (error_log_printer & LOG_NONE)
  {
1040
    error_log_handler_list[0]= 0;
1041 1042 1043 1044 1045
    return;
  }

  switch (error_log_printer) {
  case LOG_FILE:
1046 1047 1048 1049
    error_log_handler_list[0]= file_log_handler;
    error_log_handler_list[1]= 0;
    break;
    /* these two are disabled for now */
1050
  case LOG_TABLE:
1051 1052
    DBUG_ASSERT(0);
    break;
1053
  case LOG_TABLE|LOG_FILE:
1054 1055 1056 1057 1058
    DBUG_ASSERT(0);
    break;
  }
}

1059
void LOGGER::init_slow_log(uint slow_log_printer)
1060
{
1061 1062
  if (slow_log_printer & LOG_NONE)
  {
1063
    slow_log_handler_list[0]= 0;
1064 1065 1066 1067 1068
    return;
  }

  switch (slow_log_printer) {
  case LOG_FILE:
1069 1070 1071
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= 0;
    break;
1072
  case LOG_TABLE:
1073 1074 1075
    slow_log_handler_list[0]= table_log_handler;
    slow_log_handler_list[1]= 0;
    break;
1076
  case LOG_TABLE|LOG_FILE:
1077 1078 1079 1080 1081 1082 1083
    slow_log_handler_list[0]= file_log_handler;
    slow_log_handler_list[1]= table_log_handler;
    slow_log_handler_list[2]= 0;
    break;
  }
}

1084
void LOGGER::init_general_log(uint general_log_printer)
1085
{
1086 1087
  if (general_log_printer & LOG_NONE)
  {
1088
    general_log_handler_list[0]= 0;
1089 1090 1091 1092 1093
    return;
  }

  switch (general_log_printer) {
  case LOG_FILE:
1094 1095 1096
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= 0;
    break;
1097
  case LOG_TABLE:
1098 1099 1100
    general_log_handler_list[0]= table_log_handler;
    general_log_handler_list[1]= 0;
    break;
1101
  case LOG_TABLE|LOG_FILE:
1102 1103 1104 1105 1106 1107 1108 1109
    general_log_handler_list[0]= file_log_handler;
    general_log_handler_list[1]= table_log_handler;
    general_log_handler_list[2]= 0;
    break;
  }
}


1110 1111
bool LOGGER::activate_log_handler(THD* thd, uint log_type)
{
1112
  MYSQL_QUERY_LOG *file_log;
1113 1114
  bool res= FALSE;
  lock_exclusive();
1115 1116 1117 1118
  switch (log_type) {
  case QUERY_LOG_SLOW:
    if (!opt_slow_log)
    {
1119 1120 1121
      file_log= file_log_handler->get_mysql_slow_log();

      file_log->open_slow_log(sys_var_slow_log_path.value);
1122 1123 1124 1125
      if (table_log_handler->activate_log(thd, QUERY_LOG_SLOW))
      {
        /* Error printed by open table in activate_log() */
        res= TRUE;
1126
        file_log->close(0);
1127 1128
      }
      else
1129 1130
      {
        init_slow_log(log_output_options);
1131
        opt_slow_log= TRUE;
1132
      }
1133 1134 1135 1136 1137
    }
    break;
  case QUERY_LOG_GENERAL:
    if (!opt_log)
    {
1138 1139 1140
      file_log= file_log_handler->get_mysql_log();

      file_log->open_query_log(sys_var_general_log_path.value);
1141 1142 1143 1144
      if (table_log_handler->activate_log(thd, QUERY_LOG_GENERAL))
      {
        /* Error printed by open table in activate_log() */
        res= TRUE;
1145
        file_log->close(0);
1146 1147
      }
      else
1148 1149
      {
        init_general_log(log_output_options);
1150
        opt_log= TRUE;
1151
      }
1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176
    }
    break;
  default:
    DBUG_ASSERT(0);
  }
  unlock();
  return res;
}


void LOGGER::deactivate_log_handler(THD *thd, uint log_type)
{
  my_bool *tmp_opt= 0;
  MYSQL_LOG *file_log;

  switch (log_type) {
  case QUERY_LOG_SLOW:
    tmp_opt= &opt_slow_log;
    file_log= file_log_handler->get_mysql_slow_log();
    break;
  case QUERY_LOG_GENERAL:
    tmp_opt= &opt_log;
    file_log= file_log_handler->get_mysql_log();
    break;
  default:
1177
    assert(0);                                  // Impossible
1178 1179 1180 1181 1182
  }

  if (!(*tmp_opt))
    return;

1183
  lock_exclusive();
1184 1185 1186 1187 1188 1189
  file_log->close(0);
  *tmp_opt= FALSE;
  unlock();
}


1190 1191 1192
/* the parameters are unused for the log tables */
bool Log_to_csv_event_handler::init()
{
1193
  return 0;
1194 1195
}

1196 1197 1198
int LOGGER::set_handlers(uint error_log_printer,
                         uint slow_log_printer,
                         uint general_log_printer)
1199 1200
{
  /* error log table is not supported yet */
1201
  DBUG_ASSERT(error_log_printer < LOG_TABLE);
1202

1203
  lock_exclusive();
1204

1205
  if ((slow_log_printer & LOG_TABLE || general_log_printer & LOG_TABLE) &&
1206 1207
      !is_log_tables_initialized)
  {
1208 1209
    slow_log_printer= (slow_log_printer & ~LOG_TABLE) | LOG_FILE;
    general_log_printer= (general_log_printer & ~LOG_TABLE) | LOG_FILE;
1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224

    sql_print_error("Failed to initialize log tables. "
                    "Falling back to the old-fashioned logs");
  }

  init_error_log(error_log_printer);
  init_slow_log(slow_log_printer);
  init_general_log(general_log_printer);

  unlock();

  return 0;
}


1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244
 /*
  Save position of binary log transaction cache.

  SYNPOSIS
    binlog_trans_log_savepos()

    thd      The thread to take the binlog data from
    pos      Pointer to variable where the position will be stored

  DESCRIPTION

    Save the current position in the binary log transaction cache into
    the variable pointed to by 'pos'
 */

static void
binlog_trans_log_savepos(THD *thd, my_off_t *pos)
{
  DBUG_ENTER("binlog_trans_log_savepos");
  DBUG_ASSERT(pos != NULL);
1245
  if (thd_get_ha_data(thd, binlog_hton) == NULL)
1246 1247
    thd->binlog_setup_trx_data();
  binlog_trx_data *const trx_data=
1248
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1249 1250
  DBUG_ASSERT(mysql_bin_log.is_open());
  *pos= trx_data->position();
1251
  DBUG_PRINT("return", ("*pos: %lu", (ulong) *pos));
1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274
  DBUG_VOID_RETURN;
}


/*
  Truncate the binary log transaction cache.

  SYNPOSIS
    binlog_trans_log_truncate()

    thd      The thread to take the binlog data from
    pos      Position to truncate to

  DESCRIPTION

    Truncate the binary log to the given position. Will not change
    anything else.

 */
static void
binlog_trans_log_truncate(THD *thd, my_off_t pos)
{
  DBUG_ENTER("binlog_trans_log_truncate");
1275
  DBUG_PRINT("enter", ("pos: %lu", (ulong) pos));
1276

1277
  DBUG_ASSERT(thd_get_ha_data(thd, binlog_hton) != NULL);
1278 1279 1280 1281
  /* Only true if binlog_trans_log_savepos() wasn't called before */
  DBUG_ASSERT(pos != ~(my_off_t) 0);

  binlog_trx_data *const trx_data=
1282
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1283 1284 1285 1286 1287
  trx_data->truncate(pos);
  DBUG_VOID_RETURN;
}


1288 1289
/*
  this function is mostly a placeholder.
1290
  conceptually, binlog initialization (now mostly done in MYSQL_BIN_LOG::open)
1291 1292 1293
  should be moved here.
*/

1294
int binlog_init(void *p)
1295
{
1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306
  binlog_hton= (handlerton *)p;
  binlog_hton->state=opt_bin_log ? SHOW_OPTION_YES : SHOW_OPTION_NO;
  binlog_hton->db_type=DB_TYPE_BINLOG;
  binlog_hton->savepoint_offset= sizeof(my_off_t);
  binlog_hton->close_connection= binlog_close_connection;
  binlog_hton->savepoint_set= binlog_savepoint_set;
  binlog_hton->savepoint_rollback= binlog_savepoint_rollback;
  binlog_hton->commit= binlog_commit;
  binlog_hton->rollback= binlog_rollback;
  binlog_hton->prepare= binlog_prepare;
  binlog_hton->flags= HTON_NOT_USER_SELECTABLE | HTON_HIDDEN;
1307
  return 0;
1308 1309
}

1310
static int binlog_close_connection(handlerton *hton, THD *thd)
1311
{
1312
  binlog_trx_data *const trx_data=
1313
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1314
  DBUG_ASSERT(trx_data->empty());
1315
  thd_set_ha_data(thd, binlog_hton, NULL);
1316
  trx_data->~binlog_trx_data();
1317
  my_free((uchar*)trx_data, MYF(0));
1318 1319 1320
  return 0;
}

1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343
/*
  End a transaction.

  SYNOPSIS
    binlog_end_trans()

    thd      The thread whose transaction should be ended
    trx_data Pointer to the transaction data to use
    end_ev   The end event to use, or NULL
    all      True if the entire transaction should be ended, false if
             only the statement transaction should be ended.

  DESCRIPTION

    End the currently open transaction. The transaction can be either
    a real transaction (if 'all' is true) or a statement transaction
    (if 'all' is false).

    If 'end_ev' is NULL, the transaction is a rollback of only
    transactional tables, so the transaction cache will be truncated
    to either just before the last opened statement transaction (if
    'all' is false), or reset completely (if 'all' is true).
 */
1344
static int
1345 1346
binlog_end_trans(THD *thd, binlog_trx_data *trx_data,
                 Log_event *end_ev, bool all)
1347
{
1348
  DBUG_ENTER("binlog_end_trans");
1349 1350
  int error=0;
  IO_CACHE *trans_log= &trx_data->trans_log;
1351 1352
  DBUG_PRINT("enter", ("transaction: %s  end_ev: 0x%lx",
                       all ? "all" : "stmt", (long) end_ev));
1353 1354 1355
  DBUG_PRINT("info", ("thd->options={ %s%s}",
                      FLAGSTR(thd->options, OPTION_NOT_AUTOCOMMIT),
                      FLAGSTR(thd->options, OPTION_BEGIN)));
1356

1357 1358 1359 1360 1361 1362
  /*
    NULL denotes ROLLBACK with nothing to replicate: i.e., rollback of
    only transactional tables.  If the transaction contain changes to
    any non-transactiona tables, we need write the transaction and log
    a ROLLBACK last.
  */
1363
  if (end_ev != NULL)
1364
  {
1365
    /*
1366 1367 1368 1369
      Doing a commit or a rollback including non-transactional tables,
      i.e., ending a transaction where we might write the transaction
      cache to the binary log.

1370 1371 1372 1373 1374
      We can always end the statement when ending a transaction since
      transactions are not allowed inside stored functions.  If they
      were, we would have to ensure that we're not ending a statement
      inside a stored function.
     */
1375
    thd->binlog_flush_pending_rows_event(TRUE);
1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396
    /*
      We write the transaction cache to the binary log if either we're
      committing the entire transaction, or if we are doing an
      autocommit outside a transaction.
     */
    if (all || !(thd->options & (OPTION_BEGIN | OPTION_NOT_AUTOCOMMIT)))
    {
      error= mysql_bin_log.write(thd, &trx_data->trans_log, end_ev);
      trx_data->reset();
      /*
        We need to step the table map version after writing the
        transaction cache to disk.
      */
      mysql_bin_log.update_table_map_version();
      statistic_increment(binlog_cache_use, &LOCK_status);
      if (trans_log->disk_writes != 0)
      {
        statistic_increment(binlog_cache_disk_use, &LOCK_status);
        trans_log->disk_writes= 0;
      }
    }
1397 1398 1399
  }
  else
  {
1400 1401 1402
    /*
      If rolling back an entire transaction or a single statement not
      inside a transaction, we reset the transaction cache.
1403

1404 1405 1406 1407 1408
      If rolling back a statement in a transaction, we truncate the
      transaction cache to remove the statement.
     */
    if (all || !(thd->options & (OPTION_BEGIN | OPTION_NOT_AUTOCOMMIT)))
      trx_data->reset();
1409 1410
    else                                        // ...statement
      trx_data->truncate(trx_data->before_stmt_pos);
1411 1412 1413 1414 1415 1416 1417

    /*
      We need to step the table map version on a rollback to ensure
      that a new table map event is generated instead of the one that
      was written to the thrown-away transaction cache.
    */
    mysql_bin_log.update_table_map_version();
1418
  }
1419

1420
  DBUG_RETURN(error);
1421 1422
}

1423
static int binlog_prepare(handlerton *hton, THD *thd, bool all)
1424 1425 1426 1427 1428
{
  /*
    do nothing.
    just pretend we can do 2pc, so that MySQL won't
    switch to 1pc.
1429
    real work will be done in MYSQL_BIN_LOG::log_xid()
1430 1431 1432 1433
  */
  return 0;
}

1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448
/**
  This function is called once after each statement.

  It has the responsibility to flush the transaction cache to the
  binlog file on commits.

  @param hton  The binlog handlerton.
  @param thd   The client thread that executes the transaction.
  @param all   true if this is the last statement before a COMMIT
               statement; false if either this is a statement in a
               transaction but not the last, or if this is a statement
               not inside a BEGIN block and autocommit is on.

  @see handlerton::commit
*/
1449
static int binlog_commit(handlerton *hton, THD *thd, bool all)
1450 1451
{
  DBUG_ENTER("binlog_commit");
1452
  binlog_trx_data *const trx_data=
1453
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1454

1455
  if (trx_data->empty())
1456
  {
1457
    // we're here because trans_log was flushed in MYSQL_BIN_LOG::log_xid()
1458
    trx_data->reset();
1459 1460
    DBUG_RETURN(0);
  }
1461 1462 1463 1464 1465 1466 1467 1468 1469
  /*
    Write commit event if at least one of the following holds:
     - the user sends an explicit COMMIT; or
     - the autocommit flag is on, and we are not inside a BEGIN.
    However, if the user has not sent an explicit COMMIT, and we are
    either inside a BEGIN or run with autocommit off, then this is not
    the end of a transaction and we should not write a commit event.
  */
  if (all || !(thd->options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN)))
1470 1471 1472
  {
    Query_log_event qev(thd, STRING_WITH_LEN("COMMIT"), TRUE, FALSE);
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
1473 1474
    int error= binlog_end_trans(thd, trx_data, &qev, all);
    DBUG_RETURN(error);
1475 1476
  }
  else
1477 1478 1479 1480
  {
    int error= binlog_end_trans(thd, trx_data, &invisible_commit, all);
    DBUG_RETURN(error);
  }
1481 1482
}

1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499
/**
  This function is called when a transaction involving a transactional
  table is rolled back.

  It has the responsibility to flush the transaction cache to the
  binlog file. However, if the transaction does not involve
  non-transactional tables, nothing needs to be logged.

  @param hton  The binlog handlerton.
  @param thd   The client thread that executes the transaction.
  @param all   true if this is the last statement before a COMMIT
               statement; false if either this is a statement in a
               transaction but not the last, or if this is a statement
               not inside a BEGIN block and autocommit is on.

  @see handlerton::rollback
*/
1500
static int binlog_rollback(handlerton *hton, THD *thd, bool all)
1501 1502
{
  DBUG_ENTER("binlog_rollback");
1503 1504
  int error=0;
  binlog_trx_data *const trx_data=
1505
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
1506 1507 1508 1509 1510 1511

  if (trx_data->empty()) {
    trx_data->reset();
    DBUG_RETURN(0);
  }

1512
  /*
1513 1514 1515 1516
    Update the binary log with a BEGIN/ROLLBACK block if we have
    cached some queries and we updated some non-transactional
    table. Such cases should be rare (updating a
    non-transactional table inside a transaction...)
1517
  */
1518 1519
  if (unlikely(thd->transaction.all.modified_non_trans_table || 
               (thd->options & OPTION_KEEP_LOG)))
1520
  {
1521
    Query_log_event qev(thd, STRING_WITH_LEN("ROLLBACK"), TRUE, FALSE);
1522
    qev.error_code= 0; // see comment in MYSQL_LOG::write(THD, IO_CACHE)
1523
    error= binlog_end_trans(thd, trx_data, &qev, all);
1524
  }
1525
  else
1526
    error= binlog_end_trans(thd, trx_data, 0, all);
1527 1528 1529
  DBUG_RETURN(error);
}

1530 1531 1532 1533 1534 1535
/**
  @note
  How do we handle this (unlikely but legal) case:
  @verbatim
    [transaction] + [update to non-trans table] + [rollback to savepoint] ?
  @endverbatim
1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553
  The problem occurs when a savepoint is before the update to the
  non-transactional table. Then when there's a rollback to the savepoint, if we
  simply truncate the binlog cache, we lose the part of the binlog cache where
  the update is. If we want to not lose it, we need to write the SAVEPOINT
  command and the ROLLBACK TO SAVEPOINT command to the binlog cache. The latter
  is easy: it's just write at the end of the binlog cache, but the former
  should be *inserted* to the place where the user called SAVEPOINT. The
  solution is that when the user calls SAVEPOINT, we write it to the binlog
  cache (so no need to later insert it). As transactions are never intermixed
  in the binary log (i.e. they are serialized), we won't have conflicts with
  savepoint names when using mysqlbinlog or in the slave SQL thread.
  Then when ROLLBACK TO SAVEPOINT is called, if we updated some
  non-transactional table, we don't truncate the binlog cache but instead write
  ROLLBACK TO SAVEPOINT to it; otherwise we truncate the binlog cache (which
  will chop the SAVEPOINT command from the binlog cache, which is good as in
  that case there is no need to have it in the binlog).
*/

1554
static int binlog_savepoint_set(handlerton *hton, THD *thd, void *sv)
1555 1556 1557
{
  DBUG_ENTER("binlog_savepoint_set");

1558
  binlog_trans_log_savepos(thd, (my_off_t*) sv);
1559
  /* Write it to the binary log */
1560 1561 1562 1563 1564
  
  int const error=
    thd->binlog_query(THD::STMT_QUERY_TYPE,
                      thd->query, thd->query_length, TRUE, FALSE);
  DBUG_RETURN(error);
1565 1566
}

1567
static int binlog_savepoint_rollback(handlerton *hton, THD *thd, void *sv)
1568 1569 1570 1571 1572 1573 1574 1575
{
  DBUG_ENTER("binlog_savepoint_rollback");

  /*
    Write ROLLBACK TO SAVEPOINT to the binlog cache if we have updated some
    non-transactional table. Otherwise, truncate the binlog cache starting
    from the SAVEPOINT command.
  */
1576 1577
  if (unlikely(thd->transaction.all.modified_non_trans_table || 
               (thd->options & OPTION_KEEP_LOG)))
1578
  {
1579
    int error=
1580 1581 1582
      thd->binlog_query(THD::STMT_QUERY_TYPE,
                        thd->query, thd->query_length, TRUE, FALSE);
    DBUG_RETURN(error);
1583
  }
1584
  binlog_trans_log_truncate(thd, *(my_off_t*)sv);
1585 1586 1587
  DBUG_RETURN(0);
}

1588

1589 1590 1591 1592 1593
int check_binlog_magic(IO_CACHE* log, const char** errmsg)
{
  char magic[4];
  DBUG_ASSERT(my_b_tell(log) == 0);

1594
  if (my_b_read(log, (uchar*) magic, sizeof(magic)))
1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608
  {
    *errmsg = "I/O error reading the header from the binary log";
    sql_print_error("%s, errno=%d, io cache code=%d", *errmsg, my_errno,
		    log->error);
    return 1;
  }
  if (memcmp(magic, BINLOG_MAGIC, sizeof(magic)))
  {
    *errmsg = "Binlog has bad magic number;  It's not a binary log file that can be used by this version of MySQL";
    return 1;
  }
  return 0;
}

1609

1610 1611 1612 1613 1614
File open_binlog(IO_CACHE *log, const char *log_file_name, const char **errmsg)
{
  File file;
  DBUG_ENTER("open_binlog");

1615 1616
  if ((file = my_open(log_file_name, O_RDONLY | O_BINARY | O_SHARE, 
                      MYF(MY_WME))) < 0)
1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642
  {
    sql_print_error("Failed to open log (file '%s', errno %d)",
                    log_file_name, my_errno);
    *errmsg = "Could not open log file";
    goto err;
  }
  if (init_io_cache(log, file, IO_SIZE*2, READ_CACHE, 0, 0,
                    MYF(MY_WME|MY_DONT_CHECK_FILESIZE)))
  {
    sql_print_error("Failed to create a cache on log (file '%s')",
                    log_file_name);
    *errmsg = "Could not open log file";
    goto err;
  }
  if (check_binlog_magic(log,errmsg))
    goto err;
  DBUG_RETURN(file);

err:
  if (file >= 0)
  {
    my_close(file,MYF(0));
    end_io_cache(log);
  }
  DBUG_RETURN(-1);
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1643

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1644 1645
#ifdef __NT__
static int eventSource = 0;
serg@serg.mylan's avatar
serg@serg.mylan committed
1646

1647
static void setup_windows_event_source()
monty@mysql.com's avatar
monty@mysql.com committed
1648
{
1649
  HKEY    hRegKey= NULL;
monty@mysql.com's avatar
monty@mysql.com committed
1650 1651 1652
  DWORD   dwError= 0;
  TCHAR   szPath[MAX_PATH];
  DWORD dwTypes;
1653

monty@mysql.com's avatar
monty@mysql.com committed
1654 1655 1656
  if (eventSource)               // Ensure that we are only called once
    return;
  eventSource= 1;
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1657

monty@mysql.com's avatar
monty@mysql.com committed
1658
  // Create the event source registry key
1659
  dwError= RegCreateKey(HKEY_LOCAL_MACHINE,
monty@mysql.com's avatar
monty@mysql.com committed
1660 1661
                          "SYSTEM\\CurrentControlSet\\Services\\EventLog\\Application\\MySQL", 
                          &hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1662

monty@mysql.com's avatar
monty@mysql.com committed
1663 1664
  /* Name of the PE module that contains the message resource */
  GetModuleFileName(NULL, szPath, MAX_PATH);
serg@serg.mylan's avatar
serg@serg.mylan committed
1665

rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1666
  /* Register EventMessageFile */
1667
  dwError = RegSetValueEx(hRegKey, "EventMessageFile", 0, REG_EXPAND_SZ,
1668
                          (PBYTE) szPath, (DWORD) (strlen(szPath) + 1));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1669

monty@mysql.com's avatar
monty@mysql.com committed
1670 1671 1672 1673 1674
  /* Register supported event types */
  dwTypes= (EVENTLOG_ERROR_TYPE | EVENTLOG_WARNING_TYPE |
            EVENTLOG_INFORMATION_TYPE);
  dwError= RegSetValueEx(hRegKey, "TypesSupported", 0, REG_DWORD,
                         (LPBYTE) &dwTypes, sizeof dwTypes);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1675

monty@mysql.com's avatar
monty@mysql.com committed
1676
  RegCloseKey(hRegKey);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1677 1678
}

monty@mysql.com's avatar
monty@mysql.com committed
1679
#endif /* __NT__ */
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
1680 1681


1682 1683 1684 1685 1686 1687 1688 1689
/**
  Find a unique filename for 'filename.#'.

  Set '#' to a number as low as possible.

  @return
    nonzero if not possible to get unique filename
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1690 1691 1692

static int find_uniq_filename(char *name)
{
1693 1694 1695 1696
  long                  number;
  uint                  i;
  char                  buff[FN_REFLEN];
  struct st_my_dir     *dir_info;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1697
  reg1 struct fileinfo *file_info;
1698
  ulong                 max_found=0;
1699 1700
  size_t		buf_length, length;
  char			*start, *end;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1701 1702
  DBUG_ENTER("find_uniq_filename");

1703 1704 1705
  length= dirname_part(buff, name, &buf_length);
  start=  name + length;
  end=    strend(start);
1706

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1707
  *end='.';
1708
  length= (size_t) (end-start+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1709 1710 1711 1712 1713 1714 1715 1716 1717

  if (!(dir_info = my_dir(buff,MYF(MY_DONT_SORT))))
  {						// This shouldn't happen
    strmov(end,".1");				// use name+1
    DBUG_RETURN(0);
  }
  file_info= dir_info->dir_entry;
  for (i=dir_info->number_off_files ; i-- ; file_info++)
  {
1718
    if (bcmp((uchar*) file_info->name, (uchar*) start, length) == 0 &&
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1719 1720 1721 1722 1723 1724 1725 1726
	test_if_number(file_info->name+length, &number,0))
    {
      set_if_bigger(max_found,(ulong) number);
    }
  }
  my_dirend(dir_info);

  *end++='.';
1727
  sprintf(end,"%06ld",max_found+1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1728 1729 1730
  DBUG_RETURN(0);
}

1731

1732
void MYSQL_LOG::init(enum_log_type log_type_arg,
1733
                     enum cache_type io_cache_type_arg)
1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802
{
  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;

1803
  if (log_type == LOG_NORMAL)
1804 1805
  {
    char *end;
1806
    int len=my_snprintf(buff, sizeof(buff), "%s, Version: %s (%s). "
1807
#ifdef EMBEDDED_LIBRARY
1808 1809
                        "embedded library\n",
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT
1810 1811
#elif __NT__
			"started with:\nTCP Port: %d, Named Pipe: %s\n",
1812 1813
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
1814 1815
#else
			"started with:\nTcp port: %d  Unix socket: %s\n",
1816 1817
                        my_progname, server_version, MYSQL_COMPILATION_COMMENT,
                        mysqld_port, mysqld_unix_port
1818 1819 1820 1821
#endif
                       );
    end= strnmov(buff + len, "Time                 Id Command    Argument\n",
                 sizeof(buff) - len);
1822
    if (my_b_write(&log_file, (uchar*) buff, (uint) (end-buff)) ||
1823 1824 1825 1826
	flush_io_cache(&log_file))
      goto err;
  }

1827
  log_state= LOG_OPENED;
1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838
  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);
1839
  log_state= LOG_CLOSED;
1840 1841 1842
  DBUG_RETURN(1);
}

1843
MYSQL_LOG::MYSQL_LOG()
serg@janus.mylan's avatar
serg@janus.mylan committed
1844 1845
  : name(0), write_error(FALSE), inited(FALSE), log_type(LOG_UNKNOWN),
    log_state(LOG_CLOSED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1846 1847
{
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1848 1849
    We don't want to initialize LOCK_Log here as such initialization depends on
    safe_mutex (when using safe_mutex) which depends on MY_INIT(), which is
1850
    called only in main(). Doing initialization here would make it happen
1851
    before main().
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1852
  */
1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880
  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));
1881
  if (log_state == LOG_OPENED)
1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897
  {
    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);
    }
  }

1898
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
1899 1900
  safeFree(name);
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1901 1902
}

1903
/** This is called only once. */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
1904

monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1905
void MYSQL_LOG::cleanup()
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1906
{
1907
  DBUG_ENTER("cleanup");
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1908
  if (inited)
1909
  {
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
1910
    inited= 0;
1911
    (void) pthread_mutex_destroy(&LOCK_log);
1912
    close(0);
1913
  }
1914
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1915 1916
}

1917

bk@work.mysql.com's avatar
bk@work.mysql.com committed
1918
int MYSQL_LOG::generate_new_name(char *new_name, const char *log_name)
1919
{
1920 1921
  fn_format(new_name, log_name, mysql_data_home, "", 4);
  if (log_type == LOG_BIN)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934
  {
    if (!fn_ext(log_name)[0])
    {
      if (find_uniq_filename(new_name))
      {
	sql_print_error(ER(ER_NO_UNIQUE_LOGFILE), log_name);
	return 1;
      }
    }
  }
  return 0;
}

1935

1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947
/*
  Reopen the log file

  SYNOPSIS
    reopen_file()

  DESCRIPTION
    Reopen the log file. The method is used during FLUSH LOGS
    and locks LOCK_log mutex
*/


1948
void MYSQL_QUERY_LOG::reopen_file()
1949
{
1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965
  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);

  /*
1966
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
1967 1968
  */

1969
  open(save_name, log_type, 0, io_cache_type);
1970 1971 1972 1973
  my_free(save_name, MYF(0));

  pthread_mutex_unlock(&LOCK_log);

1974
  DBUG_VOID_RETURN;
1975 1976
}

1977

1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
/*
  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
*/

2003 2004 2005 2006
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)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2007
{
2008 2009
  char buff[32];
  uint length= 0;
2010
  char local_time_buff[MAX_TIME_SIZE];
2011 2012 2013
  struct tm start;
  uint time_buff_len= 0;

2014 2015
  (void) pthread_mutex_lock(&LOCK_log);

2016 2017 2018 2019 2020 2021 2022 2023 2024 2025
  /* 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);

2026
        time_buff_len= my_snprintf(local_time_buff, MAX_TIME_SIZE,
2027 2028 2029 2030 2031
                                   "%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);

2032
        if (my_b_write(&log_file, (uchar*) local_time_buff, time_buff_len))
2033 2034 2035
          goto err;
      }
      else
2036
        if (my_b_write(&log_file, (uchar*) "\t\t" ,2) < 0)
2037 2038
          goto err;

2039
      /* command_type, thread_id */
2040
      length= my_snprintf(buff, 32, "%5ld ", (long) thread_id);
2041

2042
    if (my_b_write(&log_file, (uchar*) buff, length))
2043 2044
      goto err;

2045
    if (my_b_write(&log_file, (uchar*) command_type, command_type_len))
2046 2047
      goto err;

2048
    if (my_b_write(&log_file, (uchar*) "\t", 1))
2049 2050 2051
      goto err;

    /* sql_text */
2052
    if (my_b_write(&log_file, (uchar*) sql_text, sql_text_len))
2053 2054
      goto err;

2055
    if (my_b_write(&log_file, (uchar*) "\n", 1) ||
2056 2057 2058 2059
        flush_io_cache(&log_file))
      goto err;
  }

2060
  (void) pthread_mutex_unlock(&LOCK_log);
2061 2062 2063 2064 2065 2066 2067 2068
  return FALSE;
err:

  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
2069
  (void) pthread_mutex_unlock(&LOCK_log);
2070
  return TRUE;
2071 2072
}

2073

2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085
/*
  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
2086 2087
    query_utime       Amount of time the query took to execute (in microseconds)
    lock_utime        Amount of time the query was locked (in microseconds)
2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102
    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
*/

2103 2104
bool MYSQL_QUERY_LOG::write(THD *thd, time_t current_time,
                            time_t query_start_arg, const char *user_host,
2105 2106
                            uint user_host_len, ulonglong query_utime,
                            ulonglong lock_utime, bool is_command,
2107
                            const char *sql_text, uint sql_text_len)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2108
{
2109
  bool error= 0;
2110
  DBUG_ENTER("MYSQL_QUERY_LOG::write");
2111

2112 2113
  (void) pthread_mutex_lock(&LOCK_log);

2114
  if (!is_open())
2115 2116
  {
    (void) pthread_mutex_unlock(&LOCK_log);
2117
    DBUG_RETURN(0);
2118
  }
2119 2120 2121 2122 2123

  if (is_open())
  {						// Safety agains reopen
    int tmp_errno= 0;
    char buff[80], *end;
2124
    char query_time_buff[22+7], lock_time_buff[22+7];
2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142
    uint buff_len;
    end= buff;

    if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
    {
      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 */
2143
        if (my_b_write(&log_file, (uchar*) buff, buff_len))
2144 2145
          tmp_errno= errno;
      }
2146 2147
      const uchar uh[]= "# User@Host: ";
      if (my_b_write(&log_file, uh, sizeof(uh) - 1))
2148
        tmp_errno= errno;
2149
      if (my_b_write(&log_file, (uchar*) user_host, user_host_len))
2150
        tmp_errno= errno;
2151
      if (my_b_write(&log_file, (uchar*) "\n", 1))
2152 2153 2154
        tmp_errno= errno;
    }
    /* For slow query log */
2155 2156
    sprintf(query_time_buff, "%.6f", ulonglong2double(query_utime)/1000000.0);
    sprintf(lock_time_buff,  "%.6f", ulonglong2double(lock_utime)/1000000.0);
2157
    if (my_b_printf(&log_file,
2158
                    "# Query_time: %s  Lock_time: %s"
2159
                    " Rows_sent: %lu  Rows_examined: %lu\n",
2160
                    query_time_buff, lock_time_buff,
2161 2162 2163 2164 2165 2166 2167 2168 2169
                    (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);
    }
2170
    if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
2171 2172
    {
      end=strmov(end, ",last_insert_id=");
2173 2174 2175
      end=longlong10_to_str((longlong)
                            thd->first_successful_insert_id_in_prev_stmt_for_binlog,
                            end, -10);
2176 2177
    }
    // Save value if we do an insert.
2178
    if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
2179 2180 2181 2182
    {
      if (!(specialflag & SPECIAL_SHORT_LOG_FORMAT))
      {
        end=strmov(end,",insert_id=");
2183 2184 2185
        end=longlong10_to_str((longlong)
                              thd->auto_inc_intervals_in_cur_stmt_for_binlog.minimum(),
                              end, -10);
2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200
      }
    }

    /*
      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';
2201 2202
      if (my_b_write(&log_file, (uchar*) "SET ", 4) ||
          my_b_write(&log_file, (uchar*) buff + 1, (uint) (end-buff)))
2203 2204 2205 2206 2207 2208
        tmp_errno= errno;
    }
    if (is_command)
    {
      end= strxmov(buff, "# administrator command: ", NullS);
      buff_len= (ulong) (end - buff);
2209
      my_b_write(&log_file, (uchar*) buff, buff_len);
2210
    }
2211 2212
    if (my_b_write(&log_file, (uchar*) sql_text, sql_text_len) ||
        my_b_write(&log_file, (uchar*) ";\n",2) ||
2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224
        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);
      }
    }
  }
2225
  (void) pthread_mutex_unlock(&LOCK_log);
2226
  DBUG_RETURN(error);
2227 2228
}

2229

2230 2231 2232 2233 2234
/**
  @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.
*/
2235
const char *MYSQL_LOG::generate_name(const char *log_name,
2236 2237
                                      const char *suffix,
                                      bool strip_ext, char *buff)
2238 2239 2240
{
  if (!log_name || !log_name[0])
  {
2241 2242 2243
    strmake(buff, pidfile_name, FN_REFLEN - strlen(suffix) - 1);
    return (const char *)
      fn_format(buff, buff, "", suffix, MYF(MY_REPLACE_EXT|MY_REPLACE_DIR));
2244 2245 2246 2247
  }
  // get rid of extension if the log is binary to avoid problems
  if (strip_ext)
  {
2248 2249 2250
    char *p= fn_ext(log_name);
    uint length= (uint) (p - log_name);
    strmake(buff, log_name, min(length, FN_REFLEN));
2251 2252 2253 2254 2255
    return (const char*)buff;
  }
  return log_name;
}

2256

2257

2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313
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,
2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338
                                const char *log_name)
{
  File index_file_nr= -1;
  DBUG_ASSERT(!my_b_inited(&index_file));

  /*
    First open of this class instance
    Create an index file that will hold all file names uses for logging.
    Add new entries to the end of it.
  */
  myf opt= MY_UNPACK_FILENAME;
  if (!index_file_name_arg)
  {
    index_file_name_arg= log_name;    // Use same basename for index file
    opt= MY_UNPACK_FILENAME | MY_REPLACE_EXT;
  }
  fn_format(index_file_name, index_file_name_arg, mysql_data_home,
            ".index", opt);
  if ((index_file_nr= my_open(index_file_name,
                              O_RDWR | O_CREAT | O_BINARY ,
                              MYF(MY_WME))) < 0 ||
       my_sync(index_file_nr, MYF(MY_WME)) ||
       init_io_cache(&index_file, index_file_nr,
                     IO_SIZE, WRITE_CACHE,
                     my_seek(index_file_nr,0L,MY_SEEK_END,MYF(0)),
serg@serg.mylan's avatar
serg@serg.mylan committed
2339
			0, MYF(MY_WME | MY_WAIT_IF_FULL)))
2340 2341 2342 2343 2344 2345 2346 2347
  {
    if (index_file_nr >= 0)
      my_close(index_file_nr,MYF(0));
    return TRUE;
  }
  return FALSE;
}

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2348

2349
/**
2350
  Open a (new) binlog file.
2351

2352
  - Open the log file and the index file. Register the new
2353
  file name in it
2354
  - When calling this when the file is in use, you must have a locks
2355
  on LOCK_log and LOCK_index.
2356

2357
  @retval
2358
    0	ok
2359
  @retval
2360 2361 2362
    1	error
*/

2363 2364 2365 2366 2367 2368 2369
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)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2370
{
2371
  File file= -1;
2372
  DBUG_ENTER("MYSQL_BIN_LOG::open");
2373
  DBUG_PRINT("enter",("log_type: %d",(int) log_type_arg));
2374 2375

  write_error=0;
2376

2377 2378 2379
  /* 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 */
2380

2381
  init(no_auto_events_arg, max_size_arg);
2382

2383
  open_count++;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2384

2385
  DBUG_ASSERT(log_type == LOG_BIN);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2386 2387

  {
2388 2389 2390 2391 2392 2393 2394 2395 2396 2397
    bool write_file_name_to_index_file=0;

    if (!my_b_filelength(&log_file))
    {
      /*
	The binary log file was empty (probably newly created)
	This is the normal case and happens when the user doesn't specify
	an extension for the binary log files.
	In this case we write a standard header to it.
      */
2398
      if (my_b_safe_write(&log_file, (uchar*) BINLOG_MAGIC,
2399
			  BIN_LOG_HEADER_SIZE))
2400
        goto err;
2401 2402
      bytes_written+= BIN_LOG_HEADER_SIZE;
      write_file_name_to_index_file= 1;
2403
    }
2404

2405
    DBUG_ASSERT(my_b_inited(&index_file) != 0);
2406 2407
    reinit_io_cache(&index_file, WRITE_CACHE,
                    my_b_filelength(&index_file), 0, 0);
2408
    if (need_start_event && !no_auto_events)
2409
    {
2410 2411 2412 2413 2414
      /*
        In 4.x we set need_start_event=0 here, but in 5.0 we want a Start event
        even if this is not the very first binlog.
      */
      Format_description_log_event s(BINLOG_VERSION);
serg@serg.mylan's avatar
serg@serg.mylan committed
2415 2416 2417 2418 2419 2420
      /*
        don't set LOG_EVENT_BINLOG_IN_USE_F for SEQ_READ_APPEND io_cache
        as we won't be able to reset it later
      */
      if (io_cache_type == WRITE_CACHE)
        s.flags|= LOG_EVENT_BINLOG_IN_USE_F;
2421 2422
      if (!s.is_valid())
        goto err;
2423
      s.dont_set_created= null_created_arg;
2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438
      if (s.write(&log_file))
        goto err;
      bytes_written+= s.data_written;
    }
    if (description_event_for_queue &&
        description_event_for_queue->binlog_version>=4)
    {
      /*
        This is a relay log written to by the I/O slave thread.
        Write the event so that others can later know the format of this relay
        log.
        Note that this event is very close to the original event from the
        master (it has binlog version of the master, event types of the
        master), so this is suitable to parse the next relay log's event. It
        has been produced by
monty@mysql.com's avatar
monty@mysql.com committed
2439
        Format_description_log_event::Format_description_log_event(char* buf,).
2440 2441 2442 2443 2444 2445 2446 2447 2448
        Why don't we want to write the description_event_for_queue if this
        event is for format<4 (3.23 or 4.x): this is because in that case, the
        description_event_for_queue describes the data received from the
        master, but not the data written to the relay log (*conversion*),
        which is in format 4 (slave's).
      */
      /*
        Set 'created' to 0, so that in next relay logs this event does not
        trigger cleaning actions on the slave in
2449
        Format_description_log_event::apply_event_impl().
2450 2451 2452 2453
      */
      description_event_for_queue->created= 0;
      /* Don't set log_pos in event header */
      description_event_for_queue->artificial_event=1;
2454

2455 2456 2457
      if (description_event_for_queue->write(&log_file))
        goto err;
      bytes_written+= description_event_for_queue->data_written;
2458
    }
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2459 2460
    if (flush_io_cache(&log_file) ||
        my_sync(log_file.file, MYF(MY_WME)))
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
2461
      goto err;
2462 2463 2464

    if (write_file_name_to_index_file)
    {
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2465 2466 2467 2468
      /*
        As this is a new log file, we write the file name to the index
        file. As every time we write to the index file, we sync it.
      */
2469
      if (my_b_write(&index_file, (uchar*) log_file_name,
2470
		     strlen(log_file_name)) ||
2471
	  my_b_write(&index_file, (uchar*) "\n", 1) ||
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2472 2473
	  flush_io_cache(&index_file) ||
          my_sync(index_file.file, MYF(MY_WME)))
2474
	goto err;
monty@donna.mysql.com's avatar
monty@donna.mysql.com committed
2475
    }
2476
  }
2477 2478
  log_state= LOG_OPENED;

2479
  DBUG_RETURN(0);
2480 2481

err:
2482 2483 2484
  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, \
2485
shutdown the MySQL server and restart it.", name, errno);
2486 2487 2488
  if (file >= 0)
    my_close(file,MYF(0));
  end_io_cache(&log_file);
2489 2490
  end_io_cache(&index_file);
  safeFree(name);
2491
  log_state= LOG_CLOSED;
2492
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2493 2494
}

2495

2496
int MYSQL_BIN_LOG::get_current_log(LOG_INFO* linfo)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2497 2498
{
  pthread_mutex_lock(&LOCK_log);
2499 2500 2501 2502 2503
  int ret = raw_get_current_log(linfo);
  pthread_mutex_unlock(&LOCK_log);
  return ret;
}

jonas@perch.ndb.mysql.com's avatar
jonas@perch.ndb.mysql.com committed
2504
int MYSQL_BIN_LOG::raw_get_current_log(LOG_INFO* linfo)
2505
{
2506 2507
  strmake(linfo->log_file_name, log_file_name, sizeof(linfo->log_file_name)-1);
  linfo->pos = my_b_tell(&log_file);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2508 2509 2510
  return 0;
}

2511 2512
/**
  Move all data up in a file in an filename index file.
2513 2514 2515 2516 2517

    We do the copy outside of the IO_CACHE as the cache buffers would just
    make things slower and more complicated.
    In most cases the copy loop should only do one read.

2518 2519 2520 2521 2522 2523 2524
  @param index_file			File to move
  @param offset			Move everything from here to beginning

  @note
    File will be truncated to be 'offset' shorter or filled up with newlines

  @retval
2525 2526 2527
    0	ok
*/

2528 2529
#ifdef HAVE_REPLICATION

2530 2531 2532 2533 2534
static bool copy_up_file_and_fill(IO_CACHE *index_file, my_off_t offset)
{
  int bytes_read;
  my_off_t init_offset= offset;
  File file= index_file->file;
2535
  uchar io_buf[IO_SIZE*2];
2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546
  DBUG_ENTER("copy_up_file_and_fill");

  for (;; offset+= bytes_read)
  {
    (void) my_seek(file, offset, MY_SEEK_SET, MYF(0));
    if ((bytes_read= (int) my_read(file, io_buf, sizeof(io_buf), MYF(MY_WME)))
	< 0)
      goto err;
    if (!bytes_read)
      break;					// end of file
    (void) my_seek(file, offset-init_offset, MY_SEEK_SET, MYF(0));
2547
    if (my_write(file, io_buf, bytes_read, MYF(MY_WME | MY_NABP)))
2548 2549 2550
      goto err;
  }
  /* The following will either truncate the file or fill the end with \n' */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
2551 2552
  if (my_chsize(file, offset - init_offset, '\n', MYF(MY_WME)) ||
      my_sync(file, MYF(MY_WME)))
2553 2554 2555 2556 2557 2558 2559 2560 2561 2562
    goto err;

  /* Reset data in old index cache */
  reinit_io_cache(index_file, READ_CACHE, (my_off_t) 0, 0, 1);
  DBUG_RETURN(0);

err:
  DBUG_RETURN(1);
}

2563
#endif /* HAVE_REPLICATION */
2564

2565 2566
/**
  Find the position in the log-index-file for the given log name.
2567

2568 2569 2570 2571 2572 2573
  @param linfo		Store here the found log file name and position to
                       the NEXT log file name in the index file.
  @param log_name	Filename to find in the index file.
                       Is a null pointer if we want to read the first entry
  @param need_lock	Set this to 1 if the parent doesn't already have a
                       lock on LOCK_index
2574

2575
  @note
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
2576 2577
    On systems without the truncate function the file will end with one or
    more empty lines.  These will be ignored when reading the file.
2578

2579
  @retval
2580
    0			ok
2581 2582 2583
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
2584 2585 2586
    LOG_INFO_IO		Got IO error while reading file
*/

2587
int MYSQL_BIN_LOG::find_log_pos(LOG_INFO *linfo, const char *log_name,
2588
			    bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2589
{
2590 2591 2592 2593 2594
  int error= 0;
  char *fname= linfo->log_file_name;
  uint log_name_len= log_name ? (uint) strlen(log_name) : 0;
  DBUG_ENTER("find_log_pos");
  DBUG_PRINT("enter",("log_name: %s", log_name ? log_name : "NULL"));
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2595

2596
  /*
2597 2598
    Mutex needed because we need to make sure the file pointer does not
    move from under our feet
2599
  */
2600
  if (need_lock)
2601
    pthread_mutex_lock(&LOCK_index);
2602 2603 2604 2605 2606
  safe_mutex_assert_owner(&LOCK_index);

  /* As the file is flushed, we can't get an error here */
  (void) reinit_io_cache(&index_file, READ_CACHE, (my_off_t) 0, 0, 0);

2607
  for (;;)
2608
  {
2609
    uint length;
2610 2611 2612 2613
    my_off_t offset= my_b_tell(&index_file);
    /* If we get 0 or 1 characters, this is the end of the file */

    if ((length= my_b_gets(&index_file, fname, FN_REFLEN)) <= 1)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2614
    {
2615 2616 2617
      /* Did not find the given entry; Return not found or error */
      error= !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2618 2619
    }

2620 2621
    // if the log entry matches, null string matching anything
    if (!log_name ||
sasha@mysql.sashanet.com's avatar
sasha@mysql.sashanet.com committed
2622
	(log_name_len == length-1 && fname[log_name_len] == '\n' &&
2623
	 !memcmp(fname, log_name, log_name_len)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2624
    {
2625
      DBUG_PRINT("info",("Found log file entry"));
2626
      fname[length-1]=0;			// remove last \n
2627 2628
      linfo->index_file_start_offset= offset;
      linfo->index_file_offset = my_b_tell(&index_file);
2629
      break;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2630
    }
2631
  }
2632

2633
  if (need_lock)
2634
    pthread_mutex_unlock(&LOCK_index);
2635
  DBUG_RETURN(error);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2636
}
2637

2638

2639 2640
/**
  Find the position in the log-index-file for the given log name.
2641

2642
  @param
2643 2644
    linfo		Store here the next log file name and position to
			the file name after that.
2645
  @param
2646 2647 2648
    need_lock		Set this to 1 if the parent doesn't already have a
			lock on LOCK_index

2649
  @note
2650
    - Before calling this function, one has to call find_log_pos()
2651
    to set up 'linfo'
2652
    - Mutex needed because we need to make sure the file pointer does not move
2653
    from under our feet
2654

2655
  @retval
2656
    0			ok
2657 2658 2659
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
2660 2661 2662
    LOG_INFO_IO		Got IO error while reading file
*/

2663
int MYSQL_BIN_LOG::find_next_log(LOG_INFO* linfo, bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2664
{
2665
  int error= 0;
2666
  uint length;
2667 2668
  char *fname= linfo->log_file_name;

2669 2670
  if (need_lock)
    pthread_mutex_lock(&LOCK_index);
2671 2672 2673 2674 2675 2676 2677 2678
  safe_mutex_assert_owner(&LOCK_index);

  /* As the file is flushed, we can't get an error here */
  (void) reinit_io_cache(&index_file, READ_CACHE, linfo->index_file_offset, 0,
			 0);

  linfo->index_file_start_offset= linfo->index_file_offset;
  if ((length=my_b_gets(&index_file, fname, FN_REFLEN)) <= 1)
2679
  {
2680
    error = !index_file.error ? LOG_INFO_EOF : LOG_INFO_IO;
2681 2682
    goto err;
  }
2683
  fname[length-1]=0;				// kill \n
2684
  linfo->index_file_offset = my_b_tell(&index_file);
2685

bk@work.mysql.com's avatar
bk@work.mysql.com committed
2686
err:
2687 2688
  if (need_lock)
    pthread_mutex_unlock(&LOCK_index);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
2689 2690 2691
  return error;
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2692

2693 2694 2695
/**
  Delete all logs refered to in the index file.
  Start writing to a new log file.
2696

2697
  The new index file will only contain this file.
2698

2699
  @param thd		Thread
2700

2701 2702
  @note
    If not called from slave thread, write start event to new log
2703

2704
  @retval
2705
    0	ok
2706
  @retval
2707 2708 2709
    1   error
*/

2710
bool MYSQL_BIN_LOG::reset_logs(THD* thd)
2711 2712
{
  LOG_INFO linfo;
2713
  bool error=0;
2714
  const char* save_name;
2715
  DBUG_ENTER("reset_logs");
2716

tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2717
  ha_reset_logs(thd);
2718 2719 2720 2721
  /*
    We need to get both locks to be sure that no one is trying to
    write to the index log file.
  */
2722
  pthread_mutex_lock(&LOCK_log);
2723
  pthread_mutex_lock(&LOCK_index);
2724

2725 2726 2727 2728 2729 2730
  /*
    The following mutex is needed to ensure that no threads call
    'delete thd' as we would then risk missing a 'rollback' from this
    thread. If the transaction involved MyISAM tables, it should go
    into binlog even on rollback.
  */
2731
  VOID(pthread_mutex_lock(&LOCK_thread_count));
2732 2733 2734 2735

  /* Save variables so that we can reopen the log */
  save_name=name;
  name=0;					// Protect against free
2736
  close(LOG_CLOSE_TO_BE_OPENED);
2737 2738 2739 2740

  /* First delete all old log files */

  if (find_log_pos(&linfo, NullS, 0))
2741 2742 2743 2744
  {
    error=1;
    goto err;
  }
2745

2746
  for (;;)
2747
  {
2748
    my_delete_allow_opened(linfo.log_file_name, MYF(MY_WME));
2749
    if (find_next_log(&linfo, 0))
2750 2751
      break;
  }
2752 2753

  /* Start logging with a new file */
2754
  close(LOG_CLOSE_INDEX);
2755
  my_delete_allow_opened(index_file_name, MYF(MY_WME));	// Reset (open will update)
2756
  if (!thd->slave_thread)
2757
    need_start_event=1;
2758
  if (!open_index_file(index_file_name, 0))
2759
    open(save_name, log_type, 0, io_cache_type, no_auto_events, max_size, 0);
2760
  my_free((uchar*) save_name, MYF(0));
2761

2762
err:
2763
  VOID(pthread_mutex_unlock(&LOCK_thread_count));
2764
  pthread_mutex_unlock(&LOCK_index);
2765
  pthread_mutex_unlock(&LOCK_log);
2766
  DBUG_RETURN(error);
2767 2768
}

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2769

2770
/**
2771 2772
  Delete relay log files prior to rli->group_relay_log_name
  (i.e. all logs which are not involved in a non-finished group
2773 2774
  (transaction)), remove them from the index file and start on next
  relay log.
2775

2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789
  IMPLEMENTATION
  - Protects index file with LOCK_index
  - Delete relevant relay log files
  - Copy all file names after these ones to the front of the index file
  - If the OS has truncate, truncate the file, else fill it with \n'
  - Read the next file name from the index file and store in rli->linfo

  @param rli	       Relay log information
  @param included     If false, all relay logs that are strictly before
                      rli->group_relay_log_name are deleted ; if true, the
                      latter is deleted too (i.e. all relay logs
                      read by the SQL slave thread are deleted).

  @note
2790
    - This is only called from the slave-execute thread when it has read
2791
    all commands from a relay log and want to switch to a new relay log.
2792
    - When this happens, we can be in an active transaction as
2793 2794 2795
    a transaction can span over two relay logs
    (although it is always written as a single block to the master's binary
    log, hence cannot span over two master's binary logs).
2796

2797
  @retval
2798
    0			ok
2799 2800 2801
  @retval
    LOG_INFO_EOF	        End of log-index-file found
  @retval
2802
    LOG_INFO_SEEK	Could not allocate IO cache
2803
  @retval
2804
    LOG_INFO_IO		Got IO error while reading file
2805 2806
*/

hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
2807
#ifdef HAVE_REPLICATION
2808

2809
int MYSQL_BIN_LOG::purge_first_log(Relay_log_info* rli, bool included)
2810
{
2811
  int error;
2812 2813
  DBUG_ENTER("purge_first_log");

2814 2815
  DBUG_ASSERT(is_open());
  DBUG_ASSERT(rli->slave_running == 1);
2816
  DBUG_ASSERT(!strcmp(rli->linfo.log_file_name,rli->event_relay_log_name));
2817

2818
  pthread_mutex_lock(&LOCK_index);
2819 2820 2821 2822 2823 2824
  pthread_mutex_lock(&rli->log_space_lock);
  rli->relay_log.purge_logs(rli->group_relay_log_name, included,
                            0, 0, &rli->log_space_total);
  // Tell the I/O thread to take the relay_log_space_limit into account
  rli->ignore_log_space_limit= 0;
  pthread_mutex_unlock(&rli->log_space_lock);
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2825 2826

  /*
2827 2828 2829
    Ok to broadcast after the critical region as there is no risk of
    the mutex being destroyed by this thread later - this helps save
    context switches
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2830
  */
2831 2832 2833 2834 2835
  pthread_cond_broadcast(&rli->log_space_cond);
  
  /*
    Read the next log file name from the index file and pass it back to
    the caller
2836 2837
    If included is true, we want the first relay log;
    otherwise we want the one after event_relay_log_name.
2838
  */
2839 2840 2841 2842
  if ((included && (error=find_log_pos(&rli->linfo, NullS, 0))) ||
      (!included &&
       ((error=find_log_pos(&rli->linfo, rli->event_relay_log_name, 0)) ||
        (error=find_next_log(&rli->linfo, 0)))))
2843 2844
  {
    char buff[22];
2845 2846 2847 2848 2849
    sql_print_error("next log error: %d  offset: %s  log: %s included: %d",
                    error,
                    llstr(rli->linfo.index_file_offset,buff),
                    rli->group_relay_log_name,
                    included);
2850 2851
    goto err;
  }
2852

2853
  /*
2854
    Reset rli's coordinates to the current log.
2855
  */
2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869
  rli->event_relay_log_pos= BIN_LOG_HEADER_SIZE;
  strmake(rli->event_relay_log_name,rli->linfo.log_file_name,
	  sizeof(rli->event_relay_log_name)-1);

  /*
    If we removed the rli->group_relay_log_name file,
    we must update the rli->group* coordinates, otherwise do not touch it as the
    group's execution is not finished (e.g. COMMIT not executed)
  */
  if (included)
  {
    rli->group_relay_log_pos = BIN_LOG_HEADER_SIZE;
    strmake(rli->group_relay_log_name,rli->linfo.log_file_name,
            sizeof(rli->group_relay_log_name)-1);
2870
    rli->notify_group_relay_log_name_update();
2871
  }
2872 2873 2874

  /* Store where we are in the new file for the execution thread */
  flush_relay_log_info(rli);
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2875

2876
err:
2877
  pthread_mutex_unlock(&LOCK_index);
2878
  DBUG_RETURN(error);
2879 2880
}

2881 2882
/**
  Update log index_file.
2883 2884
*/

2885
int MYSQL_BIN_LOG::update_log_index(LOG_INFO* log_info, bool need_update_threads)
2886 2887 2888 2889 2890
{
  if (copy_up_file_and_fill(&index_file, log_info->index_file_start_offset))
    return LOG_INFO_IO;

  // now update offsets in index file for running threads
2891 2892
  if (need_update_threads)
    adjust_linfo_offsets(log_info->index_file_start_offset);
2893 2894
  return 0;
}
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2895

2896
/**
2897 2898
  Remove all logs before the given log from disk and from the index file.

2899 2900 2901 2902 2903 2904 2905
  @param to_log	      Delete all log file name before this file.
  @param included            If true, to_log is deleted too.
  @param need_mutex
  @param need_update_threads If we want to update the log coordinates of
                             all threads. False for relay logs, true otherwise.
  @param freed_log_space     If not null, decrement this variable of
                             the amount of log space freed
2906

2907
  @note
2908 2909 2910
    If any of the logs before the deleted one is in use,
    only purge logs up to this one.

2911 2912 2913
  @retval
    0			ok
  @retval
2914 2915 2916
    LOG_INFO_EOF		to_log not found
*/

2917
int MYSQL_BIN_LOG::purge_logs(const char *to_log, 
2918 2919 2920 2921
                          bool included,
                          bool need_mutex, 
                          bool need_update_threads, 
                          ulonglong *decrease_log_space)
2922 2923
{
  int error;
2924
  int ret = 0;
2925
  bool exit_loop= 0;
2926 2927
  LOG_INFO log_info;
  DBUG_ENTER("purge_logs");
2928
  DBUG_PRINT("info",("to_log= %s",to_log));
2929

2930 2931
  if (need_mutex)
    pthread_mutex_lock(&LOCK_index);
2932
  if ((error=find_log_pos(&log_info, to_log, 0 /*no mutex*/)))
2933 2934
    goto err;

2935
  /*
2936
    File name exists in index file; delete until we find this file
2937 2938 2939
    or a file that is used.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
2940
    goto err;
2941 2942
  while ((strcmp(to_log,log_info.log_file_name) || (exit_loop=included)) &&
         !log_in_use(log_info.log_file_name))
2943
  {
2944
    ulong file_size= 0;
2945 2946 2947
    if (decrease_log_space) //stat the file we want to delete
    {
      MY_STAT s;
2948 2949 2950 2951 2952 2953

      /* 
         If we could not stat, we can't know the amount
         of space that deletion will free. In most cases,
         deletion won't work either, so it's not a problem.
      */
2954
      if (my_stat(log_info.log_file_name,&s,MYF(0)))
2955
        file_size= s.st_size;
2956
      else
2957 2958
	sql_print_information("Failed to execute my_stat on file '%s'",
			      log_info.log_file_name);
2959 2960 2961 2962 2963 2964 2965
    }
    /*
      It's not fatal if we can't delete a log file ;
      if we could delete it, take its size into account
    */
    DBUG_PRINT("info",("purging %s",log_info.log_file_name));
    if (!my_delete(log_info.log_file_name, MYF(0)) && decrease_log_space)
2966
      *decrease_log_space-= file_size;
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2967 2968

    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);
2969 2970
    if (current_thd->is_slave_error) {
      DBUG_PRINT("info",("slave error: %d", current_thd->is_slave_error));
2971 2972 2973 2974 2975 2976
      if (my_errno == EMFILE) {
        DBUG_PRINT("info",("my_errno: %d, set ret = LOG_INFO_EMFILE", my_errno));
        ret = LOG_INFO_EMFILE;
        break;
      }
    }
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
2977

2978
    if (find_next_log(&log_info, 0) || exit_loop)
2979
      break;
2980
  }
2981

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
2982 2983 2984 2985
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
2986
  error= update_log_index(&log_info, need_update_threads);
2987 2988 2989
  if (error == 0) {
    error = ret;
  }
2990

2991
err:
2992 2993
  if (need_mutex)
    pthread_mutex_unlock(&LOCK_index);
2994
  DBUG_RETURN(error);
2995 2996
}

2997
/**
2998 2999 3000
  Remove all logs before the given file date from disk and from the
  index file.

3001 3002
  @param thd		Thread pointer
  @param before_date	Delete all log files before given date.
3003

3004
  @note
3005 3006 3007
    If any of the logs before the deleted one is in use,
    only purge logs up to this one.

3008
  @retval
3009
    0				ok
3010
  @retval
3011 3012 3013
    LOG_INFO_PURGE_NO_ROTATE	Binary file that can't be rotated
*/

3014
int MYSQL_BIN_LOG::purge_logs_before_date(time_t purge_time)
3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029
{
  int error;
  LOG_INFO log_info;
  MY_STAT stat_area;

  DBUG_ENTER("purge_logs_before_date");

  pthread_mutex_lock(&LOCK_index);

  /*
    Delete until we find curren file
    or a file that is used or a file
    that is older than purge_time.
  */
  if ((error=find_log_pos(&log_info, NullS, 0 /*no mutex*/)))
3030
    goto err;
3031 3032 3033 3034 3035

  while (strcmp(log_file_name, log_info.log_file_name) &&
	 !log_in_use(log_info.log_file_name))
  {
    /* It's not fatal even if we can't delete a log file */
3036 3037
    if (!my_stat(log_info.log_file_name, &stat_area, MYF(0)) ||
	stat_area.st_mtime >= purge_time)
3038
      break;
3039
    my_delete(log_info.log_file_name, MYF(0));
tomas@poseidon.ndb.mysql.com's avatar
tomas@poseidon.ndb.mysql.com committed
3040 3041 3042

    ha_binlog_index_purge_file(current_thd, log_info.log_file_name);

3043 3044
    if (find_next_log(&log_info, 0))
      break;
3045
  }
3046

3047 3048 3049 3050
  /*
    If we get killed -9 here, the sysadmin would have to edit
    the log index file after restart - otherwise, this should be safe
  */
3051
  error= update_log_index(&log_info, 1);
3052

3053 3054
err:
  pthread_mutex_unlock(&LOCK_index);
3055
  DBUG_RETURN(error);
3056
}
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3057
#endif /* HAVE_REPLICATION */
3058

3059

3060 3061
/**
  Create a new log file name.
3062

3063
  @param buf		buf of at least FN_REFLEN where new name is stored
3064

3065
  @note
3066 3067 3068
    If file name will be longer then FN_REFLEN it will be truncated
*/

3069
void MYSQL_BIN_LOG::make_log_name(char* buf, const char* log_ident)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3070
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3071
  uint dir_len = dirname_length(log_file_name); 
3072
  if (dir_len >= FN_REFLEN)
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3073 3074
    dir_len=FN_REFLEN-1;
  strnmov(buf, log_file_name, dir_len);
3075
  strmake(buf+dir_len, log_ident, FN_REFLEN - dir_len -1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3076 3077
}

3078

3079 3080
/**
  Check if we are writing/reading to the given log file.
3081 3082
*/

3083
bool MYSQL_BIN_LOG::is_active(const char *log_file_name_arg)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3084
{
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3085
  return !strcmp(log_file_name, log_file_name_arg);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3086 3087
}

3088

3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108
/*
  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);
}


3109 3110
/**
  Start writing to a new log file or reopen the old file.
3111

3112
  @param need_lock		Set to 1 if caller has not locked LOCK_log
3113

3114
  @note
3115 3116 3117
    The new file name is stored last in the index file
*/

3118
void MYSQL_BIN_LOG::new_file_impl(bool need_lock)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3119
{
3120 3121
  char new_name[FN_REFLEN], *new_name_ptr, *old_name;

3122
  DBUG_ENTER("MYSQL_BIN_LOG::new_file_impl");
3123
  if (!is_open())
3124 3125 3126 3127
  {
    DBUG_PRINT("info",("log is closed"));
    DBUG_VOID_RETURN;
  }
3128 3129 3130

  if (need_lock)
    pthread_mutex_lock(&LOCK_log);
3131 3132
  pthread_mutex_lock(&LOCK_index);

3133 3134 3135
  safe_mutex_assert_owner(&LOCK_log);
  safe_mutex_assert_owner(&LOCK_index);

3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148
  /*
    if binlog is used as tc log, be sure all xids are "unlogged",
    so that on recover we only need to scan one - latest - binlog file
    for prepared xids. As this is expected to be a rare event,
    simple wait strategy is enough. We're locking LOCK_log to be sure no
    new Xid_log_event's are added to the log (and prepared_xids is not
    increased), and waiting on COND_prep_xids for late threads to
    catch up.
  */
  if (prepared_xids)
  {
    tc_log_page_waits++;
    pthread_mutex_lock(&LOCK_prep_xids);
3149 3150
    while (prepared_xids) {
      DBUG_PRINT("info", ("prepared_xids=%lu", prepared_xids));
3151
      pthread_cond_wait(&COND_prep_xids, &LOCK_prep_xids);
3152
    }
3153 3154 3155
    pthread_mutex_unlock(&LOCK_prep_xids);
  }

3156
  /* Reuse old name if not binlog and not update log */
3157
  new_name_ptr= name;
monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3158

3159
  /*
3160 3161 3162
    If user hasn't specified an extension, generate a new log name
    We have to do this here and not in open as we want to store the
    new file name in the current binary log file.
3163
  */
3164 3165 3166
  if (generate_new_name(new_name, name))
    goto end;
  new_name_ptr=new_name;
3167

3168
  if (log_type == LOG_BIN)
3169
  {
3170
    if (!no_auto_events)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3171
    {
3172
      /*
3173 3174
        We log the whole file name for log file as the user may decide
        to change base names at some point.
3175
      */
3176
      Rotate_log_event r(new_name+dirname_length(new_name),
3177
                         0, LOG_EVENT_OFFSET, 0);
3178
      r.write(&log_file);
3179
      bytes_written += r.data_written;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3180
    }
3181 3182 3183 3184 3185
    /*
      Update needs to be signalled even if there is no rotate event
      log rotation should give the waiting thread a signal to
      discover EOF and move on to the next log.
    */
3186
    signal_update();
3187 3188 3189
  }
  old_name=name;
  name=0;				// Don't free name
3190
  close(LOG_CLOSE_TO_BE_OPENED);
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3191

3192
  /*
3193
     Note that at this point, log_state != LOG_CLOSED (important for is_open()).
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3194
  */
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3195

3196
  /*
3197
     new_file() is only used for rotation (in FLUSH LOGS or because size >
3198
     max_binlog_size or max_relay_log_size).
3199 3200 3201 3202
     If this is a binary log, the Format_description_log_event at the beginning of
     the new file should have created=0 (to distinguish with the
     Format_description_log_event written at server startup, which should
     trigger temp tables deletion on slaves.
3203
  */
3204

3205
  open(old_name, log_type, new_name_ptr,
3206
       io_cache_type, no_auto_events, max_size, 1);
3207 3208
  my_free(old_name,MYF(0));

3209
end:
3210 3211
  if (need_lock)
    pthread_mutex_unlock(&LOCK_log);
3212 3213
  pthread_mutex_unlock(&LOCK_index);

3214
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3215 3216
}

3217

3218
bool MYSQL_BIN_LOG::append(Log_event* ev)
3219 3220 3221
{
  bool error = 0;
  pthread_mutex_lock(&LOCK_log);
3222
  DBUG_ENTER("MYSQL_BIN_LOG::append");
3223

3224
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3225 3226 3227 3228
  /*
    Log_event::write() is smart enough to use my_b_write() or
    my_b_append() depending on the kind of cache we have.
  */
3229 3230 3231 3232 3233
  if (ev->write(&log_file))
  {
    error=1;
    goto err;
  }
3234
  bytes_written+= ev->data_written;
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3235
  DBUG_PRINT("info",("max_size: %lu",max_size));
3236
  if ((uint) my_b_append_tell(&log_file) > max_size)
3237
    new_file_without_locking();
3238

3239
err:
3240
  pthread_mutex_unlock(&LOCK_log);
3241
  signal_update();				// Safe as we don't call close
3242
  DBUG_RETURN(error);
3243 3244
}

3245

3246
bool MYSQL_BIN_LOG::appendv(const char* buf, uint len,...)
3247
{
3248
  bool error= 0;
3249
  DBUG_ENTER("MYSQL_BIN_LOG::appendv");
3250 3251
  va_list(args);
  va_start(args,len);
3252

3253
  DBUG_ASSERT(log_file.type == SEQ_READ_APPEND);
3254

3255
  safe_mutex_assert_owner(&LOCK_log);
3256 3257
  do
  {
3258
    if (my_b_append(&log_file,(uchar*) buf,len))
3259
    {
3260 3261
      error= 1;
      goto err;
3262
    }
3263
    bytes_written += len;
3264
  } while ((buf=va_arg(args,const char*)) && (len=va_arg(args,uint)));
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
3265
  DBUG_PRINT("info",("max_size: %lu",max_size));
3266
  if ((uint) my_b_append_tell(&log_file) > max_size)
3267
    new_file_without_locking();
3268 3269

err:
3270 3271
  if (!error)
    signal_update();
3272
  DBUG_RETURN(error);
3273
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3274

monty@hundin.mysql.fi's avatar
monty@hundin.mysql.fi committed
3275

3276
bool MYSQL_BIN_LOG::flush_and_sync()
3277
{
3278 3279 3280 3281 3282
  int err=0, fd=log_file.file;
  safe_mutex_assert_owner(&LOCK_log);
  if (flush_io_cache(&log_file))
    return 1;
  if (++sync_binlog_counter >= sync_binlog_period && sync_binlog_period)
3283 3284
  {
    sync_binlog_counter= 0;
3285
    err=my_sync(fd, MYF(MY_WME));
3286
  }
3287
  return err;
3288 3289
}

3290
void MYSQL_BIN_LOG::start_union_events(THD *thd, query_id_t query_id_param)
3291 3292 3293 3294 3295
{
  DBUG_ASSERT(!thd->binlog_evt_union.do_union);
  thd->binlog_evt_union.do_union= TRUE;
  thd->binlog_evt_union.unioned_events= FALSE;
  thd->binlog_evt_union.unioned_events_trans= FALSE;
3296
  thd->binlog_evt_union.first_query_id= query_id_param;
3297 3298
}

3299
void MYSQL_BIN_LOG::stop_union_events(THD *thd)
3300 3301 3302 3303 3304
{
  DBUG_ASSERT(thd->binlog_evt_union.do_union);
  thd->binlog_evt_union.do_union= FALSE;
}

3305
bool MYSQL_BIN_LOG::is_query_in_union(THD *thd, query_id_t query_id_param)
3306 3307 3308 3309 3310
{
  return (thd->binlog_evt_union.do_union && 
          query_id_param >= thd->binlog_evt_union.first_query_id);
}

3311 3312 3313 3314 3315 3316 3317 3318 3319 3320

/*
  These functions are placed in this file since they need access to
  binlog_hton, which has internal linkage.
*/

int THD::binlog_setup_trx_data()
{
  DBUG_ENTER("THD::binlog_setup_trx_data");
  binlog_trx_data *trx_data=
3321
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3322 3323 3324 3325

  if (trx_data)
    DBUG_RETURN(0);                             // Already set up

3326
  trx_data= (binlog_trx_data*) my_malloc(sizeof(binlog_trx_data), MYF(MY_ZEROFILL));
3327 3328 3329 3330
  if (!trx_data ||
      open_cached_file(&trx_data->trans_log, mysql_tmpdir,
                       LOG_PREFIX, binlog_cache_size, MYF(MY_WME)))
  {
3331
    my_free((uchar*)trx_data, MYF(MY_ALLOW_ZERO_PTR));
3332 3333
    DBUG_RETURN(1);                      // Didn't manage to set it up
  }
3334
  thd_set_ha_data(this, binlog_hton, trx_data);
3335

3336
  trx_data= new (thd_get_ha_data(this, binlog_hton)) binlog_trx_data;
3337

3338 3339 3340
  DBUG_RETURN(0);
}

3341
/*
3342 3343
  Function to start a statement and optionally a transaction for the
  binary log.
3344

3345 3346
  SYNOPSIS
    binlog_start_trans_and_stmt()
3347

3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371
  DESCRIPTION

    This function does three things:
    - Start a transaction if not in autocommit mode or if a BEGIN
      statement has been seen.

    - Start a statement transaction to allow us to truncate the binary
      log.

    - Save the currrent binlog position so that we can roll back the
      statement by truncating the transaction log.

      We only update the saved position if the old one was undefined,
      the reason is that there are some cases (e.g., for CREATE-SELECT)
      where the position is saved twice (e.g., both in
      select_create::prepare() and THD::binlog_write_table_map()) , but
      we should use the first. This means that calls to this function
      can be used to start the statement before the first table map
      event, to include some extra events.
 */

void
THD::binlog_start_trans_and_stmt()
{
3372
  binlog_trx_data *trx_data= (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3373 3374 3375 3376 3377 3378
  DBUG_ENTER("binlog_start_trans_and_stmt");
  DBUG_PRINT("enter", ("trx_data: 0x%lx  trx_data->before_stmt_pos: %lu",
                       (long) trx_data,
                       (trx_data ? (ulong) trx_data->before_stmt_pos :
                        (ulong) 0)));

3379
  if (trx_data == NULL ||
3380
      trx_data->before_stmt_pos == MY_OFF_T_UNDEF)
3381
  {
3382
    this->binlog_set_stmt_begin();
3383
    if (options & (OPTION_NOT_AUTOCOMMIT | OPTION_BEGIN))
mats@romeo.(none)'s avatar
mats@romeo.(none) committed
3384 3385
      trans_register_ha(this, TRUE, binlog_hton);
    trans_register_ha(this, FALSE, binlog_hton);
3386 3387 3388 3389
  }
  DBUG_VOID_RETURN;
}

3390 3391
void THD::binlog_set_stmt_begin() {
  binlog_trx_data *trx_data=
3392
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3393 3394 3395 3396 3397 3398 3399 3400 3401

  /*
    The call to binlog_trans_log_savepos() might create the trx_data
    structure, if it didn't exist before, so we save the position
    into an auto variable and then write it into the transaction
    data for the binary log (i.e., trx_data).
  */
  my_off_t pos= 0;
  binlog_trans_log_savepos(this, &pos);
3402
  trx_data= (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3403 3404 3405 3406 3407 3408
  trx_data->before_stmt_pos= pos;
}

int THD::binlog_flush_transaction_cache()
{
  DBUG_ENTER("binlog_flush_transaction_cache");
3409 3410
  binlog_trx_data *trx_data= (binlog_trx_data*)
    thd_get_ha_data(this, binlog_hton);
3411
  DBUG_PRINT("enter", ("trx_data=0x%lu", (ulong) trx_data));
3412
  if (trx_data)
3413 3414
    DBUG_PRINT("enter", ("trx_data->before_stmt_pos=%lu",
                         (ulong) trx_data->before_stmt_pos));
3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435

  /*
    Write the transaction cache to the binary log.  We don't flush and
    sync the log file since we don't know if more will be written to
    it. If the caller want the log file sync:ed, the caller has to do
    it.

    The transaction data is only reset upon a successful write of the
    cache to the binary log.
  */

  if (trx_data && likely(mysql_bin_log.is_open())) {
    if (int error= mysql_bin_log.write_cache(&trx_data->trans_log, true, true))
      DBUG_RETURN(error);
    trx_data->reset();
  }

  DBUG_RETURN(0);
}


3436 3437
/*
  Write a table map to the binary log.
3438 3439
 */

3440 3441
int THD::binlog_write_table_map(TABLE *table, bool is_trans)
{
3442
  int error;
3443
  DBUG_ENTER("THD::binlog_write_table_map");
3444 3445
  DBUG_PRINT("enter", ("table: 0x%lx  (%s: #%lu)",
                       (long) table, table->s->table_name.str,
3446
                       table->s->table_map_id));
3447 3448

  /* Pre-conditions */
3449
  DBUG_ASSERT(current_stmt_binlog_row_based && mysql_bin_log.is_open());
3450 3451 3452 3453 3454 3455 3456 3457
  DBUG_ASSERT(table->s->table_map_id != ULONG_MAX);

  Table_map_log_event::flag_set const
    flags= Table_map_log_event::TM_NO_FLAGS;

  Table_map_log_event
    the_event(this, table, table->s->table_map_id, is_trans, flags);

3458 3459
  if (is_trans && binlog_table_maps == 0)
    binlog_start_trans_and_stmt();
3460

3461
  if ((error= mysql_bin_log.write(&the_event)))
3462 3463
    DBUG_RETURN(error);

3464
  binlog_table_maps++;
3465 3466 3467 3468
  table->s->table_map_version= mysql_bin_log.table_map_version();
  DBUG_RETURN(0);
}

3469 3470 3471 3472
Rows_log_event*
THD::binlog_get_pending_rows_event() const
{
  binlog_trx_data *const trx_data=
3473
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3474 3475 3476 3477 3478 3479
  /*
    This is less than ideal, but here's the story: If there is no
    trx_data, prepare_pending_rows_event() has never been called
    (since the trx_data is set up there). In that case, we just return
    NULL.
   */
3480
  return trx_data ? trx_data->pending() : NULL;
3481 3482 3483 3484 3485
}

void
THD::binlog_set_pending_rows_event(Rows_log_event* ev)
{
3486
  if (thd_get_ha_data(this, binlog_hton) == NULL)
3487 3488
    binlog_setup_trx_data();

3489
  binlog_trx_data *const trx_data=
3490
    (binlog_trx_data*) thd_get_ha_data(this, binlog_hton);
3491

3492
  DBUG_ASSERT(trx_data);
3493
  trx_data->set_pending(ev);
3494 3495 3496 3497 3498 3499 3500 3501
}


/*
  Moves the last bunch of rows from the pending Rows event to the binlog
  (either cached binlog if transaction, or disk binlog). Sets a new pending
  event.
*/
3502 3503 3504
int
MYSQL_BIN_LOG::flush_and_set_pending_rows_event(THD *thd,
                                                Rows_log_event* event)
3505
{
3506
  DBUG_ENTER("MYSQL_BIN_LOG::flush_and_set_pending_rows_event(event)");
3507
  DBUG_ASSERT(mysql_bin_log.is_open());
3508
  DBUG_PRINT("enter", ("event: 0x%lx", (long) event));
3509 3510 3511 3512

  int error= 0;

  binlog_trx_data *const trx_data=
3513
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
3514 3515 3516

  DBUG_ASSERT(trx_data);

3517
  DBUG_PRINT("info", ("trx_data->pending(): 0x%lx", (long) trx_data->pending()));
3518

3519
  if (Rows_log_event* pending= trx_data->pending())
3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579
  {
    IO_CACHE *file= &log_file;

    /*
      Decide if we should write to the log file directly or to the
      transaction log.
    */
    if (pending->get_cache_stmt() || my_b_tell(&trx_data->trans_log))
      file= &trx_data->trans_log;

    /*
      If we are writing to the log file directly, we could avoid
      locking the log. This does not work since we need to step the
      m_table_map_version below, and that change has to be protected
      by the LOCK_log mutex.
    */
    pthread_mutex_lock(&LOCK_log);

    /*
      Write pending event to log file or transaction cache
    */
    if (pending->write(file))
    {
      pthread_mutex_unlock(&LOCK_log);
      DBUG_RETURN(1);
    }

    /*
      We step the table map version if we are writing an event
      representing the end of a statement.  We do this regardless of
      wheather we write to the transaction cache or to directly to the
      file.

      In an ideal world, we could avoid stepping the table map version
      if we were writing to a transaction cache, since we could then
      reuse the table map that was written earlier in the transaction
      cache.  This does not work since STMT_END_F implies closing all
      table mappings on the slave side.

      TODO: Find a solution so that table maps does not have to be
      written several times within a transaction.
     */
    if (pending->get_flags(Rows_log_event::STMT_END_F))
      ++m_table_map_version;

    delete pending;

    if (file == &log_file)
    {
      error= flush_and_sync();
      if (!error)
      {
        signal_update();
        rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
      }
    }

    pthread_mutex_unlock(&LOCK_log);
  }

3580
  thd->binlog_set_pending_rows_event(event);
3581 3582 3583 3584

  DBUG_RETURN(error);
}

3585 3586
/**
  Write an event to the binary log.
3587 3588
*/

3589
bool MYSQL_BIN_LOG::write(Log_event *event_info)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3590
{
3591 3592
  THD *thd= event_info->thd;
  bool error= 1;
3593
  DBUG_ENTER("MYSQL_BIN_LOG::write(Log_event *)");
3594

3595 3596
  if (thd->binlog_evt_union.do_union)
  {
3597 3598 3599 3600
    /*
      In Stored function; Remember that function call caused an update.
      We will log the function call to the binary log on function exit
    */
3601 3602 3603 3604
    thd->binlog_evt_union.unioned_events= TRUE;
    thd->binlog_evt_union.unioned_events_trans |= event_info->cache_stmt;
    DBUG_RETURN(0);
  }
3605 3606 3607 3608 3609 3610 3611

  /*
    Flush the pending rows event to the transaction cache or to the
    log file.  Since this function potentially aquire the LOCK_log
    mutex, we do this before aquiring the LOCK_log mutex in this
    function.

3612 3613 3614
    We only end the statement if we are in a top-level statement.  If
    we are inside a stored function, we do not end the statement since
    this will close all tables on the slave.
3615
  */
3616 3617 3618
  bool const end_stmt=
    thd->prelocked_mode && thd->lex->requires_prelocking();
  thd->binlog_flush_pending_rows_event(end_stmt);
3619

3620 3621
  pthread_mutex_lock(&LOCK_log);

3622
  /*
guilhem@mysql.com's avatar
guilhem@mysql.com committed
3623 3624 3625 3626
     In most cases this is only called if 'is_open()' is true; in fact this is
     mostly called if is_open() *was* true a few instructions before, but it
     could have changed since.
  */
3627
  if (likely(is_open()))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3628
  {
3629
    IO_CACHE *file= &log_file;
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3630
#ifdef HAVE_REPLICATION
3631
    /*
3632 3633 3634
      In the future we need to add to the following if tests like
      "do the involved tables match (to be implemented)
      binlog_[wild_]{do|ignore}_table?" (WL#1049)"
3635
    */
3636
    const char *local_db= event_info->get_db();
3637
    if ((thd && !(thd->options & OPTION_BIN_LOG)) ||
lars@mysql.com's avatar
Merge  
lars@mysql.com committed
3638
	(!binlog_filter->db_ok(local_db)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
3639
    {
3640
      VOID(pthread_mutex_unlock(&LOCK_log));
3641
      DBUG_RETURN(0);
3642
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
3643
#endif /* HAVE_REPLICATION */
3644

3645
#if defined(USING_TRANSACTIONS) 
3646 3647 3648 3649 3650 3651 3652
    /*
      Should we write to the binlog cache or to the binlog on disk?
      Write to the binlog cache if:
      - it is already not empty (meaning we're in a transaction; note that the
     present event could be about a non-transactional table, but still we need
     to write to the binlog cache in that case to handle updates to mixed
     trans/non-trans table types the best possible in binlogging)
3653
      - or if the event asks for it (cache_stmt == TRUE).
3654 3655 3656
    */
    if (opt_using_transactions && thd)
    {
3657 3658
      if (thd->binlog_setup_trx_data())
        goto err;
3659

3660
      binlog_trx_data *const trx_data=
3661
        (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
3662
      IO_CACHE *trans_log= &trx_data->trans_log;
3663 3664
      my_off_t trans_log_pos= my_b_tell(trans_log);
      if (event_info->get_cache_stmt() || trans_log_pos != 0)
3665
      {
3666
        DBUG_PRINT("info", ("Using trans_log: cache: %d, trans_log_pos: %lu",
3667
                            event_info->get_cache_stmt(),
3668
                            (ulong) trans_log_pos));
3669 3670
        if (trans_log_pos == 0)
          thd->binlog_start_trans_and_stmt();
3671
        file= trans_log;
3672
      }
3673
      /*
3674
        TODO as Mats suggested, for all the cases above where we write to
3675 3676
        trans_log, it sounds unnecessary to lock LOCK_log. We should rather
        test first if we want to write to trans_log, and if not, lock
3677
        LOCK_log.
3678
      */
3679
    }
3680
#endif /* USING_TRANSACTIONS */
3681
    DBUG_PRINT("info",("event type: %d",event_info->get_type_code()));
3682

3683 3684 3685 3686
    /*
      No check for auto events flag here - this write method should
      never be called if auto-events are enabled
    */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3687 3688

    /*
3689 3690
      1. Write first log events which describe the 'run environment'
      of the SQL command
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3691 3692
    */

3693 3694 3695 3696
    /*
      If row-based binlogging, Insert_id, Rand and other kind of "setting
      context" events are not needed.
    */
3697
    if (thd)
3698
    {
3699
      if (!thd->current_stmt_binlog_row_based)
3700
      {
3701
        if (thd->stmt_depends_on_first_successful_insert_id_in_prev_stmt)
3702 3703
        {
          Intvar_log_event e(thd,(uchar) LAST_INSERT_ID_EVENT,
3704
                             thd->first_successful_insert_id_in_prev_stmt_for_binlog);
3705 3706 3707
          if (e.write(file))
            goto err;
        }
3708
        if (thd->auto_inc_intervals_in_cur_stmt_for_binlog.nb_elements() > 0)
3709
        {
3710 3711 3712
          DBUG_PRINT("info",("number of auto_inc intervals: %u",
                             thd->auto_inc_intervals_in_cur_stmt_for_binlog.
                             nb_elements()));
3713 3714
          /*
            If the auto_increment was second in a table's index (possible with
3715
            MyISAM or BDB) (table->next_number_keypart != 0), such event is
3716 3717
            in fact not necessary. We could avoid logging it.
          */
3718 3719 3720
          Intvar_log_event e(thd, (uchar) INSERT_ID_EVENT,
                             thd->auto_inc_intervals_in_cur_stmt_for_binlog.
                             minimum());
3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734
          if (e.write(file))
            goto err;
        }
        if (thd->rand_used)
        {
          Rand_log_event e(thd,thd->rand_saved_seed1,thd->rand_saved_seed2);
          if (e.write(file))
            goto err;
        }
        if (thd->user_var_events.elements)
        {
          for (uint i= 0; i < thd->user_var_events.elements; i++)
          {
            BINLOG_USER_VAR_EVENT *user_var_event;
3735
            get_dynamic(&thd->user_var_events,(uchar*) &user_var_event, i);
3736 3737 3738 3739 3740 3741 3742 3743 3744 3745
            User_var_log_event e(thd, user_var_event->user_var_event->name.str,
                                 user_var_event->user_var_event->name.length,
                                 user_var_event->value,
                                 user_var_event->length,
                                 user_var_event->type,
                                 user_var_event->charset_number);
            if (e.write(file))
              goto err;
          }
        }
3746
      }
3747
    }
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3748

3749 3750
    /*
       Write the SQL command
3751
     */
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3752

monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3753
    if (event_info->write(file))
3754
      goto err;
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
3755

3756
    if (file == &log_file) // we are writing to the real log (disk)
3757
    {
3758
      if (flush_and_sync())
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3759
	goto err;
3760 3761
      signal_update();
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
3762
    }
monty@narttu.mysql.fi's avatar
monty@narttu.mysql.fi committed
3763
    error=0;
3764

3765 3766 3767 3768
err:
    if (error)
    {
      if (my_errno == EFBIG)
bell@sanja.is.com.ua's avatar
bell@sanja.is.com.ua committed
3769
	my_message(ER_TRANS_CACHE_FULL, ER(ER_TRANS_CACHE_FULL), MYF(0));
3770 3771 3772 3773 3774
      else
	my_error(ER_ERROR_ON_WRITE, MYF(0), name, errno);
      write_error=1;
    }
  }
3775

3776 3777 3778
  if (event_info->flags & LOG_EVENT_UPDATE_TABLE_MAP_VERSION_F)
    ++m_table_map_version;

3779
  pthread_mutex_unlock(&LOCK_log);
3780
  DBUG_RETURN(error);
3781 3782
}

3783 3784 3785 3786 3787 3788 3789 3790 3791

int error_log_print(enum loglevel level, const char *format,
                    va_list args)
{
  return logger.error_log_print(level, format, args);
}


bool slow_log_print(THD *thd, const char *query, uint query_length,
3792
                    ulonglong current_utime)
3793
{
3794
  return logger.slow_log_print(thd, query, query_length, current_utime);
3795 3796 3797
}


3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825
bool LOGGER::log_command(THD *thd, enum enum_server_command command)
{
#ifndef NO_EMBEDDED_ACCESS_CHECKS
  Security_context *sctx= thd->security_ctx;
#endif
  /*
    Log command if we have at least one log event handler enabled and want
    to log this king of commands
  */
  if (*general_log_handler_list && (what_to_log & (1L << (uint) command)))
  {
    if ((thd->options & OPTION_LOG_OFF)
#ifndef NO_EMBEDDED_ACCESS_CHECKS
         && (sctx->master_access & SUPER_ACL)
#endif
       )
    {
      /* No logging */
      return FALSE;
    }

    return TRUE;
  }

  return FALSE;
}


3826 3827 3828 3829 3830 3831
bool general_log_print(THD *thd, enum enum_server_command command,
                       const char *format, ...)
{
  va_list args;
  uint error= 0;

3832 3833 3834 3835
  /* Print the message to the buffer if we want to log this king of commands */
  if (! logger.log_command(thd, command))
    return FALSE;

3836 3837 3838 3839 3840 3841 3842
  va_start(args, format);
  error= logger.general_log_print(thd, command, format, args);
  va_end(args);

  return error;
}

3843 3844 3845 3846 3847 3848 3849 3850 3851 3852
bool general_log_write(THD *thd, enum enum_server_command command,
                       const char *query, uint query_length)
{
  /* Write the message to the log if we want to log this king of commands */
  if (logger.log_command(thd, command))
    return logger.general_log_write(thd, command, query, query_length);

  return FALSE;
}

3853
void MYSQL_BIN_LOG::rotate_and_purge(uint flags)
3854
{
3855 3856
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_lock(&LOCK_log);
3857 3858
  if ((flags & RP_FORCE_ROTATE) ||
      (my_b_tell(&log_file) >= (my_off_t) max_size))
3859
  {
3860
    new_file_without_locking();
3861 3862 3863
#ifdef HAVE_REPLICATION
    if (expire_logs_days)
    {
3864
      time_t purge_time= my_time(0) - expire_logs_days*24*60*60;
3865 3866 3867 3868 3869
      if (purge_time >= 0)
        purge_logs_before_date(purge_time);
    }
#endif
  }
3870 3871
  if (!(flags & RP_LOCK_LOG_IS_ALREADY_LOCKED))
    pthread_mutex_unlock(&LOCK_log);
3872
}
3873

3874
uint MYSQL_BIN_LOG::next_file_id()
3875 3876 3877 3878 3879 3880 3881 3882
{
  uint res;
  pthread_mutex_lock(&LOCK_log);
  res = file_id++;
  pthread_mutex_unlock(&LOCK_log);
  return res;
}

3883

3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903
/*
  Write the contents of a cache to the binary log.

  SYNOPSIS
    write_cache()
    cache    Cache to write to the binary log
    lock_log True if the LOCK_log mutex should be aquired, false otherwise
    sync_log True if the log should be flushed and sync:ed

  DESCRIPTION
    Write the contents of the cache to the binary log. The cache will
    be reset as a READ_CACHE to be able to read the contents from it.
 */

int MYSQL_BIN_LOG::write_cache(IO_CACHE *cache, bool lock_log, bool sync_log)
{
  Mutex_sentry sentry(lock_log ? &LOCK_log : NULL);

  if (reinit_io_cache(cache, READ_CACHE, 0, 0, 0))
    return ER_ERROR_ON_WRITE;
3904
  uint length= my_b_bytes_in_cache(cache), group, carry, hdr_offs;
3905 3906 3907
  long val;
  uchar header[LOG_EVENT_HEADER_LEN];

3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921
  /*
    The events in the buffer have incorrect end_log_pos data
    (relative to beginning of group rather than absolute),
    so we'll recalculate them in situ so the binlog is always
    correct, even in the middle of a group. This is possible
    because we now know the start position of the group (the
    offset of this cache in the log, if you will); all we need
    to do is to find all event-headers, and add the position of
    the group to the end_log_pos of each event.  This is pretty
    straight forward, except that we read the cache in segments,
    so an event-header might end up on the cache-border and get
    split.
  */

3922
  group= (uint)my_b_tell(&log_file);
3923 3924
  hdr_offs= carry= 0;

3925 3926
  do
  {
3927 3928 3929 3930 3931 3932

    /*
      if we only got a partial header in the last iteration,
      get the other half now and process a full header.
    */
    if (unlikely(carry > 0))
3933 3934 3935
    {
      DBUG_ASSERT(carry < LOG_EVENT_HEADER_LEN);

3936
      /* assemble both halves */
3937 3938
      memcpy(&header[carry], (char *)cache->read_pos, LOG_EVENT_HEADER_LEN - carry);

3939
      /* fix end_log_pos */
3940 3941 3942
      val= uint4korr(&header[LOG_POS_OFFSET]) + group;
      int4store(&header[LOG_POS_OFFSET], val);

3943
      /* write the first half of the split header */
3944 3945 3946
      if (my_b_write(&log_file, header, carry))
        return ER_ERROR_ON_WRITE;

3947 3948 3949 3950
      /*
        copy fixed second half of header to cache so the correct
        version will be written later.
      */
3951 3952
      memcpy((char *)cache->read_pos, &header[carry], LOG_EVENT_HEADER_LEN - carry);

3953
      /* next event header at ... */
3954
      hdr_offs = uint4korr(&header[EVENT_LEN_OFFSET]) - carry;
3955 3956 3957 3958

      carry= 0;
    }

3959 3960
    /* if there is anything to write, process it. */

3961
    if (likely(length > 0))
3962
    {
3963
      /*
3964 3965 3966 3967
        process all event-headers in this (partial) cache.
        if next header is beyond current read-buffer,
        we'll get it later (though not necessarily in the
        very next iteration, just "eventually").
3968
      */
3969

3970
      while (hdr_offs < length)
3971
      {
3972 3973 3974 3975
        /*
          partial header only? save what we can get, process once
          we get the rest.
        */
3976

3977
        if (hdr_offs + LOG_EVENT_HEADER_LEN > length)
3978
        {
3979
          carry= length - hdr_offs;
3980
          memcpy(header, (char *)cache->read_pos + hdr_offs, carry);
3981
          length= hdr_offs;
3982 3983 3984 3985
        }
        else
        {
          /* we've got a full event-header, and it came in one piece */
3986

3987
          uchar *log_pos= (uchar *)cache->read_pos + hdr_offs + LOG_POS_OFFSET;
3988

3989 3990 3991
          /* fix end_log_pos */
          val= uint4korr(log_pos) + group;
          int4store(log_pos, val);
3992

3993 3994 3995
          /* next event header at ... */
          log_pos= (uchar *)cache->read_pos + hdr_offs + EVENT_LEN_OFFSET;
          hdr_offs += uint4korr(log_pos);
3996

3997
        }
3998
      }
3999 4000

      /*
4001 4002 4003 4004 4005 4006
        Adjust hdr_offs. Note that it may still point beyond the segment
        read in the next iteration; if the current event is very long,
        it may take a couple of read-iterations (and subsequent adjustments
        of hdr_offs) for it to point into the then-current segment.
        If we have a split header (!carry), hdr_offs will be set at the
        beginning of the next iteration, overwriting the value we set here:
4007
      */
4008
      hdr_offs -= length;
4009 4010 4011
    }

    /* Write data to the binary log file */
4012
    if (my_b_write(&log_file, cache->read_pos, length))
4013
      return ER_ERROR_ON_WRITE;
4014
    cache->read_pos=cache->read_end;		// Mark buffer used up
4015
  } while ((length= my_b_fill(cache)));
4016

4017
  DBUG_ASSERT(carry == 0);
4018 4019 4020 4021 4022 4023 4024

  if (sync_log)
    flush_and_sync();

  return 0;                                     // All OK
}

4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035
/**
  Write a cached log entry to the binary log.
  - To support transaction over replication, we wrap the transaction
  with BEGIN/COMMIT or BEGIN/ROLLBACK in the binary log.
  We want to write a BEGIN/ROLLBACK block when a non-transactional table
  was updated in a transaction which was rolled back. This is to ensure
  that the same updates are run on the slave.

  @param thd
  @param cache		The cache to copy to the binlog
  @param commit_event   The commit event to print after writing the
4036
                        contents of the cache.
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4037

4038 4039 4040 4041 4042 4043
  @note
    We only come here if there is something in the cache.
  @note
    The thing in the cache is always a complete transaction.
  @note
    'cache' needs to be reinitialized after this functions returns.
4044 4045
*/

4046
bool MYSQL_BIN_LOG::write(THD *thd, IO_CACHE *cache, Log_event *commit_event)
4047
{
4048
  DBUG_ENTER("MYSQL_BIN_LOG::write(THD *, IO_CACHE *, Log_event *)");
4049
  VOID(pthread_mutex_lock(&LOCK_log));
4050

4051 4052 4053
  /* NULL would represent nothing to replicate after ROLLBACK */
  DBUG_ASSERT(commit_event != NULL);

4054
  DBUG_ASSERT(is_open());
4055
  if (likely(is_open()))                       // Should always be true
4056
  {
4057
    /*
4058 4059 4060 4061
      We only bother to write to the binary log if there is anything
      to write.
     */
    if (my_b_tell(cache) > 0)
4062
    {
4063
      /*
4064 4065 4066
        Log "BEGIN" at the beginning of every transaction.  Here, a
        transaction is either a BEGIN..COMMIT block or a single
        statement in autocommit mode.
4067
      */
4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099
      Query_log_event qinfo(thd, STRING_WITH_LEN("BEGIN"), TRUE, FALSE);
      /*
        Imagine this is rollback due to net timeout, after all
        statements of the transaction succeeded. Then we want a
        zero-error code in BEGIN.  In other words, if there was a
        really serious error code it's already in the statement's
        events, there is no need to put it also in this internally
        generated event, and as this event is generated late it would
        lead to false alarms.

        This is safer than thd->clear_error() against kills at shutdown.
      */
      qinfo.error_code= 0;
      /*
        Now this Query_log_event has artificial log_pos 0. It must be
        adjusted to reflect the real position in the log. Not doing it
        would confuse the slave: it would prevent this one from
        knowing where he is in the master's binlog, which would result
        in wrong positions being shown to the user, MASTER_POS_WAIT
        undue waiting etc.
      */
      if (qinfo.write(&log_file))
        goto err;

      DBUG_EXECUTE_IF("crash_before_writing_xid",
                      {
                        if ((write_error= write_cache(cache, false, true)))
                          DBUG_PRINT("info", ("error writing binlog cache: %d",
                                               write_error));
                        DBUG_PRINT("info", ("crashing before writing xid"));
                        abort();
                      });
4100

4101 4102
      if ((write_error= write_cache(cache, false, false)))
        goto err;
4103

4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115
      if (commit_event && commit_event->write(&log_file))
        goto err;
      if (flush_and_sync())
        goto err;
      DBUG_EXECUTE_IF("half_binlogged_transaction", abort(););
      if (cache->error)				// Error on read
      {
        sql_print_error(ER(ER_ERROR_ON_READ), cache->file_name, errno);
        write_error=1;				// Don't give more errors
        goto err;
      }
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4116
    }
4117

4118 4119 4120 4121 4122 4123 4124 4125
    /*
      if commit_event is Xid_log_event, increase the number of
      prepared_xids (it's decreasd in ::unlog()). Binlog cannot be rotated
      if there're prepared xids in it - see the comment in new_file() for
      an explanation.
      If the commit_event is not Xid_log_event (then it's a Query_log_event)
      rotate binlog, if necessary.
    */
4126
    if (commit_event && commit_event->get_type_code() == XID_EVENT)
4127 4128 4129 4130 4131
    {
      pthread_mutex_lock(&LOCK_prep_xids);
      prepared_xids++;
      pthread_mutex_unlock(&LOCK_prep_xids);
    }
4132 4133
    else
      rotate_and_purge(RP_LOCK_LOG_IS_ALREADY_LOCKED);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4134
  }
4135
  VOID(pthread_mutex_unlock(&LOCK_log));
heikki@hundin.mysql.fi's avatar
heikki@hundin.mysql.fi committed
4136

4137
  DBUG_RETURN(0);
4138 4139

err:
4140 4141 4142 4143 4144
  if (!write_error)
  {
    write_error= 1;
    sql_print_error(ER(ER_ERROR_ON_WRITE), name, errno);
  }
4145
  VOID(pthread_mutex_unlock(&LOCK_log));
4146
  DBUG_RETURN(1);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4147 4148
}

4149

4150 4151
/**
  Wait until we get a signal that the binary log has been updated.
4152

4153 4154 4155 4156
  @param thd		Thread variable
  @param is_slave      If 0, the caller is the Binlog_dump thread from master;
                       if 1, the caller is the SQL thread from the slave. This
                       influences only thd->proc_info.
4157

4158
  @note
4159
    One must have a lock on LOCK_log before calling this function.
4160
    This lock will be released before return! That's required by
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4161
    THD::enter_cond() (see NOTES in sql_class.h).
4162 4163
*/

4164
void MYSQL_BIN_LOG::wait_for_update(THD* thd, bool is_slave)
4165
{
monty@mysql.com's avatar
monty@mysql.com committed
4166 4167
  const char *old_msg;
  DBUG_ENTER("wait_for_update");
4168

monty@mysql.com's avatar
monty@mysql.com committed
4169
  old_msg= thd->enter_cond(&update_cond, &LOCK_log,
4170
                           is_slave ?
monty@mysql.com's avatar
monty@mysql.com committed
4171
                           "Has read all relay log; waiting for the slave I/O "
4172
                           "thread to update it" :
monty@mysql.com's avatar
monty@mysql.com committed
4173
                           "Has sent all binlog to slave; waiting for binlog "
4174
                           "to be updated");
4175 4176
  pthread_cond_wait(&update_cond, &LOCK_log);
  thd->exit_cond(old_msg);
monty@mysql.com's avatar
monty@mysql.com committed
4177
  DBUG_VOID_RETURN;
4178
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4179

4180

4181 4182
/**
  Close the log file.
4183

4184 4185 4186 4187 4188
  @param exiting     Bitmask for one or more of the following bits:
          - LOG_CLOSE_INDEX : if we should close the index file
          - LOG_CLOSE_TO_BE_OPENED : if we intend to call open
                                     at once after close.
          - LOG_CLOSE_STOP_EVENT : write a 'stop' event to the log
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
4189

4190
  @note
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
4191 4192
    One can do an open on the object at once after doing a close.
    The internal structures are not freed until cleanup() is called
4193 4194
*/

4195
void MYSQL_BIN_LOG::close(uint exiting)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4196
{					// One can't set log_type here!
4197
  DBUG_ENTER("MYSQL_BIN_LOG::close");
4198
  DBUG_PRINT("enter",("exiting: %d", (int) exiting));
4199
  if (log_state == LOG_OPENED)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4200
  {
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4201
#ifdef HAVE_REPLICATION
4202 4203
    if (log_type == LOG_BIN && !no_auto_events &&
	(exiting & LOG_CLOSE_STOP_EVENT))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4204 4205
    {
      Stop_log_event s;
4206
      s.write(&log_file);
4207
      bytes_written+= s.data_written;
4208
      signal_update();
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4209
    }
hf@deer.mysql.r18.ru's avatar
SCRUM  
hf@deer.mysql.r18.ru committed
4210
#endif /* HAVE_REPLICATION */
4211 4212

    /* don't pwrite in a file opened with O_APPEND - it doesn't work */
4213
    if (log_file.type == WRITE_CACHE && log_type == LOG_BIN)
4214 4215
    {
      my_off_t offset= BIN_LOG_HEADER_SIZE + FLAGS_OFFSET;
4216 4217
      my_off_t org_position= my_tell(log_file.file, MYF(0));
      uchar flags= 0;            // clearing LOG_EVENT_BINLOG_IN_USE_F
4218
      my_pwrite(log_file.file, &flags, 1, offset, MYF(0));
4219 4220 4221 4222 4223 4224 4225
      /*
        Restore position so that anything we have in the IO_cache is written
        to the correct position.
        We need the seek here, as my_pwrite() is not guaranteed to keep the
        original position on system that doesn't support pwrite().
      */
      my_seek(log_file.file, org_position, MY_SEEK_SET, MYF(0));
4226 4227
    }

4228 4229
    /* this will cleanup IO_CACHE, sync and close the file */
    MYSQL_LOG::close(exiting);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4230
  }
4231 4232 4233 4234 4235 4236

  /*
    The following test is needed even if is_open() is not set, as we may have
    called a not complete close earlier and the index file is still open.
  */

4237
  if ((exiting & LOG_CLOSE_INDEX) && my_b_inited(&index_file))
4238
  {
4239 4240
    end_io_cache(&index_file);
    if (my_close(index_file.file, MYF(0)) < 0 && ! write_error)
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4241
    {
4242 4243
      write_error= 1;
      sql_print_error(ER(ER_ERROR_ON_WRITE), index_file_name, errno);
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4244
    }
4245
  }
4246
  log_state= (exiting & LOG_CLOSE_TO_BE_OPENED) ? LOG_TO_BE_OPENED : LOG_CLOSED;
4247
  safeFree(name);
4248
  DBUG_VOID_RETURN;
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4249 4250 4251
}


4252
void MYSQL_BIN_LOG::set_max_size(ulong max_size_arg)
4253 4254 4255 4256 4257 4258 4259 4260
{
  /*
    We need to take locks, otherwise this may happen:
    new_file() is called, calls open(old_max_size), then before open() starts,
    set_max_size() sets max_size to max_size_arg, then open() starts and
    uses the old_max_size argument, so max_size_arg has been overwritten and
    it's like if the SET command was never run.
  */
4261
  DBUG_ENTER("MYSQL_BIN_LOG::set_max_size");
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4262
  pthread_mutex_lock(&LOCK_log);
4263 4264
  if (is_open())
    max_size= max_size_arg;
guilhem@mysql.com's avatar
guilhem@mysql.com committed
4265 4266
  pthread_mutex_unlock(&LOCK_log);
  DBUG_VOID_RETURN;
4267
}
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4268

4269

4270 4271
/**
  Check if a string is a valid number.
4272

4273 4274 4275
  @param str			String to test
  @param res			Store value here
  @param allow_wildcards	Set to 1 if we should ignore '%' and '_'
4276

4277
  @note
4278 4279 4280
    For the moment the allow_wildcards argument is not used
    Should be move to some other file.

4281
  @retval
4282
    1	String is a number
4283
  @retval
4284
    0	Error
4285
*/
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297

static bool test_if_number(register const char *str,
			   long *res, bool allow_wildcards)
{
  reg2 int flag;
  const char *start;
  DBUG_ENTER("test_if_number");

  flag=0; start=str;
  while (*str++ == ' ') ;
  if (*--str == '-' || *str == '+')
    str++;
4298 4299
  while (my_isdigit(files_charset_info,*str) ||
	 (allow_wildcards && (*str == wild_many || *str == wild_one)))
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4300 4301 4302 4303 4304 4305 4306
  {
    flag=1;
    str++;
  }
  if (*str == '.')
  {
    for (str++ ;
4307
	 my_isdigit(files_charset_info,*str) ||
bk@work.mysql.com's avatar
bk@work.mysql.com committed
4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326
	   (allow_wildcards && (*str == wild_many || *str == wild_one)) ;
	 str++, flag=1) ;
  }
  if (*str != 0 || flag == 0)
    DBUG_RETURN(0);
  if (res)
    *res=atol(start);
  DBUG_RETURN(1);			/* Number ok */
} /* test_if_number */


void sql_perror(const char *message)
{
#ifdef HAVE_STRERROR
  sql_print_error("%s: %s",message, strerror(errno));
#else
  perror(message);
#endif
}
4327

monty@mysql.com's avatar
monty@mysql.com committed
4328

4329 4330 4331
bool flush_error_log()
{
  bool result=0;
4332
  if (opt_error_log)
4333 4334 4335 4336
  {
    char err_renamed[FN_REFLEN], *end;
    end= strmake(err_renamed,log_error_file,FN_REFLEN-4);
    strmov(end, "-old");
4337
    VOID(pthread_mutex_lock(&LOCK_error_log));
4338 4339 4340 4341 4342 4343
#ifdef __WIN__
    char err_temp[FN_REFLEN+4];
    /*
     On Windows is necessary a temporary file for to rename
     the current error file.
    */
4344
    strxmov(err_temp, err_renamed,"-tmp",NullS);
4345 4346 4347
    (void) my_delete(err_temp, MYF(0)); 
    if (freopen(err_temp,"a+",stdout))
    {
4348 4349 4350 4351
      int fd;
      size_t bytes;
      uchar buf[IO_SIZE];

4352 4353 4354 4355 4356
      freopen(err_temp,"a+",stderr);
      (void) my_delete(err_renamed, MYF(0));
      my_rename(log_error_file,err_renamed,MYF(0));
      if (freopen(log_error_file,"a+",stdout))
        freopen(log_error_file,"a+",stderr);
4357

4358 4359
      if ((fd = my_open(err_temp, O_RDONLY, MYF(0))) >= 0)
      {
4360 4361 4362
        while ((bytes= my_read(fd, buf, IO_SIZE, MYF(0))) &&
               bytes != MY_FILE_ERROR)
          my_fwrite(stderr, buf, bytes, MYF(0));
4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375
        my_close(fd, MYF(0));
      }
      (void) my_delete(err_temp, MYF(0)); 
    }
    else
     result= 1;
#else
   my_rename(log_error_file,err_renamed,MYF(0));
   if (freopen(log_error_file,"a+",stdout))
     freopen(log_error_file,"a+",stderr);
   else
     result= 1;
#endif
4376
    VOID(pthread_mutex_unlock(&LOCK_error_log));
4377 4378 4379
  }
   return result;
}
monty@mashka.mysql.fi's avatar
monty@mashka.mysql.fi committed
4380

4381
void MYSQL_BIN_LOG::signal_update()
monty@mysql.com's avatar
monty@mysql.com committed
4382
{
4383
  DBUG_ENTER("MYSQL_BIN_LOG::signal_update");
monty@mysql.com's avatar
monty@mysql.com committed
4384 4385 4386 4387
  pthread_cond_broadcast(&update_cond);
  DBUG_VOID_RETURN;
}

4388
#ifdef __NT__
4389
static void print_buffer_to_nt_eventlog(enum loglevel level, char *buff,
4390
                                        size_t length, size_t buffLen)
4391 4392
{
  HANDLE event;
4393
  char   *buffptr= buff;
serg@serg.mylan's avatar
serg@serg.mylan committed
4394
  DBUG_ENTER("print_buffer_to_nt_eventlog");
4395

4396 4397
  /* Add ending CR/LF's to string, overwrite last chars if necessary */
  strmov(buffptr+min(length, buffLen-5), "\r\n\r\n");
4398

monty@mysql.com's avatar
monty@mysql.com committed
4399 4400
  setup_windows_event_source();
  if ((event= RegisterEventSource(NULL,"MySQL")))
4401
  {
monty@mysql.com's avatar
monty@mysql.com committed
4402
    switch (level) {
4403
      case ERROR_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4404
        ReportEvent(event, EVENTLOG_ERROR_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
4405
                    (LPCSTR*)&buffptr, NULL);
4406 4407
        break;
      case WARNING_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4408
        ReportEvent(event, EVENTLOG_WARNING_TYPE, 0, MSG_DEFAULT, NULL, 1, 0,
4409
                    (LPCSTR*) &buffptr, NULL);
4410 4411
        break;
      case INFORMATION_LEVEL:
monty@mysql.com's avatar
monty@mysql.com committed
4412
        ReportEvent(event, EVENTLOG_INFORMATION_TYPE, 0, MSG_DEFAULT, NULL, 1,
4413
                    0, (LPCSTR*) &buffptr, NULL);
4414 4415 4416 4417 4418 4419 4420
        break;
    }
    DeregisterEventSource(event);
  }

  DBUG_VOID_RETURN;
}
monty@mysql.com's avatar
monty@mysql.com committed
4421 4422
#endif /* __NT__ */

4423

4424
/**
monty@mysql.com's avatar
monty@mysql.com committed
4425 4426
  Prints a printf style message to the error log and, under NT, to the
  Windows event log.
4427

4428 4429
  This function prints the message into a buffer and then sends that buffer
  to other functions to write that message to other logging sources.
4430

4431 4432 4433
  @param event_type          Type of event to write (Error, Warning, or Info)
  @param format              Printf style format of message
  @param args                va_list list of arguments for the message
4434

4435
  @returns
4436 4437 4438
    The function always returns 0. The return value is present in the
    signature to be compatible with other logging routines, which could
    return an error (e.g. logging to the log tables)
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4439
*/
monty@mysql.com's avatar
monty@mysql.com committed
4440

4441
#ifdef EMBEDDED_LIBRARY
rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
4442
int vprint_msg_to_log(enum loglevel level __attribute__((unused)),
4443 4444
                       const char *format __attribute__((unused)),
                       va_list argsi __attribute__((unused)))
rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
4445 4446 4447 4448
{
  DBUG_ENTER("vprint_msg_to_log");
  DBUG_RETURN(0);
}
4449
#else /*!EMBEDDED_LIBRARY*/
4450 4451 4452 4453 4454 4455 4456 4457 4458 4459
static void print_buffer_to_file(enum loglevel level, const char *buffer)
{
  time_t skr;
  struct tm tm_tmp;
  struct tm *start;
  DBUG_ENTER("print_buffer_to_file");
  DBUG_PRINT("enter",("buffer: %s", buffer));

  VOID(pthread_mutex_lock(&LOCK_error_log));

4460
  skr= my_time(0);
4461 4462
  localtime_r(&skr, &tm_tmp);
  start=&tm_tmp;
4463

4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481
  fprintf(stderr, "%02d%02d%02d %2d:%02d:%02d [%s] %s\n",
          start->tm_year % 100,
          start->tm_mon+1,
          start->tm_mday,
          start->tm_hour,
          start->tm_min,
          start->tm_sec,
          (level == ERROR_LEVEL ? "ERROR" : level == WARNING_LEVEL ?
           "Warning" : "Note"),
          buffer);

  fflush(stderr);

  VOID(pthread_mutex_unlock(&LOCK_error_log));
  DBUG_VOID_RETURN;
}


rburnett@production.mysql.com's avatar
rburnett@production.mysql.com committed
4482
int vprint_msg_to_log(enum loglevel level, const char *format, va_list args)
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4483 4484
{
  char   buff[1024];
4485
  size_t length;
4486
  DBUG_ENTER("vprint_msg_to_log");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4487

4488
  length= my_vsnprintf(buff, sizeof(buff), format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4489
  print_buffer_to_file(level, buff);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4490 4491

#ifdef __NT__
monty@mysql.com's avatar
monty@mysql.com committed
4492
  print_buffer_to_nt_eventlog(level, buff, length, sizeof(buff));
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4493
#endif
4494

4495
  DBUG_RETURN(0);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4496
}
4497
#endif /*EMBEDDED_LIBRARY*/
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4498

4499

monty@mysql.com's avatar
monty@mysql.com committed
4500
void sql_print_error(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4501 4502
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4503
  DBUG_ENTER("sql_print_error");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4504

serg@serg.mylan's avatar
serg@serg.mylan committed
4505
  va_start(args, format);
4506
  error_log_print(ERROR_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4507
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4508 4509 4510 4511 4512

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
4513
void sql_print_warning(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4514 4515
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4516
  DBUG_ENTER("sql_print_warning");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4517

serg@serg.mylan's avatar
serg@serg.mylan committed
4518
  va_start(args, format);
4519
  error_log_print(WARNING_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4520
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4521 4522 4523 4524 4525

  DBUG_VOID_RETURN;
}


monty@mysql.com's avatar
monty@mysql.com committed
4526
void sql_print_information(const char *format, ...) 
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4527 4528
{
  va_list args;
serg@serg.mylan's avatar
serg@serg.mylan committed
4529
  DBUG_ENTER("sql_print_information");
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4530

serg@serg.mylan's avatar
serg@serg.mylan committed
4531
  va_start(args, format);
4532
  error_log_print(INFORMATION_LEVEL, format, args);
serg@serg.mylan's avatar
serg@serg.mylan committed
4533
  va_end(args);
rburnett@build.mysql.com's avatar
rburnett@build.mysql.com committed
4534

guilhem@mysql.com's avatar
guilhem@mysql.com committed
4535 4536
  DBUG_VOID_RETURN;
}
4537

4538

4539 4540 4541 4542 4543 4544 4545 4546
/********* transaction coordinator log for 2pc - mmap() based solution *******/

/*
  the log consists of a file, mmapped to a memory.
  file is divided on pages of tc_log_page_size size.
  (usable size of the first page is smaller because of log header)
  there's PAGE control structure for each page
  each page (or rather PAGE control structure) can be in one of three
4547 4548
  states - active, syncing, pool.
  there could be only one page in active or syncing states,
4549
  but many in pool - pool is fifo queue.
4550
  usual lifecycle of a page is pool->active->syncing->pool
4551
  "active" page - is a page where new xid's are logged.
4552 4553
  the page stays active as long as syncing slot is taken.
  "syncing" page is being synced to disk. no new xid can be added to it.
4554
  when the sync is done the page is moved to a pool and an active page
4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567
  becomes "syncing".

  the result of such an architecture is a natural "commit grouping" -
  If commits are coming faster than the system can sync, they do not
  stall. Instead, all commit that came since the last sync are
  logged to the same page, and they all are synced with the next -
  one - sync. Thus, thought individual commits are delayed, throughput
  is not decreasing.

  when a xid is added to an active page, the thread of this xid waits
  for a page's condition until the page is synced. when syncing slot
  becomes vacant one of these waiters is awaken to take care of syncing.
  it syncs the page and signals all waiters that the page is synced.
4568 4569 4570 4571
  PAGE::waiters is used to count these waiters, and a page may never
  become active again until waiters==0 (that is all waiters from the
  previous sync have noticed the sync was completed)

4572 4573
  note, that the page becomes "dirty" and has to be synced only when a
  new xid is added into it. Removing a xid from a page does not make it
4574 4575
  dirty - we don't sync removals to disk.
*/
4576 4577 4578 4579 4580

ulong tc_log_page_waits= 0;

#ifdef HAVE_MMAP

4581 4582
#define TC_LOG_HEADER_SIZE (sizeof(tc_log_magic)+1)

4583
static const char tc_log_magic[]={(char) 254, 0x23, 0x05, 0x74};
4584

4585
ulong opt_tc_log_size= TC_LOG_MIN_SIZE;
4586
ulong tc_log_max_pages_used=0, tc_log_page_size=0, tc_log_cur_pages_used=0;
4587 4588 4589 4590 4591 4592 4593

int TC_LOG_MMAP::open(const char *opt_name)
{
  uint i;
  bool crashed=FALSE;
  PAGE *pg;

4594
  DBUG_ASSERT(total_ha_2pc > 1);
4595 4596
  DBUG_ASSERT(opt_name && opt_name[0]);

serg@serg.mylan's avatar
serg@serg.mylan committed
4597
  tc_log_page_size= my_getpagesize();
4598 4599 4600
  DBUG_ASSERT(TC_LOG_PAGE_SIZE % tc_log_page_size == 0);

  fn_format(logname,opt_name,mysql_data_home,"",MY_UNPACK_FILENAME);
4601
  if ((fd= my_open(logname, O_RDWR, MYF(0))) < 0)
4602
  {
4603 4604
    if (my_errno != ENOENT)
      goto err;
4605 4606
    if (using_heuristic_recover())
      return 1;
4607
    if ((fd= my_create(logname, CREATE_MODE, O_RDWR, MYF(MY_WME))) < 0)
4608 4609
      goto err;
    inited=1;
4610
    file_length= opt_tc_log_size;
4611 4612 4613 4614 4615
    if (my_chsize(fd, file_length, 0, MYF(MY_WME)))
      goto err;
  }
  else
  {
4616 4617
    inited= 1;
    crashed= TRUE;
serg@serg.mylan's avatar
serg@serg.mylan committed
4618
    sql_print_information("Recovering after a crash using %s", opt_name);
4619 4620 4621 4622 4623 4624
    if (tc_heuristic_recover)
    {
      sql_print_error("Cannot perform automatic crash recovery when "
                      "--tc-heuristic-recover is used");
      goto err;
    }
4625
    file_length= my_seek(fd, 0L, MY_SEEK_END, MYF(MY_WME+MY_FAE));
4626 4627 4628 4629
    if (file_length == MY_FILEPOS_ERROR || file_length % tc_log_page_size)
      goto err;
  }

4630
  data= (uchar *)my_mmap(0, (size_t)file_length, PROT_READ|PROT_WRITE,
4631 4632 4633 4634 4635 4636 4637 4638
                        MAP_NOSYNC|MAP_SHARED, fd, 0);
  if (data == MAP_FAILED)
  {
    my_errno=errno;
    goto err;
  }
  inited=2;

4639
  npages=(uint)file_length/tc_log_page_size;
4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656
  DBUG_ASSERT(npages >= 3);             // to guarantee non-empty pool
  if (!(pages=(PAGE *)my_malloc(npages*sizeof(PAGE), MYF(MY_WME|MY_ZEROFILL))))
    goto err;
  inited=3;
  for (pg=pages, i=0; i < npages; i++, pg++)
  {
    pg->next=pg+1;
    pg->waiters=0;
    pg->state=POOL;
    pthread_mutex_init(&pg->lock, MY_MUTEX_INIT_FAST);
    pthread_cond_init (&pg->cond, 0);
    pg->start=(my_xid *)(data + i*tc_log_page_size);
    pg->end=(my_xid *)(pg->start + tc_log_page_size);
    pg->size=pg->free=tc_log_page_size/sizeof(my_xid);
  }
  pages[0].size=pages[0].free=
                (tc_log_page_size-TC_LOG_HEADER_SIZE)/sizeof(my_xid);
4657
  pages[0].start=pages[0].end-pages[0].size;
4658 4659 4660
  pages[npages-1].next=0;
  inited=4;

4661
  if (crashed && recover())
4662
      goto err;
4663

4664
  memcpy(data, tc_log_magic, sizeof(tc_log_magic));
4665
  data[sizeof(tc_log_magic)]= (uchar)total_ha_2pc;
4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676
  my_msync(fd, data, tc_log_page_size, MS_SYNC);
  inited=5;

  pthread_mutex_init(&LOCK_sync,    MY_MUTEX_INIT_FAST);
  pthread_mutex_init(&LOCK_active,  MY_MUTEX_INIT_FAST);
  pthread_mutex_init(&LOCK_pool,    MY_MUTEX_INIT_FAST);
  pthread_cond_init(&COND_active, 0);
  pthread_cond_init(&COND_pool, 0);

  inited=6;

4677
  syncing= 0;
4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688
  active=pages;
  pool=pages+1;
  pool_last=pages+npages-1;

  return 0;

err:
  close();
  return 1;
}

4689 4690
/**
  there is no active page, let's got one from the pool.
4691

4692 4693 4694
  Two strategies here:
    -# take the first from the pool
    -# if there're waiters - take the one with the most free space.
4695

4696 4697 4698
  @todo
    TODO page merging. try to allocate adjacent page first,
    so that they can be flushed both in one sync
4699
*/
4700

4701 4702 4703 4704 4705
void TC_LOG_MMAP::get_active_from_pool()
{
  PAGE **p, **best_p=0;
  int best_free;

4706 4707 4708
  if (syncing)
    pthread_mutex_lock(&LOCK_pool);

4709 4710 4711
  do
  {
    best_p= p= &pool;
4712 4713
    if ((*p)->waiters == 0) // can the first page be used ?
      break;                // yes - take it.
4714

4715
    best_free=0;            // no - trying second strategy
4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727
    for (p=&(*p)->next; *p; p=&(*p)->next)
    {
      if ((*p)->waiters == 0 && (*p)->free > best_free)
      {
        best_free=(*p)->free;
        best_p=p;
      }
    }
  }
  while ((*best_p == 0 || best_free == 0) && overflow());

  active=*best_p;
4728
  if (active->free == active->size) // we've chosen an empty page
4729 4730 4731 4732 4733
  {
    tc_log_cur_pages_used++;
    set_if_bigger(tc_log_max_pages_used, tc_log_cur_pages_used);
  }

4734
  if ((*best_p)->next)              // unlink the page from the pool
4735 4736 4737
    *best_p=(*best_p)->next;
  else
    pool_last=*best_p;
4738 4739 4740

  if (syncing)
    pthread_mutex_unlock(&LOCK_pool);
4741 4742
}

4743 4744 4745 4746
/**
  @todo
  perhaps, increase log size ?
*/
4747 4748 4749 4750 4751 4752 4753 4754 4755
int TC_LOG_MMAP::overflow()
{
  /*
    simple overflow handling - just wait
    TODO perhaps, increase log size ?
    let's check the behaviour of tc_log_page_waits first
  */
  tc_log_page_waits++;
  pthread_cond_wait(&COND_pool, &LOCK_pool);
4756
  return 1; // always return 1
4757 4758
}

4759 4760
/**
  Record that transaction XID is committed on the persistent storage.
4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775

    This function is called in the middle of two-phase commit:
    First all resources prepare the transaction, then tc_log->log() is called,
    then all resources commit the transaction, then tc_log->unlog() is called.

    All access to active page is serialized but it's not a problem, as
    we're assuming that fsync() will be a main bottleneck.
    That is, parallelizing writes to log pages we'll decrease number of
    threads waiting for a page, but then all these threads will be waiting
    for a fsync() anyway

   If tc_log == MYSQL_LOG then tc_log writes transaction to binlog and
   records XID in a special Xid_log_event.
   If tc_log = TC_LOG_MMAP then xid is written in a special memory-mapped
   log.
4776

4777 4778 4779 4780 4781 4782 4783
  @retval
    0  - error
  @retval
    \# - otherwise, "cookie", a number that will be passed as an argument
    to unlog() call. tc_log can define it any way it wants,
    and use for whatever purposes. TC_LOG_MMAP sets it
    to the position in memory where xid was logged to.
4784 4785
*/

4786
int TC_LOG_MMAP::log_xid(THD *thd, my_xid xid)
4787 4788 4789 4790 4791 4792 4793
{
  int err;
  PAGE *p;
  ulong cookie;

  pthread_mutex_lock(&LOCK_active);

4794 4795 4796 4797 4798 4799 4800
  /*
    if active page is full - just wait...
    frankly speaking, active->free here accessed outside of mutex
    protection, but it's safe, because it only means we may miss an
    unlog() for the active page, and we're not waiting for it here -
    unlog() does not signal COND_active.
  */
4801 4802 4803
  while (unlikely(active && active->free == 0))
    pthread_cond_wait(&COND_active, &LOCK_active);

4804
  /* no active page ? take one from the pool */
4805 4806 4807 4808 4809 4810
  if (active == 0)
    get_active_from_pool();

  p=active;
  pthread_mutex_lock(&p->lock);

4811
  /* searching for an empty slot */
4812 4813 4814 4815 4816 4817
  while (*p->ptr)
  {
    p->ptr++;
    DBUG_ASSERT(p->ptr < p->end);               // because p->free > 0
  }

4818 4819
  /* found! store xid there and mark the page dirty */
  cookie= (ulong)((uchar *)p->ptr - data);      // can never be zero
4820 4821 4822 4823 4824
  *p->ptr++= xid;
  p->free--;
  p->state= DIRTY;

  /* to sync or not to sync - this is the question */
4825
  pthread_mutex_unlock(&LOCK_active);
4826 4827 4828
  pthread_mutex_lock(&LOCK_sync);
  pthread_mutex_unlock(&p->lock);

4829
  if (syncing)
4830 4831
  {                                          // somebody's syncing. let's wait
    p->waiters++;
4832
    /*
monty@mishka.local's avatar
monty@mishka.local committed
4833
      note - it must be while (), not do ... while () here
4834 4835 4836
      as p->state may be not DIRTY when we come here
    */
    while (p->state == DIRTY && syncing)
4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847
      pthread_cond_wait(&p->cond, &LOCK_sync);
    p->waiters--;
    err= p->state == ERROR;
    if (p->state != DIRTY)                   // page was synced
    {
      if (p->waiters == 0)
        pthread_cond_signal(&COND_pool);     // in case somebody's waiting
      pthread_mutex_unlock(&LOCK_sync);
      goto done;                             // we're done
    }
  }                                          // page was not synced! do it now
4848 4849 4850 4851
  DBUG_ASSERT(active == p && syncing == 0);
  pthread_mutex_lock(&LOCK_active);
  syncing=p;                                 // place is vacant - take it
  active=0;                                  // page is not active anymore
4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864
  pthread_cond_broadcast(&COND_active);      // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_active);
  pthread_mutex_unlock(&LOCK_sync);
  err= sync();

done:
  return err ? 0 : cookie;
}

int TC_LOG_MMAP::sync()
{
  int err;

4865
  DBUG_ASSERT(syncing != active);
4866 4867 4868 4869 4870

  /*
    sit down and relax - this can take a while...
    note - no locks are held at this point
  */
4871
  err= my_msync(fd, syncing->start, 1, MS_SYNC);
4872

4873
  /* page is synced. let's move it to the pool */
4874
  pthread_mutex_lock(&LOCK_pool);
4875 4876 4877 4878 4879
  pool_last->next=syncing;
  pool_last=syncing;
  syncing->next=0;
  syncing->state= err ? ERROR : POOL;
  pthread_cond_broadcast(&syncing->cond);    // signal "sync done"
4880 4881 4882
  pthread_cond_signal(&COND_pool);           // in case somebody's waiting
  pthread_mutex_unlock(&LOCK_pool);

4883
  /* marking 'syncing' slot free */
4884
  pthread_mutex_lock(&LOCK_sync);
4885
  syncing=0;
4886 4887 4888 4889 4890
  pthread_cond_signal(&active->cond);        // wake up a new syncer
  pthread_mutex_unlock(&LOCK_sync);
  return err;
}

4891
/**
4892
  erase xid from the page, update page free space counters/pointers.
4893
  cookie points directly to the memory where xid was logged.
4894
*/
4895

4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908
void TC_LOG_MMAP::unlog(ulong cookie, my_xid xid)
{
  PAGE *p=pages+(cookie/tc_log_page_size);
  my_xid *x=(my_xid *)(data+cookie);

  DBUG_ASSERT(*x == xid);
  DBUG_ASSERT(x >= p->start && x < p->end);
  *x=0;

  pthread_mutex_lock(&p->lock);
  p->free++;
  DBUG_ASSERT(p->free <= p->size);
  set_if_smaller(p->ptr, x);
4909
  if (p->free == p->size)               // the page is completely empty
4910 4911
    statistic_decrement(tc_log_cur_pages_used, &LOCK_status);
  if (p->waiters == 0)                 // the page is in pool and ready to rock
4912
    pthread_cond_signal(&COND_pool);   // ping ... for overflow()
4913 4914 4915 4916 4917
  pthread_mutex_unlock(&p->lock);
}

void TC_LOG_MMAP::close()
{
serg@serg.mylan's avatar
serg@serg.mylan committed
4918
  uint i;
4919 4920 4921 4922 4923 4924 4925 4926 4927
  switch (inited) {
  case 6:
    pthread_mutex_destroy(&LOCK_sync);
    pthread_mutex_destroy(&LOCK_active);
    pthread_mutex_destroy(&LOCK_pool);
    pthread_cond_destroy(&COND_pool);
  case 5:
    data[0]='A'; // garble the first (signature) byte, in case my_delete fails
  case 4:
serg@serg.mylan's avatar
serg@serg.mylan committed
4928
    for (i=0; i < npages; i++)
4929 4930 4931 4932 4933 4934 4935
    {
      if (pages[i].ptr == 0)
        break;
      pthread_mutex_destroy(&pages[i].lock);
      pthread_cond_destroy(&pages[i].cond);
    }
  case 3:
4936
    my_free((uchar*)pages, MYF(0));
4937
  case 2:
4938
    my_munmap((char*)data, (size_t)file_length);
4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963
  case 1:
    my_close(fd, MYF(0));
  }
  if (inited>=5) // cannot do in the switch because of Windows
    my_delete(logname, MYF(MY_WME));
  inited=0;
}

int TC_LOG_MMAP::recover()
{
  HASH xids;
  PAGE *p=pages, *end_p=pages+npages;

  if (memcmp(data, tc_log_magic, sizeof(tc_log_magic)))
  {
    sql_print_error("Bad magic header in tc log");
    goto err1;
  }

  /*
    the first byte after magic signature is set to current
    number of storage engines on startup
  */
  if (data[sizeof(tc_log_magic)] != total_ha_2pc)
  {
4964
    sql_print_error("Recovery failed! You must enable "
4965 4966 4967
                    "exactly %d storage engines that support "
                    "two-phase commit protocol",
                    data[sizeof(tc_log_magic)]);
4968 4969 4970 4971
    goto err1;
  }

  if (hash_init(&xids, &my_charset_bin, tc_log_page_size/3, 0,
4972
                sizeof(my_xid), 0, 0, MYF(0)))
4973 4974 4975 4976 4977
    goto err1;

  for ( ; p < end_p ; p++)
  {
    for (my_xid *x=p->start; x < p->end; x++)
4978
      if (*x && my_hash_insert(&xids, (uchar *)x))
4979 4980 4981 4982 4983 4984 4985
        goto err2; // OOM
  }

  if (ha_recover(&xids))
    goto err2;

  hash_free(&xids);
4986
  bzero(data, (size_t)file_length);
4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997
  return 0;

err2:
  hash_free(&xids);
err1:
  sql_print_error("Crash recovery failed. Either correct the problem "
                  "(if it's, for example, out of memory error) and restart, "
                  "or delete tc log and start mysqld with "
                  "--tc-heuristic-recover={commit|rollback}");
  return 1;
}
serg@serg.mylan's avatar
serg@serg.mylan committed
4998 4999 5000 5001 5002
#endif

TC_LOG *tc_log;
TC_LOG_DUMMY tc_log_dummy;
TC_LOG_MMAP  tc_log_mmap;
5003

5004 5005
/**
  Perform heuristic recovery, if --tc-heuristic-recover was used.
5006

5007
  @note
5008 5009
    no matter whether heuristic recovery was successful or not
    mysqld must exit. So, return value is the same in both cases.
5010 5011 5012 5013 5014

  @retval
    0	no heuristic recovery was requested
  @retval
    1   heuristic recovery was performed
5015 5016
*/

5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029
int TC_LOG::using_heuristic_recover()
{
  if (!tc_heuristic_recover)
    return 0;

  sql_print_information("Heuristic crash recovery mode");
  if (ha_recover(0))
    sql_print_error("Heuristic crash recovery failed");
  sql_print_information("Please restart mysqld without --tc-heuristic-recover");
  return 1;
}

/****** transaction coordinator log for 2pc - binlog() based solution ******/
5030
#define TC_LOG_BINLOG MYSQL_BIN_LOG
5031

5032 5033 5034
/**
  @todo
  keep in-memory list of prepared transactions
5035 5036 5037 5038 5039
  (add to list in log(), remove on unlog())
  and copy it to the new binlog if rotated
  but let's check the behaviour of tc_log_page_waits first!
*/

5040
int TC_LOG_BINLOG::open(const char *opt_name)
5041
{
5042 5043
  LOG_INFO log_info;
  int      error= 1;
5044 5045

  DBUG_ASSERT(total_ha_2pc > 1);
5046
  DBUG_ASSERT(opt_name && opt_name[0]);
5047 5048 5049 5050

  pthread_mutex_init(&LOCK_prep_xids, MY_MUTEX_INIT_FAST);
  pthread_cond_init (&COND_prep_xids, 0);

5051 5052 5053 5054 5055 5056 5057
  if (!my_b_inited(&index_file))
  {
    /* There was a failure to open the index file, can't open the binlog */
    cleanup();
    return 1;
  }

5058
  if (using_heuristic_recover())
5059 5060 5061 5062
  {
    /* generate a new binlog to mask a corrupted one */
    open(opt_name, LOG_BIN, 0, WRITE_CACHE, 0, max_binlog_size, 0);
    cleanup();
5063
    return 1;
5064
  }
5065

5066
  if ((error= find_log_pos(&log_info, NullS, 1)))
5067
  {
5068 5069 5070 5071 5072
    if (error != LOG_INFO_EOF)
      sql_print_error("find_log_pos() failed (error: %d)", error);
    else
      error= 0;
    goto err;
5073
  }
5074 5075 5076 5077 5078

  {
    const char *errmsg;
    IO_CACHE    log;
    File        file;
5079 5080
    Log_event  *ev=0;
    Format_description_log_event fdle(BINLOG_VERSION);
5081
    char        log_name[FN_REFLEN];
5082 5083 5084 5085

    if (! fdle.is_valid())
      goto err;

5086
    do
5087
    {
5088 5089 5090 5091 5092 5093 5094
      strmake(log_name, log_info.log_file_name, sizeof(log_name)-1);
    } while (!(error= find_next_log(&log_info, 1)));

    if (error !=  LOG_INFO_EOF)
    {
      sql_print_error("find_log_pos() failed (error: %d)", error);
      goto err;
5095
    }
5096

5097
    if ((file= open_binlog(&log, log_name, &errmsg)) < 0)
5098 5099 5100 5101 5102
    {
      sql_print_error("%s", errmsg);
      goto err;
    }

5103 5104 5105
    if ((ev= Log_event::read_log_event(&log, 0, &fdle)) &&
        ev->get_type_code() == FORMAT_DESCRIPTION_EVENT &&
        ev->flags & LOG_EVENT_BINLOG_IN_USE_F)
serg@serg.mylan's avatar
serg@serg.mylan committed
5106 5107
    {
      sql_print_information("Recovering after a crash using %s", opt_name);
5108
      error= recover(&log, (Format_description_log_event *)ev);
serg@serg.mylan's avatar
serg@serg.mylan committed
5109
    }
5110 5111
    else
      error=0;
5112 5113 5114 5115 5116 5117 5118 5119 5120 5121

    delete ev;
    end_io_cache(&log);
    my_close(file, MYF(MY_WME));

    if (error)
      goto err;
  }

err:
5122
  return error;
5123 5124
}

5125
/** This is called on shutdown, after ha_panic. */
5126
void TC_LOG_BINLOG::close()
5127 5128 5129 5130 5131 5132
{
  DBUG_ASSERT(prepared_xids==0);
  pthread_mutex_destroy(&LOCK_prep_xids);
  pthread_cond_destroy (&COND_prep_xids);
}

5133 5134 5135
/**
  @todo
  group commit
5136

5137 5138 5139 5140
  @retval
    0    error
  @retval
    1    success
5141
*/
5142
int TC_LOG_BINLOG::log_xid(THD *thd, my_xid xid)
5143
{
5144
  DBUG_ENTER("TC_LOG_BINLOG::log");
5145
  Xid_log_event xle(thd, xid);
5146
  binlog_trx_data *trx_data=
5147
    (binlog_trx_data*) thd_get_ha_data(thd, binlog_hton);
5148 5149 5150 5151 5152
  /*
    We always commit the entire transaction when writing an XID. Also
    note that the return value is inverted.
   */
  DBUG_RETURN(!binlog_end_trans(thd, trx_data, &xle, TRUE));
5153 5154
}

5155
void TC_LOG_BINLOG::unlog(ulong cookie, my_xid xid)
5156
{
5157
  pthread_mutex_lock(&LOCK_prep_xids);
5158
  DBUG_ASSERT(prepared_xids > 0);
5159 5160
  if (--prepared_xids == 0) {
    DBUG_PRINT("info", ("prepared_xids=%lu", prepared_xids));
5161
    pthread_cond_signal(&COND_prep_xids);
5162
  }
5163
  pthread_mutex_unlock(&LOCK_prep_xids);
5164
  rotate_and_purge(0);     // as ::write() did not rotate
5165 5166
}

5167
int TC_LOG_BINLOG::recover(IO_CACHE *log, Format_description_log_event *fdle)
5168 5169 5170 5171 5172 5173
{
  Log_event  *ev;
  HASH xids;
  MEM_ROOT mem_root;

  if (! fdle->is_valid() ||
5174
      hash_init(&xids, &my_charset_bin, TC_LOG_PAGE_SIZE/3, 0,
5175
                sizeof(my_xid), 0, 0, MYF(0)))
5176 5177
    goto err1;

5178
  init_alloc_root(&mem_root, TC_LOG_PAGE_SIZE, TC_LOG_PAGE_SIZE);
5179

5180 5181
  fdle->flags&= ~LOG_EVENT_BINLOG_IN_USE_F; // abort on the first error

5182 5183 5184 5185 5186
  while ((ev= Log_event::read_log_event(log,0,fdle)) && ev->is_valid())
  {
    if (ev->get_type_code() == XID_EVENT)
    {
      Xid_log_event *xev=(Xid_log_event *)ev;
5187 5188
      uchar *x= (uchar *) memdup_root(&mem_root, (uchar*) &xev->xid,
                                      sizeof(xev->xid));
5189 5190 5191 5192 5193 5194 5195 5196 5197 5198 5199 5200 5201 5202 5203 5204 5205 5206 5207 5208 5209 5210 5211 5212 5213
      if (! x)
        goto err2;
      my_hash_insert(&xids, x);
    }
    delete ev;
  }

  if (ha_recover(&xids))
    goto err2;

  free_root(&mem_root, MYF(0));
  hash_free(&xids);
  return 0;

err2:
  free_root(&mem_root, MYF(0));
  hash_free(&xids);
err1:
  sql_print_error("Crash recovery failed. Either correct the problem "
                  "(if it's, for example, out of memory error) and restart, "
                  "or delete (or rename) binary log and start mysqld with "
                  "--tc-heuristic-recover={commit|rollback}");
  return 1;
}

5214 5215 5216 5217 5218 5219 5220 5221 5222 5223 5224 5225 5226 5227 5228 5229 5230 5231 5232 5233 5234 5235 5236

#ifdef INNODB_COMPATIBILITY_HOOKS
/**
  Get the file name of the MySQL binlog.
  @return the name of the binlog file
*/
extern "C"
const char* mysql_bin_log_file_name(void)
{
  return mysql_bin_log.get_log_fname();
}
/**
  Get the current position of the MySQL binlog.
  @return byte offset from the beginning of the binlog
*/
extern "C"
ulonglong mysql_bin_log_file_pos(void)
{
  return (ulonglong) mysql_bin_log.get_log_file()->pos_in_file;
}
#endif /* INNODB_COMPATIBILITY_HOOKS */


5237
struct st_mysql_storage_engine binlog_storage_engine=
5238
{ MYSQL_HANDLERTON_INTERFACE_VERSION };
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5239 5240 5241 5242

mysql_declare_plugin(binlog)
{
  MYSQL_STORAGE_ENGINE_PLUGIN,
5243 5244
  &binlog_storage_engine,
  "binlog",
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5245
  "MySQL AB",
5246
  "This is a pseudo storage engine to represent the binlog in a transaction",
5247
  PLUGIN_LICENSE_GPL,
5248
  binlog_init, /* Plugin Init */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5249 5250
  NULL, /* Plugin Deinit */
  0x0100 /* 1.0 */,
5251 5252 5253
  NULL,                       /* status variables                */
  NULL,                       /* system variables                */
  NULL                        /* config options                  */
acurtis@xiphis.org's avatar
acurtis@xiphis.org committed
5254 5255
}
mysql_declare_plugin_end;